New scheduler.

Removed cli and sti stack in favor of tracking
number of locks held on each CPU and explicit
conditionals in spinlock.c.

1 files changed, 168 insertions, 153 deletions

diff --git a/proc.c b/proc.c
index 76ec64e..01d8f2f 100644
--- a/proc.c
+++ b/proc.c
@@ -12,6 +12,7 @@ struct spinlock proc_table_lock;
 struct proc proc[NPROC];
 struct proc *curproc[NCPU];
 int next_pid = 1;
+extern void forkret(void);
 
 /*
  * set up a process's task state and segment descriptors
@@ -96,12 +97,14 @@ newproc()
   *(np->tf) = *(op->tf);
   np->tf->tf_regs.reg_eax = 0; // so fork() returns 0 in child
 
-  // set up new jmpbuf to start executing at trapret with esp pointing at tf
+  // Set up new jmpbuf to start executing forkret (see trapasm.S)
+  // with esp pointing at tf.  Forkret will call forkret1 (below) to release
+  // the proc_table_lock and then jump into the usual trap return code.
   memset(&np->jmpbuf, 0, sizeof np->jmpbuf);
-  np->jmpbuf.jb_eip = (unsigned) trapret;
+  np->jmpbuf.jb_eip = (unsigned) forkret;
   np->jmpbuf.jb_esp = (unsigned) np->tf - 4;  // -4 for the %eip that isn't actually there
 
-  // copy file descriptors
+  // Copy file descriptors
   for(fd = 0; fd < NOFILE; fd++){
     np->fds[fd] = op->fds[fd];
     if(np->fds[fd])
@@ -112,127 +115,152 @@ newproc()
 }
 
 void
+forkret1(void)
+{
+  release(&proc_table_lock);
+}
+
+// Per-CPU process scheduler. 
+// Each CPU calls scheduler() after setting itself up.
+// Scheduler never returns.  It loops, doing:
+//  - choose a process to run
+//  - longjmp to start running that process
+//  - eventually that process transfers control back
+//      via longjmp back to the top of scheduler.
+void
 scheduler(void)
 {
-  struct proc *op, *np;
+  struct proc *p;
   int i;
 
   cprintf("start scheduler on cpu %d jmpbuf %p\n", cpu(), &cpus[cpu()].jmpbuf);
   cpus[cpu()].lastproc = &proc[0];
 
-  setjmp(&cpus[cpu()].jmpbuf);
-
-  op = curproc[cpu()];
-
-  if(op == 0 || op->mtx != &proc_table_lock)
-    acquire1(&proc_table_lock, op);
-
-  if(op){
-    if(op->newstate <= 0 || op->newstate > ZOMBIE)
-      panic("scheduler");
-    op->state = op->newstate;
-    op->newstate = -1;
-    if(op->mtx){
-      struct spinlock *mtx = op->mtx;
-      op->mtx = 0;
-      if(mtx != &proc_table_lock)
-        release1(mtx, op);
-    }
-  }
-
-  // find a runnable process and switch to it
-  curproc[cpu()] = 0;
-  np = cpus[cpu()].lastproc + 1;
-  while(1){
+  for(;;){
+    // Loop over process table looking for process to run.
+    acquire(&proc_table_lock);
     for(i = 0; i < NPROC; i++){
-      if(np >= &proc[NPROC])
-        np = &proc[0];
-      if(np->state == RUNNABLE)
-        break;
-      np++;
-    }
-
-    if(i < NPROC){
-      np->state = RUNNING;
-      release1(&proc_table_lock, op);
-      break;
+      p = &proc[i];
+      if(p->state != RUNNABLE)
+        continue;
+      
+      // Run this process.
+      // XXX move this into swtch or trapret or something.
+      // It can run on the other stack.
+      // h/w sets busy bit in TSS descriptor sometimes, and faults
+      // if it's set in LTR. so clear tss descriptor busy bit.
+      p->gdt[SEG_TSS].sd_type = STS_T32A;
+    
+      // XXX should probably have an lgdt() function in x86.h
+      // to confine all the inline assembly.
+      // XXX probably ought to lgdt on trap return too, in case
+      // a system call has moved a program or changed its size.
+      asm volatile("lgdt %0" : : "g" (p->gdt_pd.pd_lim));
+      ltr(SEG_TSS << 3);
+
+      // Switch to chosen process.  It is the process's job 
+      // to release proc_table_lock and then reacquire it
+      // before jumping back to us.
+      if(0) cprintf("cpu%d: run %d\n", cpu(), p-proc);
+      curproc[cpu()] = p;
+      p->state = RUNNING;
+      if(setjmp(&cpus[cpu()].jmpbuf) == 0)
+        longjmp(&p->jmpbuf);
+    
+      // Process is done running for now. 
+      // It should have changed its p->state before coming back.
+      curproc[cpu()] = 0;
+      if(p->state == RUNNING)
+        panic("swtch to scheduler with state=RUNNING");
+      
+      // XXX if not holding proc_table_lock panic.
     }
+    release(&proc_table_lock);
     
