drm/prime: keep a reference from the handle to exported dma-buf (v6)
Subsystems (7)
Dave Airlie
maintainerPATCH 1/2 drm/prime: keep a reference from the handle to exported dma-buf (v6)
Currently we have a problem with this:
1. i915: create gem object
2. i915: export gem object to prime
3. radeon: import gem object
4. close prime fd
5. radeon: unref object
6. i915: unref object
i915 has an imported object reference in its file priv, that isn't cleaned up properly until fd close. The reference gets added at step 2, but at step 6 we don't have enough info to clean it up.
The solution is to take a reference on the dma-buf when we export it, and drop the reference when the gem handle goes away.
So when we export a dma_buf from a gem object, we keep track of it with the handle, we take a reference to the dma_buf. When we close the handle (i.e. userspace is finished with the buffer), we drop the reference to the dma_buf, and it gets collected.
This patch isn't meant to fix any other problem or bikesheds, and it doesn't fix any races with other scenarios.
v1.1: move export symbol line back up.
v2: okay I had to do a bit more, as the first patch showed a leak on one of my tests, that I found using the dma-buf debugfs support, the problem case is exporting a buffer twice with the same handle, we'd add another export handle for it unnecessarily, however
we now fail if we try to export the same object with a different gem handle, however I'm not sure if that is a case I want to support, and I've gotten the code to WARN_ON if we hit something like that.
v2.1: rebase this patch, write better commit msg.
v3: cleanup error handling, track import vs export in linked list, these two patches were separate previously, but seem to work better like this.
v4: danvet is correct, this code is no longer useful, since the buffer better exist, so remove it.
v5: always take a reference to the dma buf object, import or export. (Imre Deak contributed this originally)
v6: square the circle, remove import vs export tracking now that there is no difference
--- drivers/gpu/drm/drm_gem.c | 4 +-- drivers/gpu/drm/drm_prime.c | 76 ++++++++++++++++++++++++--------------------- include/drm/drmP.h | 5 ++- 3 files changed, 44 insertions(+), 41 deletions(-)
drivers/gpu/drm/drm_gem.c +2 -2
@@ -205,11 +205,11 @@ static void drm_gem_remove_prime_handles(struct drm_gem_object *obj, struct drm_file *filp) {if (obj->import_attach) { -drm_prime_remove_imported_buf_handle (&filp->prime, +drm_prime_remove_buf_handle (&filp->prime, obj->import_attach->dmabuf); }if (obj->export_dma_buf) { -drm_prime_remove_imported_buf_handle (&filp->prime, +drm_prime_remove_buf_handle (&filp->prime, obj->export_dma_buf); } }
drivers/gpu/drm/drm_prime.c +40 -36
150 lines not on this page · raw message
include/drm/drmP.h +2 -3
@@ -1593,9 +1593,8 @@ extern void drm_prime_gem_destroy(struct drm_gem_object *obj, struct sg_table *svoid drm_prime_init_file_private (struct drm_prime_file_private *prime_fpriv);void drm_prime_destroy_file_private (struct drm_prime_file_private *prime_fpriv); -int drm_prime_add_imported_buf_handle (struct drm_prime_file_private *prime_fpriv,struct dma_buf *dma_buf,uint32_t handle); -int drm_prime_lookup_imported_buf_handle (struct drm_prime_file_private *prime_fpriv,struct dma_buf *dma_buf,uint32_t *handle); -void drm_prime_remove_imported_buf_handle (struct drm_prime_file_private *prime_fpriv,struct dma_buf *dma_buf); +int drm_prime_lookup_buf_handle (struct drm_prime_file_private *prime_fpriv,struct dma_buf *dma_buf,uint32_t *handle); +void drm_prime_remove_buf_handle (struct drm_prime_file_private *prime_fpriv,struct dma_buf *dma_buf);int drm_prime_add_dma_buf (struct drm_device *dev,struct drm_gem_object *obj); int drm_prime_lookup_obj(struct drm_device *dev, struct dma_buf *buf,
Signature · 1 line
1.8.2
Dave Airlie
maintainerPATCH 2/2 drm: prime: fix refcounting on the dmabuf import error path
From: Imre Deak <[email protected]>
In commit be8a42ae60 we inroduced a refcount problem, where on the drm_gem_prime_fd_to_handle() error path we'll call dma_buf_put() for self imported dma buffers.
Fix this by taking a reference on the dma buffer in the .gem_import hook instead of assuming the caller had taken one. Besides fixing the bug this is also more logical.
