2010-09-15 16:09:52 +02:00
|
|
|
/* This file contains essentially the MP handling code of the Minix kernel.
|
|
|
|
|
*
|
|
|
|
|
* Changes:
|
|
|
|
|
* Apr 1, 2008 Added SMP support.
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
#define _SMP
|
|
|
|
|
|
|
|
|
|
#include "kernel/kernel.h"
|
|
|
|
|
#include "arch_proto.h"
|
|
|
|
|
#include <unistd.h>
|
2012-07-13 00:54:27 +02:00
|
|
|
#include <assert.h>
|
2011-12-15 10:56:08 +01:00
|
|
|
#include <stdlib.h>
|
2012-08-14 18:38:03 +02:00
|
|
|
#include <string.h>
|
2012-07-13 00:54:27 +02:00
|
|
|
#include <machine/archtypes.h>
|
|
|
|
|
#include <archconst.h>
|
2010-09-15 16:09:52 +02:00
|
|
|
#include <machine/cmos.h>
|
|
|
|
|
#include <machine/bios.h>
|
|
|
|
|
#include <minix/portio.h>
|
|
|
|
|
|
|
|
|
|
#include "kernel/spinlock.h"
|
|
|
|
|
#include "kernel/smp.h"
|
|
|
|
|
#include "apic.h"
|
|
|
|
|
#include "acpi.h"
|
2011-12-15 10:56:08 +01:00
|
|
|
#include "clock.h"
|
2010-09-15 16:09:52 +02:00
|
|
|
|
|
|
|
|
#include "glo.h"
|
|
|
|
|
|
2012-03-24 16:16:34 +01:00
|
|
|
void trampoline(void);
|
2010-09-15 16:09:52 +02:00
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* arguments for trampoline. We need to pass the logical cpu id, gdt and idt.
|
|
|
|
|
* They have to be in location which is reachable using absolute addressing in
|
|
|
|
|
* 16-bit mode
|
|
|
|
|
*/
|
2012-07-13 00:54:27 +02:00
|
|
|
extern volatile u32_t __ap_id, __ap_pt;
|
No more intel/minix segments.
This commit removes all traces of Minix segments (the text/data/stack
memory map abstraction in the kernel) and significance of Intel segments
(hardware segments like CS, DS that add offsets to all addressing before
page table translation). This ultimately simplifies the memory layout
and addressing and makes the same layout possible on non-Intel
architectures.
There are only two types of addresses in the world now: virtual
and physical; even the kernel and processes have the same virtual
address space. Kernel and user processes can be distinguished at a
glance as processes won't use 0xF0000000 and above.
No static pre-allocated memory sizes exist any more.
Changes to booting:
. The pre_init.c leaves the kernel and modules exactly as
they were left by the bootloader in physical memory
. The kernel starts running using physical addressing,
loaded at a fixed location given in its linker script by the
bootloader. All code and data in this phase are linked to
this fixed low location.
. It makes a bootstrap pagetable to map itself to a
fixed high location (also in linker script) and jumps to
the high address. All code and data then use this high addressing.
. All code/data symbols linked at the low addresses is prefixed by
an objcopy step with __k_unpaged_*, so that that code cannot
reference highly-linked symbols (which aren't valid yet) or vice
versa (symbols that aren't valid any more).
. The two addressing modes are separated in the linker script by
collecting the unpaged_*.o objects and linking them with low
addresses, and linking the rest high. Some objects are linked
twice, once low and once high.
. The bootstrap phase passes a lot of information (e.g. free memory
list, physical location of the modules, etc.) using the kinfo
struct.
. After this bootstrap the low-linked part is freed.
. The kernel maps in VM into the bootstrap page table so that VM can
begin executing. Its first job is to make page tables for all other
boot processes. So VM runs before RS, and RS gets a fully dynamic,
VM-managed address space. VM gets its privilege info from RS as usual
but that happens after RS starts running.
. Both the kernel loading VM and VM organizing boot processes happen
using the libexec logic. This removes the last reason for VM to
still know much about exec() and vm/exec.c is gone.
Further Implementation:
. All segments are based at 0 and have a 4 GB limit.
. The kernel is mapped in at the top of the virtual address
space so as not to constrain the user processes.
. Processes do not use segments from the LDT at all; there are
no segments in the LDT any more, so no LLDT is needed.
. The Minix segments T/D/S are gone and so none of the
user-space or in-kernel copy functions use them. The copy
functions use a process endpoint of NONE to realize it's
a physical address, virtual otherwise.
. The umap call only makes sense to translate a virtual address
to a physical address now.
. Segments-related calls like newmap and alloc_segments are gone.
. All segments-related translation in VM is gone (vir2map etc).
. Initialization in VM is simpler as no moving around is necessary.
. VM and all other boot processes can be linked wherever they wish
and will be mapped in at the right location by the kernel and VM
respectively.
