> 2025年2月7日 16:34，Lazar, Lijo <lijo.lazar@xxxxxxx> 写道： > > > > On 2/7/2025 2:00 PM, Gerry Liu wrote: >> >> >>> 2025年2月7日 16:04，Lazar, Lijo <lijo.lazar@xxxxxxx> 写道： >>> >>> >>> >>> On 2/7/2025 12:14 PM, Jiang Liu wrote: >>>> As pwfw resets entrycount when device is suspended, so we should >>>> accmulate the gfx_off_entrycount value instead of save the last value >>>> of it. >>>> >>>> Signed-off-by: Jiang Liu <gerry@xxxxxxxxxxxxxxxxx> >>>> --- >>>> drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c | 6 +++--- >>>> 1 file changed, 3 insertions(+), 3 deletions(-) >>>> >>>> diff --git a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c >>>> index ed9dac00ebfb..70a5ab649e5f 100644 >>>> --- a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c >>>> +++ b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c >>>> @@ -2134,12 +2134,12 @@ static int smu_suspend(struct amdgpu_ip_block *ip_block) >>>> smu_set_gfx_cgpg(smu, false); >>>> >>>> /* >>>> - * pwfw resets entrycount when device is suspended, so we save the >>>> - * last value to be used when we resume to keep it consistent >>>> + * pwfw resets entrycount when device is suspended, so we accumulate >>>> + * the `gfx_off_entrycount` value. >>>> */ >>>> ret = smu_get_entrycount_gfxoff(smu, &count); >>>> if (!ret) >>>> - adev->gfx.gfx_off_entrycount = count; >>>> + adev->gfx.gfx_off_entrycount += count; >>> >>> This is slightly misleading - only Vangogh implements >>> get_gfx_off_entrycount and its implementation, >>> vangogh_get_gfxoff_entrycount, is already doing something like this - >>> >>> *entrycount = value + adev->gfx.gfx_off_entrycount; >> >> Because we have not restored the hardware counter by writing back the value on resume, adev->gfx.gfx_off_entrycount only remembers the latest value of latest suspend/resume cycle. And history information is lost, so changed it to accumulate all values. > > What I meant is that accumulation is already done in the implementation > side - check implementation of vangogh_get_gfxoff_entrycount. Aha, you are right, smu_suspend() calls smu_get_entrycount_gfxoff(), it already accumulates along the way. Will drop it. > > Thanks, > Lijo > >> >>> >>> Thanks, >>> Lijo >>> >>>> >>>> /* clear this on suspend so it will get reprogrammed on resume */ >>>> smu->workload_mask = 0;
