mirrors
/
mimalloc
mirror of https://github.com/microsoft/mimalloc
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

merge from dev

This commit is contained in:
daan 2020-04-06 12:59:29 -07:00
parent 2d52b967bc afc4f79a69
commit 19cae48556
10 changed files with 68 additions and 752 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
ide/vs2019/mimalloc-override-test.vcxproj
View File

@ -176,14 +176,14 @@
</Command>
</PostBuildEvent>
</ItemDefinitionGroup>
<ItemGroup>
<ClCompile Include="..\..\test\main-override.cpp" />
</ItemGroup>
<ItemGroup>
<ProjectReference Include="mimalloc-override.vcxproj">
<Project>{abb5eae7-b3e6-432e-b636-333449892ea7}</Project>
</ProjectReference>
</ItemGroup>
<ItemGroup>
<ClCompile Include="..\..\test\main-override.cpp" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
</ImportGroup>

2
ide/vs2019/mimalloc-override.vcxproj
View File

@ -254,4 +254,4 @@
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
</ImportGroup>
</Project>
</Project>

722
include/mimalloc-internal-tld.h
View File

@ -1,722 +0,0 @@
/* ----------------------------------------------------------------------------
Copyright (c) 2019, 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.
-----------------------------------------------------------------------------*/
#pragma once
#ifndef MIMALLOC_INTERNAL_TLD_H
#define MIMALLOC_INTERNAL_TLD_H
#include "mimalloc-types.h"
#include "mimalloc-internal.h"
#define MI_TLD_DECL 1 // thread local declaration
#define MI_TLD_PTHREAD 2 // ptrhead_get/setspecific
#define MI_TLD_DECL_GUARD 3 // thread local + recursion guard at initial load
#define MI_TLD_PTHREAD_GUARD 4 // ptrhead_get/setspecific + recursion guard at initial load
#define MI_TLD_SLOT 5 // steal slot from OS thread local predefined slots
#define MI_TLD_PTHREAD_SLOT 6 // steal slot from pthread structure (usually `retval`)
#if !defined(MI_TLD)
#if defined(_MSC_VER) || defined(__linux__) || defined(__FreeBSD__) || defined(__NetBSD__)
// on windows and linux/freeBSD/netBSD (with initial-exec) a __thread always works without recursion into malloc
#define MI_TLD MI_TLD_DECL
#elif !defined(MI_MIMALLOC_OVERRIDE)
// if not overriding, __thread declarations should be fine (use MI_TLD_PTHREAD if your OS does not have __thread)
#define MI_TLD MI_TLD_DECL
#elif // defined(MI_MALLOC_OVERRIDE)
// if overriding, some BSD variants allocate when accessing a thread local the first time
#if defined(__APPLE__)
#define MI_TLD MI_TLD_SLOT
#define MI_TLD_SLOT_NUM 89 // seems unused? (__PTK_FRAMEWORK_OLDGC_KEY9) see <https://github.com/rweichler/substrate/blob/master/include/pthread_machdep.h>
// possibly unused ones are 9, 29, __PTK_FRAMEWORK_JAVASCRIPTCORE_KEY4 (94), __PTK_FRAMEWORK_GC_KEY9 (112) and __PTK_FRAMEWORK_OLDGC_KEY9 (89)
// #define MI_TLD MI_TLD_PTHREAD_SLOT
// #define MI_TLD_PTHREAD_SLOT_OFS (2*sizeof(void*) + sizeof(long) + 2*sizeof(void*) /*TAILQ*/) // offset `tl_exit_value` <https://github.com/apple/darwin-libpthread/blob/master/src/internal.h#L184>
#elif defined(__OpenBSD__)
#define MI_TLD MI_TLD_PTHREAD_SLOT
#define MI_TLD_PTHREAD_SLOT_OFS (6*sizeof(int) + 1*sizeof(void*)) // offset `retval` <https://github.com/openbsd/src/blob/master/lib/libc/include/thread_private.h#L371>
#elif defined(__DragonFly__)
#define MI_TLD MI_TLD_PTHREAD_SLOT
#define MI_TLD_PTHREAD_SLOT_OFS (4 + 1*sizeof(void*)) // offset `uniqueid` (also used by gdb?) <https://github.com/DragonFlyBSD/DragonFlyBSD/blob/master/lib/libthread_xu/thread/thr_private.h#L458>
#endif
#endif
#endif
#if (MI_DEBUG>0)
#define mi_trace_message(...) _mi_trace_message(__VA_ARGS__)
#else
#define mi_trace_message(...)
#endif
#define MI_CACHE_LINE 64
#if defined(_MSC_VER)
#pragma warning(disable:4127) // suppress constant conditional warning (due to MI_SECURE paths)
#define mi_decl_noinline __declspec(noinline)
#define mi_decl_thread __declspec(thread)
#define mi_decl_cache_align __declspec(align(MI_CACHE_LINE))
#elif (defined(__GNUC__) && (__GNUC__>=3)) // includes clang and icc
#define mi_decl_noinline __attribute__((noinline))
#define mi_decl_thread __thread
#define mi_decl_cache_align __attribute__((aligned(MI_CACHE_LINE)))
#else
#define mi_decl_noinline
#define mi_decl_thread __thread // hope for the best :-)
#define mi_decl_cache_align
#endif
// "options.c"
void _mi_fputs(mi_output_fun* out, void* arg, const char* prefix, const char* message);
void _mi_fprintf(mi_output_fun* out, void* arg, const char* fmt, ...);
void _mi_warning_message(const char* fmt, ...);
void _mi_verbose_message(const char* fmt, ...);
void _mi_trace_message(const char* fmt, ...);
void _mi_options_init(void);
void _mi_error_message(int err, const char* fmt, ...);
// random.c
void _mi_random_init(mi_random_ctx_t* ctx);
