Push 2023 10 17

README.md

@@ -50,7 +50,7 @@ Graphical front ends:
 - **pcm-sensor** :  front-end for KDE KSysGuard
 - **pcm-service** :  front-end for Windows perfmon
 
-There are also utilities for reading/writing model specific registers (**pcm-msr**), PCI configuration registers (**pcm-pcicfg**) and memory mapped registers (**pcm-mmio**) supported on Linux, Windows, Mac OS X and FreeBSD.
+There are also utilities for reading/writing model specific registers (**pcm-msr**), PCI configuration registers (**pcm-pcicfg**), memory mapped registers (**pcm-mmio**) and TPMI registers (**pcm-tpmi**) supported on Linux, Windows, Mac OS X and FreeBSD.
 
 And finally a daemon that stores core, memory and QPI counters in shared memory that can be be accessed by non-root users.
 
@@ -87,7 +87,7 @@ Debug is default on Windows. Specify config to build Release:
 ```
 cmake --build . --config Release
 ```
-On Windows and MacOs additional drivers are required. Please find instructions here: [WINDOWS_HOWTO.md](doc/WINDOWS_HOWTO.md) and [MAC_HOWTO.txt](doc/MAC_HOWTO.txt).
+On Windows and MacOs additional drivers and steps are required. Please find instructions here: [WINDOWS_HOWTO.md](doc/WINDOWS_HOWTO.md) and [MAC_HOWTO.txt](doc/MAC_HOWTO.txt).
 
 FreeBSD/DragonFlyBSD-specific details can be found in [FREEBSD_HOWTO.txt](doc/FREEBSD_HOWTO.txt)
 
@@ -103,7 +103,7 @@ Downloading Pre-Compiled PCM Tools
   * RHEL8.5 or later: `sudo dnf install pcm` 
   * Fedora: `sudo yum install pcm`
   * RPMs and DEBs with the *latest* PCM version for RHEL/SLE/Ubuntu/Debian/openSUSE/etc distributions (binary and source) are available [here](https://software.opensuse.org/download/package?package=pcm&project=home%3Aopcm)
-- Windows: download PCM binaries as [appveyor build service](https://ci.appveyor.com/project/opcm/pcm/history) artifacts and required Visual C++ Redistributable from [www.microsoft.com](https://www.microsoft.com/en-us/download/details.aspx?id=48145). Additional drivers are needed, see [WINDOWS_HOWTO.md](doc/WINDOWS_HOWTO.md).
+- Windows: download PCM binaries as [appveyor build service](https://ci.appveyor.com/project/opcm/pcm/history) artifacts and required Visual C++ Redistributable from [www.microsoft.com](https://www.microsoft.com/en-us/download/details.aspx?id=48145). Additional steps and drivers are required, see [WINDOWS_HOWTO.md](doc/WINDOWS_HOWTO.md).
 - Docker: see [instructions on how to use pcm-sensor-server pre-compiled container from docker hub](doc/DOCKER_README.md).
 
 --------------------------------------------------------------------------------

doc/WINDOWS_HOWTO.md

@@ -13,11 +13,11 @@ _For support of systems with more than _**_64_**_ logical cores you need to comp
    ```
    alternatively you can perform `cmake -B build`, open *PCM.sln* form *build* folder in and build required project in Visual Studio.
    .exe and .dll files will be located in *build\bin\Release* folder
-3. Copy the msr.sys driver and pcm.exe into a single directory
+3. As Administrator create PCM directory in Windows "Program Files" directory (e.g. `C:\Program Files (x86)\PCM\`)
+4. As Administrator copy the msr.sys driver and pcm.exe into the PCM directory
+5. Run pcm.exe utility from the PCM directory as Administrator
 
-4. Run pcm.exe utility from this directory
-
-For Windows 7 and Windows Server 2008 R2 the PCM utilities need to be run as administrator:
+For Windows 7+ and Windows Server 2008+ R2 the PCM utilities need to be run as administrator:
 
 Alternatively you can achieve the same using the “Properties” Windows menu of the executable (“Privilege level” setting in the “Compatibility” tab): Right mouse click -> Properties -> Compatibility -> Privilege level -> Set “Run this program as an administrator”.
 
