drm/amdgpu: rework dma_resv handling v3

Drop the workaround and instead implement a better solution.

Basically we are now chaining all submissions using a dma_fence_chain
container and adding them as exclusive fence to the dma_resv object.

This way other drivers can still sync to the single exclusive fence
while amdgpu only sync to fences from different processes.

v3: add the shared fence first before the exclusive one

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20210614174536.5188-2-christian.koenig@amd.com
This commit is contained in:
Christian König
2021-06-09 13:51:36 +02:00
parent 22f0463ae6
commit 8c505bdc9c
6 changed files with 55 additions and 79 deletions

View File

@@ -42,48 +42,6 @@
#include <linux/pci-p2pdma.h>
#include <linux/pm_runtime.h>
static int
__dma_resv_make_exclusive(struct dma_resv *obj)
{
struct dma_fence **fences;
unsigned int count;
int r;
if (!dma_resv_shared_list(obj)) /* no shared fences to convert */
return 0;
r = dma_resv_get_fences(obj, NULL, &count, &fences);
if (r)
return r;
if (count == 0) {
/* Now that was unexpected. */
} else if (count == 1) {
dma_resv_add_excl_fence(obj, fences[0]);
dma_fence_put(fences[0]);
kfree(fences);
} else {
struct dma_fence_array *array;
array = dma_fence_array_create(count, fences,
dma_fence_context_alloc(1), 0,
false);
if (!array)
goto err_fences_put;
dma_resv_add_excl_fence(obj, &array->base);
dma_fence_put(&array->base);
}
return 0;
err_fences_put:
while (count--)
dma_fence_put(fences[count]);
kfree(fences);
return -ENOMEM;
}
/**
* amdgpu_dma_buf_attach - &dma_buf_ops.attach implementation
*
@@ -110,24 +68,6 @@ static int amdgpu_dma_buf_attach(struct dma_buf *dmabuf,
if (r < 0)
goto out;
r = amdgpu_bo_reserve(bo, false);
if (unlikely(r != 0))
goto out;
/*
* We only create shared fences for internal use, but importers
* of the dmabuf rely on exclusive fences for implicitly
* tracking write hazards. As any of the current fences may
* correspond to a write, we need to convert all existing
* fences on the reservation object into a single exclusive
* fence.
*/
r = __dma_resv_make_exclusive(bo->tbo.base.resv);
if (r)
goto out;
bo->prime_shared_count++;
amdgpu_bo_unreserve(bo);
return 0;
out:
@@ -150,9 +90,6 @@ static void amdgpu_dma_buf_detach(struct dma_buf *dmabuf,
struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);
struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
if (attach->dev->driver != adev->dev->driver && bo->prime_shared_count)
bo->prime_shared_count--;
pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
}
@@ -406,8 +343,6 @@ amdgpu_dma_buf_create_obj(struct drm_device *dev, struct dma_buf *dma_buf)
bo = gem_to_amdgpu_bo(gobj);
bo->allowed_domains = AMDGPU_GEM_DOMAIN_GTT;
bo->preferred_domains = AMDGPU_GEM_DOMAIN_GTT;
if (dma_buf->ops != &amdgpu_dmabuf_ops)
bo->prime_shared_count = 1;
dma_resv_unlock(resv);
return gobj;