Skip to contentSkip to conversation

futex: Fix hrtimer oops in futex_lock_pi()

PATCH In review Tony Lindgren 2 participants 4 messages Apr 14, 2017

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 futex: Fix hrtimer oops in futex_lock_pi()

Commit cfafcd117da0 ("futex: Rework futex_lock_pi() to use rt_mutex_*_proxy_lock()") caused a regression where things would occasionally randomly oops when restarting X:

Unable to handle kernel NULL pointer dereference at virtual address 00000000
...
Internal error: Oops: 80000005 [#1] SMP ARM
...
PC is at 0x0
LR is at __hrtimer_run_queues+0x138/0x58c
pc : [<00000000>]    lr : [<c01c7884>]    psr: 20000193

...
[<c01c7884>] (__hrtimer_run_queues) from [<c01c7f4c>] (hrtimer_interrupt+0xbc/0x210)
[<c01c7f4c>] (hrtimer_interrupt) from [<c010fcfc>]
...

When this happens, the hrtimer is not properly initialized and it's function is NULL. This happens because we now call hrtimer_start_expires() in futex_lock_pi() for the timer initialized with hrtimer_init_on_stack().

To fix it, let's pair the hrtimer_start_expires() with hrtimer_cancel() in the same function.

Fixes cfafcd117da0 ("futex: Rework futex_lock_pi() to use

rt_mutex_*_proxy_lock()")

---
 kernel/futex.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
kernel/futex.c +3 -1
@@ -2736,8 +2736,10 @@ static int futex_lock_pi(u32 __user *uaddr, unsigned int flags,
 out_put_key:
 	put_futex_key(&q.key);
 out:
-	if (to)
+	if (to) {
+		hrtimer_cancel(&to->timer);
 		destroy_hrtimer_on_stack(&to->timer);
+	}
 	return ret != -EINTR ? ret : -ERESTARTNOINTR;
 
 uaddr_faulted:
Signature · 1 line
2.12.2
#1 lore raw
Quoted 17 lines · Tony Lindgren

Commit cfafcd117da0 ("futex: Rework futex_lock_pi() to use rt_mutex_*_proxy_lock()") caused a regression where things would occasionally randomly oops when restarting X:

Unable to handle kernel NULL pointer dereference at virtual address 00000000
...
Internal error: Oops: 80000005 [#1] SMP ARM
...
PC is at 0x0
LR is at __hrtimer_run_queues+0x138/0x58c
pc : [<00000000>]    lr : [<c01c7884>]    psr: 20000193

...
[<c01c7884>] (__hrtimer_run_queues) from [<c01c7f4c>] (hrtimer_interrupt+0xbc/0x210)
[<c01c7f4c>] (hrtimer_interrupt) from [<c010fcfc>]
...

Quoted 6 lines

When this happens, the hrtimer is not properly initialized and it's function is NULL. This happens because we now call hrtimer_start_expires() in futex_lock_pi() for the timer initialized with hrtimer_init_on_stack().

To fix it, let's pair the hrtimer_start_expires() with hrtimer_cancel() in the same function.

Already fixed:

https://lkml.kernel.org/r/[email protected]

Thanks for the patch though.

#2 lore raw
Quoted 22 lines · Peter Zijlstra
Quoted 21 lines · Tony Lindgren

Commit cfafcd117da0 ("futex: Rework futex_lock_pi() to use rt_mutex_*_proxy_lock()") caused a regression where things would occasionally randomly oops when restarting X:

Unable to handle kernel NULL pointer dereference at virtual address 00000000
...
Internal error: Oops: 80000005 [#1] SMP ARM
...
PC is at 0x0
LR is at __hrtimer_run_queues+0x138/0x58c
pc : [<00000000>]    lr : [<c01c7884>]    psr: 20000193

...
[<c01c7884>] (__hrtimer_run_queues) from [<c01c7f4c>] (hrtimer_interrupt+0xbc/0x210)
[<c01c7f4c>] (hrtimer_interrupt) from [<c010fcfc>]
...

When this happens, the hrtimer is not properly initialized and it's function is NULL. This happens because we now call hrtimer_start_expires() in futex_lock_pi() for the timer initialized with hrtimer_init_on_stack().

Quoted 8 lines
Quoted 2 lines

To fix it, let's pair the hrtimer_start_expires() with hrtimer_cancel() in the same function.

Already fixed:

https://lkml.kernel.org/r/[email protected]

Thanks for the patch though.

Oh OK thanks. It seems to be missing in Linux next though.

Regards,

Tony

#3 lore raw
Quoted 26 lines · Tony Lindgren
Quoted 25 lines · Peter Zijlstra
Quoted 23 lines · Tony Lindgren

Commit cfafcd117da0 ("futex: Rework futex_lock_pi() to use rt_mutex_*_proxy_lock()") caused a regression where things would occasionally randomly oops when restarting X:

Unable to handle kernel NULL pointer dereference at virtual address 00000000
...
Internal error: Oops: 80000005 [#1] SMP ARM
...
PC is at 0x0
LR is at __hrtimer_run_queues+0x138/0x58c
pc : [<00000000>]    lr : [<c01c7884>]    psr: 20000193

...
[<c01c7884>] (__hrtimer_run_queues) from [<c01c7f4c>] (hrtimer_interrupt+0xbc/0x210)
[<c01c7f4c>] (hrtimer_interrupt) from [<c010fcfc>]
...

When this happens, the hrtimer is not properly initialized and it's function is NULL. This happens because we now call hrtimer_start_expires() in futex_lock_pi() for the timer initialized with hrtimer_init_on_stack().

To fix it, let's pair the hrtimer_start_expires() with hrtimer_cancel() in the same function.

Quoted 7 lines
Quoted 5 lines

Already fixed:

https://lkml.kernel.org/r/[email protected]

Thanks for the patch though.

Oh OK thanks. It seems to be missing in Linux next though.

Oh it was just committed, I see it now in tip after git fetch. So should be heading into nex soon.

Tony