#include "libc.h"
#include "atomic.h"
#include "pthread_impl.h"
+#include "malloc_impl.h"
-uintptr_t __brk(uintptr_t);
-void *__mmap(void *, size_t, int, int, int, off_t);
-int __munmap(void *, size_t);
-void *__mremap(void *, size_t, size_t, int, ...);
-int __madvise(void *, size_t, int);
-
-struct chunk {
- size_t psize, csize;
- struct chunk *next, *prev;
-};
-
-struct bin {
- int lock[2];
- struct chunk *head;
- struct chunk *tail;
-};
+#if defined(__GNUC__) && defined(__PIC__)
+#define inline inline __attribute__((always_inline))
+#endif
static struct {
- uintptr_t brk;
- size_t *heap;
- uint64_t binmap;
+ volatile uint64_t binmap;
struct bin bins[64];
- int brk_lock[2];
- int free_lock[2];
+ volatile int free_lock[2];
} mal;
-
-#define SIZE_ALIGN (4*sizeof(size_t))
-#define SIZE_MASK (-SIZE_ALIGN)
-#define OVERHEAD (2*sizeof(size_t))
-#define MMAP_THRESHOLD (0x1c00*SIZE_ALIGN)
-#define DONTCARE 16
-#define RECLAIM 163840
-
-#define CHUNK_SIZE(c) ((c)->csize & -2)
-#define CHUNK_PSIZE(c) ((c)->psize & -2)
-#define PREV_CHUNK(c) ((struct chunk *)((char *)(c) - CHUNK_PSIZE(c)))
-#define NEXT_CHUNK(c) ((struct chunk *)((char *)(c) + CHUNK_SIZE(c)))
-#define MEM_TO_CHUNK(p) (struct chunk *)((char *)(p) - OVERHEAD)
-#define CHUNK_TO_MEM(c) (void *)((char *)(c) + OVERHEAD)
-#define BIN_TO_CHUNK(i) (MEM_TO_CHUNK(&mal.bins[i].head))
-
-#define C_INUSE ((size_t)1)
-
-#define IS_MMAPPED(c) !((c)->csize & (C_INUSE))
-
+int __malloc_replaced;
/* Synchronization tools */
-static void lock(volatile int *lk)
+static inline void lock(volatile int *lk)
{
- if (!libc.threads_minus_1) return;
- while(a_swap(lk, 1)) __wait(lk, lk+1, 1, 1);
+ int need_locks = libc.need_locks;
+ if (need_locks) {
+ while(a_swap(lk, 1)) __wait(lk, lk+1, 1, 1);
+ if (need_locks < 0) libc.need_locks = 0;
+ }
}
-static void unlock(volatile int *lk)
+static inline void unlock(volatile int *lk)
{
- if (!libc.threads_minus_1) return;
- a_store(lk, 0);
- if (lk[1]) __wake(lk, 1, 1);
+ if (lk[0]) {
+ a_store(lk, 0);
+ if (lk[1]) __wake(lk, 1, 1);
+ }
}
-static void lock_bin(int i)
+static inline void lock_bin(int i)
{
- if (libc.threads_minus_1)
- lock(mal.bins[i].lock);
+ lock(mal.bins[i].lock);
if (!mal.bins[i].head)
mal.bins[i].head = mal.bins[i].tail = BIN_TO_CHUNK(i);
}
-static void unlock_bin(int i)
+static inline void unlock_bin(int i)
{
- if (!libc.threads_minus_1) return;
unlock(mal.bins[i].lock);
}
#endif
}
+static const unsigned char bin_tab[60] = {
+ 32,33,34,35,36,36,37,37,38,38,39,39,
+ 40,40,40,40,41,41,41,41,42,42,42,42,43,43,43,43,
+ 44,44,44,44,44,44,44,44,45,45,45,45,45,45,45,45,
+ 46,46,46,46,46,46,46,46,47,47,47,47,47,47,47,47,
+};
+
static int bin_index(size_t x)
{
x = x / SIZE_ALIGN - 1;
if (x <= 32) return x;
+ if (x < 512) return bin_tab[x/8-4];
if (x > 0x1c00) return 63;
- return ((union { float v; uint32_t r; }){(int)x}.r>>21) - 496;
+ return bin_tab[x/128-4] + 16;
}
static int bin_index_up(size_t x)
{
x = x / SIZE_ALIGN - 1;
if (x <= 32) return x;
- return ((union { float v; uint32_t r; }){(int)x}.r+0x1fffff>>21) - 496;
+ x--;
+ if (x < 512) return bin_tab[x/8-4] + 1;
+ return bin_tab[x/128-4] + 17;
}
#if 0
static struct chunk *expand_heap(size_t n)
{
+ static int heap_lock[2];
+ static void *end;
+ void *p;
struct chunk *w;
- uintptr_t new;
-
- lock(mal.brk_lock);
-
- if (n > SIZE_MAX - mal.brk - 2*PAGE_SIZE) goto fail;
- new = mal.brk + n + SIZE_ALIGN + PAGE_SIZE - 1 & -PAGE_SIZE;
- n = new - mal.brk;
-
- if (__brk(new) != new) goto fail;
-
- w = MEM_TO_CHUNK(new);
- w->psize = n | C_INUSE;
- w->csize = 0 | C_INUSE;
