mariadb/mem/mem0pool.c
marko 86b8525254 branches/innodb+: Merge revisions 3602:3931 from branches/zip:
------------------------------------------------------------------------
  r3607 | marko | 2008-12-30 22:33:31 +0200 (Tue, 30 Dec 2008) | 20 lines

  branches/zip: Remove the dependency on the MySQL HASH table implementation.
  Use the InnoDB hash table for keeping track of INNOBASE_SHARE objects.

  struct st_innobase_share: Make table_name const uchar*.  Add the member
  table_name_hash.

  innobase_open_tables: Change the type from HASH to hash_table_t*.

  innobase_get_key(): Remove.

  innobase_fold_name(): New function, for computing the fold value for the
  InnoDB hash table.

  get_share(), free_share(): Use the InnoDB hash functions.

  innobase_end(): Free innobase_open_tables before shutting down InnoDB.
  Shutting down InnoDB will invalidate all memory allocated via InnoDB.

  rb://65 approved by Heikki Tuuri.  This addresses Issue #104.
  ------------------------------------------------------------------------
  r3608 | marko | 2008-12-30 22:45:04 +0200 (Tue, 30 Dec 2008) | 22 lines

  branches/zip: When setting the PAGE_LEVEL of a compressed B-tree page
  from or to 0, compress the page at the same time.  This is necessary,
  because the column information stored on the compressed page will
  differ between leaf and non-leaf pages.  Leaf pages are identified by
  PAGE_LEVEL=0.  This bug was reported as Issue #150.

  Document the similarity between btr_page_create() and
  btr_page_empty().  Make the function signature of btr_page_empty()
  identical with btr_page_create().  (This will add the parameter "level".)

  btr_root_raise_and_insert(): Replace some code with a call to
  btr_page_empty().

  btr_attach_half_pages(): Assert that the page level has already been
  set on both block and new_block.  Do not set it again.

  btr_discard_only_page_on_level(): Document that this function is
  probably never called.  Make it work on any height tree.  (Tested on
  2-high tree by disabling btr_lift_page_up().)

  rb://68
  ------------------------------------------------------------------------
  r3612 | marko | 2009-01-02 11:02:44 +0200 (Fri, 02 Jan 2009) | 14 lines

  branches/zip: Merge c2998 from branches/6.0, so that the same InnoDB Plugin
  source tree will work both under 5.1 and 6.0.  Do not add the test case
  innodb_ctype_ldml.test, because it would not work under MySQL 5.1.

  Refuse to create tables whose columns contain collation IDs above 255.
  This removes an assertion failure that was introduced in WL#4164
  (Two-byte collation IDs).

  create_table_def(): Do not fail an assertion if a column contains a
  charset-collation ID greater than 256. Instead, issue an error and
  refuse to create the table.

  The original change (branches/6.0 r2998) was rb://51 approved by Calvin Sun.
  ------------------------------------------------------------------------
  r3613 | inaam | 2009-01-02 15:10:50 +0200 (Fri, 02 Jan 2009) | 6 lines

  branches/zip: Implement the parameter innodb_use_sys_malloc
  (false by default), for disabling InnoDB's internal memory allocator
  and using system malloc/free instead.

  rb://62 approved by Marko
  ------------------------------------------------------------------------
  r3614 | marko | 2009-01-02 15:55:12 +0200 (Fri, 02 Jan 2009) | 1 line

  branches/zip: ChangeLog: Document r3608 and r3613.
  ------------------------------------------------------------------------
  r3615 | marko | 2009-01-02 15:57:51 +0200 (Fri, 02 Jan 2009) | 1 line

  branches/zip: ChangeLog: Clarify the impact of r3608.
  ------------------------------------------------------------------------
  r3616 | marko | 2009-01-03 00:23:30 +0200 (Sat, 03 Jan 2009) | 1 line

  branches/zip: srv_suspend_mysql_thread(): Add some clarifying comments.
  ------------------------------------------------------------------------
  r3618 | marko | 2009-01-05 12:54:53 +0200 (Mon, 05 Jan 2009) | 15 lines

  branches/zip: Merge revisions 3598:3601 from branches/5.1:

