Add MPX trace log harvesting and on-demand support

config/mp_info.json

@@ -0,0 +1,106 @@
+[
+    {
+        "Name": "IOD0_MPART",
+        "Index": 0
+    },
+    {
+        "Name": "IOD0_MPASP",
+        "Index": 1
+    },
+    {
+        "Name": "IOD0_MP1",
+        "Index": 2
+    },
+    {
+        "Name": "IOD0_MPIO",
+        "Index": 3
+    },
+    {
+        "Name": "IOD0_MPRAS",
+        "Index": 4
+    },
+    {
+        "Name": "IOD0_MPM",
+        "Index": 5
+    },
+    {
+        "Name": "IOD0_MPDACC0",
+        "Index": 6
+    },
+    {
+        "Name": "IOD0_MPDACC1",
+        "Index": 7
+    },
+    {
+        "Name": "IOD0_MPDACC2",
+        "Index": 8
+    },
+    {
+        "Name": "IOD0_CCD0_MP5",
+        "Index": 9
+    },
+    {
+        "Name": "IOD0_CCD1_MP5",
+        "Index": 10
+    },
+    {
+        "Name": "IOD0_CCD2_MP5",
+        "Index": 11
+    },
+    {
+        "Name": "IOD0_CCD3_MP5",
+        "Index": 12
+    },
+    {
+        "Name": "IOD1_MPART",
+        "Index": 13
+    },
+    {
+        "Name": "IOD1_MPASP",
+        "Index": 14
+    },
+    {
+        "Name": "IOD1_MP1",
+        "Index": 15
+    },
+    {
+        "Name": "IOD1_MPIO",
+        "Index": 16
+    },
+    {
+        "Name": "IOD1_MPRAS",
+        "Index": 17
+    },
+    {
+        "Name": "IOD1_MPM",
+        "Index": 18
+    },
+    {
+        "Name": "IOD1_MPDACC0",
+        "Index": 19
+    },
+    {
+        "Name": "IOD1_MPDACC1",
+        "Index": 20
+    },
+    {
+        "Name": "IOD1_MPDACC2",
+        "Index": 21
+    },
+    {
+        "Name": "IOD1_CCD0_MP5",
+        "Index": 22
+    },
+    {
+        "Name": "IOD1_CCD1_MP5",
+        "Index": 23
+    },
+    {
+        "Name": "IOD1_CCD2_MP5",
+        "Index": 24
+    },
+    {
+        "Name": "IOD1_CCD3_MP5",
+        "Index": 25
+    }
+]

config/ras_config.json

@@ -139,6 +139,24 @@
                 "Description": "Error threshold count for PCIE AER errors",
                 "Value": 1
             }
+	},
+        {
+            "Soc0MPList": {
+                "Description": "List of Socket 0 MP's where tracelogs are enabled",
+                "Value": [
+                    "IOD0_MPART",
+                    "IOD0_MPASP"
+                ]
+            }
+        },
+        {
+            "Soc1MPList": {
+                "Description": "List ofSocket 1 MP's where tracelogs are enabled",
+                "Value": [
+                    "IOD0_MPART",
+                    "IOD0_MPASP"
+                ]
+            }
         }
     ]
 }

include/apml_manager.hpp

@@ -95,6 +95,7 @@ class Manager : public amd::ras::Manager
     uint8_t progId;
     size_t contextType;
     uint64_t recordId;
+    uint16_t fatalSectionCount;
     size_t watchdogTimerCounter;
     boost::asio::io_context& io;
     std::vector<uint8_t> blockId;
@@ -361,6 +362,22 @@ class Manager : public amd::ras::Manager
      */
     void getLastTransAddr(const std::shared_ptr<FatalCperRecord>&, uint8_t);
 
+    /** @brief Harvests MPX trace logs.
+     *
+     * @details This function collects MPX trace logs from the system
+     * when a fatal error occurs.
+     *
+     * @param[in] fatalPtr - Shared pointer to a FatalCperRecord.
+     * @param[in] socNum - The SoC number for which the trace logs are
+     * harvested.
+     * @param[out] mpList - List of MP's for which trace data is collected.
+     * @param[in] baseSectionCount - The base section count used for cper offset
+     * calculation
+     */
+
+    void harvestMpxTraceLog(const std::shared_ptr<FatalCperRecord>&, uint8_t,
+                            std::vector<std::string>&, uint8_t);    
+
     /** @brief Harvests debug log ID dump data
      *
      * @details This function harvests the debug log ID data for the list

include/base_manager.hpp

@@ -3,6 +3,8 @@
 #include "config_manager.hpp"
 #include "oem_cper.hpp"
 
+#include <map>
+
 constexpr size_t socket1 = 1;
 constexpr size_t socket2 = 2;
 
@@ -71,6 +73,7 @@ class Manager
     std::shared_ptr<McaRuntimeCperRecord> mcaPtr;
     std::shared_ptr<McaRuntimeCperRecord> dramPtr;
     std::shared_ptr<PcieRuntimeCperRecord> pciePtr;
+    std::vector<std::pair<std::string, int>> mpToIndexMap;
     std::string node;
     std::vector<size_t> socIndex;
 

include/config_manager.hpp

@@ -18,6 +18,14 @@ namespace config
 static constexpr auto service = "com.amd.RAS";
 static constexpr auto objectPath = "/com/amd/RAS";
 
+constexpr size_t maxErrorIndex = 256;
+constexpr auto errorCountFile = "/var/lib/amd-bmc-ras/error_count.json";
+
+extern std::array<uint64_t, maxErrorIndex> correctableCPUErrors;
+extern std::array<uint64_t, maxErrorIndex> noncorrectableCPUErrors;
+extern std::array<uint64_t, maxErrorIndex> correctableOtherErrors;
+extern std::array<uint64_t, maxErrorIndex> noncorrectableOtherErrors;
+
 using ConfigIface = sdbusplus::server::object_t<
     sdbusplus::com::amd::RAS::server::Configuration,
     sdbusplus::xyz::openbmc_project::Collection::server::DeleteAll>;
@@ -100,6 +108,10 @@ class Manager : public amd::ras::config::ConfigIface
      */
     void deleteAll() override;
 
+    void loadErrorCounts();
+
+    void saveErrorCounts();
+
   private:
     sdbusplus::asio::object_server& objServer;
     std::shared_ptr<sdbusplus::asio::connection>& systemBus;

include/oem_cper.hpp

@@ -1,5 +1,7 @@
 #pragma once
 
+#include <fstream>
+
 extern "C"
 {
 #include "libcper/Cper.h"
@@ -12,6 +14,7 @@ constexpr size_t length8 = 8;
 constexpr size_t length32 = 32;
 constexpr size_t length96 = 96;
 constexpr size_t length91 = 91;
+constexpr size_t traceBufferDataLen = 2048;
 
 struct CrashdumpData
 {
@@ -51,6 +54,13 @@ struct AmdFatalErrorData
 
 using EFI_AMD_FATAL_ERROR_DATA = AmdFatalErrorData;
 
+struct AmdMpTraceBufferData
+{
+    uint32_t TracelogData[traceBufferDataLen];
+} __attribute__((packed));
+
+using EFI_AMD_MP_TRACELOG_DATA = AmdMpTraceBufferData;
+
 struct RuntimeErrorInfo
 {
     EFI_IA32_X64_PROCESSOR_ERROR_RECORD ProcError;
@@ -66,6 +76,11 @@ struct FatalCperRecord
     EFI_COMMON_ERROR_RECORD_HEADER Header;
     EFI_ERROR_SECTION_DESCRIPTOR* SectionDescriptor;
     EFI_AMD_FATAL_ERROR_DATA* ErrorRecord;
+    EFI_AMD_MP_TRACELOG_DATA* TraceBufferRecord;
+    FatalCperRecord() :
+        SectionDescriptor(nullptr), ErrorRecord(nullptr),
+        TraceBufferRecord(nullptr)
+    {}    
 } __attribute__((packed));
 
