[Intel-gfx] v3.9-rc1 instability on Chromebook Pixel with gmbus irq

[Daniel Vetter]

On Wed, Mar 6, 2013 at 3:35 AM, Benson Leung <bleung at chromium.org> wrote:
> I'm working on touch devices Chromium OS, and I've noticed a
> regression between 3.8 and 3.9-rc1, which was posted yesterday.
>
> The hardware in question is a Chromebook Pixel. For this device, we
> have i2c input devices: atmel mxt224s touchpad and atmel mxt1664s
> touchscreen. The touchpad is on bus 1, "i915 gmbus vga" at 1-004b. The
> touchscreen is on bus 2, "i915 gmbus panel" at 2-004a.
>
> I was testing v3.9-rc1 on the Pixel and the touchscreen driver is
> being returned -110 (-ETIMEDOUT) on an i2c_transfer after several
> seconds of both touch devices working correctly. At the time of the
> failure, there are no error messages from GMBUS that I can see, but
> the bus never recovers. I can keep interacting with the touchscreen or
> touchpad, and the interrupts trigger reads, but all subsequent reads
> return -110.
>
> I noticed that between 3.8 and 3.9-rc1, your patch series to add gmbus
> irq support was merged. After bisecting, I found that this commit
> seems to be causing the timeout problem. Reverting it makes the
> problem go away, and the bus is stable.

Can you please retest with latest drm-intel-fixes merged into -rc1?
Paulo's patch fixes a race in handling PCH interrupts (where the gmbus
hw is) which matches rather well with your description here.

Thanks, Daniel

> commit 2c438c0273b76d6cb158f8bdd0aa3ebf66e48a28
> Author: Daniel Vetter <daniel.vetter at ffwll.ch>
> Date:   Sat Dec 1 13:53:46 2012 +0100
>
>     drm/i915: use gmbus irq to wait for gmbus idle
>
>     GMBUS_ACTIVE has inverted sense and so doesn't fit into the
>     wait_hw_status helper, hence create a new gmbus_wait_idle functions.
>     Also, we only care about the idle irq event and nothing else, which
>     allows us to use the wait_event_timeout helper directly without
>     jumping through hoops to catch NAKs.
>
>     Since gen2/3 don't have gmbus interrupts, handle them separately with
>     the old wait_for macro.
>
>     This shaves another few ms off reading EDID from a hdmi screen on my
>     testbox here. EDID reading with interrupt driven gmbus is now as fast
>     as with busy-looping gmbus at 28 ms here (with negligible cpu
>     overhead).
>
>     Reviewed-by: Imre Deak <imre.deak at intel.com>
>     Signed-off-by: Daniel Vetter <daniel.vetter at ffwll.ch>
>
>
> Is there anything I can do to help debug this some more?
>
> --
> Benson Leung
> Software Engineer, Chrom* OS
> bleung at chromium.org



-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch