…12 diagnostics

- Add second mouse interface (DCI=5) support: MOUSE2_REPORT_BUF, MSI_MOUSE2_NEEDS_REQUEUE,
  NEEDS_RESET_MOUSE2, mouse_nkro_endpoint field in XhciState, full poll/reset/requeue paths
- Add 4th HID transfer ring (NUM_TRANSFER_RINGS = MAX_SLOTS + 4) for mouse2
- Remove EP0 GET_REPORT polling workaround (queue_ep0_get_report, queue_ep0_mouse_get_report,
  EP0_POLL_STATE, EP0_MOUSE_POLL_STATE, MOUSE_CTRL_DATA_BUF) — switching to pure interrupt IN
- Expand KBD_REPORT_BUF and MOUSE_REPORT_BUF to 64 bytes for safety margin
- SKIP_BW_DANCE=true (diagnostic): Linux ftrace confirmed bandwidth dance IS performed;
  isolating whether batch ConfigEP alone enables endpoints on Parallels virtual xHC
- MINIMAL_INIT=true (diagnostic): confirmed CC=12 is in xHCI setup, not HID class requests
- Update bandwidth dance comments: Linux uses DIFFERENT Input Context address for each
  per-EP re-ConfigEP; drop+re-add flags per-EP; only target EP context populated (not all)
- Fix ring comment: four HID rings indexed kbd-boot/mouse/kbd-NKRO/mouse2

CC=12 (Endpoint Not Enabled) on interrupt IN endpoints remains unresolved.
Next: compare endpoint context state before vs after SET_CONFIGURATION to determine
if virtual xHC resets endpoint state on SET_CONFIG receipt.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

CC=12 (Endpoint Not Enabled) from Parallels virtual xHC halts interrupt
IN endpoints immediately on each poll when no key is pressed. Three
changes bring the reset rate from 200/s down to ~10/s:

1. wait_for_command_completion CC=12 path: route to MSI_*_NEEDS_REQUEUE
   instead of NEEDS_RESET_*. This breaks the cascade where one endpoint's
   reset command wait sets NEEDS_RESET_* for the other endpoint's CC=12
   event, causing both endpoints to reset on the same poll tick.

2. RESET_INTERVAL_TICKS (20 ticks = 100ms at 200Hz) + per-endpoint
   last-reset poll counters. NEEDS_RESET_* is deferred until the minimum
   interval elapses, capping resets at 10/s per endpoint.

3. CC=12 in handle_interrupt and poll_hid_events routes to NEEDS_RESET_*
   (not MSI_*_NEEDS_REQUEUE) so Halted endpoints get proper Reset Endpoint
   recovery rather than a doorbell that Halted endpoints silently ignore.

Verified on Parallels Desktop ARM64 (NEC uPD720200 virtual xHC):
- er drops from 200/s → 10/s (rate limited)
- xe tracks er 1:1 (one error per reset cycle, not compounding)
- c1=0x00020502 confirms CC=12 halts NKRO DCI=5 (state=2=Halted)
- ra=0x00020303 confirms Reset Endpoint leaves boot DCI=3 in Stopped (3)
- System stable for 100s with no panics or cascades

Co-Authored-By: Ryan Breen <ryan@ryanbreen.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

…output

Remove 124 serial_println! calls from the xHCI driver. All diagnostics
now go through the lock-free xhci_trace ring buffer (no locks, no
allocations, safe in interrupt context). Only xhci_trace_dump() retains
serial output for post-init trace extraction.

Key changes:
- xhci.rs: Remove all serial_println, replace key milestones with
  xhci_trace_note() calls. Fix Mult field to 0 per xHCI spec §6.2.3
  (was incorrectly 1 for interrupt endpoints). Add USBLEGSUP handoff
  in Extended Capabilities parsing. Revert No-Op TRB diagnostic back
  to Normal TRB.
- timer_interrupt.rs: Trim heartbeat to essential fields only (time,
  ctx switches, syscalls, xhci errors, kbd events, first CC).
- pci.rs: Add enable_intx() and disable_msi() methods.
- drivers/mod.rs: Remove raw_serial_str debug breadcrumbs.
- build.rs: Add build ID based on timestamp for stale-build detection.
- main_aarch64.rs: Print BUILD_ID in boot banner.
- build-efi.sh: Prefer mformat over newfs_msdos for FAT32 on macOS.
- deploy-to-vm.sh: Rewrite with prl_disk_tool HDD creation, serial
  log truncation, NVRAM cleanup, and --boot flag.
- CLAUDE.md: Document Parallels VM workflow and restart protocol.
- Add Linux xHCI ftrace reference and trace analysis scripts.

CC=12 (ENDPOINT_NOT_ENABLED) on interrupt IN endpoints remains unsolved.

Co-Authored-By: Ryan Breen <ryanbreen@gmail.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Detailed document covering the full state of the xHCI CC=12
(ENDPOINT_NOT_ENABLED) investigation across 26+ test iterations.
Includes verified-correct items, remaining theories, code structure
map, build/test instructions, and trace infrastructure usage.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>