    ------------------------------------------------------------------------
    r3601 | marko | 2008-12-22 16:05:19 +0200 (Mon, 22 Dec 2008) | 9 lines

    branches/5.1: Make
    SET SESSION TRANSACTION ISOLATION LEVEL READ COMMITTED
    a true replacement of SET GLOBAL INNODB_LOCKS_UNSAFE_FOR_BINLOG=1.
    This fixes an error that was introduced in r370, causing
    semi-consistent read not to not unlock rows in READ COMMITTED mode.
    (Bug #41671, Issue #146)

    rb://67 approved by Heikki Tuuri
    ------------------------------------------------------------------------
  ------------------------------------------------------------------------
  r3623 | vasil | 2009-01-06 09:56:32 +0200 (Tue, 06 Jan 2009) | 7 lines

  branches/zip:

  Add patch to fix the failing main.variables mysql-test. It started failing
  after the variable innodb_use_sys_malloc was added because it matches
  '%alloc%' and the test is badly written and expects that no new variables
  like that will ever be added.
  ------------------------------------------------------------------------
  r3795 | marko | 2009-01-07 16:17:47 +0200 (Wed, 07 Jan 2009) | 7 lines

  branches/zip: row_merge_tuple_cmp(): Do not report a duplicate key value
  if any of the fields are NULL.  While the tuples are equal in the
  sorting order, SQL NULL is defined to be logically inequal to
  anything else. (Bug #41904)

  rb://70 approved by Heikki Tuuri
  ------------------------------------------------------------------------
  r3796 | marko | 2009-01-07 16:19:32 +0200 (Wed, 07 Jan 2009) | 1 line

  branches/zip: Add the tests that were forgotten from r3795.
  ------------------------------------------------------------------------
  r3797 | marko | 2009-01-07 16:22:18 +0200 (Wed, 07 Jan 2009) | 22 lines

  branches/zip: Do not call trx_allocate_for_mysql() directly, but use
  helper functions that initialize some members of the transaction struct.
  (Bug #41680)

  innobase_trx_init(): New function: initialize some fields of a
  transaction struct from a MySQL THD object.

  innobase_trx_allocate(): New function: allocate and initialize a
  transaction struct.

  check_trx_exists(): Use the above two functions.

  ha_innobase::delete_table(), ha_innobase::rename_table(),
  ha_innobase::add_index(), ha_innobase::final_drop_index():
  Use innobase_trx_allocate().

  innobase_drop_database(): In the Windows plugin, initialize the trx_t
  specially, because the THD is not available.  Otherwise, use
  innobase_trx_allocate().

  rb://69 accepted by Heikki Tuuri
  ------------------------------------------------------------------------
  r3798 | marko | 2009-01-07 16:42:42 +0200 (Wed, 07 Jan 2009) | 8 lines

  branches/zip: row_merge_drop_temp_indexes(): Do not lock the rows of
  SYS_INDEXES when looking for partially created indexes.  Use the
  transaction isolation level READ UNCOMMITTED to avoid interfering with
  locks held by incomplete transactions that will be rolled back in a
  subsequent step in the recovery.  (Issue #152)

  Approved by Heikki Tuuri
  ------------------------------------------------------------------------
  r3852 | vasil | 2009-01-08 22:10:10 +0200 (Thu, 08 Jan 2009) | 4 lines

  branches/zip:

  Add ChangeLog entries for r3795 r3796 r3797 r3798.

  ------------------------------------------------------------------------
  r3866 | marko | 2009-01-09 15:09:51 +0200 (Fri, 09 Jan 2009) | 2 lines

  branches/zip: buf_flush_try_page(): Move some common code from each
  switch case before the switch block.
  ------------------------------------------------------------------------
  r3867 | marko | 2009-01-09 15:13:14 +0200 (Fri, 09 Jan 2009) | 2 lines

  branches/zip: buf_flush_try_page(): Introduce the variable is_compressed
  for caching the result of buf_page_get_state(bpage) == BUF_BLOCK_FILE_PAGE.
  ------------------------------------------------------------------------
  r3868 | marko | 2009-01-09 15:40:11 +0200 (Fri, 09 Jan 2009) | 4 lines

  branches/zip: buf_flush_insert_into_flush_list(),
  buf_flush_insert_sorted_into_flush_list(): Remove unused code.
  Change the parameter to buf_block_t* block and assert that
  block->state == BUF_BLOCK_FILE_PAGE.  This is part of Issue #155.
  ------------------------------------------------------------------------
  r3873 | marko | 2009-01-09 22:27:40 +0200 (Fri, 09 Jan 2009) | 17 lines

  branches/zip: Some non-functional changes related to Issue #155.

