From: Sarannya S <quic_sarannya@quicinc.com>

On remote susbsystem restart rproc will stop glink subdev which will
trigger qcom_glink_native_remove, any ongoing intent wait should be
aborted from there otherwise this wait delays glink send which potentially
delays glink channel removal as well. This further introduces delay in ssr
notification to other remote subsystems from rproc.

Currently qcom_glink_native_remove is not setting channel->intent_received,
so any ongoing intent wait is not aborted on remote susbsystem restart.
abort_tx flag can be used as a condition to abort in such cases.

Adding abort_tx flag check in intent wait, to abort intent wait from
qcom_glink_native_remove.

Fixes: c05dfce0b89e ("rpmsg: glink: Wait for intent, not just request ack")
Cc: stable@vger.kernel.org
Signed-off-by: Sarannya S <quic_sarannya@quicinc.com>
Signed-off-by: Deepak Kumar Singh <quic_deesin@quicinc.com>
---
 drivers/rpmsg/qcom_glink_native.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/rpmsg/qcom_glink_native.c b/drivers/rpmsg/qcom_glink_native.c
index 82d460ff4777..ff828531c36f 100644
--- a/drivers/rpmsg/qcom_glink_native.c
+++ b/drivers/rpmsg/qcom_glink_native.c
@@ -438,7 +438,6 @@ static void qcom_glink_handle_intent_req_ack(struct qcom_glink *glink,
 
 static void qcom_glink_intent_req_abort(struct glink_channel *channel)
 {
-	WRITE_ONCE(channel->intent_req_result, 0);
 	wake_up_all(&channel->intent_req_wq);
 }
 
@@ -1354,8 +1353,9 @@ static int qcom_glink_request_intent(struct qcom_glink *glink,
 		goto unlock;
 
 	ret = wait_event_timeout(channel->intent_req_wq,
-				 READ_ONCE(channel->intent_req_result) >= 0 &&
-				 READ_ONCE(channel->intent_received),
+				 (READ_ONCE(channel->intent_req_result) >= 0 &&
+				 READ_ONCE(channel->intent_received)) ||
+				 glink->abort_tx,
 				 10 * HZ);
 	if (!ret) {
 		dev_err(glink->dev, "intent request timed out\n");
-- 
2.34.1

On Wed, Sep 25, 2024 at 12:53:28PM +0530, Deepak Kumar Singh wrote:
> From: Sarannya S <quic_sarannya@quicinc.com>
> 
> On remote susbsystem restart rproc will stop glink subdev which will

"When stopping or restarting a remoteproc the glink subdev stop will
invoke qcom_glink_native_remove(). Any ..."

> trigger qcom_glink_native_remove, any ongoing intent wait should be

Please always have () on function names, to make clear they are indeed
functions.

s/intent wait/wait for intent request/

And I think "can" is a better word than "should" (we can abort the wait
to not waiting for the intents that aren't expected to come).

> aborted from there otherwise this wait delays glink send which potentially
> delays glink channel removal as well. This further introduces delay in ssr
> notification to other remote subsystems from rproc.
> 
> Currently qcom_glink_native_remove is not setting channel->intent_received,

This observation is correct, but I don't see a reason why it should. So
express this in terms of the applicable logic (i.e. we have abort_tx to
signal this scenario already, let's use it)

> so any ongoing intent wait is not aborted on remote susbsystem restart.
> abort_tx flag can be used as a condition to abort in such cases.
> 
> Adding abort_tx flag check in intent wait, to abort intent wait from
> qcom_glink_native_remove.

More () please.

> 
> Fixes: c05dfce0b89e ("rpmsg: glink: Wait for intent, not just request ack")
> Cc: stable@vger.kernel.org

I don't think the current code is broken, just suboptimal. And as such I
don't think this is a bugfix.

Perhaps I'm missing some case here? Otherwise, please drop the Fixes and
Cc...

> Signed-off-by: Sarannya S <quic_sarannya@quicinc.com>
> Signed-off-by: Deepak Kumar Singh <quic_deesin@quicinc.com>
> ---
>  drivers/rpmsg/qcom_glink_native.c | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/rpmsg/qcom_glink_native.c b/drivers/rpmsg/qcom_glink_native.c
> index 82d460ff4777..ff828531c36f 100644
> --- a/drivers/rpmsg/qcom_glink_native.c
> +++ b/drivers/rpmsg/qcom_glink_native.c
> @@ -438,7 +438,6 @@ static void qcom_glink_handle_intent_req_ack(struct qcom_glink *glink,
>  
>  static void qcom_glink_intent_req_abort(struct glink_channel *channel)
>  {
> -	WRITE_ONCE(channel->intent_req_result, 0);
>  	wake_up_all(&channel->intent_req_wq);
>  }
>  
> @@ -1354,8 +1353,9 @@ static int qcom_glink_request_intent(struct qcom_glink *glink,
>  		goto unlock;
>  
>  	ret = wait_event_timeout(channel->intent_req_wq,
> -				 READ_ONCE(channel->intent_req_result) >= 0 &&
> -				 READ_ONCE(channel->intent_received),
> +				 (READ_ONCE(channel->intent_req_result) >= 0 &&
> +				 READ_ONCE(channel->intent_received)) ||
> +				 glink->abort_tx,

This looks good. But Chris and I talked about his patches posted
yesterday, and it seems like a good idea to differentiate the cases of
aborted and granted = 0.

Chris' patch is fixing a real bug, so that should be backported, so
let's conclude that discussion (with this patch included or in mind).

Regards,
Bjorn

>  				 10 * HZ);
>  	if (!ret) {
>  		dev_err(glink->dev, "intent request timed out\n");
> -- 
> 2.34.1
>