AuroraMiddleware/mimalloc
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

added [std::]alloc_aligned/free redundency for restricted platforms, removed 'overrides'

Browse Source
This commit is contained in:
Reece Wilson 2020-10-02 10:17:12 +01:00
parent 13a4030619
commit 06042c079b
5 changed files with 35 additions and 511 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

260
src/alloc-override-osx.c
View File

@ -1,260 +0,0 @@
/* ----------------------------------------------------------------------------
Copyright (c) 2018, Microsoft Research, Daan Leijen
This is free software; you can redistribute it and/or modify it under the
terms of the MIT license. A copy of the license can be found in the file
"LICENSE" at the root of this distribution.
-----------------------------------------------------------------------------*/
#include "mimalloc.h"
#include "mimalloc-internal.h"
#if defined(MI_MALLOC_OVERRIDE)
#if !defined(__APPLE__)
#error "this file should only be included on macOS"
#endif
/* ------------------------------------------------------
Override system malloc on macOS
This is done through the malloc zone interface.
It seems we also need to interpose (see `alloc-override.c`)
or otherwise we get zone errors as there are usually
already allocations done by the time we take over the
zone. Unfortunately, that means we need to replace
the `free` with a checked free (`cfree`) impacting
performance.
------------------------------------------------------ */
#include <AvailabilityMacros.h>
#include <malloc/malloc.h>
#include <string.h> // memset
#if defined(MAC_OS_X_VERSION_10_6) && \
MAC_OS_X_VERSION_MAX_ALLOWED >= MAC_OS_X_VERSION_10_6
// only available from OSX 10.6
extern malloc_zone_t* malloc_default_purgeable_zone(void) __attribute__((weak_import));
#endif
/* ------------------------------------------------------
malloc zone members
------------------------------------------------------ */
static size_t zone_size(malloc_zone_t* zone, const void* p) {
UNUSED(zone);
if (!mi_is_in_heap_region(p))
return 0; // not our pointer, bail out
return mi_usable_size(p);
}
static void* zone_malloc(malloc_zone_t* zone, size_t size) {
UNUSED(zone);
return mi_malloc(size);
}
static void* zone_calloc(malloc_zone_t* zone, size_t count, size_t size) {
UNUSED(zone);
return mi_calloc(count, size);
}
static void* zone_valloc(malloc_zone_t* zone, size_t size) {
UNUSED(zone);
return mi_malloc_aligned(size, _mi_os_page_size());
}
static void zone_free(malloc_zone_t* zone, void* p) {
UNUSED(zone);
return mi_free(p);
}
static void* zone_realloc(malloc_zone_t* zone, void* p, size_t newsize) {
UNUSED(zone);
return mi_realloc(p, newsize);
}
static void* zone_memalign(malloc_zone_t* zone, size_t alignment, size_t size) {
UNUSED(zone);
return mi_malloc_aligned(size,alignment);
}
static void zone_destroy(malloc_zone_t* zone) {
UNUSED(zone);
// todo: ignore for now?
}
static unsigned zone_batch_malloc(malloc_zone_t* zone, size_t size, void** ps, unsigned count) {
size_t i;
for (i = 0; i < count; i++) {
ps[i] = zone_malloc(zone, size);
if (ps[i] == NULL) break;
}
return i;
}
static void zone_batch_free(malloc_zone_t* zone, void** ps, unsigned count) {
for(size_t i = 0; i < count; i++) {
zone_free(zone, ps[i]);
ps[i] = NULL;
}
}
static size_t zone_pressure_relief(malloc_zone_t* zone, size_t size) {
UNUSED(zone); UNUSED(size);
mi_collect(false);
return 0;
}
static void zone_free_definite_size(malloc_zone_t* zone, void* p, size_t size) {
UNUSED(size);
zone_free(zone,p);
}
/* ------------------------------------------------------
Introspection members
------------------------------------------------------ */
static kern_return_t intro_enumerator(task_t task, void* p,
unsigned type_mask, vm_address_t zone_address,
memory_reader_t reader,
vm_range_recorder_t recorder)
{
// todo: enumerate all memory
UNUSED(task); UNUSED(p); UNUSED(type_mask); UNUSED(zone_address);
UNUSED(reader); UNUSED(recorder);
return KERN_SUCCESS;
}
static size_t intro_good_size(malloc_zone_t* zone, size_t size) {
UNUSED(zone);
return mi_good_size(size);
}
static boolean_t intro_check(malloc_zone_t* zone) {
UNUSED(zone);
return true;
}
static void intro_print(malloc_zone_t* zone, boolean_t verbose) {
UNUSED(zone); UNUSED(verbose);
mi_stats_print(NULL);
}
static void intro_log(malloc_zone_t* zone, void* p) {
UNUSED(zone); UNUSED(p);
// todo?
}
static void intro_force_lock(malloc_zone_t* zone) {
UNUSED(zone);
// todo?
}
static void intro_force_unlock(malloc_zone_t* zone) {
UNUSED(zone);
// todo?
}
static void intro_statistics(malloc_zone_t* zone, malloc_statistics_t* stats) {
UNUSED(zone);
// todo...
stats->blocks_in_use = 0;
stats->size_in_use = 0;
stats->max_size_in_use = 0;
stats->size_allocated = 0;
}
static boolean_t intro_zone_locked(malloc_zone_t* zone) {
UNUSED(zone);
return false;
}
/* ------------------------------------------------------
At process start, override the default allocator
------------------------------------------------------ */
static malloc_zone_t* mi_get_default_zone()
{
// The first returned zone is the real default
malloc_zone_t** zones = NULL;
unsigned count = 0;
kern_return_t ret = malloc_get_all_zones(0, NULL, (vm_address_t**)&zones, &count);
if (ret == KERN_SUCCESS && count > 0) {
return zones[0];
}
else {
// fallback
return malloc_default_zone();
}
}
static void __attribute__((constructor)) _mi_macos_override_malloc()
{
static malloc_introspection_t intro;
memset(&intro, 0, sizeof(intro));
intro.enumerator = &intro_enumerator;
intro.good_size = &intro_good_size;
intro.check = &intro_check;
intro.print = &intro_print;
intro.log = &intro_log;
intro.force_lock = &intro_force_lock;
intro.force_unlock = &intro_force_unlock;
static malloc_zone_t zone;
memset(&zone, 0, sizeof(zone));
zone.version = 4;
zone.zone_name = "mimalloc";
zone.size = &zone_size;
zone.introspect = &intro;
zone.malloc = &zone_malloc;
zone.calloc = &zone_calloc;
zone.valloc = &zone_valloc;
zone.free = &zone_free;
zone.realloc = &zone_realloc;
zone.destroy = &zone_destroy;
zone.batch_malloc = &zone_batch_malloc;
zone.batch_free = &zone_batch_free;
malloc_zone_t* purgeable_zone = NULL;
#if defined(MAC_OS_X_VERSION_10_6) && \
MAC_OS_X_VERSION_MAX_ALLOWED >= MAC_OS_X_VERSION_10_6
// switch to version 9 on OSX 10.6 to support memalign.
zone.version = 9;
zone.memalign = &zone_memalign;
zone.free_definite_size = &zone_free_definite_size;
zone.pressure_relief = &zone_pressure_relief;
intro.zone_locked = &intro_zone_locked;
intro.statistics = &intro_statistics;
// force the purgeable zone to exist to avoid strange bugs
if (malloc_default_purgeable_zone) {
purgeable_zone = malloc_default_purgeable_zone();
}
#endif
// Register our zone
malloc_zone_register(&zone);
// Unregister the default zone, this makes our zone the new default
// as that was the last registered.
malloc_zone_t *default_zone = mi_get_default_zone();
malloc_zone_unregister(default_zone);
// Reregister the default zone so free and realloc in that zone keep working.
malloc_zone_register(default_zone);
// Unregister, and re-register the purgeable_zone to avoid bugs if it occurs
// earlier than the default zone.
if (purgeable_zone != NULL) {
malloc_zone_unregister(purgeable_zone);
malloc_zone_register(purgeable_zone);
}
}
#endif // MI_MALLOC_OVERRIDE

214
src/alloc-override.c
View File

@ -1,214 +0,0 @@
/* ----------------------------------------------------------------------------
Copyright (c) 2018, Microsoft Research, Daan Leijen
This is free software; you can redistribute it and/or modify it under the
terms of the MIT license. A copy of the license can be found in the file
"LICENSE" at the root of this distribution.
