Hi
Please apply this before 3.19 is released.
Mikulas
The patch e22b886a8a43b147e1994a9f970f678fc0df2033 introduced a bug in the
raid5 subsystem.
The function raid5_quiesce (and resize_stripes) calls
lock_all_device_hash_locks_irq that disables interrupts and takes a few
spinlocks, then it calls wait_event_cmd with cmd1
unlock_all_device_hash_locks_irq(conf) and cmd2
lock_all_device_hash_locks_irq(conf). cmd1 unlocks the spinlocks and
enables interrupts, cmd2 disables interrupts and locks the spinlock.
The patch e22b886a8a43b147e1994a9f970f678fc0df2033 adds might_sleep() to a
position where spinlocks are taken, thus it introduces a bug.
This patch removes might_sleep() from wait_event_cmd.
Signed-off-by: Mikulas Patocka <[email protected]>
---
include/linux/wait.h | 1 -
1 file changed, 1 deletion(-)
Index: linux-2.6/include/linux/wait.h
===================================================================
--- linux-2.6.orig/include/linux/wait.h 2014-12-30 01:19:25.564231262 +0100
+++ linux-2.6/include/linux/wait.h 2015-02-02 15:30:16.766354658 +0100
@@ -363,7 +363,6 @@ do { \
*/
#define wait_event_cmd(wq, condition, cmd1, cmd2) \
do { \
- might_sleep(); \
if (condition) \
break; \
__wait_event_cmd(wq, condition, cmd1, cmd2); \
On Mon, 2 Feb 2015 09:39:02 -0500 (EST) Mikulas Patocka <[email protected]>
wrote:
> Hi
>
> Please apply this before 3.19 is released.
>
> Mikulas
>
>
> The patch e22b886a8a43b147e1994a9f970f678fc0df2033 introduced a bug in the
> raid5 subsystem.
>
> The function raid5_quiesce (and resize_stripes) calls
> lock_all_device_hash_locks_irq that disables interrupts and takes a few
> spinlocks, then it calls wait_event_cmd with cmd1
> unlock_all_device_hash_locks_irq(conf) and cmd2
> lock_all_device_hash_locks_irq(conf). cmd1 unlocks the spinlocks and
> enables interrupts, cmd2 disables interrupts and locks the spinlock.
>
> The patch e22b886a8a43b147e1994a9f970f678fc0df2033 adds might_sleep() to a
> position where spinlocks are taken, thus it introduces a bug.
>
> This patch removes might_sleep() from wait_event_cmd.
>
> Signed-off-by: Mikulas Patocka <[email protected]>
>
> ---
> include/linux/wait.h | 1 -
> 1 file changed, 1 deletion(-)
>
> Index: linux-2.6/include/linux/wait.h
> ===================================================================
> --- linux-2.6.orig/include/linux/wait.h 2014-12-30 01:19:25.564231262 +0100
> +++ linux-2.6/include/linux/wait.h 2015-02-02 15:30:16.766354658 +0100
> @@ -363,7 +363,6 @@ do { \
> */
> #define wait_event_cmd(wq, condition, cmd1, cmd2) \
> do { \
> - might_sleep(); \
> if (condition) \
> break; \
> __wait_event_cmd(wq, condition, cmd1, cmd2); \
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
I support this patch.
However in case it doesn't get in, I've queued up a patch to change raid5.c
to use __wait_event_cmd instead...
NeilBrown
On Tue, Feb 03, 2015 at 09:12:53AM +1100, NeilBrown wrote:
> I support this patch.
It should hopefully already be en-route to tip.
> However in case it doesn't get in, I've queued up a patch to change raid5.c
> to use __wait_event_cmd instead...
So those primitives are useful in their own; their typical use would be
when you already know @cond to be false and want to avoid the extra
invocation.
Commit-ID: 3e87523897e18a3e17fc8955ed795188be737ff1
Gitweb: http://git.kernel.org/tip/3e87523897e18a3e17fc8955ed795188be737ff1
Author: Mikulas Patocka <[email protected]>
AuthorDate: Mon, 2 Feb 2015 09:39:02 -0500
Committer: Ingo Molnar <[email protected]>
CommitDate: Tue, 3 Feb 2015 12:14:25 +0100
sched/wait: Remove might_sleep() from wait_event_cmd()
The patch e22b886a8a43 ("sched/wait: Add might_sleep() checks")
introduced a bug in the raid5 subsystem.
The function raid5_quiesce() (and resize_stripes()) uses the 'cmd'
part to release and acquire a spinlock (so we call the sleep
primitives in atomic context), and therefore we cannot do the
might_sleep() check.
Remove it.
Fixes: e22b886a8a43 ("sched/wait: Add might_sleep() checks")
Signed-off-by: Mikulas Patocka <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Cc: Linus Torvalds <[email protected]>
Link: http://lkml.kernel.org/r/alpine.LRH.2.02.1502020935580.13510@file01.intranet.prod.int.rdu2.redhat.com
Signed-off-by: Ingo Molnar <[email protected]>
---
include/linux/wait.h | 1 -
1 file changed, 1 deletion(-)
diff --git a/include/linux/wait.h b/include/linux/wait.h
index 2232ed1..37423e0 100644
--- a/include/linux/wait.h
+++ b/include/linux/wait.h
@@ -363,7 +363,6 @@ do { \
*/
#define wait_event_cmd(wq, condition, cmd1, cmd2) \
do { \
- might_sleep(); \
if (condition) \
break; \
__wait_event_cmd(wq, condition, cmd1, cmd2); \
On Tue, 3 Feb 2015 12:06:30 +0100 Peter Zijlstra <[email protected]> wrote:
> On Tue, Feb 03, 2015 at 09:12:53AM +1100, NeilBrown wrote:
> > I support this patch.
>
> It should hopefully already be en-route to tip.
Hopefully that means it will be in 3.19-final...
Thanks, I'll drop my __wait_event_cmd patch then.
NeilBrown
>
> > However in case it doesn't get in, I've queued up a patch to change raid5.c
> > to use __wait_event_cmd instead...
>
> So those primitives are useful in their own; their typical use would be
> when you already know @cond to be false and want to avoid the extra
> invocation.
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/