A little infrastructure, preceding some upcoming changes

to the profiling and statistics code.

Submitted by:	DavidXu@
Reviewed by:	peter@

sys/kern/kern_clock.c

@@ -313,6 +313,10 @@ startprofclock(p)
 	 * cover psdiv, etc. as well.
 	 */
 	mtx_lock_spin(&sched_lock);
+	if (p->p_sflag & PS_STOPPROF) {
+		mtx_unlock_spin(&sched_lock);
+		return;
+	}
 	if ((p->p_sflag & PS_PROFIL) == 0) {
 		p->p_sflag |= PS_PROFIL;
 		if (++profprocs == 1)
@@ -329,9 +333,18 @@ stopprofclock(p)
 	register struct proc *p;
 {
 
+	PROC_LOCK_ASSERT(p, MA_OWNED);
+retry:
 	mtx_lock_spin(&sched_lock);
 	if (p->p_sflag & PS_PROFIL) {
-		p->p_sflag &= ~PS_PROFIL;
+		if (p->p_profthreads) {
+			p->p_sflag |= PS_STOPPROF;
+			mtx_unlock_spin(&sched_lock);
+			msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
+			       "stopprof", NULL);
+			goto retry;
+		}
+		p->p_sflag &= ~(PS_PROFIL|PS_STOPPROF);
 		if (--profprocs == 0)
 			cpu_stopprofclock();
 	}
@@ -400,7 +413,7 @@ statclock(frame)
 		}
 	}
 
-	sched_clock(ke->ke_thread);
+	sched_clock(td);
 
 	/* Update resource usage integrals and maximums. */
 	if ((pstats = p->p_stats) != NULL &&
@@ -430,9 +443,12 @@ profclock(frame)
 		/*
 		 * Came from user mode; CPU was in user state.
 		 * If this process is being profiled, record the tick.
+		 * if there is no related user location yet, don't
+		 * bother trying to count it.
 		 */
 		td = curthread;
-		if (td->td_proc->p_sflag & PS_PROFIL)
+		if ((td->td_proc->p_sflag & PS_PROFIL) &&
+		    !(td->td_flags & TDF_UPCALLING))
 			addupc_intr(td->td_kse, CLKF_PC(frame), 1);
 	}
 #ifdef GPROF

sys/kern/kern_exit.c

@@ -238,7 +238,6 @@ exit1(td, rv)
 	TAILQ_FOREACH(ep, &exit_list, next) 
 		(*ep->function)(p);
 
-	stopprofclock(p);
 
 	MALLOC(p->p_ru, struct rusage *, sizeof(struct rusage),
 		M_ZOMBIE, 0);
@@ -247,6 +246,7 @@ exit1(td, rv)
 	 * P_PPWAIT is set; we will wakeup the parent below.
 	 */
 	PROC_LOCK(p);
+	stopprofclock(p);
 	p->p_flag &= ~(P_TRACED | P_PPWAIT);
 	SIGEMPTYSET(p->p_siglist);
 	PROC_UNLOCK(p);

sys/kern/subr_prof.c

@@ -358,7 +358,9 @@ sysctl_kern_prof(SYSCTL_HANDLER_ARGS)
 			return (0);
 		if (state == GMON_PROF_OFF) {
 			gp->state = state;
+			PROC_LOCK(&proc0);
 			stopprofclock(&proc0);
+			PROC_UNLOCK(&proc0);
 			stopguprof(gp);
 		} else if (state == GMON_PROF_ON) {
 			gp->state = GMON_PROF_OFF;
@@ -369,7 +371,9 @@ sysctl_kern_prof(SYSCTL_HANDLER_ARGS)
 #ifdef GUPROF
 		} else if (state == GMON_PROF_HIRES) {
 			gp->state = GMON_PROF_OFF;
+			PROC_LOCK(&proc0);
 			stopprofclock(&proc0);
+			PROC_UNLOCK(&proc0);
 			startguprof(gp);
 			gp->state = state;
 #endif
@@ -419,7 +423,7 @@ profil(td, uap)
 	struct thread *td;
 	register struct profil_args *uap;
 {
-	register struct uprof *upp;
+	struct uprof *upp;
 	int s;
 	int error = 0;
 
