/*************************************************************//**
Creates a mutex array to protect a hash table. */
UNIV_INTERN
void
hash_create_mutexes_func(
/*=====================*/
hash_table_t* table, /*!< in: hash table */
#ifdef UNIV_SYNC_DEBUG
ulint sync_level, /*!< in: latching order level of the
mutexes: used in the debug version */
#endif /* UNIV_SYNC_DEBUG */
ulint n_mutexes) /*!< in: number of mutexes, must be a
power of 2 */
{
ulint i;
ut_ad(table);
ut_ad(table->magic_n == HASH_TABLE_MAGIC_N);
ut_a(n_mutexes > 0);
ut_a(ut_is_2pow(n_mutexes));
table->mutexes = mem_alloc(n_mutexes * sizeof(mutex_t));
for (i = 0; i < n_mutexes; i++) {
mutex_create(table->mutexes + i, sync_level);
}
table->n_mutexes = n_mutexes;
}
/*************************************************************//**
Creates a hash table with at least n array cells. The actual number
of cells is chosen to be a prime number slightly bigger than n.
@return own: created table */
UNIV_INTERN
hash_table_t*
ha_create_func(
/*===========*/
ulint n, /*!< in: number of array cells */
#ifdef UNIV_SYNC_DEBUG
ulint mutex_level, /*!< in: level of the mutexes in the latching
order: this is used in the debug version */
#endif /* UNIV_SYNC_DEBUG */
ulint n_mutexes) /*!< in: number of mutexes to protect the
hash table: must be a power of 2, or 0 */
{
hash_table_t* table;
#ifndef UNIV_HOTBACKUP
ulint i;
#endif /* !UNIV_HOTBACKUP */
ut_ad(ut_is_2pow(n_mutexes));
table = hash_create(n);
#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
# ifndef UNIV_HOTBACKUP
table->adaptive = TRUE;
# endif /* !UNIV_HOTBACKUP */
#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
/* Creating MEM_HEAP_BTR_SEARCH type heaps can potentially fail,
but in practise it never should in this case, hence the asserts. */
if (n_mutexes == 0) {
table->heap = mem_heap_create_in_btr_search(
ut_min(4096, MEM_MAX_ALLOC_IN_BUF));
ut_a(table->heap);
return(table);
}
#ifndef UNIV_HOTBACKUP
hash_create_mutexes(table, n_mutexes, mutex_level);
table->heaps = mem_alloc(n_mutexes * sizeof(void*));
for (i = 0; i < n_mutexes; i++) {
table->heaps[i] = mem_heap_create_in_btr_search(4096);
ut_a(table->heaps[i]);
}
#endif /* !UNIV_HOTBACKUP */
return(table);
}
/********************************************************************//**
Creates a cache of column prefixes of externally stored columns.
@return own: column prefix cache */
UNIV_INTERN
row_ext_t*
row_ext_create(
/*===========*/
ulint n_ext, /*!< in: number of externally stored columns */
const ulint* ext, /*!< in: col_no's of externally stored columns
in the InnoDB table object, as reported by
dict_col_get_no(); NOT relative to the records
in the clustered index */
ulint flags, /*!< in: table->flags */
const dtuple_t* tuple, /*!< in: data tuple containing the field
references of the externally stored
columns; must be indexed by col_no;
the clustered index record must be
covered by a lock or a page latch
to prevent deletion (rollback or purge). */
mem_heap_t* heap) /*!< in: heap where created */
{
ulint i;
ulint zip_size = dict_table_flags_to_zip_size(flags);
row_ext_t* ret = mem_heap_alloc(heap, (sizeof *ret)
+ (n_ext - 1) * sizeof ret->len);
ut_ad(ut_is_2pow(zip_size));
ut_ad(zip_size <= UNIV_PAGE_SIZE);
ret->n_ext = n_ext;
ret->ext = ext;
ret->max_len = DICT_MAX_FIELD_LEN_BY_FORMAT_FLAG(flags);
ret->buf = mem_heap_alloc(heap, n_ext * ret->max_len);
#ifdef UNIV_DEBUG
memset(ret->buf, 0xaa, n_ext * ret->max_len);
UNIV_MEM_ALLOC(ret->buf, n_ext * ret->max_len);
#endif
/* Fetch the BLOB prefixes */
for (i = 0; i < n_ext; i++) {
const dfield_t* dfield;
dfield = dtuple_get_nth_field(tuple, ext[i]);
row_ext_cache_fill(ret, i, zip_size, dfield);
}
return(ret);
}
/**********************************************************************//**
Get the offset of the buddy of a compressed page frame.