void _mi_random_split(mi_random_ctx_t* ctx, mi_random_ctx_t* new_ctx);
uintptr_t _mi_random_next(mi_random_ctx_t* ctx);
uintptr_t _mi_heap_random_next(mi_heap_t* heap);
uintptr_t _os_random_weak(uintptr_t extra_seed);
static inline uintptr_t _mi_random_shuffle(uintptr_t x);
// init.c
extern mi_stats_t _mi_stats_main;
extern const mi_page_t _mi_page_empty;
bool _mi_is_main_thread(void);
bool _mi_preloading(); // true while the C runtime is not ready
// os.c
size_t _mi_os_page_size(void);
void _mi_os_init(void); // called from process init
void* _mi_os_alloc(size_t size, mi_stats_t* stats); // to allocate thread local data
void _mi_os_free(void* p, size_t size, mi_stats_t* stats); // to free thread local data
size_t _mi_os_good_alloc_size(size_t size);
// memory.c
void* _mi_mem_alloc_aligned(size_t size, size_t alignment, bool* commit, bool* large, bool* is_zero, size_t* id, mi_os_tld_t* tld);
void _mi_mem_free(void* p, size_t size, size_t id, bool fully_committed, bool any_reset, mi_os_tld_t* tld);
bool _mi_mem_reset(void* p, size_t size, mi_os_tld_t* tld);
bool _mi_mem_unreset(void* p, size_t size, bool* is_zero, mi_os_tld_t* tld);
bool _mi_mem_commit(void* p, size_t size, bool* is_zero, mi_os_tld_t* tld);
bool _mi_mem_protect(void* addr, size_t size);
bool _mi_mem_unprotect(void* addr, size_t size);
void _mi_mem_collect(mi_os_tld_t* tld);
// "segment.c"
mi_page_t* _mi_segment_page_alloc(mi_heap_t* heap, size_t block_wsize, mi_segments_tld_t* tld, mi_os_tld_t* os_tld);
void _mi_segment_page_free(mi_page_t* page, bool force, mi_segments_tld_t* tld);
void _mi_segment_page_abandon(mi_page_t* page, mi_segments_tld_t* tld);
uint8_t* _mi_segment_page_start(const mi_segment_t* segment, const mi_page_t* page, size_t block_size, size_t* page_size, size_t* pre_size); // page start for any page
void _mi_segment_huge_page_free(mi_segment_t* segment, mi_page_t* page, mi_block_t* block);
void _mi_segment_thread_collect(mi_segments_tld_t* tld);
void _mi_abandoned_reclaim_all(mi_heap_t* heap, mi_segments_tld_t* tld);
void _mi_abandoned_await_readers(void);
// "page.c"
void* _mi_malloc_generic(mi_heap_t* heap, size_t size) mi_attr_noexcept mi_attr_malloc;
void _mi_page_retire(mi_page_t* page); // free the page if there are no other pages with many free blocks
void _mi_page_unfull(mi_page_t* page);
void _mi_page_free(mi_page_t* page, mi_page_queue_t* pq, bool force); // free the page
void _mi_page_abandon(mi_page_t* page, mi_page_queue_t* pq); // abandon the page, to be picked up by another thread...
void _mi_heap_delayed_free(mi_heap_t* heap);
void _mi_heap_collect_retired(mi_heap_t* heap, bool force);
void _mi_page_use_delayed_free(mi_page_t* page, mi_delayed_t delay, bool override_never);
size_t _mi_page_queue_append(mi_heap_t* heap, mi_page_queue_t* pq, mi_page_queue_t* append);
void _mi_deferred_free(mi_heap_t* heap, bool force);
void _mi_page_free_collect(mi_page_t* page,bool force);
void _mi_page_reclaim(mi_heap_t* heap, mi_page_t* page); // callback from segments
size_t _mi_bin_size(uint8_t bin); // for stats
uint8_t _mi_bin(size_t size); // for stats
uint8_t _mi_bsr(uintptr_t x); // bit-scan-right, used on BSD in "os.c"
// "heap.c"
void _mi_heap_destroy_pages(mi_heap_t* heap);
void _mi_heap_collect_abandon(mi_heap_t* heap);
void _mi_heap_set_default_direct(mi_heap_t* heap);
// "stats.c"
void _mi_stats_done(mi_stats_t* stats);
mi_msecs_t _mi_clock_now(void);
mi_msecs_t _mi_clock_end(mi_msecs_t start);
mi_msecs_t _mi_clock_start(void);
// "alloc.c"
void* _mi_page_malloc(mi_heap_t* heap, mi_page_t* page, size_t size) mi_attr_noexcept; // called from `_mi_malloc_generic`
void* _mi_heap_malloc_zero(mi_heap_t* heap, size_t size, bool zero);
void* _mi_heap_realloc_zero(mi_heap_t* heap, void* p, size_t newsize, bool zero);
mi_block_t* _mi_page_ptr_unalign(const mi_segment_t* segment, const mi_page_t* page, const void* p);
bool _mi_free_delayed_block(mi_block_t* block);
void _mi_block_zero_init(const mi_page_t* page, void* p, size_t size);
#if MI_DEBUG>1
bool _mi_page_is_valid(mi_page_t* page);
#endif
// ------------------------------------------------------
// Branches
// ------------------------------------------------------
#if defined(__GNUC__) || defined(__clang__)
#define mi_unlikely(x) __builtin_expect((x),0)
#define mi_likely(x) __builtin_expect((x),1)
#else
#define mi_unlikely(x) (x)
#define mi_likely(x) (x)
#endif
#ifndef __has_builtin
#define __has_builtin(x) 0
#endif
/* -----------------------------------------------------------
Error codes passed to `_mi_fatal_error`
All are recoverable but EFAULT is a serious error and aborts by default in secure mode.
For portability define undefined error codes using common Unix codes:
<https://www-numi.fnal.gov/offline_software/srt_public_context/WebDocs/Errors/unix_system_errors.html>
