Re: [REGRESSION] Please revert 1191285ecb42 from 7.1.y

Sasha Levin <[email protected]>
Newsgroups org.freedesktop.lists.amd-gfx,dev.linux.lists.regressions,org.kernel.vger.stable
Message-ID <anzx-jMt6_6OTdTI@laps>
On Tue, Aug 11, 2026 at 01:25:14PM -0400, Alex Deucher wrote:
>On Tue, Aug 11, 2026 at 11:25 AM Sergio Callegari
><[email protected]> wrote:
>>
>> On 11/08/2026 10:08, Greg KH wrote:
>> > We aim to be "bug compatible" with Linus's branch, otherwise the
>> > "pressure" to fix those bugs goes down.  But I defer to the maintainers
>> > of the code involved if it is being worked on, and they want it
>> > reverted, great, we can do that.  Otherwise we prefer to leave it as-is.
>>
>> I understand. Sorry for asking about things that might be obvious to
>> most people here. My expectation was that only a selected subset of the
>> patches on the Linus' tree were to enter in stable, so do we really need
>> to have stable "bug compatible" wrt this bug? The original behavior was
>> not free of issues, but they were rather minor (a reporting issue). Now
>> we have this very bad behavior where the display remains dark.
>>
>> To make the matter worse, on systems that have encrypted root the screen
>> remains dark also on that password prompt (observed with plymouth).
>>
>> In any case, I managed testing 7.2.0RC7 and it is fine again.
>>
>> Looks like the Linus' tree has more changes wrt backlight, and the
>> interesting ones are probably those in the cumulative fix
>> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=973fd9493ec21f12e760fbbedeb7b0b2317c02f4.
>>
>
>Thanks for testing.  It sounds like we should revert the change from
>stable.  There must be some dependency we are missing in stable.

Hmm...

In stable kernels, convert_custom_brightness() still scales the user value by
(max - min), but f1b5d8f9cc54 ("drm/amd/display: Fix backlight max_brightness
to match exported range") widened the exported range to max. Past ~(max - min),
scale_input_to_fw() exceeds AMDGPU_MAX_BL_LEVEL and
convert_brightness_from_user() returns >0xFFFF, which wraps in
dce_driver_set_backlight(). With the defaults the cliff lands on 64533, which
is what Sergio sees.

Upstream only hides it: mod_power_set_backlight_percent() clamps at 100%.

Does this make sense?

-- 
Thanks,
Sasha
lmpx.com only provides a reader for public news (NNTP) servers. It is not affiliated with the servers or forums shown here and is not responsible for the content of articles, which is written by their respective authors.