mesa: update to 26.0.2

[onlylunix]

Testing the changes

• I tested the changes in this PR: YES

Local build testing

• I built this PR locally for my native architecture, (x86_64-LIBC)

[emjomi]

Hey, thank you very much! I’m really looking forward to when this will be merged, because the problems with Vulkan on Intel iGPU are really annoying me

[onlylunix]

Unfortunately, Mesa 25.3.5 still contains Vulkan bugs for Intel iGPUs.
I just experienced this in WarThunder:

$ dmesg
...
[16308.490404] i915 0000:00:02.0: Using 39-bit DMA addresses
[16711.735723] i915 0000:00:02.0: [drm] Resetting rcs0 for CS error
[16711.735787] i915 0000:00:02.0: [drm] aces[8769] context reset due to GPU hang
[16712.508228] i915 0000:00:02.0: [drm] GPU HANG: ecode 9:1:98968f5e, in aces [8769]
[16721.204740] i915 0000:00:02.0: [drm] Resetting rcs0 for preemption time out
[16721.204799] i915 0000:00:02.0: [drm] aces[8769] context reset due to GPU hang
[16721.388731] i915 0000:00:02.0: [drm] GPU HANG: ecode 9:1:8ed8fff3, in aces [8769]
[16723.212455] i915 0000:00:02.0: [drm] Resetting rcs0 for CS error
[16723.212520] i915 0000:00:02.0: [drm] aces[8769] context reset due to GPU hang
[16723.490002] i915 0000:00:02.0: [drm] GPU HANG: ecode 9:1:84dffffa, in aces [8769]

@emjomi
Rolling back to mesa 25.1.9 solves the problem.

[onlylunix]

Mesa 26.0.0, iGPU Intel HD630, WarThunder, Vulkan

[ 3234.702248] i915 0000:00:02.0: [drm] Resetting rcs0 for CS error
[ 3234.702309] i915 0000:00:02.0: [drm] aces[2234] context reset due to GPU hang
[ 3235.617039] i915 0000:00:02.0: [drm] GPU HANG: ecode 9:1:84dfbffc, in aces [2234]
[ 3255.368437] i915 0000:00:02.0: [drm] Resetting rcs0 for preemption time out
[ 3255.368509] i915 0000:00:02.0: [drm] aces[2234] context reset due to GPU hang
[ 3255.843385] i915 0000:00:02.0: [drm] GPU HANG: ecode 9:1:84dfbffc, in aces [2234]
[ 3259.020771] i915 0000:00:02.0: [drm] Resetting rcs0 for CS error
[ 3259.020833] i915 0000:00:02.0: [drm] aces[2234] context reset due to GPU hang
[ 3259.428562] i915 0000:00:02.0: [drm] GPU HANG: ecode 9:1:85dffffb, in aces [2234]

[onlylunix]

