Re: [REGRESSION] PCI v5.5-rc1 breaks google kevin

[Vicente Bergas <vicencb@gmail.com>]

On Thursday, December 12, 2019 3:16:25 PM CET, Enric Balletbo Serra wrote:
> Hi Vicente,
>
> Missatge de Andrew Murray <andrew.murray@arm.com> del dia dj., 12 de
> des. 2019 a les 1:53:
>> 
>> On Thu, Dec 12, 2019 at 12:12:56AM +0000, Robin Murphy wrote:
>>> Hi Vicente,
>>> 
>>> On 2019-12-11 11:38 pm, Vicente Bergas wrote:
>>>> Hi,
>>>> since v5.5-rc1 the google kevin chromebook does not boot.
>>>> Git bisect reports 5e0c21c75e8c PCI/ASPM: Remove pcie_aspm_enabled()
>>>> unnecessary locking
>>>> as the first bad commit.
>>>> 
>>>> In order to revert it from v5.5-rc1 i had to also revert 
>>>> some dependencies:
>>>> 5e0c21c75e8c08375a69710527e4a921b897cb7e
>>>> aff5d0552da4055da3faa27ee4252e48bb1f5821
>>>> 35efea32b26f9aacc99bf07e0d2cdfba2028b099
>>>> 687aaf386aeb551130f31705ce40d1341047a936
>>>> 72ea91afbfb08619696ccde610ee4d0d29cf4a1d
>>>> 87e90283c94c76ee11d379ab5a0973382bbd0baf
>>>> After reverting all of this, still no luck.
>>>> So, either the results of git bisect are not to be trusted, or
>>>> there are more bad commits.
>>>> 
>>>> By "does not boot" i mean that the display fails to start and
>>>> the display is the only output device, so debugging is quite difficult.
>>> 
>
> Another issue that is affecting current mainline for kevin is fixed
> with [1]. As usual, I have a tracking branch for 5.5 for different
> Chromebooks with some not yet merged patches that makes things work
> while are not fixed [2]. For kevin only the mentioned ASoC patch [1]
> and the pcie fix [3] should be needed. Other than that display is
> working for me on Kevin.
>
> Cheers,
>  Enric
>
> [1] 
> https://git.kernel.org/pub/scm/linux/kernel/git/broonie/sound.git/commit/?h=for-5.5&id=4bf2e385aa59c2fae5f880aa25cfd2b470109093
> [2] 
> https://gitlab.collabora.com/eballetbo/linux/commits/topic/chromeos/somewhat-stable-5.5
> [3]  https://lkml.org/lkml/2019/12/11/199
>
>>> Assuming it's a manifestation of the same PCI breakage that Enric and
>>> Lorenzo figured out, there's a proposed fix here:
>>> https://lkml.org/lkml/2019/12/11/199
>> 
>> It's likely that any PCI driver that uses PCI IO with that controller will
>> suffer the same fate.
>> 
>> Vicente - can you try the patch that has been proposed and verify it fixes
>> the issue for you?
>> 
>> Thanks,
>> 
>> Andrew Murray
>> 
>>> 
>>> Robin.
>>> 
>>>> v5.5-rc1 as is (reverting no commits at all) works fine when 
>>>> disabling PCI:
>>>> # CONFIG_PCI is not set
>>>> 
>>>> Regards,
>>>>   Vicente.

Hi Robin, Andrew and Enric,
thank you all for the quick responses!
I can confirm that patch [3] fixes the issue reported in this email and
that [1] fixes the other issue reported on the other email.

Regards,
  Vicenç.