Add warning for unsupported NVME (commaai#23972)

* Add warning for unsupported NVME

* fix it up

* fix typo

Co-authored-by: Comma Device <device@comma.ai>

Author: 2 people

selfdrive/common/params.cc

@@ -166,6 +166,7 @@ std::unordered_map<std::string, uint32_t> keys = {
     {"ApiCache_DriveStats", PERSISTENT},
     {"ApiCache_NavDestinations", PERSISTENT},
     {"ApiCache_Owner", PERSISTENT},
+    {"Offroad_BadNvme", CLEAR_ON_MANAGER_START},
     {"Offroad_CarUnrecognized", CLEAR_ON_MANAGER_START | CLEAR_ON_IGNITION_ON},
     {"Offroad_ChargeDisabled", CLEAR_ON_MANAGER_START },
     {"Offroad_ConnectivityNeeded", CLEAR_ON_MANAGER_START},

selfdrive/controls/lib/alerts_offroad.json

@@ -41,6 +41,10 @@
     "text": "NVMe drive not mounted.",
     "severity": 1
   },
+  "Offroad_BadNvme": {
+    "text": "Unsupported NVMe drive detected. Device may draw significantly more power and overheat due to the unsupported NVMe.",
+    "severity": 1
+  },
   "Offroad_CarUnrecognized": {
     "text": "openpilot was unable to identify your car. Your car is either unsupported or its ECUs are not recognized. Please submit a pull request to add the firmware versions to the proper vehicle. Need help? Join discord.comma.ai.",
     "severity": 0

selfdrive/thermald/thermald.py

@@ -278,9 +278,21 @@ def thermald_thread(end_event, hw_queue):
     onroad_conditions["device_temp_good"] = thermal_status < ThermalStatus.danger
     set_offroad_alert_if_changed("Offroad_TemperatureTooHigh", (not onroad_conditions["device_temp_good"]))
 
+    # TODO: this should move to TICI.initialize_hardware, but we currently can't import params there
     if TICI:
-      missing = (not Path("/data/media").is_mount()) and (not os.path.isfile("/persist/comma/living-in-the-moment"))
-      set_offroad_alert_if_changed("Offroad_StorageMissing", missing)
+      if not os.path.isfile("/persist/comma/living-in-the-moment"):
+        if not Path("/data/media").is_mount():
+          set_offroad_alert_if_changed("Offroad_StorageMissing", True)
+        else:
+          # check for bad NVMe
+          try:
+            with open("/sys/block/nvme0n1/device/model") as f:
+              model = f.read().strip()
+            if not model.startswith("Samsung SSD 980") and params.get("Offroad_BadNvme") is None:
+              set_offroad_alert_if_changed("Offroad_BadNvme", True)
+              cloudlog.event("Unsupported NVMe", model=model, error=True)
+          except Exception:
+            pass
 
     # Handle offroad/onroad transition
     should_start = all(onroad_conditions.values())