
This is a bit sad, but, with the Allocators as static globals their destructors were running before some user code. Which doesn't really make much sense, as none of the members of (at least the basic one) do any real heavy lifting or have many resources to RAII. To avoid the problem, just mmap the memory for the global arrays of Allocators in __malloc_init and let the Kernel collect the memory when we're done with the process.
387 lines
12 KiB
C++
387 lines
12 KiB
C++
#include <AK/Bitmap.h>
|
|
#include <AK/InlineLinkedList.h>
|
|
#include <AK/ScopedValueRollback.h>
|
|
#include <AK/Vector.h>
|
|
#include <LibThread/Lock.h>
|
|
#include <assert.h>
|
|
#include <mallocdefs.h>
|
|
#include <serenity.h>
|
|
#include <stdio.h>
|
|
#include <stdlib.h>
|
|
#include <sys/mman.h>
|
|
|
|
// FIXME: Thread safety.
|
|
|
|
//#define MALLOC_DEBUG
|
|
#define RECYCLE_BIG_ALLOCATIONS
|
|
|
|
#define MAGIC_PAGE_HEADER 0x42657274
|
|
#define MAGIC_BIGALLOC_HEADER 0x42697267
|
|
#define PAGE_ROUND_UP(x) ((((size_t)(x)) + PAGE_SIZE - 1) & (~(PAGE_SIZE - 1)))
|
|
|
|
static LibThread::Lock& malloc_lock()
|
|
{
|
|
static u32 lock_storage[sizeof(LibThread::Lock) / sizeof(u32)];
|
|
return *reinterpret_cast<LibThread::Lock*>(&lock_storage);
|
|
}
|
|
|
|
constexpr int number_of_chunked_blocks_to_keep_around_per_size_class = 32;
|
|
constexpr int number_of_big_blocks_to_keep_around_per_size_class = 8;
|
|
|
|
static bool s_log_malloc = false;
|
|
static bool s_scrub_malloc = true;
|
|
static bool s_scrub_free = true;
|
|
static unsigned short size_classes[] = { 8, 16, 32, 64, 128, 252, 508, 1016, 2036, 0 };
|
|
static constexpr size_t num_size_classes = sizeof(size_classes) / sizeof(unsigned short);
|
|
|
|
struct CommonHeader {
|
|
size_t m_magic;
|
|
size_t m_size;
|
|
};
|
|
|
|
struct BigAllocationBlock : public CommonHeader {
|
|
BigAllocationBlock(size_t size)
|
|
{
|
|
m_magic = MAGIC_BIGALLOC_HEADER;
|
|
m_size = size;
|
|
}
|
|
unsigned char* m_slot[0];
|
|
};
|
|
|
|
struct FreelistEntry {
|
|
FreelistEntry* next;
|
|
};
|
|
|
|
struct ChunkedBlock : public CommonHeader
|
|
, public InlineLinkedListNode<ChunkedBlock> {
|
|
ChunkedBlock(size_t bytes_per_chunk)
|
|
{
|
|
m_magic = MAGIC_PAGE_HEADER;
|
|
m_size = bytes_per_chunk;
|
|
m_free_chunks = chunk_capacity();
|
|
m_freelist = (FreelistEntry*)chunk(0);
|
|
for (size_t i = 0; i < chunk_capacity(); ++i) {
|
|
auto* entry = (FreelistEntry*)chunk(i);
|
|
if (i != chunk_capacity() - 1)
|
|
entry->next = (FreelistEntry*)chunk(i + 1);
|
|
else
|
|
entry->next = nullptr;
|
|
}
|
|
}
|
|
|
|
ChunkedBlock* m_prev { nullptr };
|
|
ChunkedBlock* m_next { nullptr };
|
|
FreelistEntry* m_freelist { nullptr };
|
|
unsigned short m_free_chunks { 0 };
|
|
unsigned char m_slot[0];
|
|
|
|
void* chunk(int index)
|
|
{
|
|
return &m_slot[index * m_size];
|
|
}
|
|
bool is_full() const { return m_free_chunks == 0; }
|
|
size_t bytes_per_chunk() const { return m_size; }
|
|
size_t free_chunks() const { return m_free_chunks; }
|
|
size_t used_chunks() const { return chunk_capacity() - m_free_chunks; }
|
|
