Log UFS health

openpilot/cereal/log.capnp

@@ -178,6 +178,15 @@ struct InitData {
 
   wallTimeNanos @20 :UInt64;
 
+  ufsHealth @25 :UfsHealth;
+
+  struct UfsHealth {
+    preEolInfo @0 :UInt8;
+    lifeTimeEstimateA @1 :UInt8;
+    lifeTimeEstimateB @2 :UInt8;
+    vendorHealthReport @3 :Data;
+  }
+
   enum DeviceType {
     unknown @0;
     neo @1;

openpilot/common/hardware/base.h

@@ -1,15 +1,25 @@
 #pragma once
 
 #include <cstdlib>
+#include <cstdint>
 #include <fstream>
 #include <map>
+#include <optional>
 #include <string>
+#include <vector>
 
 #include "openpilot/cereal/gen/cpp/log.capnp.h"
 
 // no-op base hw class
 class HardwareNone {
 public:
+  struct UfsHealth {
+    uint8_t pre_eol_info;
+    uint8_t life_time_estimate_a;
+    uint8_t life_time_estimate_b;
+    std::vector<uint8_t> vendor_health_report;
+  };
+
   static std::string get_name() { return ""; }
   static cereal::InitData::DeviceType get_device_type() { return cereal::InitData::DeviceType::UNKNOWN; }
 
@@ -19,6 +29,8 @@ class HardwareNone {
     return {};
   }
 
+  static std::optional<UfsHealth> get_ufs_health() { return std::nullopt; }
+
   static void set_ir_power(int percentage) {}
 
   static bool PC() { return false; }

openpilot/common/hardware/tici/hardware.h

@@ -1,16 +1,62 @@
 #pragma once
 
 #include <cassert>
+#include <array>
+#include <cstddef>
+#include <cstdint>
+#include <fcntl.h>
 #include <fstream>
 #include <map>
 #include <string>
 #include <algorithm>  // for std::clamp
+#include <sys/ioctl.h>
+#include <unistd.h>
 
 #include "common/util.h"
 #include "common/hardware/base.h"
 
 class HardwareTici : public HardwareNone {
 public:
+  static std::optional<UfsHealth> get_ufs_health() {
+    constexpr unsigned long UFS_IOCTL_QUERY = 0x5388;
+    constexpr uint32_t UPIU_QUERY_OPCODE_READ_DESC = 0x1;
+    constexpr uint8_t QUERY_DESC_IDN_HEALTH = 0x9;
+    constexpr uint16_t QUERY_DESC_HEALTH_SIZE = 0x25;
+
+    struct UfsQuery {
+      uint32_t opcode;
+      uint8_t idn;
+      uint8_t reserved;
+      uint16_t buf_size;
+      std::array<uint8_t, QUERY_DESC_HEALTH_SIZE> buffer;
+    };
+    static_assert(offsetof(UfsQuery, buffer) == 8);
+
+    UfsQuery query = {};
+    query.opcode = UPIU_QUERY_OPCODE_READ_DESC;
+    query.idn = QUERY_DESC_IDN_HEALTH;
+    query.buf_size = QUERY_DESC_HEALTH_SIZE;
+
+    int fd = open("/dev/sda", O_RDONLY | O_CLOEXEC);
+    if (fd < 0) {
+      return std::nullopt;
+    }
+
+    int ret = ioctl(fd, UFS_IOCTL_QUERY, &query);
+    close(fd);
+    if (ret != 0 || query.buf_size < 5 || query.buf_size > query.buffer.size() ||
+        query.buffer[0] != query.buf_size || query.buffer[1] != QUERY_DESC_IDN_HEALTH) {
+      return std::nullopt;
+    }
+
+    return UfsHealth{
+      query.buffer[2],
+      query.buffer[3],
+      query.buffer[4],
+      std::vector<uint8_t>(query.buffer.begin() + 5, query.buffer.begin() + query.buf_size),
+    };
+  }
+
   static std::string get_name() {
     static const std::string name = []() {
       std::string model = util::read_file("/sys/firmware/devicetree/base/model");

openpilot/system/loggerd/logger.cc

@@ -39,6 +39,14 @@ kj::Array<capnp::word> logger_build_init_data() {
   init.setKernelVersion(util::read_file("/proc/version"));
   init.setOsVersion(util::read_file("/VERSION"));
 
+  if (auto health = Hardware::get_ufs_health()) {
+    auto ufs_health = init.initUfsHealth();
+    ufs_health.setPreEolInfo(health->pre_eol_info);
+    ufs_health.setLifeTimeEstimateA(health->life_time_estimate_a);
+    ufs_health.setLifeTimeEstimateB(health->life_time_estimate_b);
+    ufs_health.setVendorHealthReport(capnp::Data::Reader(health->vendor_health_report.data(), health->vendor_health_report.size()));
+  }
+
   // log params
   Params params(util::getenv("PARAMS_COPY_PATH", ""));
   std::map<std::string, std::string> params_map = params.readAll();

openpilot/system/loggerd/tests/test_loggerd.py

@@ -178,6 +178,15 @@ def test_init_data_values(self):
     assert initData.dirty != bool(os.environ["CLEAN"])
     assert initData.version == get_version()
 
+    if TICI:
+      assert initData._has("ufsHealth")
+      assert initData.ufsHealth.preEolInfo in (1, 2, 3)
+      assert 1 <= initData.ufsHealth.lifeTimeEstimateA <= 11
+      assert 1 <= initData.ufsHealth.lifeTimeEstimateB <= 11
+      assert len(initData.ufsHealth.vendorHealthReport) == 32
+    else:
+      assert not initData._has("ufsHealth")
+
     if os.path.isfile("/proc/cmdline"):
       with open("/proc/cmdline") as f:
         assert list(initData.kernelArgs) == f.read().strip().split(" ")