-----------------------------------------------------------------------------*/
#if !defined(MI_IN_ALLOC_C)
#error "this file should be included from 'alloc.c' (so aliases can work)"
#endif
#if defined(MI_MALLOC_OVERRIDE) && defined(_WIN32) && !(defined(MI_SHARED_LIB) && defined(_DLL))
#error "It is only possible to override "malloc" on Windows when building as a DLL (and linking the C runtime as a DLL)"
#endif
#if defined(MI_MALLOC_OVERRIDE) && !(defined(_WIN32)) // || (defined(__MACH__) && !defined(MI_INTERPOSE)))
// ------------------------------------------------------
// Override system malloc
// ------------------------------------------------------
#if (defined(__GNUC__) || defined(__clang__)) && !defined(__MACH__)
// use aliasing to alias the exported function to one of our `mi_` functions
#if (defined(__GNUC__) && __GNUC__ >= 9)
#define MI_FORWARD(fun) __attribute__((alias(#fun), used, visibility("default"), copy(fun)))
#else
#define MI_FORWARD(fun) __attribute__((alias(#fun), used, visibility("default")))
#endif
#define MI_FORWARD1(fun,x) MI_FORWARD(fun)
#define MI_FORWARD2(fun,x,y) MI_FORWARD(fun)
#define MI_FORWARD3(fun,x,y,z) MI_FORWARD(fun)
#define MI_FORWARD0(fun,x) MI_FORWARD(fun)
#define MI_FORWARD02(fun,x,y) MI_FORWARD(fun)
#else
// use forwarding by calling our `mi_` function
#define MI_FORWARD1(fun,x) { return fun(x); }
#define MI_FORWARD2(fun,x,y) { return fun(x,y); }
#define MI_FORWARD3(fun,x,y,z) { return fun(x,y,z); }
#define MI_FORWARD0(fun,x) { fun(x); }
#define MI_FORWARD02(fun,x,y) { fun(x,y); }
#endif
#if defined(__APPLE__) && defined(MI_SHARED_LIB_EXPORT) && defined(MI_INTERPOSE)
// use interposing so `DYLD_INSERT_LIBRARIES` works without `DYLD_FORCE_FLAT_NAMESPACE=1`
// See: <https://books.google.com/books?id=K8vUkpOXhN4C&pg=PA73>
struct mi_interpose_s {
const void* replacement;
const void* target;
};
#define MI_INTERPOSE_FUN(oldfun,newfun) { (const void*)&newfun, (const void*)&oldfun }
#define MI_INTERPOSE_MI(fun) MI_INTERPOSE_FUN(fun,mi_##fun)
__attribute__((used)) static struct mi_interpose_s _mi_interposes[] __attribute__((section("__DATA, __interpose"))) =
{
MI_INTERPOSE_MI(malloc),
MI_INTERPOSE_MI(calloc),
MI_INTERPOSE_MI(realloc),
MI_INTERPOSE_MI(strdup),
MI_INTERPOSE_MI(strndup),
MI_INTERPOSE_MI(realpath),
MI_INTERPOSE_MI(posix_memalign),
MI_INTERPOSE_MI(reallocf),
MI_INTERPOSE_MI(valloc),
// some code allocates from a zone but deallocates using plain free :-( (like NxHashResizeToCapacity <https://github.com/nneonneo/osx-10.9-opensource/blob/master/objc4-551.1/runtime/hashtable2.mm>)
MI_INTERPOSE_FUN(free,mi_cfree), // use safe free that checks if pointers are from us
};
#elif defined(_MSC_VER)
// cannot override malloc unless using a dll.
// we just override new/delete which does work in a static library.
#else
// On all other systems forward to our API
void* malloc(size_t size) MI_FORWARD1(mi_malloc, size);
void* calloc(size_t size, size_t n) MI_FORWARD2(mi_calloc, size, n);
void* realloc(void* p, size_t newsize) MI_FORWARD2(mi_realloc, p, newsize);
void free(void* p) MI_FORWARD0(mi_free, p);
#endif
#if (defined(__GNUC__) || defined(__clang__)) && !defined(__MACH__)
#pragma GCC visibility push(default)
#endif
// ------------------------------------------------------
// Override new/delete
// This is not really necessary as they usually call
// malloc/free anyway, but it improves performance.
// ------------------------------------------------------
#ifdef __cplusplus
// ------------------------------------------------------
// With a C++ compiler we override the new/delete operators.
// see <https://en.cppreference.com/w/cpp/memory/new/operator_new>
// ------------------------------------------------------
#include <new>
void operator delete(void* p) noexcept MI_FORWARD0(mi_free,p);
void operator delete[](void* p) noexcept MI_FORWARD0(mi_free,p);
void* operator new(std::size_t n) noexcept(false) MI_FORWARD1(mi_new,n);
void* operator new[](std::size_t n) noexcept(false) MI_FORWARD1(mi_new,n);
void* operator new (std::size_t n, const std::nothrow_t& tag) noexcept { UNUSED(tag); return mi_new_nothrow(n); }
void* operator new[](std::size_t n, const std::nothrow_t& tag) noexcept { UNUSED(tag); return mi_new_nothrow(n); }
#if (__cplusplus >= 201402L || _MSC_VER >= 1916)
void operator delete (void* p, std::size_t n) noexcept MI_FORWARD02(mi_free_size,p,n);
void operator delete[](void* p, std::size_t n) noexcept MI_FORWARD02(mi_free_size,p,n);
#endif
#if (__cplusplus > 201402L && defined(__cpp_aligned_new)) && (!defined(__GNUC__) || (__GNUC__ > 5))
void operator delete (void* p, std::align_val_t al) noexcept { mi_free_aligned(p, static_cast<size_t>(al)); }
void operator delete[](void* p, std::align_val_t al) noexcept { mi_free_aligned(p, static_cast<size_t>(al)); }
void operator delete (void* p, std::size_t n, std::align_val_t al) noexcept { mi_free_size_aligned(p, n, static_cast<size_t>(al)); };
void operator delete[](void* p, std::size_t n, std::align_val_t al) noexcept { mi_free_size_aligned(p, n, static_cast<size_t>(al)); };
void* operator new( std::size_t n, std::align_val_t al) noexcept(false) { return mi_new_aligned(n, static_cast<size_t>(al)); }
void* operator new[]( std::size_t n, std::align_val_t al) noexcept(false) { return mi_new_aligned(n, static_cast<size_t>(al)); }
void* operator new (std::size_t n, std::align_val_t al, const std::nothrow_t&) noexcept { return mi_new_aligned_nothrow(n, static_cast<size_t>(al)); }
void* operator new[](std::size_t n, std::align_val_t al, const std::nothrow_t&) noexcept { return mi_new_aligned_nothrow(n, static_cast<size_t>(al)); }
#endif
#elif (defined(__GNUC__) || defined(__clang__))
// ------------------------------------------------------
// Override by defining the mangled C++ names of the operators (as
// used by GCC and CLang).
// See <https://itanium-cxx-abi.github.io/cxx-abi/abi.html#mangling>
// ------------------------------------------------------
void _ZdlPv(void* p) MI_FORWARD0(mi_free,p); // delete
void _ZdaPv(void* p) MI_FORWARD0(mi_free,p); // delete[]
void _ZdlPvm(void* p, size_t n) MI_FORWARD02(mi_free_size,p,n);
void _ZdaPvm(void* p, size_t n) MI_FORWARD02(mi_free_size,p,n);
void _ZdlPvSt11align_val_t(void* p, size_t al) { mi_free_aligned(p,al); }
void _ZdaPvSt11align_val_t(void* p, size_t al) { mi_free_aligned(p,al); }
void _ZdlPvmSt11align_val_t(void* p, size_t n, size_t al) { mi_free_size_aligned(p,n,al); }
void _ZdaPvmSt11align_val_t(void* p, size_t n, size_t al) { mi_free_size_aligned(p,n,al); }
typedef struct mi_nothrow_s { } mi_nothrow_t;
#if (MI_INTPTR_SIZE==8)
void* _Znwm(size_t n) MI_FORWARD1(mi_new,n); // new 64-bit
void* _Znam(size_t n) MI_FORWARD1(mi_new,n); // new[] 64-bit
void* _ZnwmSt11align_val_t(size_t n, size_t al) MI_FORWARD2(mi_new_aligned, n, al);
void* _ZnamSt11align_val_t(size_t n, size_t al) MI_FORWARD2(mi_new_aligned, n, al);
void* _ZnwmRKSt9nothrow_t(size_t n, mi_nothrow_t tag) { UNUSED(tag); return mi_new_nothrow(n); }
void* _ZnamRKSt9nothrow_t(size_t n, mi_nothrow_t tag) { UNUSED(tag); return mi_new_nothrow(n); }
void* _ZnwmSt11align_val_tRKSt9nothrow_t(size_t n, size_t al, mi_nothrow_t tag) { UNUSED(tag); return mi_new_aligned_nothrow(n,al); }
void* _ZnamSt11align_val_tRKSt9nothrow_t(size_t n, size_t al, mi_nothrow_t tag) { UNUSED(tag); return mi_new_aligned_nothrow(n,al); }
