integrate ETW windows event tracing into mimalloc as another track tool

CMakeLists.txt

@@ -12,6 +12,7 @@ option(MI_XMALLOC           "Enable abort() call on memory allocation failure by
 option(MI_SHOW_ERRORS       "Show error and warning messages by default (only enabled by default in DEBUG mode)" OFF)
 option(MI_TRACK_VALGRIND    "Compile with Valgrind support (adds a small overhead)" OFF)
 option(MI_TRACK_ASAN        "Compile with address sanitizer support (adds a small overhead)" OFF)
+option(MI_TRACK_ETW         "Compile with Windows event tracing (ETW) support (adds a small overhead)" OFF)
 option(MI_USE_CXX           "Use the C++ compiler to compile the library (instead of the C compiler)" OFF)
 option(MI_SEE_ASM           "Generate assembly files" OFF)
 option(MI_OSX_INTERPOSE     "Use interpose to override standard malloc on macOS" ON)
@@ -165,6 +166,21 @@ if(MI_TRACK_ASAN)
   endif()
 endif()
 
+if(MI_TRACK_ETW)
+  if NOT WIN32
+    set(MI_TRACK_ETW OFF)
+    message(WARNING "Can only enable ETW support on Windows (MI_TRACK_ETW=OFF)")
+  endif()
+  if (MI_TRACK_VALGRIND OR MI_TRACK_ASAN)
+    set(MI_TRACK_ETW OFF)
+    message(WARNING "Cannot enable ETW support with also Valgrind or ASAN support enabled (MI_TRACK_ETW=OFF)")
+  endif()
+  if(MI_TRACK_ETW)
+    message(STATUS "Compile with Windows event tracing support (MI_TRACK_ETW=ON)")
+    list(APPEND mi_defines MI_TRACK_ETW=1)
+  endif()
+endif()
+
 if(MI_SEE_ASM)
   message(STATUS "Generate assembly listings (MI_SEE_ASM=ON)")
   list(APPEND mi_cflags -save-temps)

include/mimalloc-etw.h

@@ -1,4 +0,0 @@
-#pragma once
-#include <windows.h>
-#include "mimalloc-etw-gen.h"
-

include/mimalloc-track.h

@@ -27,10 +27,12 @@ Optional:
 
   #define mi_track_align(p,alignedp,offset,size)
   #define mi_track_resize(p,oldsize,newsize)
+  #define mi_track_init()
 
 The `mi_track_align` is called right after a `mi_track_malloc` for aligned pointers in a block.
 The corresponding `mi_track_free` still uses the block start pointer and original size (corresponding to the `mi_track_malloc`).
 The `mi_track_resize` is currently unused but could be called on reallocations within a block.
+`mi_track_init` is called at program start.
 
 The following macros are for tools like asan and valgrind to track whether memory is 
 defined, undefined, or not accessible at all:
@@ -42,6 +44,7 @@ defined, undefined, or not accessible at all:
 -------------------------------------------------------------------------------------------------------*/
 
 #if MI_TRACK_VALGRIND
+// valgrind tool
 
 #define MI_TRACK_ENABLED      1
 #define MI_TRACK_HEAP_DESTROY 1           // track free of individual blocks on heap_destroy
@@ -58,6 +61,7 @@ defined, undefined, or not accessible at all:
 #define mi_track_mem_noaccess(p,size)             VALGRIND_MAKE_MEM_NOACCESS(p,size)
 
 #elif MI_TRACK_ASAN
+// address sanitizer
 
 #define MI_TRACK_ENABLED      1
 #define MI_TRACK_HEAP_DESTROY 0
@@ -71,19 +75,23 @@ defined, undefined, or not accessible at all:
 #define mi_track_mem_undefined(p,size)            ASAN_UNPOISON_MEMORY_REGION(p,size)
 #define mi_track_mem_noaccess(p,size)             ASAN_POISON_MEMORY_REGION(p,size)
 
-#elif MI_ETW
+#elif MI_TRACK_ETW
+// windows event tracing
+
 #define MI_TRACK_ENABLED      1
+#define MI_TRACK_HEAP_DESTROY 0
 #define MI_TRACK_TOOL         "ETW"
 
-#include "mimalloc-etw.h"
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+#include "../src/prim/windows/etw.h"
 
-#define mi_track_malloc_size(p,reqsize,size,zero)        EventWriteETW_MI_ALLOC((UINT64)p, size)
+#define mi_track_init()                           EventRegistermicrosoft_windows_mimalloc();
-#define mi_track_free_size(p,size)                    EventWriteETW_MI_FREE((UINT64)p, size)
+#define mi_track_malloc_size(p,reqsize,size,zero) EventWriteETW_MI_ALLOC((UINT64)(p), size)
-#define mi_track_mem_defined(p,size)
+#define mi_track_free_size(p,size)                EventWriteETW_MI_FREE((UINT64)(p), size)
-#define mi_track_mem_undefined(p,size)
-#define mi_track_mem_noaccess(p,size)
 
 #else
+// no tracking
 
 #define MI_TRACK_ENABLED      0
 #define MI_TRACK_HEAP_DESTROY 0 
@@ -91,11 +99,6 @@ defined, undefined, or not accessible at all:
 
