On Tue, Jan 28, 2025 at 9:36 PM Mikhail Rudenko <mike.rudenko@xxxxxxxxx> wrote:
>
> When support for V4L2_FLAG_MEMORY_NON_CONSISTENT was removed in
> commit 129134e5415d ("media: media/v4l2: remove
> V4L2_FLAG_MEMORY_NON_CONSISTENT flag"),
> vb2_dc_dmabuf_ops_{begin,end}_cpu_access() functions were made
> no-ops. Later, when support for V4L2_MEMORY_FLAG_NON_COHERENT was
> introduced in commit c0acf9cfeee0 ("media: videobuf2: handle
> V4L2_MEMORY_FLAG_NON_COHERENT flag"), the above functions remained
> no-ops, making cache maintenance for non-coherent dmabufs allocated by
> dma-contig impossible.
>
> Fix this by reintroducing dma_sync_sgtable_for_{cpu,device} and
> {flush,invalidate}_kernel_vmap_range calls to
> vb2_dc_dmabuf_ops_{begin,end}_cpu_access() functions for non-coherent
> buffers.
>
> Fixes: c0acf9cfeee0 ("media: videobuf2: handle V4L2_MEMORY_FLAG_NON_COHERENT flag")
> Cc: stable@xxxxxxxxxxxxxxx
> Signed-off-by: Mikhail Rudenko <mike.rudenko@xxxxxxxxx>
> ---
> .../media/common/videobuf2/videobuf2-dma-contig.c | 22 ++++++++++++++++++++++
> 1 file changed, 22 insertions(+)
>
> diff --git a/drivers/media/common/videobuf2/videobuf2-dma-contig.c b/drivers/media/common/videobuf2/videobuf2-dma-contig.c
> index bb0b7fa67b539aa73ad5ccf3c3bc318e26f8a4cb..146d7997a0da5989fb081a6f28ce0641fe726e63 100644
> --- a/drivers/media/common/videobuf2/videobuf2-dma-contig.c
> +++ b/drivers/media/common/videobuf2/videobuf2-dma-contig.c
> @@ -427,6 +427,17 @@ static int
> vb2_dc_dmabuf_ops_begin_cpu_access(struct dma_buf *dbuf,
> enum dma_data_direction direction)
> {
> + struct vb2_dc_buf *buf = dbuf->priv;
> + struct sg_table *sgt = buf->dma_sgt;
> +
> + if (!buf->non_coherent_mem)
> + return 0;
> +
> + if (buf->vaddr)
> + invalidate_kernel_vmap_range(buf->vaddr, buf->size);
> +
> + dma_sync_sgtable_for_cpu(buf->dev, sgt, buf->dma_dir);
> +
> return 0;
> }
>
> @@ -434,6 +445,17 @@ static int
> vb2_dc_dmabuf_ops_end_cpu_access(struct dma_buf *dbuf,
> enum dma_data_direction direction)
> {
> + struct vb2_dc_buf *buf = dbuf->priv;
> + struct sg_table *sgt = buf->dma_sgt;
> +
> + if (!buf->non_coherent_mem)
> + return 0;
> +
> + if (buf->vaddr)
> + flush_kernel_vmap_range(buf->vaddr, buf->size);
> +
> + dma_sync_sgtable_for_device(buf->dev, sgt, buf->dma_dir);
> +
> return 0;
> }
I took some time (over)thinking the kernel vmap range synchronization,
because these functions can be called both from the kernel space using
respective dma_buf_*() kAPI and also from the user space using the
DMA_BUF_SYNC IOCTLs, so we could in theory have the multiple
invocations racing with each other, but then I realized that we don't
really provide any guarantees for concurrent writes and reads from the
CPU, so I believe this should work fine. Sorry for the delay.
Acked-by: Tomasz Figa <tfiga@xxxxxxxxxxxx>
Let me add @Christoph Hellwig and @Robin Murphy just in case I'm wrong
on that, though... Hans, let's give them some time to take a look
before applying this.
Best regards,
Tomasz