#elif (MI_INTPTR_SIZE==4)
void* _Znwj(size_t n) MI_FORWARD1(mi_new,n); // new 64-bit
void* _Znaj(size_t n) MI_FORWARD1(mi_new,n); // new[] 64-bit
void* _ZnwjSt11align_val_t(size_t n, size_t al) MI_FORWARD2(mi_new_aligned, n, al);
void* _ZnajSt11align_val_t(size_t n, size_t al) MI_FORWARD2(mi_new_aligned, n, al);
void* _ZnwjRKSt9nothrow_t(size_t n, mi_nothrow_t tag) { UNUSED(tag); return mi_new_nothrow(n); }
void* _ZnajRKSt9nothrow_t(size_t n, mi_nothrow_t tag) { UNUSED(tag); return mi_new_nothrow(n); }
void* _ZnwjSt11align_val_tRKSt9nothrow_t(size_t n, size_t al, mi_nothrow_t tag) { UNUSED(tag); return mi_new_aligned_nothrow(n,al); }
void* _ZnajSt11align_val_tRKSt9nothrow_t(size_t n, size_t al, mi_nothrow_t tag) { UNUSED(tag); return mi_new_aligned_nothrow(n,al); }
#else
#error "define overloads for new/delete for this platform (just for performance, can be skipped)"
#endif
#endif // __cplusplus
#ifdef __cplusplus
extern "C" {
#endif
// ------------------------------------------------------
// Posix & Unix functions definitions
// ------------------------------------------------------
void cfree(void* p) MI_FORWARD0(mi_free, p);
void* reallocf(void* p, size_t newsize) MI_FORWARD2(mi_reallocf,p,newsize);
size_t malloc_size(const void* p) MI_FORWARD1(mi_usable_size,p);
#if !defined(__ANDROID__)
size_t malloc_usable_size(void *p) MI_FORWARD1(mi_usable_size,p);
#else
size_t malloc_usable_size(const void *p) MI_FORWARD1(mi_usable_size,p);
#endif
// no forwarding here due to aliasing/name mangling issues
void* valloc(size_t size) { return mi_valloc(size); }
void* pvalloc(size_t size) { return mi_pvalloc(size); }
void* reallocarray(void* p, size_t count, size_t size) { return mi_reallocarray(p, count, size); }
void* memalign(size_t alignment, size_t size) { return mi_memalign(alignment, size); }
int posix_memalign(void** p, size_t alignment, size_t size) { return mi_posix_memalign(p, alignment, size); }
void* _aligned_malloc(size_t alignment, size_t size) { return mi_aligned_alloc(alignment, size); }
// on some glibc `aligned_alloc` is declared `static inline` so we cannot override it (e.g. Conda). This happens
// when _GLIBCXX_HAVE_ALIGNED_ALLOC is not defined. However, in those cases it will use `memalign`, `posix_memalign`,
// or `_aligned_malloc` and we can avoid overriding it ourselves.
// We should always override if using C compilation. (issue #276)
#if _GLIBCXX_HAVE_ALIGNED_ALLOC || !defined(__cplusplus)
void* aligned_alloc(size_t alignment, size_t size) { return mi_aligned_alloc(alignment, size); }
#endif
#if defined(__GLIBC__) && defined(__linux__)
// forward __libc interface (needed for glibc-based Linux distributions)
void* __libc_malloc(size_t size) MI_FORWARD1(mi_malloc,size);
void* __libc_calloc(size_t count, size_t size) MI_FORWARD2(mi_calloc,count,size);
void* __libc_realloc(void* p, size_t size) MI_FORWARD2(mi_realloc,p,size);
void __libc_free(void* p) MI_FORWARD0(mi_free,p);
void __libc_cfree(void* p) MI_FORWARD0(mi_free,p);
void* __libc_valloc(size_t size) { return mi_valloc(size); }
void* __libc_pvalloc(size_t size) { return mi_pvalloc(size); }
void* __libc_memalign(size_t alignment, size_t size) { return mi_memalign(alignment,size); }
int __posix_memalign(void** p, size_t alignment, size_t size) { return mi_posix_memalign(p,alignment,size); }
#endif
#ifdef __cplusplus
}
#endif
#if (defined(__GNUC__) || defined(__clang__)) && !defined(__MACH__)
#pragma GCC visibility pop
#endif
#endif // MI_MALLOC_OVERRIDE && !_WIN32

6
src/alloc.c
View File

@ -4,16 +4,16 @@ This is free software; you can redistribute it and/or modify it under the
terms of the MIT license. A copy of the license can be found in the file
"LICENSE" at the root of this distribution.
