From: "Christian König" <christian.koenig@amd.com>
To: Prike Liang <Prike.Liang@amd.com>, amd-gfx@lists.freedesktop.org
Cc: Alexander.Deucher@amd.com
Subject: Re: [PATCH 4/4] drm/amdgpu: free the evf when the attached bo release
Date: Wed, 16 Apr 2025 13:07:16 +0200 [thread overview]
Message-ID: <366de5ed-2234-45e6-9c2a-d21e83899b7b@amd.com> (raw)
In-Reply-To: <20250416085029.2278563-4-Prike.Liang@amd.com>
Am 16.04.25 um 10:50 schrieb Prike Liang:
> Free the evf when the attached bo released. The evf still
> be dependent on and referred to by the attached bo that is
> scheduled by the kernel queue SDMA or gfx after the evf signalled.
>
> Signed-off-by: Prike Liang <Prike.Liang@amd.com>
> ---
> .../drm/amd/amdgpu/amdgpu_eviction_fence.c | 31 ++++++++++++++++---
> .../drm/amd/amdgpu/amdgpu_eviction_fence.h | 1 +
> drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 1 +
> 3 files changed, 28 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c
> index b34225bbd85d..60be1ac5047d 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c
> @@ -27,6 +27,7 @@
>
> #define work_to_evf_mgr(w, name) container_of(w, struct amdgpu_eviction_fence_mgr, name)
> #define evf_mgr_to_fpriv(e) container_of(e, struct amdgpu_fpriv, evf_mgr)
> +#define fence_to_evf(f) container_of(f, struct amdgpu_eviction_fence, base)
>
> static const char *
> amdgpu_eviction_fence_get_driver_name(struct dma_fence *fence)
> @@ -47,7 +48,7 @@ int
> amdgpu_eviction_fence_replace_fence(struct amdgpu_eviction_fence_mgr *evf_mgr,
> struct drm_exec *exec)
> {
> - struct amdgpu_eviction_fence *old_ef, *new_ef;
> + struct amdgpu_eviction_fence *new_ef;
> struct drm_gem_object *obj;
> unsigned long index;
> int ret;
> @@ -72,7 +73,6 @@ amdgpu_eviction_fence_replace_fence(struct amdgpu_eviction_fence_mgr *evf_mgr,
>
> /* Update the eviction fence now */
> spin_lock(&evf_mgr->ev_fence_lock);
> - old_ef = evf_mgr->ev_fence;
> evf_mgr->ev_fence = new_ef;
> spin_unlock(&evf_mgr->ev_fence_lock);
>
> @@ -102,9 +102,6 @@ amdgpu_eviction_fence_replace_fence(struct amdgpu_eviction_fence_mgr *evf_mgr,
> }
> }
>
> - /* Free old fence */
> - if (old_ef)
> - dma_fence_put(&old_ef->base);
That change looks completely incorrect to me, you will now leak the old fence.
> return 0;
>
> free_err:
> @@ -237,6 +234,30 @@ void amdgpu_eviction_fence_detach(struct amdgpu_eviction_fence_mgr *evf_mgr,
> dma_fence_put(stub);
> }
>
> +void amdgpu_userq_remove_all_eviction_fences(struct amdgpu_bo *bo)
Please name that amdgpu_eviction_fence_remove_all().
Regards,
Christian.
> +{
> + struct dma_resv *resv = &bo->tbo.base._resv;
> + struct dma_fence *fence, *stub;
> + struct dma_resv_iter cursor;
> +
> + dma_resv_assert_held(resv);
> +
> + stub = dma_fence_get_stub();
> + dma_resv_for_each_fence(&cursor, resv, DMA_RESV_USAGE_BOOKKEEP, fence) {
> + struct amdgpu_eviction_fence *ev_fence;
> +
> + ev_fence = fence_to_evf(fence);
> + if (!ev_fence || !dma_fence_is_signaled(&ev_fence->base))
> + continue;
> +
> + dma_resv_replace_fences(resv, fence->context, stub,
> + DMA_RESV_USAGE_BOOKKEEP);
> +
> + }
> +
> + dma_fence_put(stub);
> +}
> +
> int amdgpu_eviction_fence_init(struct amdgpu_eviction_fence_mgr *evf_mgr)
> {
> /* This needs to be done one time per open */
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.h
> index fcd867b7147d..da99ac322a2e 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.h
> @@ -66,4 +66,5 @@ amdgpu_eviction_fence_signal(struct amdgpu_eviction_fence_mgr *evf_mgr,
> int
> amdgpu_eviction_fence_replace_fence(struct amdgpu_eviction_fence_mgr *evf_mgr,
> struct drm_exec *exec);
> +void amdgpu_userq_remove_all_eviction_fences(struct amdgpu_bo *bo);
> #endif
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> index 1e73ce30d4d7..f001018a01eb 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> @@ -1392,6 +1392,7 @@ void amdgpu_bo_release_notify(struct ttm_buffer_object *bo)
> amdgpu_vram_mgr_set_cleared(bo->resource);
> dma_resv_add_fence(&bo->base._resv, fence, DMA_RESV_USAGE_KERNEL);
> dma_fence_put(fence);
> + amdgpu_userq_remove_all_eviction_fences(abo);
>
> out:
> dma_resv_unlock(&bo->base._resv);
next prev parent reply other threads:[~2025-04-16 11:07 UTC|newest]
Thread overview: 21+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-04-16 8:50 [PATCH 1/4] drm/amdgpu: add the evf attached gem obj resv dump Prike Liang
2025-04-16 8:50 ` [PATCH 2/4] drm/amdgpu: set the evf name to identify the userq case Prike Liang
2025-04-16 11:02 ` Christian König
2025-04-16 12:56 ` Liang, Prike
2025-04-16 8:50 ` [PATCH 3/4] drm/amdgpu: trace the scheduler dependent job fence name Prike Liang
2025-04-16 11:04 ` Christian König
2025-04-16 14:16 ` Liang, Prike
2025-04-17 7:35 ` Christian König
2025-04-16 8:50 ` [PATCH 4/4] drm/amdgpu: free the evf when the attached bo release Prike Liang
2025-04-16 11:07 ` Christian König [this message]
2025-04-16 14:47 ` Liang, Prike
2025-04-17 7:39 ` Christian König
2025-04-22 9:14 ` Liang, Prike
2025-04-22 9:27 ` Christian König
2025-04-22 12:39 ` Yadav, Arvind
2025-04-22 13:09 ` Liang, Prike
2025-04-22 13:26 ` Christian König
2025-04-23 3:25 ` Liang, Prike
2025-04-16 11:01 ` [PATCH 1/4] drm/amdgpu: add the evf attached gem obj resv dump Christian König
2025-04-16 12:54 ` Liang, Prike
2025-04-16 12:58 ` Christian König
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=366de5ed-2234-45e6-9c2a-d21e83899b7b@amd.com \
--to=christian.koenig@amd.com \
--cc=Alexander.Deucher@amd.com \
--cc=Prike.Liang@amd.com \
--cc=amd-gfx@lists.freedesktop.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.