size_t chunk_capacity() const { return (PAGE_SIZE - sizeof(ChunkedBlock)) / m_size; }
|
|
};
|
|
|
|
struct Allocator {
|
|
size_t size { 0 };
|
|
size_t block_count { 0 };
|
|
size_t empty_block_count { 0 };
|
|
ChunkedBlock* empty_blocks[number_of_chunked_blocks_to_keep_around_per_size_class] { nullptr };
|
|
InlineLinkedList<ChunkedBlock> usable_blocks;
|
|
InlineLinkedList<ChunkedBlock> full_blocks;
|
|
};
|
|
|
|
struct BigAllocator {
|
|
Vector<BigAllocationBlock*, number_of_big_blocks_to_keep_around_per_size_class> blocks;
|
|
};
|
|
|
|
// Allocators will be mmapped in __malloc_init
|
|
Allocator* g_allocators = nullptr;
|
|
BigAllocator* g_big_allocators = nullptr;
|
|
|
|
static Allocator* allocator_for_size(size_t size, size_t& good_size)
|
|
{
|
|
for (int i = 0; size_classes[i]; ++i) {
|
|
if (size <= size_classes[i]) {
|
|
good_size = size_classes[i];
|
|
return &g_allocators[i];
|
|
}
|
|
}
|
|
good_size = PAGE_ROUND_UP(size);
|
|
return nullptr;
|
|
}
|
|
|
|
static BigAllocator* big_allocator_for_size(size_t size)
|
|
{
|
|
if (size == 4096)
|
|
return &g_big_allocators[0];
|
|
return nullptr;
|
|
}
|
|
|
|
extern "C" {
|
|
|
|
size_t malloc_good_size(size_t size)
|
|
{
|
|
for (int i = 0; size_classes[i]; ++i) {
|
|
if (size < size_classes[i])
|
|
return size_classes[i];
|
|
}
|
|
return PAGE_ROUND_UP(size);
|
|
}
|
|
|
|
static void* os_alloc(size_t size, const char* name)
|
|
{
|
|
return mmap_with_name(nullptr, size, PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE | MAP_PURGEABLE, 0, 0, name);
|
|
}
|
|
|
|
static void os_free(void* ptr, size_t size)
|
|
{
|
|
int rc = munmap(ptr, size);
|
|
assert(rc == 0);
|
|
}
|
|
|
|
void* malloc(size_t size)
|
|
{
|
|
LOCKER(malloc_lock());
|
|
|
|
if (s_log_malloc)
|
|
dbgprintf("LibC: malloc(%zu)\n", size);
|
|
|
|
if (!size)
|
|
return nullptr;
|
|
|
|
size_t good_size;
|
|
auto* allocator = allocator_for_size(size, good_size);
|
|
|
|
if (!allocator) {
|
|
size_t real_size = PAGE_ROUND_UP(sizeof(BigAllocationBlock) + size);
|
|
#ifdef RECYCLE_BIG_ALLOCATIONS
|
|
if (auto* allocator = big_allocator_for_size(real_size)) {
|
|
if (!allocator->blocks.is_empty()) {
|
|
auto* block = allocator->blocks.take_last();
|
|
int rc = madvise(block, real_size, MADV_SET_NONVOLATILE);
|
|
bool this_block_was_purged = rc == 1;
|
|
if (rc < 0) {
|
|
perror("madvise");
|
|
ASSERT_NOT_REACHED();
|
|
}
|
|
if (mprotect(block, real_size, PROT_READ | PROT_WRITE) < 0) {
|
|
perror("mprotect");
|
|
ASSERT_NOT_REACHED();
|
|
}
|
|
if (this_block_was_purged)
|
|
new (block) BigAllocationBlock(real_size);
|
|
set_mmap_name(block, PAGE_SIZE, "malloc: BigAllocationBlock (reused)");
|
|
return &block->m_slot[0];
|
|
}
|
|
}
|
|
#endif
|
|
auto* block = (BigAllocationBlock*)os_alloc(real_size, "malloc: BigAllocationBlock");
|
|
new (block) BigAllocationBlock(real_size);
|
|
return &block->m_slot[0];
|
|
}
|
|
|
|
ChunkedBlock* block = nullptr;
|
|
|
|
for (block = allocator->usable_blocks.head(); block; block = block->next()) {
|
|
if (block->free_chunks())
|
|
break;
|
|
}
|
|
|
|
if (!block && allocator->empty_block_count) {