  buf_page_struct: Note that space and offset are also protected by
  buf_pool_mutex.  They are only assigned to by
  buf_block_set_file_page().  Thus, it suffices for buf_flush_batch() to
  hold just buf_pool_mutex when checking these fields.

  buf_flush_try_page(): Rename "locked" to "is_s_latched", per Heikki's request.

  buf_flush_batch(): Move the common statement mutex_exit(block_mutex)
  from all if-else if-else branches before the if block.  Remove the
  redundant test (buf_pool->init_flush[flush_type] == FALSE) that was
  apparently copied from buf_flush_write_complete().

  buf_flush_write_block_low(): Note why it is safe not to hold buf_pool_mutex
  or block_mutex.  Enumerate the assumptions in debug assertions.
  ------------------------------------------------------------------------
  r3874 | marko | 2009-01-09 23:09:06 +0200 (Fri, 09 Jan 2009) | 4 lines

  branches/zip: Add comments related to Issue #155.

  buf_flush_try_page(): Note why it is safe to access bpage without
  holding buf_pool_mutex or block_mutex.
  ------------------------------------------------------------------------
  r3875 | marko | 2009-01-09 23:15:12 +0200 (Fri, 09 Jan 2009) | 11 lines

  branches/zip: Non-functional change: Tighten debug assertions and
  remove dead code.

  buf_flush_ready_for_flush(), buf_flush_try_page(): Assert that
  flush_type is one of BUF_FLUSH_LRU or BUF_FLUSH_LIST.  The flush_type
  comes from buf_flush_batch(), which already asserts this.  The
  assertion holds for all calls in the source code.

  buf_flush_try_page(): Remove the dead case BUF_FLUSH_SINGLE_PAGE
  of switch (flush_type).
  ------------------------------------------------------------------------
  r3879 | marko | 2009-01-12 12:46:44 +0200 (Mon, 12 Jan 2009) | 14 lines

  branches/zip: Simplify the flushing of dirty pages from the buffer pool.

  buf_flush_try_page(): Rename to buf_flush_page(), and change the
  return type to void.  Replace the parameters space, offset with bpage,
  and remove the second page hash lookup.  Note and assert that both
  buf_pool_mutex and block_mutex must now be held upon entering the
  function.  They will still be released by this function.

  buf_flush_try_neighbors(): Replace buf_flush_try_page() with
  buf_flush_page().  Make the logic easier to follow by not negating the
  precondition of buf_flush_page().

  rb://73 approved by Sunny Bains.  This is related to Issue #157.
  ------------------------------------------------------------------------
  r3880 | marko | 2009-01-12 13:24:37 +0200 (Mon, 12 Jan 2009) | 2 lines

  branches/zip: buf_flush_page(): Fix a comment that should have been fixed
  in r3879.  Spotted by Sunny.
  ------------------------------------------------------------------------
  r3881 | marko | 2009-01-12 14:25:22 +0200 (Mon, 12 Jan 2009) | 2 lines

  branches/zip: buf_page_get_newest_modification(): Use the block mutex
  instead of the buffer pool mutex.  This is related to Issue #157.
  ------------------------------------------------------------------------
  r3882 | marko | 2009-01-12 14:40:08 +0200 (Mon, 12 Jan 2009) | 3 lines

  branches/zip: struct mtr_struct: Remove the unused field magic_n
  unless UNIV_DEBUG is defined.  mtr->magic_n is only assigned to
  and checked in UNIV_DEBUG builds.
  ------------------------------------------------------------------------
  r3883 | marko | 2009-01-12 14:48:59 +0200 (Mon, 12 Jan 2009) | 1 line

  branches/zip: Non-functional change: Use ut_d when assigning to mtr->state.
  ------------------------------------------------------------------------
  r3884 | marko | 2009-01-12 18:56:11 +0200 (Mon, 12 Jan 2009) | 16 lines

  branches/zip: Non-functional change: Add some debug assertions and comments.

