Commit Doug White and Alan Cox's fix for the cross-ipi smp deadlock.

We were obtaining different spin mutexes (which disable interrupts after
aquisition) and spin waiting for delivery.  For example, KSE processes
do LDT operations which use smp_rendezvous, while other parts of the
system are doing things like tlb shootdowns with a different mutex.

This patch uses the common smp_rendezvous mutex for all MD home-grown
IPIs that spinwait for delivery.  Having the single mutex means that
the spinloop to aquire it will enable interrupts periodically, thus
avoiding the cross-ipi deadlock.

Obtained from: dwhite, alc
Reviewed by:   jhb

1 files changed, 8 insertions, 1 deletions

diff --git a/sys/kern/subr_smp.c b/sys/kern/subr_smp.c
index b1be3fa..945ef6f 100644
--- a/sys/kern/subr_smp.c
+++ b/sys/kern/subr_smp.c
@@ -102,7 +102,14 @@ static void (*smp_rv_action_func)(void *arg);
 static void (*smp_rv_teardown_func)(void *arg);
 static void *smp_rv_func_arg;
 static volatile int smp_rv_waiters[2];
-static struct mtx smp_rv_mtx;
+
+/* 
+ * Shared mutex to restrict busywaits between smp_rendezvous() and
+ * smp(_targeted)_tlb_shootdown().  A deadlock occurs if both of these
+ * functions trigger at once and cause multiple CPUs to busywait with
+ * interrupts disabled. 
+ */
+struct mtx smp_rv_mtx;
 
 /*
  * Let the MD SMP code initialize mp_maxid very early if it can.