@@ -430,7 +434,9 @@ profil(td, uap)
 		goto done2;
 	}
 	if (uap->scale == 0) {
+		PROC_LOCK(td->td_proc);
 		stopprofclock(td->td_proc);
+		PROC_UNLOCK(td->td_proc);
 		goto done2;
 	}
 	upp = &td->td_proc->p_stats->p_prof;
@@ -472,15 +478,12 @@ profil(td, uap)
  * inaccurate.
  */
 void
-addupc_intr(ke, pc, ticks)
+addupc_intr(struct kse *ke, uintptr_t pc, u_int ticks)
-	register struct kse *ke;
-	register uintptr_t pc;
-	u_int ticks;
 {
-	register struct uprof *prof;
+	struct uprof *prof;
-	register caddr_t addr;
+	caddr_t addr;
-	register u_int i;
+	u_int i;
-	register int v;
+	int v;
 
 	if (ticks == 0)
 		return;
@@ -502,34 +505,56 @@ addupc_intr(ke, pc, ticks)
 /*
  * Much like before, but we can afford to take faults here.  If the
  * update fails, we simply turn off profiling.
+ * XXXKSE, don't use kse unless we got sched lock.
  */
 void
-addupc_task(ke, pc, ticks)
+addupc_task(struct kse *ke, uintptr_t pc, u_int ticks)
-	register struct kse *ke;
-	register uintptr_t pc;
-	u_int ticks;
 {
-	struct proc *p = ke->ke_proc;
+	struct proc *p = ke->ke_proc; 
-	register struct uprof *prof;
+	struct uprof *prof;
-	register caddr_t addr;
+	caddr_t addr;
-	register u_int i;
+	u_int i;
 	u_short v;
+	int stop = 0;
 
 	if (ticks == 0)
 		return;
 
+	PROC_LOCK(p);
+	mtx_lock_spin(&sched_lock);
+	if (!(p->p_sflag & PS_PROFIL)) {
+		mtx_unlock_spin(&sched_lock);
+		PROC_UNLOCK(p);
+		return;
+	}
+	p->p_profthreads++;
+	mtx_unlock_spin(&sched_lock);
+	PROC_UNLOCK(p);
 	prof = &p->p_stats->p_prof;
 	if (pc < prof->pr_off ||
-	    (i = PC_TO_INDEX(pc, prof)) >= prof->pr_size)
+	    (i = PC_TO_INDEX(pc, prof)) >= prof->pr_size) {
-		return;
+		goto out;
+	}
 
 	addr = prof->pr_base + i;
 	if (copyin(addr, &v, sizeof(v)) == 0) {
 		v += ticks;
 		if (copyout(&v, addr, sizeof(v)) == 0)
-			return;
+			goto out;
 	}
-	stopprofclock(p);
+	stop = 1;
+
+out:
+	PROC_LOCK(p);
+	if (--p->p_profthreads == 0) {
+		if (p->p_sflag & PS_STOPPROF) {
+			wakeup(&p->p_profthreads);
+			stop = 0;
+		}
+	}
+	if (stop)
+		stopprofclock(p);
+	PROC_UNLOCK(p);
 }
 
 #if defined(__i386__) && __GNUC__ >= 2

sys/sys/proc.h

@@ -569,12 +569,13 @@ struct proc {
 	u_int		p_swtime;	/* (j) Time swapped in or out. */
 	struct itimerval p_realtimer;	/* (h?/k?) Alarm timer. */
 	struct bintime	p_runtime;	/* (j) Real time. */
+	int		p_profthreads;	/* (c) Num threads in addupc_task */
 	int		p_traceflag;	/* (o) Kernel trace points. */
 	struct vnode	*p_tracep;	/* (c + o) Trace to vnode. */
 	sigset_t	p_siglist;	/* (c) Sigs arrived, not delivered. */
 	struct vnode	*p_textvp;	/* (b) Vnode of executable. */
 	char		p_lock;		/* (c) Proclock (prevent swap) count. */
-	struct klist p_klist;		/* (c) Knotes attached to this proc. */
+	struct klist	p_klist;	/* (c) Knotes attached to this proc. */
 	struct sigiolst	p_sigiolst;	/* (c) List of sigio sources. */
 	int		p_sigparent;	/* (c) Signal to parent on exit. */
 	sigset_t	p_oldsigmask;	/* (c) Saved mask from pre sigpause. */
@@ -669,6 +670,7 @@ struct proc {
 #define	PS_INMEM	0x00001	/* Loaded into memory. */
 #define	PS_XCPU		0x00002 /* Exceeded CPU limit. */
 #define	PS_PROFIL	0x00004	/* Has started profiling. */
+#define	PS_STOPPROF	0x00008	/* Has thread in requesting to stop prof */
 #define	PS_ALRMPEND	0x00020	/* Pending SIGVTALRM needs to be posted. */
 #define	PS_PROFPEND	0x00040	/* Pending SIGPROF needs to be posted. */
 #define	PS_SWAPINREQ	0x00100	/* Swapin request due to wakeup. */