 #define mi_track_malloc_size(p,reqsize,size,zero)
 #define mi_track_free_size(p,_size)
-#define mi_track_align(p,alignedp,offset,size)  
-#define mi_track_resize(p,oldsize,newsize)
-#define mi_track_mem_defined(p,size)
-#define mi_track_mem_undefined(p,size)
-#define mi_track_mem_noaccess(p,size)
 
 #endif
 
@@ -110,6 +113,23 @@ defined, undefined, or not accessible at all:
 #define mi_track_align(p,alignedp,offset,size)  mi_track_mem_noaccess(p,offset)
 #endif
 
+#ifndef mi_track_init
+#define mi_track_init()
+#endif
+
+#ifndef mi_track_mem_defined
+#define mi_track_mem_defined(p,size)
+#endif
+
+#ifndef mi_track_mem_undefined
+#define mi_track_mem_undefined(p,size)
+#endif
+
+#ifndef mi_track_mem_noaccess
+#define mi_track_mem_noaccess(p,size)
+#endif
+
+
 #if MI_PADDING
 #define mi_track_malloc(p,reqsize,zero) \
   if ((p)!=NULL) { \

include/mimalloc-types.h

@@ -32,8 +32,7 @@ terms of the MIT license. A copy of the license can be found in the file
 // Define MI_TRACK_<tool> to enable tracking support
 // #define MI_TRACK_VALGRIND 1
 // #define MI_TRACK_ASAN     1
-// Define MI_ETW to enable ETW provider
+#define MI_TRACK_ETW      1
- #define MI_ETW 1
 
 // Define MI_STAT as 1 to maintain statistics; set it to 2 to have detailed statistics (but costs some performance).
 // #define MI_STAT 1
@@ -62,7 +61,7 @@ terms of the MIT license. A copy of the license can be found in the file
 
 // Reserve extra padding at the end of each block to be more resilient against heap block overflows.
 // The padding can detect buffer overflow on free.
-#if !defined(MI_PADDING) && (MI_SECURE>=3 || MI_DEBUG>=1 || MI_TRACK_VALGRIND || MI_TRACK_ASAN || MI_ETW)
+#if !defined(MI_PADDING) && (MI_SECURE>=3 || MI_DEBUG>=1 || (MI_TRACK_VALGRIND || MI_TRACK_ASAN || MI_TRACK_ETW))
 #define MI_PADDING  1
 #endif
 

src/init.c

@@ -534,6 +534,7 @@ void mi_process_init(void) mi_attr_noexcept {
   #endif
 
   mi_stats_reset();  // only call stat reset *after* thread init (or the heap tld == NULL)
+  mi_track_init();
 
   if (mi_option_is_enabled(mi_option_reserve_huge_os_pages)) {
     size_t pages = mi_option_get_clamp(mi_option_reserve_huge_os_pages, 0, 128*1024);
@@ -550,10 +551,6 @@ void mi_process_init(void) mi_attr_noexcept {
       mi_reserve_os_memory((size_t)ksize*MI_KiB, true, true);
     }
   }
-
-#ifdef MI_ETW
-  EventRegistermicrosoft_windows_mimalloc();
-#endif
 }
 
 // Called when the process is done (through `at_exit`)

src/prim/readme.md

@@ -1,6 +1,8 @@
+## Portability Primitives
+
 This is the portability layer where all primitives needed from the OS are defined.
 
 - `prim.h`: API definition
-- `prim.c`: Selects one of `prim-unix.c`, `prim-wasi.c`, or `prim-windows.c` depending on the host platform.
+- `prim.c`: Selects one of `unix/prim.c`, `wasi/prim.c`, or `windows/prim.c` depending on the host platform.
 
-Note: still work in progress, there may be other places in the sources that still depend on OS ifdef's.
+Note: still work in progress, there may still be places in the sources that still depend on OS ifdef's.

src/prim/windows/mimalloc.wprp

@@ -29,7 +29,12 @@
         </SystemCollectorId>
         <EventCollectorId Value="Mimalloc_Collector">
           <EventProviders>
-            <EventProviderId Value="MimallocEventProvider" />
+            <EventProviderId Value="MimallocEventProvider" >
+              <Keywords>
+                <Keyword Value="100"/>
+                <Keyword Value="101"/>
+              </Keywords>
+            </EventProviderId>
           </EventProviders>
         </EventCollectorId>
       </Collectors>

src/prim/windows/readme.md

@@ -0,0 +1,17 @@
+## Primitives:
+
+- `prim.c` contains Windows primitives for OS allocation.
+
+## Event Tracing for Windows (ETW)
+
+- `etw.h` is generated from `etw.man` which contains the manifest for mimalloc events.
+  (100 is an allocation, 101 is for a free)
+
+- `mimalloc.wprp` is a profile for the Windows Performance Recorder (WPR).
+  In an admin prompt, you can use:
+  ```
+  > wpr -start src\prim\windows\mimalloc.wprp -filemode
+  > <my mimalloc program>
+  > wpr -stop test.etl
+  ``` 
+  and then open `test.etl` in the Windows Performance Analyzer (WPA).