Re: [1/4] gpio: gpio-f7188x: Use mutex for access serialisation.

From: Vincent Pelletier <plr.vincent@gmail.com>
To: Simon Guinot <simon.guinot@sequanux.org>
Cc: linux-gpio@vger.kernel.org,
	Vincent Donnefort <vdonnefort@gmail.com>,
	Yoann Sculo <yoann@sculo.fr>
Subject: Re: [1/4] gpio: gpio-f7188x: Use mutex for access serialisation.
Date: Fri, 21 Aug 2015 22:48:24 +0200	[thread overview]
Message-ID: <20150821224824.3406caa0@x2> (raw)
In-Reply-To: <20150821175216.GE1729@kw.sim.vm.gnt>

Hello Simon,

Thanks for reviewing my patch.

On Fri, 21 Aug 2015 19:52:16 +0200, Simon Guinot
<simon.guinot@sequanux.org> wrote:
> IMHO, understand *clearly* the issue could be a good start in order to
> fix it efficiently.

I totally agree, but have no idea how to debug further.
Could you suggest any mechanism to debug *_resource ?

> Please, could you describe a setup (as simple as possible) allowing to
> reproduce the issue ? I'll try it on my side.
[...]
> Please try to make the module list needed to reproduce the issue as
> short as possible. Ideally only gpio_f7188x would be needed.

The simplest I could find so far needs gpio-input-polled with 20ms
polling period (I didn't try to change polling period), and a shell loop
writing to gpioXX/value.

I'm using the following platform driver to declare a single
gpio-input-polled key:
  https://github.com/vpelletier/linux/blob/free_nonexistent_resource/drivers/platform/x86/qnap-tsx51.c
Full version of this driver:
  https://github.com/vpelletier/linux/blob/ts651/drivers/platform/x86/qnap-tsx51.c
  (ignore code surrounded with '#if QNAP_TSX51_GPIOD', it is
  still broken).

It should be easy to create a variant fitting another board, as long
as you can recycle two GPIOs.

Once I've unloaded all other listed drivers for this chip and loaded
the stripped-down qnap-tsx51 version:
Terminal 0:
  echo 0 > /sys/devices/system/cpu/cpu1/online
Terminal 1:
  cd /sys/class/gpio/
  echo 62 > export
  cd gpio62
  echo "out" > direction
  while :; echo 0 > value; echo 1 > value; done
No error so far.
Terminal 1:
  echo 1 > /sys/devices/cpu/cpu1/online
After a few (10 to 30) seconds, tty is flooded with the error I
reported.

Stopping the loop stop the error.
I have seen rare cases where a few extra messages could occur within
the next minute or so, but this was before I stripped most out of the
platform driver.

With just two such loops poking at (out) GPIOs, it does not happen.

I'm testing this on v4.1.4 with the following patch applied (the error
happened before and after that patch, so it likely does not matter):
  https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=7e08117de6ee17ae6c8f2983999a98cb95eb9bc2

CPU is an intel celeron J1800, dual-core at 2.41GHz.
Exact board is this machine's motherboard:
  https://www.qnap.com/i/fr/product/model.php?II=144

> Unfortunately the f71882fg and f71889f Super-I/Os don't provide an
> I/O region dedicated to GPIOs. If it was the case, I would have used
> this way. But for this Super-I/O models, the GPIOs have to be configured
> through the global registers. That's why your patch breaks support with
> this models.

Wow, I didn't expect such difference between models otherwise handled
the same way in hwmon/f71882fg.c . I guess the GPIO function works
differently from hwmon function in these models (hwmon still having
dedicated IO range, not GPIO).

Regards,
-- 
Vincent Pelletier