Re: [PATCH 1/2] sched: Add schedule_(raw_)spin_unlock and schedule_(raw_)spin_unlock_irq

[Kirill Tkhai <tkhai@yandex.ru>]

18.06.2013, 21:28, "Peter Zijlstra" <peterz@infradead.org>:
> On Tue, Jun 18, 2013 at 07:36:52PM +0400, Kirill Tkhai wrote:
>
>>  Helpers for replacement repeating patterns:
>>
>>  1)spin_unlock(lock);
>>    schedule();
>>  2)spin_unlock_irq(lock);
>>    schedule();
>
> I just noticed this; the existing schedule_preempt_disabled() is
> equivalent to:
>
>   preempt_enable()
>   schedule()
>   preempt_disable()
>
> So I somewhat expected these new primitives to be:
>
>   spin_unlock()
>   schedule()
>   spin_lock()
>
> Now I haven't actually looked at the usage patch to see what the
> converted sites look like (thanks for adding that one though!).
>
> My OCD just triggered on the preemption and locked schedule calls having
> different semantics.

They have different semantic and different ending.

Many places (as you can see from the second patch) need additional actions
between schedule() and next spin_lock(). Several places don't do the second
lock.

Kirill