-    release1(&proc_table_lock, op);
-    op = 0;
-    acquire(&proc_table_lock);
-    np = &proc[0];
+    if(cpus[cpu()].nlock != 0)
+      panic("holding locks in scheduler");
+
+    // With proc_table_lock released, there are no 
+    // locks held on this cpu, so interrupts are enabled.
+    // Hardware interrupts can happen here.
+    // Also, releasing the lock here lets the other CPUs
+    // look for runnable processes too.
   }
+}
 
-  cpus[cpu()].lastproc = np;
-  curproc[cpu()] = np;
-
-  // h/w sets busy bit in TSS descriptor sometimes, and faults
-  // if it's set in LTR. so clear tss descriptor busy bit.
-  np->gdt[SEG_TSS].sd_type = STS_T32A;
-
-  // XXX should probably have an lgdt() function in x86.h
-  // to confine all the inline assembly.
-  // XXX probably ought to lgdt on trap return too, in case
-  // a system call has moved a program or changed its size.
-  asm volatile("lgdt %0" : : "g" (np->gdt_pd.pd_lim));
-  ltr(SEG_TSS << 3);
-
-  if(0) cprintf("cpu%d: run %d esp=%p callerpc=%p\n", cpu(), np-proc);
-  longjmp(&np->jmpbuf);
+// Enter scheduler.  Must already hold proc_table_lock
+// and have changed curproc[cpu()]->state.
+void
+sched(void)
+{
+  if(setjmp(&curproc[cpu()]->jmpbuf) == 0)
+    longjmp(&cpus[cpu()].jmpbuf);
 }
 
-// give up the cpu by switching to the scheduler,
-// which runs on the per-cpu stack.
+// Give up the CPU for one scheduling round.
 void
-swtch(int newstate)
+yield()
 {
-  struct proc *p = curproc[cpu()];
+  struct proc *p;
 
-  if(p == 0)
-    panic("swtch no proc");
-  if(p->mtx == 0 && p->locks != 0)
-    panic("swtch w/ locks");
-  if(p->mtx && p->locks != 1)
-    panic("swtch w/ locks 1");
-  if(p->mtx && p->mtx->locked == 0)
-    panic("switch w/ lock but not held");
-  if(p->locks && (read_eflags() & FL_IF))
-    panic("swtch w/ lock but FL_IF");
-
-  p->newstate = newstate; // basically an argument to scheduler()
-  if(setjmp(&p->jmpbuf) == 0)
-    longjmp(&cpus[cpu()].jmpbuf);
+  if((p=curproc[cpu()]) == 0 || curproc[cpu()]->state != RUNNING)
+    panic("yield");
+  acquire(&proc_table_lock);
+  p->state = RUNNABLE;
+  sched();
+  release(&proc_table_lock);
 }
 
+// Atomically release lock and sleep on chan.
+// Reacquires lock when reawakened.
 void
-sleep(void *chan, struct spinlock *mtx)
+sleep(void *chan, struct spinlock *lk)
 {
   struct proc *p = curproc[cpu()];
 
   if(p == 0)
     panic("sleep");
 
+  // Must acquire proc_table_lock in order to 
+  // change p->state and then call sched.
+  // 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.
   p->chan = chan;
-  p->mtx = mtx; // scheduler will release it
+  p->state = SLEEPING;
+  sched();
 
-  swtch(WAITING);
-  
-  if(mtx)
-    acquire(mtx);
+  // Tidy up.
   p->chan = 0;
+
+  // Reacquire original lock.
+  if(lk != &proc_table_lock){
+    release(&proc_table_lock);
+    acquire(lk);
+  }
 }
 
+// Wake up all processes sleeping on chan.
+// Proc_table_lock must be held.
 void
 wakeup1(void *chan)
 {
   struct proc *p;
 
   for(p = proc; p < &proc[NPROC]; p++)
-    if(p->state == WAITING && p->chan == chan)
+    if(p->state == SLEEPING && p->chan == chan)
       p->state = RUNNABLE;
 }
 