--- drivers/gpu/drm/drm_prime.c | 8 +++++++- drivers/gpu/drm/exynos/exynos_drm_dmabuf.c | 4 +++- drivers/gpu/drm/i915/i915_gem_dmabuf.c | 5 ++++- drivers/gpu/drm/omapdrm/omap_gem_dmabuf.c | 1 - drivers/gpu/drm/udl/udl_gem.c | 4 ++++ 5 files changed, 18 insertions(+), 4 deletions(-)
drivers/gpu/drm/drm_prime.c +7 -1
@@ -271,7 +271,6 @@ struct drm_gem_object *drm_gem_prime_import(struct drm_device *dev,* refcount on gem itself instead of f_count of dmabuf. */ drm_gem_object_reference (obj); -dma_buf_put (dma_buf);return obj; } }
@@ -280,6 +279,8 @@ struct drm_gem_object *drm_gem_prime_import(struct drm_device *dev,if (IS_ERR (attach))return ERR_PTR (PTR_ERR (attach)); +get_dma_buf (dma_buf); + sgt =dma_buf_map_attachment (attach,DMA_BIDIRECTIONAL );if (IS_ERR_OR_NULL (sgt)) { ret =PTR_ERR (sgt);
@@ -300,6 +301,8 @@ fail_unmap:dma_buf_unmap_attachment (attach, sgt,DMA_BIDIRECTIONAL ); fail_detach:dma_buf_detach (dma_buf, attach); +dma_buf_put (dma_buf); +return ERR_PTR (ret); }EXPORT_SYMBOL (drm_gem_prime_import);
@@ -342,6 +345,9 @@ int drm_gem_prime_fd_to_handle(struct drm_device *dev, goto fail;mutex_unlock (&file_priv->prime.lock); + +dma_buf_put (dma_buf); +return 0 ; fail:
drivers/gpu/drm/exynos/exynos_drm_dmabuf.c +3 -1
@@ -235,7 +235,6 @@ struct drm_gem_object *exynos_dmabuf_prime_import(struct drm_device *drm_dev,* refcount on gem itself instead of f_count of dmabuf. */ drm_gem_object_reference (obj); -dma_buf_put (dma_buf);return obj; } }
@@ -244,6 +243,7 @@ struct drm_gem_object *exynos_dmabuf_prime_import(struct drm_device *drm_dev,if (IS_ERR (attach))return ERR_PTR (-EINVAL ); +get_dma_buf (dma_buf); sgt =dma_buf_map_attachment (attach,DMA_BIDIRECTIONAL );if (IS_ERR_OR_NULL (sgt)) {
@@ -298,6 +298,8 @@ err_unmap_attach:dma_buf_unmap_attachment (attach, sgt,DMA_BIDIRECTIONAL ); err_buf_detach:dma_buf_detach (dma_buf, attach); +dma_buf_put (dma_buf); +return ERR_PTR (ret); }
drivers/gpu/drm/i915/i915_gem_dmabuf.c +4 -1
@@ -271,7 +271,6 @@ struct drm_gem_object *i915_gem_prime_import(struct drm_device *dev,* refcount on gem itself instead of f_count of dmabuf. */ drm_gem_object_reference (&obj->base); -dma_buf_put (dma_buf);return &obj->base; } }
@@ -281,6 +280,8 @@ struct drm_gem_object *i915_gem_prime_import(struct drm_device *dev,if (IS_ERR (attach))return ERR_CAST (attach); +get_dma_buf (dma_buf); + obj =i915_gem_object_alloc (dev);if (obj ==NULL ) { ret = -ENOMEM ;
@@ -300,5 +301,7 @@ struct drm_gem_object *i915_gem_prime_import(struct drm_device *dev, fail_detach:dma_buf_detach (dma_buf, attach); +dma_buf_put (dma_buf); +return ERR_PTR (ret); }
drivers/gpu/drm/omapdrm/omap_gem_dmabuf.c +0 -1
@@ -212,7 +212,6 @@ struct drm_gem_object *omap_gem_prime_import(struct drm_device *dev,* refcount on gem itself instead of f_count of dmabuf. */ drm_gem_object_reference (obj); -dma_buf_put (buffer);return obj; } }
drivers/gpu/drm/udl/udl_gem.c +4 -0
@@ -303,6 +303,8 @@ struct drm_gem_object *udl_gem_prime_import(struct drm_device *dev,if (IS_ERR (attach))return ERR_CAST (attach); +get_dma_buf (dma_buf); + sg =dma_buf_map_attachment (attach,DMA_BIDIRECTIONAL );if (IS_ERR (sg)) { ret =PTR_ERR (sg);
@@ -322,5 +324,7 @@ fail_unmap:dma_buf_unmap_attachment (attach, sg,DMA_BIDIRECTIONAL ); fail_detach:dma_buf_detach (dma_buf, attach); +dma_buf_put (dma_buf); +return ERR_PTR (ret); }
Signature · 1 line
1.8.2
Quoted 44 lines · Dave Airlie
Currently we have a problem with this:
1. i915: create gem object
2. i915: export gem object to prime
3. radeon: import gem object
4. close prime fd
5. radeon: unref object
6. i915: unref objecti915 has an imported object reference in its file priv, that isn't cleaned up properly until fd close. The reference gets added at step 2, but at step 6 we don't have enough info to clean it up.
The solution is to take a reference on the dma-buf when we export it, and drop the reference when the gem handle goes away.
So when we export a
dma_buffrom a gem object, we keep track of it with the handle, we take a reference to thedma_buf. When we close the handle (i.e. userspace is finished with the buffer), we drop the reference to thedma_buf, and it gets collected.This patch isn't meant to fix any other problem or bikesheds, and it doesn't fix any races with other scenarios.
v1.1: move export symbol line back up.
v2: okay I had to do a bit more, as the first patch showed a leak on one of my tests, that I found using the dma-buf debugfs support, the problem case is exporting a buffer twice with the same handle, we'd add another export handle for it unnecessarily, however
we now fail if we try to export the same object with a different gem handle, however I'm not sure if that is a case I want to support, and I've gotten the code toWARN_ONif we hit something like that.v2.1: rebase this patch, write better commit msg.
v3: cleanup error handling, track import vs export in linked list, these two patches were separate previously, but seem to work better like this.
v4: danvet is correct, this code is no longer useful, since the buffer better exist, so remove it.
v5: always take a reference to the dma buf object, import or export. (Imre Deak contributed this originally)
v6: square the circle, remove import vs export tracking now that there is no difference
Quoted 1 line
Signed-off-by Dave Airlie <[email protected]>
Reviewing the right patch is apparently recommneded ;-)
Signature · 3 lines
Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch