sanchayanmaity/minix
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity
minix/kernel/proc.c

1794 lines
54 KiB
C
Raw Normal View History

Initial revision
2005-04-21 16:53:53 +02:00
/* This file contains essentially all of the process and message handling.
Fixed clear_proc(): properly release resources (IRQ hooks, notify buffers); Cleaned up comments in proc.c (old code still present for comparison);
2005-05-30 13:05:42 +02:00
* Together with "mpx.s" it forms the lowest layer of the MINIX kernel.
* There is one entry point from the outside:
Initial revision
2005-04-21 16:53:53 +02:00
*
Formatting fixes for the book
2005-08-29 18:47:18 +02:00
* sys_call: a system call, i.e., the kernel is trapped with an INT
Initial revision
2005-04-21 16:53:53 +02:00
*
* Changes:
Scheduling updates to the kernel. Sched() function now is single point for policy. Actual policy not yet implemented. PM calculates nice values for processes in boot image. IS debug dumps improved (Shift+F1-F4).
2005-08-22 17:14:11 +02:00
* Aug 19, 2005 rewrote scheduling code (Jorrit N. Herder)
* Jul 25, 2005 rewrote system call handling (Jorrit N. Herder)
New scheduling code in kernel. Work in progress. Round-robin within one priority queue works fine. Ageing algorithm to be done.
2005-08-19 18:43:28 +02:00
* May 26, 2005 rewrote message passing functions (Jorrit N. Herder)
* May 24, 2005 new notification system call (Jorrit N. Herder)
* Oct 28, 2004 nonblocking send and receive calls (Jorrit N. Herder)
Fixed clear_proc(): properly release resources (IRQ hooks, notify buffers); Cleaned up comments in proc.c (old code still present for comparison);
2005-05-30 13:05:42 +02:00
*
* The code here is critical to make everything work and is important for the
* overall performance of the system. A large fraction of the code deals with
* list manipulation. To make this both easy to understand and fast to execute
* pointer pointers are used throughout the code. Pointer pointers prevent
* exceptions for the head or tail of a linked list.
*
* node_t *queue, *new_node; // assume these as global variables
* node_t **xpp = &queue; // get pointer pointer to head of queue
* while (*xpp != NULL) // find last pointer of the linked list
* xpp = &(*xpp)->next; // get pointer to next pointer
* *xpp = new_node; // now replace the end (the NULL pointer)
* new_node->next = NULL; // and mark the new end of the list
*
* For example, when adding a new node to the end of the list, one normally
* makes an exception for an empty list and looks up the end of the list for
* nonempty lists. As shown above, this is not required with pointer pointers.
Initial revision
2005-04-21 16:53:53 +02:00
*/
#include <minix/com.h>
'proc number' is process slot, 'endpoint' are generation-aware process instance numbers, encoded and decoded using macros in <minix/endpoint.h>. proc number -> endpoint migration . proc_nr in the interrupt hook is now an endpoint, proc_nr_e. . m_source for messages and notifies is now an endpoint, instead of proc number. . isokendpt() converts an endpoint to a process number, returns success (but fails if the process number is out of range, the process slot is not a living process, or the given endpoint number does not match the endpoint number in the process slot, indicating an old process). . okendpt() is the same as isokendpt(), but panic()s if the conversion fails. This is mainly used for decoding message.m_source endpoints, and other endpoint numbers in kernel data structures, which should always be correct. . if DEBUG_ENABLE_IPC_WARNINGS is enabled, isokendpt() and okendpt() get passed the __FILE__ and __LINE__ of the calling lines, and print messages about what is wrong with the endpoint number (out of range proc, empty proc, or inconsistent endpoint number), with the caller, making finding where the conversion failed easy without having to include code for every call to print where things went wrong. Sometimes this is harmless (wrong arg to a kernel call), sometimes it's a fatal internal inconsistency (bogus m_source). . some process table fields have been appended an _e to indicate it's become and endpoint. . process endpoint is stored in p_endpoint, without generation number. it turns out the kernel never needs the generation number, except when fork()ing, so it's decoded then. . kernel calls all take endpoints as arguments, not proc numbers. the one exception is sys_fork(), which needs to know in which slot to put the child.
2006-03-03 11:00:02 +01:00
#include <minix/endpoint.h>
kernel - prettier output for ipc errors, call names instead of trap numbers
2010-07-16 17:36:29 +02:00
#include <minix/ipcconst.h>
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
#include <stddef.h>
Allow a process to kill itself. Remove the endpoint of a process that kills itself before reporting the singal to pm.
2006-03-15 13:01:59 +01:00
#include <signal.h>
Add 'getidle' CPU utilization measurement infrastructure
2009-12-02 12:52:26 +01:00
#include <minix/syslib.h>
re-establish kernel assert()s. use the regular <assert.h> assert() instead of vmassert() in kernel. throw out some #if 0 code. fix a few assert() conditions. enable by default.
2010-03-10 14:00:05 +01:00
#include <assert.h>
Initial revision
2005-04-21 16:53:53 +02:00
Basic VM and other minor improvements. Not complete, probably not fully debugged or optimized.
2008-11-19 13:26:10 +01:00
#include "debug.h"
#include "kernel.h"
#include "proc.h"
#include "vm.h"
scheduling - time quantum in miliseconds - Currently the cpu time quantum is timer-ticks based. Thus the remaining quantum is decreased only if the processes is interrupted by a timer tick. As processes block a lot this typically does not happen for normal user processes. Also the quantum depends on the frequency of the timer. - This change makes the quantum miliseconds based. Internally the miliseconds are translated into cpu cycles. Everytime userspace execution is interrupted by kernel the cycles just consumed by the current process are deducted from the remaining quantum. - It makes the quantum system timer frequency independent. - The boot processes quantum is loosely derived from the tick-based quantas and 60Hz timer and subject to future change - the 64bit arithmetics is a little ugly, will be changes once we have compiler support for 64bit integers (soon)
2010-05-25 10:06:14 +02:00
#include "clock.h"
SMP - We boot APs - kernel detects CPUs by searching ACPI tables for local apic nodes - each CPU has its own TSS that points to its own stack. All cpus boot on the same boot stack (in sequence) but switch to its private stack as soon as they can. - final booting code in main() placed in bsp_finish_booting() which is executed only after the BSP switches to its final stack - apic functions to send startup interrupts - assembler functions to handle CPU features not needed for single cpu mode like memory barries, HT detection etc. - new files kernel/smp.[ch], kernel/arch/i386/arch_smp.c and kernel/arch/i386/include/arch_smp.h - 16-bit trampoline code for the APs. It is executed by each AP after receiving startup IPIs it brings up the CPUs to 32bit mode and let them spin in an infinite loop so they don't do any damage. - implementation of kernel spinlock - CONFIG_SMP and CONFIG_MAX_CPUS set by the build system
2010-09-15 16:09:52 +02:00
#include "spinlock.h"
Busy idle loop when profiling - the Intel architecture cycle counter (performance counter) does not count when the CPU is idle therefore we use busy loop instead of halting the cpu when there is nothing to schedule - the downside is that handling interrupts may be accounted as idle time if a sample is taken before we get out of the nested trap and pick a new process
2010-09-23 12:49:52 +02:00
#include "profile.h"
Basic VM and other minor improvements. Not complete, probably not fully debugged or optimized.
2008-11-19 13:26:10 +01:00
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
#include "arch_proto.h"
No locking in kernel code - No locking in RTS_(UN)SET macros - No lock_notify() - Removed unused lock_send() - No lock/unlock macros anymore
2010-02-09 16:26:58 +01:00
/* Scheduling and message passing functions */
Add 'getidle' CPU utilization measurement infrastructure
2009-12-02 12:52:26 +01:00
FORWARD _PROTOTYPE( void idle, (void));
Userspace scheduling - cotributed by Bjorn Swift - In this first phase, scheduling is moved from the kernel to the PM server. The next steps are to a) moving scheduling to its own server and b) include useful information in the "out of quantum" message, so that the scheduler can make use of this information. - The kernel process table now keeps record of who is responsible for scheduling each process (p_scheduler). When this pointer is NULL, the process will be scheduled by the kernel. If such a process runs out of quantum, the kernel will simply renew its quantum an requeue it. - When PM loads, it will take over scheduling of all running processes, except system processes, using sys_schedctl(). Essentially, this only results in taking over init. As children inherit a scheduler from their parent, user space programs forked by init will inherit PM (for now) as their scheduler. - Once a process has been assigned a scheduler, and runs out of quantum, its RTS_NO_QUANTUM flag will be set and the process dequeued. The kernel will send a message to the scheduler, on the process' behalf, informing the scheduler that it has run out of quantum. The scheduler can take what ever action it pleases, based on its policy, and then reschedule the process using the sys_schedule() system call. - Balance queues does not work as before. While the old in-kernel function used to renew the quantum of processes in the highest priority run queue, the user-space implementation only acts on processes that have been bumped down to a lower priority queue. This approach reacts slower to changes than the old one, but saves us sending a sys_schedule message for each process every time we balance the queues. Currently, when processes are moved up a priority queue, their quantum is also renewed, but this can be fiddled with. - do_nice has been removed from kernel. PM answers to get- and setpriority calls, updates it's own nice variable as well as the max_run_queue. This will be refactored once scheduling is moved to a separate server. We will probably have PM update it's local nice value and then send a message to whoever is scheduling the process. - changes to fix an issue in do_fork() where processes could run out of quantum but bypassing the code path that handles it correctly. The future plan is to remove the policy from do_fork() and implement it in userspace too.
2010-03-29 13:07:20 +02:00
/**
* Made public for use in clock.c (for user-space scheduling)
Use endpoint_t for the destination of mini_send and _syscall, and the source of mini_receive. Also some small cleanup.
2010-06-02 23:51:32 +02:00
FORWARD _PROTOTYPE( int mini_send, (struct proc *caller_ptr, endpoint_t dst_e,
Basic VM and other minor improvements. Not complete, probably not fully debugged or optimized.
2008-11-19 13:26:10 +01:00
message *m_ptr, int flags));
Userspace scheduling - cotributed by Bjorn Swift - In this first phase, scheduling is moved from the kernel to the PM server. The next steps are to a) moving scheduling to its own server and b) include useful information in the "out of quantum" message, so that the scheduler can make use of this information. - The kernel process table now keeps record of who is responsible for scheduling each process (p_scheduler). When this pointer is NULL, the process will be scheduled by the kernel. If such a process runs out of quantum, the kernel will simply renew its quantum an requeue it. - When PM loads, it will take over scheduling of all running processes, except system processes, using sys_schedctl(). Essentially, this only results in taking over init. As children inherit a scheduler from their parent, user space programs forked by init will inherit PM (for now) as their scheduler. - Once a process has been assigned a scheduler, and runs out of quantum, its RTS_NO_QUANTUM flag will be set and the process dequeued. The kernel will send a message to the scheduler, on the process' behalf, informing the scheduler that it has run out of quantum. The scheduler can take what ever action it pleases, based on its policy, and then reschedule the process using the sys_schedule() system call. - Balance queues does not work as before. While the old in-kernel function used to renew the quantum of processes in the highest priority run queue, the user-space implementation only acts on processes that have been bumped down to a lower priority queue. This approach reacts slower to changes than the old one, but saves us sending a sys_schedule message for each process every time we balance the queues. Currently, when processes are moved up a priority queue, their quantum is also renewed, but this can be fiddled with. - do_nice has been removed from kernel. PM answers to get- and setpriority calls, updates it's own nice variable as well as the max_run_queue. This will be refactored once scheduling is moved to a separate server. We will probably have PM update it's local nice value and then send a message to whoever is scheduling the process. - changes to fix an issue in do_fork() where processes could run out of quantum but bypassing the code path that handles it correctly. The future plan is to remove the policy from do_fork() and implement it in userspace too.
2010-03-29 13:07:20 +02:00
*/
Use endpoint_t for the destination of mini_send and _syscall, and the source of mini_receive. Also some small cleanup.
2010-06-02 23:51:32 +02:00
FORWARD _PROTOTYPE( int mini_receive, (struct proc *caller_ptr, endpoint_t src,
Basic VM and other minor improvements. Not complete, probably not fully debugged or optimized.
2008-11-19 13:26:10 +01:00
message *m_ptr, int flags));
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
FORWARD _PROTOTYPE( int mini_senda, (struct proc *caller_ptr,
asynmsg_t *table, size_t size));
New deadlock check at proc.c. New exit checks (stop IPC).
2005-10-12 17:08:23 +02:00
FORWARD _PROTOTYPE( int deadlock, (int function,
kernel: deadlock test with endpoints instead of slot numbers, slightly cleaner
2010-07-28 16:14:06 +02:00
register struct proc *caller, endpoint_t src_dst_e));
Enable sending a notification when sending of an asynchronous message was completed (successfully or not). AMF_NOTIFY_ERR can be used if the sender only wishes to be notified in case of an error (e.g., EDEADSRCDST). A new endpoint ASYNCM will be the sender of the notification.
2011-04-08 17:14:48 +02:00
FORWARD _PROTOTYPE( int try_async, (struct proc *caller_ptr) );
Use a bitmap for pending asynchronous messages instead of a global flag. That way it works similar to pending notifications.
2011-04-08 17:03:33 +02:00
FORWARD _PROTOTYPE( int try_one, (struct proc *src_ptr, struct proc *dst_ptr));
pick_proc() called only just before returning to userspace - new proc_is_runnable() macro to test whether process is runnable. All tests whether p_rts_flags == 0 converted to use this macro - pick_proc() calls removed from enqueue() and dequeue() - removed the test for recursive calls from pick_proc() as it certainly cannot be called recursively now - PREEMPTED flag to mark processes that were preempted by enqueueuing a higher priority process in enqueue() - enqueue_head() to enqueue PREEMPTED processes again at the head of their current priority queue - NO_QUANTUM flag to block and dequeue processes preempted by timer tick with exceeded quantum. They need to be enqueued again in schedcheck() - next_ptr global variable removed
2009-11-09 18:48:31 +01:00
FORWARD _PROTOTYPE( struct proc * pick_proc, (void));
FORWARD _PROTOTYPE( void enqueue_head, (struct proc *rp));
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
SMP - CPU local idle stub - each CPU has its own pseudo idle process and its structure - idle cycles accounting is agregated when exporting to userspace
2010-09-15 16:10:21 +02:00
/* all idles share the same idle_priv structure */
PRIVATE struct priv idle_priv;
PRIVATE void set_idle_name(char * name, int n)
{
int i, c;
int p_z = 0;
/*
* P_NAME_LEN limits us to 3 characters for the idle task numer. 999
* should be enough though.
*/
if (n > 999)
n = 999;
name[0] = 'i';
name[1] = 'd';
name[2] = 'l';
name[3] = 'e';
for (i = 4, c = 100; c > 0; c /= 10) {
int digit;
digit = n / c;
n -= digit * c;
if (p_z || digit != 0 || c == 1) {
p_z = 1;
name[i++] = '0' + digit;
}
}
name[i] = '\0';
}
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
#define PICK_ANY 1
#define PICK_HIGHERONLY 2
#define BuildNotifyMessage(m_ptr, src, dst_ptr) \
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
(m_ptr)->m_type = NOTIFY_FROM(src); \
(m_ptr)->NOTIFY_TIMESTAMP = get_uptime(); \
System processes can be signaled; signals are transformed in SYS_EVENT message that passes signal map along. This mechanisms is also used for nonuser signals like SIGKMESS, SIGKSTOP, SIGKSIG. Revised comments of many system call handlers. Renamed setpriority to nice.
2005-07-19 14:21:36 +02:00
switch (src) { \
case HARDWARE: \
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
(m_ptr)->NOTIFY_ARG = priv(dst_ptr)->s_int_pending; \
priv(dst_ptr)->s_int_pending = 0; \
System processes can be signaled; signals are transformed in SYS_EVENT message that passes signal map along. This mechanisms is also used for nonuser signals like SIGKMESS, SIGKSTOP, SIGKSIG. Revised comments of many system call handlers. Renamed setpriority to nice.
2005-07-19 14:21:36 +02:00
break; \
case SYSTEM: \
(m_ptr)->NOTIFY_ARG = priv(dst_ptr)->s_sig_pending; \
priv(dst_ptr)->s_sig_pending = 0; \
break; \
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
}
proc_init() - code that initializes proc.c structures removed from main() and placed in proc_init() function
2010-09-15 16:09:43 +02:00
PUBLIC void proc_init(void)
{
struct proc * rp;
struct priv *sp;
int i;
/* Clear the process table. Anounce each slot as empty and set up
* mappings for proc_addr() and proc_nr() macros. Do the same for the
* table with privilege structures for the system processes.
*/
for (rp = BEG_PROC_ADDR, i = -NR_TASKS; rp < END_PROC_ADDR; ++rp, ++i) {
rp->p_rts_flags = RTS_SLOT_FREE;/* initialize free slot */
rp->p_magic = PMAGIC;
rp->p_nr = i; /* proc number from ptr */
rp->p_endpoint = _ENDPOINT(0, rp->p_nr); /* generation no. 0 */
rp->p_scheduler = NULL; /* no user space scheduler */
rp->p_priority = 0; /* no priority */
rp->p_quantum_size_ms = 0; /* no quantum size */
}
for (sp = BEG_PRIV_ADDR, i = 0; sp < END_PRIV_ADDR; ++sp, ++i) {
sp->s_proc_nr = NONE; /* initialize as free */
sp->s_id = (sys_id_t) i; /* priv structure index */
ppriv_addr[i] = sp; /* priv ptr from number */
sp->s_sig_mgr = NONE; /* clear signal managers */
sp->s_bak_sig_mgr = NONE;
}
SMP - CPU local idle stub - each CPU has its own pseudo idle process and its structure - idle cycles accounting is agregated when exporting to userspace
2010-09-15 16:10:21 +02:00
idle_priv.s_flags = IDL_F;
/* initialize IDLE structures for every CPU */
for (i = 0; i < CONFIG_MAX_CPUS; i++) {
struct proc * ip = get_cpu_var_ptr(i, idle_proc);
SMP - trully idle APs - any cpu can use smp_schedule() to tell another cpu to reschedule - if an AP is idle, it turns off timer as there is nothing to preempt, no need to wakeup just to go back to sleep again - if a cpu makes a process runnable on an idle cpu, it must wake it up to reschedule
2010-09-15 16:10:57 +02:00
ip->p_endpoint = IDLE;
SMP - CPU local idle stub - each CPU has its own pseudo idle process and its structure - idle cycles accounting is agregated when exporting to userspace
2010-09-15 16:10:21 +02:00
ip->p_priv = &idle_priv;
SMP - proc_ptr and bill_ptr initialization - they should point somewhere
2010-09-15 16:10:24 +02:00
/* must not let idle ever get scheduled */
ip->p_rts_flags |= RTS_PROC_STOP;
SMP - CPU local idle stub - each CPU has its own pseudo idle process and its structure - idle cycles accounting is agregated when exporting to userspace
2010-09-15 16:10:21 +02:00
set_idle_name(ip->p_name, i);
}
SMP - lazy FPU - when a process is migrated to a different CPU it may have an active FPU context in the processor registers. We must save it and migrate it together with the process.
2010-09-15 16:11:25 +02:00
for (rp = BEG_PROC_ADDR; rp < END_PROC_ADDR; ++rp) {
/*
* FXSR requires 16-byte alignment of memory image, but
* unfortunately a.out does not preserve the alignment while
* linking. Thus we have to do manual alignment.
*/
phys_bytes aligned_fp_area;
aligned_fp_area =
(phys_bytes) &rp->p_fpu_state.fpu_image;
if(aligned_fp_area % FPUALIGN) {
aligned_fp_area += FPUALIGN -
(aligned_fp_area % FPUALIGN);
}
rp->p_fpu_state.fpu_save_area_p =
(void *) aligned_fp_area;
}
proc_init() - code that initializes proc.c structures removed from main() and placed in proc_init() function
2010-09-15 16:09:43 +02:00
}
SMP - CPU local run queues - each CPU has its own runqueues - processes on BSP are put on the runqueues later after a switch to the final stack when cpuid works to avoid special cases - enqueue() and dequeue() use the run queues of the cpu the process is assigned to - pick_proc() uses the local run queues - printing of per-CPU run queues ('2') on serial console
2010-09-15 16:10:18 +02:00
PRIVATE void switch_address_space_idle(void)
{
#ifdef CONFIG_SMP
/*
* currently we bet that VM is always alive and its pages available so
* when the CPU wakes up the kernel is mapped and no surprises happen.
* This is only a problem if more than 1 cpus are available
*/
switch_address_space(proc_addr(VM_PROC_NR));
#endif
}
Add 'getidle' CPU utilization measurement infrastructure
2009-12-02 12:52:26 +01:00
/*===========================================================================*
* idle *
*===========================================================================*/
Time accounting based on TSC - as thre are still KERNEL and IDLE entries, time accounting for kernel and idle time works the same as for any other process - everytime we stop accounting for the currently running process, kernel or idle, we read the TSC counter and increment the p_cycles entry. - the process cycles inherently include some of the kernel cycles as we can stop accounting for the process only after we save its context and we start accounting just before we restore its context - this assumes that the system does not scale the CPU frequency which will be true for ... long time ;-)
2010-02-10 16:36:54 +01:00
PRIVATE void idle(void)
Add 'getidle' CPU utilization measurement infrastructure
2009-12-02 12:52:26 +01:00
{
SMP - CPU local idle stub - each CPU has its own pseudo idle process and its structure - idle cycles accounting is agregated when exporting to userspace
2010-09-15 16:10:21 +02:00
struct proc * p;
Add 'getidle' CPU utilization measurement infrastructure
2009-12-02 12:52:26 +01:00
/* This function is called whenever there is no work to do.
* Halt the CPU, and measure how many timestamp counter ticks are
* spent not doing anything. This allows test setups to measure
* the CPU utiliziation of certain workloads with high precision.
*/
SMP - CPU local idle stub - each CPU has its own pseudo idle process and its structure - idle cycles accounting is agregated when exporting to userspace
2010-09-15 16:10:21 +02:00
p = get_cpulocal_var(proc_ptr) = get_cpulocal_var_ptr(idle_proc);
if (priv(p)->s_flags & BILLABLE)
get_cpulocal_var(bill_ptr) = p;
SMP - CPU local run queues - each CPU has its own runqueues - processes on BSP are put on the runqueues later after a switch to the final stack when cpuid works to avoid special cases - enqueue() and dequeue() use the run queues of the cpu the process is assigned to - pick_proc() uses the local run queues - printing of per-CPU run queues ('2') on serial console
2010-09-15 16:10:18 +02:00
switch_address_space_idle();
SMP - Compiles for both single and multi processor again - this patch adds various fixes as some of the previous patches break compilations without CONFIG_SMP being set
2010-09-15 16:11:03 +02:00
#ifdef CONFIG_SMP
APIC timer needs rearming before halting the cpu - time stops if there is no activity and the timer expired before we halted the cpu - restart_local_timer() checks if the timer has expired and if so it restarts it - we do the same when switching back to userspace
2011-05-11 11:54:23 +02:00
get_cpulocal_var(cpu_is_idle) = 1;
SMP - trully idle APs - any cpu can use smp_schedule() to tell another cpu to reschedule - if an AP is idle, it turns off timer as there is nothing to preempt, no need to wakeup just to go back to sleep again - if a cpu makes a process runnable on an idle cpu, it must wake it up to reschedule
2010-09-15 16:10:57 +02:00
/* we don't need to keep time on APs as it is handled on the BSP */
if (cpuid != bsp_cpu_id)
SMP - Single shot local timer - APIC timer always reprogrammed if expired - timer tick never happens when in kernel => never immediate return from userspace to kernel because of a buffered interrupt - renamed argument to lapic_set_timer_one_shot() - removed arch_ prefix from timer functions
2010-09-15 16:11:06 +02:00
stop_local_timer();
APIC timer needs rearming before halting the cpu - time stops if there is no activity and the timer expired before we halted the cpu - restart_local_timer() checks if the timer has expired and if so it restarts it - we do the same when switching back to userspace
2011-05-11 11:54:23 +02:00
else
SMP - Compiles for both single and multi processor again - this patch adds various fixes as some of the previous patches break compilations without CONFIG_SMP being set
2010-09-15 16:11:03 +02:00
#endif
APIC timer needs rearming before halting the cpu - time stops if there is no activity and the timer expired before we halted the cpu - restart_local_timer() checks if the timer has expired and if so it restarts it - we do the same when switching back to userspace
2011-05-11 11:54:23 +02:00
{
/*
* If the timer has expired while in kernel we must
* rearm it before we go to sleep
*/
restart_local_timer();
}
SMP - trully idle APs - any cpu can use smp_schedule() to tell another cpu to reschedule - if an AP is idle, it turns off timer as there is nothing to preempt, no need to wakeup just to go back to sleep again - if a cpu makes a process runnable on an idle cpu, it must wake it up to reschedule
2010-09-15 16:10:57 +02:00
Time accounting based on TSC - as thre are still KERNEL and IDLE entries, time accounting for kernel and idle time works the same as for any other process - everytime we stop accounting for the currently running process, kernel or idle, we read the TSC counter and increment the p_cycles entry. - the process cycles inherently include some of the kernel cycles as we can stop accounting for the process only after we save its context and we start accounting just before we restore its context - this assumes that the system does not scale the CPU frequency which will be true for ... long time ;-)
2010-02-10 16:36:54 +01:00
/* start accounting for the idle time */
rename of mode/context switching functions - this patch only renames schedcheck() to switch_to_user(), cycles_accounting_stop() to context_stop() and restart() to +restore_user_context() - the motivation is that since the introduction of schedcheck() it has been abused for many things. It deserves a better name. It should express the fact that from the moment we call the function we are in the process of switching to user. - cycles_accounting_stop() was originally a single purpose function. As this function is called at were convenient places it is used in for other things too, e.g. (un)locking the kernel. Thus it deserves a better name too. - using the old name, restart() does not call schedcheck(), however calls to restart are replaced by calls to schedcheck() [switch_to_user] and it calls restart() [restore_user_context]
2010-05-18 15:00:39 +02:00
context_stop(proc_addr(KERNEL));
Fix for SPROFILE == 0 - contributed by Antoine Leca
2011-05-25 09:42:11 +02:00
#if !SPROFILE
halt_cpu();
#else
Busy idle loop when profiling - the Intel architecture cycle counter (performance counter) does not count when the CPU is idle therefore we use busy loop instead of halting the cpu when there is nothing to schedule - the downside is that handling interrupts may be accounted as idle time if a sample is taken before we get out of the nested trap and pick a new process
2010-09-23 12:49:52 +02:00
if (!sprofiling)
halt_cpu();
else {
volatile int * v;
v = get_cpulocal_var_ptr(idle_interrupted);
interrupts_enable();
while (!*v)
arch_pause();
interrupts_disable();
*v = 0;
}
Fix for SPROFILE == 0 - contributed by Antoine Leca
2011-05-25 09:42:11 +02:00
#endif
Time accounting based on TSC - as thre are still KERNEL and IDLE entries, time accounting for kernel and idle time works the same as for any other process - everytime we stop accounting for the currently running process, kernel or idle, we read the TSC counter and increment the p_cycles entry. - the process cycles inherently include some of the kernel cycles as we can stop accounting for the process only after we save its context and we start accounting just before we restore its context - this assumes that the system does not scale the CPU frequency which will be true for ... long time ;-)
2010-02-10 16:36:54 +01:00
/*
* end of accounting for the idle task does not happen here, the kernel
* is handling stuff for quite a while before it gets back here!
*/
Add 'getidle' CPU utilization measurement infrastructure
2009-12-02 12:52:26 +01:00
}
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
/*===========================================================================*
rename of mode/context switching functions - this patch only renames schedcheck() to switch_to_user(), cycles_accounting_stop() to context_stop() and restart() to +restore_user_context() - the motivation is that since the introduction of schedcheck() it has been abused for many things. It deserves a better name. It should express the fact that from the moment we call the function we are in the process of switching to user. - cycles_accounting_stop() was originally a single purpose function. As this function is called at were convenient places it is used in for other things too, e.g. (un)locking the kernel. Thus it deserves a better name too. - using the old name, restart() does not call schedcheck(), however calls to restart are replaced by calls to schedcheck() [switch_to_user] and it calls restart() [restore_user_context]
2010-05-18 15:00:39 +02:00
* switch_to_user *
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
*===========================================================================*/
rename of mode/context switching functions - this patch only renames schedcheck() to switch_to_user(), cycles_accounting_stop() to context_stop() and restart() to +restore_user_context() - the motivation is that since the introduction of schedcheck() it has been abused for many things. It deserves a better name. It should express the fact that from the moment we call the function we are in the process of switching to user. - cycles_accounting_stop() was originally a single purpose function. As this function is called at were convenient places it is used in for other things too, e.g. (un)locking the kernel. Thus it deserves a better name too. - using the old name, restart() does not call schedcheck(), however calls to restart are replaced by calls to schedcheck() [switch_to_user] and it calls restart() [restore_user_context]
2010-05-18 15:00:39 +02:00
PUBLIC void switch_to_user(void)
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
{
/* This function is called an instant before proc_ptr is
* to be scheduled again.
*/
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
struct proc * p;
SMP - reduced TLB flushing - flush TLB of processes only if the page tables has been changed and the page tables of this process are already loaded on this cpu which means that there might be stale entries in TLB. Until now SMP was always flushing TLB to make sure everything is consistent.
2010-10-25 18:21:23 +02:00
#ifdef CONFIG_SMP
int tlb_must_refresh = 0;
#endif
pick_proc() called only just before returning to userspace - new proc_is_runnable() macro to test whether process is runnable. All tests whether p_rts_flags == 0 converted to use this macro - pick_proc() calls removed from enqueue() and dequeue() - removed the test for recursive calls from pick_proc() as it certainly cannot be called recursively now - PREEMPTED flag to mark processes that were preempted by enqueueuing a higher priority process in enqueue() - enqueue_head() to enqueue PREEMPTED processes again at the head of their current priority queue - NO_QUANTUM flag to block and dequeue processes preempted by timer tick with exceeded quantum. They need to be enqueued again in schedcheck() - next_ptr global variable removed
2009-11-09 18:48:31 +01:00
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
p = get_cpulocal_var(proc_ptr);
pick_proc() called only just before returning to userspace - new proc_is_runnable() macro to test whether process is runnable. All tests whether p_rts_flags == 0 converted to use this macro - pick_proc() calls removed from enqueue() and dequeue() - removed the test for recursive calls from pick_proc() as it certainly cannot be called recursively now - PREEMPTED flag to mark processes that were preempted by enqueueuing a higher priority process in enqueue() - enqueue_head() to enqueue PREEMPTED processes again at the head of their current priority queue - NO_QUANTUM flag to block and dequeue processes preempted by timer tick with exceeded quantum. They need to be enqueued again in schedcheck() - next_ptr global variable removed
2009-11-09 18:48:31 +01:00
/*
* if the current process is still runnable check the misc flags and let
* it run unless it becomes not runnable in the meantime
*/
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
if (proc_is_runnable(p))
pick_proc() called only just before returning to userspace - new proc_is_runnable() macro to test whether process is runnable. All tests whether p_rts_flags == 0 converted to use this macro - pick_proc() calls removed from enqueue() and dequeue() - removed the test for recursive calls from pick_proc() as it certainly cannot be called recursively now - PREEMPTED flag to mark processes that were preempted by enqueueuing a higher priority process in enqueue() - enqueue_head() to enqueue PREEMPTED processes again at the head of their current priority queue - NO_QUANTUM flag to block and dequeue processes preempted by timer tick with exceeded quantum. They need to be enqueued again in schedcheck() - next_ptr global variable removed
2009-11-09 18:48:31 +01:00
goto check_misc_flags;
/*
* if a process becomes not runnable while handling the misc flags, we
* need to pick a new one here and start from scratch. Also if the
* current process wasn' runnable, we pick a new one here
*/
not_runnable_pick_new:
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
if (proc_is_preempted(p)) {
p->p_rts_flags &= ~RTS_PREEMPTED;
if (proc_is_runnable(p)) {
if (!is_zero64(p->p_cpu_time_left))
enqueue_head(p);
scheduling - time quantum in miliseconds - Currently the cpu time quantum is timer-ticks based. Thus the remaining quantum is decreased only if the processes is interrupted by a timer tick. As processes block a lot this typically does not happen for normal user processes. Also the quantum depends on the frequency of the timer. - This change makes the quantum miliseconds based. Internally the miliseconds are translated into cpu cycles. Everytime userspace execution is interrupted by kernel the cycles just consumed by the current process are deducted from the remaining quantum. - It makes the quantum system timer frequency independent. - The boot processes quantum is loosely derived from the tick-based quantas and 60Hz timer and subject to future change - the 64bit arithmetics is a little ugly, will be changes once we have compiler support for 64bit integers (soon)
2010-05-25 10:06:14 +02:00
else
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
enqueue(p);
scheduling - time quantum in miliseconds - Currently the cpu time quantum is timer-ticks based. Thus the remaining quantum is decreased only if the processes is interrupted by a timer tick. As processes block a lot this typically does not happen for normal user processes. Also the quantum depends on the frequency of the timer. - This change makes the quantum miliseconds based. Internally the miliseconds are translated into cpu cycles. Everytime userspace execution is interrupted by kernel the cycles just consumed by the current process are deducted from the remaining quantum. - It makes the quantum system timer frequency independent. - The boot processes quantum is loosely derived from the tick-based quantas and 60Hz timer and subject to future change - the 64bit arithmetics is a little ugly, will be changes once we have compiler support for 64bit integers (soon)
2010-05-25 10:06:14 +02:00
}
Userspace scheduling - cotributed by Bjorn Swift - In this first phase, scheduling is moved from the kernel to the PM server. The next steps are to a) moving scheduling to its own server and b) include useful information in the "out of quantum" message, so that the scheduler can make use of this information. - The kernel process table now keeps record of who is responsible for scheduling each process (p_scheduler). When this pointer is NULL, the process will be scheduled by the kernel. If such a process runs out of quantum, the kernel will simply renew its quantum an requeue it. - When PM loads, it will take over scheduling of all running processes, except system processes, using sys_schedctl(). Essentially, this only results in taking over init. As children inherit a scheduler from their parent, user space programs forked by init will inherit PM (for now) as their scheduler. - Once a process has been assigned a scheduler, and runs out of quantum, its RTS_NO_QUANTUM flag will be set and the process dequeued. The kernel will send a message to the scheduler, on the process' behalf, informing the scheduler that it has run out of quantum. The scheduler can take what ever action it pleases, based on its policy, and then reschedule the process using the sys_schedule() system call. - Balance queues does not work as before. While the old in-kernel function used to renew the quantum of processes in the highest priority run queue, the user-space implementation only acts on processes that have been bumped down to a lower priority queue. This approach reacts slower to changes than the old one, but saves us sending a sys_schedule message for each process every time we balance the queues. Currently, when processes are moved up a priority queue, their quantum is also renewed, but this can be fiddled with. - do_nice has been removed from kernel. PM answers to get- and setpriority calls, updates it's own nice variable as well as the max_run_queue. This will be refactored once scheduling is moved to a separate server. We will probably have PM update it's local nice value and then send a message to whoever is scheduling the process. - changes to fix an issue in do_fork() where processes could run out of quantum but bypassing the code path that handles it correctly. The future plan is to remove the policy from do_fork() and implement it in userspace too.
2010-03-29 13:07:20 +02:00
}
Idle task never runs - idle task becomes a pseudo task which is never scheduled. It is never put on any run queue and never enters userspace. An entry for this task still remains in the process table for time accounting - Instead of panicing if there is not process to schedule, pick_proc() returns NULL which is a signal to put the cpu in an idle state and set everything in such a way that after receiving and interrupt it looks like idle task was preempted - idle task is set non-preemptible to avoid handling in the timer interrupt code which make userspace scheduling simpler as idle task does not need to be handled as a special case.
2009-11-12 09:42:18 +01:00
/*
* if we have no process to run, set IDLE as the current process for
* time accounting and put the cpu in and idle state. After the next
* timer interrupt the execution resumes here and we can pick another
* process. If there is still nothing runnable we "schedule" IDLE again
*/
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
while (!(p = pick_proc())) {
Add 'getidle' CPU utilization measurement infrastructure
2009-12-02 12:52:26 +01:00
idle();
Idle task never runs - idle task becomes a pseudo task which is never scheduled. It is never put on any run queue and never enters userspace. An entry for this task still remains in the process table for time accounting - Instead of panicing if there is not process to schedule, pick_proc() returns NULL which is a signal to put the cpu in an idle state and set everything in such a way that after receiving and interrupt it looks like idle task was preempted - idle task is set non-preemptible to avoid handling in the timer interrupt code which make userspace scheduling simpler as idle task does not need to be handled as a special case.
2009-11-12 09:42:18 +01:00
}
pick_proc() called only just before returning to userspace - new proc_is_runnable() macro to test whether process is runnable. All tests whether p_rts_flags == 0 converted to use this macro - pick_proc() calls removed from enqueue() and dequeue() - removed the test for recursive calls from pick_proc() as it certainly cannot be called recursively now - PREEMPTED flag to mark processes that were preempted by enqueueuing a higher priority process in enqueue() - enqueue_head() to enqueue PREEMPTED processes again at the head of their current priority queue - NO_QUANTUM flag to block and dequeue processes preempted by timer tick with exceeded quantum. They need to be enqueued again in schedcheck() - next_ptr global variable removed
2009-11-09 18:48:31 +01:00
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
/* update the global variable */
get_cpulocal_var(proc_ptr) = p;
SMP - reduced TLB flushing - flush TLB of processes only if the page tables has been changed and the page tables of this process are already loaded on this cpu which means that there might be stale entries in TLB. Until now SMP was always flushing TLB to make sure everything is consistent.