----------------------------------------------------------- */
#include <errno.h>
#ifndef EAGAIN // double free
#define EAGAIN (11)
#endif
#ifndef ENOMEM // out of memory
#define ENOMEM (12)
#endif
#ifndef EFAULT // corrupted free-list or meta-data
#define EFAULT (14)
#endif
#ifndef EINVAL // trying to free an invalid pointer
#define EINVAL (22)
#endif
#ifndef EOVERFLOW // count*size overflow
#define EOVERFLOW (75)
#endif
/* -----------------------------------------------------------
Inlined definitions
----------------------------------------------------------- */
#define UNUSED(x) (void)(x)
#if (MI_DEBUG>0)
#define UNUSED_RELEASE(x)
#else
#define UNUSED_RELEASE(x) UNUSED(x)
#endif
#define MI_INIT4(x) x(),x(),x(),x()
#define MI_INIT8(x) MI_INIT4(x),MI_INIT4(x)
#define MI_INIT16(x) MI_INIT8(x),MI_INIT8(x)
#define MI_INIT32(x) MI_INIT16(x),MI_INIT16(x)
#define MI_INIT64(x) MI_INIT32(x),MI_INIT32(x)
#define MI_INIT128(x) MI_INIT64(x),MI_INIT64(x)
#define MI_INIT256(x) MI_INIT128(x),MI_INIT128(x)
// Is `x` a power of two? (0 is considered a power of two)
static inline bool _mi_is_power_of_two(uintptr_t x) {
return ((x & (x - 1)) == 0);
}
// Align upwards
static inline uintptr_t _mi_align_up(uintptr_t sz, size_t alignment) {
mi_assert_internal(alignment != 0);
uintptr_t mask = alignment - 1;
if ((alignment & mask) == 0) { // power of two?
return ((sz + mask) & ~mask);
}
else {
return (((sz + mask)/alignment)*alignment);
}
}
// Divide upwards: `s <= _mi_divide_up(s,d)*d < s+d`.
static inline uintptr_t _mi_divide_up(uintptr_t size, size_t divider) {
mi_assert_internal(divider != 0);
return (divider == 0 ? size : ((size + divider - 1) / divider));
}
// Is memory zero initialized?
static inline bool mi_mem_is_zero(void* p, size_t size) {
for (size_t i = 0; i < size; i++) {
if (((uint8_t*)p)[i] != 0) return false;
}
return true;
}
// Align a byte size to a size in _machine words_,
// i.e. byte size == `wsize*sizeof(void*)`.
static inline size_t _mi_wsize_from_size(size_t size) {
mi_assert_internal(size <= SIZE_MAX - sizeof(uintptr_t));
return (size + sizeof(uintptr_t) - 1) / sizeof(uintptr_t);
}
// Overflow detecting multiply
static inline bool mi_mul_overflow(size_t count, size_t size, size_t* total) {
#if __has_builtin(__builtin_umul_overflow) || __GNUC__ >= 5
#include <limits.h> // UINT_MAX, ULONG_MAX
#if (SIZE_MAX == UINT_MAX)
return __builtin_umul_overflow(count, size, total);
#elif (SIZE_MAX == ULONG_MAX)
return __builtin_umull_overflow(count, size, total);
#else
return __builtin_umulll_overflow(count, size, total);
#endif
#else /* __builtin_umul_overflow is unavailable */
#define MI_MUL_NO_OVERFLOW ((size_t)1 << (4*sizeof(size_t))) // sqrt(SIZE_MAX)
*total = count * size;
return ((size >= MI_MUL_NO_OVERFLOW || count >= MI_MUL_NO_OVERFLOW)
&& size > 0 && (SIZE_MAX / size) < count);
#endif
}
// Safe multiply `count*size` into `total`; return `true` on overflow.
static inline bool mi_count_size_overflow(size_t count, size_t size, size_t* total) {
if (count==1) { // quick check for the case where count is one (common for C++ allocators)
*total = size;
return false;
}
else if (mi_unlikely(mi_mul_overflow(count, size, total))) {
_mi_error_message(EOVERFLOW, "allocation request too large (%zu * %zu bytes)\n", count, size);
*total = SIZE_MAX;
return true;
}
else return false;
}
/* -----------------------------------------------------------
The thread local default heap
----------------------------------------------------------- */
extern const mi_heap_t _mi_heap_empty; // read-only empty heap, initial value of the thread local default heap
extern mi_heap_t _mi_heap_main; // statically allocated main backing heap
extern bool _mi_process_is_initialized;
#if defined(MI_TLS_OSX_FAST)
#define MI_TLS_OSX_OFFSET (MI_TLS_OSX_SLOT*sizeof(void*))
static inline void* mi_tls_osx_fast_get(void) {
void* ret;
__asm__("mov %%gs:%1, %0" : "=r" (ret) : "m" (*(void**)(MI_TLS_OSX_OFFSET)));
return ret;
}
static inline void mi_tls_osx_fast_set(void* value) {
__asm__("movq %1,%%gs:%0" : "=m" (*(void**)(MI_TLS_OSX_OFFSET)) : "rn" (value));
}
#elif defined(MI_TLS_PTHREADS)
extern pthread_key_t _mi_heap_default_key;
#else
extern mi_decl_thread mi_heap_t* _mi_heap_default; // default heap to allocate from
#endif
static inline mi_heap_t* mi_get_default_heap(void) {
#if defined(MI_TLS_OSX_FAST)
// Use a fixed slot in the TSD on MacOSX to avoid recursion (since the loader calls malloc).
// We use slot 94 (__PTK_FRAMEWORK_JAVASCRIPTCORE_KEY4) <https://github.com/apportable/Foundation/blob/master/System/System/src/pthread_machdep.h>
// which seems unused except for the more recent Webkit <https://github.com/WebKit/webkit/blob/master/Source/WTF/wtf/FastTLS.h>