- w = MEM_TO_CHUNK(mal.brk);
- w->csize = n | C_INUSE;
- mal.brk = new;
-
- unlock(mal.brk_lock);
+ /* The argument n already accounts for the caller's chunk
+ * overhead needs, but if the heap can't be extended in-place,
+ * we need room for an extra zero-sized sentinel chunk. */
+ n += SIZE_ALIGN;
- return w;
-fail:
- unlock(mal.brk_lock);
- return 0;
-}
+ lock(heap_lock);
-static int init_malloc(size_t n)
-{
- static int init, waiters;
- int state;
- struct chunk *c;
-
- if (init == 2) return 0;
-
- while ((state=a_swap(&init, 1)) == 1)
- __wait(&init, &waiters, 1, 1);
- if (state) {
- a_store(&init, 2);
+ p = __expand_heap(&n);
+ if (!p) {
+ unlock(heap_lock);
return 0;
}
- mal.brk = __brk(0) + 2*SIZE_ALIGN-1 & -SIZE_ALIGN;
+ /* If not just expanding existing space, we need to make a
+ * new sentinel chunk below the allocated space. */
+ if (p != end) {
+ /* Valid/safe because of the prologue increment. */
+ n -= SIZE_ALIGN;
+ p = (char *)p + SIZE_ALIGN;
+ w = MEM_TO_CHUNK(p);
+ w->psize = 0 | C_INUSE;
+ }
- c = expand_heap(n);
+ /* Record new heap end and fill in footer. */
+ end = (char *)p + n;
+ w = MEM_TO_CHUNK(end);
+ w->psize = n | C_INUSE;
+ w->csize = 0 | C_INUSE;
- if (!c) {
- a_store(&init, 0);
- if (waiters) __wake(&init, 1, 1);
- return -1;
- }
+ /* Fill in header, which may be new or may be replacing a
+ * zero-size sentinel header at the old end-of-heap. */
+ w = MEM_TO_CHUNK(p);
+ w->csize = n | C_INUSE;
- mal.heap = (void *)c;
- c->psize = 0 | C_INUSE;
- free(CHUNK_TO_MEM(c));
+ unlock(heap_lock);
- a_store(&init, 2);
- if (waiters) __wake(&init, -1, 1);
- return 1;
+ return w;
}
static int adjust_size(size_t *n)
next->psize = n1-n | C_INUSE;
self->csize = n | C_INUSE;
- free(CHUNK_TO_MEM(split));
+ __bin_chunk(split);
}
void *malloc(size_t n)
for (;;) {
uint64_t mask = mal.binmap & -(1ULL<<i);
if (!mask) {
- if (init_malloc(n) > 0) continue;
c = expand_heap(n);
if (!c) return 0;
if (alloc_rev(c)) {
j = first_set(mask);
lock_bin(j);
c = mal.bins[j].head;
- if (c != BIN_TO_CHUNK(j) && j == bin_index(c->csize)) {
+ if (c != BIN_TO_CHUNK(j)) {
if (!pretrim(c, n, i, j)) unbin(c, j);
unlock_bin(j);
break;
return CHUNK_TO_MEM(c);
}
+static size_t mal0_clear(char *p, size_t pagesz, size_t n)
+{
+#ifdef __GNUC__
+ typedef uint64_t __attribute__((__may_alias__)) T;
+#else
+ typedef unsigned char T;
+#endif
+ char *pp = p + n;
+ size_t i = (uintptr_t)pp & (pagesz - 1);
+ for (;;) {
+ pp = memset(pp - i, 0, i);
+ if (pp - p < pagesz) return pp - p;
+ for (i = pagesz; i; i -= 2*sizeof(T), pp -= 2*sizeof(T))
+ if (((T *)pp)[-1] | ((T *)pp)[-2])
+ break;
+ }
+}
+
+void *calloc(size_t m, size_t n)
+{
+ if (n && m > (size_t)-1/n) {
+ errno = ENOMEM;
+ return 0;
+ }
+ n *= m;
+ void *p = malloc(n);
+ if (!p) return p;
+ if (!__malloc_replaced) {
+ if (IS_MMAPPED(MEM_TO_CHUNK(p)))
+ return p;
+ if (n >= PAGE_SIZE)
+ n = mal0_clear(p, PAGE_SIZE, n);
+ }
+ return memset(p, 0, n);
+}
+
void *realloc(void *p, size_t n)
{
struct chunk *self, *next;
size_t newlen = n + extra;
/* Crash on realloc of freed chunk */
if (extra & 1) a_crash();
- if (newlen < PAGE_SIZE && (new = malloc(n))) {
- memcpy(new, p, n-OVERHEAD);
- free(p);
- return new;
+ if (newlen < PAGE_SIZE && (new = malloc(n-OVERHEAD))) {
+ n0 = n;
+ goto copy_free_ret;
}
newlen = (newlen + PAGE_SIZE-1) & -PAGE_SIZE;
if (oldlen == newlen) return p;
base = __mremap(base, oldlen, newlen, MREMAP_MAYMOVE);
if (base == (void *)-1)
- return newlen < oldlen ? p : 0;
+ goto copy_realloc;
self = (void *)(base + extra);
self->csize = newlen - extra;
return CHUNK_TO_MEM(self);
