Re: [PATCH v2 1/2] sched/rtmutex/deadline: Fix a PI crash for deadline tasks

[Xunlei Pang <xpang@redhat.com>]

On 2016/04/07 at 02:14, Peter Zijlstra wrote:
> On Wed, Apr 06, 2016 at 08:59:15PM +0800, Xunlei Pang wrote:
>> A crash happened while I'm playing with deadline PI rtmutex.
>>
>>     BUG: unable to handle kernel NULL pointer dereference at 0000000000000018
>>     IP: [<ffffffff810eeb8f>] rt_mutex_get_top_task+0x1f/0x30
>>     PGD 232a75067 PUD 230947067 PMD 0
>>     Oops: 0000 [#1] SMP
>>     CPU: 1 PID: 10994 Comm: a.out Not tainted
>>
>>     Call Trace:
>>     [<ffffffff810cf8aa>] ? enqueue_task_dl+0x2a/0x320
>>     [<ffffffff810b658c>] enqueue_task+0x2c/0x80
>>     [<ffffffff810ba763>] activate_task+0x23/0x30
>>     [<ffffffff810d0ab5>] pull_dl_task+0x1d5/0x260
>>     [<ffffffff810d0be6>] pre_schedule_dl+0x16/0x20
>>     [<ffffffff8164e783>] __schedule+0xd3/0x900
>>     [<ffffffff8164efd9>] schedule+0x29/0x70
>>     [<ffffffff8165035b>] __rt_mutex_slowlock+0x4b/0xc0
>>     [<ffffffff81650501>] rt_mutex_slowlock+0xd1/0x190
>>     [<ffffffff810eeb33>] rt_mutex_timed_lock+0x53/0x60
>>     [<ffffffff810ecbfc>] futex_lock_pi.isra.18+0x28c/0x390
>>     [<ffffffff810cfa15>] ? enqueue_task_dl+0x195/0x320
>>     [<ffffffff810d0bac>] ? prio_changed_dl+0x6c/0x90
>>     [<ffffffff810ed8b0>] do_futex+0x190/0x5b0
>>     [<ffffffff810edd50>] SyS_futex+0x80/0x180
>>     [<ffffffff8165a089>] system_call_fastpath+0x16/0x1b
>>     RIP  [<ffffffff810eeb8f>] rt_mutex_get_top_task+0x1f/0x30
>>
>> This is because rt_mutex_enqueue_pi() and rt_mutex_dequeue_pi()
>> are only protected by pi_lock when operating pi waiters, while
>> rt_mutex_get_top_task() will access them with rq lock held but
>> not holding pi_lock.
>>
>> In order to tackle it, we introduce a new pointer "pi_top_task"
>> in task_struct, and update it to be the top waiter task(this waiter
>> is updated under pi_lock) in rt_mutex_setprio() which is under 
>> both pi_lock and rq lock, then ensure all its accessers be under 
>> rq lock (or pi_lock), this can safely fix the crash.
>>
>> This patch is originated from "Peter Zijlstra", with several
>> tweaks and improvements by me.
> I would suggest doing the rt_mutex_postunlock() thing as a separate
> patch, it has some merit outside of these changes and reduces the total
> amount of complexity in this patch.

I think the code change is necessary , as it avoids the invalid task_struct
access issue introduced by PATCH1.

Do you mean just making the code refactor using rt_mutex_postunlock()
as a separate patch? or do I miss something?

>
> Also, I would very much like Thomas to ack this patch before I take it,
> but since its conference season this might take a little while. Esp. the
> change marked with XXX is something that I'm not sure about.

I'm ok with this change, waiting for Thomas.

Regards,
Xunlei