push sheeet

pkgs/by-name/dc/dcgm/dynamic-libs.patch

@@ -0,0 +1,40 @@
+diff --git a/cmake/FindJsoncpp.cmake b/cmake/FindJsoncpp.cmake
+index abedf7bb85..d7a52f5c7b 100644
+--- a/cmake/FindJsoncpp.cmake
++++ b/cmake/FindJsoncpp.cmake
+@@ -16,8 +16,8 @@
+ 
+ if (NOT TARGET JsonCpp::JsonCpp)
+     find_package(jsoncpp REQUIRED CONFIG)
+-    set(JSONCPP_STATIC_LIBS jsoncpp_static)
+-    set(JSONCPP_INCLUDE_PATH $<TARGET_PROPERTY:jsoncpp_static,INTERFACE_INCLUDE_DIRECTORIES>)
++    set(JSONCPP_STATIC_LIBS jsoncpp)
++    set(JSONCPP_INCLUDE_PATH $<TARGET_PROPERTY:jsoncpp,INTERFACE_INCLUDE_DIRECTORIES>)
+ endif()
+ # set(Jsoncpp_PATH_PREFIXES /usr/local "${Jsoncpp_ROOT}" "$ENV{HOME}")
+ # foreach(prefix ${Jsoncpp_PATH_PREFIXES})
+diff --git a/cmake/FindLibevent.cmake b/cmake/FindLibevent.cmake
+index 354d6f9b99..38aca9534e 100644
+--- a/cmake/FindLibevent.cmake
++++ b/cmake/FindLibevent.cmake
+@@ -26,16 +26,12 @@
+ endforeach()
+ 
+ find_path(LIBEVENT_INCLUDE_DIR evhttp.h event.h PATHS ${Libevent_INCLUDE_PATHS})
+-find_library(LIBEVENT_STATIC_LIB NAMES libevent.a libevent_core.a libevent_extra.a PATHS ${Libevent_LIB_PATHS})
+-find_library(LIBEVENT_PTHREAD_STATIC_LIB NAMES libevent_pthreads.a PATHS ${Libevent_LIB_PATHS})
++find_library(LIBEVENT_STATIC_LIB NAMES libevent.so libevent_core.so libevent_extra.so PATHS ${Libevent_LIB_PATHS})
++find_library(LIBEVENT_PTHREAD_STATIC_LIB NAMES libevent_pthreads.so PATHS ${Libevent_LIB_PATHS})
+ 
+-if (LIBEVENT_INCLUDE_DIR AND LIBEVENT_STATIC_LIB AND LIBEVENT_PTHREAD_STATIC_LIB)
++if (LIBEVENT_INCLUDE_DIR AND LIBEVENT_STATIC_LIB)
+     set(Libevent_FOUND TRUE)
+-    add_library(libevent_event_static STATIC IMPORTED)
+-    set_target_properties(libevent_event_static PROPERTIES IMPORTED_LOCATION ${LIBEVENT_STATIC_LIB})
+-    add_library(libevent_event_pthread STATIC IMPORTED)
+-    set_target_properties(libevent_event_pthread PROPERTIES IMPORTED_LOCATION ${LIBEVENT_PTHREAD_STATIC_LIB})
+-    set(LIBEVENT_STATIC_LIBS libevent_event_static libevent_event_pthread)
++    set(LIBEVENT_STATIC_LIBS ${LIBEVENT_STATIC_LIB} ${LIBEVENT_PTHREAD_STATIC_LIB})
+ else ()
+     set(Libevent_FOUND FALSE)
+ endif ()

pkgs/by-name/dc/dcgm/fix-paths.patch

@@ -0,0 +1,252 @@
+diff --git a/common/LsHw.cpp b/common/LsHw.cpp
+index 8d0f35cd89..12fe26957f 100644
+--- a/common/LsHw.cpp
++++ b/common/LsHw.cpp
+@@ -145,7 +145,7 @@
+ {
+     static std::string const cmd = "lshw -json";
+     std::string cmdOutput;
+-    static std::array<std::string, 2> const cmdPathPrefix { "/usr/bin/", "/usr/sbin/" };
++    static std::array<std::string, 1> const cmdPathPrefix { "@lshw@/bin/" };
+ 
+     dcgmReturn_t result = DCGM_ST_OK;
+     for (auto const &prefix : cmdPathPrefix)
+@@ -162,4 +162,4 @@
+     }
+ 
+     return cmdOutput;
+-}
+\ No newline at end of file
++}
+diff --git a/common/tests/LsHwTests.cpp b/common/tests/LsHwTests.cpp
+index edb65e599e..5950fb8d9e 100644
+--- a/common/tests/LsHwTests.cpp
++++ b/common/tests/LsHwTests.cpp
+@@ -609,7 +609,7 @@
+         std::unique_ptr<MockRunningUserChecker> checker = std::make_unique<MockRunningUserChecker>();
+         checker->MockIsRoot(true);
+         std::unique_ptr<MockRunCmdHelper> runCmdHelper = std::make_unique<MockRunCmdHelper>();
+-        runCmdHelper->MockCmdOutput("/usr/bin/lshw -json", DCGM_ST_OK, lshwMultipleCpusAbridgedValidJson);
++        runCmdHelper->MockCmdOutput("@lshw@/bin/lshw -json", DCGM_ST_OK, lshwMultipleCpusAbridgedValidJson);
