dma-buf: implement vmap refcounting in the interface logic
All drivers which implement this need to have some sort of refcount to
allow concurrent vmap usage. Hence implement this in the dma-buf core.
To protect against concurrent calls we need a lock, which potentially
causes new funny locking inversions. But this shouldn't be a problem
for exporters with statically allocated backing storage, and more
dynamic drivers have decent issues already anyway.
Inspired by some refactoring patches from Aaron Plattner, who
implemented the same idea, but only for drm/prime drivers.
v2: Check in dma_buf_release that no dangling vmaps are left.
Suggested by Aaron Plattner. We might want to do similar checks for
attachments, but that's for another patch. Also fix up ERR_PTR return
for vmap.
v3: Check whether the passed-in vmap address matches with the cached
one for vunmap. Eventually we might want to remove that parameter -
compared to the kmap functions there's no need for the vaddr for
unmapping. Suggested by Chris Wilson.
v4: Fix a brown-paper-bag bug spotted by Aaron Plattner.
Cc: Aaron Plattner <aplattner@nvidia.com>
Reviewed-by: Aaron Plattner <aplattner@nvidia.com>
Tested-by: Aaron Plattner <aplattner@nvidia.com>
Reviewed-by: Rob Clark <rob@ti.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Sumit Semwal <sumit.semwal@linaro.org>
diff --git a/Documentation/dma-buf-sharing.txt b/Documentation/dma-buf-sharing.txt
index 0188903..4966b1b 100644
--- a/Documentation/dma-buf-sharing.txt
+++ b/Documentation/dma-buf-sharing.txt
@@ -302,7 +302,11 @@
void dma_buf_vunmap(struct dma_buf *dmabuf, void *vaddr)
The vmap call can fail if there is no vmap support in the exporter, or if it
- runs out of vmalloc space. Fallback to kmap should be implemented.
+ runs out of vmalloc space. Fallback to kmap should be implemented. Note that
+ the dma-buf layer keeps a reference count for all vmap access and calls down
+ into the exporter's vmap function only when no vmapping exists, and only
+ unmaps it once. Protection against concurrent vmap/vunmap calls is provided
+ by taking the dma_buf->lock mutex.
3. Finish access