linux-imx

MLK-22835: mxc-jpeg: jpeg decoder stuck due to race condition

The race condition is on ctx->dht_needed, this variable indicates if the
decoder operates in one stage (decode) or two stages (inject default
Huffman table + decode).

While decoding the current buffer, and before the IP finishes it, a new
buffer in enqueued, jpeg parse begins, initializes ctx->dht_needed with
true, and does not get to finish parsing the jpeg headers. In the
meantime, the IP finishes decoding the current buffer and checks for
ctx->dht_needed, and believes it needs to run again the IP to finish
the job, which is wrong and gets the IP stuck.

This dht_needed property should be per buffer, not per context, so, add
a custom structure for the source buffer, struct mxc_jpeg_src_buf, and
don't forget to tell v4l2 core about its size, in mxc_jpeg_queue_init.

Signed-off-by: Mirela Rabulea <mirela.rabulea@nxp.com>
Reviewed-by: Robert Chiras <robert.chiras@nxp.com>
(cherry picked from commit 4601417e95d35e995e0d8347b8464090a511db0d)