+ 
+         LsHw lshw;
+         lshw.SetChecker(std::move(checker));
+@@ -626,7 +626,7 @@
+         std::unique_ptr<MockRunningUserChecker> checker = std::make_unique<MockRunningUserChecker>();
+         checker->MockIsRoot(true);
+         std::unique_ptr<MockRunCmdHelper> runCmdHelper = std::make_unique<MockRunCmdHelper>();
+-        runCmdHelper->MockCmdOutput("/usr/bin/lshw -json", DCGM_ST_OK, lshwSingleCpuAbridgedValidJson);
++        runCmdHelper->MockCmdOutput("@lshw@/bin/lshw -json", DCGM_ST_OK, lshwSingleCpuAbridgedValidJson);
+ 
+         LsHw lshw;
+         lshw.SetChecker(std::move(checker));
+@@ -642,7 +642,7 @@
+         std::unique_ptr<MockRunningUserChecker> checker = std::make_unique<MockRunningUserChecker>();
+         checker->MockIsRoot(true);
+         std::unique_ptr<MockRunCmdHelper> runCmdHelper = std::make_unique<MockRunCmdHelper>();
+-        runCmdHelper->MockCmdOutput("/usr/bin/lshw -json", DCGM_ST_OK, lshwSingleCpuNoSerialNumberAbridgedValidJson);
++        runCmdHelper->MockCmdOutput("@lshw@/bin/lshw -json", DCGM_ST_OK, lshwSingleCpuNoSerialNumberAbridgedValidJson);
+ 
+         LsHw lshw;
+         lshw.SetChecker(std::move(checker));
+@@ -658,7 +658,7 @@
+         std::unique_ptr<MockRunningUserChecker> checker = std::make_unique<MockRunningUserChecker>();
+         checker->MockIsRoot(true);
+         std::unique_ptr<MockRunCmdHelper> runCmdHelper = std::make_unique<MockRunCmdHelper>();
+-        runCmdHelper->MockCmdOutput("/usr/bin/lshw -json", DCGM_ST_OK, lshwSingleNonNvidiaCpuAbridgedValidJson);
++        runCmdHelper->MockCmdOutput("@lshw@/bin/lshw -json", DCGM_ST_OK, lshwSingleNonNvidiaCpuAbridgedValidJson);
+ 
+         LsHw lshw;
+         lshw.SetChecker(std::move(checker));
+@@ -673,7 +673,7 @@
+         std::unique_ptr<MockRunningUserChecker> checker = std::make_unique<MockRunningUserChecker>();
+         checker->MockIsRoot(true);
+         std::unique_ptr<MockRunCmdHelper> runCmdHelper = std::make_unique<MockRunCmdHelper>();
+-        runCmdHelper->MockCmdOutput("/usr/bin/lshw -json", DCGM_ST_OK, lshwIncorrectIdValueTypeJson);
++        runCmdHelper->MockCmdOutput("@lshw@/bin/lshw -json", DCGM_ST_OK, lshwIncorrectIdValueTypeJson);
+ 
+         LsHw lshw;
+         lshw.SetChecker(std::move(checker));
+@@ -687,7 +687,7 @@
+         std::unique_ptr<MockRunningUserChecker> checker = std::make_unique<MockRunningUserChecker>();
+         checker->MockIsRoot(true);
+         std::unique_ptr<MockRunCmdHelper> runCmdHelper = std::make_unique<MockRunCmdHelper>();
+-        runCmdHelper->MockCmdOutput("/usr/bin/lshw -json", DCGM_ST_OK, lshwMissingCpuJson);
++        runCmdHelper->MockCmdOutput("@lshw@/bin/lshw -json", DCGM_ST_OK, lshwMissingCpuJson);
+ 
+         LsHw lshw;
+         lshw.SetChecker(std::move(checker));
+@@ -702,7 +702,7 @@
+         std::unique_ptr<MockRunningUserChecker> checker = std::make_unique<MockRunningUserChecker>();
+         checker->MockIsRoot(true);
+         std::unique_ptr<MockRunCmdHelper> runCmdHelper = std::make_unique<MockRunCmdHelper>();
+-        runCmdHelper->MockCmdOutput("/usr/bin/lshw -json", DCGM_ST_OK, lshwBadSyntaxJson);
++        runCmdHelper->MockCmdOutput("@lshw@/bin/lshw -json", DCGM_ST_OK, lshwBadSyntaxJson);
+ 
+         LsHw lshw;
+         lshw.SetChecker(std::move(checker));
+@@ -716,6 +716,7 @@
+ {
+     SECTION("Will try /usr/sbin/")
+     {
++        SKIP("Nixpkgs patches this out");
+         std::unique_ptr<MockRunningUserChecker> checker = std::make_unique<MockRunningUserChecker>();
+         checker->MockIsRoot(true);
+         std::unique_ptr<MockRunCmdHelper> runCmdHelper = std::make_unique<MockRunCmdHelper>();
+@@ -730,4 +731,4 @@
+         REQUIRE(cpuSerials.value()[0] == "0x000000017820B1C80400000015FF81C0");
+         REQUIRE(cpuSerials.value()[1] == "0x000000017820B1C8040000000A0200C0");
+     }
+-}
+\ No newline at end of file
++}
+diff --git a/modules/diag/DcgmDiagManager.cpp b/modules/diag/DcgmDiagManager.cpp
+index d0a75dcede..ccd4347719 100644
+--- a/modules/diag/DcgmDiagManager.cpp
++++ b/modules/diag/DcgmDiagManager.cpp
+@@ -253,7 +253,7 @@
+     int result;
+ 
+     // Default NVVS binary path
+-    cmd = "/usr/libexec/datacenter-gpu-manager-4/nvvs";
++    cmd = "@dcgm_out@/libexec/datacenter-gpu-manager-4/nvvs";
+ 
+     // Check for NVVS binary path enviroment variable
+     value = std::getenv("NVVS_BIN_PATH");
+diff --git a/modules/mndiag/dcgm_mndiag_structs.hpp b/modules/mndiag/dcgm_mndiag_structs.hpp
+index 40e61a8fd0..7e43ce8bed 100644
+--- a/modules/mndiag/dcgm_mndiag_structs.hpp
++++ b/modules/mndiag/dcgm_mndiag_structs.hpp
+@@ -33,8 +33,8 @@
+ constexpr std::string_view ENV_ALLOW_RUN_AS_ROOT = "DCGM_MPIRUN_ALLOW_RUN_AS_ROOT";
+ 
+ // Default paths
+-constexpr std::string_view DEFAULT_MPIRUN_PATH     = "/usr/bin/mpirun";
+-constexpr std::string_view DEFAULT_MNUBERGEMM_PATH = "/usr/libexec/datacenter-gpu-manager-4/plugins/cuda12/mnubergemm";
++constexpr std::string_view DEFAULT_MPIRUN_PATH     = "@mpi@/bin/mpirun";
++constexpr std::string_view DEFAULT_MNUBERGEMM_PATH = "@dcgm_out@/libexec/datacenter-gpu-manager-4/plugins/cuda12/mnubergemm";
+ } //namespace MnDiagConstants
+ 
+ // Message types
+diff --git a/modules/mndiag/tests/MnDiagManagerTests.cpp b/modules/mndiag/tests/MnDiagManagerTests.cpp
+index 40dbda3b72..5ec8dafa1c 100644
+--- a/modules/mndiag/tests/MnDiagManagerTests.cpp
++++ b/modules/mndiag/tests/MnDiagManagerTests.cpp
+@@ -2228,7 +2228,7 @@
+         mockStateMachine->SetMnubergemmPathCallback([&capturedPath](std::string const &path) { capturedPath = path; });
+ 
+         // Set env to custom path
+-        std::string customPath = "/bin/true";
++        std::string customPath = "@coreutils@/bin/true";
+         setenv(MnDiagConstants::ENV_MNUBERGEMM_PATH.data(), customPath.c_str(), 1);
+ 
+         auto mockCoreProxy = std::make_unique<MockDcgmCoreProxy>();
+@@ -3112,7 +3112,7 @@
+     {
+         // Save current environment state
+         auto savedPath         = saveEnvVar(MnDiagConstants::ENV_MNUBERGEMM_PATH.data());
+-        std::string customPath = "/bin/true";
++        std::string customPath = "@coreutils@/bin/true";
+         setenv(MnDiagConstants::ENV_MNUBERGEMM_PATH.data(), customPath.c_str(), 1);
+ 
+         // Setup mock DCGM API with callback to inspect request
+@@ -3251,7 +3251,7 @@
+         auto savedPath = saveEnvVar(MnDiagConstants::ENV_MNUBERGEMM_PATH.data());
+ 
+         // Use a known executable that exists
+-        std::string customPath = "/bin/true";
++        std::string customPath = "@coreutils@/bin/true";
+         setenv(MnDiagConstants::ENV_MNUBERGEMM_PATH.data(), customPath.c_str(), 1);
+ 
+         // Call the method and verify path
+diff --git a/modules/mndiag/tests/MnDiagProcessUtilsTests.cpp b/modules/mndiag/tests/MnDiagProcessUtilsTests.cpp
+index 633e327c42..168ed91db2 100644
+--- a/modules/mndiag/tests/MnDiagProcessUtilsTests.cpp
++++ b/modules/mndiag/tests/MnDiagProcessUtilsTests.cpp
+@@ -71,7 +71,7 @@
+     {
+         // Start a long-running process
+         DcgmNs::Common::Subprocess::ChildProcessBuilder builder;
+-        builder.SetExecutable("/bin/sleep").AddArg("0.5");
++        builder.SetExecutable("@coreutils@/bin/sleep").AddArg("0.5");
+ 
+         IoContext ioContext {};
+         auto process = std::make_unique<DcgmNs::Common::Subprocess::ChildProcess>(builder.Build(ioContext));
+@@ -155,4 +155,4 @@
+ 
+         REQUIRE(result.empty());
+     }
+-}
+\ No newline at end of file
++}
+diff --git a/modules/mndiag/tests/MpiRunnerTests.cpp b/modules/mndiag/tests/MpiRunnerTests.cpp
+index 526c80fd47..1e8596d3b7 100755
+--- a/modules/mndiag/tests/MpiRunnerTests.cpp
++++ b/modules/mndiag/tests/MpiRunnerTests.cpp
+@@ -60,7 +60,7 @@
+ 
+     std::string GetMpiBinPath() const override
+     {
+-        return "/bin/bash";
++        return "@shell@";
+     }
+ 
+ private:
+@@ -138,9 +138,9 @@
+         runner.ConstructMpiCommand(&config);
+ 
+         // Verify command construction
+-        REQUIRE(runner.GetMpiBinPath() == "/bin/bash");
++        REQUIRE(runner.GetMpiBinPath() == "@shell@");
+         std::string fullCommand = runner.GetLastCommand();
+-        REQUIRE(fullCommand.find("/bin/bash -c") != std::string::npos);
++        REQUIRE(fullCommand.find("@shell@ -c") != std::string::npos);
+         REQUIRE(fullCommand.find("sleep 1") != std::string::npos);
+         REQUIRE(fullCommand.find("Output from sleep process") != std::string::npos);
+ 
+@@ -244,4 +244,4 @@
+         // Test with invalid parameter (null pointer)
+         REQUIRE(runner.PopulateResponse(nullptr, nodeInfoMap_t()) == DCGM_ST_BADPARAM);
+     }
+-}
+\ No newline at end of file
++}
+diff --git a/modules/sysmon/DcgmCpuTopology.cpp b/modules/sysmon/DcgmCpuTopology.cpp
+index 786d3877fc..ccbeccc81d 100644
+--- a/modules/sysmon/DcgmCpuTopology.cpp
++++ b/modules/sysmon/DcgmCpuTopology.cpp
+@@ -136,7 +136,7 @@
+ {
+     static std::string cmd = "lscpu --json";
+     std::string cmdOutput;
+-    static std::array<std::string, 2> cmdPathPrefix = { "/usr/bin/", "/usr/sbin/" };
++    static std::array<std::string, 1> cmdPathPrefix = { "@util-linux@/bin/" };
+ 
+     dcgmReturn_t result = DCGM_ST_OK;
+     for (auto const &prefix : cmdPathPrefix)
+diff --git a/nvvs/plugin_src/nvbandwidth/NVBandwidthPlugin.cpp b/nvvs/plugin_src/nvbandwidth/NVBandwidthPlugin.cpp
+index 261bba4490..4d439ad452 100644
+--- a/nvvs/plugin_src/nvbandwidth/NVBandwidthPlugin.cpp
++++ b/nvvs/plugin_src/nvbandwidth/NVBandwidthPlugin.cpp
+@@ -238,7 +238,7 @@
+     std::vector<std::string> const search_paths
+         = { GetCurrentModuleLocation(),
+             fmt::format("./apps/nvvs/plugins/cuda{}", m_cudaDriverMajorVersion),
+-            fmt::format("/usr/libexec/datacenter-gpu-manager-4/plugins/cuda{}", m_cudaDriverMajorVersion),
++            fmt::format("@dcgm_out@/libexec/datacenter-gpu-manager-4/plugins/cuda{}", m_cudaDriverMajorVersion),
+             GetNvvsBinCheckPath(m_cudaDriverMajorVersion) };
+     std::stringstream path_buf;
+ 
+diff --git a/testing/TestDiagManager.cpp b/testing/TestDiagManager.cpp
+index 8087123a49..a6333b3a3d 100644
+--- a/testing/TestDiagManager.cpp
++++ b/testing/TestDiagManager.cpp
+@@ -236,7 +236,7 @@
+     if (nvvsPathEnv)
