Mitigate + instrument utility-VM boot-connect timeout (62543002)

[yeelam-gordon]

What

Utility-VM Initialize waits for the Linux guest to connect back over hvsocket within KernelBootTimeout — first mini_init, then the notify channel. When a boot is healthy but slow (loaded host, cold first boot), this watchdog trips and surfaces as .../CreateVm/0x800705b4 (ERROR_TIMEOUT). This is the reported failure in bug 62543002 on loaded SP11 ARM64 CI.

KernelBootTimeout is a guest-responsiveness watchdog (a give-up ceiling), not a boot budget — a healthy boot connects in ~1–3s. The 30s default leaves too little headroom for a contended host, and the notify-channel accept was uninstrumented, so the lost time could not be attributed to a phase.

Fix

Two small, targeted changes:

1. Mitigate — raise the default KernelBootTimeout from 30s to 90s (WslCoreConfig.h). Still bounds a genuinely stuck VM; just gives a slow-but-alive boot room to finish. User-overridable via wsl2.kernelBootTimeout.
2. Instrument — wrap the notify-channel accept in its own SlowOperationWatcher{"WaitForNotifyChannelConnect"}, mirroring the adjacent WaitForMiniInitConnect block, so slow/hung boots are attributed to the correct phase. Diagnostic-only — no control-flow change.

Why this isn't already fixed by microsoft#40519

microsoft#40519 (merged, in this PR's base) protects critical WSL processes via a guest-side wsl-user cgroup. It does not address this failure: its host-side change is only an IsolateDistroCgroup flag (no host CPU reservation/priority on the VM), and its cgroup tree is built inside the running distro after mini_init has already connected — a different phase, and intra-guest contention rather than the host starving the whole VM's vCPUs during boot.

Root cause (follow-up)

The underlying disease is host CPU-scheduling starvation of the utility VM's vCPUs during cold first boot: the VM is created with all logical processors but no CPU reservation/weight/priority. The real cure is a host-side CPU guarantee for the VM until mini_init connects — to be built after confirming the starved resource via the phase telemetry above plus a host resource snapshot.

Testing

• SlowOperationWatcher unit tests: Total=6 Passed=6, including SlowCompletionEmitsOneRecord (real elapsedMs) and HangEmitsOnceAtMaxThenStops (timedOut=true hang backstop).
• wslservice build-verified (CMake configure + build, no errors) with both changes.
• Deterministic repro confirmed on the same accept path: setting kernelBootTimeout=1 yields the exact .../CreateVm/0x800705b4.

[yeelam-gordon]

Code-review loop — outcome

Verdict: APPROVE. No must-fix items.

Verified: the new block scopes only the SlowOperationWatcher construction + the m_notifyChannel assignment (member lifetime unaffected); SlowOperationWatcher is already used throughout this TU with the string-literal ctor; purely additive, passive instrumentation mirroring the adjacent WaitForMiniInitConnect block — no control-flow, timeout, or assignment-semantics change.

Tests: SlowOperationWatcher unit tests 6/6 PASS (real elapsedMs; timedOut=true hang backstop). Composes with the elapsed-time telemetry work (build-verified).

[Copilot]

Pull request overview

This PR adds missing phase-level telemetry during utility VM initialization by instrumenting the notify-channel accept with SlowOperationWatcher, allowing slow/timeout boot attribution between the mini_init accept and the notify-channel accept phases.

Changes:

• Wrap the notify-channel AcceptConnection(...) call in a dedicated SlowOperationWatcher{"WaitForNotifyChannelConnect"} scope.
• Add explanatory in-code comments describing why this accept is separately instrumented.

[Copilot]

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated no new comments.

[Copilot]

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated no new comments.

[Copilot]

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated no new comments.

[Copilot]

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated no new comments.

[yeelam-gordon]

Superseded by upstream microsoft#41165 (same squashed change).