[PATCH] drm/amdgpu/vce1: Fix checking whether VCE1 is idle

[PATCH] drm/amdgpu/vce1: Fix checking whether VCE1 is idle

[Timur Kristóf]

It turns out that the VCE_STATUS register bits that we
used to check whether VCE1 is idle are actually not useful:
VCE_STATUS.JOB_BUSY doesn't exist until VCE3
VCE_STATUS.UENC_BUSY seems to be stuck high on Cape Verde

Instead, do the following:
Check that the read and write pointer matches for each ring.
Check VCE_STATUS.VCPU_REPORT bits to check busy status.
Finally, also check SRBM_STATUS2.VCE_BUSY just in case,
although this one seems bogus on VCE1.

Fixes: d4a640d4b9f3 ("drm/amdgpu/vce1: Implement VCE1 IP block (v2)")
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
---
 drivers/gpu/drm/amd/amdgpu/vce_v1_0.c         | 25 ++++++++++++++-----
 .../include/asic_reg/vce/vce_1_0_sh_mask.h    |  2 --
 2 files changed, 19 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/vce_v1_0.c b/drivers/gpu/drm/amd/amdgpu/vce_v1_0.c
index b1428f5bbc81..38e21d1a3ebb 100644
--- a/drivers/gpu/drm/amd/amdgpu/vce_v1_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/vce_v1_0.c
@@ -47,6 +47,9 @@
 #define VCE_V1_0_STACK_SIZE	(64 * 1024)
 #define VCE_V1_0_DATA_SIZE	(ALIGN(7808 * (AMDGPU_MAX_VCE_HANDLES + 1), VCE_V1_0_ALIGNMENT))
 #define VCE_STATUS_VCPU_REPORT_FW_LOADED_MASK	0x02
+#define VCE_STATUS_VCPU_REPORT_AUTO_BUSY_MASK   0x08
+#define VCE_STATUS_VCPU_REPORT_RB0_BUSY_MASK    0x10
+#define VCE_STATUS_VCPU_REPORT_RB1_BUSY_MASK    0x20
 
 static void vce_v1_0_set_ring_funcs(struct amdgpu_device *adev);
 static void vce_v1_0_set_irq_funcs(struct amdgpu_device *adev);
@@ -349,19 +352,29 @@ static int vce_v1_0_mc_resume(struct amdgpu_device *adev)
  *
  * @ip_block: amdgpu_ip_block pointer
  *
- * Check whether VCE is busy according to VCE_STATUS.
+ * Check whether VCE is busy according to VCE_STATUS
+ * and whether the ring pointers match.
+ *
  * Also check whether the SRBM thinks VCE is busy, although
- * SRBM_STATUS.VCE_BUSY seems to be bogus because it
+ * SRBM_STATUS2.VCE_BUSY seems to be bogus because it
  * appears to mirror the VCE_STATUS.VCPU_REPORT_FW_LOADED bit.
  */
 static bool vce_v1_0_is_idle(struct amdgpu_ip_block *ip_block)
 {
 	struct amdgpu_device *adev = ip_block->adev;
-	bool busy =
-		(RREG32(mmVCE_STATUS) & (VCE_STATUS__JOB_BUSY_MASK | VCE_STATUS__UENC_BUSY_MASK)) ||
-		(RREG32(mmSRBM_STATUS2) & SRBM_STATUS2__VCE_BUSY_MASK);
 