Mesa 25.3.6 is also buggy

[ 2552.253049] i915 0000:00:02.0: Using 39-bit DMA addresses
[ 3621.622782] Fence expiration time out i915-0000:00:02.0:aces[1704]:27524!
[ 3621.633725] Fence expiration time out i915-0000:00:02.0:aces[1704]:27526!
[ 3641.790828] Fence expiration time out i915-0000:00:02.0:aces[1704]:27584!
[ 3641.949814] i915 0000:00:02.0: [drm] Resetting rcs0 for CS error
[ 3641.949873] i915 0000:00:02.0: [drm] aces[1704] context reset due to GPU hang
[ 3642.703925] i915 0000:00:02.0: [drm] GPU HANG: ecode 9:1:85dffffb, in aces [1704]
[ 3878.536338] i915 0000:00:02.0: [drm] Resetting rcs0 for preemption time out
[ 3878.536398] i915 0000:00:02.0: [drm] aces[1704] context reset due to GPU hang
[ 3879.101272] i915 0000:00:02.0: [drm] GPU HANG: ecode 9:1:87f9ffff, in aces [1704]
[ 3900.516440] Fence expiration time out i915-0000:00:02.0:aces[1704]:10!
[ 3920.550599] Fence expiration time out i915-0000:00:02.0:aces[1704]:12!
[ 3941.414038] Fence expiration time out i915-0000:00:02.0:aces[1704]:3bc!
[ 3961.415682] Fence expiration time out i915-0000:00:02.0:aces[1704]:3be!
[ 3962.248969] i915 0000:00:02.0: [drm] Resetting rcs0 for preemption time out
[ 3962.249040] i915 0000:00:02.0: [drm] aces[1704] context reset due to GPU hang
[ 3962.616638] i915 0000:00:02.0: [drm] GPU HANG: ecode 9:1:85dffffb, in aces [1704]
[ 3983.906479] Fence expiration time out i915-0000:00:02.0:aces[1704]:10!
[ 4003.952087] Fence expiration time out i915-0000:00:02.0:aces[1704]:12!
[ 4024.152938] Fence expiration time out i915-0000:00:02.0:aces[1704]:30!
[ 4024.800527] xfce4-terminal[1695]: segfault at 90 ip 00007f82061a6f56 sp 00007ffe3d918230 error 4 in libgdk-3.so.0.2418.32[86f56,7f820614e000+7c000] likely on CPU 0 (core 0, socket 0)
[ 4024.800539] Code: fb ff 48 89 df e8 1a f9 fa ff 4c 89 ff 48 89 c3 e8 af 82 fb ff 44 8b 5c 24 10 85 c0 74 07 49 89 9f 88 00 00 00 48 8b 5c 24 08 <49> 8b b4 24 90 00 00 00 44 89 5c 24 10 48 89 df e8 d5 8c fb ff 49
[ 4024.805925] xfdesktop[1555]: segfault at 90 ip 00007f39b107ff56 sp 00007ffe341268d0 error 4 in libgdk-3.so.0.2418.32[86f56,7f39b1027000+7c000] likely on CPU 1 (core 1, socket 0)
[ 4024.805935] Code: fb ff 48 89 df e8 1a f9 fa ff 4c 89 ff 48 89 c3 e8 af 82 fb ff 44 8b 5c 24 10 85 c0 74 07 49 89 9f 88 00 00 00 48 8b 5c 24 08 <49> 8b b4 24 90 00 00 00 44 89 5c 24 10 48 89 df e8 d5 8c fb ff 49

The entire 25.3 branch contains errors for Intel iGPUs.

[onlylunix]

Mesa 26.0.1, iGPU Intel HD630, WarThunder, Vulkan

[ 2198.885028] i915 0000:00:02.0: [drm] Resetting rcs0 for CS error
[ 2198.885092] i915 0000:00:02.0: [drm] aces[2533] context reset due to GPU hang
[ 2199.575945] i915 0000:00:02.0: [drm] GPU HANG: ecode 9:1:87cabbf2, in aces [2533]
[ 2200.883985] i915 0000:00:02.0: Using 39-bit DMA addresses
[ 2304.713061] i915 0000:00:02.0: [drm] Resetting rcs0 for preemption time out
[ 2304.713132] i915 0000:00:02.0: [drm] aces[2533] context reset due to GPU hang
[ 2305.156326] i915 0000:00:02.0: [drm] GPU HANG: ecode 9:1:85dffffb, in aces [2533]
[ 2313.865088] i915 0000:00:02.0: [drm] Resetting rcs0 for preemption time out
[ 2313.865159] i915 0000:00:02.0: [drm] aces[2533] context reset due to GPU hang
[ 2314.187844] i915 0000:00:02.0: [drm] GPU HANG: ecode 9:1:87d79ffe, in aces [2533]
[ 2317.896996] i915 0000:00:02.0: [drm] Resetting rcs0 for preemption time out
[ 2317.897055] i915 0000:00:02.0: [drm] aces[2533] context reset due to GPU hang
[ 2318.167580] i915 0000:00:02.0: [drm] GPU HANG: ecode 9:1:87f9ffff, in aces [2533]

upd: The error appears after 5-45 minutes of play.