  buf_page_t: Note that the LRU fields are protected by buf_pool_mutex
  only, not block->mutex or buf_pool_zip_mutex.

  buf_page_get_freed_page_clock(): Note that this is sometimes invoked
  without mutex protection.

  buf_pool_get_oldest_modification(): Note that the result may be out of
  date.

  buf_page_get_LRU_position(), buf_page_is_old(): Assert that the buffer
  pool mutex is being held.

  buf_page_release(): Assert that dirty blocks are in the flush list.
  ------------------------------------------------------------------------
  r3896 | marko | 2009-01-13 09:30:26 +0200 (Tue, 13 Jan 2009) | 2 lines

  branches/zip: buf_flush_try_neighbors(): Fix a bug
  that was introduced in r3879 (rb://73).
  ------------------------------------------------------------------------
  r3900 | marko | 2009-01-13 10:32:24 +0200 (Tue, 13 Jan 2009) | 1 line

  branches/zip: Fix some comments to say buf_pool_mutex.
  ------------------------------------------------------------------------
  r3907 | marko | 2009-01-13 11:54:01 +0200 (Tue, 13 Jan 2009) | 3 lines

  branches/zip: row_merge_create_temporary_table(): On error,
  row_create_table_for_mysql() already frees new_table.
  Do not attempt to free it again.
  ------------------------------------------------------------------------
  r3908 | marko | 2009-01-13 12:34:32 +0200 (Tue, 13 Jan 2009) | 1 line

  branches/zip: Enable HASH_ASSERT_OWNED independently of UNIV_SYNC_DEBUG.
  ------------------------------------------------------------------------
  r3914 | marko | 2009-01-13 21:46:22 +0200 (Tue, 13 Jan 2009) | 37 lines

  branches/zip: In hash table lookups, assert that the traversed items
  satisfy some conditions when UNIV_DEBUG is defined.

  HASH_SEARCH(): New parameter: ASSERTION. All users will pass an appropriate
  ut_ad() or nothing.

  dict_table_add_to_columns(): Assert that the table being added to the data
  dictionary cache is not already being pointed to by the name_hash and
  id_hash tables.

  HASH_SEARCH_ALL(): New macro, for use in dict_table_add_to_columns().

  dict_mem_table_free(): Set ut_d(table->cached = FALSE), so that we can
  check ut_ad(table->cached) when traversing the hash tables, as in
  HASH_SEARCH(name_hash, dict_sys->table_hash, ...) and
  HASH_SEARCH(id_hash, dict_sys->table_id_hash, ...).

  dict_table_get_low(), dict_table_get_on_id_low(): Assert
  ut_ad(!table || table->cached).

  fil_space_get_by_id(): Check ut_ad(space->magic_n == FIL_SPACE_MAGIC_N)
  in HASH_SEARCH(hash, fil_system->spaces, ...).

  fil_space_get_by_name(): Check ut_ad(space->magic_n == FIL_SPACE_MAGIC_N)
  in HASH_SEARCH(name_hash, fil_system->name_hash, ...).

  buf_buddy_block_free(): Check that the blocks are in valid state in
  HASH_SEARCH(hash, buf_pool->zip_hash, ...).

  buf_page_hash_get(): Check that the blocks are in valid state in
  HASH_SEARCH(hash, buf_pool->page_hash, ...).

  get_share(), free_share(): Check ut_ad(share->use_count > 0) in
  HASH_SEARCH(table_name_hash, innobase_open_tables, ...).

  This was posted as rb://75 for tracking down errors similar to Issue #153.
  ------------------------------------------------------------------------
  r3931 | marko | 2009-01-14 16:06:22 +0200 (Wed, 14 Jan 2009) | 26 lines

  branches/zip: Merge revisions 3601:3930 from branches/5.1:

    ------------------------------------------------------------------------
    r3911 | sunny | 2009-01-13 14:15:24 +0200 (Tue, 13 Jan 2009) | 13 lines

    branches/5.1: Fix Bug#38187 Error 153 when creating savepoints
    InnoDB previously treated savepoints as a stack e.g.,
      SAVEPOINT a;
      SAVEPOINT b;
      SAVEPOINT c;
      SAVEPOINT b; <- This would delete b and c.