Other changes:
. The multiboot code is less special: it does not use mb_print
for its diagnostics any more but uses printf() as normal, saving
the output into the diagnostics buffer, only printing to the
screen using the direct print functions if a panic() occurs.
. The multiboot code uses the flexible 'free memory map list'
style to receive the list of free memory if available.
. The kernel determines the memory layout of the processes to
a degree: it tells VM where the kernel starts and ends and
where the kernel wants the top of the process to be. VM then
uses this entire range, i.e. the stack is right at the top,
and mmap()ped bits of memory are placed below that downwards,
and the break grows upwards.
Other Consequences:
. Every process gets its own page table as address spaces
can't be separated any more by segments.
. As all segments are 0-based, there is no distinction between
virtual and linear addresses, nor between userspace and
kernel addresses.
. Less work is done when context switching, leading to a net
performance increase. (8% faster on my machine for 'make servers'.)
. The layout and configuration of the GDT makes sysenter and syscall
possible.
2012-05-07 16:03:35 +02:00
|
|
|
extern volatile struct desctableptr_s __ap_gdt, __ap_idt;
|
2012-07-13 00:54:27 +02:00
|
|
|
extern u32_t __ap_gdt_tab, __ap_idt_tab;
|
2010-09-15 16:10:00 +02:00
|
|
|
extern void * __trampoline_end;
|
2010-09-15 16:09:52 +02:00
|
|
|
|
|
|
|
|
extern u32_t busclock[CONFIG_MAX_CPUS];
|
|
|
|
|
extern int panicking;
|
|
|
|
|
|
2010-09-15 16:11:21 +02:00
|
|
|
static int volatile ap_cpu_ready;
|
|
|
|
|
static int volatile cpu_down;
|
2010-09-15 16:09:52 +02:00
|
|
|
|
|
|
|
|
/* there can be at most 255 local APIC ids, each fits in 8 bits */
|
2012-03-25 20:25:53 +02:00
|
|
|
static unsigned char apicid2cpuid[255];
|
|
|
|
|
unsigned char cpuid2apicid[CONFIG_MAX_CPUS];
|
2010-09-15 16:09:52 +02:00
|
|
|
|
|
|
|
|
SPINLOCK_DEFINE(smp_cpu_lock)
|
|
|
|
|
SPINLOCK_DEFINE(dispq_lock)
|
|
|
|
|
|
2012-03-25 20:25:53 +02:00
|
|
|
static void smp_reinit_vars(void);
|
2010-09-15 16:09:52 +02:00
|
|
|
|
2012-07-13 00:54:27 +02:00
|
|
|
/* These are initialized in protect.c */
|
|
|
|
|
extern struct segdesc_s gdt[GDT_SIZE];
|
|
|
|
|
extern struct gatedesc_s idt[IDT_SIZE];
|
|
|
|
|
extern struct tss_s tss[CONFIG_MAX_CPUS];
|
|
|
|
|
extern int prot_init_done; /* Indicates they are ready */
|
|
|
|
|
|
|
|
|
|
static phys_bytes trampoline_base;
|
|
|
|
|
|
|
|
|
|
static u32_t ap_lin_addr(void *vaddr)
|
|
|
|
|
{
|
|
|
|
|
assert(trampoline_base);
|
|
|
|
|
return (u32_t) vaddr - (u32_t) &trampoline + trampoline_base;
|
|
|
|
|
}
|
|
|
|
|
|
2010-09-15 16:10:00 +02:00
|
|
|
/*
|
|
|
|
|
* copies the 16-bit AP trampoline code to the first 1M of memory
|
|
|
|
|
*/
|
2012-07-13 00:54:27 +02:00
|
|
|
void copy_trampoline(void)
|
2010-09-15 16:10:00 +02:00
|
|
|
{
|
2012-07-13 00:54:27 +02:00
|
|
|
unsigned tramp_size, tramp_start = (unsigned)&trampoline;;
|
|
|
|
|
|
|
|
|
|
/* The trampoline code/data is made to be page-aligned. */
|
|
|
|
|
assert(!(tramp_start % I386_PAGE_SIZE));
|
2010-09-15 16:10:00 +02:00
|
|
|
|
2012-07-13 00:54:27 +02:00
|
|
|
tramp_size = (unsigned) &__trampoline_end - tramp_start;
|
|
|
|
|
trampoline_base = alloc_lowest(&kinfo, tramp_size);
|
2010-09-15 16:10:00 +02:00
|
|
|
|
2012-07-13 00:54:27 +02:00
|
|
|
/* The memory allocator finds the lowest available memory..
|
|
|
|
|
* Verify it's low enough
|
|
|
|
|
*/
|
|
|
|
|
assert(trampoline_base + tramp_size < (1 << 20));
|
|
|
|
|
|
|
|
|
|
/* prepare gdt and idt for the new cpus; make copies
|
|
|
|
|
* of both the tables and the descriptors of them
|
|
|
|
|
* in their boot addressing environment.
