From: Felix Kuehling <felix.kuehling@amd.com>
To: "Christian König" <ckoenig.leichtzumerken@gmail.com>,
daniel.vetter@ffwll.ch, dri-devel@lists.freedesktop.org
Cc: "Christian König" <christian.koenig@amd.com>
Subject: Re: [PATCH 01/23] dma-buf: add dma_resv_replace_fences v2
Date: Wed, 23 Mar 2022 11:55:04 -0400 [thread overview]
Message-ID: <c25fd1c5-aef3-9abf-ab94-e901904bc4ea@amd.com> (raw)
In-Reply-To: <20220321135856.1331-1-christian.koenig@amd.com>
Am 2022-03-21 um 09:58 schrieb Christian König:
> This function allows to replace fences from the shared fence list when
> we can gurantee that the operation represented by the original fence has
> finished or no accesses to the resources protected by the dma_resv
> object any more when the new fence finishes.
>
> Then use this function in the amdkfd code when BOs are unmapped from the
> process.
>
> v2: add an example when this is usefull.
>
> Signed-off-by: Christian König <christian.koenig@amd.com>
> ---
> drivers/dma-buf/dma-resv.c | 45 +++++++++++++++++
> .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 49 +++----------------
> include/linux/dma-resv.h | 2 +
> 3 files changed, 54 insertions(+), 42 deletions(-)
>
> diff --git a/drivers/dma-buf/dma-resv.c b/drivers/dma-buf/dma-resv.c
> index b51416405e86..509060861cf3 100644
> --- a/drivers/dma-buf/dma-resv.c
> +++ b/drivers/dma-buf/dma-resv.c
> @@ -289,6 +289,51 @@ void dma_resv_add_shared_fence(struct dma_resv *obj, struct dma_fence *fence)
> }
> EXPORT_SYMBOL(dma_resv_add_shared_fence);
>
> +/**
> + * dma_resv_replace_fences - replace fences in the dma_resv obj
> + * @obj: the reservation object
> + * @context: the context of the fences to replace
> + * @replacement: the new fence to use instead
> + *
> + * Replace fences with a specified context with a new fence. Only valid if the
> + * operation represented by the original fence has no longer access to the
> + * resources represented by the dma_resv object when the new fence completes.
> + *
> + * And example for using this is replacing a preemption fence with a page table
> + * update fence which makes the resource inaccessible.
> + */
> +void dma_resv_replace_fences(struct dma_resv *obj, uint64_t context,
> + struct dma_fence *replacement)
> +{
> + struct dma_resv_list *list;
> + struct dma_fence *old;
> + unsigned int i;
> +
> + dma_resv_assert_held(obj);
> +
> + write_seqcount_begin(&obj->seq);
> +
> + old = dma_resv_excl_fence(obj);
> + if (old->context == context) {
> + RCU_INIT_POINTER(obj->fence_excl, dma_fence_get(replacement));
> + dma_fence_put(old);
> + }
> +
> + list = dma_resv_shared_list(obj);
> + for (i = 0; list && i < list->shared_count; ++i) {
> + old = rcu_dereference_protected(list->shared[i],
> + dma_resv_held(obj));
> + if (old->context != context)
> + continue;
> +
> + rcu_assign_pointer(list->shared[i], dma_fence_get(replacement));
> + dma_fence_put(old);
> + }
> +
> + write_seqcount_end(&obj->seq);
> +}
> +EXPORT_SYMBOL(dma_resv_replace_fences);
> +
> /**
> * dma_resv_add_excl_fence - Add an exclusive fence.
> * @obj: the reservation object
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
> index f9bab963a948..b6f266f612ea 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
> @@ -253,53 +253,18 @@ void amdgpu_amdkfd_release_notify(struct amdgpu_bo *bo)
> static int amdgpu_amdkfd_remove_eviction_fence(struct amdgpu_bo *bo,
> struct amdgpu_amdkfd_fence *ef)
> {
> - struct dma_resv *resv = bo->tbo.base.resv;
> - struct dma_resv_list *old, *new;
> - unsigned int i, j, k;
> + struct dma_fence *replacement;
>
> if (!ef)
> return -EINVAL;
>
> - old = dma_resv_shared_list(resv);
> - if (!old)
> - return 0;
> -
> - new = kmalloc(struct_size(new, shared, old->shared_max), GFP_KERNEL);
> - if (!new)
> - return -ENOMEM;
> -
> - /* Go through all the shared fences in the resevation object and sort
> - * the interesting ones to the end of the list.
