windows-boot-recovery

Windows Recovery Skill

Diagnose and recover non-booting Windows systems using a Linux recovery environment. Covers: NTFS filesystem repair, hibernation state clearing, GPU driver crash recovery, data backup via rsync, and BCD/boot configuration repair.

Problem

A Windows machine will not boot -- stuck in BSOD loops, hung on startup, or reporting NTFS filesystem errors -- and the only available recovery environment is a Linux live USB (e.g., SystemRescue). The user needs to diagnose the root cause, back up critical data before it is lost, and repair the system to a bootable state, all without access to Windows itself.

---

Operator Modes

This skill supports three operators:

• Human (interactive): Presents plan, waits for confirmation per the selected interrupt mode.
• Engineer (autonomous with oversight): Presents plan, runs with interrupts at dangerous steps.
• Agent (fully autonomous): Presents plan and runs end-to-end; logs all actions; flags anomalies.

---

Step 0: Present Plan and Set Interrupt Mode

Always do this first. Before running any commands:

1. State the diagnostic plan in plain language.
2. Identify which steps are DANGEROUS (irreversible or data-affecting):
   • ntfsfix — clears hibernation state, empties NTFS journal (irreversible)
   • rsync with --delete — destructive sync (only if used)
   • Any dd, mkfs, fdisk, gdisk writes
   • Registry edits via chntpw
   • BCD modifications
3. Ask the operator to choose an interrupt mode:

Interrupt modes:
  [A] Prompt before every action
  [B] Prompt before dangerous actions only  ← recommended default
  [C] Run fully — log everything, interrupt only on unexpected errors

Do not proceed until the operator selects a mode. Default to [B] if no response (agent context).

---

Step 1: Verify Recovery Environment

lsblk                        # Confirm NVMe/SATA drive is visible
fdisk -l /dev/nvme0n1        # Verify GPT partition table intact
efibootmgr -v                # Read-only: confirm UEFI boot entries

Expected: NVMe visible, GPT with EFI + NTFS + Recovery partitions. If drive not visible: Check BIOS NVMe mode (RAID vs AHCI). Run nvme list. If still missing — possible dead drive. Stop and report.

---

Step 2: Attempt Mount

mount -t ntfs-3g -o ro /dev/nvme0n1p4 /mnt/win
mount | grep nvme

Outcomes:

• Mounts read-only (fuseblk ro) → hibernation metadata present. Proceed to Step 3.
• Mounts read-write → filesystem clean. Skip to Step 5.
• Fails with BitLocker error → need recovery key. Stop unless key is available.
• Fails with other error → note exact error, proceed to Step 6 (WinRE).

⚠️ FUSE mount blocks raw device access. ntfsinfo, ntfsfix, and similar tools require the volume to be unmounted first.

---

Step 3: Diagnose

Run all diagnostics before any repairs.

3a. Check Hibernation State

xxd -l 4 /mnt/win/hiberfil.sys

• RSTR → Fast Startup hibernation (most common boot blocker)
• hibr/HIBR → Full hibernation
• wake/WAKE → Already resumed, not the issue
• File missing → Not a hibernation issue

3b. Read Registry Hives

⚠️ Copy hives to writable location first — chntpw cannot read from read-only FUSE mount.

mkdir -p /mnt/tmp/diagnostic
cp /mnt/win/Windows/System32/config/SOFTWARE /mnt/tmp/diagnostic/
cp /mnt/win/Windows/System32/config/SYSTEM /mnt/tmp/diagnostic/
chntpw -e /mnt/tmp/diagnostic/SOFTWARE   # Check Windows version/build
chntpw -e /mnt/tmp/diagnostic/SYSTEM    # Check boot state, driver config

In SYSTEM hive: cd Select → check Current (active ControlSet) and Failed value. For driver state: cd ControlSet001\Services\<drivername> → Start value: 0=boot, 1=system, 2=auto, 3=demand, 4=disabled.

3c. Check Crash Dumps

ls -la /mnt/win/Windows/Minidump/
xxd -s 0x38 -l 4 -e <file>.dmp    # Read bugcheck code

GPU crash codes: 0x116 = VIDEO_TDR_FAILURE, 0x119 = VIDEO_SCHEDULER_INTERNAL_ERROR

3d. Check Boot Files

ls -la /mnt/win/Windows/System32/ntoskrnl.exe
ls -la /mnt/win/Windows/System32/winload.efi

---

Step 4: Backup User Data

Do this before any repair that modifies the volume.

⚠️ DANGEROUS if --delete is used. Confirm exclusions with operator before running. Note: Gather ALL exclusions before starting. Restarting rsync re-scans the full tree.

# Identify large directories worth excluding first
du -sh /mnt/win/Users/<username>/*/

# Mount external drive
lsblk   # identify external drive (e.g., /dev/sdb1)
mkdir -p /mnt/backup
mount /dev/sdb1 /mnt/backup

# Run backup
rsync -av --progress \
  --exclude='AppData' \
  --exclude='NTUSER*' \
  --exclude='ntuser*' \
  /mnt/win/Users/<username>/ /mnt/backup/<backup-name>/

Known non-fatal error: rsync exit code 23 with symlink errors on exFAT target = Windows junction points. Not real user data. Safe to ignore or add --no-links.

I/O errors on specific files = possible bad NVMe sectors. Note affected files. rsync will skip and continue. Run chkdsk C: /r after Windows boots.

---

Step 5: Repair — Clear Hibernation (ntfsfix)

⚠️ DANGEROUS — irreversible. Clears hibernation state and empties NTFS journal. Only proceed if hibernation signature is RSTR or hibr and state is known corrupt.

umount /mnt/win
ntfsfix /dev/nvme0n1p4
mount -t ntfs-3g /dev/nvme0n1p4 /mnt/win
mount | grep nvme   # Verify now RW

If still read-only after first run:

ntfsfix -d /dev/nvme0n1p4

---

Step 6: Repair — GPU Driver (after ntfsfix succeeds)

If crash dumps show 0x116/0x119 and Windows boots after ntfsfix:

# From Windows elevated CMD after booting Safe Mode:
bcdedit /set {current} safeboot minimal
# Reboot → Safe Mode → uninstall NVIDIA from Device Manager
bcdedit /deletevalue {current} safeboot
# Reboot normally → install fresh driver from nvidia.com

Also: always disable Fast Startup after recovery:

powercfg /h off

---

Step 7: Repair — WinRE (if ntfsfix insufficient)

Boot to recovery (F12 → recovery partition, or Shift+Restart from Windows):

Troubleshoot → Advanced Options → Command Prompt
bcdedit /set {current} safeboot minimal
powercfg /h off
chkdsk C: /f

---

Step 8: Repair — System Restore (last resort before reinstall)

WinRE → Troubleshoot → Advanced Options → System Restore. Choose restore point from before crash cascade. If no restore points or restore fails → Windows Reset or clean install. Ensure backup from Step 4 is complete first.

---

Decision Tree (Quick Reference)

Drive visible? → NO → Check BIOS NVMe mode. Stop if still missing.
     ↓ YES
Partition table intact? → NO → Run gdisk, testdisk. Stop if unrecoverable.
     ↓ YES
BitLocker? → YES → Need recovery key. Stop.
     ↓ NO
hiberfil.sys signature?
  RSTR/hibr → Step 5 (ntfsfix) [after Step 4 backup]
  wake/missing → Step 3c crash dumps
     ↓
Crash dumps show 0x116/0x119? → YES → Step 6 (GPU driver)
     ↓ NO
BCD/boot files missing? → YES → Step 7 (WinRE)
     ↓ NO
ntfsfix + WinRE both fail → Step 8 (System Restore / reinstall)

---

Hardware Notes: Razer Blade 15 (RZ09-0409, Mid 2021)

• BIOS: F1/DEL to enter setup, F12 for one-time boot menu
• No option to disable boot logo on this model
• GPU: RTX 3080 Mobile — known TDR instability in thin chassis; thermal paste degrades after 2-3 years
• NVIDIA driver: demand-start (Start=3), loaded from DriverStore — failure alone should not prevent boot
• Fast Startup: enabled by default on all Razer OEM installs — always disable after recovery
• Partition layout: p1=Factory Recovery (18.3GB), p2=EFI (100MB), p3=MSR (16MB), p4=Windows (934.5GB), p5=WinRE (1GB)

---

Post-Recovery Checklist

[ ] powercfg /h off                          # Disable Fast Startup permanently
[ ] Uninstall NVIDIA driver (Safe Mode)      # If GPU BSODs were occurring
[ ] Install latest NVIDIA driver             # nvidia.com
[ ] sfc /scannow                             # System file check
[ ] DISM /Online /Cleanup-Image /RestoreHealth
[ ] chkdsk C: /f  (add /r if I/O errors seen)
[ ] Windows Update
[ ] Razer firmware update (support.razer.com, search RZ09-0409)
[ ] Install HWiNFO64, monitor GPU temps under load (>90C sustained = thermal issue)
[ ] Verify backup integrity on external drive

---

Verification

1. After ntfsfix, confirm the NTFS volume mounts read-write: mount -t ntfs-3g /dev/nvme0n1p4 /mnt/win && mount | grep nvme shows rw (not ro).
2. Reboot the machine and confirm Windows reaches the login screen or desktop without BSOD or boot loop.
3. Run chkdsk C: /f from within Windows and verify it reports a clean filesystem with no errors.
4. Verify backed-up user data on the external drive is intact: spot-check file counts and open representative files.
5. Confirm Fast Startup is disabled: powercfg /a should not list "Fast Startup" as available, or powercfg /h confirms hibernation is off.

---

Known Pitfalls

Pitfall	Symptom	Fix
chntpw on FUSE mount	"Access denied"	Copy hive to writable path first
ntfsinfo on mounted volume	"Exclusively opened"	Unmount first
ntfsfix on mounted volume	Fails/no-op	Unmount first
exFAT rsync symlink errors	Exit code 23, symlink failed	Non-fatal; add --no-links
I/O errors on specific files	Input/output error (5)	Note files; run chkdsk /r after boot
rsync restart rescans tree	Slow restart	Gather all exclusions before first run
Razer splash hides POST	Can't see boot errors	Use F12 to interrupt splash

---

Runbook derived from live incident 2026-02-28 (Razer Blade 15 RZ09-0409). Generalized for Windows NTFS boot failures diagnosed from Linux.