-	return !busy;
+	/* Check whether the rings are busy. */
+	if (RREG32(mmVCE_RB_RPTR) != RREG32(mmVCE_RB_WPTR) ||
+	    RREG32(mmVCE_RB_RPTR2) != RREG32(mmVCE_RB_WPTR2))
+		return false;
+
+	/* Check whether the VCPU is busy. */
+	if (RREG32(mmVCE_STATUS) & (VCE_STATUS_VCPU_REPORT_AUTO_BUSY_MASK |
+				    VCE_STATUS_VCPU_REPORT_RB0_BUSY_MASK |
+				    VCE_STATUS_VCPU_REPORT_RB1_BUSY_MASK))
+		return false;
+
+	return !(RREG32(mmSRBM_STATUS2) & SRBM_STATUS2__VCE_BUSY_MASK);
 }
 
 static int vce_v1_0_wait_for_idle(struct amdgpu_ip_block *ip_block)
diff --git a/drivers/gpu/drm/amd/include/asic_reg/vce/vce_1_0_sh_mask.h b/drivers/gpu/drm/amd/include/asic_reg/vce/vce_1_0_sh_mask.h
index 1f82d6f5abde..13c15c2138b3 100644
--- a/drivers/gpu/drm/amd/include/asic_reg/vce/vce_1_0_sh_mask.h
+++ b/drivers/gpu/drm/amd/include/asic_reg/vce/vce_1_0_sh_mask.h
@@ -63,8 +63,6 @@
 #define VCE_SOFT_RESET__ECPU_SOFT_RESET__SHIFT 0x00000000
 #define VCE_SOFT_RESET__FME_SOFT_RESET_MASK 0x00000004L
 #define VCE_SOFT_RESET__FME_SOFT_RESET__SHIFT 0x00000002
-#define VCE_STATUS__JOB_BUSY_MASK 0x00000001L
-#define VCE_STATUS__JOB_BUSY__SHIFT 0x00000000
 #define VCE_STATUS__UENC_BUSY_MASK 0x00000100L
 #define VCE_STATUS__UENC_BUSY__SHIFT 0x00000008
 #define VCE_STATUS__VCPU_REPORT_MASK 0x000000feL
-- 
2.55.0

Re: [PATCH] drm/amdgpu/vce1: Fix checking whether VCE1 is idle

[David Rosca]

On 10/6/26 19:03, Timur Kristóf wrote:
> It turns out that the VCE_STATUS register bits that we
> used to check whether VCE1 is idle are actually not useful:
> VCE_STATUS.JOB_BUSY doesn't exist until VCE3
> VCE_STATUS.UENC_BUSY seems to be stuck high on Cape Verde
>
> Instead, do the following:
> Check that the read and write pointer matches for each ring.
> Check VCE_STATUS.VCPU_REPORT bits to check busy status.
> Finally, also check SRBM_STATUS2.VCE_BUSY just in case,
> although this one seems bogus on VCE1.
>
> Fixes: d4a640d4b9f3 ("drm/amdgpu/vce1: Implement VCE1 IP block (v2)")
> Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>

There's still occasional warning that VCE is not idle on my cape verde, 
but it's much better than the current state where it prints it every time.