|
|
block = allocator->empty_blocks[--allocator->empty_block_count];
|
|
int rc = madvise(block, PAGE_SIZE, MADV_SET_NONVOLATILE);
|
|
bool this_block_was_purged = rc == 1;
|
|
if (rc < 0) {
|
|
perror("madvise");
|
|
ASSERT_NOT_REACHED();
|
|
}
|
|
rc = mprotect(block, PAGE_SIZE, PROT_READ | PROT_WRITE);
|
|
if (rc < 0) {
|
|
perror("mprotect");
|
|
ASSERT_NOT_REACHED();
|
|
}
|
|
if (this_block_was_purged)
|
|
new (block) ChunkedBlock(good_size);
|
|
char buffer[64];
|
|
snprintf(buffer, sizeof(buffer), "malloc: ChunkedBlock(%zu) (reused)", good_size);
|
|
set_mmap_name(block, PAGE_SIZE, buffer);
|
|
allocator->usable_blocks.append(block);
|
|
}
|
|
|
|
if (!block) {
|
|
char buffer[64];
|
|
snprintf(buffer, sizeof(buffer), "malloc: ChunkedBlock(%zu)", good_size);
|
|
block = (ChunkedBlock*)os_alloc(PAGE_SIZE, buffer);
|
|
new (block) ChunkedBlock(good_size);
|
|
allocator->usable_blocks.append(block);
|
|
++allocator->block_count;
|
|
}
|
|
|
|
--block->m_free_chunks;
|
|
void* ptr = block->m_freelist;
|
|
block->m_freelist = block->m_freelist->next;
|
|
if (block->is_full()) {
|
|
#ifdef MALLOC_DEBUG
|
|
dbgprintf("Block %p is now full in size class %zu\n", block, good_size);
|
|
#endif
|
|
allocator->usable_blocks.remove(block);
|
|
allocator->full_blocks.append(block);
|
|
}
|
|
#ifdef MALLOC_DEBUG
|
|
dbgprintf("LibC: allocated %p (chunk in block %p, size %zu)\n", ptr, block, block->bytes_per_chunk());
|
|
#endif
|
|
if (s_scrub_malloc)
|
|
memset(ptr, MALLOC_SCRUB_BYTE, block->m_size);
|
|
return ptr;
|
|
}
|
|
|
|
void free(void* ptr)
|
|
{
|
|
ScopedValueRollback rollback(errno);
|
|
|
|
if (!ptr)
|
|
return;
|
|
|
|
LOCKER(malloc_lock());
|
|
|
|
void* page_base = (void*)((uintptr_t)ptr & (uintptr_t)~0xfff);
|
|
size_t magic = *(size_t*)page_base;
|
|
|
|
if (magic == MAGIC_BIGALLOC_HEADER) {
|
|
auto* block = (BigAllocationBlock*)page_base;
|
|
#ifdef RECYCLE_BIG_ALLOCATIONS
|
|
if (auto* allocator = big_allocator_for_size(block->m_size)) {
|
|
if (allocator->blocks.size() < number_of_big_blocks_to_keep_around_per_size_class) {
|
|
allocator->blocks.append(block);
|
|
set_mmap_name(block, PAGE_SIZE, "malloc: BigAllocationBlock (free)");
|
|
if (mprotect(block, PAGE_SIZE, PROT_NONE) < 0) {
|
|
perror("mprotect");
|
|
ASSERT_NOT_REACHED();
|
|
}
|
|
if (madvise(block, PAGE_SIZE, MADV_SET_VOLATILE) != 0) {
|
|
perror("madvise");
|
|
ASSERT_NOT_REACHED();
|
|
}
|
|
return;
|
|
}
|
|
}
|
|
#endif
|
|
os_free(block, block->m_size);
|
|
return;
|
|
}
|
|
|
|
assert(magic == MAGIC_PAGE_HEADER);
|
|
auto* block = (ChunkedBlock*)page_base;
|
|
|
|
#ifdef MALLOC_DEBUG
|
|
dbgprintf("LibC: freeing %p in allocator %p (size=%u, used=%u)\n", ptr, block, block->bytes_per_chunk(), block->used_chunks());
|
|
#endif
|
|
|
|
if (s_scrub_free)
|
|
memset(ptr, FREE_SCRUB_BYTE, block->bytes_per_chunk());
|
|
|
|
auto* entry = (FreelistEntry*)ptr;
|
|
entry->next = block->m_freelist;
|
|
block->m_freelist = entry;
|
|
|
|
if (block->is_full()) {
|
|
size_t good_size;
|
|
auto* allocator = allocator_for_size(block->m_size, good_size);
|
|
#ifdef MALLOC_DEBUG
