Use low level constructs borrowed from interrupt threads to wait for

work in proc0.
Remove the TDP_WAKEPROC0 workaround.

sys/kern/kern_exit.c

@@ -503,15 +503,7 @@ exit1(struct thread *td, int rv)
 	critical_enter();
 	mtx_unlock_spin(&sched_lock);
 	wakeup(p->p_pptr);
-	/*
-	 * XXX hack, swap in parent process, please see TDP_WAKEPROC0
-	 * code, because TDP_WAKEPROC0 is only useful if thread is
-	 * leaving critical region, but here we never leave and
-	 * thread_exit() will call cpu_throw(), TDP_WAKEPROC0 is never
-	 * cleared.
-	 */
-	if (p->p_pptr->p_sflag & PS_SWAPINREQ)
-		wakeup(&proc0);
+	
 	PROC_UNLOCK(p->p_pptr);
 	mtx_lock_spin(&sched_lock);
 	critical_exit();

sys/kern/kern_switch.c

@@ -593,15 +593,9 @@ critical_exit(void)
 	td = curthread;
 	KASSERT(td->td_critnest != 0,
 	    ("critical_exit: td_critnest == 0"));
-	if (td->td_critnest == 1) {
-		if (td->td_pflags & TDP_WAKEPROC0) {
-			td->td_pflags &= ~TDP_WAKEPROC0;
-			wakeup(&proc0);
-		}
-		
-		td->td_critnest = 0;
-
 #ifdef PREEMPTION
+	if (td->td_critnest == 1) {
+		td->td_critnest = 0;
 		mtx_assert(&sched_lock, MA_NOTOWNED);
 		if (td->td_owepreempt) {
 			td->td_critnest = 1;
@@ -610,12 +604,11 @@ critical_exit(void)
 			mi_switch(SW_INVOL, NULL);
 			mtx_unlock_spin(&sched_lock);
 		}
-	             
+	} else 
 #endif
-
-	} else {
 		td->td_critnest--;
-	}
+	
+	
 	CTR4(KTR_CRITICAL, "critical_exit by thread %p (%ld, %s) to %d", td,
 	    (long)td->td_proc->p_pid, td->td_proc->p_comm, td->td_critnest);
 }

sys/kern/kern_synch.c

@@ -410,11 +410,10 @@ setrunnable(struct thread *td)
 			p->p_sflag |= PS_SWAPINREQ;
 			/*
 			 * due to a LOR between sched_lock and
-			 * the sleepqueue chain locks, delay
-			 * wakeup proc0 until thread leaves
-			 * critical region.
+			 * the sleepqueue chain locks, use
+			 * lower level scheduling functions.
 			 */
-			curthread->td_pflags |= TDP_WAKEPROC0;
+			kick_proc0();
 		}
 	} else
 		sched_wakeup(td);

sys/sys/proc.h

@@ -370,7 +370,7 @@ struct thread {
 #define	TDP_SA		0x00000080 /* A scheduler activation based thread. */
 #define	TDP_UNUSED8	0x00000100 /* --available -- */
 #define	TDP_OWEUPC	0x00000200 /* Call addupc() at next AST. */
-#define	TDP_WAKEPROC0	0x00000400 /* Wants caller to wakeup(&proc0) */
+#define	TDP_UNUSED10	0x00000400 /* --available -- */
 #define	TDP_CAN_UNBIND	0x00000800 /* Only temporarily bound. */
 #define	TDP_SCHED1	0x00001000 /* Reserved for scheduler private use */
 #define	TDP_SCHED2	0x00002000 /* Reserved for scheduler private use */
@@ -848,6 +848,7 @@ void	fork_exit(void (*)(void *, struct trapframe *), void *,
 	    struct trapframe *);
 void	fork_return(struct thread *, struct trapframe *);
 int	inferior(struct proc *p);
+void 	kick_proc0(void);
 int	leavepgrp(struct proc *p);
 int	maybe_preempt(struct thread *td);
 void	mi_switch(int flags, struct thread *newtd);

sys/vm/vm_glue.c

@@ -113,6 +113,10 @@ SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_ANY, scheduler, NULL)
 static void swapout(struct proc *);
 #endif
 
+
+static volatile int proc0_rescan;
+
+
 /*
  * MPSAFE
  *
@@ -596,6 +600,9 @@ scheduler(dummy)
 loop:
 	if (vm_page_count_min()) {
 		VM_WAIT;
+		mtx_lock_spin(&sched_lock);
+		proc0_rescan = 0;
+		mtx_unlock_spin(&sched_lock);
 		goto loop;
 	}
 
@@ -641,7 +648,13 @@ scheduler(dummy)
 	 * Nothing to do, back to sleep.
 	 */
 	if ((p = pp) == NULL) {
-		tsleep(&proc0, PVM, "sched", maxslp * hz / 2);
+		mtx_lock_spin(&sched_lock);
+		if (!proc0_rescan) {
+			TD_SET_IWAIT(&thread0);
+			mi_switch(SW_VOL, NULL);
+		}
+		proc0_rescan = 0;
+		mtx_unlock_spin(&sched_lock);
 		goto loop;
 	}
 	PROC_LOCK(p);
@@ -653,6 +666,9 @@ scheduler(dummy)
 	 */
 	if (p->p_sflag & (PS_INMEM | PS_SWAPPINGOUT | PS_SWAPPINGIN)) {
 		PROC_UNLOCK(p);
+		mtx_lock_spin(&sched_lock);
+		proc0_rescan = 0;
+		mtx_unlock_spin(&sched_lock);
 		goto loop;
 	}
 
@@ -668,10 +684,29 @@ scheduler(dummy)
 	PROC_UNLOCK(p);
 	mtx_lock_spin(&sched_lock);
 	p->p_swtime = 0;
+	proc0_rescan = 0;
 	mtx_unlock_spin(&sched_lock);
 	goto loop;
 }
 
+void kick_proc0(void)
+{
+	struct thread *td = &thread0;
+
+		
+	if (TD_AWAITING_INTR(td)) {
+		CTR2(KTR_INTR, "%s: setrunqueue %d", __func__, 0);
+		TD_CLR_IWAIT(td);
+		setrunqueue(td, SRQ_INTR);
+	} else {
+		proc0_rescan = 1;
+		CTR2(KTR_INTR, "%s: state %d",
+		    __func__, td->td_state);
+	}
+	
+}
+
+
 #ifndef NO_SWAPPING
 
 /*