On Wed, Feb 2, 2022 at 10:29 AM Limonciello, Mario <mario.limonciello@xxxxxxx> wrote: > > On 2/2/2022 08:16, Alex Deucher wrote: > > On Wed, Feb 2, 2022 at 4:07 AM Rajib Mahapatra <rajib.mahapatra@xxxxxxx> wrote: > >> > >> [Why] > >> amdgpu error observed if suspend is aborted during S0i3 > >> resume. > >> > >> [How] > >> If suspend is aborted for some reason during S0i3 resume > >> cycle, it follows amdgpu errors in resume. > >> Skipping SDMA ip in suspend solves the issue on RENOIR > >> (green sardine apu) chip. This time, the system is > >> able to resume gracefully even the suspend is aborted. > >> > >> Signed-off-by: Rajib Mahapatra <rajib.mahapatra@xxxxxxx> > >> --- > >> drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 10 ++++++++++ > >> 1 file changed, 10 insertions(+) > >> > >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c > >> index 7931132ce6e3..f01b1dffff7f 100644 > >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c > >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c > >> @@ -2927,6 +2927,16 @@ static int amdgpu_device_ip_suspend_phase2(struct amdgpu_device *adev) > >> adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GFX)) > >> continue; > >> > >> + /* skip suspend of sdma for S0ix > >> + * Resume has issues if the suspend is aborted during S0i3 cycle. > >> + * Skipping sdma for RN/CZN/BRC chip - green sardine apu. > >> + */ > >> + if (adev->in_s0ix && > >> + (adev->asic_type == CHIP_RENOIR && > >> + (adev->pdev->device == 0x15e7 || adev->pdev->device == 0x1638) && > > > > The check here seems to contradict the comment above. Is this all > > Renoir based APUs or just green sardine? If it's just green sardine, > > you can check the APU flags rather than the PCI ids. E.g., > > (adev->apu_flags & AMD_APU_IS_GREEN_SARDINE) > > Also move this to sdma 4 code as Mario suggested. > > Both RN and green sardine share the same flows for SMU, I would think it > should just be match (adev->in_s0xi && (adev->flags & AMD_IS_APU)) when > it's moved to skip suspend. The SDMA 4.0 code is shared with Raven1/2 and Picasso as well. We should verify that it's required for them as well. Alex > > > > > Alex > > > >> + adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SDMA)) > >> + continue; > >> + > >> /* XXX handle errors */ > >> r = adev->ip_blocks[i].version->funcs->suspend(adev); > >> /* XXX handle errors */ > >> -- > >> 2.25.1 > >> >