-----------------------------------------------------------------------------*/
#include "AuroraEnvrionment.hpp"
#include "mimalloc.h"
#include "mimalloc-internal.h"
#include "mimalloc-atomic.h"
#include <string.h> // memset, memcpy, strlen
#include <stdlib.h> // malloc, exit
#define MI_IN_ALLOC_C
#include "alloc-override.c"
#undef MI_IN_ALLOC_C
// ------------------------------------------------------
// Allocation

34
src/os.c
View File

@ -4,6 +4,8 @@ This is free software; you can redistribute it and/or modify it under the
terms of the MIT license. A copy of the license can be found in the file
"LICENSE" at the root of this distribution.
-----------------------------------------------------------------------------*/
#include "AuroraEnvrionment.hpp"
#ifndef _DEFAULT_SOURCE
#define _DEFAULT_SOURCE // ensure mmap flags are defined
#endif
@ -20,6 +22,10 @@ terms of the MIT license. A copy of the license can be found in the file
#include "mimalloc-internal.h"
#include "mimalloc-atomic.h"
#if defined(_AURORA_MEM_STRICT)
#include <stdlib.h> // std allocation
#endif
#include <string.h> // strerror
#ifdef _MSC_VER
@ -227,7 +233,10 @@ static bool mi_os_mem_free(void* addr, size_t size, bool was_committed, mi_stats
{
if (addr == NULL || size == 0) return true; // || _mi_os_is_huge_reserved(addr)
bool err = false;
#if defined(_WIN32)
#if defined(_AURORA_MEM_STRICT)
free(addr);
err = 0;
#elif defined(_WIN32)
err = (VirtualFree(addr, 0, MEM_RELEASE) == 0);
#elif defined(__wasi__)
err = 0; // WebAssembly's heap cannot be shrunk
@ -247,7 +256,9 @@ static bool mi_os_mem_free(void* addr, size_t size, bool was_committed, mi_stats
static void* mi_os_get_aligned_hint(size_t try_alignment, size_t size);
#ifdef _WIN32
#if defined(_AURORA_MEM_STRICT)
#elif defined(_WIN32)
static void* mi_win_virtual_allocx(void* addr, size_t size, size_t try_alignment, DWORD flags) {
#if (MI_INTPTR_SIZE >= 8)
// on 64-bit systems, try to use the virtual address area after 4TiB for 4MiB aligned allocations
@ -508,8 +519,9 @@ static void* mi_os_mem_alloc(size_t size, size_t try_alignment, bool commit, boo
}
}
*/
#if defined(_WIN32)
#if defined(_AURORA_MEM_STRICT)
return aligned_alloc(try_alignment, size);
#elif defined(_WIN32)
int flags = MEM_RESERVE;
if (commit) flags |= MEM_COMMIT;
p = mi_win_virtual_alloc(NULL, size, try_alignment, flags, false, allow_large, is_large);
@ -698,7 +710,9 @@ static bool mi_os_commitx(void* addr, size_t size, bool commit, bool conservativ
_mi_stat_decrease(&stats->committed, size);
}
#if defined(_WIN32)
#if defined(_AURORA_MEM_STRICT)
// optional low level control. we can't control the MMIO from the CRT
#elif defined(_WIN32)
if (commit) {
// if the memory was already committed, the call succeeds but it is not zero'd
// *is_zero = true;
@ -770,7 +784,9 @@ static bool mi_os_resetx(void* addr, size_t size, bool reset, mi_stats_t* stats)
}
#endif
#if defined(_WIN32)
#if defined(_AURORA_MEM_STRICT)
// optional low level control. we can't control the MMIO from the CRT
#elif defined(_WIN32)
// Testing shows that for us (on `malloc-large`) MEM_RESET is 2x faster than DiscardVirtualMemory
void* p = VirtualAlloc(start, csize, MEM_RESET, PAGE_READWRITE);
mi_assert_internal(p == start);
@ -843,7 +859,9 @@ static bool mi_os_protectx(void* addr, size_t size, bool protect) {
}
*/
int err = 0;
#ifdef _WIN32
#if defined(_AURORA_MEM_STRICT)
// optional low level control. we can't control the MMIO from the CRT
#elif defined(_WIN32)
DWORD oldprotect = 0;
BOOL ok = VirtualProtect(start, csize, protect ? PAGE_NOACCESS : PAGE_READWRITE, &oldprotect);
err = (ok ? 0 : GetLastError());
@ -882,7 +900,7 @@ bool _mi_os_shrink(void* p, size_t oldsize, size_t newsize, mi_stats_t* stats) {
void* start = mi_os_page_align_area_conservative(addr, oldsize - newsize, &size);
if (size == 0 || start != addr) return false;
#ifdef _WIN32
#if defined(_WIN32) && !defined(_AURORA_MEM_STRICT)
// we cannot shrink on windows, but we can decommit
return _mi_os_decommit(start, size, stats);
#else

32
src/stats.c
View File

@ -4,6 +4,8 @@ This is free software; you can redistribute it and/or modify it under the
terms of the MIT license. A copy of the license can be found in the file
"LICENSE" at the root of this distribution.
