branches/innodb+: Merge revisions 4072:4150 from branches/zip:

  ------------------------------------------------------------------------
  r4074 | vasil | 2009-01-31 08:05:24 +0200 (Sat, 31 Jan 2009) | 4 lines

  branches/zip:

  Adjust the failing patch patches/information_schema.diff.
  ------------------------------------------------------------------------
  r4076 | vasil | 2009-02-02 09:32:04 +0200 (Mon, 02 Feb 2009) | 4 lines

  branches/zip:

  Add ChangeLog entry for the change in r4072.
  ------------------------------------------------------------------------
  r4077 | marko | 2009-02-02 10:48:05 +0200 (Mon, 02 Feb 2009) | 2 lines

  branches/zip: innobase_start_or_create_for_mysql(): Remove a factual error
  in the function comment.  Parameters are not read from a file "srv_init".
  ------------------------------------------------------------------------
  r4081 | marko | 2009-02-02 14:28:17 +0200 (Mon, 02 Feb 2009) | 4 lines

  branches/zip: Enclose some backup functions in #ifdef UNIV_HOTBACKUP.

  recv_read_cp_info_for_backup(), recv_scan_log_seg_for_backup():
  These functions are only called by InnoDB Hot Backup.
  ------------------------------------------------------------------------
  r4082 | vasil | 2009-02-02 18:24:08 +0200 (Mon, 02 Feb 2009) | 10 lines

  branches/zip:

  Fix a mysql-test failure in innodb-zip:

  main.innodb-zip                          [ fail ]
          Test ended at 2009-02-02 18:13:25

  CURRENT_TEST: main.innodb-zip
  mysqltest: At line 160: Found line beginning with --  that didn't contain a valid mysqltest command, check your syntax or use # if you intended to write a comment
  ------------------------------------------------------------------------
  r4083 | vasil | 2009-02-02 18:33:20 +0200 (Mon, 02 Feb 2009) | 6 lines

  branches/zip:

  Fix the failing innodb-zip test to restore the environment as it was before
  the test execution because a newly added feature in the mysql-test framework
  does check for this.
  ------------------------------------------------------------------------
  r4088 | calvin | 2009-02-03 02:35:56 +0200 (Tue, 03 Feb 2009) | 8 lines

  branches/zip: fix a compiler error and a warning

  Both are minor changes:
  1) Compiler error introduced in r4072: double ';' at the end.
  2) Warning introduced in r3613: \mem\mem0pool.c(481) :
  warning C4098: 'mem_area_free' : 'void' function returning a value

  Approved by: Sunny (IM)
  ------------------------------------------------------------------------
  r4098 | marko | 2009-02-03 09:52:45 +0200 (Tue, 03 Feb 2009) | 4 lines

  branches/zip: mem_area_free(): Correct a bug that was introduced in r4088.
  free() is not the same as ut_free().  ut_free() pairs with ut_malloc(),
  not malloc().  free() pairs with malloc() and some other functions.
  ------------------------------------------------------------------------
  r4114 | marko | 2009-02-04 16:09:24 +0200 (Wed, 04 Feb 2009) | 2 lines

  branches/zip: buf_block_align(): Fix a bogus debug assertion
  that was introduced in r4036, to address Issue #161.
  ------------------------------------------------------------------------
  r4139 | vasil | 2009-02-09 13:47:16 +0200 (Mon, 09 Feb 2009) | 5 lines

  branches/zip:

  Remove mysql-test/patches/bug35261.diff because that bug has been fixed
  in the MySQL repository.
  ------------------------------------------------------------------------
  r4141 | marko | 2009-02-09 15:35:50 +0200 (Mon, 09 Feb 2009) | 1 line

  branches/zip: fil_write_lsn_and_arch_no_to_file(): Plug a memory leak.
  ------------------------------------------------------------------------
  r4144 | inaam | 2009-02-10 01:36:25 +0200 (Tue, 10 Feb 2009) | 9 lines

  branches/zip            rb://30

  This patch changes the innodb mutexes and rw_locks implementation.
  On supported platforms it uses GCC builtin atomics. These changes
  are based on the patch sent by Mark Callaghan of Google under BSD
  license. More technical discussion can be found at rb://30

  Approved by: Heikki
  ------------------------------------------------------------------------
  r4145 | vasil | 2009-02-10 07:34:43 +0200 (Tue, 10 Feb 2009) | 9 lines

  branches/zip:

  Non-functional change: Fix a compilation warning introduced in r4144:

  gcc -DHAVE_CONFIG_H -I. -I../../include -I../../include -I../../include -I../../regex -I../../storage/innobase/include -I../../sql -I.   -Werror         -Wall -g   -MT libinnobase_a-sync0arr.o -MD -MP -MF .deps/libinnobase_a-sync0arr.Tpo -c -o libinnobase_a-sync0arr.o `test -f 'sync/sync0arr.c' || echo './'`sync/sync0arr.c
  cc1: warnings being treated as errors
  sync/sync0arr.c: In function 'sync_array_object_signalled':
  sync/sync0arr.c:869: warning: pointer targets in passing argument 1 of 'os_atomic_increment' differ in signedness
  ------------------------------------------------------------------------
  r4148 | marko | 2009-02-10 10:38:41 +0200 (Tue, 10 Feb 2009) | 12 lines

  branches/zip: Map ut_malloc(), ut_realloc(), ut_free() to
  malloc(), realloc(), free() when innodb_use_sys_malloc is set.

  ut_free_all_mem(): If innodb_use_sys_malloc is set, do nothing,
  because then ut_mem_block_list_inited will never be set.

  log_init(): Use mem_alloc() instead of ut_malloc(), so that the
  memory will be freed.  (Tested with Valgrind, although it is not
  clear why the memory would be freed.)

  rb://86 approved by Heikki Tuuri and Ken Jacobs.  This addresses Issue #168.
  ------------------------------------------------------------------------
  r4149 | marko | 2009-02-10 11:09:15 +0200 (Tue, 10 Feb 2009) | 1 line

  branches/zip: ChangeLog: Document recent changes.
  ------------------------------------------------------------------------
  r4150 | marko | 2009-02-10 11:51:43 +0200 (Tue, 10 Feb 2009) | 6 lines

  branches/zip: get_share(), free_share(): Make table locking case sensitive.
  If lower_case_table_names=1, MySQL will pass the table names in lower case.
  Thus, we can use a binary comparison (strcmp) in the hash table.

  rb://87 approved by Heikki Tuuri, to address Bug #41676 and Issue #167.
  ------------------------------------------------------------------------