@return the buddy relative of page */
UNIV_INLINE
byte*
buf_buddy_get(
/*==========*/
byte* page, /*!< in: compressed page */
ulint size) /*!< in: page size in bytes */
{
ut_ad(ut_is_2pow(size));
ut_ad(size >= BUF_BUDDY_LOW);
ut_ad(size < BUF_BUDDY_HIGH);
ut_ad(!ut_align_offset(page, size));
if (((ulint) page) & size) {
return(page - size);
} else {
return(page + size);
}
}
/********************************************************************//**
Frees memory to a pool. */
UNIV_INTERN
void
mem_area_free(
/*==========*/
void* ptr, /*!< in, own: pointer to allocated memory
buffer */
mem_pool_t* pool) /*!< in: memory pool */
{
mem_area_t* area;
mem_area_t* buddy;
void* new_ptr;
ulint size;
ulint n;
if (UNIV_LIKELY(srv_use_sys_malloc)) {
free(ptr);
return;
}
/* It may be that the area was really allocated from the OS with
regular malloc: check if ptr points within our memory pool */
if ((byte*)ptr < pool->buf || (byte*)ptr >= pool->buf + pool->size) {
ut_free(ptr);
return;
}
area = (mem_area_t*) (((byte*)ptr) - MEM_AREA_EXTRA_SIZE);
if (mem_area_get_free(area)) {
fprintf(stderr,
"InnoDB: Error: Freeing element to mem pool"
" free list though the\n"
"InnoDB: element is marked free!\n");
mem_analyze_corruption(area);
ut_error;
}
size = mem_area_get_size(area);
UNIV_MEM_FREE(ptr, size - MEM_AREA_EXTRA_SIZE);
if (size == 0) {
fprintf(stderr,
"InnoDB: Error: Mem area size is 0. Possibly a"
" memory overrun of the\n"
"InnoDB: previous allocated area!\n");
mem_analyze_corruption(area);
ut_error;
}
#ifdef UNIV_LIGHT_MEM_DEBUG
if (((byte*)area) + size < pool->buf + pool->size) {
ulint next_size;
next_size = mem_area_get_size(
(mem_area_t*)(((byte*)area) + size));
if (UNIV_UNLIKELY(!next_size || !ut_is_2pow(next_size))) {
fprintf(stderr,
"InnoDB: Error: Memory area size %lu,"
" next area size %lu not a power of 2!\n"
"InnoDB: Possibly a memory overrun of"
" the buffer being freed here.\n",
(ulong) size, (ulong) next_size);
mem_analyze_corruption(area);
ut_error;
}
}
#endif
buddy = mem_area_get_buddy(area, size, pool);
n = ut_2_log(size);
mem_pool_mutex_enter(pool);
mem_n_threads_inside++;
ut_a(mem_n_threads_inside == 1);
if (buddy && mem_area_get_free(buddy)
&& (size == mem_area_get_size(buddy))) {
/* The buddy is in a free list */
if ((byte*)buddy < (byte*)area) {
new_ptr = ((byte*)buddy) + MEM_AREA_EXTRA_SIZE;
mem_area_set_size(buddy, 2 * size);
mem_area_set_free(buddy, FALSE);
} else {
new_ptr = ptr;
mem_area_set_size(area, 2 * size);
}
/* Remove the buddy from its free list and merge it to area */
UT_LIST_REMOVE(free_list, pool->free_list[n], buddy);
pool->reserved += ut_2_exp(n);
mem_n_threads_inside--;
mem_pool_mutex_exit(pool);
mem_area_free(new_ptr, pool);
return;
} else {
UT_LIST_ADD_FIRST(free_list, pool->free_list[n], area);
mem_area_set_free(area, TRUE);
ut_ad(pool->reserved >= size);
pool->reserved -= size;
}
mem_n_threads_inside--;
mem_pool_mutex_exit(pool);
ut_ad(mem_pool_validate(pool));
}
/****************************************************************//**
Allocates large pages memory.