Acked-by: David Rosca <david.rosca@amd.com>

> ---
>   drivers/gpu/drm/amd/amdgpu/vce_v1_0.c         | 25 ++++++++++++++-----
>   .../include/asic_reg/vce/vce_1_0_sh_mask.h    |  2 --
>   2 files changed, 19 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/vce_v1_0.c b/drivers/gpu/drm/amd/amdgpu/vce_v1_0.c
> index b1428f5bbc81..38e21d1a3ebb 100644
> --- a/drivers/gpu/drm/amd/amdgpu/vce_v1_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/vce_v1_0.c
> @@ -47,6 +47,9 @@
>   #define VCE_V1_0_STACK_SIZE	(64 * 1024)
>   #define VCE_V1_0_DATA_SIZE	(ALIGN(7808 * (AMDGPU_MAX_VCE_HANDLES + 1), VCE_V1_0_ALIGNMENT))
>   #define VCE_STATUS_VCPU_REPORT_FW_LOADED_MASK	0x02
> +#define VCE_STATUS_VCPU_REPORT_AUTO_BUSY_MASK   0x08
> +#define VCE_STATUS_VCPU_REPORT_RB0_BUSY_MASK    0x10
> +#define VCE_STATUS_VCPU_REPORT_RB1_BUSY_MASK    0x20
>   
>   static void vce_v1_0_set_ring_funcs(struct amdgpu_device *adev);
>   static void vce_v1_0_set_irq_funcs(struct amdgpu_device *adev);
> @@ -349,19 +352,29 @@ static int vce_v1_0_mc_resume(struct amdgpu_device *adev)
>    *
>    * @ip_block: amdgpu_ip_block pointer
>    *
> - * Check whether VCE is busy according to VCE_STATUS.
> + * Check whether VCE is busy according to VCE_STATUS
> + * and whether the ring pointers match.
> + *
>    * Also check whether the SRBM thinks VCE is busy, although
> - * SRBM_STATUS.VCE_BUSY seems to be bogus because it
> + * SRBM_STATUS2.VCE_BUSY seems to be bogus because it
>    * appears to mirror the VCE_STATUS.VCPU_REPORT_FW_LOADED bit.
>    */
>   static bool vce_v1_0_is_idle(struct amdgpu_ip_block *ip_block)
>   {
>   	struct amdgpu_device *adev = ip_block->adev;
> -	bool busy =
> -		(RREG32(mmVCE_STATUS) & (VCE_STATUS__JOB_BUSY_MASK | VCE_STATUS__UENC_BUSY_MASK)) ||
> -		(RREG32(mmSRBM_STATUS2) & SRBM_STATUS2__VCE_BUSY_MASK);
>   
> -	return !busy;
> +	/* Check whether the rings are busy. */
> +	if (RREG32(mmVCE_RB_RPTR) != RREG32(mmVCE_RB_WPTR) ||
> +	    RREG32(mmVCE_RB_RPTR2) != RREG32(mmVCE_RB_WPTR2))
> +		return false;
> +
> +	/* Check whether the VCPU is busy. */
> +	if (RREG32(mmVCE_STATUS) & (VCE_STATUS_VCPU_REPORT_AUTO_BUSY_MASK |
> +				    VCE_STATUS_VCPU_REPORT_RB0_BUSY_MASK |
> +				    VCE_STATUS_VCPU_REPORT_RB1_BUSY_MASK))
> +		return false;
> +
> +	return !(RREG32(mmSRBM_STATUS2) & SRBM_STATUS2__VCE_BUSY_MASK);
>   }
>   
>   static int vce_v1_0_wait_for_idle(struct amdgpu_ip_block *ip_block)
> diff --git a/drivers/gpu/drm/amd/include/asic_reg/vce/vce_1_0_sh_mask.h b/drivers/gpu/drm/amd/include/asic_reg/vce/vce_1_0_sh_mask.h
> index 1f82d6f5abde..13c15c2138b3 100644
> --- a/drivers/gpu/drm/amd/include/asic_reg/vce/vce_1_0_sh_mask.h
> +++ b/drivers/gpu/drm/amd/include/asic_reg/vce/vce_1_0_sh_mask.h
> @@ -63,8 +63,6 @@
>   #define VCE_SOFT_RESET__ECPU_SOFT_RESET__SHIFT 0x00000000
>   #define VCE_SOFT_RESET__FME_SOFT_RESET_MASK 0x00000004L
>   #define VCE_SOFT_RESET__FME_SOFT_RESET__SHIFT 0x00000002
> -#define VCE_STATUS__JOB_BUSY_MASK 0x00000001L
> -#define VCE_STATUS__JOB_BUSY__SHIFT 0x00000000
>   #define VCE_STATUS__UENC_BUSY_MASK 0x00000100L
>   #define VCE_STATUS__UENC_BUSY__SHIFT 0x00000008
>   #define VCE_STATUS__VCPU_REPORT_MASK 0x000000feL

Re: [PATCH] drm/amdgpu/vce1: Fix checking whether VCE1 is idle

[Timur Kristóf]

