[PATCH 00/18] i2c: remove printout on handled timeouts
Andy Shevchenko
andy.shevchenko at gmail.com
Wed Apr 24 05:44:19 PDT 2024
On Wed, Apr 24, 2024 at 3:41 PM Andy Shevchenko
<andy.shevchenko at gmail.com> wrote:
> On Wed, Apr 24, 2024 at 12:00 PM Wolfram Sang
> <wsa+renesas at sang-engineering.com> wrote:
> > On Wed, Apr 24, 2024 at 03:08:14AM +0300, Andy Shevchenko wrote:
> > > Wed, Apr 10, 2024 at 01:24:14PM +0200, Wolfram Sang kirjoitti:
> > > > While working on another cleanup series, I stumbled over the fact that
> > > > some drivers print an error on I2C or SMBus related timeouts. This is
> > > > wrong because it may be an expected state. The client driver on top
> > > > knows this, so let's keep error handling on this level and remove the
> > > > prinouts from controller drivers.
> > > >
> > > > Looking forward to comments,
> > >
> > > I do not see an equivalent change in I²C core.
> >
> > There shouldn't be. The core neither knows if it is okay or not. The
> > client driver knows.
> >
> > > IIRC in our case (DW or i801 or iSMT) we often have this message as the only
> >
> > Often? How often?
>
> Once in a couple of months I assume. Last time it was a few weeks ago
> that there was a report and they pointed to this very message which
> was helpful.
>
> > > one that points to the issues (on non-debug level), it will be much harder to
> > > debug for our customers with this going away.
> >
> > The proper fix is to print the error in the client driver. Maybe this
> > needs a helper for client drivers which they can use like:
> >
> > i2c_report_error(client, retval, flags);
> >
> > The other thing which is also more helpful IMO is that we have
> > trace_events for __i2c_transfer. There, you can see what was happening
> > on the bus before the timeout. It can easily be that, if device X
> > times out, it was because of the transfer before to device Y which locks
> > up the bus. A simple "Bus timed out" message will not help you a lot
> > there.
>
> The trace events are good to have, not sure if production kernels have
> them enabled, though.
Ah, and to accent on the usefulness of the message that happens before
one thinks about enabling trace events. How usual is that we _expect_
something to fail? Deeper debugging usually happens after we have
noticed the issue. I'm not sure if there is an equivalent to signal
about a problem without expecting it to happen. Is that -ETIMEDOUT
being converted to some message somewhere?
> > And, keep in mind the false positives I mentioned in the coverletter.
--
With Best Regards,
Andy Shevchenko
More information about the Linux-rockchip
mailing list