minix/kernel/main.c
2006-03-10 16:10:05 +00:00

258 lines
9.9 KiB
C
Executable file

/* This file contains the main program of MINIX as well as its shutdown code.
* The routine main() initializes the system and starts the ball rolling by
* setting up the process table, interrupt vectors, and scheduling each task
* to run to initialize itself.
* The routine shutdown() does the opposite and brings down MINIX.
*
* The entries into this file are:
* main: MINIX main program
* prepare_shutdown: prepare to take MINIX down
*
* Changes:
* Nov 24, 2004 simplified main() with system image (Jorrit N. Herder)
* Aug 20, 2004 new prepare_shutdown() and shutdown() (Jorrit N. Herder)
*/
#include "kernel.h"
#include <signal.h>
#include <string.h>
#include <unistd.h>
#include <a.out.h>
#include <minix/callnr.h>
#include <minix/com.h>
#include <minix/endpoint.h>
#include "proc.h"
/* Prototype declarations for PRIVATE functions. */
FORWARD _PROTOTYPE( void announce, (void));
FORWARD _PROTOTYPE( void shutdown, (timer_t *tp));
/*===========================================================================*
* main *
*===========================================================================*/
PUBLIC void main()
{
/* Start the ball rolling. */
struct boot_image *ip; /* boot image pointer */
register struct proc *rp; /* process pointer */
register struct priv *sp; /* privilege structure pointer */
register int i, s;
int hdrindex; /* index to array of a.out headers */
phys_clicks text_base;
vir_clicks text_clicks, data_clicks;
reg_t ktsb; /* kernel task stack base */
struct exec e_hdr; /* for a copy of an a.out header */
/* Initialize the interrupt controller. */
intr_init(1);
/* 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 = SLOT_FREE; /* initialize free slot */
rp->p_nr = i; /* proc number from ptr */
rp->p_endpoint = _ENDPOINT(0, rp->p_nr); /* generation no. 0 */
(pproc_addr + NR_TASKS)[i] = rp; /* proc ptr from number */
}
for (sp = BEG_PRIV_ADDR, i = 0; sp < END_PRIV_ADDR; ++sp, ++i) {
sp->s_proc_nr = NONE; /* initialize as free */
sp->s_id = i; /* priv structure index */
ppriv_addr[i] = sp; /* priv ptr from number */
}
/* Set up proc table entries for processes in boot image. The stacks of the
* kernel tasks are initialized to an array in data space. The stacks
* of the servers have been added to the data segment by the monitor, so
* the stack pointer is set to the end of the data segment. All the
* processes are in low memory on the 8086. On the 386 only the kernel
* is in low memory, the rest is loaded in extended memory.
*/
/* Task stacks. */
ktsb = (reg_t) t_stack;
for (i=0; i < NR_BOOT_PROCS; ++i) {
ip = &image[i]; /* process' attributes */
rp = proc_addr(ip->proc_nr); /* get process pointer */
ip->endpoint = rp->p_endpoint; /* ipc endpoint */
rp->p_max_priority = ip->priority; /* max scheduling priority */
rp->p_priority = ip->priority; /* current priority */
rp->p_quantum_size = ip->quantum; /* quantum size in ticks */
rp->p_ticks_left = ip->quantum; /* current credit */
strncpy(rp->p_name, ip->proc_name, P_NAME_LEN); /* set process name */
(void) get_priv(rp, (ip->flags & SYS_PROC)); /* assign structure */
priv(rp)->s_flags = ip->flags; /* process flags */
priv(rp)->s_trap_mask = ip->trap_mask; /* allowed traps */
priv(rp)->s_call_mask = ip->call_mask; /* kernel call mask */
priv(rp)->s_ipc_to.chunk[0] = ip->ipc_to; /* restrict targets */
if (iskerneln(proc_nr(rp))) { /* part of the kernel? */
if (ip->stksize > 0) { /* HARDWARE stack size is 0 */
rp->p_priv->s_stack_guard = (reg_t *) ktsb;
*rp->p_priv->s_stack_guard = STACK_GUARD;
}
ktsb += ip->stksize; /* point to high end of stack */
rp->p_reg.sp = ktsb; /* this task's initial stack ptr */
text_base = kinfo.code_base >> CLICK_SHIFT;
/* processes that are in the kernel */
hdrindex = 0; /* all use the first a.out header */
} else {
hdrindex = 1 + i-NR_TASKS; /* servers, drivers, INIT */
}
/* The bootstrap loader created an array of the a.out headers at
* absolute address 'aout'. Get one element to e_hdr.
*/
phys_copy(aout + hdrindex * A_MINHDR, vir2phys(&e_hdr),
(phys_bytes) A_MINHDR);
/* Convert addresses to clicks and build process memory map */
text_base = e_hdr.a_syms >> CLICK_SHIFT;
text_clicks = (e_hdr.a_text + CLICK_SIZE-1) >> CLICK_SHIFT;
if (!(e_hdr.a_flags & A_SEP)) text_clicks = 0; /* common I&D */
data_clicks = (e_hdr.a_total + CLICK_SIZE-1) >> CLICK_SHIFT;
rp->p_memmap[T].mem_phys = text_base;
rp->p_memmap[T].mem_len = text_clicks;
rp->p_memmap[D].mem_phys = text_base + text_clicks;
rp->p_memmap[D].mem_len = data_clicks;
rp->p_memmap[S].mem_phys = text_base + text_clicks + data_clicks;
rp->p_memmap[S].mem_vir = data_clicks; /* empty - stack is in data */
/* Set initial register values. The processor status word for tasks
* is different from that of other processes because tasks can
* access I/O; this is not allowed to less-privileged processes
*/
rp->p_reg.pc = (reg_t) ip->initial_pc;
rp->p_reg.psw = (iskernelp(rp)) ? INIT_TASK_PSW : INIT_PSW;
/* Initialize the server stack pointer. Take it down one word
* to give crtso.s something to use as "argc".
