Re: [PATCH 2/4] drm/i915: protect backlight registers and data with a spinlock

From: Daniel Vetter <daniel@ffwll.ch>
To: Chris Wilson <chris@chris-wilson.co.uk>,
	Daniel Vetter <daniel@ffwll.ch>,
	Jani Nikula <jani.nikula@intel.com>,
	intel-gfx <intel-gfx@lists.freedesktop.org>
Subject: Re: [PATCH 2/4] drm/i915: protect backlight registers and data with a spinlock
Date: Mon, 15 Apr 2013 23:40:30 +0200	[thread overview]
Message-ID: <CAKMK7uFVbmwygKEU6b5CsrS0YY8yy-EY-UASUJFHrFknZBK6Ow@mail.gmail.com> (raw)
In-Reply-To: <20130415204938.GC4106@cantiga.alporthouse.com>

On Mon, Apr 15, 2013 at 10:49 PM, Chris Wilson <chris@chris-wilson.co.uk> wrote:
> On Mon, Apr 15, 2013 at 06:59:58PM +0200, Daniel Vetter wrote:
>> On Mon, Apr 15, 2013 at 6:38 PM, Jani Nikula <jani.nikula@intel.com> wrote:
>> > On Mon, 15 Apr 2013, Chris Wilson <chris@chris-wilson.co.uk> wrote:
>> >> On Fri, Apr 12, 2013 at 03:18:37PM +0300, Jani Nikula wrote:
>> >>> Backlight data and registers are fiddled through LVDS/eDP modeset
>> >>> enable/disable hooks, backlight sysfs files, asle interrupts, and register
>> >>> save/restore. Protect the backlight related registers and driver private
>> >>> fields using a spinlock.
>> >>>
>> >>> The locking in register save/restore covers a little more than is strictly
>> >>> necessary, including non-modeset case, for simplicity.
>> >>>
>> >>> v2: Cover register access, save/restore, i915_read_blc_pwm_ctl() and code
>> >>>     paths leading there.
>> >>>
>> >>> Signed-off-by: Jani Nikula <jani.nikula@intel.com>
>> >>
>> >> Looks reasonable.
>> >>
>> >> intel_panel_actually_set_backlight() should have a WARN_ON(!spinlocked);
>> >>
>> >> The irqness of the register writes scares me slightly - since the IRQ in
>> >> question is from ACPI and we have a few bug reports along the lines of
>> >> "backlight makes the entire system sluggish" i.e. commonly associated
>> >> with bad interrupt handling. Whilst you are looking at updating the
>> >> backlight programming, can you look at pushing the writes from out
>> >> of the interrupt handler?
>> >
>> > So, add a work to do the register writes, and change the spinlock into a
>> > mutex while at it? Should be fairly simple, if you think that's the way
>> > to go.
>>
>> I think I'll go ahead with the spinlock fix here for 3.10 and we can
>> look into offloading this all for 3.11. Also, Chris do you remember
>> one of these reports - I've kinda never noticed that particular kind
>> of suck?
>
> I maybe way off the mark and the cause is likely just to be a sucky ACPI
> firmware:
> https://bugs.launchpad.net/ubuntu/+source/xserver-xorg-video-intel/+bug/1019370
>
> spinlock -> mutex + workqueue would mitigate against any bad firmware
> being run under irq context.


Hm, that's a funny one indeed. Although it's strange that it doesn't
happen when not using X. Might be that the user doesn't notice it
(since all the interactive stuff - blinking cursor - is done in irq
context), but if this is indeed ACPI doing something crazy I have no
idea what's it doing ...
-Daniel
--
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch