[PATCH] i2c: exynos5: Initialise Samsung High Speed I2C controller early

[Naveen Krishna Ch]

Hello Wolfram,

On 21 May 2014 15:55, Wolfram Sang <wsa at the-dreams.de> wrote:
> On Fri, May 09, 2014 at 03:54:25PM +0100, Mark Brown wrote:
>> On Fri, May 09, 2014 at 08:12:47PM +0530, Naveen Krishna Ch wrote:
>> > On 9 May 2014 19:21, Mark Brown <broonie at kernel.org> wrote:
>> > > On Fri, May 09, 2014 at 05:50:00PM +0530, Naveen Krishna Ch wrote:
>>
>> > >> DRM related drivers like DP, FIMD, HDMI, Mixer wants to be probed ASAP
>> > >> during the boot.
>> > >> The real problem comes when, one of these drivers do a regulator_get().
>>
>> > >> If the physical supply  is not enabled/hookedup the regulator_get() call
>> > >> assumes that physical supply is present and returns a
>> > >> "dummy_regulator" (But, not an error).
>>
>> > >> Because of which, Display and several other devices fails to work.
>>
>> > > These drivers are buggy, if they geniunely expect and handle a missing
>> > > supply then they should be using regulator_get_optional() to request the
>> > > regulator and even if they don't the use of subsys_initcall() is not
>> > > going to fix anything here - if a dummy regulator is going to be
>> > > returned the time things are probed won't make a difference.
>>
>> ...
>>
>> > If all the I2C, SPI, DMA, I2C_TUNNEL, DRM based LCD are all mod_probes()
>> > DRM drivers are probing a head of the PMIC probe. Which is causing the
>> > display drivers to get "dummy_regulators" instead of the real supplies.
>>
>> No, it really won't - I have no idea what you are doing but it's not
>> mainline.  If you are getting dummy regulators then you don't have a
>> supply present at all and probe ordering isn't going to make a blind
>> bit of difference.
>>
>> Please provide a specific technical description of the problem you are
>> seeing in mainline.
>
> +1 Unless we have a clear understanding that this is the only acceptable
> solution in mainline, this is really an out-of-tree patch.

Few of the DRM based devices need regulators during their probe and
DRM subsystem
wants to bring up LCD as fast as it could to give a better user experience.

Exynos DRM (few others for that matter) does platform_driver_register()
of devices in a sequence and finally binds them all together into one
blob or /dev/dri-0 device.
Which makes it hard to implement -EPROBE_DEFER. Even if -EPROBE_DEFER is
implemented in DRM, it adds lot of overhead during the boot up.

I've proposed to make exynos_drm_init() as late_initcall() instead of making
I2C, SPI, DMA and I2C-arbitrator and I2C-TUNNEL drivers as subsys_initcall()

http://www.spinics.net/lists/linux-samsung-soc/msg30858.html

Kindly, suggest a workable approach for all the subsystems.
>

-- 
Shine bright,
(: Nav :)