On Sunday, October 25, 2026 7:22:04 PM Central European Standard Time David 
Rosca wrote:
> On 10/6/26 19:03, Timur Kristóf wrote:
> > It turns out that the VCE_STATUS register bits that we
> > used to check whether VCE1 is idle are actually not useful:
> > VCE_STATUS.JOB_BUSY doesn't exist until VCE3
> > VCE_STATUS.UENC_BUSY seems to be stuck high on Cape Verde
> > 
> > Instead, do the following:
> > Check that the read and write pointer matches for each ring.
> > Check VCE_STATUS.VCPU_REPORT bits to check busy status.
> > Finally, also check SRBM_STATUS2.VCE_BUSY just in case,
> > although this one seems bogus on VCE1.
> > 
> > Fixes: d4a640d4b9f3 ("drm/amdgpu/vce1: Implement VCE1 IP block (v2)")
> > Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
> 
> There's still occasional warning that VCE is not idle on my cape verde,
> but it's much better than the current state where it prints it every time.
> 

Hi David,

Can you say what the remaining warning says?
Do you think we could fix the remaining warning by removing the check for the 
SRBM_STATUS bit entirely?

Thanks,
Timur


> 
> > ---
> > 
> >   drivers/gpu/drm/amd/amdgpu/vce_v1_0.c         | 25 ++++++++++++++-----
> >   .../include/asic_reg/vce/vce_1_0_sh_mask.h    |  2 --
> >   2 files changed, 19 insertions(+), 8 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/amd/amdgpu/vce_v1_0.c
> > b/drivers/gpu/drm/amd/amdgpu/vce_v1_0.c index b1428f5bbc81..38e21d1a3ebb
> > 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/vce_v1_0.c
> > +++ b/drivers/gpu/drm/amd/amdgpu/vce_v1_0.c
> > @@ -47,6 +47,9 @@
> > 
> >   #define VCE_V1_0_STACK_SIZE	(64 * 1024)
> >   #define VCE_V1_0_DATA_SIZE	(ALIGN(7808 * (AMDGPU_MAX_VCE_HANDLES + 
1),
> >   VCE_V1_0_ALIGNMENT)) #define VCE_STATUS_VCPU_REPORT_FW_LOADED_MASK	
0x02
> > 
> > +#define VCE_STATUS_VCPU_REPORT_AUTO_BUSY_MASK   0x08
> > +#define VCE_STATUS_VCPU_REPORT_RB0_BUSY_MASK    0x10
> > +#define VCE_STATUS_VCPU_REPORT_RB1_BUSY_MASK    0x20
> > 
> >   static void vce_v1_0_set_ring_funcs(struct amdgpu_device *adev);
> >   static void vce_v1_0_set_irq_funcs(struct amdgpu_device *adev);
> > 
> > @@ -349,19 +352,29 @@ static int vce_v1_0_mc_resume(struct amdgpu_device
> > *adev)> 
> >    *
> >    * @ip_block: amdgpu_ip_block pointer
> >    *
> > 
> > - * Check whether VCE is busy according to VCE_STATUS.
> > + * Check whether VCE is busy according to VCE_STATUS
> > + * and whether the ring pointers match.
> > + *
> > 
> >    * Also check whether the SRBM thinks VCE is busy, although
> > 
> > - * SRBM_STATUS.VCE_BUSY seems to be bogus because it
> > + * SRBM_STATUS2.VCE_BUSY seems to be bogus because it
> > 
> >    * appears to mirror the VCE_STATUS.VCPU_REPORT_FW_LOADED bit.
> >    */
> >   
> >   static bool vce_v1_0_is_idle(struct amdgpu_ip_block *ip_block)
> >   {
> >   
> >   	struct amdgpu_device *adev = ip_block->adev;
> > 
> > -	bool busy =
> > -		(RREG32(mmVCE_STATUS) & (VCE_STATUS__JOB_BUSY_MASK |
> > VCE_STATUS__UENC_BUSY_MASK)) || -		
(RREG32(mmSRBM_STATUS2) &
> > SRBM_STATUS2__VCE_BUSY_MASK);
> > 
> > -	return !busy;
> > +	/* Check whether the rings are busy. */
> > +	if (RREG32(mmVCE_RB_RPTR) != RREG32(mmVCE_RB_WPTR) ||
> > +	    RREG32(mmVCE_RB_RPTR2) != RREG32(mmVCE_RB_WPTR2))
> > +		return false;
> > +
> > +	/* Check whether the VCPU is busy. */
> > +	if (RREG32(mmVCE_STATUS) & (VCE_STATUS_VCPU_REPORT_AUTO_BUSY_MASK 
|
> > +				    
VCE_STATUS_VCPU_REPORT_RB0_BUSY_MASK |
> > +				    
VCE_STATUS_VCPU_REPORT_RB1_BUSY_MASK))
> > +		return false;
> > +
> > +	return !(RREG32(mmSRBM_STATUS2) & SRBM_STATUS2__VCE_BUSY_MASK);
> > 
> >   }
> >   
> >   static int vce_v1_0_wait_for_idle(struct amdgpu_ip_block *ip_block)
> > 
> > diff --git a/drivers/gpu/drm/amd/include/asic_reg/vce/vce_1_0_sh_mask.h
> > b/drivers/gpu/drm/amd/include/asic_reg/vce/vce_1_0_sh_mask.h index
> > 1f82d6f5abde..13c15c2138b3 100644
> > --- a/drivers/gpu/drm/amd/include/asic_reg/vce/vce_1_0_sh_mask.h
> > +++ b/drivers/gpu/drm/amd/include/asic_reg/vce/vce_1_0_sh_mask.h
> > @@ -63,8 +63,6 @@
> > 
> >   #define VCE_SOFT_RESET__ECPU_SOFT_RESET__SHIFT 0x00000000
> >   #define VCE_SOFT_RESET__FME_SOFT_RESET_MASK 0x00000004L
> >   #define VCE_SOFT_RESET__FME_SOFT_RESET__SHIFT 0x00000002
> > 
> > -#define VCE_STATUS__JOB_BUSY_MASK 0x00000001L
> > -#define VCE_STATUS__JOB_BUSY__SHIFT 0x00000000
> > 
> >   #define VCE_STATUS__UENC_BUSY_MASK 0x00000100L
> >   #define VCE_STATUS__UENC_BUSY__SHIFT 0x00000008
> >   #define VCE_STATUS__VCPU_REPORT_MASK 0x000000feL

