6217375 CPU unconfigure, t_disp_queue, and restore_mstate() duke it out

6294464 machine hung with threads spining in thread_lock()on a cpu dispatch lock that never gets released.

7 files changed, 34 insertions, 26 deletions

diff --git a/usr/src/uts/common/disp/disp.c b/usr/src/uts/common/disp/disp.c
index 6bb9debbe2..aee7512c08 100644
--- a/usr/src/uts/common/disp/disp.c
+++ b/usr/src/uts/common/disp/disp.c
@@ -132,7 +132,6 @@ id_t	defaultcid;	/* system "default" class; see dispadmin(1M) */
 
 disp_lock_t	transition_lock;	/* lock on transitioning threads */
 disp_lock_t	stop_lock;		/* lock on stopped threads */
-disp_lock_t	shuttle_lock;		/* lock on shuttle objects */
 
 static void		cpu_dispqalloc(int numpris);
 
@@ -202,7 +201,6 @@ dispinit(void)
 	DISP_LOCK_INIT(&transition_lock);
 	disp_lock_enter_high(&transition_lock);
 	DISP_LOCK_INIT(&stop_lock);
-	DISP_LOCK_INIT(&shuttle_lock);
 
 	mutex_enter(&cpu_lock);
 	CPU->cpu_disp->disp_maxrunpri = -1;
diff --git a/usr/src/uts/common/disp/fss.c b/usr/src/uts/common/disp/fss.c
index 2fc5fe6e2f..f2b77f7ea7 100644
--- a/usr/src/uts/common/disp/fss.c
+++ b/usr/src/uts/common/disp/fss.c
@@ -2167,7 +2167,8 @@ fss_tick(kthread_t *t)
 				fssproc->fss_flags |= FSSBACKQ;
 				cpu_surrender(t);
 			}
-		} else if (t->t_pri < t->t_disp_queue->disp_maxrunpri) {
+		} else if (t->t_state == TS_ONPROC &&
+			    t->t_pri < t->t_disp_queue->disp_maxrunpri) {
 			/*
 			 * If there is a higher-priority thread which is
 			 * waiting for a processor, then thread surrenders
diff --git a/usr/src/uts/common/disp/fx.c b/usr/src/uts/common/disp/fx.c
index 2e7f27c399..62b5eb1da2 100644
--- a/usr/src/uts/common/disp/fx.c
+++ b/usr/src/uts/common/disp/fx.c
@@ -1382,7 +1382,8 @@ fx_tick(kthread_t *t)
 			fxpp->fx_flags |= FXBACKQ;
 			cpu_surrender(t);
 		}
-	} else if (t->t_pri < t->t_disp_queue->disp_maxrunpri) {
+	} else if (t->t_state == TS_ONPROC &&
+		    t->t_pri < t->t_disp_queue->disp_maxrunpri) {
 		fxpp->fx_flags |= FXBACKQ;
 		cpu_surrender(t);
 	}
diff --git a/usr/src/uts/common/disp/rt.c b/usr/src/uts/common/disp/rt.c
index 128cdeef1b..2b60fbe24e 100644
--- a/usr/src/uts/common/disp/rt.c
+++ b/usr/src/uts/common/disp/rt.c
@@ -994,7 +994,7 @@ rt_tick(kthread_t *t)
 
 	thread_lock(t);
 	if ((rtpp->rt_pquantum != RT_TQINF && --rtpp->rt_timeleft == 0) ||
-	    (DISP_MUST_SURRENDER(t))) {
+	    (t->t_state == TS_ONPROC && DISP_MUST_SURRENDER(t))) {
 		if (rtpp->rt_timeleft == 0 && rtpp->rt_tqsignal) {
 			thread_unlock(t);
 			sigtoproc(ttoproc(t), t, rtpp->rt_tqsignal);
diff --git a/usr/src/uts/common/disp/shuttle.c b/usr/src/uts/common/disp/shuttle.c
index 1c0dd9e65e..47b6e17e66 100644
--- a/usr/src/uts/common/disp/shuttle.c
+++ b/usr/src/uts/common/disp/shuttle.c
@@ -20,7 +20,7 @@
  * CDDL HEADER END
  */
 /*
- * Copyright 2004 Sun Microsystems, Inc.  All rights reserved.
+ * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  */
 
@@ -39,6 +39,7 @@
 #include <sys/cpuvar.h>
 #include <sys/sdt.h>
 
+static	disp_lock_t	shuttle_lock;	/* lock on shuttle objects */
 
 /*
  * Place the thread in question on the run q.
@@ -86,7 +87,7 @@ shuttle_resume(kthread_t *t, kmutex_t *l)
 {
 	klwp_t	*lwp = ttolwp(curthread);
 	cpu_t	*cp;
-	extern disp_lock_t	shuttle_lock;
+	disp_lock_t *oldtlp;
 
 	thread_lock(curthread);
 	disp_lock_enter_high(&shuttle_lock);
@@ -115,16 +116,31 @@ shuttle_resume(kthread_t *t, kmutex_t *l)
 	DTRACE_SCHED1(wakeup, kthread_t *, t);
 	DTRACE_SCHED(sleep);
 	THREAD_SLEEP(curthread, &shuttle_lock);
+	disp_lock_exit_high(&shuttle_lock);
 