|
|
|
|
|
*/
|
|
|
|
|
assert(prot_init_done);
|
|
|
|
|
memcpy(&__ap_gdt_tab, gdt, sizeof(gdt));
|
|
|
|
|
memcpy(&__ap_idt_tab, gdt, sizeof(idt));
|
|
|
|
|
__ap_gdt.base = ap_lin_addr(&__ap_gdt_tab);
|
|
|
|
|
__ap_gdt.limit = sizeof(gdt)-1;
|
|
|
|
|
__ap_idt.base = ap_lin_addr(&__ap_idt_tab);
|
|
|
|
|
__ap_idt.limit = sizeof(idt)-1;
|
|
|
|
|
|
2012-08-14 18:38:03 +02:00
|
|
|
phys_copy((phys_bytes) trampoline, trampoline_base, tramp_size);
|
2010-09-15 16:10:00 +02:00
|
|
|
}
|
|
|
|
|
|
2012-07-13 00:54:27 +02:00
|
|
|
extern int booting_cpu; /* tell protect.c what to do */
|
No more intel/minix segments.
This commit removes all traces of Minix segments (the text/data/stack
memory map abstraction in the kernel) and significance of Intel segments
(hardware segments like CS, DS that add offsets to all addressing before
page table translation). This ultimately simplifies the memory layout
and addressing and makes the same layout possible on non-Intel
architectures.
There are only two types of addresses in the world now: virtual
and physical; even the kernel and processes have the same virtual
address space. Kernel and user processes can be distinguished at a
glance as processes won't use 0xF0000000 and above.
No static pre-allocated memory sizes exist any more.
Changes to booting:
. The pre_init.c leaves the kernel and modules exactly as
they were left by the bootloader in physical memory
. The kernel starts running using physical addressing,
loaded at a fixed location given in its linker script by the
bootloader. All code and data in this phase are linked to
this fixed low location.
. It makes a bootstrap pagetable to map itself to a
fixed high location (also in linker script) and jumps to
the high address. All code and data then use this high addressing.
. All code/data symbols linked at the low addresses is prefixed by
an objcopy step with __k_unpaged_*, so that that code cannot
reference highly-linked symbols (which aren't valid yet) or vice
versa (symbols that aren't valid any more).
. The two addressing modes are separated in the linker script by
collecting the unpaged_*.o objects and linking them with low
addresses, and linking the rest high. Some objects are linked
twice, once low and once high.
. The bootstrap phase passes a lot of information (e.g. free memory
list, physical location of the modules, etc.) using the kinfo
struct.
. After this bootstrap the low-linked part is freed.
. The kernel maps in VM into the bootstrap page table so that VM can
begin executing. Its first job is to make page tables for all other
boot processes. So VM runs before RS, and RS gets a fully dynamic,
VM-managed address space. VM gets its privilege info from RS as usual
but that happens after RS starts running.
. Both the kernel loading VM and VM organizing boot processes happen
using the libexec logic. This removes the last reason for VM to
still know much about exec() and vm/exec.c is gone.
Further Implementation:
. All segments are based at 0 and have a 4 GB limit.
. The kernel is mapped in at the top of the virtual address
space so as not to constrain the user processes.
. Processes do not use segments from the LDT at all; there are
no segments in the LDT any more, so no LLDT is needed.
. The Minix segments T/D/S are gone and so none of the
user-space or in-kernel copy functions use them. The copy
functions use a process endpoint of NONE to realize it's
a physical address, virtual otherwise.
. The umap call only makes sense to translate a virtual address
to a physical address now.
. Segments-related calls like newmap and alloc_segments are gone.
. All segments-related translation in VM is gone (vir2map etc).
. Initialization in VM is simpler as no moving around is necessary.
. VM and all other boot processes can be linked wherever they wish
and will be mapped in at the right location by the kernel and VM
respectively.
Other changes:
. The multiboot code is less special: it does not use mb_print
for its diagnostics any more but uses printf() as normal, saving
the output into the diagnostics buffer, only printing to the
screen using the direct print functions if a panic() occurs.
. The multiboot code uses the flexible 'free memory map list'
style to receive the list of free memory if available.
. The kernel determines the memory layout of the processes to
a degree: it tells VM where the kernel starts and ends and
where the kernel wants the top of the process to be. VM then
uses this entire range, i.e. the stack is right at the top,
and mmap()ped bits of memory are placed below that downwards,
and the break grows upwards.
Other Consequences:
. Every process gets its own page table as address spaces
can't be separated any more by segments.
. As all segments are 0-based, there is no distinction between
virtual and linear addresses, nor between userspace and
kernel addresses.
. Less work is done when context switching, leading to a net
performance increase. (8% faster on my machine for 'make servers'.)
