Re: [PATCH] rpmsg: qcom_smd: Access APCS through mailbox framework

[Bjorn Andersson]

On Wed 06 Dec 11:54 PST 2017, Stephen Boyd wrote:

> On 12/06/2017 11:02 AM, Bjorn Andersson wrote:
> > On Wed 06 Dec 04:08 PST 2017, Jitendra Sharma wrote:
> >
> >> Hi Bjorn,
> >>
> > Hi Jitendra,
> >
> >> On 11/16/2017 12:38 PM, Bjorn Andersson wrote:
> > [..]
> >>> @@ -365,7 +371,12 @@ static void qcom_smd_signal_channel(struct qcom_smd_channel *channel)
> >>>   {
> >>>   	struct qcom_smd_edge *edge = channel->edge;
> >>> -	regmap_write(edge->ipc_regmap, edge->ipc_offset, BIT(edge->ipc_bit));
> >>> +	if (edge->mbox_chan) {
> >>> +		mbox_send_message(edge->mbox_chan, NULL);
> >> mbox_send_message could fail. So return value should be checked
> > qcom_apcs_ipc_send_data() can't fail, so the case when
> > mbox_send_message() would fail is if more than MBOX_TX_QUEUE_LEN (20)
> > callers that has managed to put their data in the queue but not yet
> > execute msg_submit().
> >
> > As each bit in the APCS IPC register is modelled as it's own mailbox
> > channel this error case would mean that as mbox_send_message() returns
> > with an error there will soon be 20 callers entering
> > qcom_apcs_ipc_send_data() and trigger this very bit.
> >
> >
> > When this happens mbox_send_message() will print an error in the log, so
> > there's no point in having the caller also print an error.
> >
> > When it comes to dealing with a failing call to mbox_send_message() we
> > have already posted the message in the FIFO, so we have no way to abort
> > the transmission, as such the only way to deal with this is to either
> > retry or ignore the problem; and the mailbox queue will ensure that we
> > retry 20 times.
> >
> 
> Maybe you should wrap this up into a comment in the code? Then we don't
> have to dig this out of the mail list archives to figure out why we
> aren't checking for an error.
> 

Sounds very reasonable, I will do so.

Regards,
Bjorn