> Please amend the test plan with some form of testing in WSL
environments, otherwise this SRU has no meaning if we only care about
non-WSL.
Hi Andreas,
Thanks for pushing on this, we now have concrete WSL test coverage and
I've updated the Test Plan section of the bug description accordingly
(item 4).
Summary of what we tested, before/after the fix, on real Windows/WSL2:
- WSL2 (Ubuntu 26.04.1, kernel 6.18.40.1-microsoft-standard-WSL2) with an
NVIDIA GeForce MX130 present (unsupported by ROCm, so hsa_init() fails
the same way it would with no GPU at all).
Before (7.1.1-0ubuntu1): rocminfo fails on hsa_init() with
HSA_STATUS_ERROR_OUT_OF_RESOURCES, and rocm_agent_enumerator then hangs
for about a minute before giving up with "Timeout querying rocminfo." -
this is exactly the bug being fixed.
After (7.2.4-1~exp1): rocminfo prints "WSL environment detected." /
"hsa_init Failed, possibly no supported GPU devices" and exits
immediately (exit 1); rocm_agent_enumerator returns immediately (exit
0), no hang.
- We also re-confirmed the no-GPU WSL VM scenario: no regression, both
versions exit promptly.
We asked AMD about getting a WSL+AMD-GPU rig into the loop; they don't
consider that combination a blocker for this SRU and aren't committing
further test time beyond best-effort, since the fix triggers on any
hsa_init() failure (no GPU or unsupported vendor alike) rather than on
AMD hardware specifically, so the NVIDIA-GPU and no-GPU cases above
exercise the exact code path the fix changes.
Let me know if this addresses the concern or if you'd like anything else
added before moving this back to In Progress.
** Description changed:
[ Impact ]
* rocminfo 7.2.4 carries one upstream fix: an early-exit check in WSL
environments when `hsa_init()` fails due to no supported GPU being
present. Without this fix, `rocm_agent_enumerator` would silently keep
retrying `rocminfo` for up to one minute before giving up, causing
unnecessary startup delays for any caller (e.g. ROCm runtime detection
scripts) on Windows Subsystem for Linux systems without a supported GPU.
The fix adds a five-line guard that prints a clear error message and
returns immediately when in WSL and `hsa_init` returns a non-success
status.
* Packaging changes in this upload (no user-visible behaviour change):
- d/watch, d/copyright: track rocminfo from the ROCm/rocm-systems
monorepo release assets instead of the individual upstream repository.
- d/copyright: add dedicated Expat stanza for rocm_agent_enumerator
(SPDX: MIT), which has a distinct licence from the main Illinois-NCSA
body. Bump upstream copyright year range to 2014-2025.
[ Test Plan ]
1. Build:
- sbuild or dpkg-buildpackage succeeds.
- dpkg --compare-versions 7.2.4-1~exp1 gt 7.1.1-0ubuntu1 confirms
the new version is greater.
- No debian/*.symbols file exists for this package (binaries only,
no shared library); dpkg-gensymbols check is not applicable.
2. Installability:
- apt install rocminfo succeeds.
- Confirm reverse dependencies remain installable without rebuild.
3. Autopkgtest:
- Run autopkgtest suite (rocminfo, enumerator).
- All tests pass. Output:
autopkgtest [13:47:48]: test rocminfo: [-----------------------
Running rocminfo and expecting GPU information
autopkgtest [13:47:48]: test rocminfo: -----------------------]
autopkgtest [13:47:49]: test rocminfo: - - - - - - - - - - results - -
- - - - - - - -
rocminfo PASS (superficial)
autopkgtest [13:47:50]: test enumerator: preparing testbed
autopkgtest [13:47:51]: test enumerator: [-----------------------
autopkgtest [13:47:52]: test enumerator: -----------------------]
autopkgtest [13:47:52]: test enumerator: - - - - - - - - - - results -
- - - - - - - - -
enumerator PASS (superficial)
autopkgtest [13:47:53]: @@@@@@@@@@@@@@@@@@@@ summary
rocminfo PASS (superficial)
enumerator PASS (superficial)
2026-05-21 13:47:54 - Autopkg tests ended for rocminfo in
ppa:igorluppi/rocminfo-7.2.3.
Tests took: 0h 9m 30s. Logs saved in ./rocminfo_20260521_133824.log
+ 4. WSL testing (the environment this fix specifically targets):
+
+ Manually tested on real Windows/WSL2, before and after the fix, on a
+ system with an unsupported (non-AMD) GPU present, to exercise the
+ exact `hsa_init()`-failure path the fix changes:
+
+ - Host: WSL2, kernel 6.18.40.1-microsoft-standard-WSL2, guest
+ Ubuntu 26.04.1 (resolute), GPU: NVIDIA GeForce MX130 (no ROCm
+ support, so hsa_init() fails as it would with no GPU at all).
+
+ - Before (rocminfo 7.1.1-0ubuntu1, archive):
+ $ rocminfo
+ WSL environment detected.
+ hsa api call failure at: ./rocminfo.cc:1324
+ Call returned HSA_STATUS_ERROR_OUT_OF_RESOURCES: ...
+ $ rocm_agent_enumerator
+ Timeout querying rocminfo. Are you compiling with more than 254
threads?
+ (rocm_agent_enumerator hangs for ~1 minute before giving up -
+ this is the bug being fixed.)
+
+ - After (rocminfo 7.2.4-1~exp1, PPA):
+ $ rocminfo
+ WSL environment detected.
+ hsa_init Failed, possibly no supported GPU devices
+ $ echo $?
+ 1
+ $ rocm_agent_enumerator
+ $ echo $?
+ 0
+ (both commands return immediately, no hang.)
+
+ - Also tested in a WSL environment with no GPU at all: same
+ no-regression result, 7.1.1 and 7.2.4 both exit promptly.
+
+ - AMD has been consulted; they do not consider dedicated WSL+AMD-GPU
+ testing a hard requirement for this SRU and are not committing
+ further test coverage beyond best-effort (an AMD-GPU WSL test was
+ attempted on available hardware but that combination was not
+ reachable). Since the fix triggers on any `hsa_init()` failure
+ (absent GPU or unsupported vendor alike), not on AMD hardware
+ specifically, the no-GPU and NVIDIA-GPU scenarios above exercise
+ the exact code path the fix changes and demonstrate the fixed
+ behaviour with no regression versus 7.1.1.
+
[ Where problems could occur ]
* The only code change is the WSL early-exit path, which is guarded by an
existing `wsl_env` flag and only reached when `hsa_init()` fails. On
non-WSL systems and on WSL systems with a supported GPU the new code is
never executed. Risk is minimal and limited to WSL environments with no
- supported GPU.
+ supported GPU; this path has now been manually exercised and confirmed
+ fixed (see Test Plan item 4), so residual risk is limited to WSL/AMD-GPU
+ combinations that could not be directly tested.
[ Other Info ]
* No shared library is shipped; ABI/symbols check is not applicable.
* This update is part of the coordinated ROCm 7.2.4 stack release.
* PPA: https://launchpad.net/~igorluppi/+archive/ubuntu/rocminfo-7.2.3
* Upstream version comparison:
https://github.com/ROCm/rocminfo/compare/rocm-7.1.1...rocm-7.2.4
* target: resolute
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2153792
Title:
SRU: New upstream version 7.2.4
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/rocminfo/+bug/2153792/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
