Re: 5.14-rc failure to resume

[Jens Axboe <axboe@kernel.dk>]

On 7/24/21 3:35 PM, Jens Axboe wrote:
> On 7/24/21 3:15 PM, Andy Shevchenko wrote:
>> On Sat, Jul 24, 2021 at 11:26 PM Jens Axboe <axboe@kernel.dk> wrote:
>>>
>>> On 7/24/21 2:05 PM, Andy Shevchenko wrote:
>>>> On Sat, Jul 24, 2021 at 10:43 PM Jens Axboe <axboe@kernel.dk> wrote:
>>>>> On 7/24/21 11:56 AM, Jens Axboe wrote:
>>>>>> On 7/24/21 9:57 AM, Jens Axboe wrote:
>>>>>>> Hi,
>>>>>>>
>>>>>>> I ran into this when doing the last bit of testing on pending changes
>>>>>>> for this release on the laptop. Outside of running testing on these
>>>>>>> changes, I always build and boot current -git and my changes on my
>>>>>>> laptop as well.
>>>>>>>
>>>>>>> 5.14-rc1 + changes works fine, current -git and changes fail to resume
>>>>>>> every single time. I just get a black screen. Tip of tree before merging
>>>>>>> fixes is:
>>>>>>>
>>>>>>> commit 704f4cba43d4ed31ef4beb422313f1263d87bc55 (origin/master, origin/HEAD, master)
>>>>>>> Merge: 05daae0fb033 0077a5008272
>>>>>>> Author: Linus Torvalds <torvalds@linux-foundation.org>
>>>>>>> Date:   Fri Jul 23 11:30:12 2021 -0700
>>>>>>>
>>>>>>>     Merge tag 'ceph-for-5.14-rc3' of git://github.com/ceph/ceph-client
>>>>>>>
>>>>>>> Since bisection takes forever on the laptop (gen7 x1 carbon), I
>>>>>>> opportunistically reverted some of the most recent git pulls:
>>>>>>>
>>>>>>> - ec6badfbe1cde0eb2bec4a0b8f6e738171156b5b (acpi changes)
>>>>>>> - 1d597682d3e669ec7021aa33d088ed3d136a5149 (driver-core changes)
>>>>>>> - 74738c556db6c7f780a8b98340937e55b72c896a (usb changes)
>>>>>>> - e7562a00c1f54116f5a058e7e3ddd500188f60b2 (sound changes)
>>>>>>> - 8baef6386baaefb776bdd09b5c7630cf057c51c6 (drm changes)
>>>>>>>
>>>>>>> as they could potentially be involved, but even with all of those
>>>>>>> reverted it still won't resume.
>>>>>>>
>>>>>>> Sending this out in case someone has already reported this and I just
>>>>>>> couldn't find it. If this is a new/unknown issues, I'll go ahead and
>>>>>>> bisect it.
>>>>>>
>>>>>> Ran a bisect, and it pinpoints:
>>>>>>
>>>>>> 71f6428332844f38c7cb10461d9f29e9c9b983a0 is the first bad commit
>>>>>> commit 71f6428332844f38c7cb10461d9f29e9c9b983a0
>>>>>> Author: Andy Shevchenko <andy.shevchenko@gmail.com>
>>>>>> Date:   Mon Jul 12 21:21:21 2021 +0300
>>>>>>
>>>>>>     ACPI: utils: Fix reference counting in for_each_acpi_dev_match()
>>>>>>
>>>>>> which seems odd, as it worked for me with the acpi changes reverted. It
>>>>>> could be that it _sometimes_ works with that commit, not sure. Adding
>>>>>> relevant folks to the CC.
>>>>>>
>>>>>> I'm going to revert this on top of current master and run with that
>>>>>> and see if it does 10 successful resumes.
>>>>>
>>>>> This does appear to be the culprit. With it reverted on top of current
>>>>> master (and with the block and io_uring changes pulled in too), the
>>>>> kernel survives many resumes without issue.
>>>>
>>>> If I read correctly it seems at one of the local rebase I might have
>>>> missed something.
>>>>
>>>> Can you replace sensor->adev with adev here
>>>>
>>>> https://elixir.bootlin.com/linux/v5.14-rc2/source/drivers/media/pci/intel/ipu3/cio2-bridge.c#L231
>>>>
>>>> and retest?
>>>
>>> Tried that, doesn't work unfortunately. Hung on first resume, just like
>>> current -git with that single line edit.
>>
>> So, does it work along with the acpi_dev_put() converted to be
>>
>>   if (adev)
>>     put_device(&adev->dev);
>>
>> ?
> 
> Building, will test when it's done.

Yep, works with acpi_dev_put() checking for != NULL before doing the
put.

-- 
Jens Axboe