Remove NVMe logging (#35740)

cereal/log.capnp

@@ -492,7 +492,6 @@ struct DeviceState @0xa4d8b5af2aa492eb {
   gpuTempC @27 :List(Float32);
   dspTempC @49 :Float32;
   memoryTempC @28 :Float32;
-  nvmeTempC @35 :List(Float32);
   modemTempC @36 :List(Float32);
   pmicTempC @39 :List(Float32);
   intakeTempC @46 :Float32;
@@ -568,6 +567,7 @@ struct DeviceState @0xa4d8b5af2aa492eb {
   chargingDisabledDEPRECATED @18 :Bool;
   usbOnlineDEPRECATED @12 :Bool;
   ambientTempCDEPRECATED @30 :Float32;
+  nvmeTempCDEPRECATED @35 :List(Float32);
 }
 
 struct PandaState @0xa7649e2575e4591e {

common/params_keys.h

@@ -82,7 +82,6 @@ inline static std::unordered_map<std::string, uint32_t> keys = {
     {"NetworkMetered", PERSISTENT},
     {"ObdMultiplexingChanged", CLEAR_ON_MANAGER_START | CLEAR_ON_ONROAD_TRANSITION},
     {"ObdMultiplexingEnabled", CLEAR_ON_MANAGER_START | CLEAR_ON_ONROAD_TRANSITION},
-    {"Offroad_BadNvme", CLEAR_ON_MANAGER_START},
     {"Offroad_CarUnrecognized", CLEAR_ON_MANAGER_START | CLEAR_ON_ONROAD_TRANSITION},
     {"Offroad_ConnectivityNeeded", CLEAR_ON_MANAGER_START},
     {"Offroad_ConnectivityNeededPrompt", CLEAR_ON_MANAGER_START},

selfdrive/selfdrived/alerts_offroad.json

@@ -33,10 +33,6 @@
     "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

system/hardware/base.py

@@ -199,9 +199,6 @@ class HardwareBase(ABC):
   def get_modem_temperatures(self):
     pass
 
-  @abstractmethod
-  def get_nvme_temperatures(self):
-    pass
 
   @abstractmethod
   def initialize_hardware(self):

system/hardware/hardwared.py

@@ -39,7 +39,7 @@ ONROAD_CYCLE_TIME = 1  # seconds to wait offroad after requesting an onroad cycl
 
 ThermalBand = namedtuple("ThermalBand", ['min_temp', 'max_temp'])
 HardwareState = namedtuple("HardwareState", ['network_type', 'network_info', 'network_strength', 'network_stats',
-                                             'network_metered', 'nvme_temps', 'modem_temps'])
+                                             'network_metered', 'modem_temps'])
 
 # List of thermal bands. We will stay within this region as long as we are within the bounds.
 # When exiting the bounds, we'll jump to the lower or higher band. Bands are ordered in the dict.
@@ -142,7 +142,6 @@ def hw_state_thread(end_event, hw_queue):
           network_strength=HARDWARE.get_network_strength(network_type),
           network_stats={'wwanTx': tx, 'wwanRx': rx},
           network_metered=HARDWARE.get_network_metered(network_type),
-          nvme_temps=HARDWARE.get_nvme_temperatures(),
           modem_temps=modem_temps,
         )
 
@@ -189,7 +188,6 @@ def hardware_thread(end_event, hw_queue) -> None:
     network_metered=False,
     network_strength=NetworkStrength.unknown,
     network_stats={'wwanTx': -1, 'wwanRx': -1},
-    nvme_temps=[],
     modem_temps=[],
   )
 
@@ -268,7 +266,6 @@ def hardware_thread(end_event, hw_queue) -> None:
     if last_hw_state.network_info is not None:
       msg.deviceState.networkInfo = last_hw_state.network_info
 
-    msg.deviceState.nvmeTempC = last_hw_state.nvme_temps
     msg.deviceState.modemTempC = last_hw_state.modem_temps
 
     msg.deviceState.screenBrightnessPercent = HARDWARE.get_screen_brightness()
@@ -337,16 +334,6 @@ def hardware_thread(end_event, hw_queue) -> None:
       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())
@@ -431,8 +418,6 @@ def hardware_thread(end_event, hw_queue) -> None:
     statlog.gauge("memory_temperature", msg.deviceState.memoryTempC)
     for i, temp in enumerate(msg.deviceState.pmicTempC):
       statlog.gauge(f"pmic{i}_temperature", temp)
-    for i, temp in enumerate(last_hw_state.nvme_temps):
-      statlog.gauge(f"nvme_temperature{i}", temp)
     for i, temp in enumerate(last_hw_state.modem_temps):
       statlog.gauge(f"modem_temperature{i}", temp)
     statlog.gauge("fan_speed_percent_desired", msg.deviceState.fanSpeedPercentDesired)

system/hardware/pc/hardware.py

@@ -70,8 +70,6 @@ class Pc(HardwareBase):
   def get_modem_temperatures(self):
     return []
 
-  def get_nvme_temperatures(self):
-    return []
 
   def initialize_hardware(self):
     pass

system/hardware/tici/hardware.py

@@ -1,4 +1,3 @@
-import json
 import math
 import os
 import subprocess
@@ -291,15 +290,6 @@ class Tici(HardwareBase):
     except Exception:
       return []
 
-  def get_nvme_temperatures(self):
-    ret = []
-    try:
-      out = subprocess.check_output("sudo smartctl -aj /dev/nvme0", shell=True)
-      dat = json.loads(out)
-      ret = list(map(int, dat["nvme_smart_health_information_log"]["temperature_sensors"]))
-    except Exception:
-      pass
-    return ret
 
   def get_current_power_draw(self):
     return (self.read_param_file("/sys/class/hwmon/hwmon1/power1_input", int) / 1e6)

system/loggerd/bootlog.cc

@@ -31,9 +31,6 @@ static kj::Array<capnp::word> build_boot_log() {
     "[ -x \"$(command -v journalctl)\" ] && journalctl -o short-monotonic",
   };
 
-  if (Hardware::TICI()) {
-    bootlog_commands.push_back("[ -e /dev/nvme0 ] && sudo nvme smart-log --output-format=json /dev/nvme0");
-  }
 
   auto commands = boot.initCommands().initEntries(bootlog_commands.size());
   for (int j = 0; j < bootlog_commands.size(); j++) {