Re: v4l2-async: regression due to endpoint matching

["Linux regression tracking (Thorsten Leemhuis)" <regressions@leemhuis.info>]

On 22.05.23 13:59, Sakari Ailus wrote:
> On Mon, May 22, 2023 at 12:53:52PM +0200, Linux regression tracking (Thorsten Leemhuis) wrote:
>> On 28.04.23 09:16, Sakari Ailus wrote:
>>> On Fri, Apr 28, 2023 at 08:43:21AM +0200, Jacopo Mondi wrote:
>>>> On Fri, Apr 28, 2023 at 08:33:30AM +0200, Alexander Stein wrote:
>>>>> Am Freitag, 28. April 2023, 08:31:54 CEST schrieb Jacopo Mondi:
>>>>>> On Fri, Apr 28, 2023 at 08:24:22AM +0200, Alexander Stein wrote:
>>>>>>> Am Donnerstag, 27. April 2023, 18:01:38 CEST schrieb Jacopo Mondi:
>>>>>>>> On Thu, Apr 27, 2023 at 04:40:46PM +0200, Alexander Stein wrote:
>>>>>>>>> I have a setup on my TQMa6x (imx6q-mba6a.dts) with a tc358743 attached
>>>>>>>>> to
>>>>>>>>> the MIPI CSI input.
>>>>>>>>> I noticed that since commit 1f391df44607 ("media: v4l2-async: Use
>>>>>>>>> endpoints in __v4l2_async_nf_add_fwnode_remote()") the async subdevice
>>>>>>>>> probing does not work anymore. If I revert that, it is working again,
>>>>>>>>> even on next-20230425.
>>>>>>>>
>>>>>>>> A similar issue has been discussed at
>>>>>>>> https://www.spinics.net/lists/linux-media/msg223351.html
>>>>>>>>
>>>>>>>> Unfortunately there was no conclusion as far as I can tell if not that
>>>>>>>> imx6 is now broken
>>>>>>>
>>>>>>> Thanks for the link, seems like a non-trivial thing :(
>>>>>>>
>>>>>>> From a glimpse, this series seems to deal with multiple async subdevs:
>>>>>>> https://lore.kernel.org/all/20230330115853.1628216-1-sakari.ailus@linux.in
>>>>>>> tel.com/
>>>>>>>
>>>>>>> So imx-media-csi should be adjusted as well, no?
>>>>>>
>>>>>> It would really be helpful if you can give that series a spin on imx6
>>>>>> if you already have a test setup.
>>>>>
>>>>> I tried, but it failed to apply on my current development tree. What base does
>>>>> this series apply to? Is there also a repository available I can fetch from?
>>>>
>>>> Sakari could tell, for me it applied on v6.3-rc2 but I recall I had to
>>>> manually fix a few things.
>>>
>>> Don't try v1, it won't work. I missed some object relation changes in the
>>> linked lists. I'll post v2, hopefully some time next week, to address these
>>> issues.
>>
>> Hi, Thorsten here, the Linux kernel's regression tracker.
>>
>> I see that v2[1] got a lot of ACKs, but is not even yet in next. And
>> it's a lot of patches, so maybe too much for backporting to stable
>> kernels. Which leads to the question: Will the regression this thread is
>> about (introduced in 5.19 afaics) ever be fixed in v6.1?
>> Normally/Ideally it should be.
> 
> We'll need v3 (at least), a problem that's not trivial to fix was
> identified with v2. There patches aren't really fixes either: it's new
> functionality that wasn't there previously. I.MX6 just happened to work due
> to missing checks in the V4L2 async framework, what it needs was never
> supported (without this set).
> 
> Dropping endpoint matching will break adv748x driver that relies on it.
> 
> So I'd expect i.MX6 to work again once we have this set in, but I wouldn't
> try to backport the set.

Thx for the update. Makes me wonder if reverting the culprit[1] is an
option. Assuming the problem still happens. Alexander, is that the case?

Ciao, Thorsten

[1] 1f391df4460 ("media: v4l2-async: Use endpoints in
__v4l2_async_nf_add_fwnode_remote()") (v5.19-rc1; authored by Laurent,
commited by Mauro (both now CCed))