Enable FEATURE_PERFMAP on OSX, and update perfjitdump.cpp to work on …

…OSX (#99986)

* Enable FEATURE_PERFMAP on OSX, and update perfjitdump.cpp to work on OSX

* Update PerfMapEnabled documentation

* Enable only on OSX, and use PlatformGetCurrentThreadId

* Manual mach_absolute_time calls

* Use QueryPerformanceCounter

* Cleaner QueryPerformanceFrequency verification

* Use FEATURE_PERFMAP

* Put back conditional, but all __APPLE__

* Fix logic error when disabling jitdump

src/coreclr/clrdefinitions.cmake

@@ -155,6 +155,9 @@ endif(CLR_CMAKE_TARGET_LINUX AND CLR_CMAKE_HOST_LINUX)
 if(CLR_CMAKE_TARGET_FREEBSD)
     add_compile_definitions(FEATURE_PERFMAP)
 endif(CLR_CMAKE_TARGET_FREEBSD)
+if(CLR_CMAKE_TARGET_APPLE)
+    add_compile_definitions(FEATURE_PERFMAP)
+endif(CLR_CMAKE_TARGET_APPLE)
 
 if(FEATURE_COMWRAPPERS)
     add_compile_definitions(FEATURE_COMWRAPPERS)

src/coreclr/inc/clrconfigvalues.h

@@ -480,7 +480,7 @@ RETAIL_CONFIG_STRING_INFO(UNSUPPORTED_ETW_ObjectAllocationEventsPerTypePerSec, W
 RETAIL_CONFIG_DWORD_INFO(UNSUPPORTED_ProfAPI_ValidateNGENInstrumentation, W("ProfAPI_ValidateNGENInstrumentation"), 0, "This flag enables additional validations when using the IMetaDataEmit APIs for NGEN'ed images to ensure only supported edits are made.")
 
 #ifdef FEATURE_PERFMAP
-RETAIL_CONFIG_DWORD_INFO(EXTERNAL_PerfMapEnabled, W("PerfMapEnabled"), 0, "This flag is used on Linux to enable writing /tmp/perf-$pid.map. It is disabled by default")
+RETAIL_CONFIG_DWORD_INFO(EXTERNAL_PerfMapEnabled, W("PerfMapEnabled"), 0, "This flag is used on Linux and macOS to enable writing /tmp/perf-$pid.map. It is disabled by default")
 RETAIL_CONFIG_STRING_INFO_EX(EXTERNAL_PerfMapJitDumpPath, W("PerfMapJitDumpPath"), "Specifies a path to write the perf jitdump file. Defaults to /tmp", CLRConfig::LookupOptions::TrimWhiteSpaceFromStringValue)
 RETAIL_CONFIG_DWORD_INFO(EXTERNAL_PerfMapIgnoreSignal, W("PerfMapIgnoreSignal"), 0, "When perf map is enabled, this option will configure the specified signal to be accepted and ignored as a marker in the perf logs.  It is disabled by default")
 RETAIL_CONFIG_DWORD_INFO(EXTERNAL_PerfMapShowOptimizationTiers, W("PerfMapShowOptimizationTiers"), 1, "Shows optimization tiers in the perf map for methods, as part of the symbol name. Useful for seeing separate stack frames for different optimization tiers of each method.")

src/coreclr/pal/src/misc/perfjitdump.cpp

@@ -2,15 +2,15 @@
 // The .NET Foundation licenses this file to you under the MIT license.
 // ===========================================================================
 
-#if defined(__linux__)
-#define JITDUMP_SUPPORTED
-#endif
-
 #include "pal/palinternal.h"
 #include "pal/dbgmsg.h"
 
 #include <cstddef>
 
+#if defined(__linux__) || defined(__APPLE__)
+#define JITDUMP_SUPPORTED
+#endif
+
 #ifdef JITDUMP_SUPPORTED
 
 #include <fcntl.h>
@@ -61,24 +61,11 @@ namespace
         JIT_CODE_LOAD = 0,
     };
 
-    uint64_t GetTimeStampNS()
+    static uint64_t GetTimeStampNS()
     {
-#if HAVE_CLOCK_MONOTONIC
-        struct timespec ts;
-        int result = clock_gettime(CLOCK_MONOTONIC, &ts);
-
-        if (result != 0)
-        {
-            ASSERT("clock_gettime(CLOCK_MONOTONIC) failed: %d\n", result);
-            return 0;
-        }
-        else
-        {
-            return  ts.tv_sec * 1000000000ULL + ts.tv_nsec;
-        }
-#else
-    #error "The PAL jitdump requires clock_gettime(CLOCK_MONOTONIC) to be supported."
-#endif
+        LARGE_INTEGER result;
+        QueryPerformanceCounter(&result);
+        return result.QuadPart;
     }
 
     struct FileHeader
@@ -115,7 +102,7 @@ namespace
     {
         JitCodeLoadRecord() :
             pid(getpid()),
-            tid(syscall(SYS_gettid))
+            tid((uint32_t)PlatformGetCurrentThreadId())
         {
             header.id = JIT_CODE_LOAD;
             header.timestamp = GetTimeStampNS();
@@ -170,6 +157,19 @@ struct PerfJitDumpState
     {
         int result = 0;
 
+        // On platforms where JITDUMP is used, the PAL QueryPerformanceFrequency
+        // returns tccSecondsToNanoSeconds, meaning QueryPerformanceCounter
+        // will return a direct nanosecond value. If this isn't true,
+        // then some other method will need to be used to implement GetTimeStampNS.
+        // Validate this is true once in Start here.
+        LARGE_INTEGER freq;
+        QueryPerformanceFrequency(&freq);
+        if (freq.QuadPart != tccSecondsToNanoSeconds)
+        {
+            _ASSERTE(!"QueryPerformanceFrequency does not return tccSecondsToNanoSeconds. Implement JITDUMP GetTimeStampNS directly for this platform.\n");
+            FatalError();
+        }
+
         // Write file header
         FileHeader header;
 
@@ -203,12 +203,18 @@ struct PerfJitDumpState
         if (result == -1)
             return FatalError();
 
+#if !defined(__APPLE__)
         // mmap jitdump file
-        // this is a marker for perf inject to find the jitdumpfile
+        // this is a marker for perf inject to find the jitdumpfile on linux.
+        // On OSX, samply and others hook open and mmap is not needed. It also fails on OSX,
+        // likely because of PROT_EXEC and hardened runtime
         mmapAddr = mmap(nullptr, sizeof(FileHeader), PROT_READ | PROT_EXEC, MAP_PRIVATE, fd, 0);
 
         if (mmapAddr == MAP_FAILED)
             return FatalError();
+#else
+        mmapAddr = NULL;
+#endif
 
         enabled = true;
 
@@ -308,16 +314,13 @@ struct PerfJitDumpState
         {
             enabled = false;
 
-            if (result != 0)
-                return FatalError();
-
-            if (!enabled)
-                goto exit;
-
-            result = munmap(mmapAddr, sizeof(FileHeader));
+            if (mmapAddr != NULL)
+            {
+                result = munmap(mmapAddr, sizeof(FileHeader));
 
-            if (result == -1)
-                return FatalError();
+                if (result == -1)
+                    return FatalError();
+            }
 
             mmapAddr = MAP_FAILED;
 
@@ -333,7 +336,7 @@ struct PerfJitDumpState
 
             fd = -1;
         }
-exit:
+
         return 0;
     }
 };