Hi Loic,
On 3/10/21 3:38 AM, Loic Poulain wrote:
Some buggy hardwares (e.g sdx24) may report the current command
ring wp pointer instead of the command completion pointer. It's
obviously wrong, causing completion timeout. We can however deal
with that situation by completing the cmd n-1 element, which is
what the device actually completes.
Signed-off-by: Loic Poulain <loic.poulain@xxxxxxxxxx>
---
drivers/bus/mhi/core/main.c | 18 ++++++++++++++++++
1 file changed, 18 insertions(+)
diff --git a/drivers/bus/mhi/core/main.c b/drivers/bus/mhi/core/main.c
index 16b9640..3e3c520 100644
--- a/drivers/bus/mhi/core/main.c
+++ b/drivers/bus/mhi/core/main.c
@@ -707,6 +707,7 @@ static void mhi_process_cmd_completion(struct mhi_controller *mhi_cntrl,
{
dma_addr_t ptr = MHI_TRE_GET_EV_PTR(tre);
struct mhi_cmd *cmd_ring = &mhi_cntrl->mhi_cmd[PRIMARY_CMD_RING];
+ struct device *dev = &mhi_cntrl->mhi_dev->dev;
struct mhi_ring *mhi_ring = &cmd_ring->ring;
struct mhi_tre *cmd_pkt;
struct mhi_chan *mhi_chan;
@@ -714,6 +715,23 @@ static void mhi_process_cmd_completion(struct mhi_controller *mhi_cntrl,
cmd_pkt = mhi_to_virtual(mhi_ring, ptr);
+ if (unlikely(cmd_pkt == mhi_ring->wp)) {
As per spec : The location of the command ring read pointer is reported
to the host on the command completion events in the primary event ring.
If device is buggy and updates with WP instead of Rp, we should not
workaround it by processing Wp - 1. We can print a warning if cmd_pkt !=
mhi_ring->rp and let the command completion timeout. This needs to be
fixed by device. We can not accommodate device side bug in host side.
+ /* Some buggy hardwares (e.g sdx24) sometimes report the current
+ * command ring wp pointer instead of the command completion
+ * pointer. It's obviously wrong, causing completion timeout. We
+ * can however deal with that situation by completing the cmd
+ * n-1 element.
+ */
+ void *ring_ptr = (void *)cmd_pkt - mhi_ring->el_size;
+
+ if (ring_ptr < mhi_ring->base)
+ ring_ptr += mhi_ring->len;
+
+ cmd_pkt = ring_ptr;
+
+ dev_warn(dev, "Bad completion pointer (ptr == ring_wp)\n");
+ }
+
chan = MHI_TRE_GET_CMD_CHID(cmd_pkt);
mhi_chan = &mhi_cntrl->mhi_chan[chan];
write_lock_bh(&mhi_chan->lock);
Hi Mani,
What do you think about this workaround ?
Thanks,
Hemant
--
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project
