fix(xhci): bounded EnableSlot/AddressDevice retry — transient timeout no longer kills the mouse for the whole session

[ryanbreen]

The bug

xHCI port enumeration issued EnableSlot/AddressDevice single-shot, with no retry. A single transient timeout on port 1 (the mouse's port) meant the mouse silently enumerated onto slot 0 (or failed to enumerate at all) — the pointer would be frozen at 0,0 for the rest of the session, with no recovery path short of a reboot.

This was found live by the operator immediately after PR #411 landed, but xhci.rs was untouched by that PR — this is pre-existing latent fragility that PR #411 merely made visible (or was coincidental timing), not a regression it introduced.

The fix

Bounded 3-attempt retry around EnableSlot/AddressDevice, restricted to the cold enumeration path only (no changes to steady-state HID polling or any interrupt/syscall-hot-path code). A transient controller timeout on one attempt no longer permanently dooms that port's device for the session.

Harness assertion

scripts/parallels/launcher-smoke.sh now asserts on mouse=slotN where N != 0 — smoke runs FAIL if the mouse ends up bound to slot 0 (the failure signature this bug produces), so this class of regression can't silently reappear.

Validation — 3/3 smoke boots PASS with mouse enumerated

[
  {
    "boot": 1,
    "result": "PASS",
    "mouseEnumerated": true,
    "mouseEvidence": "[xhci] start_hid_polling: kbd=slot2/dci3 nkro=dci5 mouse=slot1/dci3 mouse2=dci5\n(no \"EnableSlot retry\" lines this run — all ports enumerated cleanly on first attempt: \"[xhci] port 1 EnableSlot -> slot 1\", \"[xhci] port 2 EnableSlot -> slot 2\", \"[xhci] port 3 EnableSlot -> slot 3\")",
    "evidencePath": "/Users/wrb/fun/code/breenix/logs/parallels-launcher-test/run-20260704-165209/"
  },
  {
    "boot": 2,
    "result": "PASS",
    "mouseEnumerated": true,
    "mouseEvidence": "[smoke 16:55:57] USB mouse enumerated: slot 1 ([xhci] start_hid_polling: kbd=slot2/dci3 nkro=dci5 mouse=slot1/dci3 mouse2=dci5)",
    "evidencePath": "/private/tmp/claude-501/-Users-wrb-fun-code-breenix/94deec68-57ce-4666-8cb5-544f45bb0f62/scratchpad/launcher-smoke-boot2.log"
  },
  {
    "boot": 3,
    "result": "PASS",
    "mouseEnumerated": true,
    "mouseEvidence": "[smoke 16:59:02] USB mouse enumerated: slot 1 ([xhci] start_hid_polling: kbd=slot2/dci3 nkro=dci5 mouse=slot1/dci3 mouse2=dci5)",
    "evidencePath": "/private/tmp/claude-501/-Users-wrb-fun-code-breenix/94deec68-57ce-4666-8cb5-544f45bb0f62/scratchpad/smoke-run3.log"
  }
]

Note boot 1 had no retry lines at all (clean first-attempt enumeration on every port) — the harness assertion (mouse != slot0) holds regardless of whether the retry path was actually exercised in a given boot, which is the correct behavior for a bounded-retry safety net.

🤖 Generated with Claude Code