+// Wake up all processes sleeping on chan.
+// Proc_table_lock is acquired and released.
 void
 wakeup(void *chan)
 {
@@ -241,15 +269,32 @@ wakeup(void *chan)
   release(&proc_table_lock);
 }
 
-// give up the CPU but stay marked as RUNNABLE
-void
-yield()
+// Kill the process with the given pid.
+// Process won't actually exit until it returns
+// to user space (see trap in trap.c).
+int
+proc_kill(int pid)
 {
-  if(curproc[cpu()] == 0 || curproc[cpu()]->state != RUNNING)
-    panic("yield");
-  swtch(RUNNABLE);
+  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.
+// Exited processes remain in the zombie state 
+// until their parent calls wait() to find out they exited.
 void
 proc_exit()
 {
@@ -257,6 +302,7 @@ proc_exit()
   struct proc *cp = curproc[cpu()];
   int fd;
 
+  // Close all open files.
   for(fd = 0; fd < NOFILE; fd++){
     if(cp->fds[fd]){
       fd_close(cp->fds[fd]);
@@ -266,91 +312,60 @@ proc_exit()
 
   acquire(&proc_table_lock);
 
-  // wake up parent
+  // Wake up our parent.
   for(p = proc; p < &proc[NPROC]; p++)
     if(p->pid == cp->ppid)
       wakeup1(p);
 
-  // abandon children
+  // Reparent our children to process 1.
   for(p = proc; p < &proc[NPROC]; p++)
     if(p->ppid == cp->pid)
-      p->pid = 1;
+      p->ppid = 1;
   
-  cp->mtx = &proc_table_lock;
-  swtch(ZOMBIE);
-  panic("a zombie revived");
+  // Jump into the scheduler, never to return.
+  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
 proc_wait(void)
 {
   struct proc *p;
   struct proc *cp = curproc[cpu()];
-  int any, pid;
+  int i, havekids, pid;
 
   acquire(&proc_table_lock);
-
-  while(1){
-    any = 0;
-    for(p = proc; p < &proc[NPROC]; p++){
-      if(p->state == ZOMBIE && p->ppid == cp->pid){
-        kfree(p->mem, p->sz);
-        kfree(p->kstack, KSTACKSIZE);
-        pid = p->pid;
-        p->state = UNUSED;
-        release(&proc_table_lock);
-        return pid;
+  for(;;){
+    // Scan through table looking zombie children.
+    havekids = 0;
+    for(i = 0; i < NPROC; i++){
+      p = &proc[i];
+      if(p->ppid == cp->pid){
+        if(p->state == ZOMBIE){
+          // Found one.
+          kfree(p->mem, p->sz);
+          kfree(p->kstack, KSTACKSIZE);
+          pid = p->pid;
+          p->state = UNUSED;
+          p->pid = 0;
+          release(&proc_table_lock);
+          return pid;
+        }
+        havekids = 1;
       }
-      if(p->state != UNUSED && p->ppid == cp->pid)
-        any = 1;
     }
-    if(any == 0){
+    
+    // No point waiting if we don't have any children.
+    if(!havekids){
       release(&proc_table_lock);
       return -1;
     }
+    
+    // Wait for children to exit.  (See wakeup1 call in proc_exit.)
     sleep(cp, &proc_table_lock);
   }
 }
 
-int
-proc_kill(int pid)
-{
-  struct proc *p;
-
-  acquire(&proc_table_lock);
-  for(p = proc; p < &proc[NPROC]; p++){
-    if(p->pid == pid && p->state != UNUSED){
-      p->killed = 1;
-      if(p->state == WAITING)
-        p->state = RUNNABLE;
-      release(&proc_table_lock);
-      return 0;
-    }
-  }
-  release(&proc_table_lock);
-  return -1;
-}
-
-// disable interrupts
-void
-cli(void)
-{
-  if(cpus[cpu()].clis == 0)
-    __asm __volatile("cli");
-  cpus[cpu()].clis += 1;
-  if((read_eflags() & FL_IF) != 0)
-    panic("cli but enabled");
-}
-
-// enable interrupts
-void
-sti(void)
-{
-  if((read_eflags() & FL_IF) != 0)
-    panic("sti but enabled");
-  if(cpus[cpu()].clis < 1)
-    panic("sti");
-  cpus[cpu()].clis -= 1;
-  if(cpus[cpu()].clis < 1)
-    __asm __volatile("sti");
-}