Get kernel profiling on SMP systems closer to working by replacing the

mcount spin mutex with a very simple non-recursive spinlock implemented
using atomic operations.

4 files changed, 8 insertions, 25 deletions

diff --git a/sys/amd64/amd64/mp_machdep.c b/sys/amd64/amd64/mp_machdep.c
index 787a472..0a0de69 100644
--- a/sys/amd64/amd64/mp_machdep.c
+++ b/sys/amd64/amd64/mp_machdep.c
@@ -325,7 +325,7 @@ static void	release_aps(void *dummy);
 struct mtx		imen_mtx;
 
 /* lock region used by kernel profiling */
-struct mtx		mcount_mtx;
+int	mcount_lock;
 
 #ifdef USE_COMLOCK
 /* locks com (tty) data/hardware accesses: a FASTINTR() */
@@ -335,12 +335,6 @@ struct mtx		com_mtx;
 static void
 init_locks(void)
 {
-	/*
-	 * XXX The mcount mutex probably needs to be statically initialized,
-	 * since it will be used even in the function calls that get us to this
-	 * point.
-	 */
-	mtx_init(&mcount_mtx, "mcount", MTX_DEF);
 
 #ifdef USE_COMLOCK
 	mtx_init(&com_mtx, "com", MTX_SPIN);
diff --git a/sys/amd64/amd64/mptable.c b/sys/amd64/amd64/mptable.c
index 787a472..0a0de69 100644
--- a/sys/amd64/amd64/mptable.c
+++ b/sys/amd64/amd64/mptable.c
@@ -325,7 +325,7 @@ static void	release_aps(void *dummy);
 struct mtx		imen_mtx;
 
 /* lock region used by kernel profiling */
-struct mtx		mcount_mtx;
+int	mcount_lock;
 
 #ifdef USE_COMLOCK
 /* locks com (tty) data/hardware accesses: a FASTINTR() */
@@ -335,12 +335,6 @@ struct mtx		com_mtx;
 static void
 init_locks(void)
 {
-	/*
-	 * XXX The mcount mutex probably needs to be statically initialized,
-	 * since it will be used even in the function calls that get us to this
-	 * point.
-	 */
-	mtx_init(&mcount_mtx, "mcount", MTX_DEF);
 
 #ifdef USE_COMLOCK
 	mtx_init(&com_mtx, "com", MTX_SPIN);
diff --git a/sys/amd64/include/mptable.h b/sys/amd64/include/mptable.h
index 787a472..0a0de69 100644
--- a/sys/amd64/include/mptable.h
+++ b/sys/amd64/include/mptable.h
@@ -325,7 +325,7 @@ static void	release_aps(void *dummy);
 struct mtx		imen_mtx;
 
 /* lock region used by kernel profiling */
-struct mtx		mcount_mtx;
+int	mcount_lock;
 
 #ifdef USE_COMLOCK
 /* locks com (tty) data/hardware accesses: a FASTINTR() */
@@ -335,12 +335,6 @@ struct mtx		com_mtx;
 static void
 init_locks(void)
 {
-	/*
-	 * XXX The mcount mutex probably needs to be statically initialized,
-	 * since it will be used even in the function calls that get us to this
-	 * point.
-	 */
-	mtx_init(&mcount_mtx, "mcount", MTX_DEF);
 
 #ifdef USE_COMLOCK
 	mtx_init(&com_mtx, "com", MTX_SPIN);
diff --git a/sys/amd64/include/profile.h b/sys/amd64/include/profile.h
index 5f2a780..883daf5 100644
--- a/sys/amd64/include/profile.h
+++ b/sys/amd64/include/profile.h
@@ -64,10 +64,11 @@
 #else
 #define	MCOUNT_DECL(s)	u_long s;
 #ifdef SMP
-#define	MCOUNT_ENTER(s)	{ s = read_eflags(); \
- 			  __asm __volatile("cli" : : : "memory"); \
-			  mtx_lock(&mcount_mtx); }
-#define	MCOUNT_EXIT(s)	{ mtx_unlock(&mcount_mtx); write_eflags(s); }
+#define	MCOUNT_ENTER(s)	{ s = read_eflags(); disable_intr(); \
+ 			  while (!atomic_cmpset_acq_int(&mcount_lock, 0, 1) \
+			  	/* nothing */ ; }
+#define	MCOUNT_EXIT(s)	{ atomic_store_rel_int(&mcount_lock, 0); \
+			  write_eflags(s); }
 #else
 #define	MCOUNT_ENTER(s)	{ s = read_eflags(); disable_intr(); }
 #define	MCOUNT_EXIT(s)	(write_eflags(s))