@@ -35,7 +35,7 @@ If you are getting the error `Starting MSR service failed with error 3 The syste
 
 4. Build 'PCM-Service.exe' using Microsoft Visual Studio or cmake
 
-5. Copy PCM-Service.exe, PCM-Service.exe.config, and pcm-lib.dll files into a single directory
+5. Copy PCM-Service.exe, PCM-Service.exe.config, and pcm-lib.dll files into the PCM sub-directory in Windows "Program Files" directory (see above)
 
 The config file enables support for legacy security policy. Without this configuration switch, you will get an exception like this:
 

src/CMakeLists.txt

@@ -3,7 +3,7 @@
 
 
 # All pcm-* executables
-set(PROJECT_NAMES pcm pcm-numa pcm-latency pcm-power pcm-msr pcm-memory pcm-tsx pcm-pcie pcm-core pcm-iio pcm-lspci pcm-pcicfg pcm-mmio pcm-raw pcm-accel)
+set(PROJECT_NAMES pcm pcm-numa pcm-latency pcm-power pcm-msr pcm-memory pcm-tsx pcm-pcie pcm-core pcm-iio pcm-lspci pcm-pcicfg pcm-mmio pcm-tpmi pcm-raw pcm-accel)
 
 file(GLOB COMMON_SOURCES msr.cpp cpucounters.cpp pci.cpp mmio.cpp bw.cpp utils.cpp topology.cpp debug.cpp threadpool.cpp uncore_pmu_discovery.cpp)
 
@@ -54,26 +54,22 @@ if(MSVC)
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /O2 /wd4251 /wd4273 /EHa /Zi")
     add_definitions(/W3)
 
-    # https://cmake.org/cmake/help/latest/prop_tgt/MSVC_RUNTIME_LIBRARY.html
-    # windows/restrictDriverAccess.cpp is building separately
-    add_library(restrictDriverAccess OBJECT windows/restrictDriverAccess.cpp)
-    set_target_properties(restrictDriverAccess PROPERTIES COMMON_LANGUAGE_RUNTIME "")
-    target_compile_definitions(restrictDriverAccess PRIVATE _CONSOLE _UNICODE UNICODE)
-
-    # Rest of windows/* files + restrictDriverAccess.cpp -> PCM_STATIC
+    # windows/* files -> PCM_STATIC
     file(GLOB WINDOWS_SOURCES winpmem/winpmem.cpp windows/stdafx.cpp freegetopt/getopt.cpp)
-    add_library(PCM_STATIC STATIC $<TARGET_OBJECTS:restrictDriverAccess> ${COMMON_SOURCES} ${WINDOWS_SOURCES})
+    add_library(PCM_STATIC STATIC ${COMMON_SOURCES} ${WINDOWS_SOURCES})
     target_compile_definitions(PCM_STATIC PRIVATE UNICODE _UNICODE _CONSOLE)
+    target_compile_options(PCM_STATIC PRIVATE "/MT$<$<CONFIG:Debug>:d>")
 
     # Graphical perfmon front-end: pcm-lib, pcm-service
     # Files: COMMON_FILES() + pcm-lib.cpp winpmem\winpmem.cpp dllmain.cpp
     file(GLOB PCM_LIB_SOURCES winpmem/winpmem.cpp dllmain.cpp pcm-lib.cpp )
-    add_library(pcm-lib SHARED $<TARGET_OBJECTS:restrictDriverAccess> ${COMMON_SOURCES} ${PCM_LIB_SOURCES})
+    add_library(pcm-lib SHARED ${COMMON_SOURCES} ${PCM_LIB_SOURCES})
     target_compile_definitions(pcm-lib PRIVATE _WINDOWS _USRDLL PCM_EXPORTS _WINDLL _UNICODE UNICODE)
+    target_compile_options(pcm-lib PRIVATE "/MT$<$<CONFIG:Debug>:d>")
 
