[PATCH] drm/amdgpu: stop/resume fb access when gpu reset V2

[zhoucm1]

On 2016年06月23日 00:40, Alex Deucher wrote:
> On Wed, Jun 22, 2016 at 2:04 AM, Chunming Zhou <David1.Zhou at amd.com> wrote:
>> V2: add wait_for_mc_idle after stopping fb access
>>
>> Change-Id: I86b5c2b8c78a17ef43ade2a97ef8a33853650be0
>> Signed-off-by: Chunming Zhou <David1.Zhou at amd.com>
>> Reviewed-by: Christian König <christian.koenig at amd.com> (V1)
>> ---
>>   drivers/gpu/drm/amd/amdgpu/cik.c | 12 +++++++++++-
>>   drivers/gpu/drm/amd/amdgpu/vi.c  | 10 +++++++++-
>>   2 files changed, 20 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/cik.c b/drivers/gpu/drm/amd/amdgpu/cik.c
>> index 40f4fda..7f2e3db 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/cik.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/cik.c
>> @@ -1213,10 +1213,20 @@ static void cik_set_bios_scratch_engine_hung(struct amdgpu_device *adev, bool hu
>>   static int cik_asic_reset(struct amdgpu_device *adev)
>>   {
>>          int r;
>> -       cik_set_bios_scratch_engine_hung(adev, true);
>> +       struct amdgpu_mode_mc_save save;
>>
>> +       cik_set_bios_scratch_engine_hung(adev, true);
>> +       /* Disable fb access */
>> +       if (adev->mode_info.num_crtc) {
>> +               amdgpu_display_stop_mc_access(adev, &save);
>> +               amdgpu_asic_wait_for_mc_idle(adev);
> Please use the GMC IP block wait_for_idle callback as I just removed
> amdgpu_asic_wait_for_mc_idle() in another patch set sent out
> yesterday.
Yeah, have you added wait_for_mc_idle to GMC_IP block?
And seems you still didn't push those patches, I cannot update V3 
accordingly at hand.
> Actually why not do this in amdgpu_gpu_reset() before
> calling amdgpu_asic_reset() since it's the same for all asics.
good idea.
>
>> +       }
>>          r = cik_gpu_pci_config_reset(adev);
>>
>> +       /* resume fb access */
>> +       if (adev->mode_info.num_crtc)
>> +               amdgpu_display_resume_mc_access(adev, &save);
>> +
> I think this part should be dropped.  There's no need to unblank the
> displays.  After the reset, the entire display pipeline is
> theoretically reset so this won't really do anything useful.  Same
> with vi.c
accept.

Regards,
David Zhou
>
> Alex
>
>>          cik_set_bios_scratch_engine_hung(adev, false);
>>
>>          return r;
>> diff --git a/drivers/gpu/drm/amd/amdgpu/vi.c b/drivers/gpu/drm/amd/amdgpu/vi.c
>> index 1ac0c91..5337c4f 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/vi.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/vi.c
>> @@ -633,10 +633,18 @@ static void vi_set_bios_scratch_engine_hung(struct amdgpu_device *adev, bool hun
>>   static int vi_asic_reset(struct amdgpu_device *adev)
>>   {
>>          int r;
>> +       struct amdgpu_mode_mc_save save;
>>
>>          vi_set_bios_scratch_engine_hung(adev, true);
>> -
>> +       /* Disable fb access */
>> +       if (adev->mode_info.num_crtc) {
>> +               amdgpu_display_stop_mc_access(adev, &save);
>> +               amdgpu_asic_wait_for_mc_idle(adev);
>> +       }
>>          r = vi_gpu_pci_config_reset(adev);
>> +       /* resume fb access */
>> +       if (adev->mode_info.num_crtc)
>> +               amdgpu_display_resume_mc_access(adev, &save);
>>
>>          vi_set_bios_scratch_engine_hung(adev, false);
>>
>> --
>> 1.9.1
>>
>> _______________________________________________
>> amd-gfx mailing list
>> amd-gfx at lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/amd-gfx