xv6-cs450/proc.c

488 lines
10 KiB
C
Raw Normal View History

2006-06-12 17:22:12 +02:00
#include "types.h"
2007-08-28 01:26:33 +02:00
#include "defs.h"
#include "param.h"
2006-06-12 17:22:12 +02:00
#include "mmu.h"
#include "x86.h"
2006-06-22 22:47:23 +02:00
#include "proc.h"
#include "spinlock.h"
struct spinlock proc_table_lock;
2006-06-12 17:22:12 +02:00
struct proc proc[NPROC];
2007-08-23 16:35:28 +02:00
static struct proc *initproc;
int nextpid = 1;
extern void forkret(void);
extern void forkret1(struct trapframe*);
2006-06-12 17:22:12 +02:00
void
pinit(void)
{
initlock(&proc_table_lock, "proc_table");
}
// Look in the process table for an UNUSED proc.
// If found, change state to EMBRYO and return it.
// Otherwise return 0.
static struct proc*
allocproc(void)
2006-06-12 17:22:12 +02:00
{
int i;
struct proc *p;
2006-06-12 17:22:12 +02:00
acquire(&proc_table_lock);
for(i = 0; i < NPROC; i++){
p = &proc[i];
if(p->state == UNUSED){
p->state = EMBRYO;
p->pid = nextpid++;
release(&proc_table_lock);
return p;
}
}
release(&proc_table_lock);
return 0;
2006-06-12 17:22:12 +02:00
}
2006-09-08 16:26:51 +02:00
// Grow current process's memory by n bytes.
// Return old size on success, -1 on failure.
int
growproc(int n)
{
2008-08-28 19:57:47 +02:00
char *newmem;
2006-09-08 16:26:51 +02:00
newmem = kalloc(cp->sz + n);
if(newmem == 0)
2007-08-24 22:22:55 +02:00
return -1;
2006-09-08 16:26:51 +02:00
memmove(newmem, cp->mem, cp->sz);
memset(newmem + cp->sz, 0, n);
2008-08-28 19:57:47 +02:00
kfree(cp->mem, cp->sz);
2006-09-08 16:26:51 +02:00
cp->mem = newmem;
cp->sz += n;
kernel SMP interruptibility fixes. Last year, right before I sent xv6 to the printer, I changed the SETGATE calls so that interrupts would be disabled on entry to interrupt handlers, and I added the nlock++ / nlock-- in trap() so that interrupts would stay disabled while the hw handlers (but not the syscall handler) did their work. I did this because the kernel was otherwise causing Bochs to triple-fault in SMP mode, and time was short. Robert observed yesterday that something was keeping the SMP preemption user test from working. It turned out that when I simplified the lapic code I swapped the order of two register writes that I didn't realize were order dependent. I fixed that and then since I had everything paged in kept going and tried to figure out why you can't leave interrupts on during interrupt handlers. There are a few issues. First, there must be some way to keep interrupts from "stacking up" and overflowing the stack. Keeping interrupts off the whole time solves this problem -- even if the clock tick handler runs long enough that the next clock tick is waiting when it finishes, keeping interrupts off means that the handler runs all the way through the "iret" before the next handler begins. This is not really a problem unless you are putting too many prints in trap -- if the OS is doing its job right, the handlers should run quickly and not stack up. Second, if xv6 had page faults, then it would be important to keep interrupts disabled between the start of the interrupt and the time that cr2 was read, to avoid a scenario like: p1 page faults [cr2 set to faulting address] p1 starts executing trapasm.S clock interrupt, p1 preempted, p2 starts executing p2 page faults [cr2 set to another faulting address] p2 starts, finishes fault handler p1 rescheduled, reads cr2, sees wrong fault address Alternately p1 could be rescheduled on the other cpu, in which case it would still see the wrong cr2. That said, I think cr2 is the only interrupt state that isn't pushed onto the interrupt stack atomically at fault time, and xv6 doesn't care. (This isn't entirely hypothetical -- I debugged this problem on Plan 9.) Third, and this is the big one, it is not safe to call cpu() unless interrupts are disabled. If interrupts are enabled then there is no guarantee that, between the time cpu() looks up the cpu id and the time that it the result gets used, the process has not been rescheduled to the other cpu. For example, the very commonly-used expression curproc[cpu()] (aka the macro cp) can end up referring to the wrong proc: the code stores the result of cpu() in %eax, gets rescheduled to the other cpu at just the wrong instant, and then reads curproc[%eax]. We use curproc[cpu()] to get the current process a LOT. In that particular case, if we arranged for the current curproc entry to be addressed by %fs:0 and just use a different %fs on each CPU, then we could safely get at curproc even with interrupts disabled, since the read of %fs would be atomic with the read of %fs:0. Alternately, we could have a curproc() function that disables interrupts while computing curproc[cpu()]. I've done that last one. Even in the current kernel, with interrupts off on entry to trap, interrupts are enabled inside release if there are no locks held. Also, the scheduler's idle loop must be interruptible at times so that the clock and disk interrupts (which might make processes runnable) can be handled. In addition to the rampant use of curproc[cpu()], this little snippet from acquire is wrong on smp: if(cpus[cpu()].nlock == 0) cli(); cpus[cpu()].nlock++; because if interrupts are off then we might call cpu(), get rescheduled to a different cpu, look at cpus[oldcpu].nlock, and wrongly decide not to disable interrupts on the new cpu. The fix is to always call cli(). But this is wrong too: if(holding(lock)) panic("acquire"); cli(); cpus[cpu()].nlock++; because holding looks at cpu(). The fix is: cli(); if(holding(lock)) panic("acquire"); cpus[cpu()].nlock++; I've done that, and I changed cpu() to complain the first time it gets called with interrupts disabled. (It gets called too much to complain every time.) I added new functions splhi and spllo that are like acquire and release but without the locking: void splhi(void) { cli(); cpus[cpu()].nsplhi++; } void spllo(void) { if(--cpus[cpu()].nsplhi == 0) sti(); } and I've used those to protect other sections of code that refer to cpu() when interrupts would otherwise be disabled (basically just curproc and setupsegs). I also use them in acquire/release and got rid of nlock. I'm not thrilled with the names, but I think the concept -- a counted cli/sti -- is sound. Having them also replaces the nlock++/nlock-- in trap.c and main.c, which is nice. Final note: it's still not safe to enable interrupts in the middle of trap() between lapic_eoi and returning to user space. I don't understand why, but we get a fault on pop %es because 0x10 is a bad segment descriptor (!) and then the fault faults trying to go into a new interrupt because 0x8 is a bad segment descriptor too! Triple fault. I haven't debugged this yet.
