Re: [PATCH v1] drm/amdgpu/userq: move comment to definition for amdgpu_userq_ensure_ev_fence

2026-04-08 Thread Khatri, Sunil



On 08-04-2026 05:43 pm, Christian König wrote:


On 4/8/26 11:42, Sunil Khatri wrote:

Move the comment for the caller to the definition for
amdgpu_userq_ensure_ev_fence.

Signed-off-by: Sunil Khatri 
---
  drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c | 14 +++---
  1 file changed, 7 insertions(+), 7 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c
index ae973c611972..b921346faa99 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c
@@ -460,6 +460,13 @@ static void amdgpu_userq_cleanup(struct 
amdgpu_usermode_queue *queue)
up_read(&adev->reset_domain->sem);
  }
  
+/*

+ * There could be a situation that we are creating a new queue while
+ * the other queues under this UQ_mgr are suspended. So if there is any
+ * resume work pending, wait for it to get done.

This is actually not necessary here.


+ *
+ * This will also make sure we have a valid eviction fence ready to be used.
+ */

That's the important part, maybe even extend that a bit:

/* This will make sure that we have a valid and not signaled eviction fence and 
user queues are ready to be used. */

Then using kerneldoc format would be even better then just a comment.

Sure, will do that way.

Regards
Sunil khatri


Thanks,
Christian.


  void
  amdgpu_userq_ensure_ev_fence(struct amdgpu_userq_mgr *uq_mgr,
 struct amdgpu_eviction_fence_mgr *evf_mgr)
@@ -786,13 +793,6 @@ amdgpu_userq_create(struct drm_file *filp, union 
drm_amdgpu_userq *args)
goto clean_mapping;
}
  
-	/*

-* There could be a situation that we are creating a new queue while
-* the other queues under this UQ_mgr are suspended. So if there is any
-* resume work pending, wait for it to get done.
-*
-* This will also make sure we have a valid eviction fence ready to be 
used.
-*/
amdgpu_userq_ensure_ev_fence(&fpriv->userq_mgr, &fpriv->evf_mgr);
  
  	r = uq_funcs->mqd_create(queue, &args->in);


Re: [PATCH v1] drm/amdgpu/userq: move comment to definition for amdgpu_userq_ensure_ev_fence

2026-04-08 Thread Christian König



On 4/8/26 11:42, Sunil Khatri wrote:
> Move the comment for the caller to the definition for
> amdgpu_userq_ensure_ev_fence.
> 
> Signed-off-by: Sunil Khatri 
> ---
>  drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c | 14 +++---
>  1 file changed, 7 insertions(+), 7 deletions(-)
> 
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c 
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c
> index ae973c611972..b921346faa99 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c
> @@ -460,6 +460,13 @@ static void amdgpu_userq_cleanup(struct 
> amdgpu_usermode_queue *queue)
>   up_read(&adev->reset_domain->sem);
>  }
>  
> +/*
> + * There could be a situation that we are creating a new queue while
> + * the other queues under this UQ_mgr are suspended. So if there is any
> + * resume work pending, wait for it to get done.

This is actually not necessary here.

> + *
> + * This will also make sure we have a valid eviction fence ready to be used.
> + */

That's the important part, maybe even extend that a bit:

/* This will make sure that we have a valid and not signaled eviction fence and 
user queues are ready to be used. */

Then using kerneldoc format would be even better then just a comment.

Thanks,
Christian.

>  void
>  amdgpu_userq_ensure_ev_fence(struct amdgpu_userq_mgr *uq_mgr,
>struct amdgpu_eviction_fence_mgr *evf_mgr)
> @@ -786,13 +793,6 @@ amdgpu_userq_create(struct drm_file *filp, union 
> drm_amdgpu_userq *args)
>   goto clean_mapping;
>   }
>  
> - /*
> -  * There could be a situation that we are creating a new queue while
> -  * the other queues under this UQ_mgr are suspended. So if there is any
> -  * resume work pending, wait for it to get done.
> -  *
> -  * This will also make sure we have a valid eviction fence ready to be 
> used.
> -  */
>   amdgpu_userq_ensure_ev_fence(&fpriv->userq_mgr, &fpriv->evf_mgr);
>  
>   r = uq_funcs->mqd_create(queue, &args->in);