> + /* TODO: Instead of block before we should use the fence of the page
> + * table update and TLB flush here directly.
I don't understand this comment. Which block are you referring to? And
what does this have to do with TLB flushing? Maybe you're thinking of a
specific use case of amdgpu_amdkfd_remove_eviction_fence for page table
BOs. But that's not the only case where we need to remove eviction fences.
Other than that, this patch is
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
> */
> - for (i = 0, j = old->shared_count, k = 0; i < old->shared_count; ++i) {
> - struct dma_fence *f;
> -
> - f = rcu_dereference_protected(old->shared[i],
> - dma_resv_held(resv));
> -
> - if (f->context == ef->base.context)
> - RCU_INIT_POINTER(new->shared[--j], f);
> - else
> - RCU_INIT_POINTER(new->shared[k++], f);
> - }
> - new->shared_max = old->shared_max;
> - new->shared_count = k;
> -
> - /* Install the new fence list, seqcount provides the barriers */
> - write_seqcount_begin(&resv->seq);
> - RCU_INIT_POINTER(resv->fence, new);
> - write_seqcount_end(&resv->seq);
> -
> - /* Drop the references to the removed fences or move them to ef_list */
> - for (i = j; i < old->shared_count; ++i) {
> - struct dma_fence *f;
> -
> - f = rcu_dereference_protected(new->shared[i],
> - dma_resv_held(resv));
> - dma_fence_put(f);
> - }
> - kfree_rcu(old, rcu);
> -
> + replacement = dma_fence_get_stub();
> + dma_resv_replace_fences(bo->tbo.base.resv, ef->base.context,
> + replacement);
> + dma_fence_put(replacement);
> return 0;
> }
>
> diff --git a/include/linux/dma-resv.h b/include/linux/dma-resv.h
> index afdfdfac729f..3f53177bdb46 100644
> --- a/include/linux/dma-resv.h
> +++ b/include/linux/dma-resv.h
> @@ -468,6 +468,8 @@ void dma_resv_init(struct dma_resv *obj);
> void dma_resv_fini(struct dma_resv *obj);
> int dma_resv_reserve_shared(struct dma_resv *obj, unsigned int num_fences);
> void dma_resv_add_shared_fence(struct dma_resv *obj, struct dma_fence *fence);
> +void dma_resv_replace_fences(struct dma_resv *obj, uint64_t context,
> + struct dma_fence *fence);
> void dma_resv_add_excl_fence(struct dma_resv *obj, struct dma_fence *fence);
> int dma_resv_get_fences(struct dma_resv *obj, bool write,
> unsigned int *num_fences, struct dma_fence ***fences);
next prev parent reply other threads:[~2022-03-23 15:55 UTC|newest]
Thread overview: 62+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-03-21 13:58 [PATCH 01/23] dma-buf: add dma_resv_replace_fences v2 Christian König
2022-03-21 13:58 ` [PATCH 02/23] dma-buf: finally make the dma_resv_list private v2 Christian König
2022-03-21 13:58 ` [PATCH 03/23] dma-buf: add dma_resv_get_singleton v2 Christian König
2022-04-01 8:21 ` Christian König
2022-04-01 15:57 ` Daniel Vetter
2022-03-21 13:58 ` [PATCH 04/23] RDMA: use dma_resv_wait() instead of extracting the fence Christian König
2022-03-23 13:22 ` Daniel Vetter
2022-03-23 16:32 ` Jason Gunthorpe
2022-03-23 17:34 ` Daniel Vetter
2022-03-21 13:58 ` [PATCH 05/23] drm/etnaviv: stop using dma_resv_excl_fence Christian König
2022-03-23 15:58 ` Lucas Stach
2022-03-21 13:58 ` [PATCH 06/23] drm/nouveau: " Christian König