2007-09-27 14:58:42 +02:00
setupsegs(cp);
2006-09-08 16:26:51 +02:00
return cp->sz - n;
}
2007-08-24 22:28:08 +02:00
// Set up CPU's segment descriptors and task state for a given process.
// If p==0, set up for "idle" state for when scheduler() is running.
void
setupsegs(struct proc *p)
{
2007-08-24 22:28:08 +02:00
struct cpu *c;
2007-09-27 22:09:40 +02:00
pushcli();
2007-08-24 22:28:08 +02:00
c = &cpus[cpu()];
c->ts.ss0 = SEG_KDATA << 3;
if(p)
2007-09-27 22:32:45 +02:00
c->ts.esp0 = (uint)(p->kstack + KSTACKSIZE);
else
c->ts.esp0 = 0xffffffff;
c->gdt[0] = SEG_NULL;
c->gdt[SEG_KCODE] = SEG(STA_X|STA_R, 0, 0x100000 + 64*1024-1, 0);
c->gdt[SEG_KDATA] = SEG(STA_W, 0, 0xffffffff, 0);
c->gdt[SEG_TSS] = SEG16(STS_T32A, (uint)&c->ts, sizeof(c->ts)-1, 0);
c->gdt[SEG_TSS].s = 0;
if(p){
c->gdt[SEG_UCODE] = SEG(STA_X|STA_R, (uint)p->mem, p->sz-1, DPL_USER);
c->gdt[SEG_UDATA] = SEG(STA_W, (uint)p->mem, p->sz-1, DPL_USER);
2007-08-28 20:37:41 +02:00
} else {
c->gdt[SEG_UCODE] = SEG_NULL;
c->gdt[SEG_UDATA] = SEG_NULL;
}
lgdt(c->gdt, sizeof(c->gdt));
ltr(SEG_TSS << 3);
2007-09-27 22:09:40 +02:00
popcli();
}
2006-06-12 17:22:12 +02:00
// Create a new process copying p as the parent.
// Sets up stack to return as if from system call.
// Caller must set state of returned proc to RUNNABLE.
2006-09-06 19:27:19 +02:00
struct proc*
copyproc(struct proc *p)
2006-06-12 17:22:12 +02:00
{
int i;
2006-06-12 17:22:12 +02:00
struct proc *np;
// Allocate process.
if((np = allocproc()) == 0)
2006-06-12 17:22:12 +02:00
return 0;
// Allocate kernel stack.
if((np->kstack = kalloc(KSTACKSIZE)) == 0){
np->state = UNUSED;
2006-06-12 17:22:12 +02:00
return 0;
}
2007-09-27 22:32:45 +02:00
np->tf = (struct trapframe*)(np->kstack + KSTACKSIZE) - 1;
2006-09-06 19:27:19 +02:00
if(p){ // Copy process state from p.
2007-08-23 16:40:30 +02:00
np->parent = p;
memmove(np->tf, p->tf, sizeof(*np->tf));
np->sz = p->sz;
if((np->mem = kalloc(np->sz)) == 0){
kfree(np->kstack, KSTACKSIZE);
np->kstack = 0;
np->state = UNUSED;
2008-08-28 02:53:24 +02:00
np->parent = 0;
return 0;
}
memmove(np->mem, p->mem, np->sz);
for(i = 0; i < NOFILE; i++)
if(p->ofile[i])
np->ofile[i] = filedup(p->ofile[i]);
np->cwd = idup(p->cwd);
}
2007-08-28 14:48:33 +02:00
// Set up new context to start executing at forkret (see below).
2008-10-15 07:14:10 +02:00
np->context = (struct context *)np->tf - 1;
memset(np->context, 0, sizeof(*np->context));
np->context->eip = (uint)forkret;
2006-06-12 17:22:12 +02:00
// Clear %eax so that fork system call returns 0 in child.
np->tf->eax = 0;
2006-06-12 17:22:12 +02:00
return np;
}
// Set up first user process.
void
userinit(void)
{
struct proc *p;
extern uchar _binary_initcode_start[], _binary_initcode_size[];
p = copyproc(0);
p->sz = PAGE;
p->mem = kalloc(p->sz);
p->cwd = namei("/");
2007-08-22 19:45:52 +02:00
memset(p->tf, 0, sizeof(*p->tf));
p->tf->cs = (SEG_UCODE << 3) | DPL_USER;
2007-08-24 22:22:55 +02:00
p->tf->ds = (SEG_UDATA << 3) | DPL_USER;
p->tf->es = p->tf->ds;
p->tf->ss = p->tf->ds;
p->tf->eflags = FL_IF;
p->tf->esp = p->sz;
2007-08-23 16:35:28 +02:00
// Make return address readable; needed for some gcc.
p->tf->esp -= 4;
*(uint*)(p->mem + p->tf->esp) = 0xefefefef;
2007-08-24 22:22:55 +02:00
// On entry to user space, start executing at beginning of initcode.S.
