commit ee4c68e21f83ddd8b40d058a8caf16cd1381760e
Author: Andriy Gapon <avg@icyb.net.ua>
Date:   Mon May 9 00:08:44 2011 +0300

    change generic_stop_cpus to use the same approach as hard_stop_cpus

diff --git a/sys/amd64/amd64/mp_machdep.c b/sys/amd64/amd64/mp_machdep.c
index 35c8a14..7125c75 100644
--- a/sys/amd64/amd64/mp_machdep.c
+++ b/sys/amd64/amd64/mp_machdep.c
@@ -1410,6 +1410,10 @@ cpustop_handler(void)
 
 	cpu = PCPU_GET(cpuid);
 
+	/* Just return if this is a belated NMI */
+	if (!CPU_ISSET(cpu, &stopping_cpus))
+		return;
+
 	savectx(&stoppcbs[cpu]);
 
 	/* Indicate that we are stopped */
diff --git a/sys/i386/i386/mp_machdep.c b/sys/i386/i386/mp_machdep.c
index 607c0ef..1ba7f08 100644
--- a/sys/i386/i386/mp_machdep.c
+++ b/sys/i386/i386/mp_machdep.c
@@ -1509,6 +1509,10 @@ cpustop_handler(void)
 
 	cpu = PCPU_GET(cpuid);
 
+	/* Just return if this is a belated NMI */
+	if (!CPU_ISSET(cpu, &stopping_cpus))
+		return;
+
 	savectx(&stoppcbs[cpu]);
 
 	/* Indicate that we are stopped */
diff --git a/sys/kern/subr_smp.c b/sys/kern/subr_smp.c
index 288ee4a..1046c9b 100644
--- a/sys/kern/subr_smp.c
+++ b/sys/kern/subr_smp.c
@@ -56,6 +56,7 @@ __FBSDID("$FreeBSD$");
 volatile cpuset_t stopped_cpus;
 volatile cpuset_t started_cpus;
 volatile cpuset_t suspended_cpus;
+volatile cpuset_t stopping_cpus;
 volatile cpuset_t hard_stopped_cpus;
 volatile cpuset_t hard_started_cpus;
 volatile cpuset_t hard_stopping_cpus;
@@ -209,7 +210,8 @@ generic_stop_cpus(cpuset_t map, u_int type)
 #ifdef KTR
 	char cpusetbuf[CPUSETBUFSIZ];
 #endif
-	static volatile u_int stopping_cpu = NOCPU;
+	static volatile u_int stopper_cpu = NOCPU;
+	int cpu;
 	int i;
 	volatile cpuset_t *cpus;
 
@@ -227,13 +229,28 @@ generic_stop_cpus(cpuset_t map, u_int type)
 	CTR2(KTR_SMP, "stop_cpus(%s) with %u type",
 	    cpusetobj_strprint(cpusetbuf, &map), type);
 
-	if (stopping_cpu != PCPU_GET(cpuid))
-		while (atomic_cmpset_int(&stopping_cpu, NOCPU,
-		    PCPU_GET(cpuid)) == 0)
-			while (stopping_cpu != NOCPU)
-				cpu_spinwait(); /* spin */
+	/* Ensure non-preemtable context, just in case. */
+	spinlock_enter();
+
+	cpu = PCPU_GET(cpuid);
+
+	if (cpu != stopper_cpu) {
+		while (atomic_cmpset_int(&stopper_cpu, NOCPU, cpu) == 0)
+			while (stopper_cpu != NOCPU) {
+				if (CPU_ISSET(cpu, &stopping_cpus))
+					cpustop_handler();
+				else
+					cpu_spinwait();
+			}
+	} else {
+		/*
+		 * Recursion here is not expected.
+		 */
+		panic("cpu stop recursion\n");
+	}
 
 	/* send the stop IPI to all CPUs in map */
+	stopping_cpus = map;
 	ipi_selected(map, type);
 
 #if defined(__amd64__) || defined(__i386__)
@@ -254,7 +271,8 @@ generic_stop_cpus(cpuset_t map, u_int type)
 		}
 	}
 
-	stopping_cpu = NOCPU;
+	stopper_cpu = NOCPU;
+	spinlock_exit();
 	return (1);
 }
 
diff --git a/sys/sys/smp.h b/sys/sys/smp.h
index 267dafb..a9c37e4 100644
--- a/sys/sys/smp.h
+++ b/sys/sys/smp.h
@@ -76,6 +76,7 @@ extern int smp_cpus;
 extern volatile cpuset_t started_cpus;
 extern volatile cpuset_t stopped_cpus;
 extern volatile cpuset_t suspended_cpus;
+extern volatile cpuset_t stopping_cpus;
 extern volatile cpuset_t hard_started_cpus;
 extern volatile cpuset_t hard_stopped_cpus;
 extern volatile cpuset_t hard_stopping_cpus;