[PATCH v2] drm/amdgpu: only keep most recent fence for each context

Christian König christian.koenig at amd.com
Wed Apr 23 12:13:03 UTC 2025


On 4/23/25 12:52, Khatri, Sunil wrote:

> [AMD Official Use Only - AMD Internal Distribution Only]
>
> You need to split this in a dma_buf patch and amd userq patch separately.

Yeah agree, and the dma_buf patch should re-use and re-structure the 
existing code and not just add a new function.


Regards,
Christian.
>
> Regards
> Sunil Khatri
>
> -----Original Message-----
> From: Yadav, Arvind <Arvind.Yadav at amd.com>
> Sent: Wednesday, April 23, 2025 4:18 PM
> To: Koenig, Christian <Christian.Koenig at amd.com>; Deucher, Alexander <Alexander.Deucher at amd.com>; Khatri, Sunil <Sunil.Khatri at amd.com>; Paneer Selvam, Arunpravin <Arunpravin.PaneerSelvam at amd.com>
> Cc: amd-gfx at lists.freedesktop.org; Yadav, Arvind <Arvind.Yadav at amd.com>; Koenig, Christian <Christian.Koenig at amd.com>; Yadav, Arvind <Arvind.Yadav at amd.com>
> Subject: [PATCH v2] drm/amdgpu: only keep most recent fence for each context
>
> Mesa passes shared bo, fence syncobj to userq_ioctl.
> There can be duplicates here or some fences that are old.
> This patch is remove duplicates fence and only keep the most recent fence for each context.
>
> v2: Export this code from dma-fence-unwrap.c(by Christian).
>
> Cc: Alex Deucher <alexander.deucher at amd.com>
> Cc: Christian König <christian.koenig at amd.com>
> Cc: Sunil Khatri <sunil.khatri at amd.com>
> Cc: Arunpravin Paneer Selvam <Arunpravin.PaneerSelvam at amd.com>
> Signed-off-by: Arvind Yadav <arvind.yadav at amd.com>
> ---
>   drivers/dma-buf/dma-fence-unwrap.c            | 29 +++++++++++++++++++
>   .../gpu/drm/amd/amdgpu/amdgpu_userq_fence.c   |  2 ++
>   include/linux/dma-fence-unwrap.h              |  3 ++
>   3 files changed, 34 insertions(+)
>
> diff --git a/drivers/dma-buf/dma-fence-unwrap.c b/drivers/dma-buf/dma-fence-unwrap.c
> index 2a059ac0ed27..18874eedf091 100644
> --- a/drivers/dma-buf/dma-fence-unwrap.c
> +++ b/drivers/dma-buf/dma-fence-unwrap.c
> @@ -179,3 +179,32 @@ struct dma_fence *__dma_fence_unwrap_merge(unsigned int num_fences,
>          return tmp;
>   }
>   EXPORT_SYMBOL_GPL(__dma_fence_unwrap_merge);
> +
> +/**
> + * dma_fence_shrink_array - return the shrink fences from fence
> +containers
> + * @fences: array of fence
> + * @num_fences: number of fence in fence array @fences
> + *
> + * only keep most recent fence for each context. It will remove
> +duplicate
> + * and old fence of same context.
> + */
> +unsigned short dma_fence_shrink_array(struct dma_fence **fences,
> +                                     unsigned short num_fences)
> +{
> +       unsigned short count = 0;
> +
> +       if (num_fences <= 1)
> +               return num_fences;
> +
> +       sort(fences, num_fences, sizeof(*fences), fence_cmp, NULL);
> +
> +       for (int i = 1; i < num_fences; i++) {
> +               if (fences[i]->context == fences[count]->context)
> +                       dma_fence_put(fences[i]);
> +               else
> +                       fences[++count] = fences[i];
> +       }
> +
> +       return ++count;
> +}
> +EXPORT_SYMBOL_GPL(dma_fence_shrink_array);
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.c
> index 3288c2ff692e..cc74bb0ed06f 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_userq_fence.c
> @@ -851,6 +851,8 @@ int amdgpu_userq_wait_ioctl(struct drm_device *dev, void *data,
>                          fences[num_fences++] = fence;
>                  }
>
> +               num_fences = dma_fence_shrink_array(fences, num_fences);
> +
>                  waitq = idr_find(&userq_mgr->userq_idr, wait_info->waitq_id);
>                  if (!waitq)
>                          goto free_fences;
> diff --git a/include/linux/dma-fence-unwrap.h b/include/linux/dma-fence-unwrap.h
> index 66b1e56fbb81..94801feeee4b 100644
> --- a/include/linux/dma-fence-unwrap.h
> +++ b/include/linux/dma-fence-unwrap.h
> @@ -72,4 +72,7 @@ struct dma_fence *__dma_fence_unwrap_merge(unsigned int num_fences,
>                  __dma_fence_unwrap_merge(ARRAY_SIZE(__f), __f, __c);    \
>          })
>
> +
> +unsigned short dma_fence_shrink_array(struct dma_fence **fences,
> +                                     unsigned short num_fences);
>   #endif
> --
> 2.34.1
>


More information about the amd-gfx mailing list