     # Pcm-service files: PCM_SHARED + AssemblyInfo.cpp PCMInstaller.cpp PCMService.cpp
     file(GLOB PCM_SERVICE_SOURCES windows/PCMInstaller.cpp  windows/PCMService.cpp windows/AssemblyInfo.cpp winddows/utils.cpp)
-    add_executable(pcm-service $<TARGET_OBJECTS:restrictDriverAccess> ${PCM_SERVICE_SOURCES})
+    add_executable(pcm-service ${PCM_SERVICE_SOURCES})
     target_compile_definitions(pcm-service PRIVATE _UNICODE UNICODE _CONSOLE)
     set_target_properties(pcm-service PROPERTIES LINK_FLAGS "/INCREMENTAL:NO" COMMON_LANGUAGE_RUNTIME "")
     set_property(TARGET pcm-service PROPERTY VS_DOTNET_REFERENCES "System;System.Configuration.Install;System.Data;System.Management;System.ServiceProcess;System.Xml")
@@ -134,6 +130,10 @@ foreach(PROJECT_NAME ${PROJECT_NAMES})
 
     add_executable(${PROJECT_NAME} ${PROJECT_FILE})
 
+    if(MSVC)
+        target_compile_options(${PROJECT_NAME} PRIVATE "/MT$<$<CONFIG:Debug>:d>")
+    endif(MSVC)
+
     # specific file for pcm-raw project
     if(${PROJECT_NAME} STREQUAL pcm-raw)
         set(LIBS ${LIBS} PCM_SIMDJSON)

src/cpucounters.cpp

@@ -89,6 +89,11 @@ int convertUnknownToInt(size_t size, char* value);
 
 #ifdef _MSC_VER
 
+void PCM_API restrictDriverAccess(LPCTSTR path)
+{
+    restrictDriverAccessNative(path);
+}
+
 HMODULE hOpenLibSys = NULL;
 
 #ifndef NO_WINRING
@@ -1984,6 +1989,19 @@ void PCM::initUncoreObjects()
     {
         initUncorePMUsDirect();
     }
+
+    std::cerr << "Info: " << uboxPMUs.size() << " UBOX units detected.\n";
+    for (uint32 s = 0; s < (uint32)num_sockets; ++s)
+    {
+        std::cerr << "Socket " << s << ":" <<
+            " " << ((s < pcuPMUs.size()) ? pcuPMUs[s].size() : 0) << " PCU units detected."
+            " " << ((s < iioPMUs.size()) ? iioPMUs[s].size() : 0) << " IIO units detected."
+            " " << ((s < irpPMUs.size()) ? irpPMUs[s].size() : 0) << " IRP units detected."
+            " " << ((s < cboPMUs.size()) ? cboPMUs[s].size() : 0) << " CHA/CBO units detected."
+            " " << ((s < mdfPMUs.size()) ? mdfPMUs[s].size() : 0) << " MDF units detected."
+            " " << ((s < cxlPMUs.size()) ? cxlPMUs[s].size() : 0) << " CXL units detected."
+            "\n";
+    }
 }
 
 void PCM::globalFreezeUncoreCounters()
@@ -2026,6 +2044,7 @@ void PCM::globalFreezeUncoreCountersInternal(const unsigned long long int freeze
 