2010-10-25 18:21:23 +02:00
#ifdef CONFIG_SMP
if (p->p_misc_flags & MF_FLUSH_TLB && get_cpulocal_var(ptproc) == p)
tlb_must_refresh = 1;
#endif
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
switch_address_space(p);
Early address space switch - switch_address_space() implements a switch of the user address space for the destination process - this makes memory of this process easily accessible, e.g. a pointer valid in the userspace can be used with a little complexity to access the process's memory - the switch does not happed only just before we return to userspace, however, it happens right after we know which process we are going to schedule. This happens before we start processing the misc flags of this process so its memory is available - if the process becomes not runnable while processing the mics flags we pick a new process and we switch the address space again which introduces possibly a little bit more overhead, however, it is hopefully hidden by reducing the overheads when we actually access the memory
2010-02-09 16:13:52 +01:00
pick_proc() called only just before returning to userspace - new proc_is_runnable() macro to test whether process is runnable. All tests whether p_rts_flags == 0 converted to use this macro - pick_proc() calls removed from enqueue() and dequeue() - removed the test for recursive calls from pick_proc() as it certainly cannot be called recursively now - PREEMPTED flag to mark processes that were preempted by enqueueuing a higher priority process in enqueue() - enqueue_head() to enqueue PREEMPTED processes again at the head of their current priority queue - NO_QUANTUM flag to block and dequeue processes preempted by timer tick with exceeded quantum. They need to be enqueued again in schedcheck() - next_ptr global variable removed
2009-11-09 18:48:31 +01:00
check_misc_flags:
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
assert(p);
assert(proc_is_runnable(p));
while (p->p_misc_flags &
Removal of the system task * Userspace change to use the new kernel calls - _taskcall(SYSTASK...) changed to _kernel_call(...) - int 32 reused for the kernel calls - _do_kernel_call() to make the trap to kernel - kernel_call() to make the actuall kernel call from C using _do_kernel_call() - unlike ipc call the kernel call always succeeds as kernel is always available, however, kernel may return an error * Kernel side implementation of kernel calls - the SYSTEm task does not run, only the proc table entry is preserved - every data_copy(SYSTEM is no data_copy(KERNEL - "locking" is an empty operation now as everything runs in kernel - sys_task() is replaced by kernel_call() which copies the message into kernel, dispatches the call to its handler and finishes by either copying the results back to userspace (if need be) or by suspending the process because of VM - suspended processes are later made runnable once the memory issue is resolved, picked up by the scheduler and only at this time the call is resumed (in fact restarted) which does not need to copy the message from userspace as the message is already saved in the process structure. - no ned for the vmrestart queue, the scheduler will restart the system calls - no special case in do_vmctl(), all requests remove the RTS_VMREQUEST flag
2010-02-09 16:20:09 +01:00
(MF_KCALL_RESUME | MF_DELIVERMSG |
MF_SC_DEFER | MF_SC_TRACE | MF_SC_ACTIVE)) {
Merge of David's ptrace branch. Summary: o Support for ptrace T_ATTACH/T_DETACH and T_SYSCALL o PM signal handling logic should now work properly, even with debuggers being present o Asynchronous PM/VFS protocol, full IPC support for senda(), and AMF_NOREPLY senda() flag DETAILS Process stop and delay call handling of PM: o Added sys_runctl() kernel call with sys_stop() and sys_resume() aliases, for PM to stop and resume a process o Added exception for sending/syscall-traced processes to sys_runctl(), and matching SIGKREADY pseudo-signal to PM o Fixed PM signal logic to deal with requests from a process after stopping it (so-called "delay calls"), using the SIGKREADY facility o Fixed various PM panics due to race conditions with delay calls versus VFS calls o Removed special PRIO_STOP priority value o Added SYS_LOCK RTS kernel flag, to stop an individual process from running while modifying its process structure Signal and debugger handling in PM: o Fixed debugger signals being dropped if a second signal arrives when the debugger has not retrieved the first one o Fixed debugger signals being sent to the debugger more than once o Fixed debugger signals unpausing process in VFS; removed PM_UNPAUSE_TR protocol message o Detached debugger signals from general signal logic and from being blocked on VFS calls, meaning that even VFS can now be traced o Fixed debugger being unable to receive more than one pending signal in one process stop o Fixed signal delivery being delayed needlessly when multiple signals are pending o Fixed wait test for tracer, which was returning for children that were not waited for o Removed second parallel pending call from PM to VFS for any process o Fixed process becoming runnable between exec() and debugger trap o Added support for notifying the debugger before the parent when a debugged child exits o Fixed debugger death causing child to remain stopped forever o Fixed consistently incorrect use of _NSIG Extensions to ptrace(): o Added T_ATTACH and T_DETACH ptrace request, to attach and detach a debugger to and from a process o Added T_SYSCALL ptrace request, to trace system calls o Added T_SETOPT ptrace request, to set trace options o Added TO_TRACEFORK trace option, to attach automatically to children of a traced process o Added TO_ALTEXEC trace option, to send SIGSTOP instead of SIGTRAP upon a successful exec() of the tracee o Extended T_GETUSER ptrace support to allow retrieving a process's priv structure o Removed T_STOP ptrace request again, as it does not help implementing debuggers properly o Added MINIX3-specific ptrace test (test42) o Added proper manual page for ptrace(2) Asynchronous PM/VFS interface: o Fixed asynchronous messages not being checked when receive() is called with an endpoint other than ANY o Added AMF_NOREPLY senda() flag, preventing such messages from satisfying the receive part of a sendrec() o Added asynsend3() that takes optional flags; asynsend() is now a #define passing in 0 as third parameter o Made PM/VFS protocol asynchronous; reintroduced tell_fs() o Made PM_BASE request/reply number range unique o Hacked in a horrible temporary workaround into RS to deal with newly revealed RS-PM-VFS race condition triangle until VFS is asynchronous System signal handling: o Fixed shutdown logic of device drivers; removed old SIGKSTOP signal o Removed is-superuser check from PM's do_procstat() (aka getsigset()) o Added sigset macros to allow system processes to deal with the full signal set, rather than just the POSIX subset Miscellaneous PM fixes: o Split do_getset into do_get and do_set, merging common code and making structure clearer o Fixed setpriority() being able to put to sleep processes using an invalid parameter, or revive zombie processes o Made find_proc() global; removed obsolete proc_from_pid() o Cleanup here and there Also included: o Fixed false-positive boot order kernel warning o Removed last traces of old NOTIFY_FROM code THINGS OF POSSIBLE INTEREST o It should now be possible to run PM at any priority, even lower than user processes o No assumptions are made about communication speed between PM and VFS, although communication must be FIFO o A debugger will now receive incoming debuggee signals at kill time only; the process may not yet be fully stopped o A first step has been made towards making the SYSTEM task preemptible
2009-09-30 11:57:22 +02:00
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
assert(proc_is_runnable(p));
if (p->p_misc_flags & MF_KCALL_RESUME) {
kernel_call_resume(p);
Removal of the system task * Userspace change to use the new kernel calls - _taskcall(SYSTASK...) changed to _kernel_call(...) - int 32 reused for the kernel calls - _do_kernel_call() to make the trap to kernel - kernel_call() to make the actuall kernel call from C using _do_kernel_call() - unlike ipc call the kernel call always succeeds as kernel is always available, however, kernel may return an error * Kernel side implementation of kernel calls - the SYSTEm task does not run, only the proc table entry is preserved - every data_copy(SYSTEM is no data_copy(KERNEL - "locking" is an empty operation now as everything runs in kernel - sys_task() is replaced by kernel_call() which copies the message into kernel, dispatches the call to its handler and finishes by either copying the results back to userspace (if need be) or by suspending the process because of VM - suspended processes are later made runnable once the memory issue is resolved, picked up by the scheduler and only at this time the call is resumed (in fact restarted) which does not need to copy the message from userspace as the message is already saved in the process structure. - no ned for the vmrestart queue, the scheduler will restart the system calls - no special case in do_vmctl(), all requests remove the RTS_VMREQUEST flag
2010-02-09 16:20:09 +01:00
}
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
else if (p->p_misc_flags & MF_DELIVERMSG) {
Merge of David's ptrace branch. Summary: o Support for ptrace T_ATTACH/T_DETACH and T_SYSCALL o PM signal handling logic should now work properly, even with debuggers being present o Asynchronous PM/VFS protocol, full IPC support for senda(), and AMF_NOREPLY senda() flag DETAILS Process stop and delay call handling of PM: o Added sys_runctl() kernel call with sys_stop() and sys_resume() aliases, for PM to stop and resume a process o Added exception for sending/syscall-traced processes to sys_runctl(), and matching SIGKREADY pseudo-signal to PM o Fixed PM signal logic to deal with requests from a process after stopping it (so-called "delay calls"), using the SIGKREADY facility o Fixed various PM panics due to race conditions with delay calls versus VFS calls o Removed special PRIO_STOP priority value o Added SYS_LOCK RTS kernel flag, to stop an individual process from running while modifying its process structure Signal and debugger handling in PM: o Fixed debugger signals being dropped if a second signal arrives when the debugger has not retrieved the first one o Fixed debugger signals being sent to the debugger more than once o Fixed debugger signals unpausing process in VFS; removed PM_UNPAUSE_TR protocol message o Detached debugger signals from general signal logic and from being blocked on VFS calls, meaning that even VFS can now be traced o Fixed debugger being unable to receive more than one pending signal in one process stop o Fixed signal delivery being delayed needlessly when multiple signals are pending o Fixed wait test for tracer, which was returning for children that were not waited for o Removed second parallel pending call from PM to VFS for any process o Fixed process becoming runnable between exec() and debugger trap o Added support for notifying the debugger before the parent when a debugged child exits o Fixed debugger death causing child to remain stopped forever o Fixed consistently incorrect use of _NSIG Extensions to ptrace(): o Added T_ATTACH and T_DETACH ptrace request, to attach and detach a debugger to and from a process o Added T_SYSCALL ptrace request, to trace system calls o Added T_SETOPT ptrace request, to set trace options o Added TO_TRACEFORK trace option, to attach automatically to children of a traced process o Added TO_ALTEXEC trace option, to send SIGSTOP instead of SIGTRAP upon a successful exec() of the tracee o Extended T_GETUSER ptrace support to allow retrieving a process's priv structure o Removed T_STOP ptrace request again, as it does not help implementing debuggers properly o Added MINIX3-specific ptrace test (test42) o Added proper manual page for ptrace(2) Asynchronous PM/VFS interface: o Fixed asynchronous messages not being checked when receive() is called with an endpoint other than ANY o Added AMF_NOREPLY senda() flag, preventing such messages from satisfying the receive part of a sendrec() o Added asynsend3() that takes optional flags; asynsend() is now a #define passing in 0 as third parameter o Made PM/VFS protocol asynchronous; reintroduced tell_fs() o Made PM_BASE request/reply number range unique o Hacked in a horrible temporary workaround into RS to deal with newly revealed RS-PM-VFS race condition triangle until VFS is asynchronous System signal handling: o Fixed shutdown logic of device drivers; removed old SIGKSTOP signal o Removed is-superuser check from PM's do_procstat() (aka getsigset()) o Added sigset macros to allow system processes to deal with the full signal set, rather than just the POSIX subset Miscellaneous PM fixes: o Split do_getset into do_get and do_set, merging common code and making structure clearer o Fixed setpriority() being able to put to sleep processes using an invalid parameter, or revive zombie processes o Made find_proc() global; removed obsolete proc_from_pid() o Cleanup here and there Also included: o Fixed false-positive boot order kernel warning o Removed last traces of old NOTIFY_FROM code THINGS OF POSSIBLE INTEREST o It should now be possible to run PM at any priority, even lower than user processes o No assumptions are made about communication speed between PM and VFS, although communication must be FIFO o A debugger will now receive incoming debuggee signals at kill time only; the process may not yet be fully stopped o A first step has been made towards making the SYSTEM task preemptible
2009-09-30 11:57:22 +02:00
TRACE(VF_SCHEDULING, printf("delivering to %s / %d\n",
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
p->p_name, p->p_endpoint););
delivermsg(p);
Merge of David's ptrace branch. Summary: o Support for ptrace T_ATTACH/T_DETACH and T_SYSCALL o PM signal handling logic should now work properly, even with debuggers being present o Asynchronous PM/VFS protocol, full IPC support for senda(), and AMF_NOREPLY senda() flag DETAILS Process stop and delay call handling of PM: o Added sys_runctl() kernel call with sys_stop() and sys_resume() aliases, for PM to stop and resume a process o Added exception for sending/syscall-traced processes to sys_runctl(), and matching SIGKREADY pseudo-signal to PM o Fixed PM signal logic to deal with requests from a process after stopping it (so-called "delay calls"), using the SIGKREADY facility o Fixed various PM panics due to race conditions with delay calls versus VFS calls o Removed special PRIO_STOP priority value o Added SYS_LOCK RTS kernel flag, to stop an individual process from running while modifying its process structure Signal and debugger handling in PM: o Fixed debugger signals being dropped if a second signal arrives when the debugger has not retrieved the first one o Fixed debugger signals being sent to the debugger more than once o Fixed debugger signals unpausing process in VFS; removed PM_UNPAUSE_TR protocol message o Detached debugger signals from general signal logic and from being blocked on VFS calls, meaning that even VFS can now be traced o Fixed debugger being unable to receive more than one pending signal in one process stop o Fixed signal delivery being delayed needlessly when multiple signals are pending o Fixed wait test for tracer, which was returning for children that were not waited for o Removed second parallel pending call from PM to VFS for any process o Fixed process becoming runnable between exec() and debugger trap o Added support for notifying the debugger before the parent when a debugged child exits o Fixed debugger death causing child to remain stopped forever o Fixed consistently incorrect use of _NSIG Extensions to ptrace(): o Added T_ATTACH and T_DETACH ptrace request, to attach and detach a debugger to and from a process o Added T_SYSCALL ptrace request, to trace system calls o Added T_SETOPT ptrace request, to set trace options o Added TO_TRACEFORK trace option, to attach automatically to children of a traced process o Added TO_ALTEXEC trace option, to send SIGSTOP instead of SIGTRAP upon a successful exec() of the tracee o Extended T_GETUSER ptrace support to allow retrieving a process's priv structure o Removed T_STOP ptrace request again, as it does not help implementing debuggers properly o Added MINIX3-specific ptrace test (test42) o Added proper manual page for ptrace(2) Asynchronous PM/VFS interface: o Fixed asynchronous messages not being checked when receive() is called with an endpoint other than ANY o Added AMF_NOREPLY senda() flag, preventing such messages from satisfying the receive part of a sendrec() o Added asynsend3() that takes optional flags; asynsend() is now a #define passing in 0 as third parameter o Made PM/VFS protocol asynchronous; reintroduced tell_fs() o Made PM_BASE request/reply number range unique o Hacked in a horrible temporary workaround into RS to deal with newly revealed RS-PM-VFS race condition triangle until VFS is asynchronous System signal handling: o Fixed shutdown logic of device drivers; removed old SIGKSTOP signal o Removed is-superuser check from PM's do_procstat() (aka getsigset()) o Added sigset macros to allow system processes to deal with the full signal set, rather than just the POSIX subset Miscellaneous PM fixes: o Split do_getset into do_get and do_set, merging common code and making structure clearer o Fixed setpriority() being able to put to sleep processes using an invalid parameter, or revive zombie processes o Made find_proc() global; removed obsolete proc_from_pid() o Cleanup here and there Also included: o Fixed false-positive boot order kernel warning o Removed last traces of old NOTIFY_FROM code THINGS OF POSSIBLE INTEREST o It should now be possible to run PM at any priority, even lower than user processes o No assumptions are made about communication speed between PM and VFS, although communication must be FIFO o A debugger will now receive incoming debuggee signals at kill time only; the process may not yet be fully stopped o A first step has been made towards making the SYSTEM task preemptible
2009-09-30 11:57:22 +02:00
}
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
else if (p->p_misc_flags & MF_SC_DEFER) {
Merge of David's ptrace branch. Summary: o Support for ptrace T_ATTACH/T_DETACH and T_SYSCALL o PM signal handling logic should now work properly, even with debuggers being present o Asynchronous PM/VFS protocol, full IPC support for senda(), and AMF_NOREPLY senda() flag DETAILS Process stop and delay call handling of PM: o Added sys_runctl() kernel call with sys_stop() and sys_resume() aliases, for PM to stop and resume a process o Added exception for sending/syscall-traced processes to sys_runctl(), and matching SIGKREADY pseudo-signal to PM o Fixed PM signal logic to deal with requests from a process after stopping it (so-called "delay calls"), using the SIGKREADY facility o Fixed various PM panics due to race conditions with delay calls versus VFS calls o Removed special PRIO_STOP priority value o Added SYS_LOCK RTS kernel flag, to stop an individual process from running while modifying its process structure Signal and debugger handling in PM: o Fixed debugger signals being dropped if a second signal arrives when the debugger has not retrieved the first one o Fixed debugger signals being sent to the debugger more than once o Fixed debugger signals unpausing process in VFS; removed PM_UNPAUSE_TR protocol message o Detached debugger signals from general signal logic and from being blocked on VFS calls, meaning that even VFS can now be traced o Fixed debugger being unable to receive more than one pending signal in one process stop o Fixed signal delivery being delayed needlessly when multiple signals are pending o Fixed wait test for tracer, which was returning for children that were not waited for o Removed second parallel pending call from PM to VFS for any process o Fixed process becoming runnable between exec() and debugger trap o Added support for notifying the debugger before the parent when a debugged child exits o Fixed debugger death causing child to remain stopped forever o Fixed consistently incorrect use of _NSIG Extensions to ptrace(): o Added T_ATTACH and T_DETACH ptrace request, to attach and detach a debugger to and from a process o Added T_SYSCALL ptrace request, to trace system calls o Added T_SETOPT ptrace request, to set trace options o Added TO_TRACEFORK trace option, to attach automatically to children of a traced process o Added TO_ALTEXEC trace option, to send SIGSTOP instead of SIGTRAP upon a successful exec() of the tracee o Extended T_GETUSER ptrace support to allow retrieving a process's priv structure o Removed T_STOP ptrace request again, as it does not help implementing debuggers properly o Added MINIX3-specific ptrace test (test42) o Added proper manual page for ptrace(2) Asynchronous PM/VFS interface: o Fixed asynchronous messages not being checked when receive() is called with an endpoint other than ANY o Added AMF_NOREPLY senda() flag, preventing such messages from satisfying the receive part of a sendrec() o Added asynsend3() that takes optional flags; asynsend() is now a #define passing in 0 as third parameter o Made PM/VFS protocol asynchronous; reintroduced tell_fs() o Made PM_BASE request/reply number range unique o Hacked in a horrible temporary workaround into RS to deal with newly revealed RS-PM-VFS race condition triangle until VFS is asynchronous System signal handling: o Fixed shutdown logic of device drivers; removed old SIGKSTOP signal o Removed is-superuser check from PM's do_procstat() (aka getsigset()) o Added sigset macros to allow system processes to deal with the full signal set, rather than just the POSIX subset Miscellaneous PM fixes: o Split do_getset into do_get and do_set, merging common code and making structure clearer o Fixed setpriority() being able to put to sleep processes using an invalid parameter, or revive zombie processes o Made find_proc() global; removed obsolete proc_from_pid() o Cleanup here and there Also included: o Fixed false-positive boot order kernel warning o Removed last traces of old NOTIFY_FROM code THINGS OF POSSIBLE INTEREST o It should now be possible to run PM at any priority, even lower than user processes o No assumptions are made about communication speed between PM and VFS, although communication must be FIFO o A debugger will now receive incoming debuggee signals at kill time only; the process may not yet be fully stopped o A first step has been made towards making the SYSTEM task preemptible
2009-09-30 11:57:22 +02:00
/* Perform the system call that we deferred earlier. */
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
assert (!(p->p_misc_flags & MF_SC_ACTIVE));
Merge of David's ptrace branch. Summary: o Support for ptrace T_ATTACH/T_DETACH and T_SYSCALL o PM signal handling logic should now work properly, even with debuggers being present o Asynchronous PM/VFS protocol, full IPC support for senda(), and AMF_NOREPLY senda() flag DETAILS Process stop and delay call handling of PM: o Added sys_runctl() kernel call with sys_stop() and sys_resume() aliases, for PM to stop and resume a process o Added exception for sending/syscall-traced processes to sys_runctl(), and matching SIGKREADY pseudo-signal to PM o Fixed PM signal logic to deal with requests from a process after stopping it (so-called "delay calls"), using the SIGKREADY facility o Fixed various PM panics due to race conditions with delay calls versus VFS calls o Removed special PRIO_STOP priority value o Added SYS_LOCK RTS kernel flag, to stop an individual process from running while modifying its process structure Signal and debugger handling in PM: o Fixed debugger signals being dropped if a second signal arrives when the debugger has not retrieved the first one o Fixed debugger signals being sent to the debugger more than once o Fixed debugger signals unpausing process in VFS; removed PM_UNPAUSE_TR protocol message o Detached debugger signals from general signal logic and from being blocked on VFS calls, meaning that even VFS can now be traced o Fixed debugger being unable to receive more than one pending signal in one process stop o Fixed signal delivery being delayed needlessly when multiple signals are pending o Fixed wait test for tracer, which was returning for children that were not waited for o Removed second parallel pending call from PM to VFS for any process o Fixed process becoming runnable between exec() and debugger trap o Added support for notifying the debugger before the parent when a debugged child exits o Fixed debugger death causing child to remain stopped forever o Fixed consistently incorrect use of _NSIG Extensions to ptrace(): o Added T_ATTACH and T_DETACH ptrace request, to attach and detach a debugger to and from a process o Added T_SYSCALL ptrace request, to trace system calls o Added T_SETOPT ptrace request, to set trace options o Added TO_TRACEFORK trace option, to attach automatically to children of a traced process o Added TO_ALTEXEC trace option, to send SIGSTOP instead of SIGTRAP upon a successful exec() of the tracee o Extended T_GETUSER ptrace support to allow retrieving a process's priv structure o Removed T_STOP ptrace request again, as it does not help implementing debuggers properly o Added MINIX3-specific ptrace test (test42) o Added proper manual page for ptrace(2) Asynchronous PM/VFS interface: o Fixed asynchronous messages not being checked when receive() is called with an endpoint other than ANY o Added AMF_NOREPLY senda() flag, preventing such messages from satisfying the receive part of a sendrec() o Added asynsend3() that takes optional flags; asynsend() is now a #define passing in 0 as third parameter o Made PM/VFS protocol asynchronous; reintroduced tell_fs() o Made PM_BASE request/reply number range unique o Hacked in a horrible temporary workaround into RS to deal with newly revealed RS-PM-VFS race condition triangle until VFS is asynchronous System signal handling: o Fixed shutdown logic of device drivers; removed old SIGKSTOP signal o Removed is-superuser check from PM's do_procstat() (aka getsigset()) o Added sigset macros to allow system processes to deal with the full signal set, rather than just the POSIX subset Miscellaneous PM fixes: o Split do_getset into do_get and do_set, merging common code and making structure clearer o Fixed setpriority() being able to put to sleep processes using an invalid parameter, or revive zombie processes o Made find_proc() global; removed obsolete proc_from_pid() o Cleanup here and there Also included: o Fixed false-positive boot order kernel warning o Removed last traces of old NOTIFY_FROM code THINGS OF POSSIBLE INTEREST o It should now be possible to run PM at any priority, even lower than user processes o No assumptions are made about communication speed between PM and VFS, although communication must be FIFO o A debugger will now receive incoming debuggee signals at kill time only; the process may not yet be fully stopped o A first step has been made towards making the SYSTEM task preemptible
2009-09-30 11:57:22 +02:00
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
arch_do_syscall(p);
Merge of David's ptrace branch. Summary: o Support for ptrace T_ATTACH/T_DETACH and T_SYSCALL o PM signal handling logic should now work properly, even with debuggers being present o Asynchronous PM/VFS protocol, full IPC support for senda(), and AMF_NOREPLY senda() flag DETAILS Process stop and delay call handling of PM: o Added sys_runctl() kernel call with sys_stop() and sys_resume() aliases, for PM to stop and resume a process o Added exception for sending/syscall-traced processes to sys_runctl(), and matching SIGKREADY pseudo-signal to PM o Fixed PM signal logic to deal with requests from a process after stopping it (so-called "delay calls"), using the SIGKREADY facility o Fixed various PM panics due to race conditions with delay calls versus VFS calls o Removed special PRIO_STOP priority value o Added SYS_LOCK RTS kernel flag, to stop an individual process from running while modifying its process structure Signal and debugger handling in PM: o Fixed debugger signals being dropped if a second signal arrives when the debugger has not retrieved the first one o Fixed debugger signals being sent to the debugger more than once o Fixed debugger signals unpausing process in VFS; removed PM_UNPAUSE_TR protocol message o Detached debugger signals from general signal logic and from being blocked on VFS calls, meaning that even VFS can now be traced o Fixed debugger being unable to receive more than one pending signal in one process stop o Fixed signal delivery being delayed needlessly when multiple signals are pending o Fixed wait test for tracer, which was returning for children that were not waited for o Removed second parallel pending call from PM to VFS for any process o Fixed process becoming runnable between exec() and debugger trap o Added support for notifying the debugger before the parent when a debugged child exits o Fixed debugger death causing child to remain stopped forever o Fixed consistently incorrect use of _NSIG Extensions to ptrace(): o Added T_ATTACH and T_DETACH ptrace request, to attach and detach a debugger to and from a process o Added T_SYSCALL ptrace request, to trace system calls o Added T_SETOPT ptrace request, to set trace options o Added TO_TRACEFORK trace option, to attach automatically to children of a traced process o Added TO_ALTEXEC trace option, to send SIGSTOP instead of SIGTRAP upon a successful exec() of the tracee o Extended T_GETUSER ptrace support to allow retrieving a process's priv structure o Removed T_STOP ptrace request again, as it does not help implementing debuggers properly o Added MINIX3-specific ptrace test (test42) o Added proper manual page for ptrace(2) Asynchronous PM/VFS interface: o Fixed asynchronous messages not being checked when receive() is called with an endpoint other than ANY o Added AMF_NOREPLY senda() flag, preventing such messages from satisfying the receive part of a sendrec() o Added asynsend3() that takes optional flags; asynsend() is now a #define passing in 0 as third parameter o Made PM/VFS protocol asynchronous; reintroduced tell_fs() o Made PM_BASE request/reply number range unique o Hacked in a horrible temporary workaround into RS to deal with newly revealed RS-PM-VFS race condition triangle until VFS is asynchronous System signal handling: o Fixed shutdown logic of device drivers; removed old SIGKSTOP signal o Removed is-superuser check from PM's do_procstat() (aka getsigset()) o Added sigset macros to allow system processes to deal with the full signal set, rather than just the POSIX subset Miscellaneous PM fixes: o Split do_getset into do_get and do_set, merging common code and making structure clearer o Fixed setpriority() being able to put to sleep processes using an invalid parameter, or revive zombie processes o Made find_proc() global; removed obsolete proc_from_pid() o Cleanup here and there Also included: o Fixed false-positive boot order kernel warning o Removed last traces of old NOTIFY_FROM code THINGS OF POSSIBLE INTEREST o It should now be possible to run PM at any priority, even lower than user processes o No assumptions are made about communication speed between PM and VFS, although communication must be FIFO o A debugger will now receive incoming debuggee signals at kill time only; the process may not yet be fully stopped o A first step has been made towards making the SYSTEM task preemptible
2009-09-30 11:57:22 +02:00
/* If the process is stopped for signal delivery, and
* not blocked sending a message after the system call,
* inform PM.
*/
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
if ((p->p_misc_flags & MF_SIG_DELAY) &&
!RTS_ISSET(p, RTS_SENDING))
sig_delay_done(p);
Merge of David's ptrace branch. Summary: o Support for ptrace T_ATTACH/T_DETACH and T_SYSCALL o PM signal handling logic should now work properly, even with debuggers being present o Asynchronous PM/VFS protocol, full IPC support for senda(), and AMF_NOREPLY senda() flag DETAILS Process stop and delay call handling of PM: o Added sys_runctl() kernel call with sys_stop() and sys_resume() aliases, for PM to stop and resume a process o Added exception for sending/syscall-traced processes to sys_runctl(), and matching SIGKREADY pseudo-signal to PM o Fixed PM signal logic to deal with requests from a process after stopping it (so-called "delay calls"), using the SIGKREADY facility o Fixed various PM panics due to race conditions with delay calls versus VFS calls o Removed special PRIO_STOP priority value o Added SYS_LOCK RTS kernel flag, to stop an individual process from running while modifying its process structure Signal and debugger handling in PM: o Fixed debugger signals being dropped if a second signal arrives when the debugger has not retrieved the first one o Fixed debugger signals being sent to the debugger more than once o Fixed debugger signals unpausing process in VFS; removed PM_UNPAUSE_TR protocol message o Detached debugger signals from general signal logic and from being blocked on VFS calls, meaning that even VFS can now be traced o Fixed debugger being unable to receive more than one pending signal in one process stop o Fixed signal delivery being delayed needlessly when multiple signals are pending o Fixed wait test for tracer, which was returning for children that were not waited for o Removed second parallel pending call from PM to VFS for any process o Fixed process becoming runnable between exec() and debugger trap o Added support for notifying the debugger before the parent when a debugged child exits o Fixed debugger death causing child to remain stopped forever o Fixed consistently incorrect use of _NSIG Extensions to ptrace(): o Added T_ATTACH and T_DETACH ptrace request, to attach and detach a debugger to and from a process o Added T_SYSCALL ptrace request, to trace system calls o Added T_SETOPT ptrace request, to set trace options o Added TO_TRACEFORK trace option, to attach automatically to children of a traced process o Added TO_ALTEXEC trace option, to send SIGSTOP instead of SIGTRAP upon a successful exec() of the tracee o Extended T_GETUSER ptrace support to allow retrieving a process's priv structure o Removed T_STOP ptrace request again, as it does not help implementing debuggers properly o Added MINIX3-specific ptrace test (test42) o Added proper manual page for ptrace(2) Asynchronous PM/VFS interface: o Fixed asynchronous messages not being checked when receive() is called with an endpoint other than ANY o Added AMF_NOREPLY senda() flag, preventing such messages from satisfying the receive part of a sendrec() o Added asynsend3() that takes optional flags; asynsend() is now a #define passing in 0 as third parameter o Made PM/VFS protocol asynchronous; reintroduced tell_fs() o Made PM_BASE request/reply number range unique o Hacked in a horrible temporary workaround into RS to deal with newly revealed RS-PM-VFS race condition triangle until VFS is asynchronous System signal handling: o Fixed shutdown logic of device drivers; removed old SIGKSTOP signal o Removed is-superuser check from PM's do_procstat() (aka getsigset()) o Added sigset macros to allow system processes to deal with the full signal set, rather than just the POSIX subset Miscellaneous PM fixes: o Split do_getset into do_get and do_set, merging common code and making structure clearer o Fixed setpriority() being able to put to sleep processes using an invalid parameter, or revive zombie processes o Made find_proc() global; removed obsolete proc_from_pid() o Cleanup here and there Also included: o Fixed false-positive boot order kernel warning o Removed last traces of old NOTIFY_FROM code THINGS OF POSSIBLE INTEREST o It should now be possible to run PM at any priority, even lower than user processes o No assumptions are made about communication speed between PM and VFS, although communication must be FIFO o A debugger will now receive incoming debuggee signals at kill time only; the process may not yet be fully stopped o A first step has been made towards making the SYSTEM task preemptible
2009-09-30 11:57:22 +02:00
}
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
else if (p->p_misc_flags & MF_SC_TRACE) {
Merge of David's ptrace branch. Summary: o Support for ptrace T_ATTACH/T_DETACH and T_SYSCALL o PM signal handling logic should now work properly, even with debuggers being present o Asynchronous PM/VFS protocol, full IPC support for senda(), and AMF_NOREPLY senda() flag DETAILS Process stop and delay call handling of PM: o Added sys_runctl() kernel call with sys_stop() and sys_resume() aliases, for PM to stop and resume a process o Added exception for sending/syscall-traced processes to sys_runctl(), and matching SIGKREADY pseudo-signal to PM o Fixed PM signal logic to deal with requests from a process after stopping it (so-called "delay calls"), using the SIGKREADY facility o Fixed various PM panics due to race conditions with delay calls versus VFS calls o Removed special PRIO_STOP priority value o Added SYS_LOCK RTS kernel flag, to stop an individual process from running while modifying its process structure Signal and debugger handling in PM: o Fixed debugger signals being dropped if a second signal arrives when the debugger has not retrieved the first one o Fixed debugger signals being sent to the debugger more than once o Fixed debugger signals unpausing process in VFS; removed PM_UNPAUSE_TR protocol message o Detached debugger signals from general signal logic and from being blocked on VFS calls, meaning that even VFS can now be traced o Fixed debugger being unable to receive more than one pending signal in one process stop o Fixed signal delivery being delayed needlessly when multiple signals are pending o Fixed wait test for tracer, which was returning for children that were not waited for o Removed second parallel pending call from PM to VFS for any process o Fixed process becoming runnable between exec() and debugger trap o Added support for notifying the debugger before the parent when a debugged child exits o Fixed debugger death causing child to remain stopped forever o Fixed consistently incorrect use of _NSIG Extensions to ptrace(): o Added T_ATTACH and T_DETACH ptrace request, to attach and detach a debugger to and from a process o Added T_SYSCALL ptrace request, to trace system calls o Added T_SETOPT ptrace request, to set trace options o Added TO_TRACEFORK trace option, to attach automatically to children of a traced process o Added TO_ALTEXEC trace option, to send SIGSTOP instead of SIGTRAP upon a successful exec() of the tracee o Extended T_GETUSER ptrace support to allow retrieving a process's priv structure o Removed T_STOP ptrace request again, as it does not help implementing debuggers properly o Added MINIX3-specific ptrace test (test42) o Added proper manual page for ptrace(2) Asynchronous PM/VFS interface: o Fixed asynchronous messages not being checked when receive() is called with an endpoint other than ANY o Added AMF_NOREPLY senda() flag, preventing such messages from satisfying the receive part of a sendrec() o Added asynsend3() that takes optional flags; asynsend() is now a #define passing in 0 as third parameter o Made PM/VFS protocol asynchronous; reintroduced tell_fs() o Made PM_BASE request/reply number range unique o Hacked in a horrible temporary workaround into RS to deal with newly revealed RS-PM-VFS race condition triangle until VFS is asynchronous System signal handling: o Fixed shutdown logic of device drivers; removed old SIGKSTOP signal o Removed is-superuser check from PM's do_procstat() (aka getsigset()) o Added sigset macros to allow system processes to deal with the full signal set, rather than just the POSIX subset Miscellaneous PM fixes: o Split do_getset into do_get and do_set, merging common code and making structure clearer o Fixed setpriority() being able to put to sleep processes using an invalid parameter, or revive zombie processes o Made find_proc() global; removed obsolete proc_from_pid() o Cleanup here and there Also included: o Fixed false-positive boot order kernel warning o Removed last traces of old NOTIFY_FROM code THINGS OF POSSIBLE INTEREST o It should now be possible to run PM at any priority, even lower than user processes o No assumptions are made about communication speed between PM and VFS, although communication must be FIFO o A debugger will now receive incoming debuggee signals at kill time only; the process may not yet be fully stopped o A first step has been made towards making the SYSTEM task preemptible
2009-09-30 11:57:22 +02:00
/* Trigger a system call leave event if this was a
* system call. We must do this after processing the
* other flags above, both for tracing correctness and
* to be able to use 'break'.
*/
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
if (!(p->p_misc_flags & MF_SC_ACTIVE))
Merge of David's ptrace branch. Summary: o Support for ptrace T_ATTACH/T_DETACH and T_SYSCALL o PM signal handling logic should now work properly, even with debuggers being present o Asynchronous PM/VFS protocol, full IPC support for senda(), and AMF_NOREPLY senda() flag DETAILS Process stop and delay call handling of PM: o Added sys_runctl() kernel call with sys_stop() and sys_resume() aliases, for PM to stop and resume a process o Added exception for sending/syscall-traced processes to sys_runctl(), and matching SIGKREADY pseudo-signal to PM o Fixed PM signal logic to deal with requests from a process after stopping it (so-called "delay calls"), using the SIGKREADY facility o Fixed various PM panics due to race conditions with delay calls versus VFS calls o Removed special PRIO_STOP priority value o Added SYS_LOCK RTS kernel flag, to stop an individual process from running while modifying its process structure Signal and debugger handling in PM: o Fixed debugger signals being dropped if a second signal arrives when the debugger has not retrieved the first one o Fixed debugger signals being sent to the debugger more than once o Fixed debugger signals unpausing process in VFS; removed PM_UNPAUSE_TR protocol message o Detached debugger signals from general signal logic and from being blocked on VFS calls, meaning that even VFS can now be traced o Fixed debugger being unable to receive more than one pending signal in one process stop o Fixed signal delivery being delayed needlessly when multiple signals are pending o Fixed wait test for tracer, which was returning for children that were not waited for o Removed second parallel pending call from PM to VFS for any process o Fixed process becoming runnable between exec() and debugger trap o Added support for notifying the debugger before the parent when a debugged child exits o Fixed debugger death causing child to remain stopped forever o Fixed consistently incorrect use of _NSIG Extensions to ptrace(): o Added T_ATTACH and T_DETACH ptrace request, to attach and detach a debugger to and from a process o Added T_SYSCALL ptrace request, to trace system calls o Added T_SETOPT ptrace request, to set trace options o Added TO_TRACEFORK trace option, to attach automatically to children of a traced process o Added TO_ALTEXEC trace option, to send SIGSTOP instead of SIGTRAP upon a successful exec() of the tracee o Extended T_GETUSER ptrace support to allow retrieving a process's priv structure o Removed T_STOP ptrace request again, as it does not help implementing debuggers properly o Added MINIX3-specific ptrace test (test42) o Added proper manual page for ptrace(2) Asynchronous PM/VFS interface: o Fixed asynchronous messages not being checked when receive() is called with an endpoint other than ANY o Added AMF_NOREPLY senda() flag, preventing such messages from satisfying the receive part of a sendrec() o Added asynsend3() that takes optional flags; asynsend() is now a #define passing in 0 as third parameter o Made PM/VFS protocol asynchronous; reintroduced tell_fs() o Made PM_BASE request/reply number range unique o Hacked in a horrible temporary workaround into RS to deal with newly revealed RS-PM-VFS race condition triangle until VFS is asynchronous System signal handling: o Fixed shutdown logic of device drivers; removed old SIGKSTOP signal o Removed is-superuser check from PM's do_procstat() (aka getsigset()) o Added sigset macros to allow system processes to deal with the full signal set, rather than just the POSIX subset Miscellaneous PM fixes: o Split do_getset into do_get and do_set, merging common code and making structure clearer o Fixed setpriority() being able to put to sleep processes using an invalid parameter, or revive zombie processes o Made find_proc() global; removed obsolete proc_from_pid() o Cleanup here and there Also included: o Fixed false-positive boot order kernel warning o Removed last traces of old NOTIFY_FROM code THINGS OF POSSIBLE INTEREST o It should now be possible to run PM at any priority, even lower than user processes o No assumptions are made about communication speed between PM and VFS, although communication must be FIFO o A debugger will now receive incoming debuggee signals at kill time only; the process may not yet be fully stopped o A first step has been made towards making the SYSTEM task preemptible
2009-09-30 11:57:22 +02:00
break;
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
p->p_misc_flags &=
Merge of David's ptrace branch. Summary: o Support for ptrace T_ATTACH/T_DETACH and T_SYSCALL o PM signal handling logic should now work properly, even with debuggers being present o Asynchronous PM/VFS protocol, full IPC support for senda(), and AMF_NOREPLY senda() flag DETAILS Process stop and delay call handling of PM: o Added sys_runctl() kernel call with sys_stop() and sys_resume() aliases, for PM to stop and resume a process o Added exception for sending/syscall-traced processes to sys_runctl(), and matching SIGKREADY pseudo-signal to PM o Fixed PM signal logic to deal with requests from a process after stopping it (so-called "delay calls"), using the SIGKREADY facility o Fixed various PM panics due to race conditions with delay calls versus VFS calls o Removed special PRIO_STOP priority value o Added SYS_LOCK RTS kernel flag, to stop an individual process from running while modifying its process structure Signal and debugger handling in PM: o Fixed debugger signals being dropped if a second signal arrives when the debugger has not retrieved the first one o Fixed debugger signals being sent to the debugger more than once o Fixed debugger signals unpausing process in VFS; removed PM_UNPAUSE_TR protocol message o Detached debugger signals from general signal logic and from being blocked on VFS calls, meaning that even VFS can now be traced o Fixed debugger being unable to receive more than one pending signal in one process stop o Fixed signal delivery being delayed needlessly when multiple signals are pending o Fixed wait test for tracer, which was returning for children that were not waited for o Removed second parallel pending call from PM to VFS for any process o Fixed process becoming runnable between exec() and debugger trap o Added support for notifying the debugger before the parent when a debugged child exits o Fixed debugger death causing child to remain stopped forever o Fixed consistently incorrect use of _NSIG Extensions to ptrace(): o Added T_ATTACH and T_DETACH ptrace request, to attach and detach a debugger to and from a process o Added T_SYSCALL ptrace request, to trace system calls o Added T_SETOPT ptrace request, to set trace options o Added TO_TRACEFORK trace option, to attach automatically to children of a traced process o Added TO_ALTEXEC trace option, to send SIGSTOP instead of SIGTRAP upon a successful exec() of the tracee o Extended T_GETUSER ptrace support to allow retrieving a process's priv structure o Removed T_STOP ptrace request again, as it does not help implementing debuggers properly o Added MINIX3-specific ptrace test (test42) o Added proper manual page for ptrace(2) Asynchronous PM/VFS interface: o Fixed asynchronous messages not being checked when receive() is called with an endpoint other than ANY o Added AMF_NOREPLY senda() flag, preventing such messages from satisfying the receive part of a sendrec() o Added asynsend3() that takes optional flags; asynsend() is now a #define passing in 0 as third parameter o Made PM/VFS protocol asynchronous; reintroduced tell_fs() o Made PM_BASE request/reply number range unique o Hacked in a horrible temporary workaround into RS to deal with newly revealed RS-PM-VFS race condition triangle until VFS is asynchronous System signal handling: o Fixed shutdown logic of device drivers; removed old SIGKSTOP signal o Removed is-superuser check from PM's do_procstat() (aka getsigset()) o Added sigset macros to allow system processes to deal with the full signal set, rather than just the POSIX subset Miscellaneous PM fixes: o Split do_getset into do_get and do_set, merging common code and making structure clearer o Fixed setpriority() being able to put to sleep processes using an invalid parameter, or revive zombie processes o Made find_proc() global; removed obsolete proc_from_pid() o Cleanup here and there Also included: o Fixed false-positive boot order kernel warning o Removed last traces of old NOTIFY_FROM code THINGS OF POSSIBLE INTEREST o It should now be possible to run PM at any priority, even lower than user processes o No assumptions are made about communication speed between PM and VFS, although communication must be FIFO o A debugger will now receive incoming debuggee signals at kill time only; the process may not yet be fully stopped o A first step has been made towards making the SYSTEM task preemptible
2009-09-30 11:57:22 +02:00
~(MF_SC_TRACE | MF_SC_ACTIVE);
/* Signal the "leave system call" event.
