Add 'Secret Menu' to HW-Diags

*Begrudgingly

scripts/wk/cfg/hw.py

@@ -14,6 +14,7 @@ ATTRIBUTE_COLORS = (
 # NOTE: Force 4K read block size for disks >= 3TB
 BADBLOCKS_EXTRA_LARGE_DISK = 15 * 1024**4
 BADBLOCKS_LARGE_DISK = 3 * 1024**4
+BADBLOCKS_MAX_ERRORS = 1
 BADBLOCKS_REGEX = re.compile(
   r'^Pass completed, (\d+) bad blocks found. .(\d+)/(\d+)/(\d+) errors',
   re.IGNORECASE,

scripts/wk/hw/diags.py

@@ -11,7 +11,7 @@ import subprocess
 from docopt import docopt
 
 from wk import cfg, debug, exe, log, osticket, std
-from wk.cfg.hw import STATUS_COLORS
+from wk.cfg.hw import CPU_TEST_MINUTES, STATUS_COLORS
 from wk.hw import benchmark as hw_benchmark
 from wk.hw import cpu as hw_cpu
 from wk.hw import disk as hw_disk
@@ -42,12 +42,6 @@ Options:
   -h --help               Show this page
   -q --quick              Skip menu and perform a quick check
   -t --test-mode      Run diags in test mode
-
-  --ignore-smart-errors   NOT RECOMMENDED!
-                          Only use if you have RTFM,
-                          know what you're doing,
-                          understand the risks,
-                          and accept responsibililty.
 '''
 LOG = logging.getLogger(__name__)
 IO_SIZE_SKIP_NAME = (
@@ -77,7 +71,13 @@ MENU_ACTIONS = (
 MENU_ACTIONS_SECRET = (
   'Matrix',
   'Tubes',
+  '?Secrets',
   )
+MENU_OPTIONS_SECRET = {
+  'Ignore SMART errors': False,
+  'Override CPU Testing Time (minutes)': CPU_TEST_MINUTES,
+  'Override Surface Scan Error Limit': 1,
+}
 MENU_OPTIONS_QUICK = ('Disk Attributes',)
 MENU_SETS = {
   'Full Diagnostic': (*TEST_GROUPS,),
@@ -107,6 +107,7 @@ class State():
     self.log_dir: pathlib.Path | None = None
     self.ost = osticket.osTicket()
     self.progress_file: pathlib.Path | None = None
+    self.secret_menu = build_secret_menu()
     self.sensors: hw_sensors.Sensors = hw_sensors.Sensors()
     self.system: hw_system.System | None = None
     self.test_groups: list[TestGroup] = []
@@ -370,6 +371,24 @@ def build_menu(cli_mode=False, quick_mode=False) -> cli.Menu:
   return menu
 
 
+def build_secret_menu() -> cli.Menu:
+  title_text = [
+    ansi.color_string(('Expert Settings', "(It's a secret to everyone!)"), ('ORANGE', None)),
+    ' ',
+    ansi.color_string(
+      [' !!', 'These settings can cause', 'MAJOR DAMAGE', 'to systems'],
+      [None, 'YELLOW', 'RED', 'YELLOW'],
+      ),
+    ' !! Please read the manual before making changes',
+    ' ',
+    ]
+  menu = cli.Menu(title='\n'.join(title_text))
+  for option, value in MENU_OPTIONS_SECRET.items():
+    menu.add_option(option, {'Selected': False, 'Value': value})
+  menu.add_action('Main Menu')
+  return menu
+
+
 def cpu_tests_init(state: State) -> None:
   """Initialize CPU tests."""
   sensors_out = pathlib.Path(f'{state.log_dir}/sensors.out')
@@ -426,6 +445,11 @@ def cpu_test_mprime(state: State, test_object, test_mode=False) -> None:
   test_minutes = cfg.hw.CPU_TEST_MINUTES
   if test_mode:
     test_minutes = cfg.hw.TEST_MODE_CPU_LIMIT
+  if state.secret_menu.options.get(
+      'Override CPU Testing Time (minutes)')['Selected']:
+    test_minutes = int(
+        state.secret_menu.options.get('Override CPU Testing Time (minutes)')['Value'],
+      )
 
