Refactor SMART self-test checks

- Preserve TimedOut status
- Adds last self-test result to notes (if present and result is unknown)

scripts/wk/hw/diags.py

@@ -118,7 +118,13 @@ class State():
     for dev in self.disks:
       disk_smart_status_check(dev, mid_run=True)
       for test in dev.tests:
-        if test.failed and 'Attributes' not in test.name:
+        if test.failed:
+          # Skip acceptable failure states
+          if 'Attributes' in test.name:
+            continue
+          if 'Self-Test' in test.name and 'TimedOut' in test.status:
+            continue
+          # Disable remaining tests
           dev.disable_disk_tests()
           break
 

scripts/wk/hw/smart.py

@@ -231,6 +231,28 @@ def get_smart_self_test_details(dev) -> dict[Any, Any]:
   return details
 
 
+def get_smart_self_test_last_result(dev) -> str:
+  """Get last SMART self-test result, returns str."""
+  result = 'Unknown'
+
+  # Parse SMART data
+  data = dev.raw_smartctl.get('ata_smart_self_test_log', {}).get('standard', {}).get('table', [])
+  if not data:
+    # No results found
+    return result
+
+  # Build result string
+  result = (
+    f'Power-on hours: {data.get("lifetime_hours", "?")}'
+    f', Type: {data.get("type", {}).get("string", "?")}'
+    f', Passed: {data.get("status", {}).get("passed", "?")}'
+    f', Result: {data.get("status", {}).get("string", "?")}'
+  )
+
+  # Done
+  return result
+
+
 def monitor_smart_self_test(test_obj, header_str, log_path) -> bool:
   """Monitor SMART self-test status and update test_obj, returns bool."""
   started = False
@@ -262,6 +284,10 @@ def monitor_smart_self_test(test_obj, header_str, log_path) -> bool:
       if _i * 5 >= SMART_SELF_TEST_START_TIMEOUT_IN_SECONDS:
         # Test didn't start within limit, stop waiting
         abort_self_test(test_obj.dev)
+        result = get_smart_self_test_last_result(test_obj.dev)
+        if result == 'Unknown':
+          result = 'SMART self-test failed to start'
+        test_obj.dev.add_note(result)
         test_obj.failed = True
         test_obj.set_status('TimedOut')
         break
@@ -290,7 +316,7 @@ def run_self_test(test_obj, log_path) -> None:
   run_smart_self_test(test_obj, log_path)
 
 
-def run_smart_self_test(test_obj, log_path) -> bool:
+def run_smart_self_test(test_obj, log_path) -> None:
   """Run SMART self-test and check if it passed, returns bool.
 
   NOTE: An exception will be raised if the disk lacks SMART support.
@@ -352,7 +378,10 @@ def run_smart_self_test(test_obj, log_path) -> bool:
     test_obj.failed = test_obj.failed or not test_obj.passed
 
   # Set status
-  if test_obj.failed and test_obj.status != 'TimedOut':
+  if test_obj.status == 'TimedOut':
+    # Preserve TimedOut status
+    pass
+  elif test_obj.failed:
     test_obj.set_status('Failed')
   elif test_obj.passed:
     test_obj.set_status('Passed')