* Block the process.
*/
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
cause_sig(proc_nr(p), SIGTRAP);
Merge of David's ptrace branch. Summary: o Support for ptrace T_ATTACH/T_DETACH and T_SYSCALL o PM signal handling logic should now work properly, even with debuggers being present o Asynchronous PM/VFS protocol, full IPC support for senda(), and AMF_NOREPLY senda() flag DETAILS Process stop and delay call handling of PM: o Added sys_runctl() kernel call with sys_stop() and sys_resume() aliases, for PM to stop and resume a process o Added exception for sending/syscall-traced processes to sys_runctl(), and matching SIGKREADY pseudo-signal to PM o Fixed PM signal logic to deal with requests from a process after stopping it (so-called "delay calls"), using the SIGKREADY facility o Fixed various PM panics due to race conditions with delay calls versus VFS calls o Removed special PRIO_STOP priority value o Added SYS_LOCK RTS kernel flag, to stop an individual process from running while modifying its process structure Signal and debugger handling in PM: o Fixed debugger signals being dropped if a second signal arrives when the debugger has not retrieved the first one o Fixed debugger signals being sent to the debugger more than once o Fixed debugger signals unpausing process in VFS; removed PM_UNPAUSE_TR protocol message o Detached debugger signals from general signal logic and from being blocked on VFS calls, meaning that even VFS can now be traced o Fixed debugger being unable to receive more than one pending signal in one process stop o Fixed signal delivery being delayed needlessly when multiple signals are pending o Fixed wait test for tracer, which was returning for children that were not waited for o Removed second parallel pending call from PM to VFS for any process o Fixed process becoming runnable between exec() and debugger trap o Added support for notifying the debugger before the parent when a debugged child exits o Fixed debugger death causing child to remain stopped forever o Fixed consistently incorrect use of _NSIG Extensions to ptrace(): o Added T_ATTACH and T_DETACH ptrace request, to attach and detach a debugger to and from a process o Added T_SYSCALL ptrace request, to trace system calls o Added T_SETOPT ptrace request, to set trace options o Added TO_TRACEFORK trace option, to attach automatically to children of a traced process o Added TO_ALTEXEC trace option, to send SIGSTOP instead of SIGTRAP upon a successful exec() of the tracee o Extended T_GETUSER ptrace support to allow retrieving a process's priv structure o Removed T_STOP ptrace request again, as it does not help implementing debuggers properly o Added MINIX3-specific ptrace test (test42) o Added proper manual page for ptrace(2) Asynchronous PM/VFS interface: o Fixed asynchronous messages not being checked when receive() is called with an endpoint other than ANY o Added AMF_NOREPLY senda() flag, preventing such messages from satisfying the receive part of a sendrec() o Added asynsend3() that takes optional flags; asynsend() is now a #define passing in 0 as third parameter o Made PM/VFS protocol asynchronous; reintroduced tell_fs() o Made PM_BASE request/reply number range unique o Hacked in a horrible temporary workaround into RS to deal with newly revealed RS-PM-VFS race condition triangle until VFS is asynchronous System signal handling: o Fixed shutdown logic of device drivers; removed old SIGKSTOP signal o Removed is-superuser check from PM's do_procstat() (aka getsigset()) o Added sigset macros to allow system processes to deal with the full signal set, rather than just the POSIX subset Miscellaneous PM fixes: o Split do_getset into do_get and do_set, merging common code and making structure clearer o Fixed setpriority() being able to put to sleep processes using an invalid parameter, or revive zombie processes o Made find_proc() global; removed obsolete proc_from_pid() o Cleanup here and there Also included: o Fixed false-positive boot order kernel warning o Removed last traces of old NOTIFY_FROM code THINGS OF POSSIBLE INTEREST o It should now be possible to run PM at any priority, even lower than user processes o No assumptions are made about communication speed between PM and VFS, although communication must be FIFO o A debugger will now receive incoming debuggee signals at kill time only; the process may not yet be fully stopped o A first step has been made towards making the SYSTEM task preemptible
2009-09-30 11:57:22 +02:00
}
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
else if (p->p_misc_flags & MF_SC_ACTIVE) {
Merge of David's ptrace branch. Summary: o Support for ptrace T_ATTACH/T_DETACH and T_SYSCALL o PM signal handling logic should now work properly, even with debuggers being present o Asynchronous PM/VFS protocol, full IPC support for senda(), and AMF_NOREPLY senda() flag DETAILS Process stop and delay call handling of PM: o Added sys_runctl() kernel call with sys_stop() and sys_resume() aliases, for PM to stop and resume a process o Added exception for sending/syscall-traced processes to sys_runctl(), and matching SIGKREADY pseudo-signal to PM o Fixed PM signal logic to deal with requests from a process after stopping it (so-called "delay calls"), using the SIGKREADY facility o Fixed various PM panics due to race conditions with delay calls versus VFS calls o Removed special PRIO_STOP priority value o Added SYS_LOCK RTS kernel flag, to stop an individual process from running while modifying its process structure Signal and debugger handling in PM: o Fixed debugger signals being dropped if a second signal arrives when the debugger has not retrieved the first one o Fixed debugger signals being sent to the debugger more than once o Fixed debugger signals unpausing process in VFS; removed PM_UNPAUSE_TR protocol message o Detached debugger signals from general signal logic and from being blocked on VFS calls, meaning that even VFS can now be traced o Fixed debugger being unable to receive more than one pending signal in one process stop o Fixed signal delivery being delayed needlessly when multiple signals are pending o Fixed wait test for tracer, which was returning for children that were not waited for o Removed second parallel pending call from PM to VFS for any process o Fixed process becoming runnable between exec() and debugger trap o Added support for notifying the debugger before the parent when a debugged child exits o Fixed debugger death causing child to remain stopped forever o Fixed consistently incorrect use of _NSIG Extensions to ptrace(): o Added T_ATTACH and T_DETACH ptrace request, to attach and detach a debugger to and from a process o Added T_SYSCALL ptrace request, to trace system calls o Added T_SETOPT ptrace request, to set trace options o Added TO_TRACEFORK trace option, to attach automatically to children of a traced process o Added TO_ALTEXEC trace option, to send SIGSTOP instead of SIGTRAP upon a successful exec() of the tracee o Extended T_GETUSER ptrace support to allow retrieving a process's priv structure o Removed T_STOP ptrace request again, as it does not help implementing debuggers properly o Added MINIX3-specific ptrace test (test42) o Added proper manual page for ptrace(2) Asynchronous PM/VFS interface: o Fixed asynchronous messages not being checked when receive() is called with an endpoint other than ANY o Added AMF_NOREPLY senda() flag, preventing such messages from satisfying the receive part of a sendrec() o Added asynsend3() that takes optional flags; asynsend() is now a #define passing in 0 as third parameter o Made PM/VFS protocol asynchronous; reintroduced tell_fs() o Made PM_BASE request/reply number range unique o Hacked in a horrible temporary workaround into RS to deal with newly revealed RS-PM-VFS race condition triangle until VFS is asynchronous System signal handling: o Fixed shutdown logic of device drivers; removed old SIGKSTOP signal o Removed is-superuser check from PM's do_procstat() (aka getsigset()) o Added sigset macros to allow system processes to deal with the full signal set, rather than just the POSIX subset Miscellaneous PM fixes: o Split do_getset into do_get and do_set, merging common code and making structure clearer o Fixed setpriority() being able to put to sleep processes using an invalid parameter, or revive zombie processes o Made find_proc() global; removed obsolete proc_from_pid() o Cleanup here and there Also included: o Fixed false-positive boot order kernel warning o Removed last traces of old NOTIFY_FROM code THINGS OF POSSIBLE INTEREST o It should now be possible to run PM at any priority, even lower than user processes o No assumptions are made about communication speed between PM and VFS, although communication must be FIFO o A debugger will now receive incoming debuggee signals at kill time only; the process may not yet be fully stopped o A first step has been made towards making the SYSTEM task preemptible
2009-09-30 11:57:22 +02:00
/* If MF_SC_ACTIVE was set, remove it now:
* we're leaving the system call.
*/
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
p->p_misc_flags &= ~MF_SC_ACTIVE;
Merge of David's ptrace branch. Summary: o Support for ptrace T_ATTACH/T_DETACH and T_SYSCALL o PM signal handling logic should now work properly, even with debuggers being present o Asynchronous PM/VFS protocol, full IPC support for senda(), and AMF_NOREPLY senda() flag DETAILS Process stop and delay call handling of PM: o Added sys_runctl() kernel call with sys_stop() and sys_resume() aliases, for PM to stop and resume a process o Added exception for sending/syscall-traced processes to sys_runctl(), and matching SIGKREADY pseudo-signal to PM o Fixed PM signal logic to deal with requests from a process after stopping it (so-called "delay calls"), using the SIGKREADY facility o Fixed various PM panics due to race conditions with delay calls versus VFS calls o Removed special PRIO_STOP priority value o Added SYS_LOCK RTS kernel flag, to stop an individual process from running while modifying its process structure Signal and debugger handling in PM: o Fixed debugger signals being dropped if a second signal arrives when the debugger has not retrieved the first one o Fixed debugger signals being sent to the debugger more than once o Fixed debugger signals unpausing process in VFS; removed PM_UNPAUSE_TR protocol message o Detached debugger signals from general signal logic and from being blocked on VFS calls, meaning that even VFS can now be traced o Fixed debugger being unable to receive more than one pending signal in one process stop o Fixed signal delivery being delayed needlessly when multiple signals are pending o Fixed wait test for tracer, which was returning for children that were not waited for o Removed second parallel pending call from PM to VFS for any process o Fixed process becoming runnable between exec() and debugger trap o Added support for notifying the debugger before the parent when a debugged child exits o Fixed debugger death causing child to remain stopped forever o Fixed consistently incorrect use of _NSIG Extensions to ptrace(): o Added T_ATTACH and T_DETACH ptrace request, to attach and detach a debugger to and from a process o Added T_SYSCALL ptrace request, to trace system calls o Added T_SETOPT ptrace request, to set trace options o Added TO_TRACEFORK trace option, to attach automatically to children of a traced process o Added TO_ALTEXEC trace option, to send SIGSTOP instead of SIGTRAP upon a successful exec() of the tracee o Extended T_GETUSER ptrace support to allow retrieving a process's priv structure o Removed T_STOP ptrace request again, as it does not help implementing debuggers properly o Added MINIX3-specific ptrace test (test42) o Added proper manual page for ptrace(2) Asynchronous PM/VFS interface: o Fixed asynchronous messages not being checked when receive() is called with an endpoint other than ANY o Added AMF_NOREPLY senda() flag, preventing such messages from satisfying the receive part of a sendrec() o Added asynsend3() that takes optional flags; asynsend() is now a #define passing in 0 as third parameter o Made PM/VFS protocol asynchronous; reintroduced tell_fs() o Made PM_BASE request/reply number range unique o Hacked in a horrible temporary workaround into RS to deal with newly revealed RS-PM-VFS race condition triangle until VFS is asynchronous System signal handling: o Fixed shutdown logic of device drivers; removed old SIGKSTOP signal o Removed is-superuser check from PM's do_procstat() (aka getsigset()) o Added sigset macros to allow system processes to deal with the full signal set, rather than just the POSIX subset Miscellaneous PM fixes: o Split do_getset into do_get and do_set, merging common code and making structure clearer o Fixed setpriority() being able to put to sleep processes using an invalid parameter, or revive zombie processes o Made find_proc() global; removed obsolete proc_from_pid() o Cleanup here and there Also included: o Fixed false-positive boot order kernel warning o Removed last traces of old NOTIFY_FROM code THINGS OF POSSIBLE INTEREST o It should now be possible to run PM at any priority, even lower than user processes o No assumptions are made about communication speed between PM and VFS, although communication must be FIFO o A debugger will now receive incoming debuggee signals at kill time only; the process may not yet be fully stopped o A first step has been made towards making the SYSTEM task preemptible
2009-09-30 11:57:22 +02:00
break;
}
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
/*
* the selected process might not be runnable anymore. We have
* to checkit and schedule another one
*/
if (!proc_is_runnable(p))
goto not_runnable_pick_new;
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
}
scheduling - time quantum in miliseconds - Currently the cpu time quantum is timer-ticks based. Thus the remaining quantum is decreased only if the processes is interrupted by a timer tick. As processes block a lot this typically does not happen for normal user processes. Also the quantum depends on the frequency of the timer. - This change makes the quantum miliseconds based. Internally the miliseconds are translated into cpu cycles. Everytime userspace execution is interrupted by kernel the cycles just consumed by the current process are deducted from the remaining quantum. - It makes the quantum system timer frequency independent. - The boot processes quantum is loosely derived from the tick-based quantas and 60Hz timer and subject to future change - the 64bit arithmetics is a little ugly, will be changes once we have compiler support for 64bit integers (soon)
2010-05-25 10:06:14 +02:00
/*
* check the quantum left before it runs again. We must do it only here
* as we are sure that a possible out-of-quantum message to the
* scheduler will not collide with the regular ipc
*/
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
if (is_zero64(p->p_cpu_time_left))
proc_no_time(p);
Userspace scheduling - cotributed by Bjorn Swift - In this first phase, scheduling is moved from the kernel to the PM server. The next steps are to a) moving scheduling to its own server and b) include useful information in the "out of quantum" message, so that the scheduler can make use of this information. - The kernel process table now keeps record of who is responsible for scheduling each process (p_scheduler). When this pointer is NULL, the process will be scheduled by the kernel. If such a process runs out of quantum, the kernel will simply renew its quantum an requeue it. - When PM loads, it will take over scheduling of all running processes, except system processes, using sys_schedctl(). Essentially, this only results in taking over init. As children inherit a scheduler from their parent, user space programs forked by init will inherit PM (for now) as their scheduler. - Once a process has been assigned a scheduler, and runs out of quantum, its RTS_NO_QUANTUM flag will be set and the process dequeued. The kernel will send a message to the scheduler, on the process' behalf, informing the scheduler that it has run out of quantum. The scheduler can take what ever action it pleases, based on its policy, and then reschedule the process using the sys_schedule() system call. - Balance queues does not work as before. While the old in-kernel function used to renew the quantum of processes in the highest priority run queue, the user-space implementation only acts on processes that have been bumped down to a lower priority queue. This approach reacts slower to changes than the old one, but saves us sending a sys_schedule message for each process every time we balance the queues. Currently, when processes are moved up a priority queue, their quantum is also renewed, but this can be fiddled with. - do_nice has been removed from kernel. PM answers to get- and setpriority calls, updates it's own nice variable as well as the max_run_queue. This will be refactored once scheduling is moved to a separate server. We will probably have PM update it's local nice value and then send a message to whoever is scheduling the process. - changes to fix an issue in do_fork() where processes could run out of quantum but bypassing the code path that handles it correctly. The future plan is to remove the policy from do_fork() and implement it in userspace too.
2010-03-29 13:07:20 +02:00
/*
* After handling the misc flags the selected process might not be
* runnable anymore. We have to checkit and schedule another one
*/
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
if (!proc_is_runnable(p))
Userspace scheduling - cotributed by Bjorn Swift - In this first phase, scheduling is moved from the kernel to the PM server. The next steps are to a) moving scheduling to its own server and b) include useful information in the "out of quantum" message, so that the scheduler can make use of this information. - The kernel process table now keeps record of who is responsible for scheduling each process (p_scheduler). When this pointer is NULL, the process will be scheduled by the kernel. If such a process runs out of quantum, the kernel will simply renew its quantum an requeue it. - When PM loads, it will take over scheduling of all running processes, except system processes, using sys_schedctl(). Essentially, this only results in taking over init. As children inherit a scheduler from their parent, user space programs forked by init will inherit PM (for now) as their scheduler. - Once a process has been assigned a scheduler, and runs out of quantum, its RTS_NO_QUANTUM flag will be set and the process dequeued. The kernel will send a message to the scheduler, on the process' behalf, informing the scheduler that it has run out of quantum. The scheduler can take what ever action it pleases, based on its policy, and then reschedule the process using the sys_schedule() system call. - Balance queues does not work as before. While the old in-kernel function used to renew the quantum of processes in the highest priority run queue, the user-space implementation only acts on processes that have been bumped down to a lower priority queue. This approach reacts slower to changes than the old one, but saves us sending a sys_schedule message for each process every time we balance the queues. Currently, when processes are moved up a priority queue, their quantum is also renewed, but this can be fiddled with. - do_nice has been removed from kernel. PM answers to get- and setpriority calls, updates it's own nice variable as well as the max_run_queue. This will be refactored once scheduling is moved to a separate server. We will probably have PM update it's local nice value and then send a message to whoever is scheduling the process. - changes to fix an issue in do_fork() where processes could run out of quantum but bypassing the code path that handles it correctly. The future plan is to remove the policy from do_fork() and implement it in userspace too.
2010-03-29 13:07:20 +02:00
goto not_runnable_pick_new;
SMP - Print cpu of the process - adds '4' to print processes assigned to each cpu without printing the process it is blocked on (a lightweight '1')
2010-09-15 16:11:01 +02:00
TRACE(VF_SCHEDULING, printf("cpu %d starting %s / %d "
"pc 0x%08x\n",
cpuid, p->p_name, p->p_endpoint, p->p_reg.pc););
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
#if DEBUG_TRACE
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
p->p_schedules++;
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
#endif
Complete ovehaul of mode switching code - after a trap to kernel, the code automatically switches to kernel stack, in the future local to the CPU - k_reenter variable replaced by a test whether the CS is kernel cs or not. The information is passed further if needed. Removes a global variable which would need to be cpu local - no need for global variables describing the exception or trap context. This information is kept on stack and a pointer to this structure is passed to the C code as a single structure - removed loadedcr3 variable and its use replaced by reading the %cr3 register - no need to redisable interrupts in restart() as they are already disabled. - unified handling of traps that push and don't push errorcode - removed save() function as the process context is not saved directly to process table but saved as required by the trap code. Essentially it means that save() code is inlined everywhere not only in the exception handling routine - returning from syscall is more arch independent - it sets the retger in C - top of the x86 stack contains the current CPU id and pointer to the currently scheduled process (the one right interrupted) so the mode switch code can find where to save the context without need to use proc_ptr which will be cpu local in the future and therefore difficult to access in assembler and expensive to access in general - some more clean up of level0 code. No need to read-back the argument passed in %eax from the proc structure. The mode switch code does not clobber %the general registers and hence we can just call what is in %eax - many assebly macros in sconst.h as they will be reused by the apic assembly
2009-11-06 10:08:26 +01:00
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
p = arch_finish_switch_to_user();
assert(!is_zero64(p->p_cpu_time_left));
Userspace scheduling - cotributed by Bjorn Swift - In this first phase, scheduling is moved from the kernel to the PM server. The next steps are to a) moving scheduling to its own server and b) include useful information in the "out of quantum" message, so that the scheduler can make use of this information. - The kernel process table now keeps record of who is responsible for scheduling each process (p_scheduler). When this pointer is NULL, the process will be scheduled by the kernel. If such a process runs out of quantum, the kernel will simply renew its quantum an requeue it. - When PM loads, it will take over scheduling of all running processes, except system processes, using sys_schedctl(). Essentially, this only results in taking over init. As children inherit a scheduler from their parent, user space programs forked by init will inherit PM (for now) as their scheduler. - Once a process has been assigned a scheduler, and runs out of quantum, its RTS_NO_QUANTUM flag will be set and the process dequeued. The kernel will send a message to the scheduler, on the process' behalf, informing the scheduler that it has run out of quantum. The scheduler can take what ever action it pleases, based on its policy, and then reschedule the process using the sys_schedule() system call. - Balance queues does not work as before. While the old in-kernel function used to renew the quantum of processes in the highest priority run queue, the user-space implementation only acts on processes that have been bumped down to a lower priority queue. This approach reacts slower to changes than the old one, but saves us sending a sys_schedule message for each process every time we balance the queues. Currently, when processes are moved up a priority queue, their quantum is also renewed, but this can be fiddled with. - do_nice has been removed from kernel. PM answers to get- and setpriority calls, updates it's own nice variable as well as the max_run_queue. This will be refactored once scheduling is moved to a separate server. We will probably have PM update it's local nice value and then send a message to whoever is scheduling the process. - changes to fix an issue in do_fork() where processes could run out of quantum but bypassing the code path that handles it correctly. The future plan is to remove the policy from do_fork() and implement it in userspace too.
2010-03-29 13:07:20 +02:00
rename of mode/context switching functions - this patch only renames schedcheck() to switch_to_user(), cycles_accounting_stop() to context_stop() and restart() to +restore_user_context() - the motivation is that since the introduction of schedcheck() it has been abused for many things. It deserves a better name. It should express the fact that from the moment we call the function we are in the process of switching to user. - cycles_accounting_stop() was originally a single purpose function. As this function is called at were convenient places it is used in for other things too, e.g. (un)locking the kernel. Thus it deserves a better name too. - using the old name, restart() does not call schedcheck(), however calls to restart are replaced by calls to schedcheck() [switch_to_user] and it calls restart() [restore_user_context]
2010-05-18 15:00:39 +02:00
context_stop(proc_addr(KERNEL));
Complete ovehaul of mode switching code - after a trap to kernel, the code automatically switches to kernel stack, in the future local to the CPU - k_reenter variable replaced by a test whether the CS is kernel cs or not. The information is passed further if needed. Removes a global variable which would need to be cpu local - no need for global variables describing the exception or trap context. This information is kept on stack and a pointer to this structure is passed to the C code as a single structure - removed loadedcr3 variable and its use replaced by reading the %cr3 register - no need to redisable interrupts in restart() as they are already disabled. - unified handling of traps that push and don't push errorcode - removed save() function as the process context is not saved directly to process table but saved as required by the trap code. Essentially it means that save() code is inlined everywhere not only in the exception handling routine - returning from syscall is more arch independent - it sets the retger in C - top of the x86 stack contains the current CPU id and pointer to the currently scheduled process (the one right interrupted) so the mode switch code can find where to save the context without need to use proc_ptr which will be cpu local in the future and therefore difficult to access in assembler and expensive to access in general - some more clean up of level0 code. No need to read-back the argument passed in %eax from the proc structure. The mode switch code does not clobber %the general registers and hence we can just call what is in %eax - many assebly macros in sconst.h as they will be reused by the apic assembly
2009-11-06 10:08:26 +01:00
Lazy FPU - FPU context is stored only if conflict between 2 FPU users or while exporting context of a process to userspace while it is the active user of FPU - FPU has its owner (fpu_owner) which points to the process whose state is currently loaded in FPU - the FPU exception is only turned on when scheduling a process which is not the owner of FPU - FPU state is restored for the process that generated the FPU exception. This process runs immediately without letting scheduler to pick a new process to resolve the FPU conflict asap, to minimize the FPU thrashing and FPU exception hadler execution - faster all non-FPU-exception kernel entries as FPU state is not checked nor saved - removed MF_USED_FPU flag, only MF_FPU_INITIALIZED remains to signal that a process has used FPU in the past
2010-06-07 09:43:17 +02:00
/* If the process isn't the owner of FPU, enable the FPU exception */
SMP - lazy FPU - when a process is migrated to a different CPU it may have an active FPU context in the processor registers. We must save it and migrate it together with the process.
2010-09-15 16:11:25 +02:00
if(get_cpulocal_var(fpu_owner) != p)
Lazy FPU - FPU context is stored only if conflict between 2 FPU users or while exporting context of a process to userspace while it is the active user of FPU - FPU has its owner (fpu_owner) which points to the process whose state is currently loaded in FPU - the FPU exception is only turned on when scheduling a process which is not the owner of FPU - FPU state is restored for the process that generated the FPU exception. This process runs immediately without letting scheduler to pick a new process to resolve the FPU conflict asap, to minimize the FPU thrashing and FPU exception hadler execution - faster all non-FPU-exception kernel entries as FPU state is not checked nor saved - removed MF_USED_FPU flag, only MF_FPU_INITIALIZED remains to signal that a process has used FPU in the past
2010-06-07 09:43:17 +02:00
enable_fpu_exception();
else
disable_fpu_exception();
kernel: clear MF_CONTEXT_SET on kernel exit.
2010-07-20 19:13:44 +02:00
/* If MF_CONTEXT_SET is set, don't clobber process state within
* the kernel. The next kernel entry is OK again though.
*/
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
p->p_misc_flags &= ~MF_CONTEXT_SET;
kernel: clear MF_CONTEXT_SET on kernel exit.
2010-07-20 19:13:44 +02:00
SMP - Force TLB flush before scheduling a process - this makes sure that each process always run with updated TLB - this is the simplest way how to achieve the consistency. As it means significant performace degradation when not require, this is nto the final solution and will be refined
2010-09-15 16:11:17 +02:00
assert(!(p->p_misc_flags & MF_FULLVM) || p->p_seg.p_cr3 != 0);
Remove unnecessary TLB flushes - this should be only for SMP
2010-09-22 10:01:36 +02:00
#ifdef CONFIG_SMP
SMP - reduced TLB flushing - flush TLB of processes only if the page tables has been changed and the page tables of this process are already loaded on this cpu which means that there might be stale entries in TLB. Until now SMP was always flushing TLB to make sure everything is consistent.
2010-10-25 18:21:23 +02:00
if (p->p_misc_flags & MF_FLUSH_TLB) {
if (tlb_must_refresh)
refresh_tlb();
p->p_misc_flags &= ~MF_FLUSH_TLB;
}
Remove unnecessary TLB flushes - this should be only for SMP
2010-09-22 10:01:36 +02:00
#endif
single shot timer interrupts fix - accidentaly this wasn't part of the SMP merge and the implementation remained uncomplete with the timer keeping ticking periodically - APIC timer is set for a signel shot and restarted everytime it expires. This way we can keep the AP's trully idle - the timer is restarted a little later before leaving to userspace - LAPIC_TIMER_ICR is written before LAPIC_LVTTR so the newest value is used
2010-10-21 19:07:01 +02:00
restart_local_timer();
rename of mode/context switching functions - this patch only renames schedcheck() to switch_to_user(), cycles_accounting_stop() to context_stop() and restart() to +restore_user_context() - the motivation is that since the introduction of schedcheck() it has been abused for many things. It deserves a better name. It should express the fact that from the moment we call the function we are in the process of switching to user. - cycles_accounting_stop() was originally a single purpose function. As this function is called at were convenient places it is used in for other things too, e.g. (un)locking the kernel. Thus it deserves a better name too. - using the old name, restart() does not call schedcheck(), however calls to restart are replaced by calls to schedcheck() [switch_to_user] and it calls restart() [restore_user_context]
2010-05-18 15:00:39 +02:00
/*
* restore_user_context() carries out the actual mode switch from kernel
* to userspace. This function does not return
*/
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
restore_user_context(p);
rename of mode/context switching functions - this patch only renames schedcheck() to switch_to_user(), cycles_accounting_stop() to context_stop() and restart() to +restore_user_context() - the motivation is that since the introduction of schedcheck() it has been abused for many things. It deserves a better name. It should express the fact that from the moment we call the function we are in the process of switching to user. - cycles_accounting_stop() was originally a single purpose function. As this function is called at were convenient places it is used in for other things too, e.g. (un)locking the kernel. Thus it deserves a better name too. - using the old name, restart() does not call schedcheck(), however calls to restart are replaced by calls to schedcheck() [switch_to_user] and it calls restart() [restore_user_context]
2010-05-18 15:00:39 +02:00
NOT_REACHABLE;
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
}
Initial revision
2005-04-21 16:53:53 +02:00
do_ipc() rearrangements this patch does not add or change any functionality of do_ipc(), it only makes things a little cleaner (hopefully). Until now do_ipc() was responsible for handling all ipc calls. The catch is that SENDA is fairly different which results in some ugly code like this typecasting and variables naming which does not make much sense for SENDA and makes the code hard to read. result = mini_senda(caller_ptr, (asynmsg_t *)m_ptr, (size_t)src_dst_e); As it is called directly from assembly, the new do_ipc() takes as input values of 3 registers in reg_t variables (it used to be 4, however, bit_map wasn't used so I removed it), does the checks common to all ipc calls and call the appropriate handler either for do_sync_ipc() (all except SENDA) or mini_senda() (for SENDA) while typecasting the reg_t values correctly. As a result, handling SENDA differences in do_sync_ipc() is no more needed. Also the code that uses msg_size variable is improved a little bit. arch_do_syscall() is simplified too.
2010-04-06 13:24:26 +02:00
/*
Small fixes - do_sync_ipc() is private - fixed typo in a comment
2010-04-06 13:29:31 +02:00
* handler for all synchronous IPC calls
Initial revision
2005-04-21 16:53:53 +02:00
*/
Small fixes - do_sync_ipc() is private - fixed typo in a comment
2010-04-06 13:29:31 +02:00
PRIVATE int do_sync_ipc(struct proc * caller_ptr, /* who made the call */
do_ipc() rearrangements this patch does not add or change any functionality of do_ipc(), it only makes things a little cleaner (hopefully). Until now do_ipc() was responsible for handling all ipc calls. The catch is that SENDA is fairly different which results in some ugly code like this typecasting and variables naming which does not make much sense for SENDA and makes the code hard to read. result = mini_senda(caller_ptr, (asynmsg_t *)m_ptr, (size_t)src_dst_e); As it is called directly from assembly, the new do_ipc() takes as input values of 3 registers in reg_t variables (it used to be 4, however, bit_map wasn't used so I removed it), does the checks common to all ipc calls and call the appropriate handler either for do_sync_ipc() (all except SENDA) or mini_senda() (for SENDA) while typecasting the reg_t values correctly. As a result, handling SENDA differences in do_sync_ipc() is no more needed. Also the code that uses msg_size variable is improved a little bit. arch_do_syscall() is simplified too.
2010-04-06 13:24:26 +02:00
int call_nr, /* system call number and flags */
endpoint_t src_dst_e, /* src or dst of the call */
message *m_ptr) /* users pointer to a message */
{
Initial revision
2005-04-21 16:53:53 +02:00
int result; /* the system call's result */
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
int src_dst_p; /* Process slot number */
kernel - prettier output for ipc errors, call names instead of trap numbers
2010-07-16 17:36:29 +02:00
char *callname;
Merge of David's ptrace branch. Summary: o Support for ptrace T_ATTACH/T_DETACH and T_SYSCALL o PM signal handling logic should now work properly, even with debuggers being present o Asynchronous PM/VFS protocol, full IPC support for senda(), and AMF_NOREPLY senda() flag DETAILS Process stop and delay call handling of PM: o Added sys_runctl() kernel call with sys_stop() and sys_resume() aliases, for PM to stop and resume a process o Added exception for sending/syscall-traced processes to sys_runctl(), and matching SIGKREADY pseudo-signal to PM o Fixed PM signal logic to deal with requests from a process after stopping it (so-called "delay calls"), using the SIGKREADY facility o Fixed various PM panics due to race conditions with delay calls versus VFS calls o Removed special PRIO_STOP priority value o Added SYS_LOCK RTS kernel flag, to stop an individual process from running while modifying its process structure Signal and debugger handling in PM: o Fixed debugger signals being dropped if a second signal arrives when the debugger has not retrieved the first one o Fixed debugger signals being sent to the debugger more than once o Fixed debugger signals unpausing process in VFS; removed PM_UNPAUSE_TR protocol message o Detached debugger signals from general signal logic and from being blocked on VFS calls, meaning that even VFS can now be traced o Fixed debugger being unable to receive more than one pending signal in one process stop o Fixed signal delivery being delayed needlessly when multiple signals are pending o Fixed wait test for tracer, which was returning for children that were not waited for o Removed second parallel pending call from PM to VFS for any process o Fixed process becoming runnable between exec() and debugger trap o Added support for notifying the debugger before the parent when a debugged child exits o Fixed debugger death causing child to remain stopped forever o Fixed consistently incorrect use of _NSIG Extensions to ptrace(): o Added T_ATTACH and T_DETACH ptrace request, to attach and detach a debugger to and from a process o Added T_SYSCALL ptrace request, to trace system calls o Added T_SETOPT ptrace request, to set trace options o Added TO_TRACEFORK trace option, to attach automatically to children of a traced process o Added TO_ALTEXEC trace option, to send SIGSTOP instead of SIGTRAP upon a successful exec() of the tracee o Extended T_GETUSER ptrace support to allow retrieving a process's priv structure o Removed T_STOP ptrace request again, as it does not help implementing debuggers properly o Added MINIX3-specific ptrace test (test42) o Added proper manual page for ptrace(2) Asynchronous PM/VFS interface: o Fixed asynchronous messages not being checked when receive() is called with an endpoint other than ANY o Added AMF_NOREPLY senda() flag, preventing such messages from satisfying the receive part of a sendrec() o Added asynsend3() that takes optional flags; asynsend() is now a #define passing in 0 as third parameter o Made PM/VFS protocol asynchronous; reintroduced tell_fs() o Made PM_BASE request/reply number range unique o Hacked in a horrible temporary workaround into RS to deal with newly revealed RS-PM-VFS race condition triangle until VFS is asynchronous System signal handling: o Fixed shutdown logic of device drivers; removed old SIGKSTOP signal o Removed is-superuser check from PM's do_procstat() (aka getsigset()) o Added sigset macros to allow system processes to deal with the full signal set, rather than just the POSIX subset Miscellaneous PM fixes: o Split do_getset into do_get and do_set, merging common code and making structure clearer o Fixed setpriority() being able to put to sleep processes using an invalid parameter, or revive zombie processes o Made find_proc() global; removed obsolete proc_from_pid() o Cleanup here and there Also included: o Fixed false-positive boot order kernel warning o Removed last traces of old NOTIFY_FROM code THINGS OF POSSIBLE INTEREST o It should now be possible to run PM at any priority, even lower than user processes o No assumptions are made about communication speed between PM and VFS, although communication must be FIFO o A debugger will now receive incoming debuggee signals at kill time only; the process may not yet be fully stopped o A first step has been made towards making the SYSTEM task preemptible
2009-09-30 11:57:22 +02:00
do_ipc() rearrangements this patch does not add or change any functionality of do_ipc(), it only makes things a little cleaner (hopefully). Until now do_ipc() was responsible for handling all ipc calls. The catch is that SENDA is fairly different which results in some ugly code like this typecasting and variables naming which does not make much sense for SENDA and makes the code hard to read. result = mini_senda(caller_ptr, (asynmsg_t *)m_ptr, (size_t)src_dst_e); As it is called directly from assembly, the new do_ipc() takes as input values of 3 registers in reg_t variables (it used to be 4, however, bit_map wasn't used so I removed it), does the checks common to all ipc calls and call the appropriate handler either for do_sync_ipc() (all except SENDA) or mini_senda() (for SENDA) while typecasting the reg_t values correctly. As a result, handling SENDA differences in do_sync_ipc() is no more needed. Also the code that uses msg_size variable is improved a little bit. arch_do_syscall() is simplified too.
2010-04-06 13:24:26 +02:00
/* Check destination. RECEIVE is the only call that accepts ANY (in addition
* to a real endpoint). The other calls (SEND, SENDREC, and NOTIFY) require an
* endpoint to corresponds to a process. In addition, it is necessary to check
* whether a process is allowed to send to a given destination.
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
*/
do_ipc() rearrangements this patch does not add or change any functionality of do_ipc(), it only makes things a little cleaner (hopefully). Until now do_ipc() was responsible for handling all ipc calls. The catch is that SENDA is fairly different which results in some ugly code like this typecasting and variables naming which does not make much sense for SENDA and makes the code hard to read. result = mini_senda(caller_ptr, (asynmsg_t *)m_ptr, (size_t)src_dst_e); As it is called directly from assembly, the new do_ipc() takes as input values of 3 registers in reg_t variables (it used to be 4, however, bit_map wasn't used so I removed it), does the checks common to all ipc calls and call the appropriate handler either for do_sync_ipc() (all except SENDA) or mini_senda() (for SENDA) while typecasting the reg_t values correctly. As a result, handling SENDA differences in do_sync_ipc() is no more needed. Also the code that uses msg_size variable is improved a little bit. arch_do_syscall() is simplified too.
2010-04-06 13:24:26 +02:00
assert(call_nr != SENDA);
kernel - prettier output for ipc errors, call names instead of trap numbers
2010-07-16 17:36:29 +02:00
/* Only allow non-negative call_nr values less than 32 */
if (call_nr < 0 || call_nr > IPCNO_HIGHEST || call_nr >= 32
|| !(callname = ipc_call_names[call_nr])) {
#if DEBUG_ENABLE_IPC_WARNINGS
printf("sys_call: trap %d not allowed, caller %d, src_dst %d\n",
call_nr, proc_nr(caller_ptr), src_dst_p);
#endif
return(ETRAPDENIED); /* trap denied by mask or kernel */
}
do_ipc() rearrangements this patch does not add or change any functionality of do_ipc(), it only makes things a little cleaner (hopefully). Until now do_ipc() was responsible for handling all ipc calls. The catch is that SENDA is fairly different which results in some ugly code like this typecasting and variables naming which does not make much sense for SENDA and makes the code hard to read. result = mini_senda(caller_ptr, (asynmsg_t *)m_ptr, (size_t)src_dst_e); As it is called directly from assembly, the new do_ipc() takes as input values of 3 registers in reg_t variables (it used to be 4, however, bit_map wasn't used so I removed it), does the checks common to all ipc calls and call the appropriate handler either for do_sync_ipc() (all except SENDA) or mini_senda() (for SENDA) while typecasting the reg_t values correctly. As a result, handling SENDA differences in do_sync_ipc() is no more needed. Also the code that uses msg_size variable is improved a little bit. arch_do_syscall() is simplified too.
2010-04-06 13:24:26 +02:00
if (src_dst_e == ANY)
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
{
if (call_nr != RECEIVE)
{
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
#if 0
kernel - prettier output for ipc errors, call names instead of trap numbers
2010-07-16 17:36:29 +02:00
printf("sys_call: %s by %d with bad endpoint %d\n",
callname,
proc_nr(caller_ptr), src_dst_e);
'proc number' is process slot, 'endpoint' are generation-aware process instance numbers, encoded and decoded using macros in <minix/endpoint.h>. proc number -> endpoint migration . proc_nr in the interrupt hook is now an endpoint, proc_nr_e. . m_source for messages and notifies is now an endpoint, instead of proc number. . isokendpt() converts an endpoint to a process number, returns success (but fails if the process number is out of range, the process slot is not a living process, or the given endpoint number does not match the endpoint number in the process slot, indicating an old process). . okendpt() is the same as isokendpt(), but panic()s if the conversion fails. This is mainly used for decoding message.m_source endpoints, and other endpoint numbers in kernel data structures, which should always be correct. . if DEBUG_ENABLE_IPC_WARNINGS is enabled, isokendpt() and okendpt() get passed the __FILE__ and __LINE__ of the calling lines, and print messages about what is wrong with the endpoint number (out of range proc, empty proc, or inconsistent endpoint number), with the caller, making finding where the conversion failed easy without having to include code for every call to print where things went wrong. Sometimes this is harmless (wrong arg to a kernel call), sometimes it's a fatal internal inconsistency (bogus m_source). . some process table fields have been appended an _e to indicate it's become and endpoint. . process endpoint is stored in p_endpoint, without generation number. it turns out the kernel never needs the generation number, except when fork()ing, so it's decoded then. . kernel calls all take endpoints as arguments, not proc numbers. the one exception is sys_fork(), which needs to know in which slot to put the child.
