Am 01.06.2018 um 16:02 schrieb Michel Dänzer:
> On 2018-06-01 02:11 PM, Christian König wrote:
>> Sorry, accidentally send this series without a cover letter.
>>
>> This is a cleanup to the DMA-buf interface, which is also a prerequisite
>> to unpinned DMA-buf operation.
>>
>> Patch #1 and #2 just remove unused functionality and clean up callback
>> parameters.
>>
>> Patch #3 and #4 introduce taking the reservation lock during
>> mapping/unmapping of DMA-bufs.
>>
>> This introduces a common lock where both exporter as well as importer
>> can then use in the future for unpinned DMA-buf operation.
>>
>> This of course means that exporters should now not take this reservation
>> lock manually any more. The DRM drivers don't seem to actually do that,
>> but I'm not 100% sure about other implementations.
>>
>> Patch #5 then makes use of the new lock to simplify the DMA-buf import
>> handling in amdgpu.
> Please rebase this series on top of
> https://patchwork.freedesktop.org/patch/226311/ and update the
> documentation in amdgpu_prime.c as needed in each patch.
Sure. In this case can we get your patches committed to
amd-staging-drm-next ASAP?
Thanks,
Christian.
So drivers don't need dummy functions just returning NULL.
Cc: Daniel Vetter <daniel(a)ffwll.ch>
Signed-off-by: Gerd Hoffmann <kraxel(a)redhat.com>
---
include/linux/dma-buf.h | 4 ++--
drivers/dma-buf/dma-buf.c | 4 ++++
2 files changed, 6 insertions(+), 2 deletions(-)
diff --git a/include/linux/dma-buf.h b/include/linux/dma-buf.h
index 085db2fee2..88917fa796 100644
--- a/include/linux/dma-buf.h
+++ b/include/linux/dma-buf.h
@@ -39,12 +39,12 @@ struct dma_buf_attachment;
/**
* struct dma_buf_ops - operations possible on struct dma_buf
- * @map_atomic: maps a page from the buffer into kernel address
+ * @map_atomic: [optional] maps a page from the buffer into kernel address
* space, users may not block until the subsequent unmap call.
* This callback must not sleep.
* @unmap_atomic: [optional] unmaps a atomically mapped page from the buffer.
* This Callback must not sleep.
- * @map: maps a page from the buffer into kernel address space.
+ * @map: [optional] maps a page from the buffer into kernel address space.
* @unmap: [optional] unmaps a page from the buffer.
* @vmap: [optional] creates a virtual mapping for the buffer into kernel
* address space. Same restrictions as for vmap and friends apply.
diff --git a/drivers/dma-buf/dma-buf.c b/drivers/dma-buf/dma-buf.c
index d78d5fc173..4c45e31258 100644
--- a/drivers/dma-buf/dma-buf.c
+++ b/drivers/dma-buf/dma-buf.c
@@ -872,6 +872,8 @@ void *dma_buf_kmap_atomic(struct dma_buf *dmabuf, unsigned long page_num)
{
WARN_ON(!dmabuf);
+ if (!dmabuf->ops->map_atomic)
+ return NULL;
return dmabuf->ops->map_atomic(dmabuf, page_num);
}
EXPORT_SYMBOL_GPL(dma_buf_kmap_atomic);
@@ -907,6 +909,8 @@ void *dma_buf_kmap(struct dma_buf *dmabuf, unsigned long page_num)
{
WARN_ON(!dmabuf);
+ if (!dmabuf->ops->map)
+ return NULL;
return dmabuf->ops->map(dmabuf, page_num);
}
EXPORT_SYMBOL_GPL(dma_buf_kmap);
--
2.9.3