 void PCM::initUncorePMUsDirect()
 {
+    pcuPMUs.resize(num_sockets);
     for (uint32 s = 0; s < (uint32)num_sockets; ++s)
     {
         auto & handle = MSR[socketRefCore[s]];
@@ -2121,7 +2140,7 @@ void PCM::initUncorePMUsDirect()
         {
         case IVYTOWN:
         case JAKETOWN:
-            pcuPMUs.push_back(
+            pcuPMUs[s].push_back(
                 UncorePMU(
                     std::make_shared<MSRRegister>(handle, JKTIVT_PCU_MSR_PMON_BOX_CTL_ADDR),
                     std::make_shared<MSRRegister>(handle, JKTIVT_PCU_MSR_PMON_CTL0_ADDR),
@@ -2144,7 +2163,7 @@ void PCM::initUncorePMUsDirect()
         case HASWELLX:
         case SKX:
         case ICX:
-            pcuPMUs.push_back(
+            pcuPMUs[s].push_back(
                 UncorePMU(
                     std::make_shared<MSRRegister>(handle, HSX_PCU_MSR_PMON_BOX_CTL_ADDR),
                     std::make_shared<MSRRegister>(handle, HSX_PCU_MSR_PMON_CTL0_ADDR),
@@ -2162,13 +2181,14 @@ void PCM::initUncorePMUsDirect()
             );
             break;
         case SPR:
-            addPMUsFromDiscovery(pcuPMUs, SPR_PCU_BOX_TYPE, 0xE);
-            if (pcuPMUs.empty())
+            addPMUsFromDiscovery(pcuPMUs[s], SPR_PCU_BOX_TYPE, 0xE);
+            if (pcuPMUs[s].empty())
             {
                 std::cerr << "ERROR: PCU PMU not found\n";
             }
             break;
         }
+        assert(pcuPMUs[s].size() <= ServerUncoreCounterState::maxPUnits);
 
         // add MDF PMUs
         switch (cpu_model)
@@ -2185,9 +2205,10 @@ void PCM::initUncorePMUsDirect()
     }
 
     // init IIO addresses
-    if (getCPUModel() == PCM::SKX)
+    iioPMUs.resize(num_sockets);
+    switch (getCPUModel())
     {
-        iioPMUs.resize(num_sockets);
+    case PCM::SKX:
         for (uint32 s = 0; s < (uint32)num_sockets; ++s)
         {
             auto & handle = MSR[socketRefCore[s]];
@@ -2206,10 +2227,8 @@ void PCM::initUncorePMUsDirect()
                 );
             }
         }
-    }
-    else if (getCPUModel() == PCM::ICX)
-    {
-        iioPMUs.resize(num_sockets);
+        break;
+    case PCM::ICX:
         for (uint32 s = 0; s < (uint32)num_sockets; ++s)
         {
             auto & handle = MSR[socketRefCore[s]];
@@ -2228,10 +2247,8 @@ void PCM::initUncorePMUsDirect()
                 );
             }
         }
-    }
-    else if (getCPUModel() == PCM::SNOWRIDGE)
-    {
-        iioPMUs.resize(num_sockets);
+        break;
+    case PCM::SNOWRIDGE:
         for (uint32 s = 0; s < (uint32)num_sockets; ++s)
         {
             auto & handle = MSR[socketRefCore[s]];
@@ -2250,11 +2267,8 @@ void PCM::initUncorePMUsDirect()
                 );
             }
         }
-    }
-
-    if (getCPUModel() == PCM::SPR)
-    {
-        iioPMUs.resize(num_sockets);
+        break;
+    case PCM::SPR:
         for (uint32 s = 0; s < (uint32)num_sockets; ++s)
         {
             auto & handle = MSR[socketRefCore[s]];
@@ -2273,9 +2287,9 @@ void PCM::initUncorePMUsDirect()
                 );
             }
         }
+        break;
     }
 
