Re: [patch] voluntary-preempt-2.6.8-rc4-O7

[Ingo Molnar <mingo@elte.hu>]

* Lee Revell <rlrevell@joe-job.com> wrote:

> In some of the traces, like this one:
> 
> http://krustophenia.net/testresults.php?dataset=2.6.8-rc4-bk3-O7#/var/www/2.6.8-rc4-bk3-O7/mount_reiserfs_latency_trace.txt
> 
> there are calls to voluntary_resched.  How is this possible?  Does it
> mean that we called voluntary_resched while holding a spinlock, where
> we needed to call voluntary_preempt_lock(&foo_lock), and thus failed
> to reschedule?

voluntary_resched() was probably called as part of the might_sleep_if()
done in mm/slab.c. If you have CONFIG_DEBUG_SPINLOCK_SLEEP enabled then
the kernel should have complained about 'Debug: sleeping function ...'.

but what i think happened here is that reiserfs still uses
lock_kernel()/unlock_kernel() quite alot (eg. ext3 or xfs doesnt), which
from a preemptability POV is just as much of a critical section as a
spinlock, but processes can sleep (the scheduler auto-releases and
auto-reacquires the big kernel lock).

	Ingo

--- linux/kernel/sched.c.orig
+++ linux/kernel/sched.c
@@ -3210,11 +3210,11 @@ __setup("voluntary-preempt=", voluntary_
 
 int __sched voluntary_resched(void)
 {
-	if (kernel_preemption || !voluntary_preemption)
-		return 0;
 #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP
 	__might_sleep(__FILE__, __LINE__);
 #endif
+	if (kernel_preemption || !voluntary_preemption)
+		return 0;
 	/*
 	 * The system_state check is somewhat ugly but we might be
 	 * called during early boot when we are not yet ready to reschedule.