+         nvvsBinPath = std::string(nvvsPathEnv) + "/nvvs";
+     else
+-        nvvsBinPath = "/usr/libexec/datacenter-gpu-manager-4/nvvs";
++        nvvsBinPath = "@dcgm_out@/libexec/datacenter-gpu-manager-4/nvvs";
+ 
+     std::string diagResponseVersionArg = fmt::format("--response-version {}", dcgmDiagResponse_version12);
+     expected.push_back(nvvsBinPath + " --channel-fd 3 " + diagResponseVersionArg

pkgs/by-name/dc/dcgm/package.nix

@@ -0,0 +1,177 @@
+{
+  lib,
+  stdenv,
+  fetchFromGitHub,
+  autoAddDriverRunpath,
+  catch2_3,
+  cmake,
+  ctestCheckHook,
+  coreutils,
+  mpi,
+  mpiCheckPhaseHook,
+  ninja,
+  cudaPackages_12,
+  boost186,
+  fmt_10,
+  git,
+  jsoncpp,
+  libevent,
+  lshw,
+  plog,
+  python3,
+  replaceVars,
+  symlinkJoin,
+  tclap_1_4,
+  util-linux,
+  yaml-cpp,
+}:
+let
+  # DCGM can depend on multiple versions of CUDA at the same time.
+  # The runtime closure, thankfully, is quite small as it does not
+  # include the CUDA libraries.
+  cudaPackageSets = [
+    cudaPackages_12
+  ];
+
+  # Select needed redist packages from cudaPackages
+  # C.f. https://github.com/NVIDIA/DCGM/blob/7e1012302679e4bb7496483b32dcffb56e528c92/dcgmbuild/scripts/0080_cuda.sh#L24-L39
+  getCudaPackages =
+    p: with p; [
+      cuda_cccl
+      cuda_cudart
+      cuda_nvcc
+      cuda_nvml_dev
+      libcublas
+      libcufft
+      libcurand
+    ];
+
+  # Builds CMake flags to add CUDA paths for include and lib.
+  mkCudaFlags =
+    cudaPackages:
+    let
+      version = cudaPackages.cudaMajorVersion;
+      # The DCGM CMake assumes that the folder containing cuda.h contains all headers, so we must
+      # combine everything together for headers to work.
+      headers = symlinkJoin {
+        name = "cuda-headers-combined-${version}";
+        paths = lib.map (pkg: "${lib.getInclude pkg}/include") (getCudaPackages cudaPackages);
+      };
+    in
+    [
+      (lib.cmakeFeature "CUDA${version}_INCLUDE_DIR" "${headers}")
+      (lib.cmakeFeature "CUDA${version}_LIBS" "${cudaPackages.cuda_cudart.stubs}/lib/stubs/libcuda.so")
+      (lib.cmakeFeature "CUDA${version}_STATIC_LIBS" "${lib.getLib cudaPackages.cuda_cudart}/lib/libcudart.so")
+      (lib.cmakeFeature "CUDA${version}_STATIC_CUBLAS_LIBS" (
+        lib.concatStringsSep ";" [
+          "${lib.getLib cudaPackages.libcublas}/lib/libcublas.so"
+          "${lib.getLib cudaPackages.libcublas}/lib/libcublasLt.so"
+        ]
+      ))
+    ];
+in
+stdenv.mkDerivation {
+  pname = "dcgm";
+  version = "4.3.1"; # N.B: If you change this, be sure prometheus-dcgm-exporter supports this version.
+
+  src = fetchFromGitHub {
+    owner = "NVIDIA";
+    repo = "DCGM";
+    # No tag for 4.3.1 yet.
+    #tag = "v${version}";
+    rev = "1477d8785e899ab3450fdff2b486102e9bed096b";
+    hash = "sha256-FebqG28aodENGLNBBbiGpckzzeuP+y44dCALtYnN1yU=";
+  };
+
+  patches = [
+    ./remove-cuda-11.patch
+    ./dynamic-libs.patch
+    (replaceVars ./fix-paths.patch {
+      inherit coreutils;
+      inherit util-linux;
+      inherit lshw;
+      inherit mpi;
+      inherit (stdenv) shell;
+      dcgm_out = null;
+    })
+  ];
+
+  hardeningDisable = [ "all" ];
+
+  strictDeps = true;
+
+  nativeBuildInputs = [
+    # autoAddDriverRunpath does not actually depend on or incur any dependency
+    # of cudaPackages. It merely adds an impure, non-Nix PATH to the RPATHs of
+    # executables that need to use cuda at runtime.