next = NEXT_CHUNK(self);
+ /* Crash on corrupted footer (likely from buffer overflow) */
+ if (next->psize != self->csize) a_crash();
+
/* Merge adjacent chunks if we need more space. This is not
* a waste of time even if we fail to get enough space, because our
* subsequent call to free would otherwise have to do the merge. */
return CHUNK_TO_MEM(self);
}
+copy_realloc:
/* As a last resort, allocate a new chunk and copy to it. */
new = malloc(n-OVERHEAD);
if (!new) return 0;
+copy_free_ret:
memcpy(new, p, n0-OVERHEAD);
free(CHUNK_TO_MEM(self));
return new;
}
-void free(void *p)
+void __bin_chunk(struct chunk *self)
{
- struct chunk *self = MEM_TO_CHUNK(p);
- struct chunk *next;
+ struct chunk *next = NEXT_CHUNK(self);
size_t final_size, new_size, size;
int reclaim=0;
int i;
- if (!p) return;
-
- if (IS_MMAPPED(self)) {
- size_t extra = self->psize;
- char *base = (char *)self - extra;
- size_t len = CHUNK_SIZE(self) + extra;
- /* Crash on double free */
- if (extra & 1) a_crash();
- __munmap(base, len);
- return;
- }
-
final_size = new_size = CHUNK_SIZE(self);
- next = NEXT_CHUNK(self);
- for (;;) {
- /* Replace middle of large chunks with fresh zero pages */
- if (reclaim && (self->psize & next->csize & C_INUSE)) {
- uintptr_t a = (uintptr_t)self + SIZE_ALIGN+PAGE_SIZE-1 & -PAGE_SIZE;
- uintptr_t b = (uintptr_t)next - SIZE_ALIGN & -PAGE_SIZE;
-#if 1
- __madvise((void *)a, b-a, MADV_DONTNEED);
-#else
- __mmap((void *)a, b-a, PROT_READ|PROT_WRITE,
- MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, -1, 0);
-#endif
- }
+ /* Crash on corrupted footer (likely from buffer overflow) */
+ if (next->psize != self->csize) a_crash();
+ for (;;) {
if (self->psize & next->csize & C_INUSE) {
self->csize = final_size | C_INUSE;
next->psize = final_size | C_INUSE;
}
}
+ if (!(mal.binmap & 1ULL<<i))
+ a_or_64(&mal.binmap, 1ULL<<i);
+
self->csize = final_size;
next->psize = final_size;
unlock(mal.free_lock);
self->next->prev = self;
self->prev->next = self;
- if (!(mal.binmap & 1ULL<<i))
- a_or_64(&mal.binmap, 1ULL<<i);
+ /* Replace middle of large chunks with fresh zero pages */
+ if (reclaim) {
+ uintptr_t a = (uintptr_t)self + SIZE_ALIGN+PAGE_SIZE-1 & -PAGE_SIZE;
+ uintptr_t b = (uintptr_t)next - SIZE_ALIGN & -PAGE_SIZE;
+#if 1
+ __madvise((void *)a, b-a, MADV_DONTNEED);
+#else
+ __mmap((void *)a, b-a, PROT_READ|PROT_WRITE,
+ MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED, -1, 0);
+#endif
+ }
unlock_bin(i);
}
+
+static void unmap_chunk(struct chunk *self)
+{
+ size_t extra = self->psize;
+ char *base = (char *)self - extra;
+ size_t len = CHUNK_SIZE(self) + extra;
+ /* Crash on double free */
+ if (extra & 1) a_crash();
+ __munmap(base, len);
+}
+
+void free(void *p)
+{
+ if (!p) return;
+
+ struct chunk *self = MEM_TO_CHUNK(p);
+
+ if (IS_MMAPPED(self))
+ unmap_chunk(self);
+ else
+ __bin_chunk(self);
+}
+
+void __malloc_donate(char *start, char *end)
+{
+ size_t align_start_up = (SIZE_ALIGN-1) & (-(uintptr_t)start - OVERHEAD);
+ size_t align_end_down = (SIZE_ALIGN-1) & (uintptr_t)end;
+
+ /* Getting past this condition ensures that the padding for alignment
+ * and header overhead will not overflow and will leave a nonzero
+ * multiple of SIZE_ALIGN bytes between start and end. */
+ if (end - start <= OVERHEAD + align_start_up + align_end_down)
+ return;
+ start += align_start_up + OVERHEAD;
+ end -= align_end_down;
+
+ struct chunk *c = MEM_TO_CHUNK(start), *n = MEM_TO_CHUNK(end);
+ c->psize = n->csize = C_INUSE;
+ c->csize = n->psize = C_INUSE | (end-start);
+ __bin_chunk(c);
+}
projects / musl / blobdiff