p->tf->eip = 0;
memmove(p->mem, _binary_initcode_start, (int)_binary_initcode_size);
safestrcpy(p->name, "initcode", sizeof(p->name));
p->state = RUNNABLE;
2007-08-23 16:35:28 +02:00
initproc = p;
}
kernel SMP interruptibility fixes. Last year, right before I sent xv6 to the printer, I changed the SETGATE calls so that interrupts would be disabled on entry to interrupt handlers, and I added the nlock++ / nlock-- in trap() so that interrupts would stay disabled while the hw handlers (but not the syscall handler) did their work. I did this because the kernel was otherwise causing Bochs to triple-fault in SMP mode, and time was short. Robert observed yesterday that something was keeping the SMP preemption user test from working. It turned out that when I simplified the lapic code I swapped the order of two register writes that I didn't realize were order dependent. I fixed that and then since I had everything paged in kept going and tried to figure out why you can't leave interrupts on during interrupt handlers. There are a few issues. First, there must be some way to keep interrupts from "stacking up" and overflowing the stack. Keeping interrupts off the whole time solves this problem -- even if the clock tick handler runs long enough that the next clock tick is waiting when it finishes, keeping interrupts off means that the handler runs all the way through the "iret" before the next handler begins. This is not really a problem unless you are putting too many prints in trap -- if the OS is doing its job right, the handlers should run quickly and not stack up. Second, if xv6 had page faults, then it would be important to keep interrupts disabled between the start of the interrupt and the time that cr2 was read, to avoid a scenario like: p1 page faults [cr2 set to faulting address] p1 starts executing trapasm.S clock interrupt, p1 preempted, p2 starts executing p2 page faults [cr2 set to another faulting address] p2 starts, finishes fault handler p1 rescheduled, reads cr2, sees wrong fault address Alternately p1 could be rescheduled on the other cpu, in which case it would still see the wrong cr2. That said, I think cr2 is the only interrupt state that isn't pushed onto the interrupt stack atomically at fault time, and xv6 doesn't care. (This isn't entirely hypothetical -- I debugged this problem on Plan 9.) Third, and this is the big one, it is not safe to call cpu() unless interrupts are disabled. If interrupts are enabled then there is no guarantee that, between the time cpu() looks up the cpu id and the time that it the result gets used, the process has not been rescheduled to the other cpu. For example, the very commonly-used expression curproc[cpu()] (aka the macro cp) can end up referring to the wrong proc: the code stores the result of cpu() in %eax, gets rescheduled to the other cpu at just the wrong instant, and then reads curproc[%eax]. We use curproc[cpu()] to get the current process a LOT. In that particular case, if we arranged for the current curproc entry to be addressed by %fs:0 and just use a different %fs on each CPU, then we could safely get at curproc even with interrupts disabled, since the read of %fs would be atomic with the read of %fs:0. Alternately, we could have a curproc() function that disables interrupts while computing curproc[cpu()]. I've done that last one. Even in the current kernel, with interrupts off on entry to trap, interrupts are enabled inside release if there are no locks held. Also, the scheduler's idle loop must be interruptible at times so that the clock and disk interrupts (which might make processes runnable) can be handled. In addition to the rampant use of curproc[cpu()], this little snippet from acquire is wrong on smp: if(cpus[cpu()].nlock == 0) cli(); cpus[cpu()].nlock++; because if interrupts are off then we might call cpu(), get rescheduled to a different cpu, look at cpus[oldcpu].nlock, and wrongly decide not to disable interrupts on the new cpu. The fix is to always call cli(). But this is wrong too: if(holding(lock)) panic("acquire"); cli(); cpus[cpu()].nlock++; because holding looks at cpu(). The fix is: cli(); if(holding(lock)) panic("acquire"); cpus[cpu()].nlock++; I've done that, and I changed cpu() to complain the first time it gets called with interrupts disabled. (It gets called too much to complain every time.) I added new functions splhi and spllo that are like acquire and release but without the locking: void splhi(void) { cli(); cpus[cpu()].nsplhi++; } void spllo(void) { if(--cpus[cpu()].nsplhi == 0) sti(); } and I've used those to protect other sections of code that refer to cpu() when interrupts would otherwise be disabled (basically just curproc and setupsegs). I also use them in acquire/release and got rid of nlock. I'm not thrilled with the names, but I think the concept -- a counted cli/sti -- is sound. Having them also replaces the nlock++/nlock-- in trap.c and main.c, which is nice. Final note: it's still not safe to enable interrupts in the middle of trap() between lapic_eoi and returning to user space. I don't understand why, but we get a fault on pop %es because 0x10 is a bad segment descriptor (!) and then the fault faults trying to go into a new interrupt because 0x8 is a bad segment descriptor too! Triple fault. I haven't debugged this yet.
2007-09-27 14:58:42 +02:00
// Return currently running process.
struct proc*
curproc(void)
{
2007-09-27 22:32:45 +02:00
struct proc *p;
kernel SMP interruptibility fixes. Last year, right before I sent xv6 to the printer, I changed the SETGATE calls so that interrupts would be disabled on entry to interrupt handlers, and I added the nlock++ / nlock-- in trap() so that interrupts would stay disabled while the hw handlers (but not the syscall handler) did their work. I did this because the kernel was otherwise causing Bochs to triple-fault in SMP mode, and time was short. Robert observed yesterday that something was keeping the SMP preemption user test from working. It turned out that when I simplified the lapic code I swapped the order of two register writes that I didn't realize were order dependent. I fixed that and then since I had everything paged in kept going and tried to figure out why you can't leave interrupts on during interrupt handlers. There are a few issues. First, there must be some way to keep interrupts from "stacking up" and overflowing the stack. Keeping interrupts off the whole time solves this problem -- even if the clock tick handler runs long enough that the next clock tick is waiting when it finishes, keeping interrupts off means that the handler runs all the way through the "iret" before the next handler begins. This is not really a problem unless you are putting too many prints in trap -- if the OS is doing its job right, the handlers should run quickly and not stack up. Second, if xv6 had page faults, then it would be important to keep interrupts disabled between the start of the interrupt and the time that cr2 was read, to avoid a scenario like: p1 page faults [cr2 set to faulting address] p1 starts executing trapasm.S clock interrupt, p1 preempted, p2 starts executing p2 page faults [cr2 set to another faulting address] p2 starts, finishes fault handler p1 rescheduled, reads cr2, sees wrong fault address Alternately p1 could be rescheduled on the other cpu, in which case it would still see the wrong cr2. That said, I think cr2 is the only interrupt state that isn't pushed onto the interrupt stack atomically at fault time, and xv6 doesn't care. (This isn't entirely hypothetical -- I debugged this problem on Plan 9.) Third, and this is the big one, it is not safe to call cpu() unless interrupts are disabled. If interrupts are enabled then there is no guarantee that, between the time cpu() looks up the cpu id and the time that it the result gets used, the process has not been rescheduled to the other cpu. For