@return allocated memory */
UNIV_INTERN
void*
os_mem_alloc_large(
/*===============*/
ulint* n) /*!< in/out: number of bytes */
{
void* ptr;
ulint size;
#if defined HAVE_LARGE_PAGES && defined UNIV_LINUX
int shmid;
struct shmid_ds buf;
if (!os_use_large_pages || !os_large_page_size) {
goto skip;
}
/* Align block size to os_large_page_size */
ut_ad(ut_is_2pow(os_large_page_size));
size = ut_2pow_round(*n + (os_large_page_size - 1),
os_large_page_size);
shmid = shmget(IPC_PRIVATE, (size_t)size, SHM_HUGETLB | SHM_R | SHM_W);
if (shmid < 0) {
fprintf(stderr, "InnoDB: HugeTLB: Warning: Failed to allocate"
" %lu bytes. errno %d\n", size, errno);
ptr = NULL;
} else {
ptr = shmat(shmid, NULL, 0);
if (ptr == (void *)-1) {
fprintf(stderr, "InnoDB: HugeTLB: Warning: Failed to"
" attach shared memory segment, errno %d\n",
errno);
ptr = NULL;
}
/* Remove the shared memory segment so that it will be
automatically freed after memory is detached or
process exits */
shmctl(shmid, IPC_RMID, &buf);
}
if (ptr) {
*n = size;
os_fast_mutex_lock(&ut_list_mutex);
ut_total_allocated_memory += size;
os_fast_mutex_unlock(&ut_list_mutex);
UNIV_MEM_ALLOC(ptr, size);
return(ptr);
}
fprintf(stderr, "InnoDB HugeTLB: Warning: Using conventional"
" memory pool\n");
skip:
#endif /* HAVE_LARGE_PAGES && UNIV_LINUX */
#ifdef __WIN__
SYSTEM_INFO system_info;
GetSystemInfo(&system_info);
/* Align block size to system page size */
ut_ad(ut_is_2pow(system_info.dwPageSize));
/* system_info.dwPageSize is only 32-bit. Casting to ulint is required
on 64-bit Windows. */
size = *n = ut_2pow_round(*n + (system_info.dwPageSize - 1),
(ulint) system_info.dwPageSize);
ptr = VirtualAlloc(NULL, size, MEM_COMMIT | MEM_RESERVE,
PAGE_READWRITE);
if (!ptr) {
fprintf(stderr, "InnoDB: VirtualAlloc(%lu bytes) failed;"
" Windows error %lu\n",
(ulong) size, (ulong) GetLastError());
} else {
os_fast_mutex_lock(&ut_list_mutex);
ut_total_allocated_memory += size;
os_fast_mutex_unlock(&ut_list_mutex);
UNIV_MEM_ALLOC(ptr, size);
}
#elif !defined OS_MAP_ANON
size = *n;
ptr = ut_malloc_low(size, TRUE, FALSE);
#else
# ifdef HAVE_GETPAGESIZE
size = getpagesize();
# else
size = UNIV_PAGE_SIZE;
# endif
/* Align block size to system page size */
ut_ad(ut_is_2pow(size));
size = *n = ut_2pow_round(*n + (size - 1), size);
ptr = mmap(NULL, size, PROT_READ | PROT_WRITE,
MAP_PRIVATE | OS_MAP_ANON, -1, 0);
if (UNIV_UNLIKELY(ptr == (void*) -1)) {
fprintf(stderr, "InnoDB: mmap(%lu bytes) failed;"
" errno %lu\n",
(ulong) size, (ulong) errno);
ptr = NULL;
} else {
os_fast_mutex_lock(&ut_list_mutex);
ut_total_allocated_memory += size;
os_fast_mutex_unlock(&ut_list_mutex);
UNIV_MEM_ALLOC(ptr, size);
}
#endif
return(ptr);
}