-----------------------------------------------------------------------------*/
#include "AuroraEnvrionment.hpp"
#include "mimalloc.h"
#include "mimalloc-internal.h"
#include "mimalloc-atomic.h"
@ -442,7 +444,7 @@ mi_msecs_t _mi_clock_end(mi_msecs_t start) {
// Basic process statistics
// --------------------------------------------------------
#if defined(_WIN32)
#if defined(AURORA_PLATFORM_WIN32)
#include <Windows.h>
#include <psapi.h>
#pragma comment(lib,"psapi.lib")
@ -474,7 +476,7 @@ static void mi_stat_process_info(mi_msecs_t* elapsed, mi_msecs_t* utime, mi_msec
*page_faults = (size_t)info.PageFaultCount;
}
#elif defined(__unix__) || defined(__unix) || defined(unix) || (defined(__APPLE__) && defined(__MACH__)) || defined(__HAIKU__)
#elif defined(AURORA_PLATFORM_LINUX) || defined(AURORA_PLATFORM_APPLE)
#include <stdio.h>
#include <unistd.h>
#include <sys/resource.h>
@ -498,40 +500,18 @@ static void mi_stat_process_info(mi_msecs_t* elapsed, mi_msecs_t* utime, mi_msec
getrusage(RUSAGE_SELF, &rusage);
*utime = timeval_secs(&rusage.ru_utime);
*stime = timeval_secs(&rusage.ru_stime);
#if !defined(__HAIKU__)
*page_faults = rusage.ru_majflt;
#endif
// estimate commit using our stats
*peak_commit = (size_t)(mi_atomic_loadi64_relaxed((_Atomic(int64_t)*)&_mi_stats_main.committed.peak));
*current_commit = (size_t)(mi_atomic_loadi64_relaxed((_Atomic(int64_t)*)&_mi_stats_main.committed.current));
*current_rss = *current_commit; // estimate
#if defined(__HAIKU__)
// Haiku does not have (yet?) a way to
// get these stats per process
thread_info tid;
area_info mem;
ssize_t c;
get_thread_info(find_thread(0), &tid);
while (get_next_area_info(tid.team, &c, &mem) == B_OK) {
*peak_rss += mem.ram_size;
}
#elif defined(__APPLE__) && defined(__MACH__)
*peak_rss = rusage.ru_maxrss; // BSD reports in bytes
struct mach_task_basic_info info;
mach_msg_type_number_t infoCount = MACH_TASK_BASIC_INFO_COUNT;
if (task_info(mach_task_self(), MACH_TASK_BASIC_INFO, (task_info_t)&info, &infoCount) == KERN_SUCCESS) {
*current_rss = (size_t)info.resident_size;
}
#else
#if defined(AURORA_PLATFORM_LINUX)
*peak_rss = rusage.ru_maxrss * 1024; // Linux reports in KiB
#endif
}
#else
#ifndef __wasi__
// WebAssembly instances are not processes
#pragma message("define a way to get process info")
#endif
static void mi_stat_process_info(mi_msecs_t* elapsed, mi_msecs_t* utime, mi_msecs_t* stime, size_t* current_rss, size_t* peak_rss, size_t* current_commit, size_t* peak_commit, size_t* page_faults)
{