// Use with care.
mi_heap_t* heap = (mi_heap_t*)mi_tls_osx_fast_get();
return (mi_unlikely(heap == NULL) ? (mi_heap_t*)&_mi_heap_empty : heap);
#elif defined(MI_TLS_PTHREADS)
// Use pthreads for TLS; this is used on macOSX with interpose as the loader calls `malloc`
// to allocate TLS storage leading to recursive calls if __thread declared variables are accessed.
// Using pthreads allows us to initialize without recursive calls. (performance seems still quite good).
mi_heap_t* heap = (mi_unlikely(_mi_heap_default_key == (pthread_key_t)(-1)) ? (mi_heap_t*)&_mi_heap_empty : (mi_heap_t*)pthread_getspecific(_mi_heap_default_key));
return (mi_unlikely(heap == NULL) ? (mi_heap_t*)&_mi_heap_empty : heap);
#else
#if defined(MI_TLS_RECURSE_GUARD)
// On some BSD platforms, like openBSD, the dynamic loader calls `malloc`
// to initialize thread local data (before our module is loaded).
// To avoid recursion, we need to avoid accessing the thread local `_mi_default_heap`
// until our module is loaded and use the statically allocated main heap until that time.
// TODO: patch ourselves dynamically to avoid this check every time?
// if (mi_unlikely(!_mi_process_is_initialized)) return &_mi_heap_main;
#endif
return _mi_heap_default;
#endif
}
static inline bool mi_heap_is_default(const mi_heap_t* heap) {
return (heap == mi_get_default_heap());
}
static inline bool mi_heap_is_backing(const mi_heap_t* heap) {
return (heap->tld->heap_backing == heap);
}
static inline bool mi_heap_is_initialized(mi_heap_t* heap) {
mi_assert_internal(heap != NULL);
return (heap != &_mi_heap_empty);
}
static inline uintptr_t _mi_ptr_cookie(const void* p) {
mi_assert_internal(_mi_heap_main.cookie != 0);
return ((uintptr_t)p ^ _mi_heap_main.cookie);
}
/* -----------------------------------------------------------
Pages
----------------------------------------------------------- */
static inline mi_page_t* _mi_heap_get_free_small_page(mi_heap_t* heap, size_t size) {
mi_assert_internal(size <= (MI_SMALL_SIZE_MAX + MI_PADDING_SIZE));
const size_t idx = _mi_wsize_from_size(size);
mi_assert_internal(idx < MI_PAGES_DIRECT);
return heap->pages_free_direct[idx];
}
// Get the page belonging to a certain size class
static inline mi_page_t* _mi_get_free_small_page(size_t size) {
return _mi_heap_get_free_small_page(mi_get_default_heap(), size);
}
// Segment that contains the pointer
static inline mi_segment_t* _mi_ptr_segment(const void* p) {
// mi_assert_internal(p != NULL);
return (mi_segment_t*)((uintptr_t)p & ~MI_SEGMENT_MASK);
}
// Segment belonging to a page
static inline mi_segment_t* _mi_page_segment(const mi_page_t* page) {
mi_segment_t* segment = _mi_ptr_segment(page);
mi_assert_internal(segment == NULL || page == &segment->pages[page->segment_idx]);
return segment;
}
// used internally
static inline uintptr_t _mi_segment_page_idx_of(const mi_segment_t* segment, const void* p) {
// if (segment->page_size > MI_SEGMENT_SIZE) return &segment->pages[0]; // huge pages
ptrdiff_t diff = (uint8_t*)p - (uint8_t*)segment;
mi_assert_internal(diff >= 0 && (size_t)diff < MI_SEGMENT_SIZE);
uintptr_t idx = (uintptr_t)diff >> segment->page_shift;
mi_assert_internal(idx < segment->capacity);
mi_assert_internal(segment->page_kind <= MI_PAGE_MEDIUM || idx == 0);
return idx;
}
// Get the page containing the pointer
static inline mi_page_t* _mi_segment_page_of(const mi_segment_t* segment, const void* p) {
uintptr_t idx = _mi_segment_page_idx_of(segment, p);
return &((mi_segment_t*)segment)->pages[idx];
}
// Quick page start for initialized pages
static inline uint8_t* _mi_page_start(const mi_segment_t* segment, const mi_page_t* page, size_t* page_size) {
const size_t bsize = page->xblock_size;
mi_assert_internal(bsize > 0 && (bsize%sizeof(void*)) == 0);
return _mi_segment_page_start(segment, page, bsize, page_size, NULL);
}
// Get the page containing the pointer
static inline mi_page_t* _mi_ptr_page(void* p) {
return _mi_segment_page_of(_mi_ptr_segment(p), p);
}
// Get the block size of a page (special cased for huge objects)
static inline size_t mi_page_block_size(const mi_page_t* page) {
const size_t bsize = page->xblock_size;
mi_assert_internal(bsize > 0);
if (mi_likely(bsize < MI_HUGE_BLOCK_SIZE)) {
return bsize;
}
else {
size_t psize;
_mi_segment_page_start(_mi_page_segment(page), page, bsize, &psize, NULL);
return psize;
}
}
// Get the usable block size of a page without fixed padding.
// This may still include internal padding due to alignment and rounding up size classes.
static inline size_t mi_page_usable_block_size(const mi_page_t* page) {
return mi_page_block_size(page) - MI_PADDING_SIZE;
}
// Thread free access
static inline mi_block_t* mi_page_thread_free(const mi_page_t* page) {
return (mi_block_t*)(mi_atomic_read_relaxed(&page->xthread_free) & ~3);