+    autoAddDriverRunpath
+
+    cmake
+    ninja
+    git
+    python3
+  ];
+
+  buildInputs = [
+    # Header-only
+    boost186
+    catch2_3
+    plog.dev
+    tclap_1_4
+
+    fmt_10
+    yaml-cpp
+    jsoncpp
+    libevent
+  ];
+
+  nativeCheckInputs = [
+    mpi
+    ctestCheckHook
+    mpiCheckPhaseHook
+  ];
+
+  disabledTests = [
+    # Fail due to lack of `/sys` in the sandbox.
+    "DcgmModuleSysmon::PauseResume Module resumed after initialization"
+    "DcgmModuleSysmon PauseResume Module rejects invalid messages"
+    "DcgmModuleSysmon PauseResume Module accepts valid messages"
+    "DcgmModuleSysmon Watches"
+    "DcgmModuleSysmon maxSampleAge"
+    "DcgmModuleSysmon::CalculateCoreUtilization"
+    "DcgmModuleSysmon::ParseProcStatCpuLine"
+    "DcgmModuleSysmon::ParseThermalFileContentsAndStore"
+    "DcgmModuleSysmon::PopulateTemperatureFileMap"
+    "DcgmModuleSysmon::ReadCoreSpeed"
+    "DcgmModuleSysmon::ReadTemperature"
+    "Sysmon: initialize module"
+  ];
+
+  # Add our paths to the CMake flags so FindCuda.cmake can find them.
+  cmakeFlags = lib.concatMap mkCudaFlags cudaPackageSets;
+
+  # Lots of dodgy C++.
+  env.NIX_CFLAGS_COMPILE = "-Wno-error";
+
+  doCheck = true;
+  dontUseNinjaCheck = true;
+
+  postPatch = ''
+    while read -r -d "" file; do
+      substituteInPlace "$file" --replace-quiet @dcgm_out@ "$out"
+    done < <(find . '(' -name '*.h' -or -name '*.cpp' ')' -print0)
+  '';
+
+  disallowedReferences = lib.concatMap getCudaPackages cudaPackageSets;
+
+  __structuredAttrs = true;
+
+  meta = with lib; {
+    description = "Data Center GPU Manager (DCGM) is a daemon that allows users to monitor NVIDIA data-center GPUs";
+    homepage = "https://developer.nvidia.com/dcgm";
+    license = licenses.asl20;
+    teams = [ teams.deshaw ];
+    mainProgram = "dcgmi";
+    platforms = platforms.linux;
+  };
+}

pkgs/by-name/dc/dcgm/remove-cuda-11.patch

@@ -0,0 +1,289 @@
+diff --git a/CMakeLists.txt b/CMakeLists.txt
+index 11317ae20f..7a6b1d5b75 100755
+--- a/CMakeLists.txt
++++ b/CMakeLists.txt
+@@ -269,10 +269,8 @@
+     add_library(testing_dcgm_cublas_stubs STATIC)
+ endif()
+ 
+-add_executable(BwChecker_11)
+ add_executable(BwChecker_12)
+ add_executable(dcgmi)
+-add_executable(dcgmproftester11)
+ add_executable(dcgmproftester12)
+ add_executable(nv-hostengine)
+ add_executable(nvvs)
+@@ -327,9 +325,7 @@
+ add_library(childprocess STATIC)
+ add_library(common_watch_objects STATIC)
+ add_library(dcgm_common STATIC)
+-add_library(dcgm_cuda_worker11 STATIC)
+ add_library(dcgm_cuda_worker12 STATIC)
+-add_library(dcgm_cuda_lib11 STATIC)
+ add_library(dcgm_cuda_lib12 STATIC)
+ add_library(dcgm_entity_types STATIC)
+ add_library(dcgm_logging STATIC)
+@@ -342,7 +338,6 @@
+ add_library(nvvs_without_main_objects OBJECT)
+ add_library(nvvs_main_objects OBJECT)
+ add_library(nvvs_plugins_common_objects OBJECT)
+-add_library(pluginCudaCommon_11 STATIC)
+ add_library(pluginCudaCommon_12 STATIC)
+ add_library(remoteconn STATIC)
+ add_library(sdk_nvml_essentials_objects STATIC)
+@@ -368,24 +363,16 @@
+ add_library(dcgmmodulepolicy SHARED)
+ add_library(dcgmmodulesysmon SHARED)
+ 
+-add_library(ContextCreate_11 SHARED)
+ add_library(ContextCreate_12 SHARED)
+-add_library(Diagnostic_11 SHARED)