2006-03-03 11:00:02 +01:00
#endif
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
return EINVAL;
}
do_ipc() rearrangements this patch does not add or change any functionality of do_ipc(), it only makes things a little cleaner (hopefully). Until now do_ipc() was responsible for handling all ipc calls. The catch is that SENDA is fairly different which results in some ugly code like this typecasting and variables naming which does not make much sense for SENDA and makes the code hard to read. result = mini_senda(caller_ptr, (asynmsg_t *)m_ptr, (size_t)src_dst_e); As it is called directly from assembly, the new do_ipc() takes as input values of 3 registers in reg_t variables (it used to be 4, however, bit_map wasn't used so I removed it), does the checks common to all ipc calls and call the appropriate handler either for do_sync_ipc() (all except SENDA) or mini_senda() (for SENDA) while typecasting the reg_t values correctly. As a result, handling SENDA differences in do_sync_ipc() is no more needed. Also the code that uses msg_size variable is improved a little bit. arch_do_syscall() is simplified too.
2010-04-06 13:24:26 +02:00
src_dst_p = (int) src_dst_e;
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
}
else
{
/* Require a valid source and/or destination process. */
if(!isokendpt(src_dst_e, &src_dst_p)) {
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
#if 0
kernel - prettier output for ipc errors, call names instead of trap numbers
2010-07-16 17:36:29 +02:00
printf("sys_call: %s by %d with bad endpoint %d\n",
callname,
proc_nr(caller_ptr), src_dst_e);
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
#endif
return EDEADSRCDST;
}
IPC privileges fixes Kernel: o Remove s_ipc_sendrec, instead using s_ipc_to for all send primitives o Centralize s_ipc_to bit manipulation, - disallowing assignment of bits pointing to unused priv structs; - preventing send-to-self by not setting bit for own priv struct; - preserving send mask matrix symmetry in all cases o Add IPC send mask checks to SENDA, which were missing entirely somehow o Slightly improve IPC stats accounting for SENDA o Remove SYSTEM from user processes' send mask o Half-fix the dependency between boot image order and process numbers, - correcting the table order of the boot processes; - documenting the order requirement needed for proper send masks; - warning at boot time if the order is violated RS: o Add support in /etc/drivers.conf for servers that talk to user processes, - disallowing IPC to user processes if no "ipc" field is present - adding a special "USER" label to explicitly allow IPC to user processes o Always apply IPC masks when specified; remove -i flag from service(8) o Use kernel send mask symmetry to delay adding IPC permissions for labels that do not exist yet, adding them to that label's process upon creation o Add VM to ipc permissions list for rtl8139 and fxp in drivers.conf Left to future fixes: o Removal of the table order vs process numbers dependency altogether, possibly using per-process send list structures as used for SYSTEM calls o Proper assignment of send masks to boot processes; some of the assigned (~0) masks are much wider than necessary o Proper assignment of IPC send masks for many more servers in drivers.conf o Removal of the debugging warning about the now legitimate case where RS's add_forward_ipc cannot find the IPC destination's label yet
2009-07-02 18:25:31 +02:00
/* If the call is to send to a process, i.e., for SEND, SENDNB,
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
* SENDREC or NOTIFY, verify that the caller is allowed to send to
* the given destination.
*/
IPC privileges fixes Kernel: o Remove s_ipc_sendrec, instead using s_ipc_to for all send primitives o Centralize s_ipc_to bit manipulation, - disallowing assignment of bits pointing to unused priv structs; - preventing send-to-self by not setting bit for own priv struct; - preserving send mask matrix symmetry in all cases o Add IPC send mask checks to SENDA, which were missing entirely somehow o Slightly improve IPC stats accounting for SENDA o Remove SYSTEM from user processes' send mask o Half-fix the dependency between boot image order and process numbers, - correcting the table order of the boot processes; - documenting the order requirement needed for proper send masks; - warning at boot time if the order is violated RS: o Add support in /etc/drivers.conf for servers that talk to user processes, - disallowing IPC to user processes if no "ipc" field is present - adding a special "USER" label to explicitly allow IPC to user processes o Always apply IPC masks when specified; remove -i flag from service(8) o Use kernel send mask symmetry to delay adding IPC permissions for labels that do not exist yet, adding them to that label's process upon creation o Add VM to ipc permissions list for rtl8139 and fxp in drivers.conf Left to future fixes: o Removal of the table order vs process numbers dependency altogether, possibly using per-process send list structures as used for SYSTEM calls o Proper assignment of send masks to boot processes; some of the assigned (~0) masks are much wider than necessary o Proper assignment of IPC send masks for many more servers in drivers.conf o Removal of the debugging warning about the now legitimate case where RS's add_forward_ipc cannot find the IPC destination's label yet
2009-07-02 18:25:31 +02:00
if (call_nr != RECEIVE)
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
{
IPC privileges fixes Kernel: o Remove s_ipc_sendrec, instead using s_ipc_to for all send primitives o Centralize s_ipc_to bit manipulation, - disallowing assignment of bits pointing to unused priv structs; - preventing send-to-self by not setting bit for own priv struct; - preserving send mask matrix symmetry in all cases o Add IPC send mask checks to SENDA, which were missing entirely somehow o Slightly improve IPC stats accounting for SENDA o Remove SYSTEM from user processes' send mask o Half-fix the dependency between boot image order and process numbers, - correcting the table order of the boot processes; - documenting the order requirement needed for proper send masks; - warning at boot time if the order is violated RS: o Add support in /etc/drivers.conf for servers that talk to user processes, - disallowing IPC to user processes if no "ipc" field is present - adding a special "USER" label to explicitly allow IPC to user processes o Always apply IPC masks when specified; remove -i flag from service(8) o Use kernel send mask symmetry to delay adding IPC permissions for labels that do not exist yet, adding them to that label's process upon creation o Add VM to ipc permissions list for rtl8139 and fxp in drivers.conf Left to future fixes: o Removal of the table order vs process numbers dependency altogether, possibly using per-process send list structures as used for SYSTEM calls o Proper assignment of send masks to boot processes; some of the assigned (~0) masks are much wider than necessary o Proper assignment of IPC send masks for many more servers in drivers.conf o Removal of the debugging warning about the now legitimate case where RS's add_forward_ipc cannot find the IPC destination's label yet
2009-07-02 18:25:31 +02:00
if (!may_send_to(caller_ptr, src_dst_p)) {
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
#if DEBUG_ENABLE_IPC_WARNINGS
no more kprintf - kernel uses libsys printf now, only kputc is special to the kernel.
2010-03-03 16:45:01 +01:00
printf(
kernel - prettier output for ipc errors, call names instead of trap numbers
2010-07-16 17:36:29 +02:00
"sys_call: ipc mask denied %s from %d to %d\n",
callname,
caller_ptr->p_endpoint, src_dst_e);
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
#endif
return(ECALLDENIED); /* call denied by ipc mask */
}
}
}
Moved stime, time, times POSIX calls from FS to PM. Removed child time accounting from kernel (now in PM). Large amount of files in this commit is due to system time problems during development.
2005-05-31 11:50:51 +02:00
/* Check if the process has privileges for the requested call. Calls to the
* kernel may only be SENDREC, because tasks always reply and may not block
* if the caller doesn't do receive().
Initial revision
2005-04-21 16:53:53 +02:00
*/
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
if (!(priv(caller_ptr)->s_trap_mask & (1 << call_nr))) {
Turn IPC warning messages off by default (because inet deadlocks are normal and will confuse our users).
2005-10-18 18:13:12 +02:00
#if DEBUG_ENABLE_IPC_WARNINGS
kernel - prettier output for ipc errors, call names instead of trap numbers
2010-07-16 17:36:29 +02:00
printf("sys_call: %s not allowed, caller %d, src_dst %d\n",
callname, proc_nr(caller_ptr), src_dst_p);
Turn IPC warning messages off by default (because inet deadlocks are normal and will confuse our users).
2005-10-18 18:13:12 +02:00
#endif
Keep track of various statistics related to IPC and SYSTEM.
2008-02-22 13:36:46 +01:00
return(ETRAPDENIED); /* trap denied by mask or kernel */
Fix to device table at FS. BIOS and AT installed in /sbin. Floppy boot fixed.
2005-08-05 20:57:20 +02:00
}
Optimized scheduling code. Old code is still available withing DEAD_CODE and NEW_SCHED_Q definitions. Some minor problems are being traced at the moment. This commit is meant to backup my files. --- Jorrit
2005-05-26 15:17:57 +02:00
do_ipc() rearrangements this patch does not add or change any functionality of do_ipc(), it only makes things a little cleaner (hopefully). Until now do_ipc() was responsible for handling all ipc calls. The catch is that SENDA is fairly different which results in some ugly code like this typecasting and variables naming which does not make much sense for SENDA and makes the code hard to read. result = mini_senda(caller_ptr, (asynmsg_t *)m_ptr, (size_t)src_dst_e); As it is called directly from assembly, the new do_ipc() takes as input values of 3 registers in reg_t variables (it used to be 4, however, bit_map wasn't used so I removed it), does the checks common to all ipc calls and call the appropriate handler either for do_sync_ipc() (all except SENDA) or mini_senda() (for SENDA) while typecasting the reg_t values correctly. As a result, handling SENDA differences in do_sync_ipc() is no more needed. Also the code that uses msg_size variable is improved a little bit. arch_do_syscall() is simplified too.
2010-04-06 13:24:26 +02:00
if (call_nr != SENDREC && call_nr != RECEIVE && iskerneln(src_dst_p)) {
Turn IPC warning messages off by default (because inet deadlocks are normal and will confuse our users).
2005-10-18 18:13:12 +02:00
#if DEBUG_ENABLE_IPC_WARNINGS
no more kprintf - kernel uses libsys printf now, only kputc is special to the kernel.
2010-03-03 16:45:01 +01:00
printf("sys_call: trap %d not allowed, caller %d, src_dst %d\n",
kernel - prettier output for ipc errors, call names instead of trap numbers
2010-07-16 17:36:29 +02:00
callname, proc_nr(caller_ptr), src_dst_e);
Turn IPC warning messages off by default (because inet deadlocks are normal and will confuse our users).
2005-10-18 18:13:12 +02:00
#endif
Keep track of various statistics related to IPC and SYSTEM.
2008-02-22 13:36:46 +01:00
return(ETRAPDENIED); /* trap denied by mask or kernel */
Kernel changes: - reinstalled priority changing, now in sched() and unready() - reinstalled check on message buffer in sys_call() - reinstalled check in send masks in sys_call() - changed do_fork() to get new privilege structure for SYS_PROCs - removed some processes from boot image---will be dynamically started later
2005-07-26 14:48:34 +02:00
}
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
switch(call_nr) {
Added SENDREC_BUSY flag to indicate that process is doing sendrec() - if so, alerts are not received before reply message.
2005-07-27 16:08:59 +02:00
case SENDREC:
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
/* A flag is set so that notifications cannot interrupt SENDREC. */
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
caller_ptr->p_misc_flags |= MF_REPLY_PEND;
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
/* fall through */
Improved NOTIFY system: fixed a minor error, ignore pending notifications on SENDREC system calls. To be done: resource (buffer pool) management; make it structurally impossible to run out of buffers.
2005-05-27 14:44:14 +02:00
case SEND:
Basic VM and other minor improvements. Not complete, probably not fully debugged or optimized.
2008-11-19 13:26:10 +01:00
result = mini_send(caller_ptr, src_dst_e, m_ptr, 0);
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
if (call_nr == SEND || result != OK)
break; /* done, or SEND failed */
/* fall through for SENDREC */
Optimized scheduling code. Old code is still available withing DEAD_CODE and NEW_SCHED_Q definitions. Some minor problems are being traced at the moment. This commit is meant to backup my files. --- Jorrit
2005-05-26 15:17:57 +02:00
case RECEIVE:
Set IPC status code only for RECEIVE
2010-04-26 16:43:59 +02:00
if (call_nr == RECEIVE) {
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
caller_ptr->p_misc_flags &= ~MF_REPLY_PEND;
Set IPC status code only for RECEIVE
2010-04-26 16:43:59 +02:00
IPC_STATUS_CLEAR(caller_ptr); /* clear IPC status code */
}
Basic VM and other minor improvements. Not complete, probably not fully debugged or optimized.
2008-11-19 13:26:10 +01:00
result = mini_receive(caller_ptr, src_dst_e, m_ptr, 0);
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
break;
Optimized scheduling code. Old code is still available withing DEAD_CODE and NEW_SCHED_Q definitions. Some minor problems are being traced at the moment. This commit is meant to backup my files. --- Jorrit
2005-05-26 15:17:57 +02:00
case NOTIFY:
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
result = mini_notify(caller_ptr, src_dst_e);
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
break;
Basic VM and other minor improvements. Not complete, probably not fully debugged or optimized.
2008-11-19 13:26:10 +01:00
case SENDNB:
result = mini_send(caller_ptr, src_dst_e, m_ptr, NON_BLOCKING);
break;
Optimized scheduling code. Old code is still available withing DEAD_CODE and NEW_SCHED_Q definitions. Some minor problems are being traced at the moment. This commit is meant to backup my files. --- Jorrit
2005-05-26 15:17:57 +02:00
default:
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
result = EBADCALL; /* illegal system call */
Initial revision
2005-04-21 16:53:53 +02:00
}
/* Now, return the result of the system call to the caller. */
return(result);
}
do_ipc() rearrangements this patch does not add or change any functionality of do_ipc(), it only makes things a little cleaner (hopefully). Until now do_ipc() was responsible for handling all ipc calls. The catch is that SENDA is fairly different which results in some ugly code like this typecasting and variables naming which does not make much sense for SENDA and makes the code hard to read. result = mini_senda(caller_ptr, (asynmsg_t *)m_ptr, (size_t)src_dst_e); As it is called directly from assembly, the new do_ipc() takes as input values of 3 registers in reg_t variables (it used to be 4, however, bit_map wasn't used so I removed it), does the checks common to all ipc calls and call the appropriate handler either for do_sync_ipc() (all except SENDA) or mini_senda() (for SENDA) while typecasting the reg_t values correctly. As a result, handling SENDA differences in do_sync_ipc() is no more needed. Also the code that uses msg_size variable is improved a little bit. arch_do_syscall() is simplified too.
2010-04-06 13:24:26 +02:00
PUBLIC int do_ipc(reg_t r1, reg_t r2, reg_t r3)
{
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
struct proc *const caller_ptr = get_cpulocal_var(proc_ptr); /* get pointer to caller */
do_ipc() rearrangements this patch does not add or change any functionality of do_ipc(), it only makes things a little cleaner (hopefully). Until now do_ipc() was responsible for handling all ipc calls. The catch is that SENDA is fairly different which results in some ugly code like this typecasting and variables naming which does not make much sense for SENDA and makes the code hard to read. result = mini_senda(caller_ptr, (asynmsg_t *)m_ptr, (size_t)src_dst_e); As it is called directly from assembly, the new do_ipc() takes as input values of 3 registers in reg_t variables (it used to be 4, however, bit_map wasn't used so I removed it), does the checks common to all ipc calls and call the appropriate handler either for do_sync_ipc() (all except SENDA) or mini_senda() (for SENDA) while typecasting the reg_t values correctly. As a result, handling SENDA differences in do_sync_ipc() is no more needed. Also the code that uses msg_size variable is improved a little bit. arch_do_syscall() is simplified too.
2010-04-06 13:24:26 +02:00
int call_nr = (int) r1;
assert(!RTS_ISSET(caller_ptr, RTS_SLOT_FREE));
kernel - account for kernel cpu time (ipc, kcalls) in caller
2011-02-08 14:58:32 +01:00
/* bill kernel time to this process. */
kbill_ipc = caller_ptr;
do_ipc() rearrangements this patch does not add or change any functionality of do_ipc(), it only makes things a little cleaner (hopefully). Until now do_ipc() was responsible for handling all ipc calls. The catch is that SENDA is fairly different which results in some ugly code like this typecasting and variables naming which does not make much sense for SENDA and makes the code hard to read. result = mini_senda(caller_ptr, (asynmsg_t *)m_ptr, (size_t)src_dst_e); As it is called directly from assembly, the new do_ipc() takes as input values of 3 registers in reg_t variables (it used to be 4, however, bit_map wasn't used so I removed it), does the checks common to all ipc calls and call the appropriate handler either for do_sync_ipc() (all except SENDA) or mini_senda() (for SENDA) while typecasting the reg_t values correctly. As a result, handling SENDA differences in do_sync_ipc() is no more needed. Also the code that uses msg_size variable is improved a little bit. arch_do_syscall() is simplified too.
2010-04-06 13:24:26 +02:00
/* If this process is subject to system call tracing, handle that first. */
if (caller_ptr->p_misc_flags & (MF_SC_TRACE | MF_SC_DEFER)) {
/* Are we tracing this process, and is it the first sys_call entry? */
if ((caller_ptr->p_misc_flags & (MF_SC_TRACE | MF_SC_DEFER)) ==
MF_SC_TRACE) {
/* We must notify the tracer before processing the actual
* system call. If we don't, the tracer could not obtain the
* input message. Postpone the entire system call.
*/
caller_ptr->p_misc_flags &= ~MF_SC_TRACE;
caller_ptr->p_misc_flags |= MF_SC_DEFER;
/* Signal the "enter system call" event. Block the process. */
cause_sig(proc_nr(caller_ptr), SIGTRAP);
/* Preserve the return register's value. */
return caller_ptr->p_reg.retreg;
}
/* If the MF_SC_DEFER flag is set, the syscall is now being resumed. */
caller_ptr->p_misc_flags &= ~MF_SC_DEFER;
assert (!(caller_ptr->p_misc_flags & MF_SC_ACTIVE));
/* Set a flag to allow reliable tracing of leaving the system call. */
caller_ptr->p_misc_flags |= MF_SC_ACTIVE;
}
if(caller_ptr->p_misc_flags & MF_DELIVERMSG) {
panic("sys_call: MF_DELIVERMSG on for %s / %d\n",
caller_ptr->p_name, caller_ptr->p_endpoint);
}
/* Now check if the call is known and try to perform the request. The only
* system calls that exist in MINIX are sending and receiving messages.
* - SENDREC: combines SEND and RECEIVE in a single system call
* - SEND: sender blocks until its message has been delivered
* - RECEIVE: receiver blocks until an acceptable message has arrived
* - NOTIFY: asynchronous call; deliver notification or mark pending
* - SENDA: list of asynchronous send requests
*/
switch(call_nr) {
case SENDREC:
case SEND:
case RECEIVE:
case NOTIFY:
case SENDNB:
Userspace scheduling - exporting stats - contributed by Bjorn Swift - adds process accounting, for example counting the number of messages sent, how often the process was preemted and how much time it spent in the run queue. These statistics, along with the current cpu load, are sent back to the user-space scheduler in the Out Of Quantum message. - the user-space scheduler may choose to make use of these statistics when making scheduling decisions. For isntance the cpu load becomes especially useful when scheduling on multiple cores.
2010-09-19 17:52:12 +02:00
{
/* Process accounting for scheduling */
caller_ptr->p_accounting.ipc_sync++;
do_ipc() rearrangements this patch does not add or change any functionality of do_ipc(), it only makes things a little cleaner (hopefully). Until now do_ipc() was responsible for handling all ipc calls. The catch is that SENDA is fairly different which results in some ugly code like this typecasting and variables naming which does not make much sense for SENDA and makes the code hard to read. result = mini_senda(caller_ptr, (asynmsg_t *)m_ptr, (size_t)src_dst_e); As it is called directly from assembly, the new do_ipc() takes as input values of 3 registers in reg_t variables (it used to be 4, however, bit_map wasn't used so I removed it), does the checks common to all ipc calls and call the appropriate handler either for do_sync_ipc() (all except SENDA) or mini_senda() (for SENDA) while typecasting the reg_t values correctly. As a result, handling SENDA differences in do_sync_ipc() is no more needed. Also the code that uses msg_size variable is improved a little bit. arch_do_syscall() is simplified too.
2010-04-06 13:24:26 +02:00
return do_sync_ipc(caller_ptr, call_nr, (endpoint_t) r2,
(message *) r3);
Userspace scheduling - exporting stats - contributed by Bjorn Swift - adds process accounting, for example counting the number of messages sent, how often the process was preemted and how much time it spent in the run queue. These statistics, along with the current cpu load, are sent back to the user-space scheduler in the Out Of Quantum message. - the user-space scheduler may choose to make use of these statistics when making scheduling decisions. For isntance the cpu load becomes especially useful when scheduling on multiple cores.
2010-09-19 17:52:12 +02:00
}
do_ipc() rearrangements this patch does not add or change any functionality of do_ipc(), it only makes things a little cleaner (hopefully). Until now do_ipc() was responsible for handling all ipc calls. The catch is that SENDA is fairly different which results in some ugly code like this typecasting and variables naming which does not make much sense for SENDA and makes the code hard to read. result = mini_senda(caller_ptr, (asynmsg_t *)m_ptr, (size_t)src_dst_e); As it is called directly from assembly, the new do_ipc() takes as input values of 3 registers in reg_t variables (it used to be 4, however, bit_map wasn't used so I removed it), does the checks common to all ipc calls and call the appropriate handler either for do_sync_ipc() (all except SENDA) or mini_senda() (for SENDA) while typecasting the reg_t values correctly. As a result, handling SENDA differences in do_sync_ipc() is no more needed. Also the code that uses msg_size variable is improved a little bit. arch_do_syscall() is simplified too.
2010-04-06 13:24:26 +02:00
case SENDA:
{
Userspace scheduling - exporting stats - contributed by Bjorn Swift - adds process accounting, for example counting the number of messages sent, how often the process was preemted and how much time it spent in the run queue. These statistics, along with the current cpu load, are sent back to the user-space scheduler in the Out Of Quantum message. - the user-space scheduler may choose to make use of these statistics when making scheduling decisions. For isntance the cpu load becomes especially useful when scheduling on multiple cores.
2010-09-19 17:52:12 +02:00
/*
do_ipc() rearrangements this patch does not add or change any functionality of do_ipc(), it only makes things a little cleaner (hopefully). Until now do_ipc() was responsible for handling all ipc calls. The catch is that SENDA is fairly different which results in some ugly code like this typecasting and variables naming which does not make much sense for SENDA and makes the code hard to read. result = mini_senda(caller_ptr, (asynmsg_t *)m_ptr, (size_t)src_dst_e); As it is called directly from assembly, the new do_ipc() takes as input values of 3 registers in reg_t variables (it used to be 4, however, bit_map wasn't used so I removed it), does the checks common to all ipc calls and call the appropriate handler either for do_sync_ipc() (all except SENDA) or mini_senda() (for SENDA) while typecasting the reg_t values correctly. As a result, handling SENDA differences in do_sync_ipc() is no more needed. Also the code that uses msg_size variable is improved a little bit. arch_do_syscall() is simplified too.
2010-04-06 13:24:26 +02:00
* Get and check the size of the argument in bytes as it is a
* table
*/
size_t msg_size = (size_t) r2;
Userspace scheduling - exporting stats - contributed by Bjorn Swift - adds process accounting, for example counting the number of messages sent, how often the process was preemted and how much time it spent in the run queue. These statistics, along with the current cpu load, are sent back to the user-space scheduler in the Out Of Quantum message. - the user-space scheduler may choose to make use of these statistics when making scheduling decisions. For isntance the cpu load becomes especially useful when scheduling on multiple cores.
2010-09-19 17:52:12 +02:00
/* Process accounting for scheduling */
caller_ptr->p_accounting.ipc_async++;
do_ipc() rearrangements this patch does not add or change any functionality of do_ipc(), it only makes things a little cleaner (hopefully). Until now do_ipc() was responsible for handling all ipc calls. The catch is that SENDA is fairly different which results in some ugly code like this typecasting and variables naming which does not make much sense for SENDA and makes the code hard to read. result = mini_senda(caller_ptr, (asynmsg_t *)m_ptr, (size_t)src_dst_e); As it is called directly from assembly, the new do_ipc() takes as input values of 3 registers in reg_t variables (it used to be 4, however, bit_map wasn't used so I removed it), does the checks common to all ipc calls and call the appropriate handler either for do_sync_ipc() (all except SENDA) or mini_senda() (for SENDA) while typecasting the reg_t values correctly. As a result, handling SENDA differences in do_sync_ipc() is no more needed. Also the code that uses msg_size variable is improved a little bit. arch_do_syscall() is simplified too.
2010-04-06 13:24:26 +02:00
/* Limit size to something reasonable. An arbitrary choice is 16
* times the number of process table entries.
*/
if (msg_size > 16*(NR_TASKS + NR_PROCS))
return EDOM;
return mini_senda(caller_ptr, (asynmsg_t *) r3, msg_size);
}
default:
return EBADCALL; /* illegal system call */
}
}
New deadlock check at proc.c. New exit checks (stop IPC).
2005-10-12 17:08:23 +02:00
/*===========================================================================*
* deadlock *
*===========================================================================*/
kernel: deadlock test with endpoints instead of slot numbers, slightly cleaner
2010-07-28 16:14:06 +02:00
PRIVATE int deadlock(function, cp, src_dst_e)
New deadlock check at proc.c. New exit checks (stop IPC).
2005-10-12 17:08:23 +02:00
int function; /* trap number */
register struct proc *cp; /* pointer to caller */
kernel: deadlock test with endpoints instead of slot numbers, slightly cleaner
2010-07-28 16:14:06 +02:00
endpoint_t src_dst_e; /* src or dst process */
New deadlock check at proc.c. New exit checks (stop IPC).
2005-10-12 17:08:23 +02:00
{
/* Check for deadlock. This can happen if 'caller_ptr' and 'src_dst' have
* a cyclic dependency of blocking send and receive calls. The only cyclic
* depency that is not fatal is if the caller and target directly SEND(REC)
* and RECEIVE to each other. If a deadlock is found, the group size is
* returned. Otherwise zero is returned.
*/
register struct proc *xp; /* process pointer */
int group_size = 1; /* start with only caller */
Basic VM and other minor improvements. Not complete, probably not fully debugged or optimized.
2008-11-19 13:26:10 +01:00
#if DEBUG_ENABLE_IPC_WARNINGS
static struct proc *processes[NR_PROCS + NR_TASKS];
processes[0] = cp;
#endif
New deadlock check at proc.c. New exit checks (stop IPC).
2005-10-12 17:08:23 +02:00
kernel: deadlock test with endpoints instead of slot numbers, slightly cleaner
2010-07-28 16:14:06 +02:00
while (src_dst_e != ANY) { /* check while process nr */
int src_dst_slot;
okendpt(src_dst_e, &src_dst_slot);
xp = proc_addr(src_dst_slot); /* follow chain of processes */
assert(proc_ptr_ok(xp));
assert(!RTS_ISSET(xp, RTS_SLOT_FREE));
Basic VM and other minor improvements. Not complete, probably not fully debugged or optimized.
2008-11-19 13:26:10 +01:00
#if DEBUG_ENABLE_IPC_WARNINGS
processes[group_size] = xp;
#endif
New deadlock check at proc.c. New exit checks (stop IPC).
2005-10-12 17:08:23 +02:00
group_size ++; /* extra process in group */
Typo in comment.
2006-03-09 14:59:59 +01:00
/* Check whether the last process in the chain has a dependency. If it
New deadlock check at proc.c. New exit checks (stop IPC).
2005-10-12 17:08:23 +02:00
* has not, the cycle cannot be closed and we are done.
*/
kernel: deadlock test with endpoints instead of slot numbers, slightly cleaner
2010-07-28 16:14:06 +02:00
if((src_dst_e = P_BLOCKEDON(xp)) == NONE)
New P_BLOCKEDON for kernel - a macro that encodes the "who is this process waiting for" logic, which is duplicated a few times in the kernel. (For a new feature for top.) Introducing it and throwing out ESRCDIED and EDSTDIED (replaced by EDEADSRCDST - so we don't have to care which part of the blocking is failing in system.c) simplifies some code in the kernel and callers that check for E{DEADSRCDST,ESRCDIED,EDSTDIED}, but don't care about the difference, a fair bit, and more significantly doesn't duplicate the 'blocked-on' logic.
2010-03-03 16:32:26 +01:00
return 0;
New deadlock check at proc.c. New exit checks (stop IPC).
2005-10-12 17:08:23 +02:00
/* Now check if there is a cyclic dependency. For group sizes of two,
* a combination of SEND(REC) and RECEIVE is not fatal. Larger groups
* or other combinations indicate a deadlock.
*/
kernel: deadlock test with endpoints instead of slot numbers, slightly cleaner
2010-07-28 16:14:06 +02:00
if (src_dst_e == cp->p_endpoint) { /* possible deadlock */
New deadlock check at proc.c. New exit checks (stop IPC).
2005-10-12 17:08:23 +02:00
if (group_size == 2) { /* caller and src_dst */
/* The function number is magically converted to flags. */
All macros defining rts flags are prefixed with RTS_ - macros used with RTS_SET group of macros to define struct proc p_rts_flags are now prefixed with RTS_ to make things clear
2009-11-10 10:11:13 +01:00
if ((xp->p_rts_flags ^ (function << 2)) & RTS_SENDING) {
New deadlock check at proc.c. New exit checks (stop IPC).
2005-10-12 17:08:23 +02:00
return(0); /* not a deadlock */
}
}
Basic VM and other minor improvements. Not complete, probably not fully debugged or optimized.
2008-11-19 13:26:10 +01:00
#if DEBUG_ENABLE_IPC_WARNINGS
{
int i;
no more kprintf - kernel uses libsys printf now, only kputc is special to the kernel.
2010-03-03 16:45:01 +01:00
printf("deadlock between these processes:\n");
Basic VM and other minor improvements. Not complete, probably not fully debugged or optimized.
2008-11-19 13:26:10 +01:00
for(i = 0; i < group_size; i++) {
no more kprintf - kernel uses libsys printf now, only kputc is special to the kernel.
2010-03-03 16:45:01 +01:00
printf(" %10s ", processes[i]->p_name);
deadlock() - more info - deadlock() is more verbose in case of a detected deadlock. First, it lists all processses in the deadlock group. Then it prints the proc extra info, not only stack trace and register dump
2010-05-03 19:38:54 +02:00
}
printf("\n\n");
for(i = 0; i < group_size; i++) {
print_proc(processes[i]);
Basic VM and other minor improvements. Not complete, probably not fully debugged or optimized.
2008-11-19 13:26:10 +01:00
proc_stacktrace(processes[i]);
}
}
#endif
New deadlock check at proc.c. New exit checks (stop IPC).
2005-10-12 17:08:23 +02:00
return(group_size); /* deadlock found */
}
}
return(0); /* not a deadlock */
}
Use a bitmap for pending asynchronous messages instead of a global flag. That way it works similar to pending notifications.
2011-04-08 17:03:33 +02:00
/*===========================================================================*
* has_pending *
*===========================================================================*/
Enable sending a notification when sending of an asynchronous message was completed (successfully or not). AMF_NOTIFY_ERR can be used if the sender only wishes to be notified in case of an error (e.g., EDEADSRCDST). A new endpoint ASYNCM will be the sender of the notification.
2011-04-08 17:14:48 +02:00
PUBLIC int has_pending(sys_map_t *map, int src_p)
Use a bitmap for pending asynchronous messages instead of a global flag. That way it works similar to pending notifications.
2011-04-08 17:03:33 +02:00
{
/* Check to see if there is a pending message from the desired source
* available.
*/
int src_id;
sys_id_t id = NULL_PRIV_ID;
/* Either check a specific bit in the mask map, or find the first bit set in
* it (if any), depending on whether the receive was called on a specific
* source endpoint.
*/
if (src_p != ANY) {
src_id = nr_to_id(src_p);
if (get_sys_bit(*map, src_id))
id = src_id;
} else {
/* Find a source with a pending message */
for (src_id = 0; src_id < NR_SYS_PROCS; src_id += BITCHUNK_BITS) {
if (get_sys_bits(*map, src_id) != 0) {
while(!get_sys_bit(*map, src_id)) src_id++;
break;
}
}
if (src_id < NR_SYS_PROCS) /* Found one */
id = src_id;
}
return(id);
}
Initial revision
2005-04-21 16:53:53 +02:00
/*===========================================================================*
* mini_send *
*===========================================================================*/
Use endpoint_t for the destination of mini_send and _syscall, and the source of mini_receive. Also some small cleanup.
2010-06-02 23:51:32 +02:00
PUBLIC int mini_send(
register struct proc *caller_ptr, /* who is trying to send a message? */
endpoint_t dst_e, /* to whom is message being sent? */
message *m_ptr, /* pointer to message buffer */
const int flags
)
Initial revision
2005-04-21 16:53:53 +02:00
{
New NOTIFY system call! Queued at kernel. Duplicate messages (with same source and type) are overwritten with newer flags/ arguments. The interface from within the kernel is lock_notify(). User processes can make a system call with notify(). NOTIFY fully replaces the old notification mechanism.
2005-05-24 12:06:17 +02:00
/* Send a message from 'caller_ptr' to 'dst'. If 'dst' is blocked waiting
* for this message, copy the message to it and unblock 'dst'. If 'dst' is
Initial revision
2005-04-21 16:53:53 +02:00
* not waiting at all, or is waiting for another source, queue 'caller_ptr'.
*/
'proc number' is process slot, 'endpoint' are generation-aware process instance numbers, encoded and decoded using macros in <minix/endpoint.h>. proc number -> endpoint migration . proc_nr in the interrupt hook is now an endpoint, proc_nr_e. . m_source for messages and notifies is now an endpoint, instead of proc number. . isokendpt() converts an endpoint to a process number, returns success (but fails if the process number is out of range, the process slot is not a living process, or the given endpoint number does not match the endpoint number in the process slot, indicating an old process). . okendpt() is the same as isokendpt(), but panic()s if the conversion fails. This is mainly used for decoding message.m_source endpoints, and other endpoint numbers in kernel data structures, which should always be correct. . if DEBUG_ENABLE_IPC_WARNINGS is enabled, isokendpt() and okendpt() get passed the __FILE__ and __LINE__ of the calling lines, and print messages about what is wrong with the endpoint number (out of range proc, empty proc, or inconsistent endpoint number), with the caller, making finding where the conversion failed easy without having to include code for every call to print where things went wrong. Sometimes this is harmless (wrong arg to a kernel call), sometimes it's a fatal internal inconsistency (bogus m_source). . some process table fields have been appended an _e to indicate it's become and endpoint. . process endpoint is stored in p_endpoint, without generation number. it turns out the kernel never needs the generation number, except when fork()ing, so it's decoded then. . kernel calls all take endpoints as arguments, not proc numbers. the one exception is sys_fork(), which needs to know in which slot to put the child.
2006-03-03 11:00:02 +01:00
register struct proc *dst_ptr;
Optimized scheduling code. Old code is still available withing DEAD_CODE and NEW_SCHED_Q definitions. Some minor problems are being traced at the moment. This commit is meant to backup my files. --- Jorrit
2005-05-26 15:17:57 +02:00
register struct proc **xpp;
'proc number' is process slot, 'endpoint' are generation-aware process instance numbers, encoded and decoded using macros in <minix/endpoint.h>. proc number -> endpoint migration . proc_nr in the interrupt hook is now an endpoint, proc_nr_e. . m_source for messages and notifies is now an endpoint, instead of proc number. . isokendpt() converts an endpoint to a process number, returns success (but fails if the process number is out of range, the process slot is not a living process, or the given endpoint number does not match the endpoint number in the process slot, indicating an old process). . okendpt() is the same as isokendpt(), but panic()s if the conversion fails. This is mainly used for decoding message.m_source endpoints, and other endpoint numbers in kernel data structures, which should always be correct. . if DEBUG_ENABLE_IPC_WARNINGS is enabled, isokendpt() and okendpt() get passed the __FILE__ and __LINE__ of the calling lines, and print messages about what is wrong with the endpoint number (out of range proc, empty proc, or inconsistent endpoint number), with the caller, making finding where the conversion failed easy without having to include code for every call to print where things went wrong. Sometimes this is harmless (wrong arg to a kernel call), sometimes it's a fatal internal inconsistency (bogus m_source). . some process table fields have been appended an _e to indicate it's become and endpoint. . process endpoint is stored in p_endpoint, without generation number. it turns out the kernel never needs the generation number, except when fork()ing, so it's decoded then. . kernel calls all take endpoints as arguments, not proc numbers. the one exception is sys_fork(), which needs to know in which slot to put the child.
2006-03-03 11:00:02 +01:00
int dst_p;
dst_p = _ENDPOINT_P(dst_e);
dst_ptr = proc_addr(dst_p);
Initial revision
2005-04-21 16:53:53 +02:00
All macros defining rts flags are prefixed with RTS_ - macros used with RTS_SET group of macros to define struct proc p_rts_flags are now prefixed with RTS_ to make things clear
2009-11-10 10:11:13 +01:00
if (RTS_ISSET(dst_ptr, RTS_NO_ENDPOINT))
Keep track of various statistics related to IPC and SYSTEM.
2008-02-22 13:36:46 +01:00
{
New P_BLOCKEDON for kernel - a macro that encodes the "who is this process waiting for" logic, which is duplicated a few times in the kernel. (For a new feature for top.) Introducing it and throwing out ESRCDIED and EDSTDIED (replaced by EDEADSRCDST - so we don't have to care which part of the blocking is failing in system.c) simplifies some code in the kernel and callers that check for E{DEADSRCDST,ESRCDIED,EDSTDIED}, but don't care about the difference, a fair bit, and more significantly doesn't duplicate the 'blocked-on' logic.
2010-03-03 16:32:26 +01:00
return EDEADSRCDST;
Keep track of various statistics related to IPC and SYSTEM.
2008-02-22 13:36:46 +01:00
}
Allow a process to kill itself. Remove the endpoint of a process that kills itself before reporting the singal to pm.
2006-03-15 13:01:59 +01:00
Optimized scheduling code. Old code is still available withing DEAD_CODE and NEW_SCHED_Q definitions. Some minor problems are being traced at the moment. This commit is meant to backup my files. --- Jorrit
2005-05-26 15:17:57 +02:00
/* Check if 'dst' is blocked waiting for this message. The destination's
All macros defining rts flags are prefixed with RTS_ - macros used with RTS_SET group of macros to define struct proc p_rts_flags are now prefixed with RTS_ to make things clear
2009-11-10 10:11:13 +01:00
* RTS_SENDING flag may be set when its SENDREC call blocked while sending.
Optimized scheduling code. Old code is still available withing DEAD_CODE and NEW_SCHED_Q definitions. Some minor problems are being traced at the moment. This commit is meant to backup my files. --- Jorrit
2005-05-26 15:17:57 +02:00
*/
Basic VM and other minor improvements. Not complete, probably not fully debugged or optimized.
2008-11-19 13:26:10 +01:00
if (WILLRECEIVE(dst_ptr, caller_ptr->p_endpoint)) {
IPC status code for receive(). IPC changes: - receive() is changed to take an additional parameter, which is a pointer to a status code. - The status code is filled in by the kernel to provide additional information to the caller. For now, the kernel only fills in the IPC call used by the sender. Syslib changes: - sef_receive() has been split into sef_receive() (with the original semantics) and sef_receive_status() which exposes the status code to userland. - Ideally, every sys process should gradually switch to sef_receive_status() and use is_ipc_notify() as a dependable way to check for notify. - SEF has been modified to use is_ipc_notify() and demonstrate how to use the new status code.
2010-03-23 01:09:11 +01:00
int call;
Initial revision
2005-04-21 16:53:53 +02:00
/* Destination is indeed waiting for this message. */
re-establish kernel assert()s. use the regular <assert.h> assert() instead of vmassert() in kernel. throw out some #if 0 code. fix a few assert() conditions. enable by default.
2010-03-10 14:00:05 +01:00
assert(!(dst_ptr->p_misc_flags & MF_DELIVERMSG));
Slightly faster IPC - there are cycles wasted in the IPC call due to a fairly compliacted way of copying messages from userland to kernel. Sometimes this complicated way (generic though) is used even for copying within the kernel address space, sometimes it is used for copying in case _no_ copying is necessary. The goal of this patch is to improve this a little bit. - the places where a copy is from user to kernel use the copy_msg_from_user() kernel-kernel copies are turned into assignments and BuildNotifyMessage uses the delivery buffers to avoid copying. - copy_msg_from_user() was introduced when removing the system task and is about 2/3 faster then using the current mechanism (phys_copy). It also avoids the PHYS_COPY_CATCH macro. Assignment is also faster and no copy is the fastest ;-) so perhaps there will be some hardly noticable performance gain besides the clean up.
