* that can be shared across processes and threads
*/
#include <sys/mman.h>
-#include <stdio.h>
-#include <stdlib.h>
#include <assert.h>
#include <string.h>
-#include <unistd.h>
-#include <sys/types.h>
-#include <limits.h>
-#include "mutex.h"
+#include "fio.h"
+#include "fio_sem.h"
+#include "os/os.h"
+#include "smalloc.h"
+#include "log.h"
-#define MP_SAFE /* define to make thread safe */
#define SMALLOC_REDZONE /* define to detect memory corruption */
#define SMALLOC_BPB 32 /* block size, bytes-per-bit in bitmap */
#define SMALLOC_BPI (sizeof(unsigned int) * 8)
#define SMALLOC_BPL (SMALLOC_BPB * SMALLOC_BPI)
-#define INITIAL_SIZE 1024*1024 /* new pool size */
-#define MAX_POOLS 4 /* maximum number of pools to setup */
+#define INITIAL_SIZE 16*1024*1024 /* new pool size */
+#define INITIAL_POOLS 8 /* maximum number of pools to setup */
+
+#define MAX_POOLS 16
#define SMALLOC_PRE_RED 0xdeadbeefU
#define SMALLOC_POST_RED 0x5aa55aa5U
unsigned int smalloc_pool_size = INITIAL_SIZE;
+#ifdef SMALLOC_REDZONE
+static const int int_mask = sizeof(int) - 1;
+#endif
struct pool {
- struct fio_mutex *lock; /* protects this pool */
+ struct fio_sem *lock; /* protects this pool */
void *map; /* map of blocks */
unsigned int *bitmap; /* blocks free/busy map */
- unsigned int free_blocks; /* free blocks */
- unsigned int nr_blocks; /* total blocks */
- unsigned int next_non_full;
- int fd; /* memory backing fd */
- char file[PATH_MAX]; /* filename for fd */
- unsigned int mmap_size;
+ size_t free_blocks; /* free blocks */
+ size_t nr_blocks; /* total blocks */
+ size_t next_non_full;
+ size_t mmap_size;
};
struct block_hdr {
- unsigned int size;
+ size_t size;
#ifdef SMALLOC_REDZONE
unsigned int prered;
#endif
static struct pool mp[MAX_POOLS];
static unsigned int nr_pools;
static unsigned int last_pool;
-static struct fio_mutex *lock;
-
-static inline void pool_lock(struct pool *pool)
-{
- if (pool->lock)
- fio_mutex_down(pool->lock);
-}
-
-static inline void pool_unlock(struct pool *pool)
-{
- if (pool->lock)
- fio_mutex_up(pool->lock);
-}
-
-static inline void global_read_lock(void)
-{
- if (lock)
- fio_mutex_down_read(lock);
-}
-
-static inline void global_read_unlock(void)
-{
- if (lock)
- fio_mutex_up_read(lock);
-}
-
-static inline void global_write_lock(void)
-{
- if (lock)
- fio_mutex_down_write(lock);
-}
-
-static inline void global_write_unlock(void)
-{
- if (lock)
- fio_mutex_up_write(lock);
-}
static inline int ptr_valid(struct pool *pool, void *ptr)
{
return (ptr >= pool->map) && (ptr < pool->map + pool_size);
}
-static int blocks_iter(unsigned int *map, unsigned int idx,
- unsigned int nr_blocks,
+static inline size_t size_to_blocks(size_t size)
+{
+ return (size + SMALLOC_BPB - 1) / SMALLOC_BPB;
+}
+
+static int blocks_iter(struct pool *pool, unsigned int pool_idx,
+ unsigned int idx, size_t nr_blocks,
int (*func)(unsigned int *map, unsigned int mask))
{
+
while (nr_blocks) {
unsigned int this_blocks, mask;
+ unsigned int *map;
+
+ if (pool_idx >= pool->nr_blocks)
+ return 0;
+
+ map = &pool->bitmap[pool_idx];
this_blocks = nr_blocks;
if (this_blocks + idx > SMALLOC_BPI) {
nr_blocks -= this_blocks;
idx = 0;
- map++;
+ pool_idx++;
}
return 1;
static int mask_clear(unsigned int *map, unsigned int mask)
{
+ assert((*map & mask) == mask);
*map &= ~mask;
return 1;
}
static int mask_set(unsigned int *map, unsigned int mask)
{
+ assert(!(*map & mask));
*map |= mask;
return 1;
}
-static int blocks_free(unsigned int *map, unsigned int idx,
- unsigned int nr_blocks)
