Re: [RFC 3/6] drm/amdgpu: Fix crash on modprobe

["Christian König" <ckoenig.leichtzumerken@gmail.com>]

Am 21.12.21 um 17:03 schrieb Andrey Grodzovsky:
>
> On 2021-12-21 2:02 a.m., Christian König wrote:
>>
>>
>> Am 20.12.21 um 20:22 schrieb Andrey Grodzovsky:
>>>
>>> On 2021-12-20 2:17 a.m., Christian König wrote:
>>>> Am 17.12.21 um 23:27 schrieb Andrey Grodzovsky:
>>>>> Restrict jobs resubmission to suspend case
>>>>> only since schedulers not initialised yet on
>>>>> probe.
>>>>>
>>>>> Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
>>>>> ---
>>>>>   drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 2 +-
>>>>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c 
>>>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
>>>>> index 5527c68c51de..8ebd954e06c6 100644
>>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
>>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
>>>>> @@ -582,7 +582,7 @@ void amdgpu_fence_driver_hw_init(struct 
>>>>> amdgpu_device *adev)
>>>>>           if (!ring || !ring->fence_drv.initialized)
>>>>>               continue;
>>>>>   -        if (!ring->no_scheduler) {
>>>>> +        if (adev->in_suspend && !ring->no_scheduler) {
>>>>
>>>> Uff, why is that suddenly necessary? Because of the changed order?
>>>>
>>>> Christian.
>>>
>>>
>>> Yes.
>>
>> Mhm, that's quite bad design then.
>
>
> If you look at the original patch for this 
> https://www.spinics.net/lists/amd-gfx/msg67560.html you will
> see that that restarting scheduler here is only relevant for 
> suspend/resume case because there was
> a race to fix. There is no point in this code on driver init because 
> nothing was submitted to scheduler yet
> and so it seems to me ok to add condition that this code run only 
> in_suspend case.

Yeah, but having extra logic like this means that we have some design 
issue in the IP block handling.

We need to clean that and some other odd approaches up at some point, 
but probably not now.

Christian.

>
>
>>
>> How about we keep the order as is and allow specifying the reset work 
>> queue with drm_sched_start() ?
>
>
> As i mentioned above, the fact we even have drm_sched_start there is 
> just part of a solution to resolve a race
> during suspend/resume. It is not for device initialization and indeed, 
> other client drivers of gpu shcheduler never call
> drm_sched_start on device init. We must guarantee that reset work 
> queue already initialized before any job submission to scheduler
> and because of that IMHO the right place for this is drm_sched_init.
>
> Andrey
>
>
>>
>> Christian.
>>
>>>
>>> Andrey
>>>
>>>
>>>>
>>>>> drm_sched_resubmit_jobs(&ring->sched);
>>>>>               drm_sched_start(&ring->sched, true);
>>>>>           }
>>>>
>>