ChangeLog

+2009-02-10	The InnoDB Team
+
+	* handler/ha_innodb.h, handler/ha_innodb.cc:
+	Fix Bug#41676 Table names are case insensitive in locking
+
+2009-02-10	The InnoDB Team
+
+	* ut/ut0mem.c:
+	Map ut_malloc_low(), ut_realloc(), and ut_free() directly to
+	malloc(), realloc(), and free() when innodb_use_sys_malloc is set.
+	As a side effect, ut_total_allocated_memory ("Total memory allocated"
+	in the "BUFFER POOL AND MEMORY" section of SHOW ENGINE INNODB STATUS)
+	will exclude any memory allocated by these functions when
+	innodb_use_sys_malloc is set.
+
+2009-02-10	The InnoDB Team
+
+	* btr/btr0cur.c, btr/btr0sea.c, buf/buf0buf.c, handler/ha_innodb.cc,
+	include/buf0buf.ic, include/os0sync.h, include/os0sync.ic,
+	include/srv0srv.h, include/sync0rw.h, include/sync0rw.ic,
+	include/sync0sync.h, include/sync0sync.ic, include/univ.i,
+	row/row0sel.c, srv/srv0srv.c, srv/srv0start.c,
+	sync/sync0arr.c, sync/sync0rw.c, sync/sync0sync.c:
+	On those platforms that support it, implement the synchronization
+	primitives of InnoDB mutexes and read/write locks with GCC atomic
+	builtins instead of Pthreads mutexes and InnoDB mutexes.  These
+	changes are based on a patch supplied by Mark Callaghan of Google
+	under a BSD license.
+
+2009-01-30	The InnoDB Team
+
+	* btr/btr0cur.c, btr/btr0sea.c, buf/buf0buf.c, handler/ha_innodb.cc,
+	include/btr0sea.h, include/buf0buf.h, include/sync0sync.h,
+	sync/sync0sync.c:
+	Make the configuration parameter innodb_adaptive_hash_index dynamic,
+	so that it can be changed at runtime.
+
 2009-01-29	The InnoDB Team
 
 	* handler/ha_innodb.cc, include/ibuf0ibuf.h, include/ibuf0ibuf.ic,
@@ -107,7 +144,10 @@
 	mysql-test/innodb-use-sys-malloc.test:
 	Implement the configuration parameter innodb_use_sys_malloc
 	(false by default), for disabling InnoDB's internal memory allocator
-	and using system malloc/free instead.
+	and using system malloc/free instead.  The "BUFFER POOL AND MEMORY"
+	section of SHOW ENGINE INNODB STATUS will report
+	"in additional pool allocated allocated 0" when
+	innodb_use_sys_malloc is set.
 
 2008-12-30	The InnoDB Team
 

btr/btr0cur.c

@@ -16,6 +16,38 @@ by crashing the database and doing a roll-forward.
 
 Created 10/16/1994 Heikki Tuuri
 *******************************************************/
+/***********************************************************************
+# Copyright (c) 2008, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions
+# are met:
+#	* Redistributions of source code must retain the above copyright
+#	  notice, this list of conditions and the following disclaimer.
+#	* Redistributions in binary form must reproduce the above
+#	  copyright notice, this list of conditions and the following
+#	  disclaimer in the documentation and/or other materials
+#	  provided with the distribution.
+#	* Neither the name of the Google Inc. nor the names of its
+#	  contributors may be used to endorse or promote products
+#	  derived from this software without specific prior written
+#	  permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Note, the BSD license applies to the new code. The old code is GPL.
+***********************************************************************/
 
 #include "btr0cur.h"
 
@@ -416,7 +448,7 @@ btr_cur_search_to_nth_level(
 
 	/* Ibuf does not use adaptive hash; this is prevented by the
 	latch_mode check below. */
-	if (btr_search_latch.writer == RW_LOCK_NOT_LOCKED
+	if (rw_lock_get_writer(&btr_search_latch) == RW_LOCK_NOT_LOCKED
 	    && latch_mode <= BTR_MODIFY_LEAF
 	    && info->last_hash_succ
 	    && !estimate

btr/btr0sea.c

@@ -6,6 +6,39 @@ The index tree adaptive search
 Created 2/17/1996 Heikki Tuuri
 *************************************************************************/
 
+/***********************************************************************
+# Copyright (c) 2008, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions
+# are met:
+#	* Redistributions of source code must retain the above copyright
+#	  notice, this list of conditions and the following disclaimer.
+#	* Redistributions in binary form must reproduce the above
+#	  copyright notice, this list of conditions and the following
+#	  disclaimer in the documentation and/or other materials
+#	  provided with the distribution.
+#	* Neither the name of the Google Inc. nor the names of its
+#	  contributors may be used to endorse or promote products
+#	  derived from this software without specific prior written
+#	  permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Note, the BSD license applies to the new code. The old code is GPL.
+***********************************************************************/
+
 #include "btr0sea.h"
 #ifdef UNIV_NONINL
 #include "btr0sea.ic"
@@ -820,8 +853,8 @@ btr_search_guess_on_hash(
 		}
 	}
 
-	ut_ad(btr_search_latch.writer != RW_LOCK_EX);
-	ut_ad(btr_search_latch.reader_count > 0);
+	ut_ad(rw_lock_get_writer(&btr_search_latch) != RW_LOCK_EX);
+	ut_ad(rw_lock_get_reader_count(&btr_search_latch) > 0);
 
 	rec = ha_search_and_get_data(btr_search_sys->hash_index, fold);
 

buf/buf0buf.c

@@ -19,6 +19,38 @@ The database buffer buf_pool
 
 Created 11/5/1995 Heikki Tuuri
 *******************************************************/
+/***********************************************************************
+# Copyright (c) 2008, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions
+# are met:
+#	* Redistributions of source code must retain the above copyright
+#	  notice, this list of conditions and the following disclaimer.
+#	* Redistributions in binary form must reproduce the above
+#	  copyright notice, this list of conditions and the following
+#	  disclaimer in the documentation and/or other materials
+#	  provided with the distribution.
+#	* Neither the name of the Google Inc. nor the names of its
+#	  contributors may be used to endorse or promote products
+#	  derived from this software without specific prior written
+#	  permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Note, the BSD license applies to the new code. The old code is GPL.
+***********************************************************************/
 
 #include "buf0buf.h"
 