-{
- return blocks_iter(map, idx, nr_blocks, mask_cmp);
-}
-
-static void set_blocks(unsigned int *map, unsigned int idx,
- unsigned int nr_blocks)
+static int blocks_free(struct pool *pool, unsigned int pool_idx,
+ unsigned int idx, size_t nr_blocks)
{
- blocks_iter(map, idx, nr_blocks, mask_set);
+ return blocks_iter(pool, pool_idx, idx, nr_blocks, mask_cmp);
}
-static void clear_blocks(unsigned int *map, unsigned int idx,
- unsigned int nr_blocks)
+static void set_blocks(struct pool *pool, unsigned int pool_idx,
+ unsigned int idx, size_t nr_blocks)
{
- blocks_iter(map, idx, nr_blocks, mask_clear);
+ blocks_iter(pool, pool_idx, idx, nr_blocks, mask_set);
}
-static inline int __ffs(int word)
+static void clear_blocks(struct pool *pool, unsigned int pool_idx,
+ unsigned int idx, size_t nr_blocks)
{
- int r = 0;
-
- if (!(word & 0xffff)) {
- word >>= 16;
- r += 16;
- }
- if (!(word & 0xff)) {
- word >>= 8;
- r += 8;
- }
- if (!(word & 0xf)) {
- word >>= 4;
- r += 4;
- }
- if (!(word & 3)) {
- word >>= 2;
- r += 2;
- }
- if (!(word & 1)) {
- word >>= 1;
- r += 1;
- }
-
- return r;
+ blocks_iter(pool, pool_idx, idx, nr_blocks, mask_clear);
}
static int find_next_zero(int word, int start)
{
assert(word != -1U);
- word >>= (start + 1);
- return __ffs(~word) + start + 1;
+ word >>= start;
+ return ffz(word) + start;
}
-static int add_pool(struct pool *pool, unsigned int alloc_size)
+static bool add_pool(struct pool *pool, unsigned int alloc_size)
{
+ int bitmap_blocks;
+ int mmap_flags;
void *ptr;
- int fd, bitmap_blocks;
- strcpy(pool->file, "/tmp/.fio_smalloc.XXXXXX");
- fd = mkstemp(pool->file);
- if (fd < 0)
- goto out_close;
+ if (nr_pools == MAX_POOLS)
+ return false;
#ifdef SMALLOC_REDZONE
alloc_size += sizeof(unsigned int);
bitmap_blocks = alloc_size / SMALLOC_BPL;
alloc_size += bitmap_blocks * sizeof(unsigned int);
pool->mmap_size = alloc_size;
-
+
pool->nr_blocks = bitmap_blocks;
pool->free_blocks = bitmap_blocks * SMALLOC_BPB;
- if (ftruncate(fd, alloc_size) < 0)
- goto out_unlink;
+ mmap_flags = OS_MAP_ANON;
+#ifdef CONFIG_ESX
+ mmap_flags |= MAP_PRIVATE;
+#else
+ mmap_flags |= MAP_SHARED;
+#endif
+ ptr = mmap(NULL, alloc_size, PROT_READ|PROT_WRITE, mmap_flags, -1, 0);
- ptr = mmap(NULL, alloc_size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
if (ptr == MAP_FAILED)
- goto out_unlink;
+ goto out_fail;
- memset(ptr, 0, alloc_size);
pool->map = ptr;
- pool->bitmap = (void *) ptr + (pool->nr_blocks * SMALLOC_BPL);
+ pool->bitmap = (unsigned int *)((char *) ptr + (pool->nr_blocks * SMALLOC_BPL));
+ memset(pool->bitmap, 0, bitmap_blocks * sizeof(unsigned int));
-#ifdef MP_SAFE
- pool->lock = fio_mutex_init(1);
+ pool->lock = fio_sem_init(FIO_SEM_UNLOCKED);
if (!pool->lock)
- goto out_unlink;
-#endif
+ goto out_fail;
- pool->fd = fd;
-
- global_write_lock();
nr_pools++;
- global_write_unlock();
- return 0;
-out_unlink:
- fprintf(stderr, "smalloc: failed adding pool\n");
+ return true;
+out_fail:
+ log_err("smalloc: failed adding pool\n");
if (pool->map)
munmap(pool->map, pool->mmap_size);
- unlink(pool->file);
-out_close:
- if (fd >= 0)
- close(fd);
- return 1;
+ return false;
}
void sinit(void)
{
- int ret;
+ bool ret;
+ int i;
-#ifdef MP_SAFE
- lock = fio_mutex_rw_init();
-#endif
- ret = add_pool(&mp[0], INITIAL_SIZE);
- assert(!ret);
+ for (i = 0; i < INITIAL_POOLS; i++) {
+ ret = add_pool(&mp[nr_pools], smalloc_pool_size);
+ if (!ret)
+ break;
+ }
+
+ /*
+ * If we added at least one pool, we should be OK for most
+ * cases.
