[Intel-gfx] [PATCH v2] pwm: lpss: Make builtin so that i915 can find the pwm_backlight

[Jani Nikula]

On Wed, 08 Mar 2017, Hans de Goede <hdegoede at redhat.com> wrote:
> Hi,
>
> On 08-03-17 10:40, Jani Nikula wrote:
>> On Fri, 20 Jan 2017, Mika Westerberg <mika.westerberg at linux.intel.com> wrote:
>>> On Fri, Jan 20, 2017 at 10:02:50AM +0200, Jani Nikula wrote:
>>>> That said, I suppose there could be an alternative to handling pwm_get()
>>>> failures at probe. We could just go on with our init, but schedule a
>>>> retry later. Perhaps a bit hacky, but it would address both of the
>>>> concerns above. Again, this patch seems a simple workaround in the mean
>>>> time.
>>>
>>> Not sure if this works or how hacky it is, but can't you
>>> request_module() before you start looking up for the pwm?
>>
>> I eyeballed this a little, and noticed:
>>
>> drivers/acpi/acpi_lpss.c:
>>
>> static struct pwm_lookup bsw_pwm_lookup[] = {
>> 	PWM_LOOKUP_WITH_MODULE("80862288:00", 0, "0000:00:02.0",
>> 			       "pwm_backlight", 0, PWM_POLARITY_NORMAL,
>> 			       "pwm-lpss-platform"),
>> };
>>
>> drivers/mfd/intel_soc_pmic_core.c:
>>
>> static struct pwm_lookup crc_pwm_lookup[] = {
>> 	PWM_LOOKUP("crystal_cove_pwm", 0, "0000:00:02.0", "pwm_backlight", 0, PWM_POLARITY_NORMAL),
>> };
>>
>> Should crc_pwm_lookup also use PWM_LOOKUP_WITH_MODULE? And which module
>> exactly? pwm_get() does an automatic request_module(), if the module is
>> given.
>>
>> And will this still be enough?
>
> I was thinking about this a couple of days ago, unfortunately the
> situation with pwm_crc is more complicated as that is part of
> an i2c mfd device, so both the i2c-adapter driver and the mfd driver
> (intel_soc_pmic_crc) need to be builtin currently the mfd driver
> cannot be modular, but the i2c-adapter driver can (and on most
> Linux distribution kernels is) configured to be modular.
>
> So just doing request module for pwm-crc is not going to help
> since the i2c-adapter driver may not yet have loaded / initialized.
>
> All in all we really need to find a way to figure out if we will need
> to do a pwm_get earlier on during i915 initialization (by moving the
> VBT parsing to earlier, or at least part of it) and do the pwm_get
> before we do anything i-reversible and if it fails then return
> -EPROBE_DEFER. Then we can make pwm_crc modular as well as all of
> intel_soc_pmic* as it really should be.

The other alternatives are:

1) Handle defers using a workqueue within i915. It's a bit tedious, but
I didn't spot any show stoppers with the approach. We'd register a
non-functional backlight interface until the pwm_get() succeeds.

2) Rip out pwm backlight from i915 altogether, and turn it into a
separate platform backlight that uses pwm. I think there's ready
infrastructure for that. It's not without problems, though, as then we
lose control over the sequence in which backlight gets enabled/disabled.

BR,
Jani.

>
> Regards,
>
> Hans

-- 
Jani Nikula, Intel Open Source Technology Center