 struct McaRuntimeCperRecord

include/tbai_manager.hpp

@@ -0,0 +1,90 @@
+#pragma once
+
+#include "crashdump_manager.hpp"
+#include "oem_cper.hpp"
+
+#include <com/amd/Dump/Create/server.hpp>
+#include <sdbusplus/asio/object_server.hpp>
+#include <sdbusplus/server.hpp>
+#include <xyz/openbmc_project/Dump/Create/server.hpp>
+
+namespace amd
+{
+namespace ras
+{
+namespace tbai
+{
+static constexpr auto service = "com.amd.RAS";
+static constexpr auto objectPath = "/com/amd/TBAI";
+
+using createDumpIface = sdbusplus::server::object_t<
+    sdbusplus::xyz::openbmc_project::Dump::server::Create,
+    sdbusplus::com::amd::Dump::server::Create>;
+
+using DumpCreateParams =
+    std::map<std::string, std::variant<std::string, uint64_t>>;
+
+/**
+ * @brief Manager class which adds the RAS configuration
+ * parameter values to the D-Bus interface.
+ *
+ * @details The class pulls the default values of ras_config.json file
+ * into the D-Bus interface and overrides the getAttribute()
+ * and setAttribute() of the RAS configuration interface.
+ *
+ *  @param[in] manager - Reference to the TBAI manager.
+ *  @param[in] objectServer - The D-Bus object server.
+ *  @param[in] systemBus - Shared pointer to the D-Bus system bus connection.
+ *  @param[in] io - Boost ASIO I/O context for asynchronous operations.
+ *  @param[in] node - host node number to determine single or multi host.
+ */
+class Manager : virtual public createDumpIface
+{
+  public:
+    Manager() = delete;
+    Manager(const Manager&) = delete;
+    Manager& operator=(const Manager&) = delete;
+    Manager(Manager&&) = delete;
+    Manager& operator=(Manager&&) = delete;
+    ~Manager() = default;
+
+    Manager(sdbusplus::asio::object_server&,
+            std::shared_ptr<sdbusplus::asio::connection>&,
+            boost::asio::io_context&, std::string&);
+
+    /** @brief Implementation for CreateDump
+     *  Method to create Dump.
+     *
+     *  @return object_path - The object path of the new entry.
+     */
+    sdbusplus::message::object_path
+        createDump(DumpCreateParams params) override;
+
+  private:
+    sdbusplus::asio::object_server& objServer;
+    std::shared_ptr<sdbusplus::asio::connection>& systemBus;
+    boost::asio::io_context& io;
+
+    std::string node;
+    std::string mpName;
+
+    std::mutex tbaiMutex;
+
+    std::vector<std::pair<std::string, int>> mpToIndexMap;
+
+    size_t currentSectionCount;
+    size_t inputSocNum;
+    size_t cpuCount;
+
+    std::vector<size_t> socIndex;
+
+    std::shared_ptr<FatalCperRecord> mpxPtr;
+
+    std::map<int, std::unique_ptr<CrashdumpInterface>> tbaiRecordMgr;
+
+    void harvestMpxTraceLog(DumpCreateParams params);
+};
+
+} // namespace tbai
+} // namespace ras
+} // namespace amd

include/utils/cper.hpp

@@ -6,6 +6,7 @@ static constexpr std::string_view runtimeMcaErr = "RUNTIME_MCA_ERROR";
 static constexpr std::string_view runtimePcieErr = "RUNTIME_PCIE_ERROR";
 static constexpr std::string_view runtimeDramErr = "RUNTIME_DRAM_ERROR";
 static constexpr std::string_view fatalErr = "FATAL";
+static constexpr std::string_view mpxTracelog = "MPX_TRACE_LOG";
 
 namespace amd
 {
@@ -25,6 +26,8 @@ constexpr uint8_t addcGenNumber3 = 0x03;
 constexpr uint8_t familyId1ah = 0x1A;
 constexpr uint16_t pcieVendorId = 0x1022;
 constexpr uint8_t minorRevision = 0xB;
+constexpr size_t maxByte = 0xFF;
+constexpr size_t quadBit = 4;
 
 /** @brief Finds a filename in the RAS directory that matches a given pattern.
  *
@@ -194,7 +197,7 @@ void dumpContext(const std::shared_ptr<FatalCperRecord>&, uint16_t numbanks,
  */
 template <typename T>
 void createFile(const std::shared_ptr<T>&, const std::string_view&, uint16_t,
-                size_t&, const std::string&);
+                size_t&, const std::string&, const std::string& tbaiFileName = "");
 
 /** @brief Checks if the signature ID matches the configuration list.
  *

include/utils/util.hpp

@@ -134,6 +134,10 @@ ReturnType getProperty(sdbusplus::bus::bus&, const char*, const char*,
  */
 bool checkObjPath(std::string);
 
+void getCpuCount(size_t&, std::string&, std::vector<size_t>&);
+
+void mpTraceLogInfo(std::vector<std::pair<std::string, int>>&);
+
 } // namespace util
 } // namespace ras
 } // namespace amd