example, the very commonly-used expression curproc[cpu()] (aka the macro cp) can end up referring to the wrong proc: the code stores the result of cpu() in %eax, gets rescheduled to the other cpu at just the wrong instant, and then reads curproc[%eax]. We use curproc[cpu()] to get the current process a LOT. In that particular case, if we arranged for the current curproc entry to be addressed by %fs:0 and just use a different %fs on each CPU, then we could safely get at curproc even with interrupts disabled, since the read of %fs would be atomic with the read of %fs:0. Alternately, we could have a curproc() function that disables interrupts while computing curproc[cpu()]. I've done that last one. Even in the current kernel, with interrupts off on entry to trap, interrupts are enabled inside release if there are no locks held. Also, the scheduler's idle loop must be interruptible at times so that the clock and disk interrupts (which might make processes runnable) can be handled. In addition to the rampant use of curproc[cpu()], this little snippet from acquire is wrong on smp: if(cpus[cpu()].nlock == 0) cli(); cpus[cpu()].nlock++; because if interrupts are off then we might call cpu(), get rescheduled to a different cpu, look at cpus[oldcpu].nlock, and wrongly decide not to disable interrupts on the new cpu. The fix is to always call cli(). But this is wrong too: if(holding(lock)) panic("acquire"); cli(); cpus[cpu()].nlock++; because holding looks at cpu(). The fix is: cli(); if(holding(lock)) panic("acquire"); cpus[cpu()].nlock++; I've done that, and I changed cpu() to complain the first time it gets called with interrupts disabled. (It gets called too much to complain every time.) I added new functions splhi and spllo that are like acquire and release but without the locking: void splhi(void) { cli(); cpus[cpu()].nsplhi++; } void spllo(void) { if(--cpus[cpu()].nsplhi == 0) sti(); } and I've used those to protect other sections of code that refer to cpu() when interrupts would otherwise be disabled (basically just curproc and setupsegs). I also use them in acquire/release and got rid of nlock. I'm not thrilled with the names, but I think the concept -- a counted cli/sti -- is sound. Having them also replaces the nlock++/nlock-- in trap.c and main.c, which is nice. Final note: it's still not safe to enable interrupts in the middle of trap() between lapic_eoi and returning to user space. I don't understand why, but we get a fault on pop %es because 0x10 is a bad segment descriptor (!) and then the fault faults trying to go into a new interrupt because 0x8 is a bad segment descriptor too! Triple fault. I haven't debugged this yet.
2007-09-27 14:58:42 +02:00
2007-09-27 22:32:45 +02:00
pushcli();
p = cpus[cpu()].curproc;
popcli();
return p;
kernel SMP interruptibility fixes. Last year, right before I sent xv6 to the printer, I changed the SETGATE calls so that interrupts would be disabled on entry to interrupt handlers, and I added the nlock++ / nlock-- in trap() so that interrupts would stay disabled while the hw handlers (but not the syscall handler) did their work. I did this because the kernel was otherwise causing Bochs to triple-fault in SMP mode, and time was short. Robert observed yesterday that something was keeping the SMP preemption user test from working. It turned out that when I simplified the lapic code I swapped the order of two register writes that I didn't realize were order dependent. I fixed that and then since I had everything paged in kept going and tried to figure out why you can't leave interrupts on during interrupt handlers. There are a few issues. First, there must be some way to keep interrupts from "stacking up" and overflowing the stack. Keeping interrupts off the whole time solves this problem -- even if the clock tick handler runs long enough that the next clock tick is waiting when it finishes, keeping interrupts off means that the handler runs all the way through the "iret" before the next handler begins. This is not really a problem unless you are putting too many prints in trap -- if the OS is doing its job right, the handlers should run quickly and not stack up. Second, if xv6 had page faults, then it would be important to keep interrupts disabled between the start of the interrupt and the time that cr2 was read, to avoid a scenario like: p1 page faults [cr2 set to faulting address] p1 starts executing trapasm.S clock interrupt, p1 preempted, p2 starts executing p2 page faults [cr2 set to another faulting address] p2 starts, finishes fault handler p1 rescheduled, reads cr2, sees wrong fault address Alternately p1 could be rescheduled on the other cpu, in which case it would still see the wrong cr2. That said, I think cr2 is the only interrupt state that isn't pushed onto the interrupt stack atomically at fault time, and xv6 doesn't care. (This isn't entirely hypothetical -- I debugged this problem on Plan 9.) Third, and this is the big one, it is not safe to call cpu() unless interrupts are disabled. If interrupts are enabled then there is no guarantee that, between the time cpu() looks up the cpu id and the time that it the result gets used, the process has not been rescheduled to the other cpu. For example, the very commonly-used expression curproc[cpu()] (aka the macro cp) can end up referring to the wrong proc: the code stores the result of cpu() in %eax, gets rescheduled to the other cpu at just the wrong instant, and then reads curproc[%eax]. We use curproc[cpu()] to get the current process a LOT. In that particular case, if we arranged for the current curproc entry to be addressed by %fs:0 and just use a different %fs on each CPU, then we could safely get at curproc even with interrupts disabled, since the read of %fs would be atomic with the read of %fs:0. Alternately, we could have a curproc() function that disables interrupts while computing curproc[cpu()]. I've done that last one. Even in the current kernel, with interrupts off on entry to trap, interrupts are enabled inside release if there are no locks held. Also, the scheduler's idle loop must be interruptible at times so that the clock and disk interrupts (which might make processes runnable) can be handled. In addition to the rampant use of curproc[cpu()], this little snippet from acquire is wrong on smp: if(cpus[cpu()].nlock == 0) cli(); cpus[cpu()].nlock++; because if interrupts are off then we might call cpu(), get rescheduled to a different cpu, look at cpus[oldcpu].nlock, and wrongly decide not to disable interrupts on the new cpu. The fix is to always call cli(). But this is wrong too: if(holding(lock)) panic("acquire"); cli(); cpus[cpu()].nlock++; because holding looks at cpu(). The fix is: cli(); if(holding(lock)) panic("acquire"); cpus[cpu()].nlock++; I've done that, and I changed cpu() to complain the first time it gets called with interrupts disabled. (It gets called too much to complain every time.) I added new functions splhi and spllo that are like acquire and release but without the locking: void splhi(void) { cli(); cpus[cpu()].nsplhi++; } void spllo(void) { if(--cpus[cpu()].nsplhi == 0) sti(); } and I've used those to protect other sections of code that refer to cpu() when interrupts would otherwise be disabled (basically just curproc and setupsegs). I also use them in acquire/release and got rid of nlock. I'm not thrilled with the names, but I think the concept -- a counted cli/sti -- is sound. Having them also replaces the nlock++/nlock-- in trap.c and main.c, which is nice. Final note: it's still not safe to enable interrupts in the middle of trap() between lapic_eoi and returning to user space. I don't understand why, but we get a fault on pop %es because 0x10 is a bad segment descriptor (!) and then the fault faults trying to go into a new interrupt because 0x8 is a bad segment descriptor too! Triple fault. I haven't debugged this yet.
