gfx1201 gpu_navi4x runners have broken ROCm environments: rocminfo segfaults

[benvanik]

gfx1201 gpu_navi4x runners have broken ROCm environments: rocminfo segfaults

Summary

Several recent CI CMake Linux gfx1201 package-test jobs on the gpu_navi4x
runner pool show the same runner-health failure: packaged rocminfo resolves
from the installed public dependency package, then immediately segfaults.

This is not a normal test failure. When the workflow temporarily treated
rocminfo as a warning and continued, the installed GPU suite produced a large
cascade of process segfaults, usually:

42% tests passed, 38 tests failed out of 66
Errors while running CTest

PR #52 changes the workflow to use rocminfo as a hard ROCm environment
canary again. While gfx1201 is marked experimental, the job can report the
broken runner and skip the installed suite instead of generating the cascade.

Impacted Runners

Sampled the last 80 CI CMake Linux workflow runs on 2026-06-09.

Matching rocminfo segfaults were observed on:

Runner label	Runner name	Matching jobs
gpu_navi4x	CS-RORDMZ-DT145	7
gpu_navi4x	CS-RORDMZ-DT147	4

No matching rocminfo segfault was found on CS-RORDMZ-DT143 in this sampled
window. The latest PR #52 gfx1201 job on CS-RORDMZ-DT143 succeeded:

• https://github.com/ROCm/hrx-system/actions/runs/27224062207/job/80386770574
• rocminfo printed ROCk/HSA information.
• Installed GPU tests passed: 100% tests passed, 0 tests failed out of 66.

Failure Signature

Pre-PR #52 warning-mode example:

rocminfo path: /home/ubuntu/actions-runner/_work/hrx-system/hrx-system/build/linux-gpu/install/public-deps/bin/rocminfo
/home/ubuntu/actions-runner/_work/_temp/9ead8cff-499a-488f-a48b-fa504e5798ed.sh: line 4: 382707 Segmentation fault      (core dumped) rocminfo
Warning: rocminfo failed; continuing to installed GPU tests
42% tests passed, 38 tests failed out of 66
Errors while running CTest

PR #52 canary-mode example:

rocminfo path: /home/ubuntu/actions-runner/_work/hrx-system/hrx-system/build/linux-gpu/install/public-deps/bin/rocminfo
.github/scripts/check_rocm_environment.sh: line 19: 347376 Segmentation fault      (core dumped) rocminfo
Error: ROCm environment on this runner is broken; rocminfo failed before GPU tests.

Matching Jobs

Started UTC	Workflow run	Job	Result	Runner	Symptom
2026-06-09 06:36:42	27188339883	80262712962	failure	CS-RORDMZ-DT145	rocminfo segfaulted before installed tests
2026-06-09 06:42:45	27188655682	80263583859	failure	CS-RORDMZ-DT147	rocminfo segfaulted before installed tests
2026-06-09 15:43:43	27217788198	80364536623	failure	CS-RORDMZ-DT145	rocminfo segfaulted; workflow continued; 38/66 installed tests failed
2026-06-09 16:39:42	27220546268	80376180903	failure	CS-RORDMZ-DT145	rocminfo segfaulted; workflow continued; 38/66 installed tests failed
2026-06-09 16:54:59	27220546396	80379157442	failure	CS-RORDMZ-DT145	rocminfo segfaulted; workflow continued; 38/66 installed tests failed
2026-06-09 17:00:14	27222201440	80380071263	failure	CS-RORDMZ-DT145	rocminfo segfaulted; workflow continued; 38/66 installed tests failed
2026-06-09 17:04:45	27222459403	80381012028	failure	CS-RORDMZ-DT147	rocminfo segfaulted; workflow continued; 38/66 installed tests failed
2026-06-09 17:07:41	27222637646	80381605883	failure	CS-RORDMZ-DT145	rocminfo segfaulted; workflow continued; 38/66 installed tests failed
2026-06-09 17:18:41	27223228760	80383743191	failure	CS-RORDMZ-DT145	rocminfo segfaulted; workflow continued; 38/66 installed tests failed
2026-06-09 17:27:34	27223713966	80385468837	failure	CS-RORDMZ-DT147	rocminfo segfaulted; workflow continued; 38/66 installed tests failed
2026-06-09 17:29:49	27223853693	80385861126	success due to experimental lane	CS-RORDMZ-DT147	PR #52 canary detected rocminfo segfault and skipped installed tests

Nearby Non-Matches

These failed gfx1201 jobs in the sampled window did not match the rocminfo
segfault pattern and look like real test/application failures instead of this
runner-health issue:

• https://github.com/ROCm/hrx-system/actions/runs/27155546734/job/80157669173
• https://github.com/ROCm/hrx-system/actions/runs/27181626862/job/80242197822
• https://github.com/ROCm/hrx-system/actions/runs/27181831300/job/80242623225

Request

Please inspect or recycle the affected gpu_navi4x runners:

• CS-RORDMZ-DT145
• CS-RORDMZ-DT147

The first health check should be running the packaged/public-deps rocminfo
path used by CI:

/home/ubuntu/actions-runner/_work/hrx-system/hrx-system/build/linux-gpu/install/public-deps/bin/rocminfo