+ */
+ assert(i);
}
static void cleanup_pool(struct pool *pool)
{
- unlink(pool->file);
- close(pool->fd);
+ /*
+ * This will also remove the temporary file we used as a backing
+ * store, it was already unlinked
+ */
munmap(pool->map, pool->mmap_size);
if (pool->lock)
- fio_mutex_remove(pool->lock);
+ fio_sem_remove(pool->lock);
}
void scleanup(void)
for (i = 0; i < nr_pools; i++)
cleanup_pool(&mp[i]);
+}
+
+#ifdef SMALLOC_REDZONE
+static void *postred_ptr(struct block_hdr *hdr)
+{
+ uintptr_t ptr;
+
+ ptr = (uintptr_t) hdr + hdr->size - sizeof(unsigned int);
+ ptr = (uintptr_t) PTR_ALIGN(ptr, int_mask);
- if (lock)
- fio_mutex_remove(lock);
+ return (void *) ptr;
}
static void fill_redzone(struct block_hdr *hdr)
{
-#ifdef SMALLOC_REDZONE
- unsigned int *postred = (void *) hdr + hdr->size - sizeof(unsigned int);
+ unsigned int *postred = postred_ptr(hdr);
hdr->prered = SMALLOC_PRE_RED;
*postred = SMALLOC_POST_RED;
-#endif
}
static void sfree_check_redzone(struct block_hdr *hdr)
{
-#ifdef SMALLOC_REDZONE
- unsigned int *postred = (void *) hdr + hdr->size - sizeof(unsigned int);
+ unsigned int *postred = postred_ptr(hdr);
if (hdr->prered != SMALLOC_PRE_RED) {
- fprintf(stderr, "smalloc pre redzone destroyed!\n");
- fprintf(stderr, " ptr=%p, prered=%x, expected %x\n",
+ log_err("smalloc pre redzone destroyed!\n"
+ " ptr=%p, prered=%x, expected %x\n",
hdr, hdr->prered, SMALLOC_PRE_RED);
assert(0);
}
if (*postred != SMALLOC_POST_RED) {
- fprintf(stderr, "smalloc post redzone destroyed!\n");
- fprintf(stderr, " ptr=%p, postred=%x, expected %x\n",
+ log_err("smalloc post redzone destroyed!\n"
+ " ptr=%p, postred=%x, expected %x\n",
hdr, *postred, SMALLOC_POST_RED);
assert(0);
}
-#endif
}
+#else
+static void fill_redzone(struct block_hdr *hdr)
+{
+}
+
+static void sfree_check_redzone(struct block_hdr *hdr)
+{
+}
+#endif
static void sfree_pool(struct pool *pool, void *ptr)
{
i = offset / SMALLOC_BPL;
idx = (offset % SMALLOC_BPL) / SMALLOC_BPB;
- pool_lock(pool);
- clear_blocks(&pool->bitmap[i], idx, size_to_blocks(hdr->size));
+ fio_sem_down(pool->lock);
+ clear_blocks(pool, i, idx, size_to_blocks(hdr->size));
if (i < pool->next_non_full)
pool->next_non_full = i;
pool->free_blocks += size_to_blocks(hdr->size);
- pool_unlock(pool);
+ fio_sem_up(pool->lock);
}
void sfree(void *ptr)
if (!ptr)
return;
- global_read_lock();
-
for (i = 0; i < nr_pools; i++) {
if (ptr_valid(&mp[i], ptr)) {
pool = &mp[i];
}
}
- global_read_unlock();
+ if (pool) {
+ sfree_pool(pool, ptr);
+ return;
+ }
- assert(pool);
- sfree_pool(pool, ptr);
+ log_err("smalloc: ptr %p not from smalloc pool\n", ptr);
}
-static inline unsigned int size_to_blocks(unsigned int size)
+static void *__smalloc_pool(struct pool *pool, size_t size)
{