Re: [PATCH] drm/amdgpu/vce1: Fix checking whether VCE1 is idle

[David Rosca]

On 10/26/26 14:00, Timur Kristóf wrote:
> On Sunday, October 25, 2026 7:22:04 PM Central European Standard Time David
> Rosca wrote:
>> On 10/6/26 19:03, Timur Kristóf wrote:
>>> It turns out that the VCE_STATUS register bits that we
>>> used to check whether VCE1 is idle are actually not useful:
>>> VCE_STATUS.JOB_BUSY doesn't exist until VCE3
>>> VCE_STATUS.UENC_BUSY seems to be stuck high on Cape Verde
>>>
>>> Instead, do the following:
>>> Check that the read and write pointer matches for each ring.
>>> Check VCE_STATUS.VCPU_REPORT bits to check busy status.
>>> Finally, also check SRBM_STATUS2.VCE_BUSY just in case,
>>> although this one seems bogus on VCE1.
>>>
>>> Fixes: d4a640d4b9f3 ("drm/amdgpu/vce1: Implement VCE1 IP block (v2)")
>>> Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
>> There's still occasional warning that VCE is not idle on my cape verde,
>> but it's much better than the current state where it prints it every time.
>>
> Hi David,
>
> Can you say what the remaining warning says?
> Do you think we could fix the remaining warning by removing the check for the
> SRBM_STATUS bit entirely?
VCE busy: VCE_STATUS=0x142, SRBM_STATUS2=0x80
VCE busy: VCE_STATUS=0x152, SRBM_STATUS2=0x80

which does look like VCE is indeed busy.