@@ -1016,7 +1048,7 @@ buf_pool_drop_hash_index(void)
 	ut_ad(!btr_search_enabled);
 
 	do {
-		buf_chunk_t*	chunks	= buf_pool->chunks;;
+		buf_chunk_t*	chunks	= buf_pool->chunks;
 		buf_chunk_t*	chunk	= chunks + buf_pool->n_chunks;
 
 		released_search_latch = FALSE;
@@ -1924,12 +1956,52 @@ buf_block_align(
 			buf_block_init() so that block[n].frame ==
 			block->frame + n * UNIV_PAGE_SIZE.  Check it. */
 			ut_ad(block->frame == page_align(ptr));
-			/* The space id and page number should be
-			stamped on the page. */
-			ut_ad(block->page.space
-			      == page_get_space_id(page_align(ptr)));
-			ut_ad(block->page.offset
-			      == page_get_page_no(page_align(ptr)));
+#ifdef UNIV_DEBUG
+			/* A thread that updates these fields must
+			hold buf_pool_mutex and block->mutex.  Acquire
+			only the latter. */
+			mutex_enter(&block->mutex);
+
+			switch (buf_block_get_state(block)) {
+			case BUF_BLOCK_ZIP_FREE:
+			case BUF_BLOCK_ZIP_PAGE:
+			case BUF_BLOCK_ZIP_DIRTY:
+				/* These types should only be used in
+				the compressed buffer pool, whose
+				memory is allocated from
+				buf_pool->chunks, in UNIV_PAGE_SIZE
+				blocks flagged as BUF_BLOCK_MEMORY. */
+				ut_error;
+				break;
+			case BUF_BLOCK_NOT_USED:
+			case BUF_BLOCK_READY_FOR_USE:
+			case BUF_BLOCK_MEMORY:
+				/* Some data structures contain
+				"guess" pointers to file pages.  The
+				file pages may have been freed and
+				reused.  Do not complain. */
+				break;
+			case BUF_BLOCK_REMOVE_HASH:
+				/* buf_LRU_block_remove_hashed_page()
+				will overwrite the FIL_PAGE_OFFSET and
+				FIL_PAGE_ARCH_LOG_NO_OR_SPACE_ID with
+				0xff and set the state to
+				BUF_BLOCK_REMOVE_HASH. */
+				ut_ad(page_get_space_id(page_align(ptr))
+				      == 0xffffffff);
+				ut_ad(page_get_page_no(page_align(ptr))
+				      == 0xffffffff);
+				break;
+			case BUF_BLOCK_FILE_PAGE:
+				ut_ad(block->page.space
+				      == page_get_space_id(page_align(ptr)));
+				ut_ad(block->page.offset
+				      == page_get_page_no(page_align(ptr)));
+				break;
+			}
+
+			mutex_exit(&block->mutex);
+#endif /* UNIV_DEBUG */
 
 			return(block);
 		}
@@ -2366,8 +2438,8 @@ buf_page_optimistic_get_func(
 			   buf_block_get_page_no(block), NULL));
 
 	if (rw_latch == RW_S_LATCH) {
-		success = rw_lock_s_lock_func_nowait(&(block->lock),
-						     file, line);
+		success = rw_lock_s_lock_nowait(&(block->lock),
+						file, line);
 		fix_type = MTR_MEMO_PAGE_S_FIX;
 	} else {
 		success = rw_lock_x_lock_func_nowait(&(block->lock),
@@ -2478,8 +2550,8 @@ buf_page_get_known_nowait(
 	ut_ad(!ibuf_inside() || (mode == BUF_KEEP_OLD));
 
 	if (rw_latch == RW_S_LATCH) {
-		success = rw_lock_s_lock_func_nowait(&(block->lock),
-						     file, line);
+		success = rw_lock_s_lock_nowait(&(block->lock),
+						file, line);
 		fix_type = MTR_MEMO_PAGE_S_FIX;
 	} else {
 		success = rw_lock_x_lock_func_nowait(&(block->lock),
@@ -2556,7 +2628,7 @@ buf_page_try_get_func(
 	mutex_exit(&block->mutex);
 
 	fix_type = MTR_MEMO_PAGE_S_FIX;
-	success = rw_lock_s_lock_func_nowait(&block->lock, file, line);
+	success = rw_lock_s_lock_nowait(&block->lock, file, line);
 
 	if (!success) {
 		/* Let us try to get an X-latch. If the current thread

fil/fil0fil.c

@@ -1678,6 +1678,8 @@ fil_write_lsn_and_arch_no_to_file(
 
 	fil_write(TRUE, 0, 0, sum_of_sizes, 0, UNIV_PAGE_SIZE, buf, NULL);
 
+	mem_free(buf1);
+
 	return(DB_SUCCESS);
 }
 

handler/ha_innodb.cc

@@ -13,6 +13,38 @@
    along with this program; if not, write to the Free Software
    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307	 USA */
 
+/***********************************************************************
+# Copyright (c) 2008, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions
+# are met:
+#	* Redistributions of source code must retain the above copyright
+#	  notice, this list of conditions and the following disclaimer.
+#	* Redistributions in binary form must reproduce the above
+#	  copyright notice, this list of conditions and the following
+#	  disclaimer in the documentation and/or other materials
+#	  provided with the distribution.
+#	* Neither the name of the Google Inc. nor the names of its
+#	  contributors may be used to endorse or promote products
+#	  derived from this software without specific prior written
+#	  permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Note, the BSD license applies to the new code. The old code is GPL.
+***********************************************************************/
 /* TODO list for the InnoDB handler in 5.0:
   - Remove the flag trx->active_trans and look at trx->conc_state
   - fix savepoint functions to use savepoint storage area
@@ -466,6 +498,8 @@ static SHOW_VAR innodb_status_variables[]= {
   (char*) &export_vars.innodb_dblwr_pages_written,	  SHOW_LONG},
   {"dblwr_writes",
   (char*) &export_vars.innodb_dblwr_writes,		  SHOW_LONG},
+  {"have_atomic_builtins",
+  (char*) &export_vars.innodb_have_atomic_builtins,	  SHOW_BOOL},
   {"log_waits",
   (char*) &export_vars.innodb_log_waits,		  SHOW_LONG},
   {"log_write_requests",
@@ -8045,7 +8079,8 @@ innodb_mutex_show_status(
 	stat_print_fn*	stat_print)
 {
 	char buf1[IO_SIZE], buf2[IO_SIZE];
-	mutex_t*  mutex;
+	mutex_t*	mutex;
+	rw_lock_t*	lock;
 #ifdef UNIV_DEBUG
 	ulint	  rw_lock_count= 0;
 	ulint	  rw_lock_count_spin_loop= 0;
@@ -8116,6 +8151,29 @@ innodb_mutex_show_status(
 
 	mutex_exit(&mutex_list_mutex);
 
+	mutex_enter(&rw_lock_list_mutex);
+
+	lock = UT_LIST_GET_FIRST(rw_lock_list);
+
+	while (lock != NULL) {
+		if (lock->count_os_wait) {
+			buf1len= my_snprintf(buf1, sizeof(buf1), "%s:%lu",
+                                    lock->cfile_name, (ulong) lock->cline);
+			buf2len= my_snprintf(buf2, sizeof(buf2),
+                                    "os_waits=%lu", lock->count_os_wait);
+
+			if (stat_print(thd, innobase_hton_name,
+				       hton_name_len, buf1, buf1len,
+				       buf2, buf2len)) {
+				mutex_exit(&rw_lock_list_mutex);
+				DBUG_RETURN(1);
+			}
+		}
+		lock = UT_LIST_GET_NEXT(list, lock);
+	}
+
+	mutex_exit(&rw_lock_list_mutex);
+
 #ifdef UNIV_DEBUG
 	buf2len= my_snprintf(buf2, sizeof(buf2),
 		"count=%lu, spin_waits=%lu, spin_rounds=%lu, "
@@ -8151,56 +8209,35 @@ bool innobase_show_status(handlerton *hton, THD* thd,
 	}
 }
 
