Am 2021-02-04 um 9:37 a.m. schrieb Alex Deucher:On Wed, Feb 3, 2021 at 2:56 AM Lazar, Lijo <Lijo.Lazar@xxxxxxx> wrote:[AMD Public Use] -----Original Message----- From: amd-gfx <amd-gfx-bounces@xxxxxxxxxxxxxxxxxxxxx> On Behalf Of Alex Deucher Sent: Tuesday, February 2, 2021 10:48 PM To: amd-gfx@xxxxxxxxxxxxxxxxxxxxx Cc: Deucher, Alexander <Alexander.Deucher@xxxxxxx> Subject: [PATCH 2/2] drm/amdgpu: enable DPM_FLAG_MAY_SKIP_RESUME and DPM_FLAG_SMART_SUSPEND flags Once the device has runtime suspended, we don't need to power it back up again for system suspend. Likewise for resume, we don't to power up the device again on resume only to power it back off again via runtime pm because it's still idle. Signed-off-by: Alex Deucher <alexander.deucher@xxxxxxx> --- drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c index b4780182f990..b78847fa769b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c @@ -206,6 +206,12 @@ int amdgpu_driver_load_kms(struct amdgpu_device *adev, unsigned long flags) if (amdgpu_device_supports_atpx(dev) && !amdgpu_is_atpx_hybrid()) dev_pm_set_driver_flags(dev->dev, DPM_FLAG_NO_DIRECT_COMPLETE); + /* we want direct complete for BOCO */ + if ((amdgpu_device_supports_atpx(dev) && + amdgpu_is_atpx_hybrid()) || + amdgpu_device_supports_boco(dev)) + dev_pm_set_driver_flags(dev->dev, DPM_FLAG_SMART_SUSPEND | + DPM_FLAG_MAY_SKIP_RESUME); Device runtime suspend callback does - amdgpu_device_suspend(drm_dev, false) System suspend callback does - amdgpu_device_suspend(drm_dev, true) One of the effects of this flag is for KFD to decide whether to evict all processes. It is done during system suspend but not during runtime device suspend. Will that have an impact if the system suspend routine is skipped in this way?+ Rajneesh Can you comment on this? Idea of this patch is to not wake the device for system suspend and resume if it's already in runtime suspend.
KFD doesn't allow amdgpu driver to runtime suspend the device as long as the per process device data is valid. This patch only enables direct complete path for already runtime suspended amdgpu device so its implicit that kfd had no active process. While we may be OK to not explicitly cancel scheduled work or suspend all kfd processes (since gpu is already runtime suspended) we still leave KFD unlocked for a system wide suspend via direct complete path. I think we should still be ok with this since locking is only used for making sure that kfd is not open during a gpu reset.
Also i suggest using DPM_FLAG_SMART_PREPARE as well, since we are relying on prepare for skipping late/noirq phase during system suspend.
Acked-by:
Rajneesh Bhardwaj <rajneesh.bhardwaj@xxxxxxx>
KFD only allows runtime suspend when there are no processes using the GPU. Therefore it should be safe (in theory) to skip process eviction if you're already in runtime suspend. Just make sure all the suspend/resume calls into KFD are paired up correctly. If you skip suspend but then later call resume anyway, it will likely cause problems. For testing this, I'd suggest running some KFD application (e.g. kfdtest or an OpenCL app with ROCm-based OpenCL) before suspend, then suspend, then run the app again after resume to make sure KFD is still good. Regards, FelixAlexThanks, Lijo pm_runtime_use_autosuspend(dev->dev); pm_runtime_set_autosuspend_delay(dev->dev, 5000); pm_runtime_allow(dev->dev); -- 2.29.2 _______________________________________________ amd-gfx mailing list amd-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/amd-gfx_______________________________________________ amd-gfx mailing list amd-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/amd-gfx
_______________________________________________ amd-gfx mailing list amd-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/amd-gfx
