From: Sudeep Holla sudeep.holla@arm.com
[ Upstream commit 9dc34d635c67e57051853855c43249408641a5ab ]
Sometimes platfom may take too long to respond to the command and OS might timeout before platform transfer the ownership of the shared memory region to the OS with the response.
Since the mailbox channel associated with the channel is freed and new commands are dispatch on the same channel, OS needs to wait until it gets back the ownership. If not, either OS may end up overwriting the platform response for the last command(which is fine as OS timed out that command) or platform might overwrite the payload for the next command with the response for the old.
The latter is problematic as platform may end up interpretting the response as the payload. In order to avoid such race, let's wait until the OS gets back the ownership before we prepare the shared memory with the payload for the next command.
Reported-by: Jim Quinlan james.quinlan@broadcom.com Signed-off-by: Sudeep Holla sudeep.holla@arm.com Signed-off-by: Sasha Levin sashal@kernel.org --- drivers/firmware/arm_scmi/driver.c | 8 ++++++++ 1 file changed, 8 insertions(+)
diff --git a/drivers/firmware/arm_scmi/driver.c b/drivers/firmware/arm_scmi/driver.c index b5bc4c7a8fab2..b49c9e6f4bf10 100644 --- a/drivers/firmware/arm_scmi/driver.c +++ b/drivers/firmware/arm_scmi/driver.c @@ -271,6 +271,14 @@ static void scmi_tx_prepare(struct mbox_client *cl, void *m) struct scmi_chan_info *cinfo = client_to_scmi_chan_info(cl); struct scmi_shared_mem __iomem *mem = cinfo->payload;
+ /* + * Ideally channel must be free by now unless OS timeout last + * request and platform continued to process the same, wait + * until it releases the shared memory, otherwise we may endup + * overwriting its response with new message payload or vice-versa + */ + spin_until_cond(ioread32(&mem->channel_status) & + SCMI_SHMEM_CHAN_STAT_CHANNEL_FREE); /* Mark channel busy + clear error */ iowrite32(0x0, &mem->channel_status); iowrite32(t->hdr.poll_completion ? 0 : SCMI_SHMEM_FLAG_INTR_ENABLED,