2007-09-27 14:58:42 +02:00
}
2006-09-07 16:12:30 +02:00
//PAGEBREAK: 42
2006-09-06 19:27:19 +02:00
// Per-CPU process scheduler.
// Each CPU calls scheduler() after setting itself up.
// Scheduler never returns. It loops, doing:
// - choose a process to run
2007-08-30 19:39:56 +02:00
// - swtch to start running that process
// - eventually that process transfers control
// via swtch back to the scheduler.
2006-06-12 17:22:12 +02:00
void
2006-07-11 03:07:40 +02:00
scheduler(void)
2006-06-12 17:22:12 +02:00
{
struct proc *p;
kernel SMP interruptibility fixes. Last year, right before I sent xv6 to the printer, I changed the SETGATE calls so that interrupts would be disabled on entry to interrupt handlers, and I added the nlock++ / nlock-- in trap() so that interrupts would stay disabled while the hw handlers (but not the syscall handler) did their work. I did this because the kernel was otherwise causing Bochs to triple-fault in SMP mode, and time was short. Robert observed yesterday that something was keeping the SMP preemption user test from working. It turned out that when I simplified the lapic code I swapped the order of two register writes that I didn't realize were order dependent. I fixed that and then since I had everything paged in kept going and tried to figure out why you can't leave interrupts on during interrupt handlers. There are a few issues. First, there must be some way to keep interrupts from "stacking up" and overflowing the stack. Keeping interrupts off the whole time solves this problem -- even if the clock tick handler runs long enough that the next clock tick is waiting when it finishes, keeping interrupts off means that the handler runs all the way through the "iret" before the next handler begins. This is not really a problem unless you are putting too many prints in trap -- if the OS is doing its job right, the handlers should run quickly and not stack up. Second, if xv6 had page faults, then it would be important to keep interrupts disabled between the start of the interrupt and the time that cr2 was read, to avoid a scenario like: p1 page faults [cr2 set to faulting address] p1 starts executing trapasm.S clock interrupt, p1 preempted, p2 starts executing p2 page faults [cr2 set to another faulting address] p2 starts, finishes fault handler p1 rescheduled, reads cr2, sees wrong fault address Alternately p1 could be rescheduled on the other cpu, in which case it would still see the wrong cr2. That said, I think cr2 is the only interrupt state that isn't pushed onto the interrupt stack atomically at fault time, and xv6 doesn't care. (This isn't entirely hypothetical -- I debugged this problem on Plan 9.) Third, and this is the big one, it is not safe to call cpu() unless interrupts are disabled. If interrupts are enabled then there is no guarantee that, between the time cpu() looks up the cpu id and the time that it the result gets used, the process has not been rescheduled to the other cpu. For example, the very commonly-used expression curproc[cpu()] (aka the macro cp) can end up referring to the wrong proc: the code stores the result of cpu() in %eax, gets rescheduled to the other cpu at just the wrong instant, and then reads curproc[%eax]. We use curproc[cpu()] to get the current process a LOT. In that particular case, if we arranged for the current curproc entry to be addressed by %fs:0 and just use a different %fs on each CPU, then we could safely get at curproc even with interrupts disabled, since the read of %fs would be atomic with the read of %fs:0. Alternately, we could have a curproc() function that disables interrupts while computing curproc[cpu()]. I've done that last one. Even in the current kernel, with interrupts off on entry to trap, interrupts are enabled inside release if there are no locks held. Also, the scheduler's idle loop must be interruptible at times so that the clock and disk interrupts (which might make processes runnable) can be handled. In addition to the rampant use of curproc[cpu()], this little snippet from acquire is wrong on smp: if(cpus[cpu()].nlock == 0) cli(); cpus[cpu()].nlock++; because if interrupts are off then we might call cpu(), get rescheduled to a different cpu, look at cpus[oldcpu].nlock, and wrongly decide not to disable interrupts on the new cpu. The fix is to always call cli(). But this is wrong too: if(holding(lock)) panic("acquire"); cli(); cpus[cpu()].nlock++; because holding looks at cpu(). The fix is: cli(); if(holding(lock)) panic("acquire"); cpus[cpu()].nlock++; I've done that, and I changed cpu() to complain the first time it gets called with interrupts disabled. (It gets called too much to complain every time.) I added new functions splhi and spllo that are like acquire and release but without the locking: void splhi(void) { cli(); cpus[cpu()].nsplhi++; } void spllo(void) { if(--cpus[cpu()].nsplhi == 0) sti(); } and I've used those to protect other sections of code that refer to cpu() when interrupts would otherwise be disabled (basically just curproc and setupsegs). I also use them in acquire/release and got rid of nlock. I'm not thrilled with the names, but I think the concept -- a counted cli/sti -- is sound. Having them also replaces the nlock++/nlock-- in trap.c and main.c, which is nice. Final note: it's still not safe to enable interrupts in the middle of trap() between lapic_eoi and returning to user space. I don't understand why, but we get a fault on pop %es because 0x10 is a bad segment descriptor (!) and then the fault faults trying to go into a new interrupt because 0x8 is a bad segment descriptor too! Triple fault. I haven't debugged this yet.
2007-09-27 14:58:42 +02:00
struct cpu *c;
int i;
c = &cpus[cpu()];
for(;;){
// Enable interrupts on this processor, in lieu of saving intena.
sti();
// Loop over process table looking for process to run.
acquire(&proc_table_lock);
for(i = 0; i < NPROC; i++){
p = &proc[i];
if(p->state != RUNNABLE)
continue;
2006-09-06 19:27:19 +02:00
// Switch to chosen process. It is the process's job
// to release proc_table_lock and then reacquire it
// before jumping back to us.
