[PATCH] Revert "drm/amd/display: limit clear_update_flags to dcn32 and above"

[Melissa Wen]

On 17-07-2025 17:57, Matthew Schwartz wrote:
>> On Jul 17, 2025, at 9:58 AM, Limonciello, Mario <Mario.Limonciello at amd.com> wrote:
>> 
>> On 7/17/25 11:55 AM, Melissa Wen wrote:
>>> On 17-07-2025 13:39, Limonciello, Mario wrote:
>>>> On 7/17/25 9:36 AM, Melissa Wen wrote:
>>>>> This reverts commit e1bd5e0bb4ca0d633ad698abd3658f8265009b81.
>>>>> 
>>>>> The commit causes a regression in Steam Deck (DCN 3.01), reintroducing a
>>>>> functional issue reported in [1] that was fixed by calling the
>>>>> clear_update_flags() from commit 7671f62c10f2a.
>>>>> 
>>>>> On Steam Deck, we use multiple hw plane color caps and up to two overlay
>>>>> planes with dynamic pipe split policy. I.e. with 1 primary + 1 overlay,
>>>>> the driver split planes into two (with 4 pipes), but with 1 primary + 2
>>>>> overlays, we don't have enough pipe for splitting. Glitches appear in
>>>>> this pipe-split transition of 1-2 overlay planes, if the driver doesn't
>>>>> clear update flags.
>>>>> 
>>>>> Besides that, the issue the commit e1bd5e0bb4ca tries to address [2]
>>>>> isn't functional.
>>>>> 
>>>>> Link: https://gitlab.freedesktop.org/drm/amd/-/issues/3441 [1]
>>>>> Link: https://gitlab.freedesktop.org/drm/amd/-/issues/4129 [2]
>>>>> Signed-off-by: Melissa Wen <mwen at igalia.com>
>>>> 
>>>> Can you please double check the use case that prompted
>>>> e1bd5e0bb4ca0d633ad698abd3658f8265009b81 in the first place?  IE With
>>>> this revert you proposed do you see a traceback on unplug of external
>>>> display?
>>> 
>>> Hi Mario,
>>> 
>>> I don't see either traceback or REG_WAIT timeout when unplugging an
>>> external HDMI-connected display.
>>> I tried with the Deck Dock and USB-C to HDMI adapter. Not sure if there
>>> is a very specific scenario that triggers this that I didn't cover.
>> 
>> No; it's quite a general problem and easy to reproduce.  This makes me 
>> wonder if there should be an exception carved out for Steam Deck instead.
> 
> I just tested DCN 3.1.2 (Z2 Go) and DCN 3.1.4 (Legion Go/ROG Ally) and
> I was not able to reproduce the issue so a Steam Deck carveout might make
> sense here.

What issue are you trying to reproduce?
Glitches on Hades II + MangoHud, or traceback/REG_WAIT timeout?

And with what kernel version?

Melissa

> 
> Matt
> 
>> 
>>> 
>>> The only message on dmesg from this action is:
>>> amdgpu 0000:04:00.0: amdgpu: pp_od_clk_voltage is not accessible if
>>> power_dpm_force_performance_level is not in manual mode!
>>> 
>>> Moreover, if I move to the Desktop mode (KDE), there is no message.
>>> 
>>> Melissa
>>> 
>>>> 
>>>>> ---
>>>>>   drivers/gpu/drm/amd/display/dc/core/dc.c | 5 +++--
>>>>>   1 file changed, 3 insertions(+), 2 deletions(-)
>>>>> 
>>>>> diff --git a/drivers/gpu/drm/amd/display/dc/core/dc.c b/drivers/gpu/drm/amd/display/dc/core/dc.c
>>>>> index c31f7f8e409f..7c48f72e5917 100644
>>>>> --- a/drivers/gpu/drm/amd/display/dc/core/dc.c
>>>>> +++ b/drivers/gpu/drm/amd/display/dc/core/dc.c
>>>>> @@ -5443,7 +5443,8 @@ bool dc_update_planes_and_stream(struct dc *dc,
>>>>>    else
>>>>>    ret = update_planes_and_stream_v2(dc, srf_updates,
>>>>>    surface_count, stream, stream_update);
>>>>> - if (ret && dc->ctx->dce_version >= DCN_VERSION_3_2)
>>>>> +
>>>>> + if (ret)
>>>>>    clear_update_flags(srf_updates, surface_count, stream);
>>>>> 
>>>>>    return ret;
>>>>> @@ -5474,7 +5475,7 @@ void dc_commit_updates_for_stream(struct dc *dc,
>>>>>    ret = update_planes_and_stream_v1(dc, srf_updates, surface_count, stream,
>>>>>    stream_update, state);
>>>>> 
>>>>> - if (ret && dc->ctx->dce_version >= DCN_VERSION_3_2)
>>>>> + if (ret)
>>>>>    clear_update_flags(srf_updates, surface_count, stream);
>>>>>   }