+ add_library(Diagnostic_12 SHARED)
+-add_library(Memory_11 SHARED)
+ add_library(Memory_12 SHARED)
+-add_library(Memtest_11 SHARED)
+ add_library(Memtest_12 SHARED)
+ add_library(NVBandwidth_12 SHARED)
+-add_library(Pcie_11 SHARED)
+ add_library(Pcie_12 SHARED)
+-add_library(TargetedPower_11 SHARED)
+ add_library(TargetedPower_12 SHARED)
+-add_library(TargetedStress_11 SHARED)
+ add_library(TargetedStress_12 SHARED)
+ 
+ add_library(dcgm SHARED)
+-add_library(dcgm_cublas_proxy11 SHARED)
+ add_library(dcgm_cublas_proxy12 SHARED)
+ add_library(pluginCommon SHARED)
+ 
+@@ -395,20 +382,13 @@
+ add_library(DCGM::dcgm ALIAS dcgm)
+ 
+ set_target_properties(
+-    ContextCreate_11
+     ContextCreate_12
+-    Diagnostic_11
+     Diagnostic_12
+-    Memory_11
+     Memory_12
+-    Memtest_11
+     Memtest_12
+     NVBandwidth_12
+-    Pcie_11
+     Pcie_12
+-    TargetedPower_11
+     TargetedPower_12
+-    TargetedStress_11
+     TargetedStress_12
+     dcgm
+     dcgmmoduleconfig
+@@ -419,7 +399,6 @@
+     dcgmmodulenvswitch
+     dcgmmodulepolicy
+     dcgmmodulesysmon
+-    dcgm_cublas_proxy11
+     dcgm_cublas_proxy12
+     pluginCommon
+     PROPERTIES
+@@ -433,7 +412,6 @@
+     RUNTIME_OUTPUT_DIRECTORY nvvs)
+ 
+ set_target_properties(
+-    BwChecker_11
+     BwChecker_12
+     PROPERTIES
+         INSTALL_RPATH "${DCGM_RPATH}:$ORIGIN/../../../${DCGM_TESTS_ARCH}")
+@@ -441,27 +419,19 @@
+ set_target_properties(dcgmi PROPERTIES RUNTIME_OUTPUT_DIRECTORY dcgmi)
+ 
+ set_target_properties(
+-    ContextCreate_11
+     ContextCreate_12
+-    Diagnostic_11
+     Diagnostic_12
+-    Memory_11
+     Memory_12
+-    Memtest_11
+     Memtest_12
+     NVBandwidth_12
+-    Pcie_11
+     Pcie_12
+-    TargetedPower_11
+     TargetedPower_12
+-    TargetedStress_11
+     TargetedStress_12
+     nvml_injection
+     nvmli_public
+     nvvs_without_main_objects
+     nvvs_plugins_common_objects
+     pluginCommon
+-    pluginCudaCommon_11
+     pluginCudaCommon_12
+     PROPERTIES
+         C_VISIBILITY_PRESET default
+@@ -594,18 +564,6 @@
+         COMPONENT Core)
+ 
+ install(
+-    TARGETS
+-        dcgm_cublas_proxy11
+-        dcgmproftester11
+-    LIBRARY
+-        DESTINATION "${CMAKE_INSTALL_LIBDIR}"
+-        COMPONENT Cuda11
+-        NAMELINK_SKIP
+-    RUNTIME
+-        DESTINATION "${CMAKE_INSTALL_BINDIR}"
+-        COMPONENT Cuda11)
+-
+-install(
+     TARGETS pluginCommon
+     LIBRARY
+         DESTINATION "${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}/plugins/cudaless"
+@@ -618,29 +576,6 @@
+     LIBRARY
+         DESTINATION "${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}/plugins/cudaless"
+         COMPONENT Core)
+-
+-install(
+-    TARGETS
+-        BwChecker_11
+-        ContextCreate_11
+-        Diagnostic_11
+-        Memory_11
+-        Memtest_11
+-        Pcie_11
+-        TargetedPower_11
+-        TargetedStress_11
+-    LIBRARY
+-        DESTINATION "${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}/plugins/cuda11"
+-        PERMISSIONS
+-            OWNER_READ OWNER_WRITE OWNER_EXECUTE
+-            GROUP_READ GROUP_EXECUTE
+-            WORLD_READ WORLD_EXECUTE
+-        COMPONENT Cuda11
+-        NAMELINK_SKIP
+-    RUNTIME
+-        DESTINATION "${CMAKE_INSTALL_LIBEXECDIR}/${PROJECT_NAME}/plugins/cuda11"