- return (size + SMALLOC_BPB - 1) / SMALLOC_BPB;
-}
-
-static void *__smalloc_pool(struct pool *pool, unsigned int size)
-{
- unsigned int nr_blocks;
+ size_t nr_blocks;
unsigned int i;
unsigned int offset;
unsigned int last_idx;
void *ret = NULL;
- pool_lock(pool);
+ fio_sem_down(pool->lock);
nr_blocks = size_to_blocks(size);
if (nr_blocks > pool->free_blocks)
}
idx = find_next_zero(pool->bitmap[i], last_idx);
- if (!blocks_free(&pool->bitmap[i], idx, nr_blocks)) {
+ if (!blocks_free(pool, i, idx, nr_blocks)) {
idx += nr_blocks;
if (idx < SMALLOC_BPI)
last_idx = idx;
}
continue;
}
- set_blocks(&pool->bitmap[i], idx, nr_blocks);
+ set_blocks(pool, i, idx, nr_blocks);
offset = i * SMALLOC_BPL + idx * SMALLOC_BPB;
break;
}
ret = pool->map + offset;
}
fail:
- pool_unlock(pool);
+ fio_sem_up(pool->lock);
return ret;
}
-static void *smalloc_pool(struct pool *pool, unsigned int size)
+static void *smalloc_pool(struct pool *pool, size_t size)
{
- struct block_hdr *hdr;
- unsigned int alloc_size;
+ size_t alloc_size = size + sizeof(struct block_hdr);
void *ptr;
- alloc_size = size + sizeof(*hdr);
+ /*
+ * Round to int alignment, so that the postred pointer will
+ * be naturally aligned as well.
+ */
#ifdef SMALLOC_REDZONE
alloc_size += sizeof(unsigned int);
+ alloc_size = (alloc_size + int_mask) & ~int_mask;
#endif
ptr = __smalloc_pool(pool, alloc_size);
- if (!ptr) {
- printf("failed allocating %u\n", alloc_size);
- return NULL;
- }
+ if (ptr) {
+ struct block_hdr *hdr = ptr;
- hdr = ptr;
- hdr->size = alloc_size;
- ptr += sizeof(*hdr);
+ hdr->size = alloc_size;
+ fill_redzone(hdr);
- fill_redzone(hdr);
+ ptr += sizeof(*hdr);
+ memset(ptr, 0, size);
+ }
- memset(ptr, 0, size);
return ptr;
}
-void *smalloc(unsigned int size)
+void *smalloc(size_t size)
{
- unsigned int i;
+ unsigned int i, end_pool;
+
+ if (size != (unsigned int) size)
+ return NULL;
- global_read_lock();
i = last_pool;
+ end_pool = nr_pools;
do {
- for (; i < nr_pools; i++) {
+ for (; i < end_pool; i++) {
void *ptr = smalloc_pool(&mp[i], size);
if (ptr) {
last_pool = i;
- global_read_unlock();
return ptr;
}
}
if (last_pool) {
- last_pool = 0;
+ end_pool = last_pool;
+ last_pool = i = 0;
continue;
}
- if (nr_pools + 1 > MAX_POOLS)
- break;
- else {
- i = nr_pools;
- global_read_unlock();
- if (add_pool(&mp[nr_pools], size))
- goto out;
- global_read_lock();
- }
+ break;
} while (1);
- global_read_unlock();
-out:
+ log_err("smalloc: OOM. Consider using --alloc-size to increase the "
+ "shared memory available.\n");
return NULL;
}
+void *scalloc(size_t nmemb, size_t size)
+{
+ return smalloc(nmemb * size);
+}
+
char *smalloc_strdup(const char *str)
{
- char *ptr;
+ char *ptr = NULL;
ptr = smalloc(strlen(str) + 1);
- strcpy(ptr, str);
+ if (ptr)
+ strcpy(ptr, str);
return ptr;
}