}
static inline mi_delayed_t mi_page_thread_free_flag(const mi_page_t* page) {
return (mi_delayed_t)(mi_atomic_read_relaxed(&page->xthread_free) & 3);
}
// Heap access
static inline mi_heap_t* mi_page_heap(const mi_page_t* page) {
return (mi_heap_t*)(mi_atomic_read_relaxed(&page->xheap));
}
static inline void mi_page_set_heap(mi_page_t* page, mi_heap_t* heap) {
mi_assert_internal(mi_page_thread_free_flag(page) != MI_DELAYED_FREEING);
mi_atomic_write(&page->xheap,(uintptr_t)heap);
}
// Thread free flag helpers
static inline mi_block_t* mi_tf_block(mi_thread_free_t tf) {
return (mi_block_t*)(tf & ~0x03);
}
static inline mi_delayed_t mi_tf_delayed(mi_thread_free_t tf) {
return (mi_delayed_t)(tf & 0x03);
}
static inline mi_thread_free_t mi_tf_make(mi_block_t* block, mi_delayed_t delayed) {
return (mi_thread_free_t)((uintptr_t)block | (uintptr_t)delayed);
}
static inline mi_thread_free_t mi_tf_set_delayed(mi_thread_free_t tf, mi_delayed_t delayed) {
return mi_tf_make(mi_tf_block(tf),delayed);
}
static inline mi_thread_free_t mi_tf_set_block(mi_thread_free_t tf, mi_block_t* block) {
return mi_tf_make(block, mi_tf_delayed(tf));
}
// are all blocks in a page freed?
// note: needs up-to-date used count, (as the `xthread_free` list may not be empty). see `_mi_page_collect_free`.
static inline bool mi_page_all_free(const mi_page_t* page) {
mi_assert_internal(page != NULL);
return (page->used == 0);
}
// are there any available blocks?
static inline bool mi_page_has_any_available(const mi_page_t* page) {
mi_assert_internal(page != NULL && page->reserved > 0);
return (page->used < page->reserved || (mi_page_thread_free(page) != NULL));
}
// are there immediately available blocks, i.e. blocks available on the free list.
static inline bool mi_page_immediate_available(const mi_page_t* page) {
mi_assert_internal(page != NULL);
return (page->free != NULL);
}
// is more than 7/8th of a page in use?
static inline bool mi_page_mostly_used(const mi_page_t* page) {
if (page==NULL) return true;
uint16_t frac = page->reserved / 8U;
return (page->reserved - page->used <= frac);
}
static inline mi_page_queue_t* mi_page_queue(const mi_heap_t* heap, size_t size) {
return &((mi_heap_t*)heap)->pages[_mi_bin(size)];
}
//-----------------------------------------------------------
// Page flags
//-----------------------------------------------------------
static inline bool mi_page_is_in_full(const mi_page_t* page) {
return page->flags.x.in_full;
}
static inline void mi_page_set_in_full(mi_page_t* page, bool in_full) {
page->flags.x.in_full = in_full;
}
static inline bool mi_page_has_aligned(const mi_page_t* page) {
return page->flags.x.has_aligned;
}
static inline void mi_page_set_has_aligned(mi_page_t* page, bool has_aligned) {
page->flags.x.has_aligned = has_aligned;
}
/* -------------------------------------------------------------------
Encoding/Decoding the free list next pointers
This is to protect against buffer overflow exploits where the
free list is mutated. Many hardened allocators xor the next pointer `p`
with a secret key `k1`, as `p^k1`. This prevents overwriting with known
values but might be still too weak: if the attacker can guess
the pointer `p` this can reveal `k1` (since `p^k1^p == k1`).
Moreover, if multiple blocks can be read as well, the attacker can
xor both as `(p1^k1) ^ (p2^k1) == p1^p2` which may reveal a lot
about the pointers (and subsequently `k1`).
Instead mimalloc uses an extra key `k2` and encodes as `((p^k2)<<<k1)+k1`.
Since these operations are not associative, the above approaches do not
work so well any more even if the `p` can be guesstimated. For example,
for the read case we can subtract two entries to discard the `+k1` term,
but that leads to `((p1^k2)<<<k1) - ((p2^k2)<<<k1)` at best.
We include the left-rotation since xor and addition are otherwise linear
in the lowest bit. Finally, both keys are unique per page which reduces
the re-use of keys by a large factor.
We also pass a separate `null` value to be used as `NULL` or otherwise
`(k2<<<k1)+k1` would appear (too) often as a sentinel value.
------------------------------------------------------------------- */
static inline bool mi_is_in_same_segment(const void* p, const void* q) {
return (_mi_ptr_segment(p) == _mi_ptr_segment(q));
}
static inline bool mi_is_in_same_page(const void* p, const void* q) {
mi_segment_t* segmentp = _mi_ptr_segment(p);
mi_segment_t* segmentq = _mi_ptr_segment(q);
if (segmentp != segmentq) return false;
uintptr_t idxp = _mi_segment_page_idx_of(segmentp, p);
uintptr_t idxq = _mi_segment_page_idx_of(segmentq, q);
return (idxp == idxq);
}
static inline uintptr_t mi_rotl(uintptr_t x, uintptr_t shift) {
shift %= MI_INTPTR_BITS;