    This fix changes the behavior to:
      SAVEPOINT a;
      SAVEPOINT b;
      SAVEPOINT c;
      SAVEPOINT b; <- Does not delete savepoint c
    ------------------------------------------------------------------------
    r3930 | marko | 2009-01-14 15:51:30 +0200 (Wed, 14 Jan 2009) | 4 lines

    branches/5.1: dict_load_table(): If dict_load_indexes() fails,
    invoke dict_table_remove_from_cache() instead of dict_mem_table_free(),
    so that the data dictionary will not point to freed data.
    (Bug #42075, Issue #153, rb://76 approved by Heikki Tuuri)
    ------------------------------------------------------------------------
  ------------------------------------------------------------------------
2009-01-14 14:25:45 +00:00

688 lines
17 KiB
C

/************************************************************************
The lowest-level memory management
(c) 1997 Innobase Oy
Created 5/12/1997 Heikki Tuuri
*************************************************************************/
#include "mem0pool.h"
#ifdef UNIV_NONINL
#include "mem0pool.ic"
#endif
#include "srv0srv.h"
#include "sync0sync.h"
#include "ut0mem.h"
#include "ut0lst.h"
#include "ut0byte.h"
#include "mem0mem.h"
/* We would like to use also the buffer frames to allocate memory. This
would be desirable, because then the memory consumption of the database
would be fixed, and we might even lock the buffer pool to the main memory.
The problem here is that the buffer management routines can themselves call
memory allocation, while the buffer pool mutex is reserved.
The main components of the memory consumption are:
1. buffer pool,
2. parsed and optimized SQL statements,
3. data dictionary cache,
4. log buffer,
5. locks for each transaction,
6. hash table for the adaptive index,
7. state and buffers for each SQL query currently being executed,
8. session for each user, and
9. stack for each OS thread.
Items 1 and 2 are managed by an LRU algorithm. Items 5 and 6 can potentially
consume very much memory. Items 7 and 8 should consume quite little memory,
and the OS should take care of item 9, which too should consume little memory.
A solution to the memory management:
1. the buffer pool size is set separately;
2. log buffer size is set separately;
3. the common pool size for all the other entries, except 8, is set separately.
Problems: we may waste memory if the common pool is set too big. Another
problem is the locks, which may take very much space in big transactions.
Then the shared pool size should be set very big. We can allow locks to take
space from the buffer pool, but the SQL optimizer is then unaware of the
usable size of the buffer pool. We could also combine the objects in the
common pool and the buffers in the buffer pool into a single LRU list and
manage it uniformly, but this approach does not take into account the parsing
and other costs unique to SQL statements.
The locks for a transaction can be seen as a part of the state of the
transaction. Hence, they should be stored in the common pool. We still
have the problem of a very big update transaction, for example, which
will set very many x-locks on rows, and the locks will consume a lot
of memory, say, half of the buffer pool size.
Another problem is what to do if we are not able to malloc a requested
block of memory from the common pool. Then we can request memory from
the operating system. If it does not help, a system error results.
Because 5 and 6 may potentially consume very much memory, we let them grow
into the buffer pool. We may let the locks of a transaction take frames
from the buffer pool, when the corresponding memory heap block has grown to
the size of a buffer frame. Similarly for the hash node cells of the locks,