2007-09-27 22:32:45 +02:00
c->curproc = p;
2007-08-28 14:52:14 +02:00
setupsegs(p);
p->state = RUNNING;
kernel SMP interruptibility fixes. Last year, right before I sent xv6 to the printer, I changed the SETGATE calls so that interrupts would be disabled on entry to interrupt handlers, and I added the nlock++ / nlock-- in trap() so that interrupts would stay disabled while the hw handlers (but not the syscall handler) did their work. I did this because the kernel was otherwise causing Bochs to triple-fault in SMP mode, and time was short. Robert observed yesterday that something was keeping the SMP preemption user test from working. It turned out that when I simplified the lapic code I swapped the order of two register writes that I didn't realize were order dependent. I fixed that and then since I had everything paged in kept going and tried to figure out why you can't leave interrupts on during interrupt handlers. There are a few issues. First, there must be some way to keep interrupts from "stacking up" and overflowing the stack. Keeping interrupts off the whole time solves this problem -- even if the clock tick handler runs long enough that the next clock tick is waiting when it finishes, keeping interrupts off means that the handler runs all the way through the "iret" before the next handler begins. This is not really a problem unless you are putting too many prints in trap -- if the OS is doing its job right, the handlers should run quickly and not stack up. Second, if xv6 had page faults, then it would be important to keep interrupts disabled between the start of the interrupt and the time that cr2 was read, to avoid a scenario like: p1 page faults [cr2 set to faulting address] p1 starts executing trapasm.S clock interrupt, p1 preempted, p2 starts executing p2 page faults [cr2 set to another faulting address] p2 starts, finishes fault handler p1 rescheduled, reads cr2, sees wrong fault address Alternately p1 could be rescheduled on the other cpu, in which case it would still see the wrong cr2. That said, I think cr2 is the only interrupt state that isn't pushed onto the interrupt stack atomically at fault time, and xv6 doesn't care. (This isn't entirely hypothetical -- I debugged this problem on Plan 9.) Third, and this is the big one, it is not safe to call cpu() unless interrupts are disabled. If interrupts are enabled then there is no guarantee that, between the time cpu() looks up the cpu id and the time that it the result gets used, the process has not been rescheduled to the other cpu. For example, the very commonly-used expression curproc[cpu()] (aka the macro cp) can end up referring to the wrong proc: the code stores the result of cpu() in %eax, gets rescheduled to the other cpu at just the wrong instant, and then reads curproc[%eax]. We use curproc[cpu()] to get the current process a LOT. In that particular case, if we arranged for the current curproc entry to be addressed by %fs:0 and just use a different %fs on each CPU, then we could safely get at curproc even with interrupts disabled, since the read of %fs would be atomic with the read of %fs:0. Alternately, we could have a curproc() function that disables interrupts while computing curproc[cpu()]. I've done that last one. Even in the current kernel, with interrupts off on entry to trap, interrupts are enabled inside release if there are no locks held. Also, the scheduler's idle loop must be interruptible at times so that the clock and disk interrupts (which might make processes runnable) can be handled. In addition to the rampant use of curproc[cpu()], this little snippet from acquire is wrong on smp: if(cpus[cpu()].nlock == 0) cli(); cpus[cpu()].nlock++; because if interrupts are off then we might call cpu(), get rescheduled to a different cpu, look at cpus[oldcpu].nlock, and wrongly decide not to disable interrupts on the new cpu. The fix is to always call cli(). But this is wrong too: if(holding(lock)) panic("acquire"); cli(); cpus[cpu()].nlock++; because holding looks at cpu(). The fix is: cli(); if(holding(lock)) panic("acquire"); cpus[cpu()].nlock++; I've done that, and I changed cpu() to complain the first time it gets called with interrupts disabled. (It gets called too much to complain every time.) I added new functions splhi and spllo that are like acquire and release but without the locking: void splhi(void) { cli(); cpus[cpu()].nsplhi++; } void spllo(void) { if(--cpus[cpu()].nsplhi == 0) sti(); } and I've used those to protect other sections of code that refer to cpu() when interrupts would otherwise be disabled (basically just curproc and setupsegs). I also use them in acquire/release and got rid of nlock. I'm not thrilled with the names, but I think the concept -- a counted cli/sti -- is sound. Having them also replaces the nlock++/nlock-- in trap.c and main.c, which is nice. Final note: it's still not safe to enable interrupts in the middle of trap() between lapic_eoi and returning to user space. I don't understand why, but we get a fault on pop %es because 0x10 is a bad segment descriptor (!) and then the fault faults trying to go into a new interrupt because 0x8 is a bad segment descriptor too! Triple fault. I haven't debugged this yet.
2007-09-27 14:58:42 +02:00
swtch(&c->context, &p->context);
2006-09-06 19:27:19 +02:00
// Process is done running for now.
// It should have changed its p->state before coming back.
2007-09-27 22:32:45 +02:00
c->curproc = 0;
setupsegs(0);
}
release(&proc_table_lock);
2006-06-12 17:22:12 +02:00
}
}
2006-06-12 17:22:12 +02:00
// Enter scheduler. Must already hold proc_table_lock
2007-09-27 22:32:45 +02:00
// and have changed curproc[cpu()]->state.