return ((x << shift) | (x >> (MI_INTPTR_BITS - shift)));
}
static inline uintptr_t mi_rotr(uintptr_t x, uintptr_t shift) {
shift %= MI_INTPTR_BITS;
return ((x >> shift) | (x << (MI_INTPTR_BITS - shift)));
}
static inline void* mi_ptr_decode(const void* null, const mi_encoded_t x, const uintptr_t* keys) {
void* p = (void*)(mi_rotr(x - keys[0], keys[0]) ^ keys[1]);
return (mi_unlikely(p==null) ? NULL : p);
}
static inline mi_encoded_t mi_ptr_encode(const void* null, const void* p, const uintptr_t* keys) {
uintptr_t x = (uintptr_t)(mi_unlikely(p==NULL) ? null : p);
return mi_rotl(x ^ keys[1], keys[0]) + keys[0];
}
static inline mi_block_t* mi_block_nextx( const void* null, const mi_block_t* block, const uintptr_t* keys ) {
#ifdef MI_ENCODE_FREELIST
return (mi_block_t*)mi_ptr_decode(null, block->next, keys);
#else
UNUSED(keys); UNUSED(null);
return (mi_block_t*)block->next;
#endif
}
static inline void mi_block_set_nextx(const void* null, mi_block_t* block, const mi_block_t* next, const uintptr_t* keys) {
#ifdef MI_ENCODE_FREELIST
block->next = mi_ptr_encode(null, next, keys);
#else
UNUSED(keys); UNUSED(null);
block->next = (mi_encoded_t)next;
#endif
}
static inline mi_block_t* mi_block_next(const mi_page_t* page, const mi_block_t* block) {
#ifdef MI_ENCODE_FREELIST
mi_block_t* next = mi_block_nextx(page,block,page->keys);
// check for free list corruption: is `next` at least in the same page?
// TODO: check if `next` is `page->block_size` aligned?
if (mi_unlikely(next!=NULL && !mi_is_in_same_page(block, next))) {
_mi_error_message(EFAULT, "corrupted free list entry of size %zub at %p: value 0x%zx\n", mi_page_block_size(page), block, (uintptr_t)next);
next = NULL;
}
return next;
#else
UNUSED(page);
return mi_block_nextx(page,block,NULL);
#endif
}
static inline void mi_block_set_next(const mi_page_t* page, mi_block_t* block, const mi_block_t* next) {
#ifdef MI_ENCODE_FREELIST
mi_block_set_nextx(page,block,next, page->keys);
#else
UNUSED(page);
mi_block_set_nextx(page,block,next,NULL);
#endif
}
// -------------------------------------------------------------------
// Fast "random" shuffle
// -------------------------------------------------------------------
static inline uintptr_t _mi_random_shuffle(uintptr_t x) {
if (x==0) { x = 17; } // ensure we don't get stuck in generating zeros
#if (MI_INTPTR_SIZE==8)
// by Sebastiano Vigna, see: <http://xoshiro.di.unimi.it/splitmix64.c>
x ^= x >> 30;
x *= 0xbf58476d1ce4e5b9UL;
x ^= x >> 27;
x *= 0x94d049bb133111ebUL;
x ^= x >> 31;
#elif (MI_INTPTR_SIZE==4)
// by Chris Wellons, see: <https://nullprogram.com/blog/2018/07/31/>
x ^= x >> 16;
x *= 0x7feb352dUL;
x ^= x >> 15;
x *= 0x846ca68bUL;
x ^= x >> 16;
#endif
return x;
}
// -------------------------------------------------------------------
// Optimize numa node access for the common case (= one node)
// -------------------------------------------------------------------
int _mi_os_numa_node_get(mi_os_tld_t* tld);
size_t _mi_os_numa_node_count_get(void);
extern size_t _mi_numa_node_count;
static inline int _mi_os_numa_node(mi_os_tld_t* tld) {
if (mi_likely(_mi_numa_node_count == 1)) return 0;
else return _mi_os_numa_node_get(tld);
}
static inline size_t _mi_os_numa_node_count(void) {
if (mi_likely(_mi_numa_node_count>0)) return _mi_numa_node_count;
else return _mi_os_numa_node_count_get();
}
// -------------------------------------------------------------------
// Getting the thread id should be performant
// as it is called in the fast path of `_mi_free`,
// so we specialize for various platforms.
// -------------------------------------------------------------------
#if defined(_WIN32)
#define WIN32_LEAN_AND_MEAN
#include <windows.h>
static inline uintptr_t _mi_thread_id(void) mi_attr_noexcept {
// Windows: works on Intel and ARM in both 32- and 64-bit
return (uintptr_t)NtCurrentTeb();
}
#elif (defined(__GNUC__) || defined(__clang__)) && \
(defined(__x86_64__) || defined(__i386__) || defined(__arm__) || defined(__aarch64__))
// TLS register on x86 is in the FS or GS register
// see: https://akkadia.org/drepper/tls.pdf
static inline uintptr_t _mi_thread_id(void) mi_attr_noexcept {
uintptr_t tid;
#if defined(__i386__)
__asm__("movl %%gs:0, %0" : "=r" (tid) : : ); // 32-bit always uses GS
#elif defined(__MACH__)
__asm__("movq %%gs:0, %0" : "=r" (tid) : : ); // x86_64 macOS uses GS
#elif defined(__x86_64__)
__asm__("movq %%fs:0, %0" : "=r" (tid) : : ); // x86_64 Linux, BSD uses FS
#elif defined(__arm__)
asm volatile ("mrc p15, 0, %0, c13, c0, 3" : "=r" (tid));
#elif defined(__aarch64__)
asm volatile ("mrs %0, tpidr_el0" : "=r" (tid));
#endif
return tid;
}
#else
// otherwise use standard C
static inline uintptr_t _mi_thread_id(void) mi_attr_noexcept {
return (uintptr_t)&_mi_heap_default;
}
#endif
#endif