+-        COMPONENT Cuda11)
+-
+ install(
+     TARGETS
+         dcgm_cublas_proxy12
+@@ -686,7 +621,6 @@
+ install(
+     TARGETS
+         dcgm
+-        dcgm_cublas_proxy11
+         dcgm_cublas_proxy12
+         dcgmi
+         dcgmmoduleconfig
+@@ -697,7 +631,6 @@
+         dcgmmodulenvswitch
+         dcgmmodulepolicy
+         dcgmmodulesysmon
+-        dcgmproftester11
+         dcgmproftester12
+         nv-hostengine
+         nvml_injection
+@@ -725,28 +658,6 @@
+ 
+ install(
+     TARGETS
+-        BwChecker_11
+-        ContextCreate_11
+-        Diagnostic_11
+-        Memory_11
+-        Memtest_11
+-        Pcie_11
+-        TargetedPower_11
+-        TargetedStress_11
+-    LIBRARY
+-        DESTINATION "${CMAKE_INSTALL_DATADIR}/dcgm_tests/apps/nvvs/plugins/cuda11"
+-        PERMISSIONS
+-            OWNER_READ OWNER_WRITE OWNER_EXECUTE
+-            GROUP_READ GROUP_EXECUTE
+-            WORLD_READ WORLD_EXECUTE
+-        COMPONENT Tests
+-        NAMELINK_SKIP
+-    RUNTIME
+-        DESTINATION "${CMAKE_INSTALL_DATADIR}/dcgm_tests/apps/nvvs/plugins/cuda11"
+-        COMPONENT Tests)
+-
+-install(
+-    TARGETS
+         BwChecker_12
+         ContextCreate_12
+         Diagnostic_12
+diff --git a/cmake/FindCuda.cmake b/cmake/FindCuda.cmake
+index 3c1769597a..cf3e54d332 100644
+--- a/cmake/FindCuda.cmake
++++ b/cmake/FindCuda.cmake
+@@ -94,10 +94,6 @@
+ 
+ endmacro()
+ 
+-if (NOT DEFINED CUDA11_INCLUDE_DIR)
+-    load_cuda(11)
+-endif()
+-
+ if (NOT DEFINED CUDA12_INCLUDE_DIR)
+     load_cuda(12)
+ endif()
+diff --git a/common/CudaLib/CMakeLists.txt b/common/CudaLib/CMakeLists.txt
+index 0b2b0e0217..ea6fd17d8d 100644
+--- a/common/CudaLib/CMakeLists.txt
++++ b/common/CudaLib/CMakeLists.txt
+@@ -40,7 +40,6 @@
+         CudaLib.h)
+ endmacro()
+ 
+-define_dcgm_cuda_lib(11)
+ define_dcgm_cuda_lib(12)
+ 
+ target_include_directories(cuda_lib_base_interface INTERFACE
+diff --git a/common/CudaWorker/CMakeLists.txt b/common/CudaWorker/CMakeLists.txt
+index 958ace542f..f2c6ae748e 100644
+--- a/common/CudaWorker/CMakeLists.txt
++++ b/common/CudaWorker/CMakeLists.txt
+@@ -35,5 +35,4 @@
+         DcgmDgemm.cpp)
+ endmacro()
+ 
+-define_dcgm_cuda_worker(11)
+ define_dcgm_cuda_worker(12)
+diff --git a/cublas_proxy/CMakeLists.txt b/cublas_proxy/CMakeLists.txt
+index 90dff9e0d2..fe6dd40861 100755
+--- a/cublas_proxy/CMakeLists.txt
++++ b/cublas_proxy/CMakeLists.txt
+@@ -38,5 +38,4 @@
+         rt)
+ endmacro()
+ 
+-add_subdirectory(Cuda11)
+ add_subdirectory(Cuda12)
+diff --git a/dcgmproftester/CMakeLists.txt b/dcgmproftester/CMakeLists.txt
+index 9d18940bf1..97aa78321f 100755
+--- a/dcgmproftester/CMakeLists.txt
++++ b/dcgmproftester/CMakeLists.txt
+@@ -62,7 +62,6 @@
+             ${COMMON_SRCS})
+ endmacro()
+ 
+-define_dcgmproftester(11)
+ define_dcgmproftester(12)
+ 
+ install(
+diff --git a/nvvs/plugin_src/CMakeLists.txt b/nvvs/plugin_src/CMakeLists.txt
+index 21d0131c4a..5a3d371c29 100644
+--- a/nvvs/plugin_src/CMakeLists.txt
++++ b/nvvs/plugin_src/CMakeLists.txt
+@@ -71,7 +71,7 @@
+     target_link_libraries(nvvs_plugins INTERFACE "${PLUGIN_NAME}_${CUDA_VER}")
+ endmacro()
+ 
+-set(SUPPORTED_CUDA_VERSIONS 11 12)
++set(SUPPORTED_CUDA_VERSIONS 12)
+ 
+ add_subdirectory(common)
+ add_subdirectory(contextcreate)