dma-buf: Wait on the reservation object when sync'ing before CPU access
authorChris Wilson <chris@chris-wilson.co.uk>
Mon, 15 Aug 2016 15:42:18 +0000 (16:42 +0100)
committerSumit Semwal <sumit.semwal@linaro.org>
Tue, 16 Aug 2016 07:01:35 +0000 (12:31 +0530)
Rendering operations to the dma-buf are tracked implicitly via the
reservation_object (dmabuf->resv). This is used to allow poll() to
wait upon outstanding rendering (or just query the current status of
rendering). The dma-buf sync ioctl allows userspace to prepare the
dma-buf for CPU access, which should include waiting upon rendering.
(Some drivers may need to do more work to ensure that the dma-buf mmap
is coherent as well as complete.)

v2: Always wait upon the reservation object implicitly. We choose to do
it after the native handler in case it can do so more efficiently.

Testcase: igt/prime_vgem
Testcase: igt/gem_concurrent_blit # *vgem*
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Sumit Semwal <sumit.semwal@linaro.org>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Eric Anholt <eric@anholt.net>
Cc: linux-media@vger.kernel.org
Cc: dri-devel@lists.freedesktop.org
Cc: linaro-mm-sig@lists.linaro.org
Cc: linux-kernel@vger.kernel.org
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Sumit Semwal <sumit.semwal@linaro.org>
Link: http://patchwork.freedesktop.org/patch/msgid/1471275738-31994-1-git-send-email-chris@chris-wilson.co.uk
drivers/dma-buf/dma-buf.c

index ddaee60..cf04d24 100644 (file)
@@ -586,6 +586,22 @@ void dma_buf_unmap_attachment(struct dma_buf_attachment *attach,
 }
 EXPORT_SYMBOL_GPL(dma_buf_unmap_attachment);
 
+static int __dma_buf_begin_cpu_access(struct dma_buf *dmabuf,
+                                     enum dma_data_direction direction)
+{
+       bool write = (direction == DMA_BIDIRECTIONAL ||
+                     direction == DMA_TO_DEVICE);
+       struct reservation_object *resv = dmabuf->resv;
+       long ret;
+
+       /* Wait on any implicit rendering fences */
+       ret = reservation_object_wait_timeout_rcu(resv, write, true,
+                                                 MAX_SCHEDULE_TIMEOUT);
+       if (ret < 0)
+               return ret;
+
+       return 0;
+}
 
 /**
  * dma_buf_begin_cpu_access - Must be called before accessing a dma_buf from the
@@ -608,6 +624,13 @@ int dma_buf_begin_cpu_access(struct dma_buf *dmabuf,
        if (dmabuf->ops->begin_cpu_access)
                ret = dmabuf->ops->begin_cpu_access(dmabuf, direction);
 
+       /* Ensure that all fences are waited upon - but we first allow
+        * the native handler the chance to do so more efficiently if it
+        * chooses. A double invocation here will be reasonably cheap no-op.
+        */
+       if (ret == 0)
+               ret = __dma_buf_begin_cpu_access(dmabuf, direction);
+
        return ret;
 }
 EXPORT_SYMBOL_GPL(dma_buf_begin_cpu_access);