15
include/mimalloc.h
View File

@ -34,7 +34,7 @@ terms of the MIT license. A copy of the license can be found in the file
#define mi_decl_nodiscard
#endif
#ifdef _MSC_VER
#if defined(_MSC_VER) || defined(__MINGW32__)
#if !defined(MI_SHARED_LIB)
#define mi_decl_export
#elif defined(MI_SHARED_LIB_EXPORT)
@ -42,13 +42,18 @@ terms of the MIT license. A copy of the license can be found in the file
#else
#define mi_decl_export __declspec(dllimport)
#endif
#if (_MSC_VER >= 1900) && !defined(__EDG__)
#define mi_decl_restrict __declspec(allocator) __declspec(restrict)
#if defined(__MINGW32__)
#define mi_decl_restrict
#define mi_attr_malloc __attribute__((malloc))
#else
#define mi_decl_restrict __declspec(restrict)
#if (_MSC_VER >= 1900) && !defined(__EDG__)
#define mi_decl_restrict __declspec(allocator) __declspec(restrict)
#else
#define mi_decl_restrict __declspec(restrict)
#endif
#define mi_attr_malloc
#endif
#define mi_cdecl __cdecl
#define mi_attr_malloc
#define mi_attr_alloc_size(s)
#define mi_attr_alloc_size2(s1,s2)
#define mi_attr_alloc_align(p)

9
src/alloc.c
View File

@ -474,15 +474,16 @@ size_t mi_usable_size(const void* p) mi_attr_noexcept {
if (p==NULL) return 0;
const mi_segment_t* const segment = _mi_ptr_segment(p);
const mi_page_t* const page = _mi_segment_page_of(segment, p);
const mi_block_t* const block = (const mi_block_t*)p;
const size_t size = mi_page_usable_size_of(page, block);
const mi_block_t* block = (const mi_block_t*)p;
if (mi_unlikely(mi_page_has_aligned(page))) {
ptrdiff_t const adjust = (uint8_t*)p - (uint8_t*)_mi_page_ptr_unalign(segment,page,p);
block = _mi_page_ptr_unalign(segment, page, p);
size_t size = mi_page_usable_size_of(page, block);
ptrdiff_t const adjust = (uint8_t*)p - (uint8_t*)block;
mi_assert_internal(adjust >= 0 && (size_t)adjust <= size);
return (size - adjust);
}
else {
return size;
return mi_page_usable_size_of(page, block);
}
}

6
src/init.c
View File

@ -286,6 +286,12 @@ static void _mi_thread_done(mi_heap_t* default_heap);
// use thread local storage keys to detect thread ending
#include <windows.h>
#include <fibersapi.h>
#if (_WIN32_WINNT < 0x600) // before Windows Vista
WINBASEAPI DWORD WINAPI FlsAlloc( _In_opt_ PFLS_CALLBACK_FUNCTION lpCallback );
WINBASEAPI PVOID WINAPI FlsGetValue( _In_ DWORD dwFlsIndex );
WINBASEAPI BOOL WINAPI FlsSetValue( _In_ DWORD dwFlsIndex, _In_opt_ PVOID lpFlsData );
WINBASEAPI BOOL WINAPI FlsFree(_In_ DWORD dwFlsIndex);
#endif
static DWORD mi_fls_key = (DWORD)(-1);
static void NTAPI mi_fls_done(PVOID value) {
if (value!=NULL) _mi_thread_done((mi_heap_t*)value);