2010-03-29 13:16:37 +02:00
if (!(flags & FROM_KERNEL)) {
if(copy_msg_from_user(caller_ptr, m_ptr, &dst_ptr->p_delivermsg))
return EFAULT;
A reliable way for userspace to check if a msg is from kernel - IPC_FLG_MSG_FROM_KERNEL status flag is returned to userspace if the receive was satisfied by s message which was sent by the kernel on behalf of a process. This perfectly reliale information. - MF_SENDING_FROM_KERNEL flag added to processes to be able to set IPC_FLG_MSG_FROM_KERNEL when finishing receive if the receiver wasn't ready to receive immediately. - PM is changed to use this information to confirm that the scheduling messages are indeed from the kernel and not faked by a process. PM uses sef_receive_status() - get_work() is removed from PM to make the changes simpler
2010-03-29 13:25:01 +02:00
} else {
Slightly faster IPC - there are cycles wasted in the IPC call due to a fairly compliacted way of copying messages from userland to kernel. Sometimes this complicated way (generic though) is used even for copying within the kernel address space, sometimes it is used for copying in case _no_ copying is necessary. The goal of this patch is to improve this a little bit. - the places where a copy is from user to kernel use the copy_msg_from_user() kernel-kernel copies are turned into assignments and BuildNotifyMessage uses the delivery buffers to avoid copying. - copy_msg_from_user() was introduced when removing the system task and is about 2/3 faster then using the current mechanism (phys_copy). It also avoids the PHYS_COPY_CATCH macro. Assignment is also faster and no copy is the fastest ;-) so perhaps there will be some hardly noticable performance gain besides the clean up.
2010-03-29 13:16:37 +02:00
dst_ptr->p_delivermsg = *m_ptr;
Set IPC status code only for RECEIVE
2010-04-26 16:43:59 +02:00
IPC_STATUS_ADD_FLAGS(dst_ptr, IPC_FLG_MSG_FROM_KERNEL);
A reliable way for userspace to check if a msg is from kernel - IPC_FLG_MSG_FROM_KERNEL status flag is returned to userspace if the receive was satisfied by s message which was sent by the kernel on behalf of a process. This perfectly reliale information. - MF_SENDING_FROM_KERNEL flag added to processes to be able to set IPC_FLG_MSG_FROM_KERNEL when finishing receive if the receiver wasn't ready to receive immediately. - PM is changed to use this information to confirm that the scheduling messages are indeed from the kernel and not faked by a process. PM uses sef_receive_status() - get_work() is removed from PM to make the changes simpler
2010-03-29 13:25:01 +02:00
}
Slightly faster IPC - there are cycles wasted in the IPC call due to a fairly compliacted way of copying messages from userland to kernel. Sometimes this complicated way (generic though) is used even for copying within the kernel address space, sometimes it is used for copying in case _no_ copying is necessary. The goal of this patch is to improve this a little bit. - the places where a copy is from user to kernel use the copy_msg_from_user() kernel-kernel copies are turned into assignments and BuildNotifyMessage uses the delivery buffers to avoid copying. - copy_msg_from_user() was introduced when removing the system task and is about 2/3 faster then using the current mechanism (phys_copy). It also avoids the PHYS_COPY_CATCH macro. Assignment is also faster and no copy is the fastest ;-) so perhaps there will be some hardly noticable performance gain besides the clean up.
2010-03-29 13:16:37 +02:00
dst_ptr->p_delivermsg.m_source = caller_ptr->p_endpoint;
dst_ptr->p_misc_flags |= MF_DELIVERMSG;
IPC status code for receive(). IPC changes: - receive() is changed to take an additional parameter, which is a pointer to a status code. - The status code is filled in by the kernel to provide additional information to the caller. For now, the kernel only fills in the IPC call used by the sender. Syslib changes: - sef_receive() has been split into sef_receive() (with the original semantics) and sef_receive_status() which exposes the status code to userland. - Ideally, every sys process should gradually switch to sef_receive_status() and use is_ipc_notify() as a dependable way to check for notify. - SEF has been modified to use is_ipc_notify() and demonstrate how to use the new status code.
2010-03-23 01:09:11 +01:00
call = (caller_ptr->p_misc_flags & MF_REPLY_PEND ? SENDREC
: (flags & NON_BLOCKING ? SENDNB : SEND));
Set IPC status code only for RECEIVE
2010-04-26 16:43:59 +02:00
IPC_STATUS_ADD_CALL(dst_ptr, call);
MF_REPLY_PEND should be removed when sendrec finishes
2010-06-28 10:32:49 +02:00
if (dst_ptr->p_misc_flags & MF_REPLY_PEND)
dst_ptr->p_misc_flags &= ~MF_REPLY_PEND;
All macros defining rts flags are prefixed with RTS_ - macros used with RTS_SET group of macros to define struct proc p_rts_flags are now prefixed with RTS_ to make things clear
2009-11-10 10:11:13 +01:00
RTS_UNSET(dst_ptr, RTS_RECEIVING);
Optional IPC logging
2010-06-24 15:31:40 +02:00
kernel - change print*() functions for ipc to generic ipc hook functions. - used to implement ipc stats tracking code
2011-02-08 14:54:33 +01:00
#if DEBUG_IPC_HOOK
hook_ipc_msgsend(&dst_ptr->p_delivermsg, caller_ptr, dst_ptr);
hook_ipc_msgrecv(&dst_ptr->p_delivermsg, caller_ptr, dst_ptr);
Optional IPC logging
2010-06-24 15:31:40 +02:00
#endif
Basic VM and other minor improvements. Not complete, probably not fully debugged or optimized.
2008-11-19 13:26:10 +01:00
} else {
if(flags & NON_BLOCKING) {
return(ENOTREADY);
}
Fix bug in IPC deadlock detection code. The old deadlock code was misplaced and unable to deal with asynchronous IPC primitives (notify and senda) effectively. As an example, the following sequence of messages allowed the deadlock detection code to trigger a false positive: 1. A.notify(B) 2. A.receive(B) 3. B.receive(A) 1. B.notify(A) The solution is to run the deadlock detection routine only when a process is about to block in mini_send() or mini_receive().
2009-12-17 00:32:08 +01:00
/* Check for a possible deadlock before actually blocking. */
kernel: deadlock test with endpoints instead of slot numbers, slightly cleaner
2010-07-28 16:14:06 +02:00
if (deadlock(SEND, caller_ptr, dst_e)) {
Fix bug in IPC deadlock detection code. The old deadlock code was misplaced and unable to deal with asynchronous IPC primitives (notify and senda) effectively. As an example, the following sequence of messages allowed the deadlock detection code to trigger a false positive: 1. A.notify(B) 2. A.receive(B) 3. B.receive(A) 1. B.notify(A) The solution is to run the deadlock detection routine only when a process is about to block in mini_send() or mini_receive().
2009-12-17 00:32:08 +01:00
return(ELOCKED);
}
Comment corrections from Al
2005-10-07 15:23:18 +02:00
/* Destination is not waiting. Block and dequeue caller. */
Slightly faster IPC - there are cycles wasted in the IPC call due to a fairly compliacted way of copying messages from userland to kernel. Sometimes this complicated way (generic though) is used even for copying within the kernel address space, sometimes it is used for copying in case _no_ copying is necessary. The goal of this patch is to improve this a little bit. - the places where a copy is from user to kernel use the copy_msg_from_user() kernel-kernel copies are turned into assignments and BuildNotifyMessage uses the delivery buffers to avoid copying. - copy_msg_from_user() was introduced when removing the system task and is about 2/3 faster then using the current mechanism (phys_copy). It also avoids the PHYS_COPY_CATCH macro. Assignment is also faster and no copy is the fastest ;-) so perhaps there will be some hardly noticable performance gain besides the clean up.
2010-03-29 13:16:37 +02:00
if (!(flags & FROM_KERNEL)) {
if(copy_msg_from_user(caller_ptr, m_ptr, &caller_ptr->p_sendmsg))
return EFAULT;
A reliable way for userspace to check if a msg is from kernel - IPC_FLG_MSG_FROM_KERNEL status flag is returned to userspace if the receive was satisfied by s message which was sent by the kernel on behalf of a process. This perfectly reliale information. - MF_SENDING_FROM_KERNEL flag added to processes to be able to set IPC_FLG_MSG_FROM_KERNEL when finishing receive if the receiver wasn't ready to receive immediately. - PM is changed to use this information to confirm that the scheduling messages are indeed from the kernel and not faked by a process. PM uses sef_receive_status() - get_work() is removed from PM to make the changes simpler
2010-03-29 13:25:01 +02:00
} else {
Slightly faster IPC - there are cycles wasted in the IPC call due to a fairly compliacted way of copying messages from userland to kernel. Sometimes this complicated way (generic though) is used even for copying within the kernel address space, sometimes it is used for copying in case _no_ copying is necessary. The goal of this patch is to improve this a little bit. - the places where a copy is from user to kernel use the copy_msg_from_user() kernel-kernel copies are turned into assignments and BuildNotifyMessage uses the delivery buffers to avoid copying. - copy_msg_from_user() was introduced when removing the system task and is about 2/3 faster then using the current mechanism (phys_copy). It also avoids the PHYS_COPY_CATCH macro. Assignment is also faster and no copy is the fastest ;-) so perhaps there will be some hardly noticable performance gain besides the clean up.
2010-03-29 13:16:37 +02:00
caller_ptr->p_sendmsg = *m_ptr;
A reliable way for userspace to check if a msg is from kernel - IPC_FLG_MSG_FROM_KERNEL status flag is returned to userspace if the receive was satisfied by s message which was sent by the kernel on behalf of a process. This perfectly reliale information. - MF_SENDING_FROM_KERNEL flag added to processes to be able to set IPC_FLG_MSG_FROM_KERNEL when finishing receive if the receiver wasn't ready to receive immediately. - PM is changed to use this information to confirm that the scheduling messages are indeed from the kernel and not faked by a process. PM uses sef_receive_status() - get_work() is removed from PM to make the changes simpler
2010-03-29 13:25:01 +02:00
/*
* we need to remember that this message is from kernel so we
* can set the delivery status flags when the message is
* actually delivered
*/
caller_ptr->p_misc_flags |= MF_SENDING_FROM_KERNEL;
}
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
All macros defining rts flags are prefixed with RTS_ - macros used with RTS_SET group of macros to define struct proc p_rts_flags are now prefixed with RTS_ to make things clear
2009-11-10 10:11:13 +01:00
RTS_SET(caller_ptr, RTS_SENDING);
'proc number' is process slot, 'endpoint' are generation-aware process instance numbers, encoded and decoded using macros in <minix/endpoint.h>. proc number -> endpoint migration . proc_nr in the interrupt hook is now an endpoint, proc_nr_e. . m_source for messages and notifies is now an endpoint, instead of proc number. . isokendpt() converts an endpoint to a process number, returns success (but fails if the process number is out of range, the process slot is not a living process, or the given endpoint number does not match the endpoint number in the process slot, indicating an old process). . okendpt() is the same as isokendpt(), but panic()s if the conversion fails. This is mainly used for decoding message.m_source endpoints, and other endpoint numbers in kernel data structures, which should always be correct. . if DEBUG_ENABLE_IPC_WARNINGS is enabled, isokendpt() and okendpt() get passed the __FILE__ and __LINE__ of the calling lines, and print messages about what is wrong with the endpoint number (out of range proc, empty proc, or inconsistent endpoint number), with the caller, making finding where the conversion failed easy without having to include code for every call to print where things went wrong. Sometimes this is harmless (wrong arg to a kernel call), sometimes it's a fatal internal inconsistency (bogus m_source). . some process table fields have been appended an _e to indicate it's become and endpoint. . process endpoint is stored in p_endpoint, without generation number. it turns out the kernel never needs the generation number, except when fork()ing, so it's decoded then. . kernel calls all take endpoints as arguments, not proc numbers. the one exception is sys_fork(), which needs to know in which slot to put the child.
2006-03-03 11:00:02 +01:00
caller_ptr->p_sendto_e = dst_e;
Initial revision
2005-04-21 16:53:53 +02:00
/* Process is now blocked. Put in on the destination's queue. */
mini_receive() clean up - for better readability xpp is substitued by sender - makes sure that the dequeued sender has p_q_link == NULL and that this condition holds when enqueuing the sender again. This is a sanity check to make sure that the new sender is not enqueued already - Before this change the dequeued sender's p_q_link may not be NULL and it was only set to NULL when enqueued again
2010-06-23 12:36:19 +02:00
assert(caller_ptr->p_q_link == NULL);
Fixed clear_proc(): properly release resources (IRQ hooks, notify buffers); Cleaned up comments in proc.c (old code still present for comparison);
2005-05-30 13:05:42 +02:00
xpp = &dst_ptr->p_caller_q; /* find end of list */
Removed NIL_SYS_PROC and NIL_PROC - NIL_PROC replaced by simple NULLs
2010-03-28 11:54:32 +02:00
while (*xpp) xpp = &(*xpp)->p_q_link;
Fixed clear_proc(): properly release resources (IRQ hooks, notify buffers); Cleaned up comments in proc.c (old code still present for comparison);
2005-05-30 13:05:42 +02:00
*xpp = caller_ptr; /* add caller to end */
Optional IPC logging
2010-06-24 15:31:40 +02:00
kernel - change print*() functions for ipc to generic ipc hook functions. - used to implement ipc stats tracking code
2011-02-08 14:54:33 +01:00
#if DEBUG_IPC_HOOK
hook_ipc_msgsend(&caller_ptr->p_sendmsg, caller_ptr, dst_ptr);
Optional IPC logging
2010-06-24 15:31:40 +02:00
#endif
Initial revision
2005-04-21 16:53:53 +02:00
}
return(OK);
}
/*===========================================================================*
Kernel cleanup. Reduced kernel size by 512KB by moving private buffer into cstart() function. Replaced assertions with erroneous return values. Removed assertions relating to isuserp(rp), since all processes should become user processes; system processes are now treated similar to other processes.
2005-06-07 14:34:25 +02:00
* mini_receive *
Initial revision
2005-04-21 16:53:53 +02:00
*===========================================================================*/
No linear addresses in message delivery - removes p_delivermsg_lin item from the process structure and code related to it - as the send part, the receive does not need to use the PHYS_COPY_CATCH() and umap_local() couple. - The address space of the target process is installed before delivermsg() is called. - unlike the linear address, the virtual address does not change when paging is turned on nor after fork().
2010-06-11 10:16:10 +02:00
PRIVATE int mini_receive(struct proc * caller_ptr,
Fixed a type declaration inconsistency.
2010-06-26 23:13:36 +02:00
endpoint_t src_e, /* which message source is wanted */
No linear addresses in message delivery - removes p_delivermsg_lin item from the process structure and code related to it - as the send part, the receive does not need to use the PHYS_COPY_CATCH() and umap_local() couple. - The address space of the target process is installed before delivermsg() is called. - unlike the linear address, the virtual address does not change when paging is turned on nor after fork().
2010-06-11 10:16:10 +02:00
message * m_buff_usr, /* pointer to message buffer */
const int flags)
Initial revision
2005-04-21 16:53:53 +02:00
{
Fixed clear_proc(): properly release resources (IRQ hooks, notify buffers); Cleaned up comments in proc.c (old code still present for comparison);
2005-05-30 13:05:42 +02:00
/* A process or task wants to get a message. If a message is already queued,
Initial revision
2005-04-21 16:53:53 +02:00
* acquire it and deblock the sender. If no message from the desired source
Basic VM and other minor improvements. Not complete, probably not fully debugged or optimized.
2008-11-19 13:26:10 +01:00
* is available block the caller.
Initial revision
2005-04-21 16:53:53 +02:00
*/
Optimized scheduling code. Old code is still available withing DEAD_CODE and NEW_SCHED_Q definitions. Some minor problems are being traced at the moment. This commit is meant to backup my files. --- Jorrit
2005-05-26 15:17:57 +02:00
register struct proc **xpp;
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
sys_map_t *map;
fix clang warnings -R in kernel/ and servers/
2011-06-09 16:09:13 +02:00
int r, src_id, src_proc_nr, src_p;
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
re-establish kernel assert()s. use the regular <assert.h> assert() instead of vmassert() in kernel. throw out some #if 0 code. fix a few assert() conditions. enable by default.
2010-03-10 14:00:05 +01:00
assert(!(caller_ptr->p_misc_flags & MF_DELIVERMSG));
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
/* This is where we want our message. */
No linear addresses in message delivery - removes p_delivermsg_lin item from the process structure and code related to it - as the send part, the receive does not need to use the PHYS_COPY_CATCH() and umap_local() couple. - The address space of the target process is installed before delivermsg() is called. - unlike the linear address, the virtual address does not change when paging is turned on nor after fork().
2010-06-11 10:16:10 +02:00
caller_ptr->p_delivermsg_vir = (vir_bytes) m_buff_usr;
'proc number' is process slot, 'endpoint' are generation-aware process instance numbers, encoded and decoded using macros in <minix/endpoint.h>. proc number -> endpoint migration . proc_nr in the interrupt hook is now an endpoint, proc_nr_e. . m_source for messages and notifies is now an endpoint, instead of proc number. . isokendpt() converts an endpoint to a process number, returns success (but fails if the process number is out of range, the process slot is not a living process, or the given endpoint number does not match the endpoint number in the process slot, indicating an old process). . okendpt() is the same as isokendpt(), but panic()s if the conversion fails. This is mainly used for decoding message.m_source endpoints, and other endpoint numbers in kernel data structures, which should always be correct. . if DEBUG_ENABLE_IPC_WARNINGS is enabled, isokendpt() and okendpt() get passed the __FILE__ and __LINE__ of the calling lines, and print messages about what is wrong with the endpoint number (out of range proc, empty proc, or inconsistent endpoint number), with the caller, making finding where the conversion failed easy without having to include code for every call to print where things went wrong. Sometimes this is harmless (wrong arg to a kernel call), sometimes it's a fatal internal inconsistency (bogus m_source). . some process table fields have been appended an _e to indicate it's become and endpoint. . process endpoint is stored in p_endpoint, without generation number. it turns out the kernel never needs the generation number, except when fork()ing, so it's decoded then. . kernel calls all take endpoints as arguments, not proc numbers. the one exception is sys_fork(), which needs to know in which slot to put the child.
2006-03-03 11:00:02 +01:00
if(src_e == ANY) src_p = ANY;
Allow a process to kill itself. Remove the endpoint of a process that kills itself before reporting the singal to pm.
2006-03-15 13:01:59 +01:00
else
{
okendpt(src_e, &src_p);
All macros defining rts flags are prefixed with RTS_ - macros used with RTS_SET group of macros to define struct proc p_rts_flags are now prefixed with RTS_ to make things clear
2009-11-10 10:11:13 +01:00
if (RTS_ISSET(proc_addr(src_p), RTS_NO_ENDPOINT))
Keep track of various statistics related to IPC and SYSTEM.
2008-02-22 13:36:46 +01:00
{
New P_BLOCKEDON for kernel - a macro that encodes the "who is this process waiting for" logic, which is duplicated a few times in the kernel. (For a new feature for top.) Introducing it and throwing out ESRCDIED and EDSTDIED (replaced by EDEADSRCDST - so we don't have to care which part of the blocking is failing in system.c) simplifies some code in the kernel and callers that check for E{DEADSRCDST,ESRCDIED,EDSTDIED}, but don't care about the difference, a fair bit, and more significantly doesn't duplicate the 'blocked-on' logic.
2010-03-03 16:32:26 +01:00
return EDEADSRCDST;
Keep track of various statistics related to IPC and SYSTEM.
2008-02-22 13:36:46 +01:00
}
Allow a process to kill itself. Remove the endpoint of a process that kills itself before reporting the singal to pm.
2006-03-15 13:01:59 +01:00
}
Initial revision
2005-04-21 16:53:53 +02:00
All macros defining rts flags are prefixed with RTS_ - macros used with RTS_SET group of macros to define struct proc p_rts_flags are now prefixed with RTS_ to make things clear
2009-11-10 10:11:13 +01:00
/* Check to see if a message from desired source is already available. The
* caller's RTS_SENDING flag may be set if SENDREC couldn't send. If it is
Optimized scheduling code. Old code is still available withing DEAD_CODE and NEW_SCHED_Q definitions. Some minor problems are being traced at the moment. This commit is meant to backup my files. --- Jorrit
2005-05-26 15:17:57 +02:00
* set, the process should be blocked.
*/
All macros defining rts flags are prefixed with RTS_ - macros used with RTS_SET group of macros to define struct proc p_rts_flags are now prefixed with RTS_ to make things clear
2009-11-10 10:11:13 +01:00
if (!RTS_ISSET(caller_ptr, RTS_SENDING)) {
Initial revision
2005-04-21 16:53:53 +02:00
Fixed clear_proc(): properly release resources (IRQ hooks, notify buffers); Cleaned up comments in proc.c (old code still present for comparison);
2005-05-30 13:05:42 +02:00
/* Check if there are pending notifications, except for SENDREC. */
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
if (! (caller_ptr->p_misc_flags & MF_REPLY_PEND)) {
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
map = &priv(caller_ptr)->s_notify_pending;
Kernel: fix notification delivery to non-ANY receivers
2011-01-07 18:04:43 +01:00
Use a bitmap for pending asynchronous messages instead of a global flag. That way it works similar to pending notifications.
2011-04-08 17:03:33 +02:00
/* Check for pending notifications */
if ((src_id = has_pending(map, src_p)) != NULL_PRIV_ID) {
Kernel: fix notification delivery to non-ANY receivers
2011-01-07 18:04:43 +01:00
endpoint_t hisep;
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
src_proc_nr = id_to_nr(src_id); /* get source proc */
. check for notify() from NONE (shouldn't happen any more) . test for cleared process slots when checking processes on exiting . clear process slot first, then do cleanup on exiting
2005-10-20 22:59:02 +02:00
#if DEBUG_ENABLE_IPC_WARNINGS
if(src_proc_nr == NONE) {
no more kprintf - kernel uses libsys printf now, only kputc is special to the kernel.
2010-03-03 16:45:01 +01:00
printf("mini_receive: sending notify from NONE\n");
. check for notify() from NONE (shouldn't happen any more) . test for cleared process slots when checking processes on exiting . clear process slot first, then do cleanup on exiting
2005-10-20 22:59:02 +02:00
}
#endif
Kernel: fix notification delivery to non-ANY receivers
2011-01-07 18:04:43 +01:00
unset_sys_bit(*map, src_id); /* no longer pending */
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
/* Found a suitable source, deliver the notification message. */
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
hisep = proc_addr(src_proc_nr)->p_endpoint;
re-establish kernel assert()s. use the regular <assert.h> assert() instead of vmassert() in kernel. throw out some #if 0 code. fix a few assert() conditions. enable by default.
2010-03-10 14:00:05 +01:00
assert(!(caller_ptr->p_misc_flags & MF_DELIVERMSG));
assert(src_e == ANY || hisep == src_e);
Slightly faster IPC - there are cycles wasted in the IPC call due to a fairly compliacted way of copying messages from userland to kernel. Sometimes this complicated way (generic though) is used even for copying within the kernel address space, sometimes it is used for copying in case _no_ copying is necessary. The goal of this patch is to improve this a little bit. - the places where a copy is from user to kernel use the copy_msg_from_user() kernel-kernel copies are turned into assignments and BuildNotifyMessage uses the delivery buffers to avoid copying. - copy_msg_from_user() was introduced when removing the system task and is about 2/3 faster then using the current mechanism (phys_copy). It also avoids the PHYS_COPY_CATCH macro. Assignment is also faster and no copy is the fastest ;-) so perhaps there will be some hardly noticable performance gain besides the clean up.
2010-03-29 13:16:37 +02:00
/* assemble message */
BuildNotifyMessage(&caller_ptr->p_delivermsg, src_proc_nr, caller_ptr);
caller_ptr->p_delivermsg.m_source = hisep;
caller_ptr->p_misc_flags |= MF_DELIVERMSG;
Set IPC status code only for RECEIVE
2010-04-26 16:43:59 +02:00
IPC_STATUS_ADD_CALL(caller_ptr, NOTIFY);
Slightly faster IPC - there are cycles wasted in the IPC call due to a fairly compliacted way of copying messages from userland to kernel. Sometimes this complicated way (generic though) is used even for copying within the kernel address space, sometimes it is used for copying in case _no_ copying is necessary. The goal of this patch is to improve this a little bit. - the places where a copy is from user to kernel use the copy_msg_from_user() kernel-kernel copies are turned into assignments and BuildNotifyMessage uses the delivery buffers to avoid copying. - copy_msg_from_user() was introduced when removing the system task and is about 2/3 faster then using the current mechanism (phys_copy). It also avoids the PHYS_COPY_CATCH macro. Assignment is also faster and no copy is the fastest ;-) so perhaps there will be some hardly noticable performance gain besides the clean up.
2010-03-29 13:16:37 +02:00
MF_REPLY_PEND should be removed when sendrec finishes
2010-06-28 10:32:49 +02:00
goto receive_done;
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
}
New NOTIFY trap (IPC call) to send queued notification messages. The call works. Permission check, restriction of outstanding notifications to be added. Low level code to make it work from within interrupt handlers will be added as well.
2005-05-19 16:05:51 +02:00
}
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
Use a bitmap for pending asynchronous messages instead of a global flag. That way it works similar to pending notifications.
2011-04-08 17:03:33 +02:00
/* Check for pending asynchronous messages */
map = &priv(caller_ptr)->s_asyn_pending;
if (has_pending(map, src_p) != NULL_PRIV_ID) {
if (src_p != ANY)
r = try_one(proc_addr(src_p), caller_ptr);
else
r = try_async(caller_ptr);
Preserve the order of IPC messages between two parties. Currently a sequence of messages between a sender A and a receiver B of the form: A.asynsend(M1, B); A.send(M2, B) may result in the receiver receiving M1 first and then M2 or viceversa. This patch makes sure that the original order M1, M2 is always preserved. Note that the order of a hypotetical sequence A.asynsend(M1, B); A.asynsend(M2, B) is already guaranteed by the implementation of asynsend by design. Other senda-based wrappers can define their own semantics.
2010-03-27 01:09:22 +01:00
if (r == OK) {
Use a bitmap for pending asynchronous messages instead of a global flag. That way it works similar to pending notifications.
2011-04-08 17:03:33 +02:00
IPC_STATUS_ADD_CALL(caller_ptr, SENDA);
goto receive_done;
}
Preserve the order of IPC messages between two parties. Currently a sequence of messages between a sender A and a receiver B of the form: A.asynsend(M1, B); A.send(M2, B) may result in the receiver receiving M1 first and then M2 or viceversa. This patch makes sure that the original order M1, M2 is always preserved. Note that the order of a hypotetical sequence A.asynsend(M1, B); A.asynsend(M2, B) is already guaranteed by the implementation of asynsend by design. Other senda-based wrappers can define their own semantics.
2010-03-27 01:09:22 +01:00
}
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
/* Check caller queue. Use pointer pointers to keep code simple. */
xpp = &caller_ptr->p_caller_q;
Removed NIL_SYS_PROC and NIL_PROC - NIL_PROC replaced by simple NULLs
2010-03-28 11:54:32 +02:00
while (*xpp) {
mini_receive() clean up - for better readability xpp is substitued by sender - makes sure that the dequeued sender has p_q_link == NULL and that this condition holds when enqueuing the sender again. This is a sanity check to make sure that the new sender is not enqueued already - Before this change the dequeued sender's p_q_link may not be NULL and it was only set to NULL when enqueued again
2010-06-23 12:36:19 +02:00
struct proc * sender = *xpp;
if (src_e == ANY || src_p == proc_nr(sender)) {
IPC status code for receive(). IPC changes: - receive() is changed to take an additional parameter, which is a pointer to a status code. - The status code is filled in by the kernel to provide additional information to the caller. For now, the kernel only fills in the IPC call used by the sender. Syslib changes: - sef_receive() has been split into sef_receive() (with the original semantics) and sef_receive_status() which exposes the status code to userland. - Ideally, every sys process should gradually switch to sef_receive_status() and use is_ipc_notify() as a dependable way to check for notify. - SEF has been modified to use is_ipc_notify() and demonstrate how to use the new status code.
2010-03-23 01:09:11 +01:00
int call;
mini_receive() clean up - for better readability xpp is substitued by sender - makes sure that the dequeued sender has p_q_link == NULL and that this condition holds when enqueuing the sender again. This is a sanity check to make sure that the new sender is not enqueued already - Before this change the dequeued sender's p_q_link may not be NULL and it was only set to NULL when enqueued again
2010-06-23 12:36:19 +02:00
assert(!RTS_ISSET(sender, RTS_SLOT_FREE));
assert(!RTS_ISSET(sender, RTS_NO_ENDPOINT));
Added some disabled debug code.
2006-03-08 13:06:33 +01:00
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
/* Found acceptable message. Copy it and update status. */
re-establish kernel assert()s. use the regular <assert.h> assert() instead of vmassert() in kernel. throw out some #if 0 code. fix a few assert() conditions. enable by default.
2010-03-10 14:00:05 +01:00
assert(!(caller_ptr->p_misc_flags & MF_DELIVERMSG));
mini_receive() clean up - for better readability xpp is substitued by sender - makes sure that the dequeued sender has p_q_link == NULL and that this condition holds when enqueuing the sender again. This is a sanity check to make sure that the new sender is not enqueued already - Before this change the dequeued sender's p_q_link may not be NULL and it was only set to NULL when enqueued again
2010-06-23 12:36:19 +02:00
caller_ptr->p_delivermsg = sender->p_sendmsg;
caller_ptr->p_delivermsg.m_source = sender->p_endpoint;
Slightly faster IPC - there are cycles wasted in the IPC call due to a fairly compliacted way of copying messages from userland to kernel. Sometimes this complicated way (generic though) is used even for copying within the kernel address space, sometimes it is used for copying in case _no_ copying is necessary. The goal of this patch is to improve this a little bit. - the places where a copy is from user to kernel use the copy_msg_from_user() kernel-kernel copies are turned into assignments and BuildNotifyMessage uses the delivery buffers to avoid copying. - copy_msg_from_user() was introduced when removing the system task and is about 2/3 faster then using the current mechanism (phys_copy). It also avoids the PHYS_COPY_CATCH macro. Assignment is also faster and no copy is the fastest ;-) so perhaps there will be some hardly noticable performance gain besides the clean up.
2010-03-29 13:16:37 +02:00
caller_ptr->p_misc_flags |= MF_DELIVERMSG;
mini_receive() clean up - for better readability xpp is substitued by sender - makes sure that the dequeued sender has p_q_link == NULL and that this condition holds when enqueuing the sender again. This is a sanity check to make sure that the new sender is not enqueued already - Before this change the dequeued sender's p_q_link may not be NULL and it was only set to NULL when enqueued again
2010-06-23 12:36:19 +02:00
RTS_UNSET(sender, RTS_SENDING);
Slightly faster IPC - there are cycles wasted in the IPC call due to a fairly compliacted way of copying messages from userland to kernel. Sometimes this complicated way (generic though) is used even for copying within the kernel address space, sometimes it is used for copying in case _no_ copying is necessary. The goal of this patch is to improve this a little bit. - the places where a copy is from user to kernel use the copy_msg_from_user() kernel-kernel copies are turned into assignments and BuildNotifyMessage uses the delivery buffers to avoid copying. - copy_msg_from_user() was introduced when removing the system task and is about 2/3 faster then using the current mechanism (phys_copy). It also avoids the PHYS_COPY_CATCH macro. Assignment is also faster and no copy is the fastest ;-) so perhaps there will be some hardly noticable performance gain besides the clean up.
2010-03-29 13:16:37 +02:00
mini_receive() clean up - for better readability xpp is substitued by sender - makes sure that the dequeued sender has p_q_link == NULL and that this condition holds when enqueuing the sender again. This is a sanity check to make sure that the new sender is not enqueued already - Before this change the dequeued sender's p_q_link may not be NULL and it was only set to NULL when enqueued again
2010-06-23 12:36:19 +02:00
call = (sender->p_misc_flags & MF_REPLY_PEND ? SENDREC : SEND);
Set IPC status code only for RECEIVE
2010-04-26 16:43:59 +02:00
IPC_STATUS_ADD_CALL(caller_ptr, call);
A reliable way for userspace to check if a msg is from kernel - IPC_FLG_MSG_FROM_KERNEL status flag is returned to userspace if the receive was satisfied by s message which was sent by the kernel on behalf of a process. This perfectly reliale information. - MF_SENDING_FROM_KERNEL flag added to processes to be able to set IPC_FLG_MSG_FROM_KERNEL when finishing receive if the receiver wasn't ready to receive immediately. - PM is changed to use this information to confirm that the scheduling messages are indeed from the kernel and not faked by a process. PM uses sef_receive_status() - get_work() is removed from PM to make the changes simpler
2010-03-29 13:25:01 +02:00
/*
* if the message is originaly from the kernel on behalf of this
* process, we must send the status flags accordingly
*/
mini_receive() clean up - for better readability xpp is substitued by sender - makes sure that the dequeued sender has p_q_link == NULL and that this condition holds when enqueuing the sender again. This is a sanity check to make sure that the new sender is not enqueued already - Before this change the dequeued sender's p_q_link may not be NULL and it was only set to NULL when enqueued again
2010-06-23 12:36:19 +02:00
if (sender->p_misc_flags & MF_SENDING_FROM_KERNEL) {
Set IPC status code only for RECEIVE
2010-04-26 16:43:59 +02:00
IPC_STATUS_ADD_FLAGS(caller_ptr, IPC_FLG_MSG_FROM_KERNEL);
A reliable way for userspace to check if a msg is from kernel - IPC_FLG_MSG_FROM_KERNEL status flag is returned to userspace if the receive was satisfied by s message which was sent by the kernel on behalf of a process. This perfectly reliale information. - MF_SENDING_FROM_KERNEL flag added to processes to be able to set IPC_FLG_MSG_FROM_KERNEL when finishing receive if the receiver wasn't ready to receive immediately. - PM is changed to use this information to confirm that the scheduling messages are indeed from the kernel and not faked by a process. PM uses sef_receive_status() - get_work() is removed from PM to make the changes simpler
2010-03-29 13:25:01 +02:00
/* we can clean the flag now, not need anymore */
mini_receive() clean up - for better readability xpp is substitued by sender - makes sure that the dequeued sender has p_q_link == NULL and that this condition holds when enqueuing the sender again. This is a sanity check to make sure that the new sender is not enqueued already - Before this change the dequeued sender's p_q_link may not be NULL and it was only set to NULL when enqueued again
2010-06-23 12:36:19 +02:00
sender->p_misc_flags &= ~MF_SENDING_FROM_KERNEL;
A reliable way for userspace to check if a msg is from kernel - IPC_FLG_MSG_FROM_KERNEL status flag is returned to userspace if the receive was satisfied by s message which was sent by the kernel on behalf of a process. This perfectly reliale information. - MF_SENDING_FROM_KERNEL flag added to processes to be able to set IPC_FLG_MSG_FROM_KERNEL when finishing receive if the receiver wasn't ready to receive immediately. - PM is changed to use this information to confirm that the scheduling messages are indeed from the kernel and not faked by a process. PM uses sef_receive_status() - get_work() is removed from PM to make the changes simpler
2010-03-29 13:25:01 +02:00
}
mini_receive() clean up - for better readability xpp is substitued by sender - makes sure that the dequeued sender has p_q_link == NULL and that this condition holds when enqueuing the sender again. This is a sanity check to make sure that the new sender is not enqueued already - Before this change the dequeued sender's p_q_link may not be NULL and it was only set to NULL when enqueued again
2010-06-23 12:36:19 +02:00
if (sender->p_misc_flags & MF_SIG_DELAY)
sig_delay_done(sender);
Slightly faster IPC - there are cycles wasted in the IPC call due to a fairly compliacted way of copying messages from userland to kernel. Sometimes this complicated way (generic though) is used even for copying within the kernel address space, sometimes it is used for copying in case _no_ copying is necessary. The goal of this patch is to improve this a little bit. - the places where a copy is from user to kernel use the copy_msg_from_user() kernel-kernel copies are turned into assignments and BuildNotifyMessage uses the delivery buffers to avoid copying. - copy_msg_from_user() was introduced when removing the system task and is about 2/3 faster then using the current mechanism (phys_copy). It also avoids the PHYS_COPY_CATCH macro. Assignment is also faster and no copy is the fastest ;-) so perhaps there will be some hardly noticable performance gain besides the clean up.
2010-03-29 13:16:37 +02:00
kernel - change print*() functions for ipc to generic ipc hook functions. - used to implement ipc stats tracking code
2011-02-08 14:54:33 +01:00
#if DEBUG_IPC_HOOK
hook_ipc_msgrecv(&caller_ptr->p_delivermsg, *xpp, caller_ptr);
Optional IPC logging
2010-06-24 15:31:40 +02:00
#endif
mini_receive() clean up - for better readability xpp is substitued by sender - makes sure that the dequeued sender has p_q_link == NULL and that this condition holds when enqueuing the sender again. This is a sanity check to make sure that the new sender is not enqueued already - Before this change the dequeued sender's p_q_link may not be NULL and it was only set to NULL when enqueued again
2010-06-23 12:36:19 +02:00
*xpp = sender->p_q_link; /* remove from queue */
sender->p_q_link = NULL;
MF_REPLY_PEND should be removed when sendrec finishes
2010-06-28 10:32:49 +02:00
goto receive_done;
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
}
mini_receive() clean up - for better readability xpp is substitued by sender - makes sure that the dequeued sender has p_q_link == NULL and that this condition holds when enqueuing the sender again. This is a sanity check to make sure that the new sender is not enqueued already - Before this change the dequeued sender's p_q_link may not be NULL and it was only set to NULL when enqueued again
2010-06-23 12:36:19 +02:00
xpp = &sender->p_q_link; /* proceed to next */
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
}
Initial revision
2005-04-21 16:53:53 +02:00
}
Improved NOTIFY system: fixed a minor error, ignore pending notifications on SENDREC system calls. To be done: resource (buffer pool) management; make it structurally impossible to run out of buffers.
2005-05-27 14:44:14 +02:00
/* No suitable message is available or the caller couldn't send in SENDREC.
Optimized scheduling code. Old code is still available withing DEAD_CODE and NEW_SCHED_Q definitions. Some minor problems are being traced at the moment. This commit is meant to backup my files. --- Jorrit
2005-05-26 15:17:57 +02:00
* Block the process trying to receive, unless the flags tell otherwise.
Initial revision
2005-04-21 16:53:53 +02:00
*/
Improved NOTIFY system: fixed a minor error, ignore pending notifications on SENDREC system calls. To be done: resource (buffer pool) management; make it structurally impossible to run out of buffers.
2005-05-27 14:44:14 +02:00
if ( ! (flags & NON_BLOCKING)) {
Fix bug in IPC deadlock detection code. The old deadlock code was misplaced and unable to deal with asynchronous IPC primitives (notify and senda) effectively. As an example, the following sequence of messages allowed the deadlock detection code to trigger a false positive: 1. A.notify(B) 2. A.receive(B) 3. B.receive(A) 1. B.notify(A) The solution is to run the deadlock detection routine only when a process is about to block in mini_send() or mini_receive().
2009-12-17 00:32:08 +01:00
/* Check for a possible deadlock before actually blocking. */
kernel: deadlock test with endpoints instead of slot numbers, slightly cleaner
2010-07-28 16:14:06 +02:00
if (deadlock(RECEIVE, caller_ptr, src_e)) {
Fix bug in IPC deadlock detection code. The old deadlock code was misplaced and unable to deal with asynchronous IPC primitives (notify and senda) effectively. As an example, the following sequence of messages allowed the deadlock detection code to trigger a false positive: 1. A.notify(B) 2. A.receive(B) 3. B.receive(A) 1. B.notify(A) The solution is to run the deadlock detection routine only when a process is about to block in mini_send() or mini_receive().