and for the adaptive index. Thus, for each individual transaction, its locks
can occupy at most about the size of the buffer frame of memory in the common
pool, and after that its locks will grow into the buffer pool. */
/* Mask used to extract the free bit from area->size */
#define MEM_AREA_FREE 1
/* The smallest memory area total size */
#define MEM_AREA_MIN_SIZE (2 * MEM_AREA_EXTRA_SIZE)
/* Data structure for a memory pool. The space is allocated using the buddy
algorithm, where free list i contains areas of size 2 to power i. */
struct mem_pool_struct{
byte* buf; /* memory pool */
ulint size; /* memory common pool size */
ulint reserved; /* amount of currently allocated
memory */
mutex_t mutex; /* mutex protecting this struct */
UT_LIST_BASE_NODE_T(mem_area_t)
free_list[64]; /* lists of free memory areas: an
area is put to the list whose number
is the 2-logarithm of the area size */
};
/* The common memory pool */
UNIV_INTERN mem_pool_t* mem_comm_pool = NULL;
/* We use this counter to check that the mem pool mutex does not leak;
this is to track a strange assertion failure reported at
mysql@lists.mysql.com */
UNIV_INTERN ulint mem_n_threads_inside = 0;
/************************************************************************
Reserves the mem pool mutex. */
UNIV_INTERN
void
mem_pool_mutex_enter(void)
/*======================*/
{
mutex_enter(&(mem_comm_pool->mutex));
}
/************************************************************************
Releases the mem pool mutex. */
UNIV_INTERN
void
mem_pool_mutex_exit(void)
/*=====================*/
{
mutex_exit(&(mem_comm_pool->mutex));
}
/************************************************************************
Returns memory area size. */
UNIV_INLINE
ulint
mem_area_get_size(
/*==============*/
/* out: size */
mem_area_t* area) /* in: area */
{
return(area->size_and_free & ~MEM_AREA_FREE);
}
/************************************************************************
Sets memory area size. */
UNIV_INLINE
void
mem_area_set_size(
/*==============*/
mem_area_t* area, /* in: area */
ulint size) /* in: size */
{
area->size_and_free = (area->size_and_free & MEM_AREA_FREE)
| size;
}
/************************************************************************
Returns memory area free bit. */
UNIV_INLINE
ibool
mem_area_get_free(
/*==============*/
/* out: TRUE if free */
mem_area_t* area) /* in: area */
{
#if TRUE != MEM_AREA_FREE
# error "TRUE != MEM_AREA_FREE"
#endif
return(area->size_and_free & MEM_AREA_FREE);
}
/************************************************************************
Sets memory area free bit. */
UNIV_INLINE
void
mem_area_set_free(
/*==============*/
mem_area_t* area, /* in: area */
ibool free) /* in: free bit value */
{
#if TRUE != MEM_AREA_FREE
# error "TRUE != MEM_AREA_FREE"
#endif
area->size_and_free = (area->size_and_free & ~MEM_AREA_FREE)
| free;
}
/************************************************************************
Creates a memory pool. */
UNIV_INTERN
mem_pool_t*
mem_pool_create(
/*============*/
/* out: memory pool */
ulint size) /* in: pool size in bytes */
{
mem_pool_t* pool;
mem_area_t* area;
ulint i;
ulint used;
ut_a(size > 10000);
pool = ut_malloc(sizeof(mem_pool_t));
/* We do not set the memory to zero (FALSE) in the pool,
but only when allocated at a higher level in mem0mem.c.
This is to avoid masking useful Purify warnings. */
pool->buf = ut_malloc_low(size, FALSE, TRUE);
pool->size = size;
mutex_create(&pool->mutex, SYNC_MEM_POOL);
/* Initialize the free lists */
for (i = 0; i < 64; i++) {
UT_LIST_INIT(pool->free_list[i]);
}
used = 0;
while (size - used >= MEM_AREA_MIN_SIZE) {
i = ut_2_log(size - used);
if (ut_2_exp(i) > size - used) {
/* ut_2_log rounds upward */
i--;
}