-
 /****************************************************************************
  Handling the shared INNOBASE_SHARE structure that is needed to provide table
  locking.
 ****************************************************************************/
 
-/****************************************************************************
-Folds a string in system_charset_info. */
-static
-ulint
-innobase_fold_name(
-/*===============*/
-				/* out: fold value of the name */
-	const uchar*	name,	/* in: string to be folded */
-	size_t		length)	/* in: length of the name in bytes */
-{
-	ulong n1 = 1, n2 = 4;
-
-	system_charset_info->coll->hash_sort(system_charset_info,
-					     name, length, &n1, &n2);
-	return((ulint) n1);
-}
-
 static INNOBASE_SHARE* get_share(const char* table_name)
 {
 	INNOBASE_SHARE *share;
 	pthread_mutex_lock(&innobase_share_mutex);
-	uint length=(uint) strlen(table_name);
 
-	ulint	fold = innobase_fold_name((const uchar*) table_name, length);
+	ulint	fold = ut_fold_string(table_name);
 
 	HASH_SEARCH(table_name_hash, innobase_open_tables, fold,
 		    INNOBASE_SHARE*, share,
 		    ut_ad(share->use_count > 0),
-		    !my_strnncoll(system_charset_info,
-				  share->table_name,
-				  share->table_name_length,
-				  (const uchar*) table_name, length));
+		    !strcmp(share->table_name, table_name));
 
 	if (!share) {
 
+		uint length = (uint) strlen(table_name);
+
 		/* TODO: invoke HASH_MIGRATE if innobase_open_tables
 		grows too big */
 
 		share = (INNOBASE_SHARE *) my_malloc(sizeof(*share)+length+1,
 			MYF(MY_FAE | MY_ZEROFILL));
 
-		share->table_name_length = length;
-		share->table_name = (uchar*) memcpy(share + 1,
-						    table_name, length + 1);
+		share->table_name = (char*) memcpy(share + 1,
+						   table_name, length + 1);
 
 		HASH_INSERT(INNOBASE_SHARE, table_name_hash,
 			    innobase_open_tables, fold, share);
@@ -8221,24 +8258,18 @@ static void free_share(INNOBASE_SHARE* share)
 
 #ifdef UNIV_DEBUG
 	INNOBASE_SHARE* share2;
-	ulint fold = innobase_fold_name(share->table_name,
-					share->table_name_length);
+	ulint	fold = ut_fold_string(share->table_name);
 
 	HASH_SEARCH(table_name_hash, innobase_open_tables, fold,
 		    INNOBASE_SHARE*, share2,
 		    ut_ad(share->use_count > 0),
-		    !my_strnncoll(system_charset_info,
-				  share->table_name,
-				  share->table_name_length,
-				  share2->table_name,
-				  share2->table_name_length));
+		    !strcmp(share->table_name, share2->table_name));
 
 	ut_a(share2 == share);
 #endif /* UNIV_DEBUG */
 
 	if (!--share->use_count) {
-		ulint	fold = innobase_fold_name(share->table_name,
-						  share->table_name_length);
+		ulint	fold = ut_fold_string(share->table_name);
 
 		HASH_DELETE(INNOBASE_SHARE, table_name_hash,
 			    innobase_open_tables, fold, share);

handler/ha_innodb.h

@@ -27,8 +27,8 @@
 typedef struct st_innobase_share {
   THR_LOCK lock;
   pthread_mutex_t mutex;
-  const uchar *table_name;
-  uint table_name_length,use_count;
+  const char* table_name;
+  uint use_count;
   void* table_name_hash;
 } INNOBASE_SHARE;
 

include/buf0buf.ic

@@ -5,6 +5,38 @@ The database buffer buf_pool
 
 Created 11/5/1995 Heikki Tuuri
 *******************************************************/
+/***********************************************************************
+# Copyright (c) 2008, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions
+# are met:
+#	* Redistributions of source code must retain the above copyright
+#	  notice, this list of conditions and the following disclaimer.
+#	* Redistributions in binary form must reproduce the above
+#	  copyright notice, this list of conditions and the following
+#	  disclaimer in the documentation and/or other materials
+#	  provided with the distribution.
+#	* Neither the name of the Google Inc. nor the names of its
+#	  contributors may be used to endorse or promote products
+#	  derived from this software without specific prior written
+#	  permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Note, the BSD license applies to the new code. The old code is GPL.
+***********************************************************************/
 
 #include "buf0flu.h"
 #include "buf0lru.h"
@@ -851,7 +883,7 @@ buf_block_buf_fix_inc_func(
 #ifdef UNIV_SYNC_DEBUG
 	ibool	ret;
 
-	ret = rw_lock_s_lock_func_nowait(&(block->debug_latch), file, line);
+	ret = rw_lock_s_lock_nowait(&(block->debug_latch), file, line);
 	ut_a(ret);
 #endif /* UNIV_SYNC_DEBUG */
 	ut_ad(mutex_own(&block->mutex));

include/log0recv.h

@@ -17,7 +17,6 @@ Created 9/20/1997 Heikki Tuuri
 
 #ifdef UNIV_HOTBACKUP
 extern ibool	recv_replay_file_ops;
-#endif /* UNIV_HOTBACKUP */
 
 /***********************************************************************
 Reads the checkpoint info needed in hot backup. */
@@ -55,6 +54,7 @@ recv_scan_log_seg_for_backup(
 	ulint*		n_bytes_scanned);/* out: how much we were able to
 					scan, smaller than buf_len if log
 					data ended here */
+#endif /* UNIV_HOTBACKUP */
 /***********************************************************************
 Returns TRUE if recovery is currently running. */
 UNIV_INLINE

include/os0sync.h

@@ -6,6 +6,38 @@ synchronization primitives.
 