Rolling back to mesa 25.1.9 solves the problem.

[onlylunix]

@dogknowsnx
something broke

[dkwo]

@onlylunix Did you rebase? Also, from the release 26.0.1 announcement:

This release and 25.3.6 (already out) contain a security fix preventing
out-of-bounds memory access in WebGPU.
They should be deployed to users as quickly as possible.

[dogknowsnx]

@dogknowsnx something broke

Haven't tried building mesa in a while. What dkwo said.

EDIT: Fwiw, no issues so far running v26.0.1 (minimal build) on Intel HD 620 and Iris Xe

[dkwo]

For me, the current version failed to build, but it builds and works fine on aarch64 if I disable _have_nv=yes and _have_amd=yes.

[onlylunix]

@dogknowsnx
You misunderstood me.
Everything builds fine on the local computer.
The problem is this: https://github.com/void-linux/void-packages/actions/runs/22415897358/job/64901314718?pr=58664#step:3:190
@classabbyamp @Duncaen
who is in charge of this?

[classabbyamp]

just rerun it, that's a transient networking issue

[onlylunix]

just rerun it, that's a transient networking issue

@classabbyamp
I can't find the corresponding control in my browser.
I probably don't have permission.

[classabbyamp]

force-push

[classabbyamp]

amend so the commit hash changes

please stop pinging me in every comment, I'm subscribed to the thread

[dkwo]

For me, on native aarch64-glibc (Apple silicon) the _have_nv=yes fails to build. Not sure if it works here because of cross, perhaps someone else can check? Also, does it make sense to set _have_amd=yes on aarch64?

[onlylunix]

Mesa 26.0.2, iGPU Intel HD630, WarThunder, Vulkan

[19835.851576] i915 0000:00:02.0: Using 39-bit DMA addresses
[23758.522949] i915 0000:00:02.0: [drm] Resetting rcs0 for CS error
[23758.523018] i915 0000:00:02.0: [drm] aces[15737] context reset due to GPU hang
[23759.378398] i915 0000:00:02.0: [drm] GPU HANG: ecode 9:1:87dbfdfa, in aces [15737]
[24001.792466] i915 0000:00:02.0: [drm] Resetting rcs0 for preemption time out
[24001.792537] i915 0000:00:02.0: [drm] aces[15737] context reset due to GPU hang
[24002.351608] i915 0000:00:02.0: [drm] GPU HANG: ecode 9:1:85dffffb, in aces [15737]
[24010.496323] i915 0000:00:02.0: [drm] Resetting rcs0 for preemption time out
[24010.496383] i915 0000:00:02.0: [drm] aces[15737] context reset due to GPU hang
[24010.831283] i915 0000:00:02.0: [drm] GPU HANG: ecode 9:1:8ed9fff2, in aces [15737]
[24032.462372] Fence expiration time out i915-0000:00:02.0:aces[15737]:10!
[24039.488290] i915 0000:00:02.0: [drm] Resetting rcs0 for preemption time out
[24039.488349] i915 0000:00:02.0: [drm] aces[15737] context reset due to GPU hang
[24039.747714] i915 0000:00:02.0: [drm] GPU HANG: ecode 9:1:85ddfffa, in aces [15737]

Rolling back to mesa 25.1.9 solves the problem.

[classabbyamp]

have you reported this to the mesa devs? seems futile to just update and hope it's fixed

[onlylunix]

There are a lot of similar messages here:
https://gitlab.freedesktop.org/mesa/mesa/-/issues/?sort=updated_desc&state=opened&label_name%5B%5D=ANV
The bad thing is that the error appears after an hour of play...
If the bug had appeared immediately, then git bisect could have been used.

seems futile to just update and hope it's fixed

Besides Intel, there are other video cards.