void
sched(void)
{
int intena;
if(readeflags()&FL_IF)
kernel SMP interruptibility fixes. Last year, right before I sent xv6 to the printer, I changed the SETGATE calls so that interrupts would be disabled on entry to interrupt handlers, and I added the nlock++ / nlock-- in trap() so that interrupts would stay disabled while the hw handlers (but not the syscall handler) did their work. I did this because the kernel was otherwise causing Bochs to triple-fault in SMP mode, and time was short. Robert observed yesterday that something was keeping the SMP preemption user test from working. It turned out that when I simplified the lapic code I swapped the order of two register writes that I didn't realize were order dependent. I fixed that and then since I had everything paged in kept going and tried to figure out why you can't leave interrupts on during interrupt handlers. There are a few issues. First, there must be some way to keep interrupts from "stacking up" and overflowing the stack. Keeping interrupts off the whole time solves this problem -- even if the clock tick handler runs long enough that the next clock tick is waiting when it finishes, keeping interrupts off means that the handler runs all the way through the "iret" before the next handler begins. This is not really a problem unless you are putting too many prints in trap -- if the OS is doing its job right, the handlers should run quickly and not stack up. Second, if xv6 had page faults, then it would be important to keep interrupts disabled between the start of the interrupt and the time that cr2 was read, to avoid a scenario like: p1 page faults [cr2 set to faulting address] p1 starts executing trapasm.S clock interrupt, p1 preempted, p2 starts executing p2 page faults [cr2 set to another faulting address] p2 starts, finishes fault handler p1 rescheduled, reads cr2, sees wrong fault address Alternately p1 could be rescheduled on the other cpu, in which case it would still see the wrong cr2. That said, I think cr2 is the only interrupt state that isn't pushed onto the interrupt stack atomically at fault time, and xv6 doesn't care. (This isn't entirely hypothetical -- I debugged this problem on Plan 9.) Third, and this is the big one, it is not safe to call cpu() unless interrupts are disabled. If interrupts are enabled then there is no guarantee that, between the time cpu() looks up the cpu id and the time that it the result gets used, the process has not been rescheduled to the other cpu. For example, the very commonly-used expression curproc[cpu()] (aka the macro cp) can end up referring to the wrong proc: the code stores the result of cpu() in %eax, gets rescheduled to the other cpu at just the wrong instant, and then reads curproc[%eax]. We use curproc[cpu()] to get the current process a LOT. In that particular case, if we arranged for the current curproc entry to be addressed by %fs:0 and just use a different %fs on each CPU, then we could safely get at curproc even with interrupts disabled, since the read of %fs would be atomic with the read of %fs:0. Alternately, we could have a curproc() function that disables interrupts while computing curproc[cpu()]. I've done that last one. Even in the current kernel, with interrupts off on entry to trap, interrupts are enabled inside release if there are no locks held. Also, the scheduler's idle loop must be interruptible at times so that the clock and disk interrupts (which might make processes runnable) can be handled. In addition to the rampant use of curproc[cpu()], this little snippet from acquire is wrong on smp: if(cpus[cpu()].nlock == 0) cli(); cpus[cpu()].nlock++; because if interrupts are off then we might call cpu(), get rescheduled to a different cpu, look at cpus[oldcpu].nlock, and wrongly decide not to disable interrupts on the new cpu. The fix is to always call cli(). But this is wrong too: if(holding(lock)) panic("acquire"); cli(); cpus[cpu()].nlock++; because holding looks at cpu(). The fix is: cli(); if(holding(lock)) panic("acquire"); cpus[cpu()].nlock++; I've done that, and I changed cpu() to complain the first time it gets called with interrupts disabled. (It gets called too much to complain every time.) I added new functions splhi and spllo that are like acquire and release but without the locking: void splhi(void) { cli(); cpus[cpu()].nsplhi++; } void spllo(void) { if(--cpus[cpu()].nsplhi == 0) sti(); } and I've used those to protect other sections of code that refer to cpu() when interrupts would otherwise be disabled (basically just curproc and setupsegs). I also use them in acquire/release and got rid of nlock. I'm not thrilled with the names, but I think the concept -- a counted cli/sti -- is sound. Having them also replaces the nlock++/nlock-- in trap.c and main.c, which is nice. Final note: it's still not safe to enable interrupts in the middle of trap() between lapic_eoi and returning to user space. I don't understand why, but we get a fault on pop %es because 0x10 is a bad segment descriptor (!) and then the fault faults trying to go into a new interrupt because 0x8 is a bad segment descriptor too! Triple fault. I haven't debugged this yet.
2007-09-27 14:58:42 +02:00
panic("sched interruptible");
2007-08-09 19:32:40 +02:00
if(cp->state == RUNNING)
2007-08-08 10:57:37 +02:00
panic("sched running");
2006-09-07 18:54:00 +02:00
if(!holding(&proc_table_lock))
2007-08-08 10:57:37 +02:00
panic("sched proc_table_lock");
2007-09-27 22:09:40 +02:00
if(cpus[cpu()].ncli != 1)
2006-09-07 18:54:00 +02:00
panic("sched locks");
intena = cpus[cpu()].intena;
2007-08-28 14:48:33 +02:00
swtch(&cp->context, &cpus[cpu()].context);
cpus[cpu()].intena = intena;
2006-07-11 03:07:40 +02:00
}
// Give up the CPU for one scheduling round.
2006-07-11 03:07:40 +02:00
void
yield(void)
2006-07-11 03:07:40 +02:00
{
acquire(&proc_table_lock);
2007-08-09 19:32:40 +02:00
cp->state = RUNNABLE;
sched();
release(&proc_table_lock);
2006-06-12 17:22:12 +02:00
}
2006-06-15 21:58:01 +02:00
2006-08-29 23:35:30 +02:00
// A fork child's very first scheduling by scheduler()
2007-08-30 19:39:56 +02:00
// will swtch here. "Return" to user space.
void
forkret(void)
{
// Still holding proc_table_lock from scheduler.
release(&proc_table_lock);
2006-09-06 19:27:19 +02:00
// Jump into assembly, never to return.
2007-08-10 18:37:27 +02:00
forkret1(cp->tf);
}
// Atomically release lock and sleep on chan.
// Reacquires lock when reawakened.
2006-06-15 21:58:01 +02:00
void
sleep(void *chan, struct spinlock *lk)
2006-06-15 21:58:01 +02:00
{
2007-08-09 19:32:40 +02:00
if(cp == 0)
2006-07-11 03:07:40 +02:00
panic("sleep");
2006-07-17 07:00:25 +02:00
if(lk == 0)
panic("sleep without lk");
2006-09-06 19:27:19 +02:00
// Must acquire proc_table_lock in order to
// change p->state and then call sched.
2006-09-06 19:27:19 +02:00
// Once we hold proc_table_lock, we can be
// guaranteed that we won't miss any wakeup
// (wakeup runs with proc_table_lock locked),
// so it's okay to release lk.
if(lk != &proc_table_lock){
acquire(&proc_table_lock);
release(lk);
}
// Go to sleep.