. The layout and configuration of the GDT makes sysenter and syscall
possible.
2012-05-07 16:03:35 +02:00
|
|
|
|
2012-03-25 20:25:53 +02:00
|
|
|
static void smp_start_aps(void)
|
2010-09-15 16:09:52 +02:00
|
|
|
{
|
|
|
|
|
unsigned cpu;
|
2012-08-14 18:38:03 +02:00
|
|
|
u32_t biosresetvector;
|
2012-07-13 00:54:27 +02:00
|
|
|
phys_bytes __ap_id_phys;
|
|
|
|
|
struct proc *bootstrap_pt = get_cpulocal_var(ptproc);
|
2010-09-15 16:09:52 +02:00
|
|
|
|
|
|
|
|
/* TODO hack around the alignment problem */
|
|
|
|
|
|
2012-08-14 18:38:03 +02:00
|
|
|
phys_copy(0x467, (phys_bytes) &biosresetvector, sizeof(u32_t));
|
2010-09-15 16:09:52 +02:00
|
|
|
|
|
|
|
|
/* set the bios shutdown code to 0xA */
|
|
|
|
|
outb(RTC_INDEX, 0xF);
|
|
|
|
|
outb(RTC_IO, 0xA);
|
|
|
|
|
|
2012-07-13 00:54:27 +02:00
|
|
|
assert(bootstrap_pt);
|
|
|
|
|
assert(bootstrap_pt->p_seg.p_cr3);
|
|
|
|
|
__ap_pt = bootstrap_pt->p_seg.p_cr3;
|
|
|
|
|
assert(__ap_pt);
|
2010-09-15 16:09:52 +02:00
|
|
|
|
2012-07-13 00:54:27 +02:00
|
|
|
copy_trampoline();
|
|
|
|
|
|
|
|
|
|
/* New locations for cpu id, pagetable root */
|
2010-09-15 16:10:00 +02:00
|
|
|
__ap_id_phys = trampoline_base +
|
|
|
|
|
(phys_bytes) &__ap_id - (phys_bytes)&trampoline;
|
|
|
|
|
|
|
|
|
|
/* setup the warm reset vector */
|
2012-08-14 18:38:03 +02:00
|
|
|
phys_copy((phys_bytes) &trampoline_base, 0x467, sizeof(u32_t));
|
2010-09-15 16:10:00 +02:00
|
|
|
|
2010-09-15 16:09:52 +02:00
|
|
|
/* okay, we're ready to go. boot all of the ap's now. we loop through
|
|
|
|
|
* using the processor's apic id values.
|
|
|
|
|
*/
|
|
|
|
|
for (cpu = 0; cpu < ncpus; cpu++) {
|
|
|
|
|
ap_cpu_ready = -1;
|
|
|
|
|
/* Don't send INIT/SIPI to boot cpu. */
|
|
|
|
|
if((apicid() == cpuid2apicid[cpu]) &&
|
|
|
|
|
(apicid() == bsp_lapic_id)) {
|
|
|
|
|
continue;
|
|
|
|
|
}
|
|
|
|
|
|
2012-07-13 00:54:27 +02:00
|
|
|
__ap_id = booting_cpu = cpu;
|
2012-08-14 18:38:03 +02:00
|
|
|
phys_copy((phys_bytes) &__ap_id, __ap_id_phys, sizeof(__ap_id));
|
2010-09-15 16:10:00 +02:00
|
|
|
mfence();
|
2010-09-15 16:09:52 +02:00
|
|
|
if (apic_send_init_ipi(cpu, trampoline_base) ||
|
|
|
|
|
apic_send_startup_ipi(cpu, trampoline_base)) {
|
|
|
|
|
printf("WARNING cannot boot cpu %d\n", cpu);
|
|
|
|
|
continue;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* wait for 5 secs for the processors to boot */
|
|
|
|
|
lapic_set_timer_one_shot(5000000);
|
|
|
|
|
|
|
|
|
|
while (lapic_read(LAPIC_TIMER_CCR)) {
|
|
|
|
|
if (ap_cpu_ready == cpu) {
|
|
|
|
|
cpu_set_flag(cpu, CPU_IS_READY);
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