area = (mem_area_t*)(pool->buf + used);
mem_area_set_size(area, ut_2_exp(i));
mem_area_set_free(area, TRUE);
UNIV_MEM_FREE(MEM_AREA_EXTRA_SIZE + (byte*) area,
ut_2_exp(i) - MEM_AREA_EXTRA_SIZE);
UT_LIST_ADD_FIRST(free_list, pool->free_list[i], area);
used = used + ut_2_exp(i);
}
ut_ad(size >= used);
pool->reserved = 0;
return(pool);
}
/************************************************************************
Fills the specified free list. */
static
ibool
mem_pool_fill_free_list(
/*====================*/
/* out: TRUE if we were able to insert a
block to the free list */
ulint i, /* in: free list index */
mem_pool_t* pool) /* in: memory pool */
{
mem_area_t* area;
mem_area_t* area2;
ibool ret;
ut_ad(mutex_own(&(pool->mutex)));
if (UNIV_UNLIKELY(i >= 63)) {
/* We come here when we have run out of space in the
memory pool: */
return(FALSE);
}
area = UT_LIST_GET_FIRST(pool->free_list[i + 1]);
if (area == NULL) {
if (UT_LIST_GET_LEN(pool->free_list[i + 1]) > 0) {
ut_print_timestamp(stderr);
fprintf(stderr,
" InnoDB: Error: mem pool free list %lu"
" length is %lu\n"
"InnoDB: though the list is empty!\n",
(ulong) i + 1,
(ulong)
UT_LIST_GET_LEN(pool->free_list[i + 1]));
}
ret = mem_pool_fill_free_list(i + 1, pool);
if (ret == FALSE) {
return(FALSE);
}
area = UT_LIST_GET_FIRST(pool->free_list[i + 1]);
}
if (UNIV_UNLIKELY(UT_LIST_GET_LEN(pool->free_list[i + 1]) == 0)) {
mem_analyze_corruption(area);
ut_error;
}
UT_LIST_REMOVE(free_list, pool->free_list[i + 1], area);
area2 = (mem_area_t*)(((byte*)area) + ut_2_exp(i));
UNIV_MEM_ALLOC(area2, MEM_AREA_EXTRA_SIZE);
mem_area_set_size(area2, ut_2_exp(i));
mem_area_set_free(area2, TRUE);
UT_LIST_ADD_FIRST(free_list, pool->free_list[i], area2);
mem_area_set_size(area, ut_2_exp(i));
UT_LIST_ADD_FIRST(free_list, pool->free_list[i], area);
return(TRUE);
}
/************************************************************************
Allocates memory from a pool. NOTE: This low-level function should only be
used in mem0mem.*! */
UNIV_INTERN
void*
mem_area_alloc(
/*===========*/
/* out, own: allocated memory buffer */
ulint* psize, /* in: requested size in bytes; for optimum
space usage, the size should be a power of 2
minus MEM_AREA_EXTRA_SIZE;
out: allocated size in bytes (greater than
or equal to the requested size) */
mem_pool_t* pool) /* in: memory pool */
{
mem_area_t* area;
ulint size;
ulint n;
ibool ret;
/* If we are using os allocator just make a simple call
to malloc */
if (srv_use_sys_malloc) {
return(malloc(*psize));
}
size = *psize;
n = ut_2_log(ut_max(size + MEM_AREA_EXTRA_SIZE, MEM_AREA_MIN_SIZE));
mutex_enter(&(pool->mutex));
mem_n_threads_inside++;
ut_a(mem_n_threads_inside == 1);
area = UT_LIST_GET_FIRST(pool->free_list[n]);
if (area == NULL) {
ret = mem_pool_fill_free_list(n, pool);
if (ret == FALSE) {
/* Out of memory in memory pool: we try to allocate
from the operating system with the regular malloc: */
mem_n_threads_inside--;
mutex_exit(&(pool->mutex));
return(ut_malloc(size));
}
area = UT_LIST_GET_FIRST(pool->free_list[n]);
}
if (!mem_area_get_free(area)) {
fprintf(stderr,
"InnoDB: Error: Removing element from mem pool"
" free list %lu though the\n"
"InnoDB: element is not marked free!\n",
(ulong) n);
mem_analyze_corruption(area);
/* Try to analyze a strange assertion failure reported at
mysql@lists.mysql.com where the free bit IS 1 in the
hex dump above */
if (mem_area_get_free(area)) {
fprintf(stderr,
"InnoDB: Probably a race condition"
" because now the area is marked free!\n");
}
ut_error;
}
if (UT_LIST_GET_LEN(pool->free_list[n]) == 0) {
fprintf(stderr,
"InnoDB: Error: Removing element from mem pool"
" free list %lu\n"
"InnoDB: though the list length is 0!\n",
(ulong) n);
mem_analyze_corruption(area);
ut_error;
}
