[PATCH RT] add might_sleep in rt_spin_lock_fastlock

[Steven Rostedt <rostedt@goodmis.org>]

[-- Attachment #1: Type: text/plain, Size: 983 bytes --]

Ingo,

Every so often we get bit by a bug "scheduling in atomic", and it comes
from a rtmutex spin_lock.  The bug only happens when that lock has
contention, so we miss it a lot.

This patch adds a might_sleep() to the rt_spin_lock_fastlock to find
bugs where we can schedule in atomic.

The one place that exists now is from do_page_fault and sending a
signal.  I wrote a simple crash program that segfaults (attached) and
with this patch, I get the warning.

-- Steve

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>

Index: linux-2.6.21-rt9/kernel/rtmutex.c
===================================================================
--- linux-2.6.21-rt9.orig/kernel/rtmutex.c
+++ linux-2.6.21-rt9/kernel/rtmutex.c
@@ -602,6 +602,8 @@ static inline void
 rt_spin_lock_fastlock(struct rt_mutex *lock,
 		void fastcall (*slowfn)(struct rt_mutex *lock))
 {
+	might_sleep();
+
 	if (likely(rt_mutex_cmpxchg(lock, NULL, current)))
 		rt_mutex_deadlock_account_lock(lock, current);
 	else


[-- Attachment #2: crashme.c --]
[-- Type: text/x-csrc, Size: 231 bytes --]

#include <stdio.h>
#include <stdlib.h>
#include <string.h>

int main (int argc, char **argv)
{
	unsigned long *p = (void*)-1;

	printf("crashing...\n");

	/* nighty night */
	*p = 0xbed;

	printf("did we crash??\n");	
	return 0;
}