2009-12-17 00:32:08 +01:00
return(ELOCKED);
}
'proc number' is process slot, 'endpoint' are generation-aware process instance numbers, encoded and decoded using macros in <minix/endpoint.h>. proc number -> endpoint migration . proc_nr in the interrupt hook is now an endpoint, proc_nr_e. . m_source for messages and notifies is now an endpoint, instead of proc number. . isokendpt() converts an endpoint to a process number, returns success (but fails if the process number is out of range, the process slot is not a living process, or the given endpoint number does not match the endpoint number in the process slot, indicating an old process). . okendpt() is the same as isokendpt(), but panic()s if the conversion fails. This is mainly used for decoding message.m_source endpoints, and other endpoint numbers in kernel data structures, which should always be correct. . if DEBUG_ENABLE_IPC_WARNINGS is enabled, isokendpt() and okendpt() get passed the __FILE__ and __LINE__ of the calling lines, and print messages about what is wrong with the endpoint number (out of range proc, empty proc, or inconsistent endpoint number), with the caller, making finding where the conversion failed easy without having to include code for every call to print where things went wrong. Sometimes this is harmless (wrong arg to a kernel call), sometimes it's a fatal internal inconsistency (bogus m_source). . some process table fields have been appended an _e to indicate it's become and endpoint. . process endpoint is stored in p_endpoint, without generation number. it turns out the kernel never needs the generation number, except when fork()ing, so it's decoded then. . kernel calls all take endpoints as arguments, not proc numbers. the one exception is sys_fork(), which needs to know in which slot to put the child.
2006-03-03 11:00:02 +01:00
caller_ptr->p_getfrom_e = src_e;
All macros defining rts flags are prefixed with RTS_ - macros used with RTS_SET group of macros to define struct proc p_rts_flags are now prefixed with RTS_ to make things clear
2009-11-10 10:11:13 +01:00
RTS_SET(caller_ptr, RTS_RECEIVING);
Initial revision
2005-04-21 16:53:53 +02:00
return(OK);
} else {
Keep track of various statistics related to IPC and SYSTEM.
2008-02-22 13:36:46 +01:00
return(ENOTREADY);
Initial revision
2005-04-21 16:53:53 +02:00
}
MF_REPLY_PEND should be removed when sendrec finishes
2010-06-28 10:32:49 +02:00
receive_done:
if (caller_ptr->p_misc_flags & MF_REPLY_PEND)
caller_ptr->p_misc_flags &= ~MF_REPLY_PEND;
return OK;
Initial revision
2005-04-21 16:53:53 +02:00
}
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
/*===========================================================================*
AT driver is not modified (debugging only); TTY: select and revive with new notify and FS call back; kernel: removed old notify code; removed ugly prepare_shutdown timer kputc: don't send to FS if PRINTF_PROC fails
2005-07-27 16:32:16 +02:00
* mini_notify *
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
*===========================================================================*/
Lots of const correctness.
2010-03-27 15:31:00 +01:00
PUBLIC int mini_notify(
const struct proc *caller_ptr, /* sender of the notification */
endpoint_t dst_e /* which process to notify */
)
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
{
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
register struct proc *dst_ptr;
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
int src_id; /* source id for late delivery */
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
int dst_p;
if (!isokendpt(dst_e, &dst_p)) {
util_stacktrace();
no more kprintf - kernel uses libsys printf now, only kputc is special to the kernel.
2010-03-03 16:45:01 +01:00
printf("mini_notify: bogus endpoint %d\n", dst_e);
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
return EDEADSRCDST;
}
dst_ptr = proc_addr(dst_p);
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
/* Check to see if target is blocked waiting for this message. A process
* can be both sending and receiving during a SENDREC system call.
*/
minor cleanup, extra check
2009-01-20 16:47:00 +01:00
if (WILLRECEIVE(dst_ptr, caller_ptr->p_endpoint) &&
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
! (dst_ptr->p_misc_flags & MF_REPLY_PEND)) {
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
/* Destination is indeed waiting for a message. Assemble a notification
* message and deliver it. Copy from pseudo-source HARDWARE, since the
* message is in the kernel's address space.
*/
re-establish kernel assert()s. use the regular <assert.h> assert() instead of vmassert() in kernel. throw out some #if 0 code. fix a few assert() conditions. enable by default.
2010-03-10 14:00:05 +01:00
assert(!(dst_ptr->p_misc_flags & MF_DELIVERMSG));
Slightly faster IPC - there are cycles wasted in the IPC call due to a fairly compliacted way of copying messages from userland to kernel. Sometimes this complicated way (generic though) is used even for copying within the kernel address space, sometimes it is used for copying in case _no_ copying is necessary. The goal of this patch is to improve this a little bit. - the places where a copy is from user to kernel use the copy_msg_from_user() kernel-kernel copies are turned into assignments and BuildNotifyMessage uses the delivery buffers to avoid copying. - copy_msg_from_user() was introduced when removing the system task and is about 2/3 faster then using the current mechanism (phys_copy). It also avoids the PHYS_COPY_CATCH macro. Assignment is also faster and no copy is the fastest ;-) so perhaps there will be some hardly noticable performance gain besides the clean up.
2010-03-29 13:16:37 +02:00
BuildNotifyMessage(&dst_ptr->p_delivermsg, proc_nr(caller_ptr), dst_ptr);
dst_ptr->p_delivermsg.m_source = caller_ptr->p_endpoint;
dst_ptr->p_misc_flags |= MF_DELIVERMSG;
Set IPC status code only for RECEIVE
2010-04-26 16:43:59 +02:00
IPC_STATUS_ADD_CALL(dst_ptr, NOTIFY);
All macros defining rts flags are prefixed with RTS_ - macros used with RTS_SET group of macros to define struct proc p_rts_flags are now prefixed with RTS_ to make things clear
2009-11-10 10:11:13 +01:00
RTS_UNSET(dst_ptr, RTS_RECEIVING);
Slightly faster IPC - there are cycles wasted in the IPC call due to a fairly compliacted way of copying messages from userland to kernel. Sometimes this complicated way (generic though) is used even for copying within the kernel address space, sometimes it is used for copying in case _no_ copying is necessary. The goal of this patch is to improve this a little bit. - the places where a copy is from user to kernel use the copy_msg_from_user() kernel-kernel copies are turned into assignments and BuildNotifyMessage uses the delivery buffers to avoid copying. - copy_msg_from_user() was introduced when removing the system task and is about 2/3 faster then using the current mechanism (phys_copy). It also avoids the PHYS_COPY_CATCH macro. Assignment is also faster and no copy is the fastest ;-) so perhaps there will be some hardly noticable performance gain besides the clean up.
2010-03-29 13:16:37 +02:00
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
return(OK);
}
/* Destination is not ready to receive the notification. Add it to the
Rewrite of boot process KERNEL CHANGES: - The kernel only knows about privileges of kernel tasks and the root system process (now RS). - Kernel tasks and the root system process are the only processes that are made schedulable by the kernel at startup. All the other processes in the boot image don't get their privileges set at startup and are inhibited from running by the RTS_NO_PRIV flag. - Removed the assumption on the ordering of processes in the boot image table. System processes can now appear in any order in the boot image table. - Privilege ids can now be assigned both statically or dynamically. The kernel assigns static privilege ids to kernel tasks and the root system process. Each id is directly derived from the process number. - User processes now all share the static privilege id of the root user process (now INIT). - sys_privctl split: we have more calls now to let RS set privileges for system processes. SYS_PRIV_ALLOW / SYS_PRIV_DISALLOW are only used to flip the RTS_NO_PRIV flag and allow / disallow a process from running. SYS_PRIV_SET_SYS / SYS_PRIV_SET_USER are used to set privileges for a system / user process. - boot image table flags split: PROC_FULLVM is the only flag that has been moved out of the privilege flags and is still maintained in the boot image table. All the other privilege flags are out of the kernel now. RS CHANGES: - RS is the only user-space process who gets to run right after in-kernel startup. - RS uses the boot image table from the kernel and three additional boot image info table (priv table, sys table, dev table) to complete the initialization of the system. - RS checks that the entries in the priv table match the entries in the boot image table to make sure that every process in the boot image gets schedulable. - RS only uses static privilege ids to set privileges for system services in the boot image. - RS includes basic memory management support to allocate the boot image buffer dynamically during initialization. The buffer shall contain the executable image of all the system services we would like to restart after a crash. - First step towards decoupling between resource provisioning and resource requirements in RS: RS must know what resources it needs to restart a process and what resources it has currently available. This is useful to tradeoff reliability and resource consumption. When required resources are missing, the process cannot be restarted. In that case, in the future, a system flag will tell RS what to do. For example, if CORE_PROC is set, RS should trigger a system-wide panic because the system can no longer function correctly without a core system process. PM CHANGES: - The process tree built at initialization time is changed to have INIT as root with pid 0, RS child of INIT and all the system services children of RS. This is required to make RS in control of all the system services. - PM no longer registers labels for system services in the boot image. This is now part of RS's initialization process.
2009-12-11 01:08:19 +01:00
* bit map with pending notifications. Note the indirectness: the privilege id
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
* instead of the process number is used in the pending bit map.
*/
src_id = priv(caller_ptr)->s_id;
set_sys_bit(priv(dst_ptr)->s_notify_pending, src_id);
return(OK);
}
Basic VM and other minor improvements. Not complete, probably not fully debugged or optimized.
2008-11-19 13:26:10 +01:00
#define ASCOMPLAIN(caller, entry, field) \
no more kprintf - kernel uses libsys printf now, only kputc is special to the kernel.
2010-03-03 16:45:01 +01:00
printf("kernel:%s:%d: asyn failed for %s in %s " \
Basic VM and other minor improvements. Not complete, probably not fully debugged or optimized.
2008-11-19 13:26:10 +01:00
"(%d/%d, tab 0x%lx)\n",__FILE__,__LINE__, \
field, caller->p_name, entry, priv(caller)->s_asynsize, priv(caller)->s_asyntab)
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
#define A_RETR_FLD(entry, field) \
Basic VM and other minor improvements. Not complete, probably not fully debugged or optimized.
2008-11-19 13:26:10 +01:00
if(data_copy(caller_ptr->p_endpoint, \
table_v + (entry)*sizeof(asynmsg_t) + offsetof(struct asynmsg,field),\
Removal of the system task * Userspace change to use the new kernel calls - _taskcall(SYSTASK...) changed to _kernel_call(...) - int 32 reused for the kernel calls - _do_kernel_call() to make the trap to kernel - kernel_call() to make the actuall kernel call from C using _do_kernel_call() - unlike ipc call the kernel call always succeeds as kernel is always available, however, kernel may return an error * Kernel side implementation of kernel calls - the SYSTEm task does not run, only the proc table entry is preserved - every data_copy(SYSTEM is no data_copy(KERNEL - "locking" is an empty operation now as everything runs in kernel - sys_task() is replaced by kernel_call() which copies the message into kernel, dispatches the call to its handler and finishes by either copying the results back to userspace (if need be) or by suspending the process because of VM - suspended processes are later made runnable once the memory issue is resolved, picked up by the scheduler and only at this time the call is resumed (in fact restarted) which does not need to copy the message from userspace as the message is already saved in the process structure. - no ned for the vmrestart queue, the scheduler will restart the system calls - no special case in do_vmctl(), all requests remove the RTS_VMREQUEST flag
2010-02-09 16:20:09 +01:00
KERNEL, (vir_bytes) &tabent.field, \
Basic VM and other minor improvements. Not complete, probably not fully debugged or optimized.
2008-11-19 13:26:10 +01:00
sizeof(tabent.field)) != OK) {\
ASCOMPLAIN(caller_ptr, entry, #field); \
return EFAULT; \
}
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
#define A_RETR(entry) do { \
if (data_copy( \
caller_ptr->p_endpoint, table_v + (entry)*sizeof(asynmsg_t),\
KERNEL, (vir_bytes) &tabent, \
sizeof(tabent)) != OK) { \
ASCOMPLAIN(caller_ptr, entry, "message entry"); \
return(EFAULT); \
} \
} while(0)
#define A_INSRT_FLD(entry, field) \
Removal of the system task * Userspace change to use the new kernel calls - _taskcall(SYSTASK...) changed to _kernel_call(...) - int 32 reused for the kernel calls - _do_kernel_call() to make the trap to kernel - kernel_call() to make the actuall kernel call from C using _do_kernel_call() - unlike ipc call the kernel call always succeeds as kernel is always available, however, kernel may return an error * Kernel side implementation of kernel calls - the SYSTEm task does not run, only the proc table entry is preserved - every data_copy(SYSTEM is no data_copy(KERNEL - "locking" is an empty operation now as everything runs in kernel - sys_task() is replaced by kernel_call() which copies the message into kernel, dispatches the call to its handler and finishes by either copying the results back to userspace (if need be) or by suspending the process because of VM - suspended processes are later made runnable once the memory issue is resolved, picked up by the scheduler and only at this time the call is resumed (in fact restarted) which does not need to copy the message from userspace as the message is already saved in the process structure. - no ned for the vmrestart queue, the scheduler will restart the system calls - no special case in do_vmctl(), all requests remove the RTS_VMREQUEST flag
2010-02-09 16:20:09 +01:00
if(data_copy(KERNEL, (vir_bytes) &tabent.field, \
Basic VM and other minor improvements. Not complete, probably not fully debugged or optimized.
2008-11-19 13:26:10 +01:00
caller_ptr->p_endpoint, \
table_v + (entry)*sizeof(asynmsg_t) + offsetof(struct asynmsg,field),\
sizeof(tabent.field)) != OK) {\
ASCOMPLAIN(caller_ptr, entry, #field); \
return EFAULT; \
}
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
#define A_INSRT(entry) do { \
if (data_copy(KERNEL, (vir_bytes) &tabent, \
caller_ptr->p_endpoint, table_v + (entry)*sizeof(asynmsg_t),\
sizeof(tabent)) != OK) { \
ASCOMPLAIN(caller_ptr, entry, "message entry"); \
return(EFAULT); \
} \
} while(0)
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
/*===========================================================================*
* mini_senda *
*===========================================================================*/
Remove obsolete m_ptr calculations in try_one() and mini_senda().
2010-01-14 13:04:24 +01:00
PRIVATE int mini_senda(struct proc *caller_ptr, asynmsg_t *table, size_t size)
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
{
Use a bitmap for pending asynchronous messages instead of a global flag. That way it works similar to pending notifications.
2011-04-08 17:03:33 +02:00
int r = OK, dst_p, done, do_notify;
unsigned int i;
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
unsigned flags;
endpoint_t dst;
struct proc *dst_ptr;
struct priv *privp;
asynmsg_t tabent;
const vir_bytes table_v = (vir_bytes) table;
privp = priv(caller_ptr);
if (!(privp->s_flags & SYS_PROC)) {
printf( "mini_senda: warning caller has no privilege structure\n");
return(EPERM);
}
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
/* Clear table */
privp->s_asyntab = -1;
privp->s_asynsize = 0;
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
if (size == 0) return(OK); /* Nothing to do, just return */
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
/* Limit size to something reasonable. An arbitrary choice is 16
* times the number of process table entries.
*
* (this check has been duplicated in sys_call but is left here
* as a sanity check)
*/
if (size > 16*(NR_TASKS + NR_PROCS)) return(EDOM);
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
/* Scan the table */
Enable sending a notification when sending of an asynchronous message was completed (successfully or not). AMF_NOTIFY_ERR can be used if the sender only wishes to be notified in case of an error (e.g., EDEADSRCDST). A new endpoint ASYNCM will be the sender of the notification.
2011-04-08 17:14:48 +02:00
do_notify = FALSE;
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
done = TRUE;
for (i = 0; i < size; i++) {
/* Process each entry in the table and store the result in the table.
* If we're done handling a message, copy the result to the sender. */
int pending_recv = FALSE;
/* Copy message to kernel */
A_RETR(i);
flags = tabent.flags;
dst = tabent.dst;
if (flags == 0) continue; /* Skip empty entries */
/* 'flags' field must contain only valid bits */
Enable sending a notification when sending of an asynchronous message was completed (successfully or not). AMF_NOTIFY_ERR can be used if the sender only wishes to be notified in case of an error (e.g., EDEADSRCDST). A new endpoint ASYNCM will be the sender of the notification.
2011-04-08 17:14:48 +02:00
if(flags & ~(AMF_VALID|AMF_DONE|AMF_NOTIFY|AMF_NOREPLY|AMF_NOTIFY_ERR))
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
return(EINVAL);
if (!(flags & AMF_VALID)) return(EINVAL); /* Must contain message */
if (flags & AMF_DONE) continue; /* Already done processing */
if (!isokendpt(tabent.dst, &dst_p))
r = EDEADSRCDST; /* Bad destination, report the error */
else if (iskerneln(dst_p))
r = ECALLDENIED; /* Asyn sends to the kernel are not allowed */
else if (!may_send_to(caller_ptr, dst_p))
r = ECALLDENIED; /* Send denied by IPC mask */
else /* r == OK */
dst_ptr = proc_addr(dst_p);
Kernel: fix faulty trap check
2009-11-28 14:15:07 +01:00
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
/* XXX: RTS_NO_ENDPOINT should be removed */
if (r == OK && RTS_ISSET(dst_ptr, RTS_NO_ENDPOINT)) {
r = EDEADSRCDST;
}
IPC privileges fixes Kernel: o Remove s_ipc_sendrec, instead using s_ipc_to for all send primitives o Centralize s_ipc_to bit manipulation, - disallowing assignment of bits pointing to unused priv structs; - preventing send-to-self by not setting bit for own priv struct; - preserving send mask matrix symmetry in all cases o Add IPC send mask checks to SENDA, which were missing entirely somehow o Slightly improve IPC stats accounting for SENDA o Remove SYSTEM from user processes' send mask o Half-fix the dependency between boot image order and process numbers, - correcting the table order of the boot processes; - documenting the order requirement needed for proper send masks; - warning at boot time if the order is violated RS: o Add support in /etc/drivers.conf for servers that talk to user processes, - disallowing IPC to user processes if no "ipc" field is present - adding a special "USER" label to explicitly allow IPC to user processes o Always apply IPC masks when specified; remove -i flag from service(8) o Use kernel send mask symmetry to delay adding IPC permissions for labels that do not exist yet, adding them to that label's process upon creation o Add VM to ipc permissions list for rtl8139 and fxp in drivers.conf Left to future fixes: o Removal of the table order vs process numbers dependency altogether, possibly using per-process send list structures as used for SYSTEM calls o Proper assignment of send masks to boot processes; some of the assigned (~0) masks are much wider than necessary o Proper assignment of IPC send masks for many more servers in drivers.conf o Removal of the debugging warning about the now legitimate case where RS's add_forward_ipc cannot find the IPC destination's label yet
2009-07-02 18:25:31 +02:00
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
/* Check if 'dst' is blocked waiting for this message.
* If AMF_NOREPLY is set, do not satisfy the receiving part of
* a SENDREC.
*/
if (r == OK && WILLRECEIVE(dst_ptr, caller_ptr->p_endpoint) &&
(!(flags&AMF_NOREPLY) || !(dst_ptr->p_misc_flags&MF_REPLY_PEND))) {
/* Destination is indeed waiting for this message. */
dst_ptr->p_delivermsg = tabent.msg;
dst_ptr->p_delivermsg.m_source = caller_ptr->p_endpoint;
dst_ptr->p_misc_flags |= MF_DELIVERMSG;
IPC_STATUS_ADD_CALL(dst_ptr, SENDA);
RTS_UNSET(dst_ptr, RTS_RECEIVING);
} else if (r == OK) {
Use a bitmap for pending asynchronous messages instead of a global flag. That way it works similar to pending notifications.
2011-04-08 17:03:33 +02:00
/* Inform receiver that something is pending */
set_sys_bit(priv(dst_ptr)->s_asyn_pending,
priv(caller_ptr)->s_id);
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
pending_recv = TRUE;
}
if (!pending_recv) flags |= AMF_DONE; /* Done handling message */
/* Store results */
tabent.result = r;
tabent.flags = flags;
if (flags & AMF_DONE) {
Enable sending a notification when sending of an asynchronous message was completed (successfully or not). AMF_NOTIFY_ERR can be used if the sender only wishes to be notified in case of an error (e.g., EDEADSRCDST). A new endpoint ASYNCM will be the sender of the notification.
2011-04-08 17:14:48 +02:00
if (flags & AMF_NOTIFY)
do_notify = TRUE;
else if (r != OK && (flags & AMF_NOTIFY_ERR))
do_notify = TRUE;
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
A_INSRT(i); /* Copy results to caller */
} else
done = FALSE;
}
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
Enable sending a notification when sending of an asynchronous message was completed (successfully or not). AMF_NOTIFY_ERR can be used if the sender only wishes to be notified in case of an error (e.g., EDEADSRCDST). A new endpoint ASYNCM will be the sender of the notification.
2011-04-08 17:14:48 +02:00
if (do_notify)
mini_notify(proc_addr(ASYNCM), caller_ptr->p_endpoint);
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
if (!done) {
privp->s_asyntab = (vir_bytes) table;
privp->s_asynsize = size;
}
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
return(OK);
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
}
/*===========================================================================*
* try_async *
*===========================================================================*/
PRIVATE int try_async(caller_ptr)
struct proc *caller_ptr;
{
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
int r;
struct priv *privp;
struct proc *src_ptr;
Kernel: try_async/try_one fixes - skip processes that are not asynsending to the target - do not clear whole asynsend table upon IPC permission error - be more accepting when one table entry is bogus later on
2011-04-19 00:56:34 +02:00
sys_map_t *map;
map = &priv(caller_ptr)->s_asyn_pending;
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
/* Try all privilege structures */
for (privp = BEG_PRIV_ADDR; privp < END_PRIV_ADDR; ++privp) {
if (privp->s_proc_nr == NONE)
continue;
Kernel: try_async/try_one fixes - skip processes that are not asynsending to the target - do not clear whole asynsend table upon IPC permission error - be more accepting when one table entry is bogus later on
2011-04-19 00:56:34 +02:00
if (!get_sys_bit(*map, privp->s_id))
continue;
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
src_ptr = proc_addr(privp->s_proc_nr);
assert(!(caller_ptr->p_misc_flags & MF_DELIVERMSG));
Use a bitmap for pending asynchronous messages instead of a global flag. That way it works similar to pending notifications.
2011-04-08 17:03:33 +02:00
if ((r = try_one(src_ptr, caller_ptr)) == OK)
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
return(r);
}
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
return(ESRCH);
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
}
/*===========================================================================*
* try_one *
*===========================================================================*/
Use a bitmap for pending asynchronous messages instead of a global flag. That way it works similar to pending notifications.
2011-04-08 17:03:33 +02:00
PRIVATE int try_one(struct proc *src_ptr, struct proc *dst_ptr)
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
{
Use a bitmap for pending asynchronous messages instead of a global flag. That way it works similar to pending notifications.
2011-04-08 17:03:33 +02:00
/* Try to receive an asynchronous message from 'src_ptr' */
fix clang warnings -R in kernel/ and servers/
2011-06-09 16:09:13 +02:00
int r = EAGAIN, done, do_notify;
Use a bitmap for pending asynchronous messages instead of a global flag. That way it works similar to pending notifications.
2011-04-08 17:03:33 +02:00
unsigned int flags, i;
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
size_t size;
endpoint_t dst;
struct proc *caller_ptr;
struct priv *privp;
asynmsg_t tabent;
vir_bytes table_v;
privp = priv(src_ptr);
if (!(privp->s_flags & SYS_PROC)) return(EPERM);
size = privp->s_asynsize;
table_v = privp->s_asyntab;
Use a bitmap for pending asynchronous messages instead of a global flag. That way it works similar to pending notifications.
2011-04-08 17:03:33 +02:00
/* Clear table pending message flag. We're done unless we're not. */
unset_sys_bit(priv(dst_ptr)->s_asyn_pending, privp->s_id);
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
if (size == 0) return(EAGAIN);
Enable sending a notification when sending of an asynchronous message was completed (successfully or not). AMF_NOTIFY_ERR can be used if the sender only wishes to be notified in case of an error (e.g., EDEADSRCDST). A new endpoint ASYNCM will be the sender of the notification.
2011-04-08 17:14:48 +02:00
if (!may_send_to(src_ptr, proc_nr(dst_ptr))) return(ECALLDENIED);
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
Enable sending a notification when sending of an asynchronous message was completed (successfully or not). AMF_NOTIFY_ERR can be used if the sender only wishes to be notified in case of an error (e.g., EDEADSRCDST). A new endpoint ASYNCM will be the sender of the notification.
2011-04-08 17:14:48 +02:00
caller_ptr = src_ptr; /* Needed for A_ macros later on */
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
/* Scan the table */
Kernel: try_async/try_one fixes - skip processes that are not asynsending to the target - do not clear whole asynsend table upon IPC permission error - be more accepting when one table entry is bogus later on
2011-04-19 00:56:34 +02:00
do_notify = FALSE;
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
done = TRUE;
for (i = 0; i < size; i++) {
/* Process each entry in the table and store the result in the table.
* If we're done handling a message, copy the result to the sender.
* Some checks done in mini_senda are duplicated here, as the sender
Kernel: try_async/try_one fixes - skip processes that are not asynsending to the target - do not clear whole asynsend table upon IPC permission error - be more accepting when one table entry is bogus later on
2011-04-19 00:56:34 +02:00
* could've altered the contents of the table in the meantime.
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
*/
/* Copy message to kernel */
A_RETR(i);
flags = tabent.flags;
dst = tabent.dst;
if (flags == 0) continue; /* Skip empty entries */
/* 'flags' field must contain only valid bits */
Enable sending a notification when sending of an asynchronous message was completed (successfully or not). AMF_NOTIFY_ERR can be used if the sender only wishes to be notified in case of an error (e.g., EDEADSRCDST). A new endpoint ASYNCM will be the sender of the notification.
2011-04-08 17:14:48 +02:00
if(flags & ~(AMF_VALID|AMF_DONE|AMF_NOTIFY|AMF_NOREPLY|AMF_NOTIFY_ERR))
r = EINVAL;
else if (!(flags & AMF_VALID)) /* Must contain message */
r = EINVAL;
else if (flags & AMF_DONE) continue; /* Already done processing */
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
/* Clear done flag. The sender is done sending when all messages in the
* table are marked done or empty. However, we will know that only
* the next time we enter this function or when the sender decides to
* send additional asynchronous messages and manages to deliver them
* all.
*/
done = FALSE;
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
Kernel: try_async/try_one fixes - skip processes that are not asynsending to the target - do not clear whole asynsend table upon IPC permission error - be more accepting when one table entry is bogus later on
2011-04-19 00:56:34 +02:00
if (r == EINVAL)
goto store_result;
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
/* Message must be directed at receiving end */
if (dst != dst_ptr->p_endpoint) continue;
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
/* If AMF_NOREPLY is set, then this message is not a reply to a
* SENDREC and thus should not satisfy the receiving part of the
* SENDREC. This message is to be delivered later.
*/
Use a bitmap for pending asynchronous messages instead of a global flag. That way it works similar to pending notifications.
2011-04-08 17:03:33 +02:00
if ((flags & AMF_NOREPLY) && (dst_ptr->p_misc_flags & MF_REPLY_PEND))
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
continue;
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
/* Destination is ready to receive the message; deliver it */
Enable sending a notification when sending of an asynchronous message was completed (successfully or not). AMF_NOTIFY_ERR can be used if the sender only wishes to be notified in case of an error (e.g., EDEADSRCDST). A new endpoint ASYNCM will be the sender of the notification.
2011-04-08 17:14:48 +02:00
r = OK;
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
dst_ptr->p_delivermsg = tabent.msg;
dst_ptr->p_delivermsg.m_source = src_ptr->p_endpoint;
dst_ptr->p_misc_flags |= MF_DELIVERMSG;
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
Enable sending a notification when sending of an asynchronous message was completed (successfully or not). AMF_NOTIFY_ERR can be used if the sender only wishes to be notified in case of an error (e.g., EDEADSRCDST). A new endpoint ASYNCM will be the sender of the notification.
2011-04-08 17:14:48 +02:00
store_result:
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
/* Store results for sender */
Enable sending a notification when sending of an asynchronous message was completed (successfully or not). AMF_NOTIFY_ERR can be used if the sender only wishes to be notified in case of an error (e.g., EDEADSRCDST). A new endpoint ASYNCM will be the sender of the notification.
2011-04-08 17:14:48 +02:00
tabent.result = r;
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
tabent.flags = flags | AMF_DONE;
Enable sending a notification when sending of an asynchronous message was completed (successfully or not). AMF_NOTIFY_ERR can be used if the sender only wishes to be notified in case of an error (e.g., EDEADSRCDST). A new endpoint ASYNCM will be the sender of the notification.
2011-04-08 17:14:48 +02:00
if (flags & AMF_NOTIFY) do_notify = TRUE;
else if (r != OK && (flags & AMF_NOTIFY_ERR)) do_notify = TRUE;
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
A_INSRT(i); /* Copy results to sender */
Merge of David's ptrace branch. Summary: o Support for ptrace T_ATTACH/T_DETACH and T_SYSCALL o PM signal handling logic should now work properly, even with debuggers being present o Asynchronous PM/VFS protocol, full IPC support for senda(), and AMF_NOREPLY senda() flag DETAILS Process stop and delay call handling of PM: o Added sys_runctl() kernel call with sys_stop() and sys_resume() aliases, for PM to stop and resume a process o Added exception for sending/syscall-traced processes to sys_runctl(), and matching SIGKREADY pseudo-signal to PM o Fixed PM signal logic to deal with requests from a process after stopping it (so-called "delay calls"), using the SIGKREADY facility o Fixed various PM panics due to race conditions with delay calls versus VFS calls o Removed special PRIO_STOP priority value o Added SYS_LOCK RTS kernel flag, to stop an individual process from running while modifying its process structure Signal and debugger handling in PM: o Fixed debugger signals being dropped if a second signal arrives when the debugger has not retrieved the first one o Fixed debugger signals being sent to the debugger more than once o Fixed debugger signals unpausing process in VFS; removed PM_UNPAUSE_TR protocol message o Detached debugger signals from general signal logic and from being blocked on VFS calls, meaning that even VFS can now be traced o Fixed debugger being unable to receive more than one pending signal in one process stop o Fixed signal delivery being delayed needlessly when multiple signals are pending o Fixed wait test for tracer, which was returning for children that were not waited for o Removed second parallel pending call from PM to VFS for any process o Fixed process becoming runnable between exec() and debugger trap o Added support for notifying the debugger before the parent when a debugged child exits o Fixed debugger death causing child to remain stopped forever o Fixed consistently incorrect use of _NSIG Extensions to ptrace(): o Added T_ATTACH and T_DETACH ptrace request, to attach and detach a debugger to and from a process o Added T_SYSCALL ptrace request, to trace system calls o Added T_SETOPT ptrace request, to set trace options o Added TO_TRACEFORK trace option, to attach automatically to children of a traced process o Added TO_ALTEXEC trace option, to send SIGSTOP instead of SIGTRAP upon a successful exec() of the tracee o Extended T_GETUSER ptrace support to allow retrieving a process's priv structure o Removed T_STOP ptrace request again, as it does not help implementing debuggers properly o Added MINIX3-specific ptrace test (test42) o Added proper manual page for ptrace(2) Asynchronous PM/VFS interface: o Fixed asynchronous messages not being checked when receive() is called with an endpoint other than ANY o Added AMF_NOREPLY senda() flag, preventing such messages from satisfying the receive part of a sendrec() o Added asynsend3() that takes optional flags; asynsend() is now a #define passing in 0 as third parameter o Made PM/VFS protocol asynchronous; reintroduced tell_fs() o Made PM_BASE request/reply number range unique o Hacked in a horrible temporary workaround into RS to deal with newly revealed RS-PM-VFS race condition triangle until VFS is asynchronous System signal handling: o Fixed shutdown logic of device drivers; removed old SIGKSTOP signal o Removed is-superuser check from PM's do_procstat() (aka getsigset()) o Added sigset macros to allow system processes to deal with the full signal set, rather than just the POSIX subset Miscellaneous PM fixes: o Split do_getset into do_get and do_set, merging common code and making structure clearer o Fixed setpriority() being able to put to sleep processes using an invalid parameter, or revive zombie processes o Made find_proc() global; removed obsolete proc_from_pid() o Cleanup here and there Also included: o Fixed false-positive boot order kernel warning o Removed last traces of old NOTIFY_FROM code THINGS OF POSSIBLE INTEREST o It should now be possible to run PM at any priority, even lower than user processes o No assumptions are made about communication speed between PM and VFS, although communication must be FIFO o A debugger will now receive incoming debuggee signals at kill time only; the process may not yet be fully stopped o A first step has been made towards making the SYSTEM task preemptible
2009-09-30 11:57:22 +02:00
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
break;
}
Merge of David's ptrace branch. Summary: o Support for ptrace T_ATTACH/T_DETACH and T_SYSCALL o PM signal handling logic should now work properly, even with debuggers being present o Asynchronous PM/VFS protocol, full IPC support for senda(), and AMF_NOREPLY senda() flag DETAILS Process stop and delay call handling of PM: o Added sys_runctl() kernel call with sys_stop() and sys_resume() aliases, for PM to stop and resume a process o Added exception for sending/syscall-traced processes to sys_runctl(), and matching SIGKREADY pseudo-signal to PM o Fixed PM signal logic to deal with requests from a process after stopping it (so-called "delay calls"), using the SIGKREADY facility o Fixed various PM panics due to race conditions with delay calls versus VFS calls o Removed special PRIO_STOP priority value o Added SYS_LOCK RTS kernel flag, to stop an individual process from running while modifying its process structure Signal and debugger handling in PM: o Fixed debugger signals being dropped if a second signal arrives when the debugger has not retrieved the first one o Fixed debugger signals being sent to the debugger more than once o Fixed debugger signals unpausing process in VFS; removed PM_UNPAUSE_TR protocol message o Detached debugger signals from general signal logic and from being blocked on VFS calls, meaning that even VFS can now be traced o Fixed debugger being unable to receive more than one pending signal in one process stop o Fixed signal delivery being delayed needlessly when multiple signals are pending o Fixed wait test for tracer, which was returning for children that were not waited for o Removed second parallel pending call from PM to VFS for any process o Fixed process becoming runnable between exec() and debugger trap o Added support for notifying the debugger before the parent when a debugged child exits o Fixed debugger death causing child to remain stopped forever o Fixed consistently incorrect use of _NSIG Extensions to ptrace(): o Added T_ATTACH and T_DETACH ptrace request, to attach and detach a debugger to and from a process o Added T_SYSCALL ptrace request, to trace system calls o Added T_SETOPT ptrace request, to set trace options o Added TO_TRACEFORK trace option, to attach automatically to children of a traced process o Added TO_ALTEXEC trace option, to send SIGSTOP instead of SIGTRAP upon a successful exec() of the tracee o Extended T_GETUSER ptrace support to allow retrieving a process's priv structure o Removed T_STOP ptrace request again, as it does not help implementing debuggers properly o Added MINIX3-specific ptrace test (test42) o Added proper manual page for ptrace(2) Asynchronous PM/VFS interface: o Fixed asynchronous messages not being checked when receive() is called with an endpoint other than ANY o Added AMF_NOREPLY senda() flag, preventing such messages from satisfying the receive part of a sendrec() o Added asynsend3() that takes optional flags; asynsend() is now a #define passing in 0 as third parameter o Made PM/VFS protocol asynchronous; reintroduced tell_fs() o Made PM_BASE request/reply number range unique o Hacked in a horrible temporary workaround into RS to deal with newly revealed RS-PM-VFS race condition triangle until VFS is asynchronous System signal handling: o Fixed shutdown logic of device drivers; removed old SIGKSTOP signal o Removed is-superuser check from PM's do_procstat() (aka getsigset()) o Added sigset macros to allow system processes to deal with the full signal set, rather than just the POSIX subset Miscellaneous PM fixes: o Split do_getset into do_get and do_set, merging common code and making structure clearer o Fixed setpriority() being able to put to sleep processes using an invalid parameter, or revive zombie processes o Made find_proc() global; removed obsolete proc_from_pid() o Cleanup here and there Also included: o Fixed false-positive boot order kernel warning o Removed last traces of old NOTIFY_FROM code THINGS OF POSSIBLE INTEREST o It should now be possible to run PM at any priority, even lower than user processes o No assumptions are made about communication speed between PM and VFS, although communication must be FIFO o A debugger will now receive incoming debuggee signals at kill time only; the process may not yet be fully stopped o A first step has been made towards making the SYSTEM task preemptible
2009-09-30 11:57:22 +02:00
Enable sending a notification when sending of an asynchronous message was completed (successfully or not). AMF_NOTIFY_ERR can be used if the sender only wishes to be notified in case of an error (e.g., EDEADSRCDST). A new endpoint ASYNCM will be the sender of the notification.
2011-04-08 17:14:48 +02:00
if (do_notify)
mini_notify(proc_addr(ASYNCM), src_ptr->p_endpoint);
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
Kernel: try_async/try_one fixes - skip processes that are not asynsending to the target - do not clear whole asynsend table upon IPC permission error - be more accepting when one table entry is bogus later on
2011-04-19 00:56:34 +02:00
if (done) {
privp->s_asyntab = -1;
privp->s_asynsize = 0;
} else {
set_sys_bit(priv(dst_ptr)->s_asyn_pending, privp->s_id);
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
}
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
Reduce the number of copies done by the kernel when handling retrieval and delivery of asynchronous messages.
2011-04-08 16:53:55 +02:00
return(r);
Clean and support for asynchronous sends.
2007-04-23 16:24:30 +02:00
}
Enable sending a notification when sending of an asynchronous message was completed (successfully or not). AMF_NOTIFY_ERR can be used if the sender only wishes to be notified in case of an error (e.g., EDEADSRCDST). A new endpoint ASYNCM will be the sender of the notification.
2011-04-08 17:14:48 +02:00
/*===========================================================================*
* cancel_async *
*===========================================================================*/
PUBLIC int cancel_async(struct proc *src_ptr, struct proc *dst_ptr)
{
/* Cancel asynchronous messages from src to dst, because dst is not interested
* in them (e.g., dst has been restarted) */
fix clang warnings -R in kernel/ and servers/
2011-06-09 16:09:13 +02:00
int done, do_notify;
Enable sending a notification when sending of an asynchronous message was completed (successfully or not). AMF_NOTIFY_ERR can be used if the sender only wishes to be notified in case of an error (e.g., EDEADSRCDST). A new endpoint ASYNCM will be the sender of the notification.
2011-04-08 17:14:48 +02:00
unsigned int flags, i;
size_t size;
endpoint_t dst;
struct proc *caller_ptr;
struct priv *privp;
asynmsg_t tabent;
vir_bytes table_v;
privp = priv(src_ptr);
if (!(privp->s_flags & SYS_PROC)) return(EPERM);
size = privp->s_asynsize;
table_v = privp->s_asyntab;
/* Clear table pending message flag. We're done unless we're not. */
privp->s_asyntab = -1;
privp->s_asynsize = 0;
unset_sys_bit(priv(dst_ptr)->s_asyn_pending, privp->s_id);
if (size == 0) return(EAGAIN);
if (!may_send_to(src_ptr, proc_nr(dst_ptr))) return(ECALLDENIED);
caller_ptr = src_ptr; /* Needed for A_ macros later on */
/* Scan the table */
do_notify = FALSE;
done = TRUE;
for (i = 0; i < size; i++) {
/* Process each entry in the table and store the result in the table.
* If we're done handling a message, copy the result to the sender.
* Some checks done in mini_senda are duplicated here, as the sender
* could've altered the contents of the table in the mean time.
*/
int r = EDEADSRCDST; /* Cancel delivery due to dead dst */
/* Copy message to kernel */
A_RETR(i);
flags = tabent.flags;
dst = tabent.dst;
if (flags == 0) continue; /* Skip empty entries */
/* 'flags' field must contain only valid bits */
if(flags & ~(AMF_VALID|AMF_DONE|AMF_NOTIFY|AMF_NOREPLY|AMF_NOTIFY_ERR))
r = EINVAL;
else if (!(flags & AMF_VALID)) /* Must contain message */
r = EINVAL;
else if (flags & AMF_DONE) continue; /* Already done processing */
/* Message must be directed at receiving end */
if (dst != dst_ptr->p_endpoint) {
done = FALSE;
continue;
}
/* Store results for sender */
tabent.result = r;
tabent.flags = flags | AMF_DONE;
if (flags & AMF_NOTIFY) do_notify = TRUE;
else if (r != OK && (flags & AMF_NOTIFY_ERR)) do_notify = TRUE;
A_INSRT(i); /* Copy results to sender */
}
if (do_notify)
mini_notify(proc_addr(ASYNCM), src_ptr->p_endpoint);
if (!done) {
privp->s_asyntab = table_v;
privp->s_asynsize = size;
}
return(OK);
}
Initial revision
2005-04-21 16:53:53 +02:00
/*===========================================================================*
New scheduling code in kernel. Work in progress. Round-robin within one priority queue works fine. Ageing algorithm to be done.