 Created 9/6/1995 Heikki Tuuri
 *******************************************************/
+/***********************************************************************
+# Copyright (c) 2008, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions
+# are met:
+#	* Redistributions of source code must retain the above copyright
+#	  notice, this list of conditions and the following disclaimer.
+#	* Redistributions in binary form must reproduce the above
+#	  copyright notice, this list of conditions and the following
+#	  disclaimer in the documentation and/or other materials
+#	  provided with the distribution.
+#	* Neither the name of the Google Inc. nor the names of its
+#	  contributors may be used to endorse or promote products
+#	  derived from this software without specific prior written
+#	  permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Note, the BSD license applies to the new code. The old code is GPL.
+***********************************************************************/
 #ifndef os0sync_h
 #define os0sync_h
 
@@ -261,6 +293,29 @@ os_fast_mutex_free(
 /*===============*/
 	os_fast_mutex_t*	fast_mutex);	/* in: mutex to free */
 
+#ifdef HAVE_GCC_ATOMIC_BUILTINS
+/**************************************************************
+Atomic compare-and-swap for InnoDB. Currently requires GCC atomic builtins. */
+UNIV_INLINE
+ibool
+os_compare_and_swap(
+/*================*/
+						/* out: true if swapped */
+	volatile lint*		ptr,		/* in: pointer to target */
+	lint			oldVal,		/* in: value to compare to */
+	lint			newVal);	/* in: value to swap in */
+/**************************************************************
+Atomic increment for InnoDB. Currently requires GCC atomic builtins. */
+UNIV_INLINE
+lint
+os_atomic_increment(
+/*================*/
+						/* out: resulting value */
+	volatile lint*		ptr,		/* in: pointer to target */
+	lint			amount);	/* in: amount of increment */
+
+#endif /* HAVE_GCC_ATOMIC_BUILTINS */
+
 #ifndef UNIV_NONINL
 #include "os0sync.ic"
 #endif

include/os0sync.ic

@@ -5,6 +5,38 @@ The interface to the operating system synchronization primitives.
 
 Created 9/6/1995 Heikki Tuuri
 *******************************************************/
+/***********************************************************************
+# Copyright (c) 2008, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions
+# are met:
+#	* Redistributions of source code must retain the above copyright
+#	  notice, this list of conditions and the following disclaimer.
+#	* Redistributions in binary form must reproduce the above
+#	  copyright notice, this list of conditions and the following
+#	  disclaimer in the documentation and/or other materials
+#	  provided with the distribution.
+#	* Neither the name of the Google Inc. nor the names of its
+#	  contributors may be used to endorse or promote products
+#	  derived from this software without specific prior written
+#	  permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Note, the BSD license applies to the new code. The old code is GPL.
+***********************************************************************/
 
 #ifdef __WIN__
 #include <winbase.h>
@@ -44,3 +76,38 @@ os_fast_mutex_trylock(
 #endif
 #endif
 }
+
+#ifdef HAVE_GCC_ATOMIC_BUILTINS
+/**************************************************************
+Atomic compare-and-swap for InnoDB. Currently requires GCC atomic builtins. */
+UNIV_INLINE
+ibool
+os_compare_and_swap(
+/*================*/
+					/* out: true if swapped */
+	volatile lint*	ptr,		/* in: pointer to target */
+	lint		oldVal,		/* in: value to compare to */
+	lint		newVal)		/* in: value to swap in */
+{
+	if(__sync_bool_compare_and_swap(ptr, oldVal, newVal)) {
+		return(TRUE);
+	}
+
+	return(FALSE);
+}
+
+/**************************************************************
+Atomic increment for InnoDB. Currently requires GCC atomic builtins. */
+UNIV_INLINE
+lint
+os_atomic_increment(
+/*================*/
+					/* out: resulting value */
+	volatile lint*	ptr,		/* in: pointer to target */
+	lint		amount)		/* in: amount of increment */
+{
+	lint newVal = __sync_add_and_fetch(ptr, amount);
+	return newVal;
+}
+
+#endif /* HAVE_GCC_ATOMIC_BUILTINS */

include/srv0srv.h

@@ -5,7 +5,38 @@ The server main program
 
 Created 10/10/1995 Heikki Tuuri
 *******************************************************/
-
+/***********************************************************************
+# Copyright (c) 2008, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions
+# are met:
+#	* Redistributions of source code must retain the above copyright
+#	  notice, this list of conditions and the following disclaimer.
+#	* Redistributions in binary form must reproduce the above
+#	  copyright notice, this list of conditions and the following
+#	  disclaimer in the documentation and/or other materials
+#	  provided with the distribution.
+#	* Neither the name of the Google Inc. nor the names of its
+#	  contributors may be used to endorse or promote products
+#	  derived from this software without specific prior written
+#	  permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Note, the BSD license applies to the new code. The old code is GPL.
+***********************************************************************/
 
 #ifndef srv0srv_h
 #define srv0srv_h
@@ -533,6 +564,7 @@ struct export_var_struct{
 	ulint innodb_buffer_pool_read_ahead_rnd;
 	ulint innodb_dblwr_pages_written;
 	ulint innodb_dblwr_writes;
+	ibool innodb_have_atomic_builtins;
 	ulint innodb_log_waits;
 	ulint innodb_log_write_requests;
 	ulint innodb_log_writes;
@@ -569,4 +601,3 @@ struct srv_sys_struct{
 extern ulint	srv_n_threads_active[];
 
 #endif
-

include/srv0start.h

@@ -65,8 +65,7 @@ srv_add_path_separator_if_needed(
 	char*	str);	/* in: null-terminated character string */
 /********************************************************************
 Starts Innobase and creates a new database if database files
-are not found and the user wants. Server parameters are
-read from a file of name "srv_init" in the ib_home directory. */
+are not found and the user wants. */
 UNIV_INTERN
 int
 innobase_start_or_create_for_mysql(void);

include/sync0rw.h

@@ -5,6 +5,38 @@ The read-write lock (for threads, not for database transactions)
 
 Created 9/11/1995 Heikki Tuuri
 *******************************************************/
+/***********************************************************************
+# Copyright (c) 2008, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions
+# are met:
+#	* Redistributions of source code must retain the above copyright
+#	  notice, this list of conditions and the following disclaimer.
+#	* Redistributions in binary form must reproduce the above
+#	  copyright notice, this list of conditions and the following
+#	  disclaimer in the documentation and/or other materials
+#	  provided with the distribution.
+#	* Neither the name of the Google Inc. nor the names of its
+#	  contributors may be used to endorse or promote products
+#	  derived from this software without specific prior written
+#	  permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Note, the BSD license applies to the new code. The old code is GPL.
+***********************************************************************/
 
 #ifndef sync0rw_h
 #define sync0rw_h
@@ -24,6 +56,12 @@ smaller than 30 and the order of the numerical values like below! */
 #define	RW_X_LATCH	2
 #define	RW_NO_LATCH	3
 
+/* We decrement lock_word by this amount for each x_lock. It is also the
+start value for the lock_word, meaning that it limits the maximum number
+of concurrent read locks before the rw_lock breaks. The current value of
+0x00100000 allows 1,048,575 concurrent readers and 2047 recursive writers.*/
+#define X_LOCK_DECR		0x00100000
+
 typedef struct rw_lock_struct		rw_lock_t;
 #ifdef UNIV_SYNC_DEBUG
 typedef struct rw_lock_debug_struct	rw_lock_debug_t;
@@ -47,14 +85,14 @@ extern ibool		rw_lock_debug_waiters;	/* This is set to TRUE, if
 					there may be waiters for the event */
 #endif /* UNIV_SYNC_DEBUG */
 
