[PATCH] SWDEV-476969 - dm/amdgpu: Fail dm_atomic_check if cursor overlay is required at MAX_SURFACES

[Mohamed, Zaeem]

[AMD Official Use Only - AMD Internal Distribution Only]

Hi Melissa,

Could you share your configuration that experienced this pagefault?

Thanks,
Zaeem

-----Original Message-----
From: Melissa Wen <mwen at igalia.com>
Sent: Monday, November 18, 2024 7:52 AM
To: Limonciello, Mario <Mario.Limonciello at amd.com>; Mohamed, Zaeem <Zaeem.Mohamed at amd.com>; amd-gfx at lists.freedesktop.org
Subject: Re: [PATCH] SWDEV-476969 - dm/amdgpu: Fail dm_atomic_check if cursor overlay is required at MAX_SURFACES




On 14/11/2024 16:04, Mario Limonciello wrote:
> Although it's really useful information for AMD people, the Jira
> shouldn't be in the "title" of the commit message.
>
> "If" we want to get into the habit of including this information for
> display code we should come up with a prescriptive field that goes
> into the commit message during promotion and it should be part of all
> patches in the promotion that have it.
>
> Something like this:
>
> AMD-Jira: SWDEV-476969
>
> Probably need to align that with other stakeholders though before
> starting that way.
>
> On 11/14/2024 08:37, Zaeem Mohamed wrote:
>> [why]
>> Prevent index-out-of-bounds due to requiring cursor overlay when
>> plane_count is MAX_SURFACES.
>>
>> [how]
>> Bounds check on plane_count when requiring overlay cursor.
>>
>
> Any link to failing bugs or anything like that you can include?
Hi Mario,

About this question, these are the related issues:
- Link: https://gitlab.freedesktop.org/drm/amd/-/issues/3693
- Link: https://gitlab.freedesktop.org/drm/amd/-/issues/3594

You can find more details in previous iterations related to this bug:
-
https://lore.kernel.org/amd-gfx/575d66c7-e77d-42ea-acbf-412d6e508a0b@igalia.com/
- https://lore.kernel.org/amd-gfx/20240925154324.348774-1-mwen@igalia.com/
>
>> Co-developed-by: Melissa Wen <mwen at igalia.com>
>> Signed-off-by: Zaeem Mohamed <zaeem.mohamed at amd.com>
>
> You're missing Melisaa's SoB for a co-developed patch.
> IIRC this should fail checkpatch.
I already mentioned before, I don't think I actually co-dev this code, so Zaeem can remove it in the next iteration.

---

BTW, about the implementation:

As I don't have the proper environment, I asked reporters to check this patch up and it doesn't help prevent interface freeze.
It seems to prevent the out of bounds bug but is causing a page fault now:

kernel: BUG: unable to handle page fault for address: 000000000174e354

 From their feedback, I found curious that MAX_SURFACES -> 4 prevents the freeze but not completely solve the problem.
MAX_SURFACES -> 6 solves it, what let me consider that the MAX_SURFACES vs MAX_SURFACE_NUM vs MAX_PLANE mismatch might be related.
I'm going to analyzing the logs but you can find more details here:
https://gitlab.freedesktop.org/drm/amd/-/issues/3693#note_2658994

BR,

Melissa Wen
>
>> ---
>>   amdgpu_dm/amdgpu_dm.c | 10 +++++++++-
>>   1 file changed, 9 insertions(+), 1 deletion(-)
>>
>> diff --git a/amdgpu_dm/amdgpu_dm.c b/amdgpu_dm/amdgpu_dm.c index
>> 97e0a1bbba..964497c613 100644
>> --- a/amdgpu_dm/amdgpu_dm.c
>> +++ b/amdgpu_dm/amdgpu_dm.c
>> @@ -11821,8 +11821,16 @@ static int amdgpu_dm_atomic_check(struct
>> drm_device *dev,
>>             /* Overlay cusor not subject to native cursor
>> restrictions */
>>           dm_new_crtc_state = to_dm_crtc_state(new_crtc_state);
>> -        if (dm_new_crtc_state->cursor_mode ==
>> DM_CURSOR_OVERLAY_MODE)
>> +        if (dm_new_crtc_state->cursor_mode ==
>> +DM_CURSOR_OVERLAY_MODE) {
>> +            if (dc->current_state->stream_status->plane_count >
>> MAX_SURFACES) {
>> +                drm_dbg_driver(crtc->dev,
>> +                       "Can't enable cursor plane with %d planes\n",
>> MAX_SURFACES);
>> +                ret = -EINVAL;
>> +                goto fail;
>> +            }
>> +
>>               continue;
>> +        }
>>             /* Check if rotation or scaling is enabled on DCN401 */
>>           if ((drm_plane_mask(crtc->cursor) &
>> new_crtc_state->plane_mask) &&
>