2005-08-19 18:43:28 +02:00
* enqueue *
Initial revision
2005-04-21 16:53:53 +02:00
*===========================================================================*/
Lots of const correctness.
2010-03-27 15:31:00 +01:00
PUBLIC void enqueue(
register struct proc *rp /* this process is now runnable */
)
Initial revision
2005-04-21 16:53:53 +02:00
{
Scheduling updates to the kernel. Sched() function now is single point for policy. Actual policy not yet implemented. PM calculates nice values for processes in boot image. IS debug dumps improved (Shift+F1-F4).
2005-08-22 17:14:11 +02:00
/* Add 'rp' to one of the queues of runnable processes. This function is
* responsible for inserting a process into one of the scheduling queues.
* The mechanism is implemented here. The actual scheduling policy is
* defined in sched() and pick_proc().
SMP - CPU local run queues - each CPU has its own runqueues - processes on BSP are put on the runqueues later after a switch to the final stack when cpuid works to avoid special cases - enqueue() and dequeue() use the run queues of the cpu the process is assigned to - pick_proc() uses the local run queues - printing of per-CPU run queues ('2') on serial console
2010-09-15 16:10:18 +02:00
*
* This function can be used x-cpu as it always uses the queues of the cpu the
* process is assigned to.
New scheduling code in kernel. Work in progress. Round-robin within one priority queue works fine. Ageing algorithm to be done.
2005-08-19 18:43:28 +02:00
*/
Userspace scheduling - cotributed by Bjorn Swift - In this first phase, scheduling is moved from the kernel to the PM server. The next steps are to a) moving scheduling to its own server and b) include useful information in the "out of quantum" message, so that the scheduler can make use of this information. - The kernel process table now keeps record of who is responsible for scheduling each process (p_scheduler). When this pointer is NULL, the process will be scheduled by the kernel. If such a process runs out of quantum, the kernel will simply renew its quantum an requeue it. - When PM loads, it will take over scheduling of all running processes, except system processes, using sys_schedctl(). Essentially, this only results in taking over init. As children inherit a scheduler from their parent, user space programs forked by init will inherit PM (for now) as their scheduler. - Once a process has been assigned a scheduler, and runs out of quantum, its RTS_NO_QUANTUM flag will be set and the process dequeued. The kernel will send a message to the scheduler, on the process' behalf, informing the scheduler that it has run out of quantum. The scheduler can take what ever action it pleases, based on its policy, and then reschedule the process using the sys_schedule() system call. - Balance queues does not work as before. While the old in-kernel function used to renew the quantum of processes in the highest priority run queue, the user-space implementation only acts on processes that have been bumped down to a lower priority queue. This approach reacts slower to changes than the old one, but saves us sending a sys_schedule message for each process every time we balance the queues. Currently, when processes are moved up a priority queue, their quantum is also renewed, but this can be fiddled with. - do_nice has been removed from kernel. PM answers to get- and setpriority calls, updates it's own nice variable as well as the max_run_queue. This will be refactored once scheduling is moved to a separate server. We will probably have PM update it's local nice value and then send a message to whoever is scheduling the process. - changes to fix an issue in do_fork() where processes could run out of quantum but bypassing the code path that handles it correctly. The future plan is to remove the policy from do_fork() and implement it in userspace too.
2010-03-29 13:07:20 +02:00
int q = rp->p_priority; /* scheduling queue to use */
SMP - CPU local run queues - each CPU has its own runqueues - processes on BSP are put on the runqueues later after a switch to the final stack when cpuid works to avoid special cases - enqueue() and dequeue() use the run queues of the cpu the process is assigned to - pick_proc() uses the local run queues - printing of per-CPU run queues ('2') on serial console
2010-09-15 16:10:18 +02:00
struct proc **rdy_head, **rdy_tail;
re-establish kernel assert()s. use the regular <assert.h> assert() instead of vmassert() in kernel. throw out some #if 0 code. fix a few assert() conditions. enable by default.
2010-03-10 14:00:05 +01:00
assert(proc_is_runnable(rp));
. fixed a recently introduced scheduler (run-queue) bug in proc.c . a little (optional) debugging code added to proc.c and associated data in proc.h
2005-05-24 14:32:34 +02:00
re-establish kernel assert()s. use the regular <assert.h> assert() instead of vmassert() in kernel. throw out some #if 0 code. fix a few assert() conditions. enable by default.
2010-03-10 14:00:05 +01:00
assert(q >= 0);
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
SMP - CPU local run queues - each CPU has its own runqueues - processes on BSP are put on the runqueues later after a switch to the final stack when cpuid works to avoid special cases - enqueue() and dequeue() use the run queues of the cpu the process is assigned to - pick_proc() uses the local run queues - printing of per-CPU run queues ('2') on serial console
2010-09-15 16:10:18 +02:00
rdy_head = get_cpu_var(rp->p_cpu, run_q_head);
rdy_tail = get_cpu_var(rp->p_cpu, run_q_tail);
New scheduling code in kernel. Work in progress. Round-robin within one priority queue works fine. Ageing algorithm to be done.
2005-08-19 18:43:28 +02:00
/* Now add the process to the queue. */
Removed NIL_SYS_PROC and NIL_PROC - NIL_PROC replaced by simple NULLs
2010-03-28 11:54:32 +02:00
if (!rdy_head[q]) { /* add to empty queue */
Removed unused code in proc.c. New message passing code with pointer pointers is now in use. Rewrote ready() and unready() fuctions.
2005-06-20 13:26:48 +02:00
rdy_head[q] = rdy_tail[q] = rp; /* create a new queue */
Removed NIL_SYS_PROC and NIL_PROC - NIL_PROC replaced by simple NULLs
2010-03-28 11:54:32 +02:00
rp->p_nextready = NULL; /* mark new end */
Initial revision
2005-04-21 16:53:53 +02:00
}
Removed unused code in proc.c. New message passing code with pointer pointers is now in use. Rewrote ready() and unready() fuctions.
2005-06-20 13:26:48 +02:00
else { /* add to tail of queue */
rdy_tail[q]->p_nextready = rp; /* chain tail of queue */
rdy_tail[q] = rp; /* set new queue tail */
Removed NIL_SYS_PROC and NIL_PROC - NIL_PROC replaced by simple NULLs
2010-03-28 11:54:32 +02:00
rp->p_nextready = NULL; /* mark new end */
Initial revision
2005-04-21 16:53:53 +02:00
}
Scheduling updates to the kernel. Sched() function now is single point for policy. Actual policy not yet implemented. PM calculates nice values for processes in boot image. IS debug dumps improved (Shift+F1-F4).
2005-08-22 17:14:11 +02:00
SMP - trully idle APs - any cpu can use smp_schedule() to tell another cpu to reschedule - if an AP is idle, it turns off timer as there is nothing to preempt, no need to wakeup just to go back to sleep again - if a cpu makes a process runnable on an idle cpu, it must wake it up to reschedule
2010-09-15 16:10:57 +02:00
if (cpuid == rp->p_cpu) {
/*
* enqueueing a process with a higher priority than the current one,
* it gets preempted. The current process must be preemptible. Testing
* the priority also makes sure that a process does not preempt itself
*/
struct proc * p;
p = get_cpulocal_var(proc_ptr);
assert(p);
if((p->p_priority > rp->p_priority) &&
(priv(p)->s_flags & PREEMPTIBLE))
RTS_SET(p, RTS_PREEMPTED); /* calls dequeue() */
}
#ifdef CONFIG_SMP
pick_proc() called only just before returning to userspace - new proc_is_runnable() macro to test whether process is runnable. All tests whether p_rts_flags == 0 converted to use this macro - pick_proc() calls removed from enqueue() and dequeue() - removed the test for recursive calls from pick_proc() as it certainly cannot be called recursively now - PREEMPTED flag to mark processes that were preempted by enqueueuing a higher priority process in enqueue() - enqueue_head() to enqueue PREEMPTED processes again at the head of their current priority queue - NO_QUANTUM flag to block and dequeue processes preempted by timer tick with exceeded quantum. They need to be enqueued again in schedcheck() - next_ptr global variable removed
2009-11-09 18:48:31 +01:00
/*
SMP - trully idle APs - any cpu can use smp_schedule() to tell another cpu to reschedule - if an AP is idle, it turns off timer as there is nothing to preempt, no need to wakeup just to go back to sleep again - if a cpu makes a process runnable on an idle cpu, it must wake it up to reschedule
2010-09-15 16:10:57 +02:00
* if the process was enqueued on a different cpu and the cpu is idle, i.e.
* the time is off, we need to wake up that cpu and let it schedule this new
* process
after enqueue()ing a process, only pick_proc() a new one if the current process is not PREEMPTIBLE (or it's not ready, or there isn't a current process yet). This fixes a case where a process that isn't PREEMPTIBLE actually gets preempted. (This solves a race condition between CLOCK and SYSTEM.)
2007-03-21 10:45:01 +01:00
*/
SMP - trully idle APs - any cpu can use smp_schedule() to tell another cpu to reschedule - if an AP is idle, it turns off timer as there is nothing to preempt, no need to wakeup just to go back to sleep again - if a cpu makes a process runnable on an idle cpu, it must wake it up to reschedule
2010-09-15 16:10:57 +02:00
else if (get_cpu_var(rp->p_cpu, cpu_is_idle)) {
smp_schedule(rp->p_cpu);
}
#endif
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
Userspace scheduling - exporting stats - contributed by Bjorn Swift - adds process accounting, for example counting the number of messages sent, how often the process was preemted and how much time it spent in the run queue. These statistics, along with the current cpu load, are sent back to the user-space scheduler in the Out Of Quantum message. - the user-space scheduler may choose to make use of these statistics when making scheduling decisions. For isntance the cpu load becomes especially useful when scheduling on multiple cores.
2010-09-19 17:52:12 +02:00
/* Make note of when this process was added to queue */
read_tsc_64(&(get_cpulocal_var(proc_ptr)->p_accounting.enter_queue));
re-establish kernel assert()s. use the regular <assert.h> assert() instead of vmassert() in kernel. throw out some #if 0 code. fix a few assert() conditions. enable by default.
2010-03-10 14:00:05 +01:00
#if DEBUG_SANITYCHECKS
SMP - CPU local run queues - each CPU has its own runqueues - processes on BSP are put on the runqueues later after a switch to the final stack when cpuid works to avoid special cases - enqueue() and dequeue() use the run queues of the cpu the process is assigned to - pick_proc() uses the local run queues - printing of per-CPU run queues ('2') on serial console
2010-09-15 16:10:18 +02:00
assert(runqueues_ok_local());
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
#endif
Initial revision
2005-04-21 16:53:53 +02:00
}
pick_proc() called only just before returning to userspace - new proc_is_runnable() macro to test whether process is runnable. All tests whether p_rts_flags == 0 converted to use this macro - pick_proc() calls removed from enqueue() and dequeue() - removed the test for recursive calls from pick_proc() as it certainly cannot be called recursively now - PREEMPTED flag to mark processes that were preempted by enqueueuing a higher priority process in enqueue() - enqueue_head() to enqueue PREEMPTED processes again at the head of their current priority queue - NO_QUANTUM flag to block and dequeue processes preempted by timer tick with exceeded quantum. They need to be enqueued again in schedcheck() - next_ptr global variable removed
2009-11-09 18:48:31 +01:00
/*===========================================================================*
* enqueue_head *
*===========================================================================*/
/*
* put a process at the front of its run queue. It comes handy when a process is
* preempted and removed from run queue to not to have a currently not-runnable
* process on a run queue. We have to put this process back at the fron to be
* fair
*/
PRIVATE void enqueue_head(struct proc *rp)
{
Lots of const correctness.
2010-03-27 15:31:00 +01:00
const int q = rp->p_priority; /* scheduling queue to use */
pick_proc() called only just before returning to userspace - new proc_is_runnable() macro to test whether process is runnable. All tests whether p_rts_flags == 0 converted to use this macro - pick_proc() calls removed from enqueue() and dequeue() - removed the test for recursive calls from pick_proc() as it certainly cannot be called recursively now - PREEMPTED flag to mark processes that were preempted by enqueueuing a higher priority process in enqueue() - enqueue_head() to enqueue PREEMPTED processes again at the head of their current priority queue - NO_QUANTUM flag to block and dequeue processes preempted by timer tick with exceeded quantum. They need to be enqueued again in schedcheck() - next_ptr global variable removed
2009-11-09 18:48:31 +01:00
SMP - CPU local run queues - each CPU has its own runqueues - processes on BSP are put on the runqueues later after a switch to the final stack when cpuid works to avoid special cases - enqueue() and dequeue() use the run queues of the cpu the process is assigned to - pick_proc() uses the local run queues - printing of per-CPU run queues ('2') on serial console
2010-09-15 16:10:18 +02:00
struct proc **rdy_head, **rdy_tail;
re-establish kernel assert()s. use the regular <assert.h> assert() instead of vmassert() in kernel. throw out some #if 0 code. fix a few assert() conditions. enable by default.
2010-03-10 14:00:05 +01:00
assert(proc_ptr_ok(rp));
assert(proc_is_runnable(rp));
pick_proc() called only just before returning to userspace - new proc_is_runnable() macro to test whether process is runnable. All tests whether p_rts_flags == 0 converted to use this macro - pick_proc() calls removed from enqueue() and dequeue() - removed the test for recursive calls from pick_proc() as it certainly cannot be called recursively now - PREEMPTED flag to mark processes that were preempted by enqueueuing a higher priority process in enqueue() - enqueue_head() to enqueue PREEMPTED processes again at the head of their current priority queue - NO_QUANTUM flag to block and dequeue processes preempted by timer tick with exceeded quantum. They need to be enqueued again in schedcheck() - next_ptr global variable removed
2009-11-09 18:48:31 +01:00
/*
* the process was runnable without its quantum expired when dequeued. A
* process with no time left should vahe been handled else and differently
*/
scheduling - time quantum in miliseconds - Currently the cpu time quantum is timer-ticks based. Thus the remaining quantum is decreased only if the processes is interrupted by a timer tick. As processes block a lot this typically does not happen for normal user processes. Also the quantum depends on the frequency of the timer. - This change makes the quantum miliseconds based. Internally the miliseconds are translated into cpu cycles. Everytime userspace execution is interrupted by kernel the cycles just consumed by the current process are deducted from the remaining quantum. - It makes the quantum system timer frequency independent. - The boot processes quantum is loosely derived from the tick-based quantas and 60Hz timer and subject to future change - the 64bit arithmetics is a little ugly, will be changes once we have compiler support for 64bit integers (soon)
2010-05-25 10:06:14 +02:00
assert(!is_zero64(rp->p_cpu_time_left));
pick_proc() called only just before returning to userspace - new proc_is_runnable() macro to test whether process is runnable. All tests whether p_rts_flags == 0 converted to use this macro - pick_proc() calls removed from enqueue() and dequeue() - removed the test for recursive calls from pick_proc() as it certainly cannot be called recursively now - PREEMPTED flag to mark processes that were preempted by enqueueuing a higher priority process in enqueue() - enqueue_head() to enqueue PREEMPTED processes again at the head of their current priority queue - NO_QUANTUM flag to block and dequeue processes preempted by timer tick with exceeded quantum. They need to be enqueued again in schedcheck() - next_ptr global variable removed
2009-11-09 18:48:31 +01:00
re-establish kernel assert()s. use the regular <assert.h> assert() instead of vmassert() in kernel. throw out some #if 0 code. fix a few assert() conditions. enable by default.
2010-03-10 14:00:05 +01:00
assert(q >= 0);
pick_proc() called only just before returning to userspace - new proc_is_runnable() macro to test whether process is runnable. All tests whether p_rts_flags == 0 converted to use this macro - pick_proc() calls removed from enqueue() and dequeue() - removed the test for recursive calls from pick_proc() as it certainly cannot be called recursively now - PREEMPTED flag to mark processes that were preempted by enqueueuing a higher priority process in enqueue() - enqueue_head() to enqueue PREEMPTED processes again at the head of their current priority queue - NO_QUANTUM flag to block and dequeue processes preempted by timer tick with exceeded quantum. They need to be enqueued again in schedcheck() - next_ptr global variable removed
2009-11-09 18:48:31 +01:00
SMP - CPU local run queues - each CPU has its own runqueues - processes on BSP are put on the runqueues later after a switch to the final stack when cpuid works to avoid special cases - enqueue() and dequeue() use the run queues of the cpu the process is assigned to - pick_proc() uses the local run queues - printing of per-CPU run queues ('2') on serial console
2010-09-15 16:10:18 +02:00
rdy_head = get_cpu_var(rp->p_cpu, run_q_head);
rdy_tail = get_cpu_var(rp->p_cpu, run_q_tail);
pick_proc() called only just before returning to userspace - new proc_is_runnable() macro to test whether process is runnable. All tests whether p_rts_flags == 0 converted to use this macro - pick_proc() calls removed from enqueue() and dequeue() - removed the test for recursive calls from pick_proc() as it certainly cannot be called recursively now - PREEMPTED flag to mark processes that were preempted by enqueueuing a higher priority process in enqueue() - enqueue_head() to enqueue PREEMPTED processes again at the head of their current priority queue - NO_QUANTUM flag to block and dequeue processes preempted by timer tick with exceeded quantum. They need to be enqueued again in schedcheck() - next_ptr global variable removed
2009-11-09 18:48:31 +01:00
/* Now add the process to the queue. */
Removed NIL_SYS_PROC and NIL_PROC - NIL_PROC replaced by simple NULLs
2010-03-28 11:54:32 +02:00
if (!rdy_head[q]) { /* add to empty queue */
pick_proc() called only just before returning to userspace - new proc_is_runnable() macro to test whether process is runnable. All tests whether p_rts_flags == 0 converted to use this macro - pick_proc() calls removed from enqueue() and dequeue() - removed the test for recursive calls from pick_proc() as it certainly cannot be called recursively now - PREEMPTED flag to mark processes that were preempted by enqueueuing a higher priority process in enqueue() - enqueue_head() to enqueue PREEMPTED processes again at the head of their current priority queue - NO_QUANTUM flag to block and dequeue processes preempted by timer tick with exceeded quantum. They need to be enqueued again in schedcheck() - next_ptr global variable removed
2009-11-09 18:48:31 +01:00
rdy_head[q] = rdy_tail[q] = rp; /* create a new queue */
Removed NIL_SYS_PROC and NIL_PROC - NIL_PROC replaced by simple NULLs
2010-03-28 11:54:32 +02:00
rp->p_nextready = NULL; /* mark new end */
pick_proc() called only just before returning to userspace - new proc_is_runnable() macro to test whether process is runnable. All tests whether p_rts_flags == 0 converted to use this macro - pick_proc() calls removed from enqueue() and dequeue() - removed the test for recursive calls from pick_proc() as it certainly cannot be called recursively now - PREEMPTED flag to mark processes that were preempted by enqueueuing a higher priority process in enqueue() - enqueue_head() to enqueue PREEMPTED processes again at the head of their current priority queue - NO_QUANTUM flag to block and dequeue processes preempted by timer tick with exceeded quantum. They need to be enqueued again in schedcheck() - next_ptr global variable removed
2009-11-09 18:48:31 +01:00
}
else /* add to head of queue */
rp->p_nextready = rdy_head[q]; /* chain head of queue */
rdy_head[q] = rp; /* set new queue head */
Userspace scheduling - exporting stats - contributed by Bjorn Swift - adds process accounting, for example counting the number of messages sent, how often the process was preemted and how much time it spent in the run queue. These statistics, along with the current cpu load, are sent back to the user-space scheduler in the Out Of Quantum message. - the user-space scheduler may choose to make use of these statistics when making scheduling decisions. For isntance the cpu load becomes especially useful when scheduling on multiple cores.
2010-09-19 17:52:12 +02:00
/* Make note of when this process was added to queue */
read_tsc_64(&(get_cpulocal_var(proc_ptr->p_accounting.enter_queue)));
/* Process accounting for scheduling */
rp->p_accounting.dequeues--;
rp->p_accounting.preempted++;
re-establish kernel assert()s. use the regular <assert.h> assert() instead of vmassert() in kernel. throw out some #if 0 code. fix a few assert() conditions. enable by default.
2010-03-10 14:00:05 +01:00
#if DEBUG_SANITYCHECKS
SMP - CPU local run queues - each CPU has its own runqueues - processes on BSP are put on the runqueues later after a switch to the final stack when cpuid works to avoid special cases - enqueue() and dequeue() use the run queues of the cpu the process is assigned to - pick_proc() uses the local run queues - printing of per-CPU run queues ('2') on serial console
2010-09-15 16:10:18 +02:00
assert(runqueues_ok_local());
pick_proc() called only just before returning to userspace - new proc_is_runnable() macro to test whether process is runnable. All tests whether p_rts_flags == 0 converted to use this macro - pick_proc() calls removed from enqueue() and dequeue() - removed the test for recursive calls from pick_proc() as it certainly cannot be called recursively now - PREEMPTED flag to mark processes that were preempted by enqueueuing a higher priority process in enqueue() - enqueue_head() to enqueue PREEMPTED processes again at the head of their current priority queue - NO_QUANTUM flag to block and dequeue processes preempted by timer tick with exceeded quantum. They need to be enqueued again in schedcheck() - next_ptr global variable removed
2009-11-09 18:48:31 +01:00
#endif
}
Initial revision
2005-04-21 16:53:53 +02:00
/*===========================================================================*
New scheduling code in kernel. Work in progress. Round-robin within one priority queue works fine. Ageing algorithm to be done.
2005-08-19 18:43:28 +02:00
* dequeue *
Initial revision
2005-04-21 16:53:53 +02:00
*===========================================================================*/
Userspace scheduling - exporting stats - contributed by Bjorn Swift - adds process accounting, for example counting the number of messages sent, how often the process was preemted and how much time it spent in the run queue. These statistics, along with the current cpu load, are sent back to the user-space scheduler in the Out Of Quantum message. - the user-space scheduler may choose to make use of these statistics when making scheduling decisions. For isntance the cpu load becomes especially useful when scheduling on multiple cores.
2010-09-19 17:52:12 +02:00
PUBLIC void dequeue(struct proc *rp)
Lots of const correctness.
2010-03-27 15:31:00 +01:00
/* this process is no longer runnable */
Initial revision
2005-04-21 16:53:53 +02:00
{
Scheduling updates to the kernel. Sched() function now is single point for policy. Actual policy not yet implemented. PM calculates nice values for processes in boot image. IS debug dumps improved (Shift+F1-F4).
2005-08-22 17:14:11 +02:00
/* A process must be removed from the scheduling queues, for example, because
* it has blocked. If the currently active process is removed, a new process
* is picked to run by calling pick_proc().
SMP - CPU local run queues - each CPU has its own runqueues - processes on BSP are put on the runqueues later after a switch to the final stack when cpuid works to avoid special cases - enqueue() and dequeue() use the run queues of the cpu the process is assigned to - pick_proc() uses the local run queues - printing of per-CPU run queues ('2') on serial console
2010-09-15 16:10:18 +02:00
*
* This function can operate x-cpu as it always removes the process from the
* queue of the cpu the process is currently assigned to.
Scheduling updates to the kernel. Sched() function now is single point for policy. Actual policy not yet implemented. PM calculates nice values for processes in boot image. IS debug dumps improved (Shift+F1-F4).
2005-08-22 17:14:11 +02:00
*/
Userspace scheduling - exporting stats - contributed by Bjorn Swift - adds process accounting, for example counting the number of messages sent, how often the process was preemted and how much time it spent in the run queue. These statistics, along with the current cpu load, are sent back to the user-space scheduler in the Out Of Quantum message. - the user-space scheduler may choose to make use of these statistics when making scheduling decisions. For isntance the cpu load becomes especially useful when scheduling on multiple cores.
2010-09-19 17:52:12 +02:00
int q = rp->p_priority; /* queue to use */
struct proc **xpp; /* iterate over queue */
struct proc *prev_xp;
u64_t tsc, tsc_delta;
Initial revision
2005-04-21 16:53:53 +02:00
SMP - CPU local run queues - each CPU has its own runqueues - processes on BSP are put on the runqueues later after a switch to the final stack when cpuid works to avoid special cases - enqueue() and dequeue() use the run queues of the cpu the process is assigned to - pick_proc() uses the local run queues - printing of per-CPU run queues ('2') on serial console
2010-09-15 16:10:18 +02:00
struct proc **rdy_tail;
re-establish kernel assert()s. use the regular <assert.h> assert() instead of vmassert() in kernel. throw out some #if 0 code. fix a few assert() conditions. enable by default.
2010-03-10 14:00:05 +01:00
assert(proc_ptr_ok(rp));
assert(!proc_is_runnable(rp));
Initial revision
2005-04-21 16:53:53 +02:00
re-establish kernel assert()s. use the regular <assert.h> assert() instead of vmassert() in kernel. throw out some #if 0 code. fix a few assert() conditions. enable by default.
2010-03-10 14:00:05 +01:00
/* Side-effect for kernel: check if the task's stack still is ok? */
assert (!iskernelp(rp) || *priv(rp)->s_stack_guard == STACK_GUARD);
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
SMP - CPU local run queues - each CPU has its own runqueues - processes on BSP are put on the runqueues later after a switch to the final stack when cpuid works to avoid special cases - enqueue() and dequeue() use the run queues of the cpu the process is assigned to - pick_proc() uses the local run queues - printing of per-CPU run queues ('2') on serial console
2010-09-15 16:10:18 +02:00
rdy_tail = get_cpu_var(rp->p_cpu, run_q_tail);
Initial revision
2005-04-21 16:53:53 +02:00
/* Now make sure that the process is not in its ready queue. Remove the
New system call added to PM: getprocnr(int *proc_nr) (like getpid); Minor update to scheduling code (unready().
2005-05-20 11:37:43 +02:00
* process if it is found. A process can be made unready even if it is not
* running by being sent a signal that kills it.
Initial revision
2005-04-21 16:53:53 +02:00
*/
Removed NIL_SYS_PROC and NIL_PROC - NIL_PROC replaced by simple NULLs
2010-03-28 11:54:32 +02:00
prev_xp = NULL;
SMP - CPU local run queues - each CPU has its own runqueues - processes on BSP are put on the runqueues later after a switch to the final stack when cpuid works to avoid special cases - enqueue() and dequeue() use the run queues of the cpu the process is assigned to - pick_proc() uses the local run queues - printing of per-CPU run queues ('2') on serial console
2010-09-15 16:10:18 +02:00
for (xpp = get_cpu_var_ptr(rp->p_cpu, run_q_head[q]); *xpp;
xpp = &(*xpp)->p_nextready) {
Removed unused code in proc.c. New message passing code with pointer pointers is now in use. Rewrote ready() and unready() fuctions.
2005-06-20 13:26:48 +02:00
if (*xpp == rp) { /* found process to remove */
*xpp = (*xpp)->p_nextready; /* replace with next chain */
re-establish kernel assert()s. use the regular <assert.h> assert() instead of vmassert() in kernel. throw out some #if 0 code. fix a few assert() conditions. enable by default.
2010-03-10 14:00:05 +01:00
if (rp == rdy_tail[q]) { /* queue tail removed */
Removed unused code in proc.c. New message passing code with pointer pointers is now in use. Rewrote ready() and unready() fuctions.
2005-06-20 13:26:48 +02:00
rdy_tail[q] = prev_xp; /* set new tail */
re-establish kernel assert()s. use the regular <assert.h> assert() instead of vmassert() in kernel. throw out some #if 0 code. fix a few assert() conditions. enable by default.
2010-03-10 14:00:05 +01:00
}
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
Optimized scheduling code. Old code is still available withing DEAD_CODE and NEW_SCHED_Q definitions. Some minor problems are being traced at the moment. This commit is meant to backup my files. --- Jorrit
2005-05-26 15:17:57 +02:00
break;
}
Removed unused code in proc.c. New message passing code with pointer pointers is now in use. Rewrote ready() and unready() fuctions.
2005-06-20 13:26:48 +02:00
prev_xp = *xpp; /* save previous in chain */
Initial revision
2005-04-21 16:53:53 +02:00
}
load average measurement implementation, accessable through getloadavg() system call in the library.
2005-11-14 16:50:46 +01:00
Userspace scheduling - exporting stats - contributed by Bjorn Swift - adds process accounting, for example counting the number of messages sent, how often the process was preemted and how much time it spent in the run queue. These statistics, along with the current cpu load, are sent back to the user-space scheduler in the Out Of Quantum message. - the user-space scheduler may choose to make use of these statistics when making scheduling decisions. For isntance the cpu load becomes especially useful when scheduling on multiple cores.
2010-09-19 17:52:12 +02:00
/* Process accounting for scheduling */
rp->p_accounting.dequeues++;
/* this is not all that accurate on virtual machines, especially with
IO bound processes that only spend a short amount of time in the queue
at a time. */
if (!is_zero64(rp->p_accounting.enter_queue)) {
read_tsc_64(&tsc);
tsc_delta = sub64(tsc, rp->p_accounting.enter_queue);
rp->p_accounting.time_in_queue = add64(rp->p_accounting.time_in_queue,
tsc_delta);
make_zero64(rp->p_accounting.enter_queue);
}
re-establish kernel assert()s. use the regular <assert.h> assert() instead of vmassert() in kernel. throw out some #if 0 code. fix a few assert() conditions. enable by default.
2010-03-10 14:00:05 +01:00
#if DEBUG_SANITYCHECKS
SMP - CPU local run queues - each CPU has its own runqueues - processes on BSP are put on the runqueues later after a switch to the final stack when cpuid works to avoid special cases - enqueue() and dequeue() use the run queues of the cpu the process is assigned to - pick_proc() uses the local run queues - printing of per-CPU run queues ('2') on serial console
2010-09-15 16:10:18 +02:00
assert(runqueues_ok_local());
Reorganized system call library; uses separate file per call now. New configuration header file to include/ exclude functionality. Extracted privileged features from struct proc and create new struct priv. Renamed various system calls for readability.
2005-07-14 17:12:12 +02:00
#endif
Initial revision
2005-04-21 16:53:53 +02:00
}
Rewrite of process scheduling: - current and maximum priority per process; - quantum size and current ticks left per process; - max number of full quantums in a row allow (otherwise current priority is decremented)
2005-06-30 17:55:19 +02:00
/*===========================================================================*
* pick_proc *
*===========================================================================*/
pick_proc() called only just before returning to userspace - new proc_is_runnable() macro to test whether process is runnable. All tests whether p_rts_flags == 0 converted to use this macro - pick_proc() calls removed from enqueue() and dequeue() - removed the test for recursive calls from pick_proc() as it certainly cannot be called recursively now - PREEMPTED flag to mark processes that were preempted by enqueueuing a higher priority process in enqueue() - enqueue_head() to enqueue PREEMPTED processes again at the head of their current priority queue - NO_QUANTUM flag to block and dequeue processes preempted by timer tick with exceeded quantum. They need to be enqueued again in schedcheck() - next_ptr global variable removed
2009-11-09 18:48:31 +01:00
PRIVATE struct proc * pick_proc(void)
Rewrite of process scheduling: - current and maximum priority per process; - quantum size and current ticks left per process; - max number of full quantums in a row allow (otherwise current priority is decremented)
2005-06-30 17:55:19 +02:00
{
pick_proc() called only just before returning to userspace - new proc_is_runnable() macro to test whether process is runnable. All tests whether p_rts_flags == 0 converted to use this macro - pick_proc() calls removed from enqueue() and dequeue() - removed the test for recursive calls from pick_proc() as it certainly cannot be called recursively now - PREEMPTED flag to mark processes that were preempted by enqueueuing a higher priority process in enqueue() - enqueue_head() to enqueue PREEMPTED processes again at the head of their current priority queue - NO_QUANTUM flag to block and dequeue processes preempted by timer tick with exceeded quantum. They need to be enqueued again in schedcheck() - next_ptr global variable removed
2009-11-09 18:48:31 +01:00
/* Decide who to run now. A new process is selected an returned.
Rewrite of process scheduling: - current and maximum priority per process; - quantum size and current ticks left per process; - max number of full quantums in a row allow (otherwise current priority is decremented)
2005-06-30 17:55:19 +02:00
* When a billable process is selected, record it in 'bill_ptr', so that the
* clock task can tell who to bill for system time.
SMP - CPU local run queues - each CPU has its own runqueues - processes on BSP are put on the runqueues later after a switch to the final stack when cpuid works to avoid special cases - enqueue() and dequeue() use the run queues of the cpu the process is assigned to - pick_proc() uses the local run queues - printing of per-CPU run queues ('2') on serial console
2010-09-15 16:10:18 +02:00
*
Kernel: small comment fix (Bug#562, reported by Ryan Riley)
2011-02-25 17:46:30 +01:00
* This function always uses the run queues of the local cpu!
Rewrite of process scheduling: - current and maximum priority per process; - quantum size and current ticks left per process; - max number of full quantums in a row allow (otherwise current priority is decremented)
2005-06-30 17:55:19 +02:00
*/
register struct proc *rp; /* process to run */
SMP - CPU local run queues - each CPU has its own runqueues - processes on BSP are put on the runqueues later after a switch to the final stack when cpuid works to avoid special cases - enqueue() and dequeue() use the run queues of the cpu the process is assigned to - pick_proc() uses the local run queues - printing of per-CPU run queues ('2') on serial console
2010-09-15 16:10:18 +02:00
struct proc **rdy_head;
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
int q; /* iterate over queues */
Rewrite of process scheduling: - current and maximum priority per process; - quantum size and current ticks left per process; - max number of full quantums in a row allow (otherwise current priority is decremented)
2005-06-30 17:55:19 +02:00
/* Check each of the scheduling queues for ready processes. The number of
* queues is defined in proc.h, and priorities are set in the task table.
Kernel: small comment fix (Bug#562, reported by Ryan Riley)
2011-02-25 17:46:30 +01:00
* If there are no processes ready to run, return NULL.
Rewrite of process scheduling: - current and maximum priority per process; - quantum size and current ticks left per process; - max number of full quantums in a row allow (otherwise current priority is decremented)
2005-06-30 17:55:19 +02:00
*/
SMP - CPU local run queues - each CPU has its own runqueues - processes on BSP are put on the runqueues later after a switch to the final stack when cpuid works to avoid special cases - enqueue() and dequeue() use the run queues of the cpu the process is assigned to - pick_proc() uses the local run queues - printing of per-CPU run queues ('2') on serial console
2010-09-15 16:10:18 +02:00
rdy_head = get_cpulocal_var(run_q_head);
Rewrite of process scheduling: - current and maximum priority per process; - quantum size and current ticks left per process; - max number of full quantums in a row allow (otherwise current priority is decremented)
2005-06-30 17:55:19 +02:00
for (q=0; q < NR_SCHED_QUEUES; q++) {
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
if(!(rp = rdy_head[q])) {
SMP - Print cpu of the process - adds '4' to print processes assigned to each cpu without printing the process it is blocked on (a lightweight '1')
2010-09-15 16:11:01 +02:00
TRACE(VF_PICKPROC, printf("cpu %d queue %d empty\n", cpuid, q););
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
continue;
}
re-establish kernel assert()s. use the regular <assert.h> assert() instead of vmassert() in kernel. throw out some #if 0 code. fix a few assert() conditions. enable by default.
2010-03-10 14:00:05 +01:00
assert(proc_is_runnable(rp));
Primary goal for these changes is: - no longer have kernel have its own page table that is loaded on every kernel entry (trap, interrupt, exception). the primary purpose is to reduce the number of required reloads. Result: - kernel can only access memory of process that was running when kernel was entered - kernel must be mapped into every process page table, so traps to kernel keep working Problem: - kernel must often access memory of arbitrary processes (e.g. send arbitrary processes messages); this can't happen directly any more; usually because that process' page table isn't loaded at all, sometimes because that memory isn't mapped in at all, sometimes because it isn't mapped in read-write. So: - kernel must be able to map in memory of any process, in its own address space. Implementation: - VM and kernel share a range of memory in which addresses of all page tables of all processes are available. This has two purposes: . Kernel has to know what data to copy in order to map in a range . Kernel has to know where to write the data in order to map it in That last point is because kernel has to write in the currently loaded page table. - Processes and kernel are separated through segments; kernel segments haven't changed. - The kernel keeps the process whose page table is currently loaded in 'ptproc.' - If it wants to map in a range of memory, it writes the value of the page directory entry for that range into the page directory entry in the currently loaded map. There is a slot reserved for such purposes. The kernel can then access this memory directly. - In order to do this, its segment has been increased (and the segments of processes start where it ends). - In the pagefault handler, detect if the kernel is doing 'trappable' memory access (i.e. a pagefault isn't a fatal error) and if so, - set the saved instruction pointer to phys_copy_fault, breaking out of phys_copy - set the saved eax register to the address of the page fault, both for sanity checking and for checking in which of the two ranges that phys_copy was called with the fault occured - Some boot-time processes do not have their own page table, and are mapped in with the kernel, and separated with segments. The kernel detects this using HASPT. If such a process has to be scheduled, any page table will work and no page table switch is done. Major changes in kernel are - When accessing user processes memory, kernel no longer explicitly checks before it does so if that memory is OK. It simply makes the mapping (if necessary), tries to do the operation, and traps the pagefault if that memory isn't present; if that happens, the copy function returns EFAULT. So all of the CHECKRANGE_OR_SUSPEND macros are gone. - Kernel no longer has to copy/read and parse page tables. - A message copying optimisation: when messages are copied, and the recipient isn't mapped in, they are copied into a buffer in the kernel. This is done in QueueMess. The next time the recipient is scheduled, this message is copied into its memory. This happens in schedcheck(). This eliminates the mapping/copying step for messages, and makes it easier to deliver messages. This eliminates soft_notify. - Kernel no longer creates a page table at all, so the vm_setbuf and pagetable writing in memory.c is gone. Minor changes in kernel are - ipc_stats thrown out, wasn't used - misc flags all renamed to MF_* - NOREC_* macros to enter and leave functions that should not be called recursively; just sanity checks really - code to fully decode segment selectors and descriptors to print on exceptions - lots of vmassert()s added, only executed if DEBUG_VMASSERT is 1
2009-09-21 16:31:52 +02:00
if (priv(rp)->s_flags & BILLABLE)
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
get_cpulocal_var(bill_ptr) = rp; /* bill for system time */
pick_proc() called only just before returning to userspace - new proc_is_runnable() macro to test whether process is runnable. All tests whether p_rts_flags == 0 converted to use this macro - pick_proc() calls removed from enqueue() and dequeue() - removed the test for recursive calls from pick_proc() as it certainly cannot be called recursively now - PREEMPTED flag to mark processes that were preempted by enqueueuing a higher priority process in enqueue() - enqueue_head() to enqueue PREEMPTED processes again at the head of their current priority queue - NO_QUANTUM flag to block and dequeue processes preempted by timer tick with exceeded quantum. They need to be enqueued again in schedcheck() - next_ptr global variable removed
2009-11-09 18:48:31 +01:00
return rp;
Rewrite of process scheduling: - current and maximum priority per process; - quantum size and current ticks left per process; - max number of full quantums in a row allow (otherwise current priority is decremented)
2005-06-30 17:55:19 +02:00
}
pick_proc() called only just before returning to userspace - new proc_is_runnable() macro to test whether process is runnable. All tests whether p_rts_flags == 0 converted to use this macro - pick_proc() calls removed from enqueue() and dequeue() - removed the test for recursive calls from pick_proc() as it certainly cannot be called recursively now - PREEMPTED flag to mark processes that were preempted by enqueueuing a higher priority process in enqueue() - enqueue_head() to enqueue PREEMPTED processes again at the head of their current priority queue - NO_QUANTUM flag to block and dequeue processes preempted by timer tick with exceeded quantum. They need to be enqueued again in schedcheck() - next_ptr global variable removed
2009-11-09 18:48:31 +01:00
return NULL;
Initial revision
2005-04-21 16:53:53 +02:00
}
Basic VM and other minor improvements. Not complete, probably not fully debugged or optimized.