David


>
> Thanks,
> Timur
>
>
>>> ---
>>>
>>>    drivers/gpu/drm/amd/amdgpu/vce_v1_0.c         | 25 ++++++++++++++-----
>>>    .../include/asic_reg/vce/vce_1_0_sh_mask.h    |  2 --
>>>    2 files changed, 19 insertions(+), 8 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/vce_v1_0.c
>>> b/drivers/gpu/drm/amd/amdgpu/vce_v1_0.c index b1428f5bbc81..38e21d1a3ebb
>>> 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/vce_v1_0.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/vce_v1_0.c
>>> @@ -47,6 +47,9 @@
>>>
>>>    #define VCE_V1_0_STACK_SIZE	(64 * 1024)
>>>    #define VCE_V1_0_DATA_SIZE	(ALIGN(7808 * (AMDGPU_MAX_VCE_HANDLES +
> 1),
>>>    VCE_V1_0_ALIGNMENT)) #define VCE_STATUS_VCPU_REPORT_FW_LOADED_MASK	
> 0x02
>>> +#define VCE_STATUS_VCPU_REPORT_AUTO_BUSY_MASK   0x08
>>> +#define VCE_STATUS_VCPU_REPORT_RB0_BUSY_MASK    0x10
>>> +#define VCE_STATUS_VCPU_REPORT_RB1_BUSY_MASK    0x20
>>>
>>>    static void vce_v1_0_set_ring_funcs(struct amdgpu_device *adev);
>>>    static void vce_v1_0_set_irq_funcs(struct amdgpu_device *adev);
>>>
>>> @@ -349,19 +352,29 @@ static int vce_v1_0_mc_resume(struct amdgpu_device
>>> *adev)>
>>>     *
>>>     * @ip_block: amdgpu_ip_block pointer
>>>     *
>>>
>>> - * Check whether VCE is busy according to VCE_STATUS.
>>> + * Check whether VCE is busy according to VCE_STATUS
>>> + * and whether the ring pointers match.
>>> + *
>>>
>>>     * Also check whether the SRBM thinks VCE is busy, although
>>>
>>> - * SRBM_STATUS.VCE_BUSY seems to be bogus because it
>>> + * SRBM_STATUS2.VCE_BUSY seems to be bogus because it
>>>
>>>     * appears to mirror the VCE_STATUS.VCPU_REPORT_FW_LOADED bit.
>>>     */
>>>    
>>>    static bool vce_v1_0_is_idle(struct amdgpu_ip_block *ip_block)
>>>    {
>>>    
>>>    	struct amdgpu_device *adev = ip_block->adev;
>>>
>>> -	bool busy =
>>> -		(RREG32(mmVCE_STATUS) & (VCE_STATUS__JOB_BUSY_MASK |
>>> VCE_STATUS__UENC_BUSY_MASK)) || -		
> (RREG32(mmSRBM_STATUS2) &
>>> SRBM_STATUS2__VCE_BUSY_MASK);
>>>
>>> -	return !busy;
>>> +	/* Check whether the rings are busy. */
>>> +	if (RREG32(mmVCE_RB_RPTR) != RREG32(mmVCE_RB_WPTR) ||
>>> +	    RREG32(mmVCE_RB_RPTR2) != RREG32(mmVCE_RB_WPTR2))
>>> +		return false;
>>> +
>>> +	/* Check whether the VCPU is busy. */
>>> +	if (RREG32(mmVCE_STATUS) & (VCE_STATUS_VCPU_REPORT_AUTO_BUSY_MASK
> |
>>> +				
> VCE_STATUS_VCPU_REPORT_RB0_BUSY_MASK |
>>> +				
> VCE_STATUS_VCPU_REPORT_RB1_BUSY_MASK))
>>> +		return false;
>>> +
>>> +	return !(RREG32(mmSRBM_STATUS2) & SRBM_STATUS2__VCE_BUSY_MASK);
>>>
>>>    }
>>>    
>>>    static int vce_v1_0_wait_for_idle(struct amdgpu_ip_block *ip_block)
>>>
>>> diff --git a/drivers/gpu/drm/amd/include/asic_reg/vce/vce_1_0_sh_mask.h
>>> b/drivers/gpu/drm/amd/include/asic_reg/vce/vce_1_0_sh_mask.h index
>>> 1f82d6f5abde..13c15c2138b3 100644
>>> --- a/drivers/gpu/drm/amd/include/asic_reg/vce/vce_1_0_sh_mask.h
>>> +++ b/drivers/gpu/drm/amd/include/asic_reg/vce/vce_1_0_sh_mask.h
>>> @@ -63,8 +63,6 @@
>>>
>>>    #define VCE_SOFT_RESET__ECPU_SOFT_RESET__SHIFT 0x00000000
>>>    #define VCE_SOFT_RESET__FME_SOFT_RESET_MASK 0x00000004L
>>>    #define VCE_SOFT_RESET__FME_SOFT_RESET__SHIFT 0x00000002
>>>
>>> -#define VCE_STATUS__JOB_BUSY_MASK 0x00000001L
>>> -#define VCE_STATUS__JOB_BUSY__SHIFT 0x00000000
>>>
>>>    #define VCE_STATUS__UENC_BUSY_MASK 0x00000100L
>>>    #define VCE_STATUS__UENC_BUSY__SHIFT 0x00000008
>>>    #define VCE_STATUS__VCPU_REPORT_MASK 0x000000feL
>
>
>