-extern	ulint	rw_s_system_call_count;
-extern	ulint	rw_s_spin_wait_count;
-extern	ulint	rw_s_exit_count;
-extern	ulint	rw_s_os_wait_count;
-extern	ulint	rw_x_system_call_count;
-extern	ulint	rw_x_spin_wait_count;
-extern	ulint	rw_x_os_wait_count;
-extern	ulint	rw_x_exit_count;
+extern	ib_int64_t	rw_s_spin_wait_count;
+extern	ib_int64_t	rw_s_spin_round_count;
+extern	ib_int64_t	rw_s_exit_count;
+extern	ib_int64_t	rw_s_os_wait_count;
+extern	ib_int64_t	rw_x_spin_wait_count;
+extern	ib_int64_t	rw_x_spin_round_count;
+extern	ib_int64_t	rw_x_os_wait_count;
+extern	ib_int64_t	rw_x_exit_count;
 
 /**********************************************************************
 Creates, or rather, initializes an rw-lock object in a specified memory
@@ -127,8 +165,22 @@ corresponding function. */
 NOTE! The following macros should be used in rw s-locking, not the
 corresponding function. */
 
-#define rw_lock_s_lock_nowait(M)	rw_lock_s_lock_func_nowait(\
-		(M), __FILE__, __LINE__)
+#define rw_lock_s_lock_nowait(M, F, L)    rw_lock_s_lock_low(\
+					  (M), 0, (F), (L))
+/**********************************************************************
+Low-level function which tries to lock an rw-lock in s-mode. Performs no
+spinning. */
+UNIV_INLINE
+ibool
+rw_lock_s_lock_low(
+/*===============*/
+				/* out: TRUE if success */
+	rw_lock_t*	lock,	/* in: pointer to rw-lock */
+	ulint		pass __attribute__((unused)),
+				/* in: pass value; != 0, if the lock will be
+				passed to another thread to unlock */
+	const char*	file_name, /* in: file name where lock requested */
+	ulint		line);	/* in: line where requested */
 /**********************************************************************
 NOTE! Use the corresponding macro, not directly this function, except if
 you supply the file name and line number. Lock an rw-lock in shared mode
@@ -146,18 +198,6 @@ rw_lock_s_lock_func(
 	const char*	file_name,/* in: file name where lock requested */
 	ulint		line);	/* in: line where requested */
 /**********************************************************************
-NOTE! Use the corresponding macro, not directly this function, except if
-you supply the file name and line number. Lock an rw-lock in shared mode
-for the current thread if the lock can be acquired immediately. */
-UNIV_INLINE
-ibool
-rw_lock_s_lock_func_nowait(
-/*=======================*/
-				/* out: TRUE if success */
-	rw_lock_t*	lock,	/* in: pointer to rw-lock */
-	const char*	file_name,/* in: file name where lock requested */
-	ulint		line);	/* in: line where requested */
-/**********************************************************************
 NOTE! Use the corresponding macro, not directly this function! Lock an
 rw-lock in exclusive mode for the current thread if the lock can be
 obtained immediately. */
@@ -341,6 +381,41 @@ ulint
 rw_lock_get_reader_count(
 /*=====================*/
 	rw_lock_t*	lock);
+/**********************************************************************
+Decrements lock_word the specified amount if it is greater than 0.
+This is used by both s_lock and x_lock operations. */
+UNIV_INLINE
+ibool
+rw_lock_lock_word_decr(
+/*===================*/
+					/* out: TRUE if decr occurs */
+	rw_lock_t*	lock,		/* in: rw-lock */
+	ulint		amount);	/* in: amount to decrement */
+/**********************************************************************
+Increments lock_word the specified amount and returns new value. */
+UNIV_INLINE
+lint
+rw_lock_lock_word_incr(
+/*===================*/
+					/* out: TRUE if decr occurs */
+	rw_lock_t*	lock,
+	ulint		amount);	/* in: rw-lock */
+/**********************************************************************
+This function sets the lock->writer_thread and lock->recursive fields.
+For platforms where we are using atomic builtins instead of lock->mutex
+it sets the lock->writer_thread field using atomics to ensure memory
+ordering. Note that it is assumed that the caller of this function
+effectively owns the lock i.e.: nobody else is allowed to modify
+lock->writer_thread at this point in time.
+The protocol is that lock->writer_thread MUST be updated BEFORE the
+lock->recursive flag is set. */
+UNIV_INLINE
+void
+rw_lock_set_writer_id_and_recursion_flag(
+/*=====================================*/
+	rw_lock_t*	lock,		/* in/out: lock to work on */
+	ibool		recursive);	/* in: TRUE if recursion
+					allowed */
 #ifdef UNIV_SYNC_DEBUG
 /**********************************************************************
 Checks if the thread has locked the rw-lock in the specified mode, with
@@ -417,44 +492,37 @@ Do not use its fields directly! The structure used in the spin lock
 implementation of a read-write lock. Several threads may have a shared lock
 simultaneously in this lock, but only one writer may have an exclusive lock,
 in which case no shared locks are allowed. To prevent starving of a writer
-blocked by readers, a writer may queue for the lock by setting the writer
-field. Then no new readers are allowed in. */
+blocked by readers, a writer may queue for x-lock by decrementing lock_word:
+no new readers will be let in while the thread waits for readers to exit. */
 