ut_ad(mem_area_get_size(area) == ut_2_exp(n));
mem_area_set_free(area, FALSE);
UT_LIST_REMOVE(free_list, pool->free_list[n], area);
pool->reserved += mem_area_get_size(area);
mem_n_threads_inside--;
mutex_exit(&(pool->mutex));
ut_ad(mem_pool_validate(pool));
*psize = ut_2_exp(n) - MEM_AREA_EXTRA_SIZE;
UNIV_MEM_ALLOC(MEM_AREA_EXTRA_SIZE + (byte*)area, *psize);
return((void*)(MEM_AREA_EXTRA_SIZE + ((byte*)area)));
}
/************************************************************************
Gets the buddy of an area, if it exists in pool. */
UNIV_INLINE
mem_area_t*
mem_area_get_buddy(
/*===============*/
/* out: the buddy, NULL if no buddy in pool */
mem_area_t* area, /* in: memory area */
ulint size, /* in: memory area size */
mem_pool_t* pool) /* in: memory pool */
{
mem_area_t* buddy;
ut_ad(size != 0);
if (((((byte*)area) - pool->buf) % (2 * size)) == 0) {
/* The buddy is in a higher address */
buddy = (mem_area_t*)(((byte*)area) + size);
if ((((byte*)buddy) - pool->buf) + size > pool->size) {
/* The buddy is not wholly contained in the pool:
there is no buddy */
buddy = NULL;
}
} else {
/* The buddy is in a lower address; NOTE that area cannot
be at the pool lower end, because then we would end up to
the upper branch in this if-clause: the remainder would be
0 */
buddy = (mem_area_t*)(((byte*)area) - size);
}
return(buddy);
}
/************************************************************************
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 (srv_use_sys_malloc) {
return(free(ptr));
}
/* 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);
mutex_enter(&(pool->mutex));
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--;
mutex_exit(&(pool->mutex));
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--;
mutex_exit(&(pool->mutex));
ut_ad(mem_pool_validate(pool));
}
/************************************************************************
Validates a memory pool. */
UNIV_INTERN
ibool
mem_pool_validate(
/*==============*/
/* out: TRUE if ok */
mem_pool_t* pool) /* in: memory pool */
{
mem_area_t* area;
mem_area_t* buddy;
ulint free;
ulint i;
mutex_enter(&(pool->mutex));
free = 0;
for (i = 0; i < 64; i++) {
UT_LIST_VALIDATE(free_list, mem_area_t, pool->free_list[i]);
area = UT_LIST_GET_FIRST(pool->free_list[i]);
while (area != NULL) {
ut_a(mem_area_get_free(area));
ut_a(mem_area_get_size(area) == ut_2_exp(i));
buddy = mem_area_get_buddy(area, ut_2_exp(i), pool);
ut_a(!buddy || !mem_area_get_free(buddy)
|| (ut_2_exp(i) != mem_area_get_size(buddy)));
area = UT_LIST_GET_NEXT(free_list, area);
free += ut_2_exp(i);
}
}
ut_a(free + pool->reserved == pool->size);
mutex_exit(&(pool->mutex));
return(TRUE);
}
/************************************************************************
Prints info of a memory pool. */
UNIV_INTERN
void
mem_pool_print_info(
/*================*/
FILE* outfile,/* in: output file to write to */
mem_pool_t* pool) /* in: memory pool */
{
ulint i;
mem_pool_validate(pool);
fprintf(outfile, "INFO OF A MEMORY POOL\n");
mutex_enter(&(pool->mutex));
for (i = 0; i < 64; i++) {
if (UT_LIST_GET_LEN(pool->free_list[i]) > 0) {
fprintf(outfile,
"Free list length %lu for"
" blocks of size %lu\n",
(ulong) UT_LIST_GET_LEN(pool->free_list[i]),
(ulong) ut_2_exp(i));
}
}
fprintf(outfile, "Pool size %lu, reserved %lu.\n", (ulong) pool->size,
(ulong) pool->reserved);
mutex_exit(&(pool->mutex));
}
/************************************************************************
Returns the amount of reserved memory. */
UNIV_INTERN
ulint
mem_pool_get_reserved(
/*==================*/
/* out: reserved memory in bytes */
mem_pool_t* pool) /* in: memory pool */
{
ulint reserved;
mutex_enter(&(pool->mutex));
reserved = pool->reserved;
mutex_exit(&(pool->mutex));
return(reserved);
}