13
src/os.c
View File

@ -823,7 +823,7 @@ and possibly associated with a specific NUMA node. (use `numa_node>=0`)
-----------------------------------------------------------------------------*/
#define MI_HUGE_OS_PAGE_SIZE (GiB)
#if defined(WIN32) && (MI_INTPTR_SIZE >= 8)
#if defined(_WIN32) && (MI_INTPTR_SIZE >= 8)
static void* mi_os_alloc_huge_os_pagesx(void* addr, size_t size, int numa_node)
{
mi_assert_internal(size%GiB == 0);
@ -866,6 +866,8 @@ static void* mi_os_alloc_huge_os_pagesx(void* addr, size_t size, int numa_node)
params[0].ULong = (unsigned)numa_node;
return (*pVirtualAlloc2)(GetCurrentProcess(), addr, size, flags, PAGE_READWRITE, params, 1);
}
#else
UNUSED(numa_node);
#endif
// otherwise use regular virtual alloc on older windows
return VirtualAlloc(addr, size, flags, PAGE_READWRITE);
@ -905,6 +907,7 @@ static void* mi_os_alloc_huge_os_pagesx(void* addr, size_t size, int numa_node)
}
#else
static void* mi_os_alloc_huge_os_pagesx(void* addr, size_t size, int numa_node) {
UNUSED(addr); UNUSED(size); UNUSED(numa_node);
return NULL;
}
#endif
@ -940,6 +943,7 @@ static uint8_t* mi_os_claim_huge_pages(size_t pages, size_t* total_size) {
}
#else
static uint8_t* mi_os_claim_huge_pages(size_t pages, size_t* total_size) {
UNUSED(pages);
if (total_size != NULL) *total_size = 0;
return NULL;
}
@ -1012,7 +1016,12 @@ void _mi_os_free_huge_pages(void* p, size_t size, mi_stats_t* stats) {
/* ----------------------------------------------------------------------------
Support NUMA aware allocation
-----------------------------------------------------------------------------*/
#ifdef WIN32
#ifdef _WIN32
#if (_WIN32_WINNT < 0x601) // before Win7
typedef struct _PROCESSOR_NUMBER { WORD Group; BYTE Number; BYTE Reserved; } PROCESSOR_NUMBER, *PPROCESSOR_NUMBER;
WINBASEAPI VOID WINAPI GetCurrentProcessorNumberEx(_Out_ PPROCESSOR_NUMBER ProcNumber);
WINBASEAPI BOOL WINAPI GetNumaProcessorNodeEx(_In_ PPROCESSOR_NUMBER Processor, _Out_ PUSHORT NodeNumber);
#endif
static size_t mi_os_numa_nodex() {
PROCESSOR_NUMBER pnum;
USHORT numa_node = 0;

12
src/segment.c
View File

@ -1290,7 +1290,7 @@ void _mi_segment_huge_page_free(mi_segment_t* segment, mi_page_t* page, mi_block
mi_assert_internal(mi_atomic_read_relaxed(&segment->thread_id)==0);
// claim it and free
mi_heap_t* heap = mi_get_default_heap();
mi_heap_t* heap = mi_heap_get_default(); // issue #221; don't use the internal get_default_heap as we need to ensure the thread is initialized.
// paranoia: if this it the last reference, the cas should always succeed
if (mi_atomic_cas_strong(&segment->thread_id, heap->thread_id, 0)) {
mi_block_set_next(page, block, page->free);
@ -1298,16 +1298,16 @@ void _mi_segment_huge_page_free(mi_segment_t* segment, mi_page_t* page, mi_block
page->used--;
page->is_zero = false;
mi_assert(page->used == 0);
mi_segments_tld_t* tld = &heap->tld->segments;
mi_tld_t* tld = heap->tld;
const size_t bsize = mi_page_usable_block_size(page);
if (bsize > MI_HUGE_OBJ_SIZE_MAX) {
_mi_stat_decrease(&tld->stats->giant, bsize);
_mi_stat_decrease(&tld->stats.giant, bsize);
}
else {
_mi_stat_decrease(&tld->stats->huge, bsize);
_mi_stat_decrease(&tld->stats.huge, bsize);
}
mi_segments_track_size((long)segment->segment_size, tld);
_mi_segment_page_free(page, true, tld);
mi_segments_track_size((long)segment->segment_size, &tld->segments);
_mi_segment_page_free(page, true, &tld->segments);
}
}

32
test/main-override.cpp
View File

@ -18,17 +18,18 @@ static void msleep(unsigned long msecs) { Sleep(msecs); }
static void msleep(unsigned long msecs) { usleep(msecs * 1000UL); }
#endif
static void heap_no_delete();
static void heap_late_free();
static void various_tests();
static void dangling_ptr_write();
void heap_thread_free_large(); // issue #221
void heap_no_delete(); // issue #202
void heap_late_free(); // issue #204
void padding_shrink(); // issue #209
void various_tests();
int main() {
mi_stats_reset(); // ignore earlier allocations
// heap_no_delete(); // issue #202
// heap_late_free(); // issue #204
// dangling_ptr_write();
// padding_shrink(); // issue #209
heap_thread_free_large();
heap_no_delete();
heap_late_free();
padding_shrink();
various_tests();
mi_stats_print(NULL);
return 0;
@ -70,7 +71,7 @@ static void various_tests() {
free(p1);
free(p2);
mi_free(s);
s[0] = 0;
// s[0] = 0; // test corruption
Test* t = new Test(42);
delete t;
// t = new(std::nothrow) Test(42); // does not work with overriding :-(
@ -175,3 +176,16 @@ void padding_shrink(void)
mi_free(shared_p);
}
// Issue #221
void heap_thread_free_large_worker() {
mi_free(shared_p);
}
void heap_thread_free_large() {
for (int i = 0; i < 100; i++) {
shared_p = mi_malloc_aligned(2*1024*1024 + 1, 8);
auto t1 = std::thread(heap_thread_free_large_worker);
t1.join();
}
}

3
test/test-api.c
View File

@ -152,6 +152,9 @@ int main() {
}
result = ok;
});
CHECK_BODY("malloc-aligned5", {
void* p = mi_malloc_aligned(4097,4096); size_t usable = mi_usable_size(p); result = usable >= 4097 && usable < 10000; mi_free(p);
});
CHECK_BODY("malloc-aligned-at1", {
void* p = mi_malloc_aligned_at(48,32,0); result = (p != NULL && ((uintptr_t)(p) + 0) % 32 == 0); mi_free(p);
});