Initial perf jitdump implementation (#26897)

* Initial perf jitdump implementation

When COMPlus_PerfMapEnabled is enabled, generate jitdump file.

Usage:

export COMPlus_PerfMapEnabled=1
perf record -k 1 -g corerun helloworld.dll
perf --debug verbose=10 inject --input perf.data --jit --output perfjit.data
perf report --input perfjit.data

Works best with latest perf with patches:

Repo: git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git 
Branch: perf/urgent
Patches:
- https://lkml.org/lkml/2019/10/2/875

cd tools/perf && make DEBUG=1

Documentation/project-docs/clr-configuration-knobs.md

@@ -709,6 +709,7 @@ Name | Description | Type | Class | Default Value | Flags
 `ETWEnabled` | This flag is used on OSes < Vista to enable/disable ETW. It is disabled by default | `DWORD` | `EXTERNAL` | `0` | REGUTIL_default
 `MsBetweenAttachCheck` |  | `DWORD` | `EXTERNAL` | `500` |
 `PerfMapEnabled` | This flag is used on Linux to enable writing /tmp/perf-$pid.map. It is disabled by default | `DWORD` | `EXTERNAL` | `0` | REGUTIL_default
+`PerfMapJitDumpPath` | Specifies a path to write the perf jitdump file. Defaults to GetTempPathA() | `STRING` | `EXTERNAL` | | REGUTIL_default
 `PerfMapIgnoreSignal` | 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 | `DWORD` | `EXTERNAL` | `0` | REGUTIL_default
 `ProfAPI_AttachProfilerMinTimeoutInMs` | Timeout in ms for the minimum time out value of AttachProfiler | `DWORD` | `EXTERNAL` | `10*1000` |
 `ProfAPI_DetachMaxSleepMs` | The maximum time, in milliseconds, the CLR will wait before checking whether a profiler that is in the process of detaching is ready to be unloaded. | `DWORD` | `EXTERNAL` | `0` |

src/inc/clrconfigvalues.h

@@ -562,6 +562,7 @@ RETAIL_CONFIG_DWORD_INFO(UNSUPPORTED_ProfAPI_ValidateNGENInstrumentation, W("Pro
 
 #ifdef FEATURE_PERFMAP
 RETAIL_CONFIG_DWORD_INFO_EX(EXTERNAL_PerfMapEnabled, W("PerfMapEnabled"), 0, "This flag is used on Linux to enable writing /tmp/perf-$pid.map. It is disabled by default", CLRConfig::REGUTIL_default)
+RETAIL_CONFIG_STRING_INFO_EX(EXTERNAL_PerfMapJitDumpPath, W("PerfMapJitDumpPath"), "Specifies a path to write the perf jitdump file. Defaults to GetTempPathA()", CLRConfig::REGUTIL_default)
 RETAIL_CONFIG_DWORD_INFO_EX(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", CLRConfig::REGUTIL_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.")
 RETAIL_CONFIG_STRING_INFO(EXTERNAL_NativeImagePerfMapFormat, W("NativeImagePerfMapFormat"), "Specifies the format of native image perfmap files generated by crossgen.  Valid options are RVA or OFFSET.")

src/pal/inc/pal.h

@@ -529,6 +529,24 @@ PAL_ProbeMemory(
     DWORD cbBuffer,
     BOOL fWriteAccess);
 
+PALIMPORT
+int
+PALAPI
+// Start the jitdump file
+PAL_PerfJitDump_Start(const char* path);
+
+PALIMPORT
+int
+PALAPI
+// Log a method to the jitdump file.
+PAL_PerfJitDump_LogMethod(void* pCode, size_t codeSize, const char* symbol, void* debugInfo, void* unwindInfo);
+
+PALIMPORT
+int
+PALAPI
+// Finish the jitdump file
+PAL_PerfJitDump_Finish();
+
 /******************* winuser.h Entrypoints *******************************/
 PALIMPORT
 LPSTR

src/pal/src/CMakeLists.txt

@@ -226,6 +226,7 @@ set(SOURCES
   misc/jitsupport.cpp
   misc/miscpalapi.cpp
   misc/msgbox.cpp
+  misc/perfjitdump.cpp
   misc/strutil.cpp
   misc/sysinfo.cpp
   misc/time.cpp