*/
if (isusern(proc_nr(rp))) { /* user-space process? */
rp->p_reg.sp = (rp->p_memmap[S].mem_vir +
rp->p_memmap[S].mem_len) << CLICK_SHIFT;
rp->p_reg.sp -= sizeof(reg_t);
}
/* Set ready. The HARDWARE task is never ready. */
if (rp->p_nr != HARDWARE) {
rp->p_rts_flags = 0; /* runnable if no flags */
lock_enqueue(rp); /* add to scheduling queues */
} else {
rp->p_rts_flags = NO_MAP; /* prevent from running */
}
/* Code and data segments must be allocated in protected mode. */
alloc_segments(rp);
}
/* Special compensation for IDLE - don't let it count in the load average. */
kloadinfo.procs_enqueued--;
#if ENABLE_BOOTDEV
/* Expect an image of the boot device to be loaded into memory as well.
* The boot device is the last module that is loaded into memory, and,
* for example, can contain the root FS (useful for embedded systems).
*/
hdrindex ++;
phys_copy(aout + hdrindex * A_MINHDR,vir2phys(&e_hdr),(phys_bytes) A_MINHDR);
if (e_hdr.a_flags & A_IMG) {
kinfo.bootdev_base = e_hdr.a_syms;
kinfo.bootdev_size = e_hdr.a_data;
}
#endif
/* MINIX is now ready. All boot image processes are on the ready queue.
* Return to the assembly code to start running the current process.
*/
bill_ptr = proc_addr(IDLE); /* it has to point somewhere */
announce(); /* print MINIX startup banner */
restart();
}
/*===========================================================================*
* announce *
*===========================================================================*/
PRIVATE void announce(void)
{
/* Display the MINIX startup banner. */
kprintf("\nMINIX %s.%s. "
"Copyright 2006, Vrije Universiteit, Amsterdam, The Netherlands\n",
OS_RELEASE, OS_VERSION);
#if (CHIP == INTEL)
/* Real mode, or 16/32-bit protected mode? */
kprintf("Executing in %s mode.\n\n",
machine.protected ? "32-bit protected" : "real");
#endif
}
/*===========================================================================*
* prepare_shutdown *
*===========================================================================*/
PUBLIC void prepare_shutdown(how)
int how;
{
/* This function prepares to shutdown MINIX. */
static timer_t shutdown_timer;
register struct proc *rp;
message m;
/* Send a signal to all system processes that are still alive to inform
* them that the MINIX kernel is shutting down. A proper shutdown sequence
* should be implemented by a user-space server. This mechanism is useful
* as a backup in case of system panics, so that system processes can still
* run their shutdown code, e.g, to synchronize the FS or to let the TTY
* switch to the first console.
*/
#if DEAD_CODE
kprintf("Sending SIGKSTOP to system processes ...\n");
for (rp=BEG_PROC_ADDR; rp<END_PROC_ADDR; rp++) {
if (!isemptyp(rp) && (priv(rp)->s_flags & SYS_PROC) && !iskernelp(rp))
send_sig(proc_nr(rp), SIGKSTOP);
}
#endif
/* Continue after 1 second, to give processes a chance to get scheduled to
* do shutdown work. Set a watchog timer to call shutdown(). The timer
* argument passes the shutdown status.
*/
kprintf("MINIX will now be shut down ...\n");
tmr_arg(&shutdown_timer)->ta_int = how;
set_timer(&shutdown_timer, get_uptime() + HZ, shutdown);
}
/*===========================================================================*
* shutdown *
*===========================================================================*/
PRIVATE void shutdown(tp)
timer_t *tp;
{
/* This function is called from prepare_shutdown or stop_sequence to bring
* down MINIX. How to shutdown is in the argument: RBT_HALT (return to the
* monitor), RBT_MONITOR (execute given code), RBT_RESET (hard reset).
*/
int how = tmr_arg(tp)->ta_int;
u16_t magic;
/* Now mask all interrupts, including the clock, and stop the clock. */
outb(INT_CTLMASK, ~0);
clock_stop();
if (mon_return && how != RBT_RESET) {
/* Reinitialize the interrupt controllers to the BIOS defaults. */
intr_init(0);
outb(INT_CTLMASK, 0);
outb(INT2_CTLMASK, 0);
/* Return to the boot monitor. Set the program if not already done. */
if (how != RBT_MONITOR) phys_copy(vir2phys(""), kinfo.params_base, 1);
level0(monitor);
}
/* Reset the system by jumping to the reset address (real mode), or by
* forcing a processor shutdown (protected mode). First stop the BIOS
* memory test by setting a soft reset flag.
*/
magic = STOP_MEM_CHECK;
phys_copy(vir2phys(&magic), SOFT_RESET_FLAG_ADDR, SOFT_RESET_FLAG_SIZE);
level0(reset);
}