Re: [PATCH] drm/amdgpu/vce1: Fix checking whether VCE1 is idle

[Timur Kristóf]

On Mon, 2026-10-26 at 14:25 +0100, David Rosca wrote:
> 
> On 10/26/26 14:00, Timur Kristóf wrote:
> > On Sunday, October 25, 2026 7:22:04 PM Central European Standard
> > Time David
> > Rosca wrote:
> > > On 10/6/26 19:03, Timur Kristóf wrote:
> > > > It turns out that the VCE_STATUS register bits that we
> > > > used to check whether VCE1 is idle are actually not useful:
> > > > VCE_STATUS.JOB_BUSY doesn't exist until VCE3
> > > > VCE_STATUS.UENC_BUSY seems to be stuck high on Cape Verde
> > > > 
> > > > Instead, do the following:
> > > > Check that the read and write pointer matches for each ring.
> > > > Check VCE_STATUS.VCPU_REPORT bits to check busy status.
> > > > Finally, also check SRBM_STATUS2.VCE_BUSY just in case,
> > > > although this one seems bogus on VCE1.
> > > > 
> > > > Fixes: d4a640d4b9f3 ("drm/amdgpu/vce1: Implement VCE1 IP block
> > > > (v2)")
> > > > Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
> > > There's still occasional warning that VCE is not idle on my cape
> > > verde,
> > > but it's much better than the current state where it prints it
> > > every time.
> > > 
> > Hi David,
> > 
> > Can you say what the remaining warning says?
> > Do you think we could fix the remaining warning by removing the
> > check for the
> > SRBM_STATUS bit entirely?
> VCE busy: VCE_STATUS=0x142, SRBM_STATUS2=0x80
> VCE busy: VCE_STATUS=0x152, SRBM_STATUS2=0x80
> 
> which does look like VCE is indeed busy.
> 
> David

It looks like 0x152 has the RB0_BUSY bit set, but 0x142 seems like it
doesn't have any of the busy bits set and only the SRBM thinks that the
VCE is busy but it may not be actually busy. It's hard to judge without
more information.

When do you see the warning? Does it happen after an encoding job or
after suspend/resume? Or...?

Thanks,
Timur