2022-04-03 15:59 ` Christian König
2022-04-04 11:47 ` Karol Herbst
2022-04-05 7:25 ` Christian König
2022-03-21 13:58 ` [PATCH 07/23] drm/vmwgfx: " Christian König
2022-03-21 14:02 ` Zack Rusin
2022-03-21 14:12 ` Christian König
2022-03-21 15:11 ` Zack Rusin
2022-03-22 7:13 ` Christian König
2022-03-22 12:40 ` Zack Rusin
2022-03-21 13:58 ` [PATCH 08/23] drm/radeon: " Christian König
2022-03-21 13:58 ` [PATCH 09/23] drm/amdgpu: use dma_resv_for_each_fence for CS workaround Christian König
2022-03-21 13:58 ` [PATCH 10/23] dma-buf: finally make dma_resv_excl_fence private v2 Christian König
2022-03-23 13:36 ` Daniel Vetter
2022-04-05 7:58 ` Christian König
2022-03-21 13:58 ` [PATCH 11/23] dma-buf: drop the DAG approach for the dma_resv object v2 Christian König
2022-03-23 13:40 ` Daniel Vetter
2022-03-21 13:58 ` [PATCH 12/23] dma-buf/drivers: make reserving a shared slot mandatory v3 Christian König
2022-03-28 17:14 ` Daniel Vetter
2022-03-31 12:07 ` Christian König
2022-03-31 16:42 ` Daniel Vetter
2022-03-21 13:58 ` [PATCH 13/23] drm/atomic-helper: support more than one write fence in drm_gem_plane_helper_prepare_fb Christian König
2022-03-21 13:58 ` [PATCH 14/23] drm/nouveau: support more than one write fence in fenv50_wndw_prepare_fb Christian König
2022-03-21 13:58 ` [PATCH 15/23] drm/amdgpu: use dma_resv_get_singleton in amdgpu_pasid_free_cb Christian König
2022-03-21 13:58 ` [PATCH 16/23] dma-buf: add enum dma_resv_usage v3 Christian König
2022-03-29 15:24 ` Daniel Vetter
2022-04-04 1:13 ` Bas Nieuwenhuizen
2022-03-21 13:58 ` [PATCH 17/23] dma-buf: specify usage while adding fences to dma_resv obj v5 Christian König
2022-03-29 15:43 ` Daniel Vetter
2022-04-01 15:01 ` Christian König
2022-04-01 16:16 ` Daniel Vetter
2022-04-02 22:16 ` Bas Nieuwenhuizen
2022-04-03 17:48 ` Christian König
2022-03-21 13:58 ` [PATCH 18/23] drm/amdgpu: remove dma_resv workaround Christian König
2022-03-29 15:47 ` Daniel Vetter
2022-03-21 13:58 ` [PATCH 19/23] dma-buf: add DMA_RESV_USAGE_KERNEL v2 Christian König
2022-03-29 15:53 ` Daniel Vetter
2022-03-21 13:58 ` [PATCH 20/23] dma-buf: add DMA_RESV_USAGE_BOOKKEEP v2 Christian König
2022-03-29 16:06 ` Daniel Vetter
2022-03-21 13:58 ` [PATCH 21/23] dma-buf: wait for map to complete for static attachments Christian König
2022-03-29 15:56 ` Daniel Vetter
2022-03-21 13:58 ` [PATCH 22/23] drm/i915: drop bo->moving dependency Christian König
2022-03-29 15:57 ` Daniel Vetter
2022-03-21 13:58 ` [PATCH 23/23] drm/ttm: remove bo->moving Christian König
2022-03-29 16:02 ` Daniel Vetter
2022-04-05 9:25 ` Christian König
2022-04-05 12:51 ` Daniel Vetter
2022-03-21 14:03 ` [PATCH 01/23] dma-buf: add dma_resv_replace_fences v2 Christian König
2022-03-23 13:09 ` Daniel Vetter
2022-03-23 15:55 ` Felix Kuehling [this message]
2022-03-23 15:57 ` 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=c25fd1c5-aef3-9abf-ab94-e901904bc4ea@amd.com \
--to=felix.kuehling@amd.com \
--cc=christian.koenig@amd.com \
--cc=ckoenig.leichtzumerken@gmail.com \
--cc=daniel.vetter@ffwll.ch \
--cc=dri-devel@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 a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox