This patch fixes an issue that a race condition happens between a client
driver and the rcar-dmac driver:
- The rcar_dmac_isr_transfer_end() is called.
- The done list appears, and desc.running is the next active list.
- rcar_dmac_chan_get_residue() is called by a client driver before
rcar_dmac_isr_channel_thread() is called.
- The rcar_dmac_chan_get_residue() will not find any descriptors.
- And, the following WARNING happens:
WARN(1, "No descriptor for cookie!");
The sh-sci driver with HSCIF (921,600bps) on R-Car H3 can cause this
situation.
So, this patch checks the done lists in rcar_dmac_chan_get_residue()
and returns zero if the done lists has the argument cookie.
Tested-by: Nguyen Viet Dung <dung.nguyen.aj@xxxxxxxxxxx>
Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@xxxxxxxxxxx>
---
drivers/dma/sh/rcar-dmac.c | 9 +++++++++
1 file changed, 9 insertions(+)
diff --git a/drivers/dma/sh/rcar-dmac.c b/drivers/dma/sh/rcar-dmac.c
index 2b2c7db..f748be6 100644
--- a/drivers/dma/sh/rcar-dmac.c
+++ b/drivers/dma/sh/rcar-dmac.c
@@ -1264,8 +1264,17 @@ static unsigned int rcar_dmac_chan_get_residue(struct rcar_dmac_chan *chan,
* If the cookie doesn't correspond to the currently running transfer
* then the descriptor hasn't been processed yet, and the residue is
* equal to the full descriptor size.
+ * Also, a client driver is possible to call this function before
+ * rcar_dmac_isr_channel_thread() runs. In this case, the "desc.running"
+ * will be the next descriptor, and the done list will appear. So, if
+ * the argument cookie matches the done list's cookie, we can assume
+ * the residue is zero.
*/
if (cookie != desc->async_tx.cookie) {
+ list_for_each_entry(desc, &chan->desc.done, node) {
+ if (cookie == desc->async_tx.cookie)
+ return 0;
+ }
list_for_each_entry(desc, &chan->desc.pending, node) {
if (cookie == desc->async_tx.cookie)
return desc->size;
--
1.9.1
