Re: [PATCH] sched: correct testing need_resched in mutex_spin_on_owner()

[Hillf Danton <dhillf@gmail.com>]

On Tue, Jun 7, 2011 at 9:59 PM, Peter Zijlstra <peterz@infradead.org> wrote:
> On Tue, 2011-06-07 at 15:47 +0200, Peter Zijlstra wrote:
>> On Tue, 2011-06-07 at 21:41 +0800, Hillf Danton wrote:
>> > It is suppose to check the owner task that is not absolutly running on the
>> > local CPU,
>>
>> Oh, why do you think so?
>>
>> > and if NEED_RESCHED is happenly set on the current task of local
>> > CPU, we get incorrect result.
>>
>> Only if your above assumption holds, which it doesn't. It explicitly
>> checks to see if _this_ cpu needs a resched while spinning, if so it
>> bails the spinning and calls schedule in the lock slow path.
>>
>> If the owner cpu reschedules, owner will leave the rq and
>> owner_running() will return false, also breaking the loop.
>>
>> >
>> > Signed-off-by: Hillf Danton <dhillf@gmail.com>
>> > ---
>> >  kernel/sched.c |    2 +-
>> >  1 files changed, 1 insertions(+), 1 deletions(-)
>> >
>> > diff --git a/kernel/sched.c b/kernel/sched.c
>> > index fd18f39..3ea64fe 100644
>> > --- a/kernel/sched.c
>> > +++ b/kernel/sched.c
>> > @@ -4326,7 +4326,7 @@ int mutex_spin_on_owner(struct mutex *lock,
>> > struct task_struct *owner)
>> >             return 0;
>> >
>> >     while (owner_running(lock, owner)) {
>> > -           if (need_resched())
>> > +           if (test_tsk_need_resched(owner))
>> >                     return 0;
>> >
>
> Furthermore, that can crash the machine, as there's no guarantee owner
> is a sane pointer at this point.
>

Well, why is owner_running looped safely?

thanks
           Hillf