 struct rw_lock_struct {
+	volatile lint	lock_word;
+				/* Holds the state of the lock. */
+	volatile ulint	waiters;/* 1: there are waiters */
+	volatile ibool	recursive;/* Default value FALSE which means the lock
+				is non-recursive. The value is typically set
+				to TRUE making normal rw_locks recursive. In
+				case of asynchronous IO, when a non-zero
+				value of 'pass' is passed then we keep the
+				lock non-recursive.
+				This flag also tells us about the state of
+				writer_thread field. If this flag is set
+				then writer_thread MUST contain the thread
+				id of the current x-holder or wait-x thread.
+				This flag must be reset in x_unlock
+				functions before incrementing the lock_word */
+	volatile os_thread_id_t	writer_thread;
+				/* Thread id of writer thread. Is only
+				guaranteed to have sane and non-stale
+				value iff recursive flag is set. */
 	os_event_t	event;	/* Used by sync0arr.c for thread queueing */
-
-#ifdef __WIN__
-	os_event_t	wait_ex_event;	/* This windows specific event is
-				used by the thread which has set the
-				lock state to RW_LOCK_WAIT_EX. The
-				rw_lock design guarantees that this
-				thread will be the next one to proceed
-				once the current the event gets
-				signalled. See LEMMA 2 in sync0sync.c */
-#endif
-
-	ulint	reader_count;	/* Number of readers who have locked this
-				lock in the shared mode */
-	ulint	writer;		/* This field is set to RW_LOCK_EX if there
-				is a writer owning the lock (in exclusive
-				mode), RW_LOCK_WAIT_EX if a writer is
-				queueing for the lock, and
-				RW_LOCK_NOT_LOCKED, otherwise. */
-	os_thread_id_t	writer_thread;
-				/* Thread id of a possible writer thread */
-	ulint	writer_count;	/* Number of times the same thread has
-				recursively locked the lock in the exclusive
-				mode */
+	os_event_t	wait_ex_event;
+				/* Event for next-writer to wait on. A thread
+				must decrement lock_word before waiting. */
+#ifndef INNODB_RW_LOCKS_USE_ATOMICS
 	mutex_t	mutex;		/* The mutex protecting rw_lock_struct */
-	ulint	pass;		/* Default value 0. This is set to some
-				value != 0 given by the caller of an x-lock
-				operation, if the x-lock is to be passed to
-				another thread to unlock (which happens in
-				asynchronous i/o). */
-	ulint	waiters;	/* This ulint is set to 1 if there are
-				waiters (readers or writers) in the global
-				wait array, waiting for this rw_lock.
-				Otherwise, == 0. */
+#endif /* INNODB_RW_LOCKS_USE_ATOMICS */
+
 	UT_LIST_NODE_T(rw_lock_t) list;
 				/* All allocated rw locks are put into a
 				list */
@@ -464,7 +532,9 @@ struct rw_lock_struct {
 				info list of the lock */
 	ulint	level;		/* Level in the global latching order. */
 #endif /* UNIV_SYNC_DEBUG */
+	ulint count_os_wait;	/* Count of os_waits. May not be accurate */
 	const char*	cfile_name;/* File name where lock created */
+        /* last s-lock file/line is not guaranteed to be correct */
 	const char*	last_s_file_name;/* File name where last s-locked */
 	const char*	last_x_file_name;/* File name where last x-locked */
 	ibool		writer_is_wait_ex;

include/sync0rw.ic

@@ -5,6 +5,38 @@ The read-write lock (for threads)
 
 Created 9/11/1995 Heikki Tuuri
 *******************************************************/
+/***********************************************************************
+# Copyright (c) 2008, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions
+# are met:
+#	* Redistributions of source code must retain the above copyright
+#	  notice, this list of conditions and the following disclaimer.
+#	* Redistributions in binary form must reproduce the above
+#	  copyright notice, this list of conditions and the following
+#	  disclaimer in the documentation and/or other materials
+#	  provided with the distribution.
+#	* Neither the name of the Google Inc. nor the names of its
+#	  contributors may be used to endorse or promote products
+#	  derived from this software without specific prior written
+#	  permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Note, the BSD license applies to the new code. The old code is GPL.
+***********************************************************************/
 
 /**********************************************************************
 Lock an rw-lock in shared mode for the current thread. If the rw-lock is
@@ -49,53 +81,88 @@ UNIV_INLINE
 ulint
 rw_lock_get_waiters(
 /*================*/
-	rw_lock_t*	lock)
+				/* out: 1 if waiters, 0 otherwise */
+	rw_lock_t*	lock)	/* in: rw-lock */
 {
 	return(lock->waiters);
 }
+
+/************************************************************************
+Sets lock->waiters to 1. It is not an error if lock->waiters is already
+1. On platforms where ATOMIC builtins are used this function enforces a
+memory barrier. */
 UNIV_INLINE
 void
-rw_lock_set_waiters(
-/*================*/
-	rw_lock_t*	lock,
-	ulint		flag)
+rw_lock_set_waiter_flag(
+/*====================*/
+	rw_lock_t*	lock)	/* in: rw-lock */
 {
-	lock->waiters = flag;
+#ifdef INNODB_RW_LOCKS_USE_ATOMICS
+	os_compare_and_swap((lint*)&(lock->waiters), 0, 1);
+#else /* INNODB_RW_LOCKS_USE_ATOMICS */
+	lock->waiters = 1;
+#endif /* INNODB_RW_LOCKS_USE_ATOMICS */
 }
+
+/************************************************************************
+Resets lock->waiters to 0. It is not an error if lock->waiters is already
+0. On platforms where ATOMIC builtins are used this function enforces a
+memory barrier. */
 UNIV_INLINE
-ulint
-rw_lock_get_writer(
-/*===============*/
-	rw_lock_t*	lock)
+void
+rw_lock_reset_waiter_flag(
+/*======================*/
+	rw_lock_t*	lock)	/* in: rw-lock */
 {
-	return(lock->writer);
+#ifdef INNODB_RW_LOCKS_USE_ATOMICS
+	os_compare_and_swap((lint*)&(lock->waiters), 1, 0);
+#else /* INNODB_RW_LOCKS_USE_ATOMICS */
+	lock->waiters = 0;
+#endif /* INNODB_RW_LOCKS_USE_ATOMICS */
 }
+
+/**********************************************************************
+Returns the write-status of the lock - this function made more sense
+with the old rw_lock implementation. */
 UNIV_INLINE
-void
-rw_lock_set_writer(
+ulint
+rw_lock_get_writer(
 /*===============*/
-	rw_lock_t*	lock,
-	ulint		flag)
+	rw_lock_t*	lock)
 {
-	lock->writer = flag;
+	lint lock_word = lock->lock_word;
+	if(lock_word > 0) {
+		/* return NOT_LOCKED in s-lock state, like the writer
+		member of the old lock implementation. */
+		return(RW_LOCK_NOT_LOCKED);
+	} else if (((-lock_word) % X_LOCK_DECR) == 0) {
+		return(RW_LOCK_EX);
+	} else {
+                ut_ad(lock_word > -X_LOCK_DECR);
+		return(RW_LOCK_WAIT_EX);
+	}
 }
+
+/**********************************************************************
+Returns number of readers. */
 UNIV_INLINE
 ulint
 rw_lock_get_reader_count(
 /*=====================*/
 	rw_lock_t*	lock)
 {
-	return(lock->reader_count);
-}
-UNIV_INLINE
-void
-rw_lock_set_reader_count(
-/*=====================*/
-	rw_lock_t*	lock,
-	ulint		count)
-{
-	lock->reader_count = count;
+	lint lock_word = lock->lock_word;
+	if(lock_word > 0) {
+		/* s-locked, no x-waiters */
+		return(X_LOCK_DECR - lock_word);
+	} else if (lock_word < 0 && lock_word > -X_LOCK_DECR) {
+		/* s-locked, with x-waiters */
+		return((ulint)(-lock_word));
+	}
+	return(0);
 }
+
+#ifndef INNODB_RW_LOCKS_USE_ATOMICS
 UNIV_INLINE
 mutex_t*
 rw_lock_get_mutex(
@@ -104,6 +171,7 @@ rw_lock_get_mutex(
 {
 	return(&(lock->mutex));
 }
+#endif
 
