[PATCH v2] drm/amd: Check that VPE has reached DPM0 in idle handler

[PATCH v2] drm/amd: Check that VPE has reached DPM0 in idle handler

[Mario Limonciello]

[Why]
Newer VPE microcode has functionality that will decrease DPM level
only when a workload has run for 2 or more seconds.  If VPE is turned
off before this DPM decrease, the SOC can get stuck with a higher
DPM level.

This can happen from amdgpu's ring buffer test because it's a short
quick workload for VPE and VPE is turned off after 1s.

[How]
In idle handler besides checking fences are drained check that VPE DPM
level is really is at DPM0. If not, schedule delayed work again until
it is.

Cc: Peyton.Lee@amd.com
Reported-by: Sultan Alsawaf <sultan@kerneltoast.com>
Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4615
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
---
Drop unnecessary spurious dpm_level variable
 drivers/gpu/drm/amd/amdgpu/amdgpu_vpe.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vpe.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vpe.c
index 474bfe36c0c2..9065d799c2c4 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vpe.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vpe.c
@@ -326,11 +326,16 @@ static void vpe_idle_work_handler(struct work_struct *work)
 {
 	struct amdgpu_device *adev =
 		container_of(work, struct amdgpu_device, vpe.idle_work.work);
+	struct amdgpu_vpe *vpe = &adev->vpe;
 	unsigned int fences = 0;
+	uint32_t dpm_level;
 
 	fences += amdgpu_fence_count_emitted(&adev->vpe.ring);
 
-	if (fences == 0)
+	dpm_level = adev->pm.dpm_enabled ?
+		    RREG32(vpe_get_reg_offset(vpe, 0, vpe->regs.dpm_request_lv)) : 0;
+
+	if (fences == 0 && dpm_level == 0)
 		amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VPE, AMD_PG_STATE_GATE);
 	else
 		schedule_delayed_work(&adev->vpe.idle_work, VPE_IDLE_TIMEOUT);
-- 
2.51.0

Re: [PATCH v2] drm/amd: Check that VPE has reached DPM0 in idle handler

[Sultan Alsawaf]

Hi Mario,

On Fri, Oct 10, 2025 at 07:37:22PM -0500, Mario Limonciello wrote:
> [Why]
> Newer VPE microcode has functionality that will decrease DPM level
> only when a workload has run for 2 or more seconds.  If VPE is turned
> off before this DPM decrease, the SOC can get stuck with a higher
> DPM level.
> 
> This can happen from amdgpu's ring buffer test because it's a short
> quick workload for VPE and VPE is turned off after 1s.
> 
> [How]
> In idle handler besides checking fences are drained check that VPE DPM
> level is really is at DPM0. If not, schedule delayed work again until
> it is.
> 
> Cc: Peyton.Lee@amd.com
> Reported-by: Sultan Alsawaf <sultan@kerneltoast.com>
> Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4615
> Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
> ---
> Drop unnecessary spurious dpm_level variable
>  drivers/gpu/drm/amd/amdgpu/amdgpu_vpe.c | 7 ++++++-
>  1 file changed, 6 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vpe.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vpe.c
> index 474bfe36c0c2..9065d799c2c4 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vpe.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vpe.c
> @@ -326,11 +326,16 @@ static void vpe_idle_work_handler(struct work_struct *work)
>  {
>  	struct amdgpu_device *adev =
>  		container_of(work, struct amdgpu_device, vpe.idle_work.work);
> +	struct amdgpu_vpe *vpe = &adev->vpe;
>  	unsigned int fences = 0;
> +	uint32_t dpm_level;
>  
>  	fences += amdgpu_fence_count_emitted(&adev->vpe.ring);
>  
> -	if (fences == 0)
> +	dpm_level = adev->pm.dpm_enabled ?
> +		    RREG32(vpe_get_reg_offset(vpe, 0, vpe->regs.dpm_request_lv)) : 0;

This register read can be skipped when fences != 0. Perhaps with a goto label?

> +
> +	if (fences == 0 && dpm_level == 0)
>  		amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VPE, AMD_PG_STATE_GATE);
>  	else
>  		schedule_delayed_work(&adev->vpe.idle_work, VPE_IDLE_TIMEOUT);
> -- 
> 2.51.0
> 

Sultan