2008-11-19 13:26:10 +01:00
/*===========================================================================*
* endpoint_lookup *
*===========================================================================*/
PUBLIC struct proc *endpoint_lookup(endpoint_t e)
{
int n;
if(!isokendpt(e, &n)) return NULL;
return proc_addr(n);
}
'proc number' is process slot, 'endpoint' are generation-aware process instance numbers, encoded and decoded using macros in <minix/endpoint.h>. proc number -> endpoint migration . proc_nr in the interrupt hook is now an endpoint, proc_nr_e. . m_source for messages and notifies is now an endpoint, instead of proc number. . isokendpt() converts an endpoint to a process number, returns success (but fails if the process number is out of range, the process slot is not a living process, or the given endpoint number does not match the endpoint number in the process slot, indicating an old process). . okendpt() is the same as isokendpt(), but panic()s if the conversion fails. This is mainly used for decoding message.m_source endpoints, and other endpoint numbers in kernel data structures, which should always be correct. . if DEBUG_ENABLE_IPC_WARNINGS is enabled, isokendpt() and okendpt() get passed the __FILE__ and __LINE__ of the calling lines, and print messages about what is wrong with the endpoint number (out of range proc, empty proc, or inconsistent endpoint number), with the caller, making finding where the conversion failed easy without having to include code for every call to print where things went wrong. Sometimes this is harmless (wrong arg to a kernel call), sometimes it's a fatal internal inconsistency (bogus m_source). . some process table fields have been appended an _e to indicate it's become and endpoint. . process endpoint is stored in p_endpoint, without generation number. it turns out the kernel never needs the generation number, except when fork()ing, so it's decoded then. . kernel calls all take endpoints as arguments, not proc numbers. the one exception is sys_fork(), which needs to know in which slot to put the child.
2006-03-03 11:00:02 +01:00
/*===========================================================================*
* isokendpt_f *
*===========================================================================*/
#if DEBUG_ENABLE_IPC_WARNINGS
PUBLIC int isokendpt_f(file, line, e, p, fatalflag)
Lots of const correctness.
2010-03-27 15:31:00 +01:00
const char *file;
'proc number' is process slot, 'endpoint' are generation-aware process instance numbers, encoded and decoded using macros in <minix/endpoint.h>. proc number -> endpoint migration . proc_nr in the interrupt hook is now an endpoint, proc_nr_e. . m_source for messages and notifies is now an endpoint, instead of proc number. . isokendpt() converts an endpoint to a process number, returns success (but fails if the process number is out of range, the process slot is not a living process, or the given endpoint number does not match the endpoint number in the process slot, indicating an old process). . okendpt() is the same as isokendpt(), but panic()s if the conversion fails. This is mainly used for decoding message.m_source endpoints, and other endpoint numbers in kernel data structures, which should always be correct. . if DEBUG_ENABLE_IPC_WARNINGS is enabled, isokendpt() and okendpt() get passed the __FILE__ and __LINE__ of the calling lines, and print messages about what is wrong with the endpoint number (out of range proc, empty proc, or inconsistent endpoint number), with the caller, making finding where the conversion failed easy without having to include code for every call to print where things went wrong. Sometimes this is harmless (wrong arg to a kernel call), sometimes it's a fatal internal inconsistency (bogus m_source). . some process table fields have been appended an _e to indicate it's become and endpoint. . process endpoint is stored in p_endpoint, without generation number. it turns out the kernel never needs the generation number, except when fork()ing, so it's decoded then. . kernel calls all take endpoints as arguments, not proc numbers. the one exception is sys_fork(), which needs to know in which slot to put the child.
2006-03-03 11:00:02 +01:00
int line;
#else
PUBLIC int isokendpt_f(e, p, fatalflag)
#endif
Use endpoint_t. New prototypes for related to grants and safecopy functions.
2006-06-20 11:57:00 +02:00
endpoint_t e;
Lots of const correctness.
2010-03-27 15:31:00 +01:00
int *p;
const int fatalflag;
'proc number' is process slot, 'endpoint' are generation-aware process instance numbers, encoded and decoded using macros in <minix/endpoint.h>. proc number -> endpoint migration . proc_nr in the interrupt hook is now an endpoint, proc_nr_e. . m_source for messages and notifies is now an endpoint, instead of proc number. . isokendpt() converts an endpoint to a process number, returns success (but fails if the process number is out of range, the process slot is not a living process, or the given endpoint number does not match the endpoint number in the process slot, indicating an old process). . okendpt() is the same as isokendpt(), but panic()s if the conversion fails. This is mainly used for decoding message.m_source endpoints, and other endpoint numbers in kernel data structures, which should always be correct. . if DEBUG_ENABLE_IPC_WARNINGS is enabled, isokendpt() and okendpt() get passed the __FILE__ and __LINE__ of the calling lines, and print messages about what is wrong with the endpoint number (out of range proc, empty proc, or inconsistent endpoint number), with the caller, making finding where the conversion failed easy without having to include code for every call to print where things went wrong. Sometimes this is harmless (wrong arg to a kernel call), sometimes it's a fatal internal inconsistency (bogus m_source). . some process table fields have been appended an _e to indicate it's become and endpoint. . process endpoint is stored in p_endpoint, without generation number. it turns out the kernel never needs the generation number, except when fork()ing, so it's decoded then. . kernel calls all take endpoints as arguments, not proc numbers. the one exception is sys_fork(), which needs to know in which slot to put the child.
2006-03-03 11:00:02 +01:00
{
int ok = 0;
/* Convert an endpoint number into a process number.
* Return nonzero if the process is alive with the corresponding
* generation number, zero otherwise.
*
* This function is called with file and line number by the
* isokendpt_d macro if DEBUG_ENABLE_IPC_WARNINGS is defined,
* otherwise without. This allows us to print the where the
* conversion was attempted, making the errors verbose without
* adding code for that at every call.
*
* If fatalflag is nonzero, we must panic if the conversion doesn't
* succeed.
*/
*p = _ENDPOINT_P(e);
if(!isokprocn(*p)) {
#if DEBUG_ENABLE_IPC_WARNINGS
no more kprintf - kernel uses libsys printf now, only kputc is special to the kernel.
2010-03-03 16:45:01 +01:00
printf("kernel:%s:%d: bad endpoint %d: proc %d out of range\n",
'proc number' is process slot, 'endpoint' are generation-aware process instance numbers, encoded and decoded using macros in <minix/endpoint.h>. proc number -> endpoint migration . proc_nr in the interrupt hook is now an endpoint, proc_nr_e. . m_source for messages and notifies is now an endpoint, instead of proc number. . isokendpt() converts an endpoint to a process number, returns success (but fails if the process number is out of range, the process slot is not a living process, or the given endpoint number does not match the endpoint number in the process slot, indicating an old process). . okendpt() is the same as isokendpt(), but panic()s if the conversion fails. This is mainly used for decoding message.m_source endpoints, and other endpoint numbers in kernel data structures, which should always be correct. . if DEBUG_ENABLE_IPC_WARNINGS is enabled, isokendpt() and okendpt() get passed the __FILE__ and __LINE__ of the calling lines, and print messages about what is wrong with the endpoint number (out of range proc, empty proc, or inconsistent endpoint number), with the caller, making finding where the conversion failed easy without having to include code for every call to print where things went wrong. Sometimes this is harmless (wrong arg to a kernel call), sometimes it's a fatal internal inconsistency (bogus m_source). . some process table fields have been appended an _e to indicate it's become and endpoint. . process endpoint is stored in p_endpoint, without generation number. it turns out the kernel never needs the generation number, except when fork()ing, so it's decoded then. . kernel calls all take endpoints as arguments, not proc numbers. the one exception is sys_fork(), which needs to know in which slot to put the child.
2006-03-03 11:00:02 +01:00
file, line, e, *p);
#endif
} else if(isemptyn(*p)) {
Basic VM and other minor improvements. Not complete, probably not fully debugged or optimized.
2008-11-19 13:26:10 +01:00
#if 0
no more kprintf - kernel uses libsys printf now, only kputc is special to the kernel.
2010-03-03 16:45:01 +01:00
printf("kernel:%s:%d: bad endpoint %d: proc %d empty\n", file, line, e, *p);
'proc number' is process slot, 'endpoint' are generation-aware process instance numbers, encoded and decoded using macros in <minix/endpoint.h>. proc number -> endpoint migration . proc_nr in the interrupt hook is now an endpoint, proc_nr_e. . m_source for messages and notifies is now an endpoint, instead of proc number. . isokendpt() converts an endpoint to a process number, returns success (but fails if the process number is out of range, the process slot is not a living process, or the given endpoint number does not match the endpoint number in the process slot, indicating an old process). . okendpt() is the same as isokendpt(), but panic()s if the conversion fails. This is mainly used for decoding message.m_source endpoints, and other endpoint numbers in kernel data structures, which should always be correct. . if DEBUG_ENABLE_IPC_WARNINGS is enabled, isokendpt() and okendpt() get passed the __FILE__ and __LINE__ of the calling lines, and print messages about what is wrong with the endpoint number (out of range proc, empty proc, or inconsistent endpoint number), with the caller, making finding where the conversion failed easy without having to include code for every call to print where things went wrong. Sometimes this is harmless (wrong arg to a kernel call), sometimes it's a fatal internal inconsistency (bogus m_source). . some process table fields have been appended an _e to indicate it's become and endpoint. . process endpoint is stored in p_endpoint, without generation number. it turns out the kernel never needs the generation number, except when fork()ing, so it's decoded then. . kernel calls all take endpoints as arguments, not proc numbers. the one exception is sys_fork(), which needs to know in which slot to put the child.
2006-03-03 11:00:02 +01:00
#endif
} else if(proc_addr(*p)->p_endpoint != e) {
#if DEBUG_ENABLE_IPC_WARNINGS
no more kprintf - kernel uses libsys printf now, only kputc is special to the kernel.
2010-03-03 16:45:01 +01:00
printf("kernel:%s:%d: bad endpoint %d: proc %d has ept %d (generation %d vs. %d)\n", file, line,
'proc number' is process slot, 'endpoint' are generation-aware process instance numbers, encoded and decoded using macros in <minix/endpoint.h>. proc number -> endpoint migration . proc_nr in the interrupt hook is now an endpoint, proc_nr_e. . m_source for messages and notifies is now an endpoint, instead of proc number. . isokendpt() converts an endpoint to a process number, returns success (but fails if the process number is out of range, the process slot is not a living process, or the given endpoint number does not match the endpoint number in the process slot, indicating an old process). . okendpt() is the same as isokendpt(), but panic()s if the conversion fails. This is mainly used for decoding message.m_source endpoints, and other endpoint numbers in kernel data structures, which should always be correct. . if DEBUG_ENABLE_IPC_WARNINGS is enabled, isokendpt() and okendpt() get passed the __FILE__ and __LINE__ of the calling lines, and print messages about what is wrong with the endpoint number (out of range proc, empty proc, or inconsistent endpoint number), with the caller, making finding where the conversion failed easy without having to include code for every call to print where things went wrong. Sometimes this is harmless (wrong arg to a kernel call), sometimes it's a fatal internal inconsistency (bogus m_source). . some process table fields have been appended an _e to indicate it's become and endpoint. . process endpoint is stored in p_endpoint, without generation number. it turns out the kernel never needs the generation number, except when fork()ing, so it's decoded then. . kernel calls all take endpoints as arguments, not proc numbers. the one exception is sys_fork(), which needs to know in which slot to put the child.
2006-03-03 11:00:02 +01:00
e, *p, proc_addr(*p)->p_endpoint,
_ENDPOINT_G(e), _ENDPOINT_G(proc_addr(*p)->p_endpoint));
#endif
} else ok = 1;
if(!ok && fatalflag) {
panic() cleanup. this change - makes panic() variadic, doing full printf() formatting - no more NO_NUM, and no more separate printf() statements needed to print extra info (or something in hex) before panicing - unifies panic() - same panic() name and usage for everyone - vm, kernel and rest have different names/syntax currently in order to implement their own luxuries, but no longer - throws out the 1st argument, to make source less noisy. the panic() in syslib retrieves the server name from the kernel so it should be clear enough who is panicing; e.g. panic("sigaction failed: %d", errno); looks like: at_wini(73130): panic: sigaction failed: 0 syslib:panic.c: stacktrace: 0x74dc 0x2025 0x100a - throws out report() - printf() is more convenient and powerful - harmonizes/fixes the use of panic() - there were a few places that used printf-style formatting (didn't work) and newlines (messes up the formatting) in panic() - throws out a few per-server panic() functions - cleans up a tie-in of tty with panic() merging printf() and panic() statements to be done incrementally.
2010-03-05 16:05:11 +01:00
panic("invalid endpoint: %d", e);
'proc number' is process slot, 'endpoint' are generation-aware process instance numbers, encoded and decoded using macros in <minix/endpoint.h>. proc number -> endpoint migration . proc_nr in the interrupt hook is now an endpoint, proc_nr_e. . m_source for messages and notifies is now an endpoint, instead of proc number. . isokendpt() converts an endpoint to a process number, returns success (but fails if the process number is out of range, the process slot is not a living process, or the given endpoint number does not match the endpoint number in the process slot, indicating an old process). . okendpt() is the same as isokendpt(), but panic()s if the conversion fails. This is mainly used for decoding message.m_source endpoints, and other endpoint numbers in kernel data structures, which should always be correct. . if DEBUG_ENABLE_IPC_WARNINGS is enabled, isokendpt() and okendpt() get passed the __FILE__ and __LINE__ of the calling lines, and print messages about what is wrong with the endpoint number (out of range proc, empty proc, or inconsistent endpoint number), with the caller, making finding where the conversion failed easy without having to include code for every call to print where things went wrong. Sometimes this is harmless (wrong arg to a kernel call), sometimes it's a fatal internal inconsistency (bogus m_source). . some process table fields have been appended an _e to indicate it's become and endpoint. . process endpoint is stored in p_endpoint, without generation number. it turns out the kernel never needs the generation number, except when fork()ing, so it's decoded then. . kernel calls all take endpoints as arguments, not proc numbers. the one exception is sys_fork(), which needs to know in which slot to put the child.
2006-03-03 11:00:02 +01:00
}
return ok;
}
Basic VM and other minor improvements. Not complete, probably not fully debugged or optimized.
2008-11-19 13:26:10 +01:00
Userspace scheduling - cotributed by Bjorn Swift - In this first phase, scheduling is moved from the kernel to the PM server. The next steps are to a) moving scheduling to its own server and b) include useful information in the "out of quantum" message, so that the scheduler can make use of this information. - The kernel process table now keeps record of who is responsible for scheduling each process (p_scheduler). When this pointer is NULL, the process will be scheduled by the kernel. If such a process runs out of quantum, the kernel will simply renew its quantum an requeue it. - When PM loads, it will take over scheduling of all running processes, except system processes, using sys_schedctl(). Essentially, this only results in taking over init. As children inherit a scheduler from their parent, user space programs forked by init will inherit PM (for now) as their scheduler. - Once a process has been assigned a scheduler, and runs out of quantum, its RTS_NO_QUANTUM flag will be set and the process dequeued. The kernel will send a message to the scheduler, on the process' behalf, informing the scheduler that it has run out of quantum. The scheduler can take what ever action it pleases, based on its policy, and then reschedule the process using the sys_schedule() system call. - Balance queues does not work as before. While the old in-kernel function used to renew the quantum of processes in the highest priority run queue, the user-space implementation only acts on processes that have been bumped down to a lower priority queue. This approach reacts slower to changes than the old one, but saves us sending a sys_schedule message for each process every time we balance the queues. Currently, when processes are moved up a priority queue, their quantum is also renewed, but this can be fiddled with. - do_nice has been removed from kernel. PM answers to get- and setpriority calls, updates it's own nice variable as well as the max_run_queue. This will be refactored once scheduling is moved to a separate server. We will probably have PM update it's local nice value and then send a message to whoever is scheduling the process. - changes to fix an issue in do_fork() where processes could run out of quantum but bypassing the code path that handles it correctly. The future plan is to remove the policy from do_fork() and implement it in userspace too.
2010-03-29 13:07:20 +02:00
PRIVATE void notify_scheduler(struct proc *p)
{
scheduling - time quantum in miliseconds - Currently the cpu time quantum is timer-ticks based. Thus the remaining quantum is decreased only if the processes is interrupted by a timer tick. As processes block a lot this typically does not happen for normal user processes. Also the quantum depends on the frequency of the timer. - This change makes the quantum miliseconds based. Internally the miliseconds are translated into cpu cycles. Everytime userspace execution is interrupted by kernel the cycles just consumed by the current process are deducted from the remaining quantum. - It makes the quantum system timer frequency independent. - The boot processes quantum is loosely derived from the tick-based quantas and 60Hz timer and subject to future change - the 64bit arithmetics is a little ugly, will be changes once we have compiler support for 64bit integers (soon)
2010-05-25 10:06:14 +02:00
message m_no_quantum;
int err;
assert(!proc_kernel_scheduler(p));
Userspace scheduling - cotributed by Bjorn Swift - In this first phase, scheduling is moved from the kernel to the PM server. The next steps are to a) moving scheduling to its own server and b) include useful information in the "out of quantum" message, so that the scheduler can make use of this information. - The kernel process table now keeps record of who is responsible for scheduling each process (p_scheduler). When this pointer is NULL, the process will be scheduled by the kernel. If such a process runs out of quantum, the kernel will simply renew its quantum an requeue it. - When PM loads, it will take over scheduling of all running processes, except system processes, using sys_schedctl(). Essentially, this only results in taking over init. As children inherit a scheduler from their parent, user space programs forked by init will inherit PM (for now) as their scheduler. - Once a process has been assigned a scheduler, and runs out of quantum, its RTS_NO_QUANTUM flag will be set and the process dequeued. The kernel will send a message to the scheduler, on the process' behalf, informing the scheduler that it has run out of quantum. The scheduler can take what ever action it pleases, based on its policy, and then reschedule the process using the sys_schedule() system call. - Balance queues does not work as before. While the old in-kernel function used to renew the quantum of processes in the highest priority run queue, the user-space implementation only acts on processes that have been bumped down to a lower priority queue. This approach reacts slower to changes than the old one, but saves us sending a sys_schedule message for each process every time we balance the queues. Currently, when processes are moved up a priority queue, their quantum is also renewed, but this can be fiddled with. - do_nice has been removed from kernel. PM answers to get- and setpriority calls, updates it's own nice variable as well as the max_run_queue. This will be refactored once scheduling is moved to a separate server. We will probably have PM update it's local nice value and then send a message to whoever is scheduling the process. - changes to fix an issue in do_fork() where processes could run out of quantum but bypassing the code path that handles it correctly. The future plan is to remove the policy from do_fork() and implement it in userspace too.
2010-03-29 13:07:20 +02:00
/* dequeue the process */
RTS_SET(p, RTS_NO_QUANTUM);
/*
* Notify the process's scheduler that it has run out of
* quantum. This is done by sending a message to the scheduler
* on the process's behalf
*/
scheduling - time quantum in miliseconds - Currently the cpu time quantum is timer-ticks based. Thus the remaining quantum is decreased only if the processes is interrupted by a timer tick. As processes block a lot this typically does not happen for normal user processes. Also the quantum depends on the frequency of the timer. - This change makes the quantum miliseconds based. Internally the miliseconds are translated into cpu cycles. Everytime userspace execution is interrupted by kernel the cycles just consumed by the current process are deducted from the remaining quantum. - It makes the quantum system timer frequency independent. - The boot processes quantum is loosely derived from the tick-based quantas and 60Hz timer and subject to future change - the 64bit arithmetics is a little ugly, will be changes once we have compiler support for 64bit integers (soon)
2010-05-25 10:06:14 +02:00
m_no_quantum.m_source = p->p_endpoint;
m_no_quantum.m_type = SCHEDULING_NO_QUANTUM;
Userspace scheduling - exporting stats - contributed by Bjorn Swift - adds process accounting, for example counting the number of messages sent, how often the process was preemted and how much time it spent in the run queue. These statistics, along with the current cpu load, are sent back to the user-space scheduler in the Out Of Quantum message. - the user-space scheduler may choose to make use of these statistics when making scheduling decisions. For isntance the cpu load becomes especially useful when scheduling on multiple cores.
2010-09-19 17:52:12 +02:00
m_no_quantum.SCHEDULING_ACNT_QUEUE = cpu_time_2_ms(p->p_accounting.time_in_queue);
m_no_quantum.SCHEDULING_ACNT_DEQS = p->p_accounting.dequeues;
m_no_quantum.SCHEDULING_ACNT_IPC_SYNC = p->p_accounting.ipc_sync;
m_no_quantum.SCHEDULING_ACNT_IPC_ASYNC = p->p_accounting.ipc_async;
m_no_quantum.SCHEDULING_ACNT_PREEMPT = p->p_accounting.preempted;
m_no_quantum.SCHEDULING_ACNT_CPU = cpuid;
m_no_quantum.SCHEDULING_ACNT_CPU_LOAD = cpu_load();
/* Reset accounting */
reset_proc_accounting(p);
Userspace scheduling - cotributed by Bjorn Swift - In this first phase, scheduling is moved from the kernel to the PM server. The next steps are to a) moving scheduling to its own server and b) include useful information in the "out of quantum" message, so that the scheduler can make use of this information. - The kernel process table now keeps record of who is responsible for scheduling each process (p_scheduler). When this pointer is NULL, the process will be scheduled by the kernel. If such a process runs out of quantum, the kernel will simply renew its quantum an requeue it. - When PM loads, it will take over scheduling of all running processes, except system processes, using sys_schedctl(). Essentially, this only results in taking over init. As children inherit a scheduler from their parent, user space programs forked by init will inherit PM (for now) as their scheduler. - Once a process has been assigned a scheduler, and runs out of quantum, its RTS_NO_QUANTUM flag will be set and the process dequeued. The kernel will send a message to the scheduler, on the process' behalf, informing the scheduler that it has run out of quantum. The scheduler can take what ever action it pleases, based on its policy, and then reschedule the process using the sys_schedule() system call. - Balance queues does not work as before. While the old in-kernel function used to renew the quantum of processes in the highest priority run queue, the user-space implementation only acts on processes that have been bumped down to a lower priority queue. This approach reacts slower to changes than the old one, but saves us sending a sys_schedule message for each process every time we balance the queues. Currently, when processes are moved up a priority queue, their quantum is also renewed, but this can be fiddled with. - do_nice has been removed from kernel. PM answers to get- and setpriority calls, updates it's own nice variable as well as the max_run_queue. This will be refactored once scheduling is moved to a separate server. We will probably have PM update it's local nice value and then send a message to whoever is scheduling the process. - changes to fix an issue in do_fork() where processes could run out of quantum but bypassing the code path that handles it correctly. The future plan is to remove the policy from do_fork() and implement it in userspace too.
2010-03-29 13:07:20 +02:00
scheduling - time quantum in miliseconds - Currently the cpu time quantum is timer-ticks based. Thus the remaining quantum is decreased only if the processes is interrupted by a timer tick. As processes block a lot this typically does not happen for normal user processes. Also the quantum depends on the frequency of the timer. - This change makes the quantum miliseconds based. Internally the miliseconds are translated into cpu cycles. Everytime userspace execution is interrupted by kernel the cycles just consumed by the current process are deducted from the remaining quantum. - It makes the quantum system timer frequency independent. - The boot processes quantum is loosely derived from the tick-based quantas and 60Hz timer and subject to future change - the 64bit arithmetics is a little ugly, will be changes once we have compiler support for 64bit integers (soon)
2010-05-25 10:06:14 +02:00
if ((err = mini_send(p, p->p_scheduler->p_endpoint,
&m_no_quantum, FROM_KERNEL))) {
panic("WARNING: Scheduling: mini_send returned %d\n", err);
Userspace scheduling - cotributed by Bjorn Swift - In this first phase, scheduling is moved from the kernel to the PM server. The next steps are to a) moving scheduling to its own server and b) include useful information in the "out of quantum" message, so that the scheduler can make use of this information. - The kernel process table now keeps record of who is responsible for scheduling each process (p_scheduler). When this pointer is NULL, the process will be scheduled by the kernel. If such a process runs out of quantum, the kernel will simply renew its quantum an requeue it. - When PM loads, it will take over scheduling of all running processes, except system processes, using sys_schedctl(). Essentially, this only results in taking over init. As children inherit a scheduler from their parent, user space programs forked by init will inherit PM (for now) as their scheduler. - Once a process has been assigned a scheduler, and runs out of quantum, its RTS_NO_QUANTUM flag will be set and the process dequeued. The kernel will send a message to the scheduler, on the process' behalf, informing the scheduler that it has run out of quantum. The scheduler can take what ever action it pleases, based on its policy, and then reschedule the process using the sys_schedule() system call. - Balance queues does not work as before. While the old in-kernel function used to renew the quantum of processes in the highest priority run queue, the user-space implementation only acts on processes that have been bumped down to a lower priority queue. This approach reacts slower to changes than the old one, but saves us sending a sys_schedule message for each process every time we balance the queues. Currently, when processes are moved up a priority queue, their quantum is also renewed, but this can be fiddled with. - do_nice has been removed from kernel. PM answers to get- and setpriority calls, updates it's own nice variable as well as the max_run_queue. This will be refactored once scheduling is moved to a separate server. We will probably have PM update it's local nice value and then send a message to whoever is scheduling the process. - changes to fix an issue in do_fork() where processes could run out of quantum but bypassing the code path that handles it correctly. The future plan is to remove the policy from do_fork() and implement it in userspace too.
2010-03-29 13:07:20 +02:00
}
}
scheduling - time quantum in miliseconds - Currently the cpu time quantum is timer-ticks based. Thus the remaining quantum is decreased only if the processes is interrupted by a timer tick. As processes block a lot this typically does not happen for normal user processes. Also the quantum depends on the frequency of the timer. - This change makes the quantum miliseconds based. Internally the miliseconds are translated into cpu cycles. Everytime userspace execution is interrupted by kernel the cycles just consumed by the current process are deducted from the remaining quantum. - It makes the quantum system timer frequency independent. - The boot processes quantum is loosely derived from the tick-based quantas and 60Hz timer and subject to future change - the 64bit arithmetics is a little ugly, will be changes once we have compiler support for 64bit integers (soon)
2010-05-25 10:06:14 +02:00
PUBLIC void proc_no_time(struct proc * p)
Userspace scheduling - cotributed by Bjorn Swift - In this first phase, scheduling is moved from the kernel to the PM server. The next steps are to a) moving scheduling to its own server and b) include useful information in the "out of quantum" message, so that the scheduler can make use of this information. - The kernel process table now keeps record of who is responsible for scheduling each process (p_scheduler). When this pointer is NULL, the process will be scheduled by the kernel. If such a process runs out of quantum, the kernel will simply renew its quantum an requeue it. - When PM loads, it will take over scheduling of all running processes, except system processes, using sys_schedctl(). Essentially, this only results in taking over init. As children inherit a scheduler from their parent, user space programs forked by init will inherit PM (for now) as their scheduler. - Once a process has been assigned a scheduler, and runs out of quantum, its RTS_NO_QUANTUM flag will be set and the process dequeued. The kernel will send a message to the scheduler, on the process' behalf, informing the scheduler that it has run out of quantum. The scheduler can take what ever action it pleases, based on its policy, and then reschedule the process using the sys_schedule() system call. - Balance queues does not work as before. While the old in-kernel function used to renew the quantum of processes in the highest priority run queue, the user-space implementation only acts on processes that have been bumped down to a lower priority queue. This approach reacts slower to changes than the old one, but saves us sending a sys_schedule message for each process every time we balance the queues. Currently, when processes are moved up a priority queue, their quantum is also renewed, but this can be fiddled with. - do_nice has been removed from kernel. PM answers to get- and setpriority calls, updates it's own nice variable as well as the max_run_queue. This will be refactored once scheduling is moved to a separate server. We will probably have PM update it's local nice value and then send a message to whoever is scheduling the process. - changes to fix an issue in do_fork() where processes could run out of quantum but bypassing the code path that handles it correctly. The future plan is to remove the policy from do_fork() and implement it in userspace too.
2010-03-29 13:07:20 +02:00
{
scheduling - time quantum in miliseconds - Currently the cpu time quantum is timer-ticks based. Thus the remaining quantum is decreased only if the processes is interrupted by a timer tick. As processes block a lot this typically does not happen for normal user processes. Also the quantum depends on the frequency of the timer. - This change makes the quantum miliseconds based. Internally the miliseconds are translated into cpu cycles. Everytime userspace execution is interrupted by kernel the cycles just consumed by the current process are deducted from the remaining quantum. - It makes the quantum system timer frequency independent. - The boot processes quantum is loosely derived from the tick-based quantas and 60Hz timer and subject to future change - the 64bit arithmetics is a little ugly, will be changes once we have compiler support for 64bit integers (soon)
2010-05-25 10:06:14 +02:00
if (!proc_kernel_scheduler(p) && priv(p)->s_flags & PREEMPTIBLE) {
/* this dequeues the process */
notify_scheduler(p);
}
else {
/*
* non-preemptible processes only need their quantum to
* be renewed. In fact, they by pass scheduling
*/
p->p_cpu_time_left = ms_2_cpu_time(p->p_quantum_size_ms);
kernel: new DEBUG_RACE option. try to provoke race conditions between processes. it does this by - making all processes interruptible by running out of quantum - giving all processes a single tick of quantum - picking a random runnable process instead of in order, and from a single pool of runnable processes (no priorities) This together with very high HZ values currently provokes some race conditions seen earlier only when running with SMP.
2010-05-08 20:00:03 +02:00
#if DEBUG_RACE
scheduling - time quantum in miliseconds - Currently the cpu time quantum is timer-ticks based. Thus the remaining quantum is decreased only if the processes is interrupted by a timer tick. As processes block a lot this typically does not happen for normal user processes. Also the quantum depends on the frequency of the timer. - This change makes the quantum miliseconds based. Internally the miliseconds are translated into cpu cycles. Everytime userspace execution is interrupted by kernel the cycles just consumed by the current process are deducted from the remaining quantum. - It makes the quantum system timer frequency independent. - The boot processes quantum is loosely derived from the tick-based quantas and 60Hz timer and subject to future change - the 64bit arithmetics is a little ugly, will be changes once we have compiler support for 64bit integers (soon)
2010-05-25 10:06:14 +02:00
RTS_SET(proc_ptr, RTS_PREEMPTED);
RTS_UNSET(proc_ptr, RTS_PREEMPTED);
kernel: new DEBUG_RACE option. try to provoke race conditions between processes. it does this by - making all processes interruptible by running out of quantum - giving all processes a single tick of quantum - picking a random runnable process instead of in order, and from a single pool of runnable processes (no priorities) This together with very high HZ values currently provokes some race conditions seen earlier only when running with SMP.
2010-05-08 20:00:03 +02:00
#endif
Userspace scheduling - cotributed by Bjorn Swift - In this first phase, scheduling is moved from the kernel to the PM server. The next steps are to a) moving scheduling to its own server and b) include useful information in the "out of quantum" message, so that the scheduler can make use of this information. - The kernel process table now keeps record of who is responsible for scheduling each process (p_scheduler). When this pointer is NULL, the process will be scheduled by the kernel. If such a process runs out of quantum, the kernel will simply renew its quantum an requeue it. - When PM loads, it will take over scheduling of all running processes, except system processes, using sys_schedctl(). Essentially, this only results in taking over init. As children inherit a scheduler from their parent, user space programs forked by init will inherit PM (for now) as their scheduler. - Once a process has been assigned a scheduler, and runs out of quantum, its RTS_NO_QUANTUM flag will be set and the process dequeued. The kernel will send a message to the scheduler, on the process' behalf, informing the scheduler that it has run out of quantum. The scheduler can take what ever action it pleases, based on its policy, and then reschedule the process using the sys_schedule() system call. - Balance queues does not work as before. While the old in-kernel function used to renew the quantum of processes in the highest priority run queue, the user-space implementation only acts on processes that have been bumped down to a lower priority queue. This approach reacts slower to changes than the old one, but saves us sending a sys_schedule message for each process every time we balance the queues. Currently, when processes are moved up a priority queue, their quantum is also renewed, but this can be fiddled with. - do_nice has been removed from kernel. PM answers to get- and setpriority calls, updates it's own nice variable as well as the max_run_queue. This will be refactored once scheduling is moved to a separate server. We will probably have PM update it's local nice value and then send a message to whoever is scheduling the process. - changes to fix an issue in do_fork() where processes could run out of quantum but bypassing the code path that handles it correctly. The future plan is to remove the policy from do_fork() and implement it in userspace too.
2010-03-29 13:07:20 +02:00
}
}
Userspace scheduling - exporting stats - contributed by Bjorn Swift - adds process accounting, for example counting the number of messages sent, how often the process was preemted and how much time it spent in the run queue. These statistics, along with the current cpu load, are sent back to the user-space scheduler in the Out Of Quantum message. - the user-space scheduler may choose to make use of these statistics when making scheduling decisions. For isntance the cpu load becomes especially useful when scheduling on multiple cores.
2010-09-19 17:52:12 +02:00
PUBLIC void reset_proc_accounting(struct proc *p)
{
p->p_accounting.preempted = 0;
p->p_accounting.ipc_sync = 0;
p->p_accounting.ipc_async = 0;
p->p_accounting.dequeues = 0;
make_zero64(p->p_accounting.time_in_queue);
make_zero64(p->p_accounting.enter_queue);
}
Lazy FPU - FPU context is stored only if conflict between 2 FPU users or while exporting context of a process to userspace while it is the active user of FPU - FPU has its owner (fpu_owner) which points to the process whose state is currently loaded in FPU - the FPU exception is only turned on when scheduling a process which is not the owner of FPU - FPU state is restored for the process that generated the FPU exception. This process runs immediately without letting scheduler to pick a new process to resolve the FPU conflict asap, to minimize the FPU thrashing and FPU exception hadler execution - faster all non-FPU-exception kernel entries as FPU state is not checked nor saved - removed MF_USED_FPU flag, only MF_FPU_INITIALIZED remains to signal that a process has used FPU in the past
2010-06-07 09:43:17 +02:00
PUBLIC void copr_not_available_handler(void)
{
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
struct proc * p;
SMP - lazy FPU - when a process is migrated to a different CPU it may have an active FPU context in the processor registers. We must save it and migrate it together with the process.
2010-09-15 16:11:25 +02:00
struct proc ** local_fpu_owner;
Lazy FPU - FPU context is stored only if conflict between 2 FPU users or while exporting context of a process to userspace while it is the active user of FPU - FPU has its owner (fpu_owner) which points to the process whose state is currently loaded in FPU - the FPU exception is only turned on when scheduling a process which is not the owner of FPU - FPU state is restored for the process that generated the FPU exception. This process runs immediately without letting scheduler to pick a new process to resolve the FPU conflict asap, to minimize the FPU thrashing and FPU exception hadler execution - faster all non-FPU-exception kernel entries as FPU state is not checked nor saved - removed MF_USED_FPU flag, only MF_FPU_INITIALIZED remains to signal that a process has used FPU in the past
2010-06-07 09:43:17 +02:00
/*
* Disable the FPU exception (both for the kernel and for the process
* once it's scheduled), and initialize or restore the FPU state.
*/
disable_fpu_exception();
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
p = get_cpulocal_var(proc_ptr);
Lazy FPU - FPU context is stored only if conflict between 2 FPU users or while exporting context of a process to userspace while it is the active user of FPU - FPU has its owner (fpu_owner) which points to the process whose state is currently loaded in FPU - the FPU exception is only turned on when scheduling a process which is not the owner of FPU - FPU state is restored for the process that generated the FPU exception. This process runs immediately without letting scheduler to pick a new process to resolve the FPU conflict asap, to minimize the FPU thrashing and FPU exception hadler execution - faster all non-FPU-exception kernel entries as FPU state is not checked nor saved - removed MF_USED_FPU flag, only MF_FPU_INITIALIZED remains to signal that a process has used FPU in the past
2010-06-07 09:43:17 +02:00
/* if FPU is not owned by anyone, do not store anything */
SMP - lazy FPU - when a process is migrated to a different CPU it may have an active FPU context in the processor registers. We must save it and migrate it together with the process.
2010-09-15 16:11:25 +02:00
local_fpu_owner = get_cpulocal_var_ptr(fpu_owner);
if (*local_fpu_owner != NULL) {
assert(*local_fpu_owner != p);
save_local_fpu(*local_fpu_owner);
Lazy FPU - FPU context is stored only if conflict between 2 FPU users or while exporting context of a process to userspace while it is the active user of FPU - FPU has its owner (fpu_owner) which points to the process whose state is currently loaded in FPU - the FPU exception is only turned on when scheduling a process which is not the owner of FPU - FPU state is restored for the process that generated the FPU exception. This process runs immediately without letting scheduler to pick a new process to resolve the FPU conflict asap, to minimize the FPU thrashing and FPU exception hadler execution - faster all non-FPU-exception kernel entries as FPU state is not checked nor saved - removed MF_USED_FPU flag, only MF_FPU_INITIALIZED remains to signal that a process has used FPU in the past
2010-06-07 09:43:17 +02:00
}
/*
* restore the current process' state and let it run again, do not
* schedule!
*/
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
restore_fpu(p);
SMP - lazy FPU - when a process is migrated to a different CPU it may have an active FPU context in the processor registers. We must save it and migrate it together with the process.
2010-09-15 16:11:25 +02:00
*local_fpu_owner = p;
Lazy FPU - FPU context is stored only if conflict between 2 FPU users or while exporting context of a process to userspace while it is the active user of FPU - FPU has its owner (fpu_owner) which points to the process whose state is currently loaded in FPU - the FPU exception is only turned on when scheduling a process which is not the owner of FPU - FPU state is restored for the process that generated the FPU exception. This process runs immediately without letting scheduler to pick a new process to resolve the FPU conflict asap, to minimize the FPU thrashing and FPU exception hadler execution - faster all non-FPU-exception kernel entries as FPU state is not checked nor saved - removed MF_USED_FPU flag, only MF_FPU_INITIALIZED remains to signal that a process has used FPU in the past
2010-06-07 09:43:17 +02:00
context_stop(proc_addr(KERNEL));
SMP - Cpu local variables - most global variables carry information which is specific to the local CPU and each CPU must have its own copy - cpu local variable must be declared in cpulocal.h between DECLARE_CPULOCAL_START and DECLARE_CPULOCAL_END markers using DECLARE_CPULOCAL macro - to access the cpu local data the provided macros must be used get_cpu_var(cpu, name) get_cpu_var_ptr(cpu, name) get_cpulocal_var(name) get_cpulocal_var_ptr(name) - using this macros makes future changes in the implementation possible - switching to ELF will make the declaration of cpu local data much simpler, e.g. CPULOCAL int blah; anywhere in the kernel source code
2010-09-15 16:09:46 +02:00
restore_user_context(p);
Lazy FPU - FPU context is stored only if conflict between 2 FPU users or while exporting context of a process to userspace while it is the active user of FPU - FPU has its owner (fpu_owner) which points to the process whose state is currently loaded in FPU - the FPU exception is only turned on when scheduling a process which is not the owner of FPU - FPU state is restored for the process that generated the FPU exception. This process runs immediately without letting scheduler to pick a new process to resolve the FPU conflict asap, to minimize the FPU thrashing and FPU exception hadler execution - faster all non-FPU-exception kernel entries as FPU state is not checked nor saved - removed MF_USED_FPU flag, only MF_FPU_INITIALIZED remains to signal that a process has used FPU in the past
2010-06-07 09:43:17 +02:00
NOT_REACHABLE;
}
SMP - lazy FPU - when a process is migrated to a different CPU it may have an active FPU context in the processor registers. We must save it and migrate it together with the process.
2010-09-15 16:11:25 +02:00
PUBLIC void release_fpu(struct proc * p) {
struct proc ** fpu_owner_ptr;
fpu_owner_ptr = get_cpu_var_ptr(p->p_cpu, fpu_owner);
if (*fpu_owner_ptr == p)
*fpu_owner_ptr = NULL;
Lazy FPU - FPU context is stored only if conflict between 2 FPU users or while exporting context of a process to userspace while it is the active user of FPU - FPU has its owner (fpu_owner) which points to the process whose state is currently loaded in FPU - the FPU exception is only turned on when scheduling a process which is not the owner of FPU - FPU state is restored for the process that generated the FPU exception. This process runs immediately without letting scheduler to pick a new process to resolve the FPU conflict asap, to minimize the FPU thrashing and FPU exception hadler execution - faster all non-FPU-exception kernel entries as FPU state is not checked nor saved - removed MF_USED_FPU flag, only MF_FPU_INITIALIZED remains to signal that a process has used FPU in the past
2010-06-07 09:43:17 +02:00
}
Reference in a new issue Copy permalink