if (ap_cpu_ready == -1) {
|
|
|
|
|
printf("WARNING : CPU %d didn't boot\n", cpu);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2012-08-14 18:38:03 +02:00
|
|
|
phys_copy((phys_bytes) &biosresetvector, 0x467, sizeof(u32_t));
|
2010-09-15 16:09:52 +02:00
|
|
|
|
|
|
|
|
outb(RTC_INDEX, 0xF);
|
|
|
|
|
outb(RTC_IO, 0);
|
|
|
|
|
|
|
|
|
|
bsp_finish_booting();
|
|
|
|
|
NOT_REACHABLE;
|
|
|
|
|
}
|
|
|
|
|
|
2012-03-25 20:25:53 +02:00
|
|
|
void smp_halt_cpu (void)
|
2010-09-15 16:09:52 +02:00
|
|
|
{
|
|
|
|
|
NOT_IMPLEMENTED;
|
|
|
|
|
}
|
|
|
|
|
|
2012-03-25 20:25:53 +02:00
|
|
|
void smp_shutdown_aps(void)
|
2010-09-15 16:09:52 +02:00
|
|
|
{
|
2010-09-15 16:10:54 +02:00
|
|
|
unsigned cpu;
|
2011-12-15 10:56:08 +01:00
|
|
|
|
2010-09-15 16:10:30 +02:00
|
|
|
if (ncpus == 1)
|
|
|
|
|
goto exit_shutdown_aps;
|
2011-12-15 10:56:08 +01:00
|
|
|
|
2010-09-15 16:10:54 +02:00
|
|
|
/* we must let the other cpus enter the kernel mode */
|
|
|
|
|
BKL_UNLOCK();
|
2010-09-15 16:10:30 +02:00
|
|
|
|
|
|
|
|
for (cpu = 0; cpu < ncpus; cpu++) {
|
2010-09-15 16:10:54 +02:00
|
|
|
if (cpu == cpuid)
|
2010-09-15 16:10:30 +02:00
|
|
|
continue;
|
2010-09-15 16:11:21 +02:00
|
|
|
if (!cpu_test_flag(cpu, CPU_IS_READY)) {
|
|
|
|
|
printf("CPU %d didn't boot\n", cpu);
|
|
|
|
|
continue;
|
|
|
|
|
}
|
|
|
|
|
|
2010-09-15 16:10:54 +02:00
|
|
|
cpu_down = -1;
|
|
|
|
|
barrier();
|
2010-09-15 16:10:30 +02:00
|
|
|
apic_send_ipi(APIC_SMP_CPU_HALT_VECTOR, cpu, APIC_IPI_DEST);
|
2010-09-15 16:10:54 +02:00
|
|
|
/* wait for the cpu to be down */
|
|
|
|
|
while (cpu_down != cpu);
|
|
|
|
|
printf("CPU %d is down\n", cpu);
|
|
|
|
|
cpu_clear_flag(cpu, CPU_IS_READY);
|
2010-09-15 16:10:30 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
exit_shutdown_aps:
|
|
|
|
|
ioapic_disable_all();
|
|
|
|
|
|
|
|
|
|
lapic_disable();
|
|
|
|
|
|
|
|
|
|
ncpus = 1; /* hopefully !!! */
|
|
|
|
|
lapic_addr = lapic_eoi_addr = 0;
|
|
|
|
|
return;
|
2010-09-15 16:09:52 +02:00
|
|
|
}
|
|
|
|
|
|
2012-03-25 20:25:53 +02:00
|
|
|
static void ap_finish_booting(void)
|
2010-09-15 16:09:52 +02:00
|
|
|
{
|
|
|
|
|
unsigned cpu = cpuid;
|
|
|
|
|
|
|
|
|
|
/* inform the world of our presence. */
|
|
|
|
|
ap_cpu_ready = cpu;
|
|
|
|
|
|
2010-09-15 16:10:12 +02:00
|
|
|
/*
|
|
|
|
|
* Finish processor initialisation. CPUs must be excluded from running.
|
|
|
|
|
* lapic timer calibration locks and unlocks the BKL because of the
|
|
|
|
|
* nested interrupts used for calibration. Therefore BKL is not good
|
|
|
|
|
* enough, the boot_lock must be held.