2007-08-09 19:32:40 +02:00
cp->chan = chan;
cp->state = SLEEPING;
sched();
// Tidy up.
2007-08-09 19:32:40 +02:00
cp->chan = 0;
// Reacquire original lock.
if(lk != &proc_table_lock){
release(&proc_table_lock);
acquire(lk);
}
2006-06-15 21:58:01 +02:00
}
//PAGEBREAK!
// Wake up all processes sleeping on chan.
// Proc_table_lock must be held.
2007-08-24 22:22:55 +02:00
static void
wakeup1(void *chan)
2006-06-15 21:58:01 +02:00
{
struct proc *p;
for(p = proc; p < &proc[NPROC]; p++)
if(p->state == SLEEPING && p->chan == chan)
2006-06-15 21:58:01 +02:00
p->state = RUNNABLE;
}
// Wake up all processes sleeping on chan.
void
wakeup(void *chan)
{
acquire(&proc_table_lock);
wakeup1(chan);
release(&proc_table_lock);
2006-06-15 21:58:01 +02:00
}
// Kill the process with the given pid.
// Process won't actually exit until it returns
// to user space (see trap in trap.c).
int
kill(int pid)
{
struct proc *p;
acquire(&proc_table_lock);
for(p = proc; p < &proc[NPROC]; p++){
if(p->pid == pid){
p->killed = 1;
// Wake process from sleep if necessary.
if(p->state == SLEEPING)
p->state = RUNNABLE;
release(&proc_table_lock);
return 0;
}
}
release(&proc_table_lock);
return -1;
}
// Exit the current process. Does not return.
2006-09-06 19:27:19 +02:00
// Exited processes remain in the zombie state
// until their parent calls wait() to find out they exited.
void
exit(void)
{
struct proc *p;
int fd;
2007-08-23 16:35:28 +02:00
if(cp == initproc)
2007-08-08 10:57:37 +02:00
panic("init exiting");
// Close all open files.
for(fd = 0; fd < NOFILE; fd++){
2006-09-06 20:38:56 +02:00
if(cp->ofile[fd]){
2006-09-06 20:43:45 +02:00
fileclose(cp->ofile[fd]);
2006-09-06 20:38:56 +02:00
cp->ofile[fd] = 0;
}
}
2006-09-06 19:27:19 +02:00
iput(cp->cwd);
cp->cwd = 0;
acquire(&proc_table_lock);
2007-10-20 20:25:38 +02:00
// Parent might be sleeping in wait().
2007-08-23 16:40:30 +02:00
wakeup1(cp->parent);
2007-08-23 16:35:28 +02:00
// Pass abandoned children to init.
for(p = proc; p < &proc[NPROC]; p++){
2007-08-23 16:40:30 +02:00
if(p->parent == cp){
p->parent = initproc;
2007-08-23 16:35:28 +02:00
if(p->state == ZOMBIE)
wakeup1(initproc);
2007-08-08 10:57:37 +02:00
}
2007-08-23 16:35:28 +02:00
}
2006-09-06 19:27:19 +02:00
// Jump into the scheduler, never to return.
cp->killed = 0;
cp->state = ZOMBIE;
sched();
panic("zombie exit");
}
// Wait for a child process to exit and return its pid.
// Return -1 if this process has no children.
int
wait(void)
{
struct proc *p;
int i, havekids, pid;
acquire(&proc_table_lock);
for(;;){
2006-08-29 23:35:30 +02:00
// Scan through table looking for zombie children.
havekids = 0;
for(i = 0; i < NPROC; i++){
p = &proc[i];
2006-09-07 03:56:22 +02:00
if(p->state == UNUSED)
continue;
2007-08-23 16:40:30 +02:00
if(p->parent == cp){
if(p->state == ZOMBIE){
// Found one.
kfree(p->mem, p->sz);
kfree(p->kstack, KSTACKSIZE);
pid = p->pid;
p->state = UNUSED;
p->pid = 0;
2007-08-23 16:40:30 +02:00
p->parent = 0;
p->name[0] = 0;
release(&proc_table_lock);
return pid;
}
havekids = 1;
}
}
// No point waiting if we don't have any children.
2007-08-08 12:29:42 +02:00
if(!havekids || cp->killed){
release(&proc_table_lock);
return -1;
}
2006-09-06 19:27:19 +02:00
// Wait for children to exit. (See wakeup1 call in proc_exit.)
sleep(cp, &proc_table_lock);
}
}
2006-09-07 17:45:38 +02:00
// Print a process listing to console. For debugging.
// Runs when user types ^P on console.
// No lock to avoid wedging a stuck machine further.
void
procdump(void)
{
static char *states[] = {
2007-08-08 11:42:36 +02:00
[UNUSED] "unused",
[EMBRYO] "embryo",
[SLEEPING] "sleep ",
[RUNNABLE] "runble",
[RUNNING] "run ",
[ZOMBIE] "zombie"
};
2007-08-20 21:37:15 +02:00
int i, j;
2006-09-07 17:45:38 +02:00
struct proc *p;
char *state;
2007-08-20 21:37:15 +02:00
uint pc[10];
2006-09-07 17:45:38 +02:00
2007-08-28 20:32:08 +02:00
for(i = 0; i < NPROC; i++){
2006-09-07 17:45:38 +02:00
p = &proc[i];
if(p->state == UNUSED)
continue;
2007-08-08 11:43:07 +02:00
if(p->state >= 0 && p->state < NELEM(states) && states[p->state])
state = states[p->state];
2007-08-08 11:42:36 +02:00
else
state = "???";
2007-08-20 21:37:15 +02:00
cprintf("%d %s %s", p->pid, state, p->name);
2007-08-28 20:32:08 +02:00
if(p->state == SLEEPING){
2008-10-15 07:14:10 +02:00
getcallerpcs((uint*)p->context->ebp+2, pc);
2007-08-20 21:37:15 +02:00
for(j=0; j<10 && pc[j] != 0; j++)
cprintf(" %p", pc[j]);
}
cprintf("\n");
2006-09-07 17:45:38 +02:00
}
}
2006-09-08 16:26:51 +02:00