|
|
dbgprintf("Block %p no longer full in size class %u\n", block, good_size);
|
|
#endif
|
|
allocator->full_blocks.remove(block);
|
|
allocator->usable_blocks.prepend(block);
|
|
}
|
|
|
|
++block->m_free_chunks;
|
|
|
|
if (!block->used_chunks()) {
|
|
size_t good_size;
|
|
auto* allocator = allocator_for_size(block->m_size, good_size);
|
|
if (allocator->block_count < number_of_chunked_blocks_to_keep_around_per_size_class) {
|
|
#ifdef MALLOC_DEBUG
|
|
dbgprintf("Keeping block %p around for size class %u\n", block, good_size);
|
|
#endif
|
|
allocator->usable_blocks.remove(block);
|
|
allocator->empty_blocks[allocator->empty_block_count++] = block;
|
|
char buffer[64];
|
|
snprintf(buffer, sizeof(buffer), "malloc: ChunkedBlock(%zu) (free)", good_size);
|
|
set_mmap_name(block, PAGE_SIZE, buffer);
|
|
mprotect(block, PAGE_SIZE, PROT_NONE);
|
|
madvise(block, PAGE_SIZE, MADV_SET_VOLATILE);
|
|
return;
|
|
}
|
|
#ifdef MALLOC_DEBUG
|
|
dbgprintf("Releasing block %p for size class %u\n", block, good_size);
|
|
#endif
|
|
allocator->usable_blocks.remove(block);
|
|
--allocator->block_count;
|
|
os_free(block, PAGE_SIZE);
|
|
}
|
|
}
|
|
|
|
void* calloc(size_t count, size_t size)
|
|
{
|
|
size_t new_size = count * size;
|
|
auto* ptr = malloc(new_size);
|
|
memset(ptr, 0, new_size);
|
|
return ptr;
|
|
}
|
|
|
|
size_t malloc_size(void* ptr)
|
|
{
|
|
if (!ptr)
|
|
return 0;
|
|
LOCKER(malloc_lock());
|
|
void* page_base = (void*)((uintptr_t)ptr & (uintptr_t)~0xfff);
|
|
auto* header = (const CommonHeader*)page_base;
|
|
auto size = header->m_size;
|
|
if (header->m_magic == MAGIC_BIGALLOC_HEADER)
|
|
size -= sizeof(CommonHeader);
|
|
return size;
|
|
}
|
|
|
|
void* realloc(void* ptr, size_t size)
|
|
{
|
|
if (!ptr)
|
|
return malloc(size);
|
|
LOCKER(malloc_lock());
|
|
auto existing_allocation_size = malloc_size(ptr);
|
|
if (size <= existing_allocation_size)
|
|
return ptr;
|
|
auto* new_ptr = malloc(size);
|
|
memcpy(new_ptr, ptr, min(existing_allocation_size, size));
|
|
free(ptr);
|
|
return new_ptr;
|
|
}
|
|
|
|
void __malloc_init()
|
|
{
|
|
new (&malloc_lock()) LibThread::Lock();
|
|
if (getenv("LIBC_NOSCRUB_MALLOC"))
|
|
s_scrub_malloc = false;
|
|
if (getenv("LIBC_NOSCRUB_FREE"))
|
|
s_scrub_free = false;
|
|
if (getenv("LIBC_LOG_MALLOC"))
|
|
s_log_malloc = true;
|
|
|
|
g_allocators = (Allocator*)mmap_with_name(nullptr, sizeof(Allocator) * num_size_classes, PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE, 0, 0, "LibC Allocators");
|
|
for (size_t i = 0; i < num_size_classes; ++i) {
|
|
new (&g_allocators[i]) Allocator();
|
|
g_allocators[i].size = size_classes[i];
|
|
}
|
|
|
|
g_big_allocators = (BigAllocator*)mmap_with_name(nullptr, sizeof(BigAllocator), PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE, 0, 0, "LibC BigAllocators");
|
|
new (g_big_allocators) (BigAllocator);
|
|
|
|
// We could mprotect the mmaps here with atexit, but, since this method is called in _start before
|
|
// _init and __init_array entries, our mprotect method would always be the last thing run before _exit.
|
|
}
|
|
|
|
}
|