|
|
|
|
|
*/
|
|
|
|
|
spinlock_lock(&boot_lock);
|
|
|
|
|
BKL_LOCK();
|
|
|
|
|
|
2012-07-13 00:54:27 +02:00
|
|
|
printf("CPU %d is up\n", cpu);
|
2010-09-15 16:10:12 +02:00
|
|
|
|
2010-10-26 23:07:27 +02:00
|
|
|
cpu_identify();
|
|
|
|
|
|
2010-09-15 16:10:12 +02:00
|
|
|
lapic_enable(cpu);
|
2010-09-15 16:11:25 +02:00
|
|
|
fpu_init();
|
2010-09-15 16:10:12 +02:00
|
|
|
|
|
|
|
|
if (app_cpu_init_timer(system_hz)) {
|
|
|
|
|
panic("FATAL : failed to initialize timer interrupts CPU %d, "
|
2010-09-15 16:10:27 +02:00
|
|
|
"cannot continue without any clock source!", cpu);
|
2010-09-15 16:10:12 +02:00
|
|
|
}
|
2010-09-15 16:10:07 +02:00
|
|
|
|
2010-09-15 16:10:18 +02:00
|
|
|
/* FIXME assign CPU local idle structure */
|
2010-09-15 16:10:24 +02:00
|
|
|
get_cpulocal_var(proc_ptr) = get_cpulocal_var_ptr(idle_proc);
|
|
|
|
|
get_cpulocal_var(bill_ptr) = get_cpulocal_var_ptr(idle_proc);
|
2010-09-15 16:10:18 +02:00
|
|
|
|
2010-09-15 16:10:12 +02:00
|
|
|
ap_boot_finished(cpu);
|
|
|
|
|
spinlock_unlock(&boot_lock);
|
2010-09-15 16:09:52 +02:00
|
|
|
|
|
|
|
|
switch_to_user();
|
|
|
|
|
NOT_REACHABLE;
|
|
|
|
|
}
|
|
|
|
|
|
2012-03-25 20:25:53 +02:00
|
|
|
void smp_ap_boot(void)
|
2010-09-15 16:09:52 +02:00
|
|
|
{
|
|
|
|
|
switch_k_stack((char *)get_k_stack_top(__ap_id) -
|
|
|
|
|
X86_STACK_TOP_RESERVED, ap_finish_booting);
|
|
|
|
|
}
|
|
|
|
|
|
2012-03-25 20:25:53 +02:00
|
|
|
static void smp_reinit_vars(void)
|
2010-09-15 16:09:52 +02:00
|
|
|
{
|
|
|
|
|
lapic_addr = lapic_eoi_addr = 0;
|
|
|
|
|
ioapic_enabled = 0;
|
|
|
|
|
|
|
|
|
|
ncpus = 1;
|
|
|
|
|
}
|
|
|
|
|
|
2012-03-25 20:25:53 +02:00
|
|
|
static void tss_init_all(void)
|
2010-09-15 16:09:52 +02:00
|
|
|
{
|
|
|
|
|
unsigned cpu;
|
|
|
|
|
|
|
|
|
|
for(cpu = 0; cpu < ncpus ; cpu++)
|
|
|
|
|
tss_init(cpu, get_k_stack_top(cpu));
|
|
|
|
|
}
|
|
|
|
|
|
2012-03-25 20:25:53 +02:00
|
|
|
static int discover_cpus(void)
|
2010-09-15 16:09:52 +02:00
|
|
|
{
|
|
|
|
|
struct acpi_madt_lapic * cpu;
|
|
|
|
|
|
|
|
|
|
while (ncpus < CONFIG_MAX_CPUS && (cpu = acpi_get_lapic_next())) {
|
|
|
|
|
apicid2cpuid[cpu->apic_id] = ncpus;
|
|
|
|
|
cpuid2apicid[ncpus] = cpu->apic_id;
|
|
|
|
|
printf("CPU %3d local APIC id %3d\n", ncpus, cpu->apic_id);
|
|
|
|
|
ncpus++;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return ncpus;
|
|
|
|
|
}
|
|
|
|
|
|
2012-03-25 20:25:53 +02:00
|
|
|
void smp_init (void)
|
2010-09-15 16:09:52 +02:00
|
|
|
{
|
|
|
|
|
/* read the MP configuration */
|
|
|
|
|
if (!discover_cpus()) {
|
|
|
|
|
ncpus = 1;
|
|
|
|
|
goto uniproc_fallback;
|
|
|
|
|
}
|
|
|
|
|
|
No more intel/minix segments.
This commit removes all traces of Minix segments (the text/data/stack
memory map abstraction in the kernel) and significance of Intel segments
(hardware segments like CS, DS that add offsets to all addressing before
page table translation). This ultimately simplifies the memory layout
and addressing and makes the same layout possible on non-Intel
architectures.
There are only two types of addresses in the world now: virtual
and physical; even the kernel and processes have the same virtual
address space. Kernel and user processes can be distinguished at a
glance as processes won't use 0xF0000000 and above.
No static pre-allocated memory sizes exist any more.
Changes to booting:
. The pre_init.c leaves the kernel and modules exactly as
they were left by the bootloader in physical memory
. The kernel starts running using physical addressing,
loaded at a fixed location given in its linker script by the
bootloader. All code and data in this phase are linked to
this fixed low location.
. It makes a bootstrap pagetable to map itself to a
fixed high location (also in linker script) and jumps to
the high address. All code and data then use this high addressing.
. All code/data symbols linked at the low addresses is prefixed by
an objcopy step with __k_unpaged_*, so that that code cannot
reference highly-linked symbols (which aren't valid yet) or vice
versa (symbols that aren't valid any more).
. The two addressing modes are separated in the linker script by
collecting the unpaged_*.o objects and linking them with low
addresses, and linking the rest high. Some objects are linked
twice, once low and once high.
. The bootstrap phase passes a lot of information (e.g. free memory
list, physical location of the modules, etc.) using the kinfo
struct.
. After this bootstrap the low-linked part is freed.
