[RFC] amdgpu: fix compressed buffer config routine waiting time

[RFC] amdgpu: fix compressed buffer config routine waiting time

[Antonio Quartulli]

Starting with commit 592c5b80110d5 ("drm/amd/display: Migrate HUBBUB register access from hwseq to hubbub component.")
the amdgpu driver is reporting the following WARNING in the kernel log
during boot time:

[   15.464476] amdgpu 0000:0c:00.0: [drm] REG_WAIT timeout 1us * 100 tries - dcn31_program_compbuf_size line:141
[   15.464522] ------------[ cut here ]------------
[   15.464523] !(generic_reg_get(hubbub2->base.ctx, hubbub2->regs->DCHUBBUB_COMPBUF_CTRL, hubbub2->shifts->CONFIG_ERROR, hubbub2->masks->CONFIG_ERROR, &compbuf_size_segments) && !compbuf_size_segments)
[   15.464524] WARNING: drivers/gpu/drm/amd/amdgpu/../display/dc/hubbub/dcn31/dcn31_hubbub.c:151 at dcn31_program_compbuf_size+0x20e/0x220 [amdgpu], CPU#2: kworker/2:2/184
[   15.464906] Workqueue: events drm_mode_rmfb_work_fn
[   15.464910] RIP: 0010:dcn31_program_compbuf_size+0x20e/0x220 [amdgpu]
[   15.465200]  <TASK>
[   15.465202]  dcn20_optimize_bandwidth+0x110/0x210 [amdgpu]
[   15.465556]  dc_commit_state_no_check+0x14ff/0x18a0 [amdgpu]
[   15.465925]  dc_commit_streams+0x471/0x640 [amdgpu]
[   15.466252]  amdgpu_dm_atomic_commit_tail+0x903/0x4430 [amdgpu]
[   15.467401]  commit_tail+0x242/0x2e0
[   15.467405]  drm_atomic_helper_commit+0x28b/0x2a0
[   15.467407]  drm_atomic_commit+0xc3/0xf0
[   15.467416]  drm_mode_rmfb_work_fn+0x85/0xb0
[   15.467436]  ret_from_fork_asm+0x1a/0x30
[   15.467441]  </TASK>

After boot this results in unstable video output, specifically after
resuming from screen sleep. The video may not come back at all
or may come up partly messed up.

This problem seems to affect various people on AMD iGPU, as discussed
on the related GitHub issue.

Giuseppe Ranieri digged up the issue and suggested the proposed code
change.

There was also a previous report on the amd-gfx ml (see related link below)
but it got nowhere.

By looking at the offending commit I am not truly able to understand why
it is breaking. Hence I am posting this patch as RFC only.

What I can say is that this fix has been solid on my platform running
7.1.0-rc3 with this patch.

I am on:
Gigabyte Technology Co., Ltd. B850M DS3H/B850M DS3H + AMD Ryzen 9 9950X

Fixes: 592c5b80110d5 ("drm/amd/display: Migrate HUBBUB register access from hwseq to hubbub component.")
Link: https://github.com/CachyOS/linux-cachyos/issues/810
Link: https://lore.kernel.org/amd-gfx/20260316094232.6bb6f0bf@schienar/
Suggested-by: Giuseppe Ranieri
Signed-off-by: Antonio Quartulli <antonio@mandelbit.com>
---
 .../gpu/drm/amd/display/dc/hubbub/dcn31/dcn31_hubbub.c    | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/amd/display/dc/hubbub/dcn31/dcn31_hubbub.c b/drivers/gpu/drm/amd/display/dc/hubbub/dcn31/dcn31_hubbub.c
index 3c298192f3596..9dc1db4524500 100644
--- a/drivers/gpu/drm/amd/display/dc/hubbub/dcn31/dcn31_hubbub.c
+++ b/drivers/gpu/drm/amd/display/dc/hubbub/dcn31/dcn31_hubbub.c
@@ -138,10 +138,10 @@ static void dcn31_program_compbuf_size(struct hubbub *hubbub, unsigned int compb
 
 	if (safe_to_increase || compbuf_size_segments <= hubbub2->compbuf_size_segments) {
 		if (compbuf_size_segments > hubbub2->compbuf_size_segments) {
-			REG_WAIT(DCHUBBUB_DET0_CTRL, DET0_SIZE_CURRENT, hubbub2->det0_size, 1, 100);
-			REG_WAIT(DCHUBBUB_DET1_CTRL, DET1_SIZE_CURRENT, hubbub2->det1_size, 1, 100);
-			REG_WAIT(DCHUBBUB_DET2_CTRL, DET2_SIZE_CURRENT, hubbub2->det2_size, 1, 100);
-			REG_WAIT(DCHUBBUB_DET3_CTRL, DET3_SIZE_CURRENT, hubbub2->det3_size, 1, 100);
+			dcn31_wait_for_det_apply(hubbub, 0);
+			dcn31_wait_for_det_apply(hubbub, 1);
+			dcn31_wait_for_det_apply(hubbub, 2);
+			dcn31_wait_for_det_apply(hubbub, 3);
 		}
 		/* Should never be hit, if it is we have an erroneous hw config*/
 		ASSERT(hubbub2->det0_size + hubbub2->det1_size + hubbub2->det2_size
-- 
2.53.0

Re: [RFC] amdgpu: fix compressed buffer config routine waiting time

[Antonio Quartulli]

On 19/05/2026 17:57, Pillai, Aurabindo wrote:
> AMD General
> 
> Hi Antonio,
> 
> Thanks for the patch. We'll add it to our weekly cycle for testing.

Hi there,

just checking if this RFC had a chance to be considered.
If there is anything I can do on my side, please let me know.

As of now the only option for many users is to patch the kernel 
manually, because otherwise the video is basically unstable.


Thanks a lot!
Best Regards,


-- 
Antonio Quartulli

CEO and Co-Founder
Mandelbit Srl
https://www.mandelbit.com

Re: [RFC] amdgpu: fix compressed buffer config routine waiting time

[Aurabindo Pillai]

On 5/31/26 4:40 PM, Antonio Quartulli wrote:
> [Some people who received this message don't often get email from 
> antonio@mandelbit.com. Learn why this is important at https://aka.ms/ 
> LearnAboutSenderIdentification ]
> 
> On 19/05/2026 17:57, Pillai, Aurabindo wrote:
>> AMD General
>>
>> Hi Antonio,
>>
>> Thanks for the patch. We'll add it to our weekly cycle for testing.
> 
> Hi there,
> 
> just checking if this RFC had a chance to be considered.
> If there is anything I can do on my side, please let me know.
> 
> As of now the only option for many users is to patch the kernel
> manually, because otherwise the video is basically unstable.

Thanks for the patch. We'll add it to this week's testing candidates.

> 
> 
> Thanks a lot!
> Best Regards,
> 
> 
> -- 
> Antonio Quartulli
> 
> CEO and Co-Founder
> Mandelbit Srl
> https://www.mandelbit.com
>