Bug 2417631

Summary: vkcube --validate crashes on Rawhide
Product: [Fedora] Fedora Reporter: Adam Williamson (Red Hat non-Fedora) <awilliam>
Component: mesaAssignee: Adam Jackson <ajax>
Status: CLOSED RAWHIDE QA Contact: Fedora Extras Quality Assurance <extras-qa>
Severity: high Docs Contact:
Priority: unspecified    
Version: rawhideCC: airlied, ajanulgu, ajax, asrivats, igor.raits, jexposit, j, lyude, marcandre.lureau, mpenttil, pemensik, philip.wyett, psklenar, rstrode, suraj.ghimire7, tstellar
Target Milestone: ---Keywords: Regression
Target Release: ---   
Hardware: All   
OS: Linux   
Whiteboard:
Fixed In Version: Doc Type: ---
Doc Text:
Story Points: ---
Clone Of: Environment:
Last Closed: 2025-12-01 18:20:08 UTC Type: Bug
Regression: --- Mount Type: ---
Documentation: --- CRM:
Verified Versions: Category: ---
oVirt Team: --- RHEL 7.3 requirements from Atomic Host:
Cloudforms Team: --- Target Upstream Version:
Embargoed:
Attachments:
Description Flags
extremely long backtrace none

Description Adam Williamson (Red Hat non-Fedora) 2025-11-28 00:29:37 UTC 
As a consequence of the recent F43 mesa update kerfuffle - https://discussion.fedoraproject.org/t/is-most-recent-mesa-push-to-stable-mesa-25-2-7-2-fc43-reasonable/173585 - I wrote a new test to run `vkcube --validate`. As I hoped, on F43, even in a VM with llvmpipe, that crashes with the known-broken mesa (mesa-25.2.7-2.fc43), but passes with the previous build (mesa-25.2.7-1.fc43) and the fixed build (mesa-25.2.7-3.fc43).

However, an interesting plot twist: it also crashes on Rawhide, with mesa-25.3.0-4.fc44 , which does not (AFAICT) have any version of airlied's backported fix that caused the trouble on F43. It also crashes on Rawhide if I downgrade mesa to mesa-25.2.6-5.fc44 . It also still crashes if I downgrade vulkan-validation-layers and vulkan-validation-tools to the previous f43 builds, 1.4.321.0-3.fc43 and 1.4.321.0-4.fc43 .

So...that seems bad. The crash is caused by a segfault, it seems:

Nov 27 16:21:05 fedora kernel: vkcube[5803]: segfault at 7ffd71c51ff8 ip 00007f470365f1e6 sp 00007ffd71c52000 error 6 in libVkLayer_khronos_validation.so[65f1e6,7f4703000000+a4d000] likely on CPU 2 (core 0, socket 2)
Nov 27 16:21:05 fedora kernel: Code: 06 89 da c1 ea 0c 31 d0 31 c3 89 d8 83 e0 3f 89 85 60 ff ff ff 83 e3 3f 48 89 d8 48 c1 e0 06 4c 8d ac 06 40 0e 00 00 4c 89 ef <e8> e5 21 9a ff 83 f8 0b 74 f3 83 f8 23 74 37 85 c0 75 51 8b b5 60

The backtrace is *huge* and has like 50,000 frames in it, not sure what's going on there. I'll attach it.

This should be easily reproducible by installing a current Fedora Workstation Rawhide in a virt-manager or Boxes VM, install vulkan-tools and vulkan-validation-layers , and running `vkcube --validate`.
Comment 1 Adam Williamson (Red Hat non-Fedora) 2025-11-28 00:34:56 UTC 
Created attachment 2116454 [details]
extremely long backtrace
Comment 2 Adam Williamson (Red Hat non-Fedora) 2025-11-28 01:02:19 UTC 
Downgrading kernel to 6.17 doesn't help. Neither does downgrading glibc to 2.42-4.fc43 (which required a lot of dep hell solving). Bit baffled.
Comment 3 Adam Williamson (Red Hat non-Fedora) 2025-11-28 01:49:53 UTC 
Ah, it looks like the mesa I downgraded to wasn't old enough; Dave actually did an earlier backport for Rawhide, in https://src.fedoraproject.org/rpms/mesa/c/c69e2123eea0cd311d3b7715ec74acefd36418f9?branch=rawhide , and the build I tested came after that commit. He's updating Rawhide to have the latest version with the layer fix now, so hopefully that will fix this.
Comment 4 Adam Williamson (Red Hat non-Fedora) 2025-11-28 01:52:24 UTC 
Aha, yeah, downgrading all the way to 25.2.5-1.fc44 makes the crash go away. Mystery solved.
Comment 5 Petr Sklenar 2025-11-28 11:34:19 UTC 
FYI 
I could see segfault running at system without any GPU, like cloud
https://bugzilla.redhat.com/show_bug.cgi?id=2416951
But its segfaulted for longer time, it seems like not a regression.


With my personal machine with GPU it was segfaulting only with "problematic MESA", now it looks fine.
Comment 6 Petr Sklenar 2025-11-28 12:25:32 UTC 
@Adam
I thing this is good reproducer of the original mesa issue, with -2 version at headless machines:

by xwfb-run - Run X11 clients on a dedicated Xwayland server headless

SETUP:
koji download-build --arch x86_64 mesa-25.2.7-2.fc43
dnf install ./mesa-vulkan-drivers-25.2.7-2.fc43.x86_64.rpm
dnf install ./mesa-vulkan-drivers-25.2.7-2.fc43.x86_64.rpm ./mesa-filesystem-25.2.7-2.fc43.x86_64.rpm
dnf install ./mesa-dri-drivers* ./mesa-libi* -y
dnf install vulkan-loader vulkan-validation-layers vulkan-tools
dnf install mutter xwayland-run


REPRODUCER:
xwfb-run -c mutter -- vkcube --validate

# coredumpctl list 
mesa-vulkan-drivers-25.2.7-2.fc43.x86_64
segfault

mesa-vulkan-drivers-25.2.7-3.fc43.x86_64
no segfault

---
Still I can see segfault of 'vkcube --validate', in terminal + defining DISPLAY=:0 (not in xwfb-run )
Comment 7 Adam Williamson (Red Hat non-Fedora) 2025-11-28 16:46:39 UTC 
Thanks! That might certainly help the mesa maintainers add this as a package-level test. Maintainers, are you interested in doing that?
Comment 8 Adam Williamson (Red Hat non-Fedora) 2025-12-01 18:20:08 UTC 
The updated mesa for Rawhide definitely did solve this, so closing.