. The kernel maps in VM into the bootstrap page table so that VM can
begin executing. Its first job is to make page tables for all other
boot processes. So VM runs before RS, and RS gets a fully dynamic,
VM-managed address space. VM gets its privilege info from RS as usual
but that happens after RS starts running.
. Both the kernel loading VM and VM organizing boot processes happen
using the libexec logic. This removes the last reason for VM to
still know much about exec() and vm/exec.c is gone.
Further Implementation:
. All segments are based at 0 and have a 4 GB limit.
. The kernel is mapped in at the top of the virtual address
space so as not to constrain the user processes.
. Processes do not use segments from the LDT at all; there are
no segments in the LDT any more, so no LLDT is needed.
. The Minix segments T/D/S are gone and so none of the
user-space or in-kernel copy functions use them. The copy
functions use a process endpoint of NONE to realize it's
a physical address, virtual otherwise.
. The umap call only makes sense to translate a virtual address
to a physical address now.
. Segments-related calls like newmap and alloc_segments are gone.
. All segments-related translation in VM is gone (vir2map etc).
. Initialization in VM is simpler as no moving around is necessary.
. VM and all other boot processes can be linked wherever they wish
and will be mapped in at the right location by the kernel and VM
respectively.
Other changes:
. The multiboot code is less special: it does not use mb_print
for its diagnostics any more but uses printf() as normal, saving
the output into the diagnostics buffer, only printing to the
screen using the direct print functions if a panic() occurs.
. The multiboot code uses the flexible 'free memory map list'
style to receive the list of free memory if available.
. The kernel determines the memory layout of the processes to
a degree: it tells VM where the kernel starts and ends and
where the kernel wants the top of the process to be. VM then
uses this entire range, i.e. the stack is right at the top,
and mmap()ped bits of memory are placed below that downwards,
and the break grows upwards.
Other Consequences:
. Every process gets its own page table as address spaces
can't be separated any more by segments.
. As all segments are 0-based, there is no distinction between
virtual and linear addresses, nor between userspace and
kernel addresses.
. Less work is done when context switching, leading to a net
performance increase. (8% faster on my machine for 'make servers'.)
. The layout and configuration of the GDT makes sysenter and syscall
possible.
2012-05-07 16:03:35 +02:00
|
|
|
lapic_addr = LOCAL_APIC_DEF_ADDR;
|
2010-09-15 16:09:52 +02:00
|
|
|
ioapic_enabled = 0;
|
|
|
|
|
|
|
|
|
|
tss_init_all();
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* we still run on the boot stack and we cannot use cpuid as its value
|
|
|
|
|
* wasn't set yet. apicid2cpuid initialized in mps_init()
|
|
|
|
|
*/
|
|
|
|
|
bsp_cpu_id = apicid2cpuid[apicid()];
|
|
|
|
|
|
|
|
|
|
if (!lapic_enable(bsp_cpu_id)) {
|
|
|
|
|
printf("ERROR : failed to initialize BSP Local APIC\n");
|
|
|
|
|
goto uniproc_fallback;
|
|
|
|
|
}
|
2010-10-19 19:07:19 +02:00
|
|
|
|
|
|
|
|
bsp_lapic_id = apicid();
|
2010-09-15 16:09:52 +02:00
|
|
|
|
|
|
|
|
acpi_init();
|
|
|
|
|
|
|
|
|
|
if (!detect_ioapics()) {
|
|
|
|
|
lapic_disable();
|
|
|
|
|
lapic_addr = 0x0;
|
|
|
|
|
goto uniproc_fallback;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
ioapic_enable_all();
|
|
|
|
|
|
|
|
|
|
if (ioapic_enabled)
|
|
|
|
|
machine.apic_enabled = 1;
|
|
|
|
|
|
|
|
|
|
/* set smp idt entries. */
|
|
|
|
|
apic_idt_init(0); /* Not a reset ! */
|
|
|
|
|
idt_reload();
|
2010-09-15 16:10:00 +02:00
|
|
|
|
2010-09-15 16:09:52 +02:00
|
|
|
BOOT_VERBOSE(printf("SMP initialized\n"));
|
|
|
|
|
|
|
|
|
|
switch_k_stack((char *)get_k_stack_top(bsp_cpu_id) -
|
|
|
|
|
X86_STACK_TOP_RESERVED, smp_start_aps);
|
|
|
|
|
|
|
|
|
|
return;
|
|
|
|
|
|
|
|
|
|
uniproc_fallback:
|
|
|
|
|
apic_idt_init(1); /* Reset to PIC idt ! */
|
|
|
|
|
idt_reload();
|
|
|
|
|
smp_reinit_vars (); /* revert to a single proc system. */
|
No more intel/minix segments.
This commit removes all traces of Minix segments (the text/data/stack
memory map abstraction in the kernel) and significance of Intel segments
(hardware segments like CS, DS that add offsets to all addressing before
page table translation). This ultimately simplifies the memory layout
and addressing and makes the same layout possible on non-Intel
architectures.
