Skip to content

[DO NOT MERGE] docs(amdsmi): changelog entry for the process list fix (exp C2) - #14707

Closed
adam360x wants to merge 1 commit into
users/adam360x/ccr-exp-C-basefrom
users/adam360x/ccr-exp-C-head
Closed

adam360x wants to merge 1 commit into
users/adam360x/ccr-exp-C-basefrom
users/adam360x/ccr-exp-C-head

Conversation

@adam360x

Copy link
Copy Markdown
Contributor

Draft experiment for review tooling. Please do not review or merge.

Targets a throwaway branch and will be closed within a day.

Signed-off-by: Adam Pryor <Adam.pryor@amd.com>

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🟡 Changes recommended

The changelog claims the cross-GPU crash is resolved, but shared process-cache access still uses different per-GPU locks, allowing concurrent queries to invalidate map nodes.

1 open issue
What changed in this PR

This draft tooling experiment adds an AMD SMI changelog entry describing a concurrent GPU process-list crash fix. It changes documentation only, not implementation or tests.

Changes:

  • Adds a resolved-issue summary for amdsmi_get_gpu_process_list().
  • Describes cache synchronization and regression-test coverage.
File Description
projects/​amdsmi/​CHANGELOG.md Adds the process-list crash resolution entry.

Review effort: Balanced

Comment on lines +174 to +175
- **Fixed a crash in `amdsmi_get_gpu_process_list()` when several threads query GPUs at once (AILITOOLS-1234)**.
- `AMDSmiGPUDevice::get_compute_process_list_impl()` now holds `ComputeProcessCache::mutex_` while it copies `process_info_cache_map`, so callers no longer utilize a freed map node and abort with `SIGSEGV`; a regression guard test covers the race.
@adam360x adam360x closed this Oct 30, 2026
@adam360x
adam360x deleted the users/adam360x/ccr-exp-C-head branch October 30, 2026 20:00
adam360x added a commit that referenced this pull request Nov 3, 2026
…14718)

## Motivation

Copilot code review runs on every `develop` PR but never sees the
amd-smi review agents (maintainer, skeptic, security, tests and the
rest): it reads instruction files only from the repository root and
cannot run custom agents, so `projects/amdsmi/.github/agents/` is
ignored.

## Technical Details

**Copilot instructions**
(`.github/instructions/amdsmi-review*.instructions.md`):
- One symlink per review agent (the orchestrator and all 10 lenses) to
its file in `projects/amdsmi/.github/agents/`, so Copilot code review
always reads the current agents. Nothing is copied.

**Review agents**
(`projects/amdsmi/.github/agents/amdsmi-review*.agent.md`):
- Add `applyTo: "projects/amdsmi/**"`, so only PRs that change amd-smi
files get the agents, and `excludeAgent: "cloud-agent"`, which keeps
them out of the Copilot cloud agent. Copilot reads both through the
symlinks; VS Code doesn't use them for agent files.

**Code owners** (`.github/CODEOWNERS`):
- Give the symlinks to @ROCm/smi-reviewers, like
`/.github/workflows/amdsmi-*.yml`.

**Side effect**: VS Code also reads root `.github/instructions/`. With
the whole repository open, it attaches the agents to agent-mode edits of
files under `projects/amdsmi/`.

## Issue Tracking

JIRA ID: TBD

## Test Plan

- Opened draft PRs into throwaway branches, each with the same
CHANGELOG-only diff and the review agents in a different place, and
requested a Copilot review on each.
- Load test: appended a test-only "flag the word X" rule to the end of
each of the 11 agents (a different word per agent) and changed one
CHANGELOG line that contains all 11 words.
- Ran this branch against a true CHANGELOG entry rewritten with internal
names, and an entry claiming a fix that `develop` doesn't have.
- Root pre-commit hooks; GitHub CODEOWNERS errors API.

## Test Result

- Copilot applied the agents from root `.github/instructions/` files,
copied (#14719) or symlinked (#14750, #14798): all 11 test words each
time, so it reads every agent to the end. It ignored them in
`projects/amdsmi/REVIEW.md` (#14703, #14708), behind a root file
pointing into `projects/amdsmi/` (#14702, #14707), and behind a root
review skill pointing there (#14751, #14752).
- This branch: Copilot flagged a CHANGELOG entry that swapped the
user-visible result for internal details (#14799) and an entry claiming
a fix that `develop` doesn't have (#14800).
- Pre-commit: pass. CODEOWNERS: no new errors (the only error, an
unknown owner on the `rocm-core/rdhc` line, is on `develop` too).

## Submission Checklist

- [x] Look over the contributing guidelines at
https://github.com/ROCm/rocm-systems/blob/develop/CONTRIBUTING.md.

---------

Signed-off-by: Adam360x <Adam.pryor@amd.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants