Re: [PATCH 0/2] pwm: lpss: Force runtime-resume on suspend on CHT devices

[Thierry Reding <thierry.reding@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 1730 bytes --]

On Sun, Oct 14, 2018 at 05:12:00PM +0200, Hans de Goede wrote:
> Hi All,
> 
> Here are 2 more / new pwm-lpss fixes.
> 
> I've received a bug report that the recent ACPI-LPSS code changes which
> properly order device resume on Cherry Trail so that the GFX0 _PS0 method
> no longer exits with an error causes the backlight to flicker on some
> Cherry Trail devices.
> 
> This series is the result of investigating this issue. On Cherry Trail
> devices under Windows the PWM controller used for the backlight is
> considered part of the GPU even though it is part of the LPSS block and
> thus is an entirely different independent hardware unit.
> 
> Because of this on Cherry Trail the GPU's (GFX0 ACPI node) _PS3 and _PS0
> ACPI methods save and restore the PWM controller registers.
> 
> After the recent fixes this save + restore of the PWM controller registers
> is actually also happening under Linux.
> 
> If userspace blanks the screen before suspending, such as e.g. GNOME
> does, then the PWM controller will be runtime-suspended when the suspend
> starts. This causes the GFX0 _PS3 method to save a value of 0xffffffff
> for the PWM control register and to restore this value on resume.
> Note that the GPU's _PS0 method also puts the PWM controller in D0, so
> this write of 0xffffffff does actually stick.
> 
> The first patch in this patch-set fixes this. This fix exposes another
> issue, the pwm-lpss driver unconditionally sets the update bit when
> pwm_apply gets called, even if nothing has changed. At least on
> Cherry Trail this seems to cause the next update (which actually changes
> something) to be ignored. The second patch fixes this issue.

Applied, thanks.

Thierry

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]