There are only two types of addresses in the world now: virtual
and physical; even the kernel and processes have the same virtual
address space. Kernel and user processes can be distinguished at a
glance as processes won't use 0xF0000000 and above.
No static pre-allocated memory sizes exist any more.
Changes to booting:
. The pre_init.c leaves the kernel and modules exactly as
they were left by the bootloader in physical memory
. The kernel starts running using physical addressing,
loaded at a fixed location given in its linker script by the
bootloader. All code and data in this phase are linked to
this fixed low location.
. It makes a bootstrap pagetable to map itself to a
fixed high location (also in linker script) and jumps to
the high address. All code and data then use this high addressing.
. All code/data symbols linked at the low addresses is prefixed by
an objcopy step with __k_unpaged_*, so that that code cannot
reference highly-linked symbols (which aren't valid yet) or vice
versa (symbols that aren't valid any more).
. The two addressing modes are separated in the linker script by
collecting the unpaged_*.o objects and linking them with low
addresses, and linking the rest high. Some objects are linked
twice, once low and once high.
. The bootstrap phase passes a lot of information (e.g. free memory
list, physical location of the modules, etc.) using the kinfo
struct.
. After this bootstrap the low-linked part is freed.
. The kernel maps in VM into the bootstrap page table so that VM can
begin executing. Its first job is to make page tables for all other
boot processes. So VM runs before RS, and RS gets a fully dynamic,
VM-managed address space. VM gets its privilege info from RS as usual
but that happens after RS starts running.
. Both the kernel loading VM and VM organizing boot processes happen
using the libexec logic. This removes the last reason for VM to
still know much about exec() and vm/exec.c is gone.
Further Implementation:
. All segments are based at 0 and have a 4 GB limit.
. The kernel is mapped in at the top of the virtual address
space so as not to constrain the user processes.
. Processes do not use segments from the LDT at all; there are
no segments in the LDT any more, so no LLDT is needed.
. The Minix segments T/D/S are gone and so none of the
user-space or in-kernel copy functions use them. The copy
functions use a process endpoint of NONE to realize it's
a physical address, virtual otherwise.
. The umap call only makes sense to translate a virtual address
to a physical address now.
. Segments-related calls like newmap and alloc_segments are gone.
. All segments-related translation in VM is gone (vir2map etc).
. Initialization in VM is simpler as no moving around is necessary.
. VM and all other boot processes can be linked wherever they wish
and will be mapped in at the right location by the kernel and VM
respectively.
Other changes:
. The multiboot code is less special: it does not use mb_print
for its diagnostics any more but uses printf() as normal, saving
the output into the diagnostics buffer, only printing to the
screen using the direct print functions if a panic() occurs.
. The multiboot code uses the flexible 'free memory map list'
style to receive the list of free memory if available.
. The kernel determines the memory layout of the processes to
a degree: it tells VM where the kernel starts and ends and
where the kernel wants the top of the process to be. VM then
uses this entire range, i.e. the stack is right at the top,
and mmap()ped bits of memory are placed below that downwards,
and the break grows upwards.
Other Consequences:
. Every process gets its own page table as address spaces
can't be separated any more by segments.
. As all segments are 0-based, there is no distinction between
virtual and linear addresses, nor between userspace and
kernel addresses.
. Less work is done when context switching, leading to a net
performance increase. (8% faster on my machine for 'make servers'.)
. The layout and configuration of the GDT makes sysenter and syscall
possible.
2012-05-07 16:03:35 +02:00
|
|
|
intr_init(0); /* no auto eoi */
|
2010-09-15 16:09:52 +02:00
|
|
|
printf("WARNING : SMP initialization failed\n");
|
|
|
|
|
}
|
2010-09-15 16:10:54 +02:00
|
|
|
|
2012-03-25 20:25:53 +02:00
|
|
|
void arch_smp_halt_cpu(void)
|
2010-09-15 16:10:54 +02:00
|
|
|
{
|
|
|
|
|
/* say that we are down */
|
|
|
|
|
cpu_down = cpuid;
|
|
|
|
|
barrier();
|
|
|
|
|
/* unlock the BKL and don't continue */
|
|
|
|
|
BKL_UNLOCK();
|
|
|
|
|
for(;;);
|
|
|
|
|
}
|
2010-09-15 16:10:57 +02:00
|
|
|
|
2012-03-25 20:25:53 +02:00
|
|
|
void arch_send_smp_schedule_ipi(unsigned cpu)
|
2010-09-15 16:10:57 +02:00
|
|
|
{
|
|
|
|
|
apic_send_ipi(APIC_SMP_SCHED_PROC_VECTOR, cpu, APIC_IPI_DEST);
|
|
|
|
|
}
|