   # Bail early
   if test_object.disabled:
@@ -502,6 +526,11 @@ def cpu_test_sysbench(state: State, test_object, test_mode=False) -> None:
   test_minutes = cfg.hw.CPU_TEST_MINUTES
   if test_mode:
     test_minutes = cfg.hw.TEST_MODE_CPU_LIMIT
+  if state.secret_menu.options.get(
+      'Override CPU Testing Time (minutes)')['Selected']:
+    test_minutes = int(
+        state.secret_menu.options.get('Override CPU Testing Time (minutes)')['Value'],
+      )
 
   # Bail early
   if test_object.disabled:
@@ -758,7 +787,12 @@ def disk_surface_scan(state: State, test_objects, test_mode=False) -> None:
   """Read-only disk surface scan using badblocks."""
   LOG.info('Disk Surface Scan (badblocks)')
   aborted = False
+  max_errors = 1
   threads = []
+  if state.secret_menu.options.get(
+      'Override Surface Scan Error Limit')['Selected']:
+    max_errors = state.secret_menu.options.get(
+      'Override Surface Scan Error Limit')['Value']
 
   # Update panes
   state.update_title_text(
@@ -775,7 +809,7 @@ def disk_surface_scan(state: State, test_objects, test_mode=False) -> None:
     # Start thread
     test_log = f'{state.log_dir}/{test.dev.path.name}_badblocks.log'
     threads.append(exe.start_thread(
-      hw_surface_scan.run_scan, args=(test, test_log, test_mode),
+      hw_surface_scan.run_scan, args=(test, test_log, test_mode, max_errors),
       ))
 
     # Show progress
@@ -832,7 +866,6 @@ def main() -> None:
   # Init
   menu = build_menu(cli_mode=args['--cli'], quick_mode=args['--quick'])
   state = State(test_mode=args['--test-mode'])
-  state.override_all_smart_errors = args['--ignore-smart-errors']
 
   # Quick Mode
   if args['--quick']:
@@ -868,9 +901,11 @@ def main() -> None:
         state.ui.update_clock()
 
     # Secrets
+    if '?Secrets' in selection:
+      state.secret_menu.settings_select()
     if 'Matrix' in selection:
       screensaver('matrix')
-    elif 'Tubes' in selection:
+    if 'Tubes' in selection:
       # Tubes ≈≈ Pipes?
       screensaver('pipes')
 

scripts/wk/hw/surface_scan.py

@@ -70,7 +70,7 @@ def check_surface_scan_results(test_obj, log_path) -> None:
     test_obj.set_status('Unknown')
 
 
-def run_scan(test_obj, log_path, test_mode=False) -> None:
+def run_scan(test_obj, log_path, test_mode=False, max_errors=1) -> None:
   """Run surface scan and handle exceptions."""
   block_size = '1024'
   dev = test_obj.dev
@@ -89,8 +89,17 @@ def run_scan(test_obj, log_path, test_mode=False) -> None:
       or dev.size >= BADBLOCKS_LARGE_DISK):
     block_size = '4096'
 
+  # Max errors
+  if int(max_errors) <= 0:
+    max_errors = ''
+  else:
+    max_errors = f'-e{max_errors}'
+
   # Start scan
-  cmd = ['sudo', 'badblocks', '-sv', '-b', block_size, '-e', '1', dev_path]
+  cmd = ['sudo', 'badblocks', '-sv', '-b', block_size]
+  if max_errors:
+    cmd.append(max_errors)
+  cmd.append(dev_path)
   if test_mode:
     # Only test a limited scope instead of the whole device
     cmd.append(TEST_MODE_BADBLOCKS_LIMIT)