Pin configuration conflict and NULL pointer dereference on RPi 3

[Michael Zoran]

On Mon, 2017-02-13 at 12:27 -0800, Eric Anholt wrote:
> Stefan Wahren <stefan.wahren at i2se.com> writes:
> 
> > Hi Gerd,
> > 
> > > Gerd Hoffmann <kraxel at redhat.com> hat am 2. Juni 2016 um 18:07
> > > geschrieben:
> > > 
> > > 
> > > On Do, 2016-06-02 at 11:31 +0200, Gerd Hoffmann wrote:
> > > >   Hi,
> > > > 
> > > > > > Patches 20-32 are for the wifi.  Mostly from Eric's
> > > > > > bcm2837-wireless
> > > > > > branch.  They clearly need some more love, but do the job
> > > > > > for now.
> > > > > 
> > > > > I just came up with a fix that should replace patch #29,
> > > > > which means
> > > > > I'll trust the expander stuff a lot more.  With the expander
> > > > > working,
> > > > > you should be able to use pin 4 on it for HDMI hotplug
> > > > > detect, at which
> > > > > point graphics should work.
> > > > 
> > > > Tried that.  i2c not working.  There seems to be a disagreement
> > > > who owns
> > > > gpio pin 45:
> > > > 
> > > > [root at pi-three linux]# dmesg | grep i2c
> > > > [    2.354380] i2c /dev entries driver
> > > > [    3.496734] pinctrl-bcm2835 3f200000.gpio: pin gpio45
> > > > already
> > > > requested by 3f20c000.pwm; cannot claim for 3f205000.i2c
> > > > [    3.512506] pinctrl-bcm2835 3f200000.gpio: pin-45
> > > > (3f205000.i2c)
> > > > status -22
> > > > [    3.540488] i2c-bcm2835 3f205000.i2c: Error applying
> > > > setting, reverse
> > > > things back
> > > > [    4.551683] i2c-bcm2835 3f205000.i2c: i2c transfer timed out
> > > > [    4.574599] i2c-bcm2835 3f805000.i2c: Could not read clock-
> > > > frequency
> > > > property
> > > 
> > > Tried to disable pwm in dt:
> > > 
> > >    -       status = "okay";
> > >    +       status = "disabled";
> > > 
> > > Now i2c is happy, and fxl6408 seems to work too:
> > > 
> > >    [root at pi-two ~]# dmesg | grep -i FXL6408
> > >    [    3.580660] fxl6408 0-0043: FXL6408 probe returned DID:
> > > 0x87
> > > 
> > 
> > i stumbled yesterday on this patch [1] which is related to this
> > (still open) issue. I wrote to the patch author and he said that
> > sometimes the i2c-bcm2835 on his RPi 3 (Aarch64 Kernel) crashed
> > with a NULL pointer dereference. He also said that the random I2C
> > timeouts like above
> > 
> > i2c-bcm2835 3f805000.i2c: i2c transfer timed out
> > 
> > still occurs. Also reported here [2].
> > 
> > Can you confirm any of these issues (requires PWM_BCM2835 and
> > I2C_BCM2835 enabled)?
> > 
> > I've the suspicion that the unexpected interrupts are triggered by
> > the pin muxing conflict at pin 45. Please refer to the firmware DT
> > blob [3].
> > 
> > In order to fix the NULL pointer dereference i suggest the
> > following:
> 
> Just a reminder: This FXL6408 driver will not be usable on the
> Raspberry
> Pi.  On Pi3, the firmware is looping hitting the only I2C controller
> that can talk to the pins attached to the FXL6408 (for its
> undervoltage
> detection).  If Linux tries to drive that controller on its own,
> everything will go wrong.

FYI: I don't understand why I'm getting some of the messages posted to
linux-rpi-kernel but not all of them.  Is it a moderated e-mail list or
something? I see many more messages on the archive then I'm actually
getting.

If HDMI hotplug is important to upstream, then you have several
solutions to this issue. They range from friendly say as upstreaming
the firmware based GPIO expander driver, to unfriendly such as somehow
disabling the firmware's access to the expander.  I personally wouldn't
recommend the unfriendly solution since I suspect it's going to be
very, very difficult to make the firmware completely disappear.