-	/* Update ustate records (there is no waitrq obviously) */
-
+	/*
+	 * Update ustate records (there is no waitrq obviously)
+	 */
 	(void) new_mstate(curthread, LMS_SLEEP);
+
+	thread_lock_high(t);
+	oldtlp = t->t_lockp;
+
 	restore_mstate(t);
 	t->t_flag &= ~T_WAKEABLE;
 	t->t_wchan0 = NULL;
 	t->t_sobj_ops = NULL;
 
 	/*
+	 * Make sure we end up on the right CPU if we are dealing with bound
+	 * CPU's or processor partitions.
+	 */
+	if (t->t_bound_cpu != NULL || t->t_cpupart != cp->cpu_part) {
+		aston(t);
+		cp->cpu_runrun = 1;
+	}
+
+	/*
 	 * We re-assign t_disp_queue and t_lockp of 't' here because
 	 * 't' could have been preempted.
 	 */
@@ -134,15 +150,12 @@ shuttle_resume(kthread_t *t, kmutex_t *l)
 	}
 
 	/*
-	 * Make sure we end up on the right CPU if we are dealing with bound
-	 * CPU's or processor partitions.
+	 * We can't call thread_unlock_high() here because t's thread lock
+	 * could have changed by thread_onproc() call above to point to
+	 * CPU->cpu_thread_lock.
 	 */
-	if (t->t_bound_cpu != NULL || t->t_cpupart != cp->cpu_part) {
-		aston(t);
-		cp->cpu_runrun = 1;
-	}
+	disp_lock_exit_high(oldtlp);
 
-	disp_lock_exit_high(&shuttle_lock);
 	mutex_exit(l);
 	/*
 	 * Make sure we didn't receive any important events while
@@ -151,6 +164,7 @@ shuttle_resume(kthread_t *t, kmutex_t *l)
 	if (lwp &&
 	    (ISSIG(curthread, JUSTLOOKING) || MUSTRETURN(curproc, curthread)))
 		setrun(curthread);
+
 	swtch_to(t);
 	/*
 	 * Caller must check for ISSIG/lwp_sysabort conditions
@@ -167,7 +181,6 @@ void
 shuttle_swtch(kmutex_t *l)
 {
 	klwp_t	*lwp = ttolwp(curthread);
-	extern disp_lock_t	shuttle_lock;
 
 	thread_lock(curthread);
 	disp_lock_enter_high(&shuttle_lock);
@@ -205,8 +218,6 @@ shuttle_sleep(kthread_t *t)
 	klwp_t	*lwp = ttolwp(t);
 	proc_t	*p = ttoproc(t);
 
-	extern disp_lock_t	shuttle_lock;
-
 	thread_lock(t);
 	disp_lock_enter_high(&shuttle_lock);
 	if (lwp != NULL) {
diff --git a/usr/src/uts/common/disp/ts.c b/usr/src/uts/common/disp/ts.c
index d0f09f5817..9a10dac7ad 100644
--- a/usr/src/uts/common/disp/ts.c
+++ b/usr/src/uts/common/disp/ts.c
@@ -20,7 +20,7 @@
  * CDDL HEADER END
  */
 /*
- * Copyright 2004 Sun Microsystems, Inc.  All rights reserved.
+ * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  */
 
@@ -1701,7 +1701,8 @@ ts_tick(kthread_t *t)
 			}
 			TRACE_2(TR_FAC_DISP, TR_TICK,
 			    "tick:tid %p old pri %d", t, oldpri);
-		} else if (t->t_pri < t->t_disp_queue->disp_maxrunpri) {
+		} else if (t->t_state == TS_ONPROC &&
+			    t->t_pri < t->t_disp_queue->disp_maxrunpri) {
 			tspp->ts_flags |= TSBACKQ;
 			cpu_surrender(t);
 		}
diff --git a/usr/src/uts/common/os/msacct.c b/usr/src/uts/common/os/msacct.c
index 1ca2347c15..475913ac76 100644
--- a/usr/src/uts/common/os/msacct.c
+++ b/usr/src/uts/common/os/msacct.c
@@ -534,7 +534,6 @@ restore_mstate(kthread_t *t)
 	hrtime_t waitrq;
 	hrtime_t newtime;
 	hrtime_t oldtime;
-	struct cpu *cpup;
 
 	if ((lwp = ttolwp(t)) == NULL)
 		return;
@@ -601,11 +600,8 @@ restore_mstate(kthread_t *t)
 	/*
 	 * Update the WAIT_CPU timer and per-cpu waitrq total.
 	 */
-	cpup = t->t_disp_queue->disp_cpu;
-	if (cpup == NULL)
-		cpup = t->t_cpu;
 	ms->ms_acct[LMS_WAIT_CPU] += (curtime - waitrq);
-	cpup->cpu_waitrq += (curtime - waitrq);
+	CPU->cpu_waitrq += (curtime - waitrq);
 	ms->ms_state_start = curtime;
 }