[PATCH 2/2] drm/amdgpu: stop disabling irqs when it isn't neccessary

[Daniel Vetter]

On Mon, Jun 13, 2016 at 7:37 PM, Christian König
<deathsimple at vodafone.de> wrote:
> Am 13.06.2016 um 18:33 schrieb Daniel Vetter:
>>
>> On Mon, Jun 13, 2016 at 04:12:43PM +0200, Christian König wrote:
>>>
>>> From: Christian König <christian.koenig at amd.com>
>>>
>>> A regular spin_lock/unlock should do here as well.
>>>
>>> Signed-off-by: Christian König <christian.koenig at amd.com>
>>
>> Just drive-by comment, but you can't mix up irq spinlocks with normal
>> ones. And there's places where this is still irqsave, but some of these
>> functions can be called directly from the scheduler kthread. You can only
>> drop the _irqsave if you know you're always called from hardirq context
>> (softirq isn't enough), or when you know someone already disabled
>> interrupts. And you can simplify the _irqsave to just _irq when you are in
>> always in process context and irqs are always still enabled.
>>
>> On a super-quick look seems fishy.
>
>
> The point is there isn't even any IRQ involved in all of this.
>
> The spin is either locked from a work item or the kthread, never from irq
> context.

In that case git grep job_list_lock finds more of those in sched_fence.c.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch