[PATCH] drm/amdgpu: avoid over-handle of fence driver fini in s3 test

[Christian König]

Hi Guchun,

see below.

Am 29.07.21 um 14:39 schrieb Chen, Guchun:
> [Public]
>
> Hi Christian,
>
> Thanks for your feedback.
>
> Originally, drm_sched_fini is part of amdgpu_fence_driver_hw_fini, I did not move it.

Yeah, not saying that this is your fault, you should just clean that up 
more thoughtfully.

> Former patch " cd87a6dcf6af drm/amdgpu: adjust fence driver enable sequence " has dropped amdgpu_fence_driver_suspend, and called amdgpu_fence_driver_hw_fini instead in function amdgpu_device_suspend. I checked the code difference between  amdgpu_fence_driver_hw_fini and amdgpu_device_suspend, they are almost the same except drm_sched_fini part, so I think we can leave it as it is, while skipping the execution of drm_sched_fini in suspend/resume case.

And exactly that's a bad idea and the reason why I already said during 
the review of patch "cd87a6dcf6af drm/amdgpu: adjust fence driver enable 
sequence" that the callers of those functions need to be adjusted instead.

1. If not already done please rename the functions as Hawking suggested 
as well, they should be amdgpu_fence_driver_hw_(init|fini) and 
amdgpu_fence_driver_sw_(init|fini).

2. Remove drm_sched_fini() from amdgpu_fence_driver_hw_fini() and add 
that to sw_fini instead.

3. Adjust the callers so that we have the same functionality as before. 
E.g. by calling both hw_fini and sw_fini at the same time.

Regards,
Christian.

>
> Regards,
> Guchun
>
> -----Original Message-----
> From: Koenig, Christian <Christian.Koenig at amd.com>
> Sent: Thursday, July 29, 2021 7:11 PM
> To: Chen, Guchun <Guchun.Chen at amd.com>; amd-gfx at lists.freedesktop.org; Gao, Likun <Likun.Gao at amd.com>; Zhang, Hawking <Hawking.Zhang at amd.com>; Deucher, Alexander <Alexander.Deucher at amd.com>
> Subject: Re: [PATCH] drm/amdgpu: avoid over-handle of fence driver fini in s3 test
>
> Am 29.07.21 um 12:49 schrieb Guchun Chen:
>> In amdgpu_fence_driver_hw_fini, no need to call drm_sched_fini to stop
>> scheduler in s3 test, otherwise, fence errors will occur after resume.
>> So introduce a new parameter to distingiush the case in this API.
> NAK, the problem is rather that drm_sched_fini() is part of the sw shutdown and should never be called during hw_fini.
>
> Christian.
>
>> Fixes: cd87a6dcf6af drm/amdgpu: adjust fence driver enable sequence
>> Signed-off-by: Guchun Chen <guchun.chen at amd.com>
>> ---
>>    drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 4 ++--
>>    drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c  | 8 +++++---
>>    drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h   | 2 +-
>>    3 files changed, 8 insertions(+), 6 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> index b1d2dc39e8be..aaff8ebbb7dc 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> @@ -3844,7 +3844,7 @@ void amdgpu_device_fini_hw(struct amdgpu_device *adev)
>>    		else
>>    			drm_atomic_helper_shutdown(adev_to_drm(adev));
>>    	}
>> -	amdgpu_fence_driver_hw_fini(adev);
>> +	amdgpu_fence_driver_hw_fini(adev, false);
>>    
>>    	if (adev->pm_sysfs_en)
>>    		amdgpu_pm_sysfs_fini(adev);
>> @@ -3941,7 +3941,7 @@ int amdgpu_device_suspend(struct drm_device *dev, bool fbcon)
>>    	/* evict vram memory */
>>    	amdgpu_bo_evict_vram(adev);
>>    
>> -	amdgpu_fence_driver_hw_fini(adev);
>> +	amdgpu_fence_driver_hw_fini(adev, adev->in_suspend);
>>    
>>    	amdgpu_device_ip_suspend_phase2(adev);
>>    	/* evict remaining vram memory
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
>> index 49c5c7331c53..7e6a73c2919d 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
>> @@ -515,14 +515,15 @@ int amdgpu_fence_driver_init(struct amdgpu_device *adev)
>>    }
>>    
>>    /**
>> - * amdgpu_fence_driver_fini - tear down the fence driver
>> + * amdgpu_fence_driver_hw_fini - tear down the fence driver
>>     * for all possible rings.
>>     *
>>     * @adev: amdgpu device pointer
>> + * @in_reset: indicator to distingiush device removal case or s3 case
>>     *
>>     * Tear down the fence driver for all possible rings (all asics).
>>     */
>> -void amdgpu_fence_driver_hw_fini(struct amdgpu_device *adev)
>> +void amdgpu_fence_driver_hw_fini(struct amdgpu_device *adev, bool
>> +in_reset)
>>    {
>>    	int i, r;
>>    
>> @@ -531,8 +532,9 @@ void amdgpu_fence_driver_hw_fini(struct
>> amdgpu_device *adev)
>>    
>>    		if (!ring || !ring->fence_drv.initialized)
>>    			continue;
>> -		if (!ring->no_scheduler)
>> +		if (!ring->no_scheduler && !in_reset)
>>    			drm_sched_fini(&ring->sched);
>> +
>>    		/* You can't wait for HW to signal if it's gone */
>>    		if (!drm_dev_is_unplugged(&adev->ddev))
>>    			r = amdgpu_fence_wait_empty(ring); diff --git
>> a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
>> index 27adffa7658d..42cbecfc26a3 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
>> @@ -115,7 +115,7 @@ int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring,
>>    int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring,
>>    				   struct amdgpu_irq_src *irq_src,
>>    				   unsigned irq_type);
>> -void amdgpu_fence_driver_hw_fini(struct amdgpu_device *adev);
>> +void amdgpu_fence_driver_hw_fini(struct amdgpu_device *adev, bool
>> +in_reset);
>>    void amdgpu_fence_driver_sw_fini(struct amdgpu_device *adev);
>>    void amdgpu_fence_driver_hw_init(struct amdgpu_device *adev);
>>    int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence
>> **fence,
> _______________________________________________
> amd-gfx mailing list
> amd-gfx at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx