with per-proc locks, we don't need the intr_on() inside the loop.

kernel/proc.c

@@ -419,11 +419,11 @@ scheduler(void)
   struct proc *p;
   struct cpu *c = mycpu();
 
+  // Let devices interrupt when no lock is held.
+  intr_on();
+  
   c->proc = 0;
   for(;;){
-    // Give devices a brief chance to interrupt.
-    intr_on();
-
     for(p = proc; p < &proc[NPROC]; p++) {
       acquire(&p->lock);
       if(p->state == RUNNABLE) {