-
     //init the IDX accelerator
     auto createIDXPMU = [](const size_t addr, const size_t mapSize, const size_t numaNode, const size_t socketId) -> IDX_PMU
     {
@@ -2562,9 +2576,10 @@ void PCM::initUncorePMUsPerf()
     irpPMUs.resize(num_sockets);
     cboPMUs.resize(num_sockets);
     mdfPMUs.resize(num_sockets);
+    pcuPMUs.resize(num_sockets);
     for (uint32 s = 0; s < (uint32)num_sockets; ++s)
     {
-        populatePerfPMUs(s, enumeratePerfPMUs("pcu", 100), pcuPMUs, false, true);
+        populatePerfPMUs(s, enumeratePerfPMUs("pcu", 100), pcuPMUs[s], false, true);
         populatePerfPMUs(s, enumeratePerfPMUs("ubox", 100), uboxPMUs, true);
         populatePerfPMUs(s, enumeratePerfPMUs("cbox", 100), cboPMUs[s], false, true, true);
         populatePerfPMUs(s, enumeratePerfPMUs("cha", 200), cboPMUs[s], false, true, true);
@@ -4673,9 +4688,12 @@ void PCM::cleanupUncorePMUs(const bool silent)
             pmu.cleanup();
         }
     }
-    for (auto & pmu : pcuPMUs)
+    for (auto& spcuPMUs : pcuPMUs)
     {
-        pmu.cleanup();
+        for (auto& pmu : spcuPMUs)
+        {
+            pmu.cleanup();
+        }
     }
     for (auto& sPMUs : cxlPMUs)
     {
@@ -5418,14 +5436,17 @@ void PCM::programPCU(uint32* PCUCntConf, const uint64 filter)
         uint32 refCore = socketRefCore[i];
         TemporalThreadAffinity tempThreadAffinity(refCore); // speedup trick for Linux
 
-        pcuPMUs[i].initFreeze(UNC_PMON_UNIT_CTL_FRZ_EN);
-
-        if (pcuPMUs[i].filter[0].get())
+        for (auto& pmu : pcuPMUs[i])
         {
-            *pcuPMUs[i].filter[0] = filter;
-        }
+            pmu.initFreeze(UNC_PMON_UNIT_CTL_FRZ_EN);
 
-        program(pcuPMUs[i], &PCUCntConf[0], &PCUCntConf[4], UNC_PMON_UNIT_CTL_FRZ_EN);
+            if (pmu.filter[0].get())
+            {
+                *pmu.filter[0] = filter;
+            }
+
+            program(pmu, &PCUCntConf[0], &PCUCntConf[4], UNC_PMON_UNIT_CTL_FRZ_EN);
+        }
     }
 }
 
@@ -5731,7 +5752,10 @@ void PCM::freezeServerUncoreCounters()
         const auto refCore = socketRefCore[i];
         TemporalThreadAffinity tempThreadAffinity(refCore); // speedup trick for Linux
 
-        pcuPMUs[i].freeze(UNC_PMON_UNIT_CTL_FRZ_EN);
+        for (auto& pmu : pcuPMUs[i])
+        {
+            pmu.freeze(UNC_PMON_UNIT_CTL_FRZ_EN);
+        }
 
         if (IIOEventsAvailable())
         {
@@ -5783,7 +5807,10 @@ void PCM::unfreezeServerUncoreCounters()
         const auto refCore = socketRefCore[i];
         TemporalThreadAffinity tempThreadAffinity(refCore); // speedup trick for Linux
 
-        pcuPMUs[i].unfreeze(UNC_PMON_UNIT_CTL_FRZ_EN);
+        for (auto& pmu : pcuPMUs[i])
+        {
+            pmu.unfreeze(UNC_PMON_UNIT_CTL_FRZ_EN);
+        }
 
         if (IIOEventsAvailable())
         {
@@ -6588,9 +6615,13 @@ ServerUncoreCounterState PCM::getServerUncoreCounterState(uint32 socket)
             result.UBOXCounter[i] = *(uboxPMUs[socket].counterValue[i]);
             result.UncClocks = getUncoreClocks(socket);
         }
-        for (int i = 0; i < ServerUncoreCounterState::maxCounters && socket < pcuPMUs.size() && size_t(i) < pcuPMUs[socket].size(); ++i)
+        for (size_t u = 0; socket < pcuPMUs.size() && u < pcuPMUs[socket].size(); ++u)
         {
-            result.PCUCounter[i] = *pcuPMUs[socket].counterValue[i];
+            for (int i = 0; i < ServerUncoreCounterState::maxCounters && size_t(i) < pcuPMUs[socket][u].size(); ++i)
+            {
+                assert(u < ServerUncoreCounterState::maxPUnits);
+                result.PCUCounter[u][i] = *pcuPMUs[socket][u].counterValue[i];
+            }
         }
         for (size_t p = 0; p < getNumCXLPorts(socket); ++p)
         {