 /**********************************************************************
 Returns the value of writer_count for the lock. Does not reserve the lock
@@ -115,7 +183,127 @@ rw_lock_get_x_lock_count(
 				/* out: value of writer_count */
 	rw_lock_t*	lock)	/* in: rw-lock */
 {
-	return(lock->writer_count);
+	lint lock_copy = lock->lock_word;
+	/* If there is a reader, lock_word is not divisible by X_LOCK_DECR */
+	if(lock_copy > 0 || (-lock_copy) % X_LOCK_DECR != 0) {
+		return(0);
+	}
+	return(((-lock_copy) / X_LOCK_DECR) + 1);
+}
+
+/**********************************************************************
+Two different implementations for decrementing the lock_word of a rw_lock:
+one for systems supporting atomic operations, one for others. This does
+does not support recusive x-locks: they should be handled by the caller and
+need not be atomic since they are performed by the current lock holder.
+Returns true if the decrement was made, false if not. */
+UNIV_INLINE
+ibool
+rw_lock_lock_word_decr(
+/*===================*/
+				/* out: TRUE if decr occurs */
+	rw_lock_t*	lock,	/* in: rw-lock */
+	ulint		amount)	/* in: amount of decrement */
+{
+
+#ifdef INNODB_RW_LOCKS_USE_ATOMICS
+
+        lint local_lock_word = lock->lock_word;
+	while (local_lock_word > 0) {
+		if(os_compare_and_swap(&(lock->lock_word),
+                                       local_lock_word,
+                                       local_lock_word - amount)) {
+			return(TRUE);
+		}
+		local_lock_word = lock->lock_word;
+	}
+	return(FALSE);
+
+#else /* INNODB_RW_LOCKS_USE_ATOMICS */
+
+	ibool success = FALSE;
+	mutex_enter(&(lock->mutex));
+	if(lock->lock_word > 0) {
+		lock->lock_word -= amount;
+		success = TRUE;
+	}
+	mutex_exit(&(lock->mutex));
+	return(success);
+
+#endif /* INNODB_RW_LOCKS_USE_ATOMICS */
+}
+
+/**********************************************************************
+Two different implementations for incrementing the lock_word of a rw_lock:
+one for systems supporting atomic operations, one for others.
+Returns the value of lock_word after increment. */
+UNIV_INLINE
+lint
+rw_lock_lock_word_incr(
+/*===================*/
+				/* out: lock->lock_word after increment */
+	rw_lock_t*	lock,	/* in: rw-lock */
+	ulint		amount)	/* in: amount of increment */
+{
+
+#ifdef INNODB_RW_LOCKS_USE_ATOMICS
+
+	return(os_atomic_increment(&(lock->lock_word), amount));
+
+#else /* INNODB_RW_LOCKS_USE_ATOMICS */
+
+	lint local_lock_word;
+
+	mutex_enter(&(lock->mutex));
+
+	lock->lock_word += amount;
+	local_lock_word = lock->lock_word;
+
+	mutex_exit(&(lock->mutex));
+
+        return(local_lock_word);
+
+#endif /* INNODB_RW_LOCKS_USE_ATOMICS */
+}
+
+/**********************************************************************
+This function sets the lock->writer_thread and lock->recursive fields.
+For platforms where we are using atomic builtins instead of lock->mutex
+it sets the lock->writer_thread field using atomics to ensure memory
+ordering. Note that it is assumed that the caller of this function
+effectively owns the lock i.e.: nobody else is allowed to modify
+lock->writer_thread at this point in time.
+The protocol is that lock->writer_thread MUST be updated BEFORE the
+lock->recursive flag is set. */
+UNIV_INLINE
+void
+rw_lock_set_writer_id_and_recursion_flag(
+/*=====================================*/
+	rw_lock_t*	lock,		/* in/out: lock to work on */
+	ibool		recursive)	/* in: TRUE if recursion
+					allowed */
+{
+	os_thread_id_t	curr_thread	= os_thread_get_curr_id();
+
+	ut_ad(lock);
+
+#ifdef INNODB_RW_LOCKS_USE_ATOMICS
+
+	os_thread_id_t	local_thread = lock->writer_thread;
+	ibool success = os_compare_and_swap((lint*)&(lock->writer_thread),
+					    (lint)local_thread,
+					    (lint)curr_thread);
+	ut_a(success);
+	lock->recursive = recursive;
+
+#else /* INNODB_RW_LOCKS_USE_ATOMICS */
+
+	mutex_enter(&lock->mutex);
+	lock->writer_thread = curr_thread;
+	lock->recursive = recursive;
+	mutex_exit(&lock->mutex);
+
+#endif /* INNODB_RW_LOCKS_USE_ATOMICS */
 }
 
 /**********************************************************************
@@ -133,25 +321,21 @@ rw_lock_s_lock_low(
 	const char*	file_name, /* in: file name where lock requested */
 	ulint		line)	/* in: line where requested */
 {
-	ut_ad(mutex_own(rw_lock_get_mutex(lock)));
-
-	/* Check if the writer field is free */
-
-	if (UNIV_LIKELY(lock->writer == RW_LOCK_NOT_LOCKED)) {
-		/* Set the shared lock by incrementing the reader count */
-		lock->reader_count++;
+	/* TODO: study performance of UNIV_LIKELY branch prediction hints. */
+	if (!rw_lock_lock_word_decr(lock, 1)) {
+		/* Locking did not succeed */
+		return(FALSE);
+	}
 
 #ifdef UNIV_SYNC_DEBUG
-		rw_lock_add_debug_info(lock, pass, RW_LOCK_SHARED, file_name,
-				       line);
+	rw_lock_add_debug_info(lock, pass, RW_LOCK_SHARED, file_name, line);
 #endif
-		lock->last_s_file_name = file_name;
-		lock->last_s_line = line;
-
-		return(TRUE);	/* locking succeeded */
-	}
+	/* These debugging values are not set safely: they may be incorrect
+        or even refer to a line that is invalid for the file name. */
+	lock->last_s_file_name = file_name;
+	lock->last_s_line = line;
 
-	return(FALSE);	/* locking did not succeed */
+	return(TRUE);	/* locking succeeded */
 }
 
 /**********************************************************************
@@ -166,11 +350,10 @@ rw_lock_s_lock_direct(
 	const char*	file_name,	/* in: file name where requested */
 	ulint		line)		/* in: line where lock requested */
 {
-	ut_ad(lock->writer == RW_LOCK_NOT_LOCKED);
-	ut_ad(rw_lock_get_reader_count(lock) == 0);
+	ut_ad(lock->lock_word == X_LOCK_DECR);
 
-	/* Set the shared lock by incrementing the reader count */
-	lock->reader_count++;
+	/* Indicate there is a new reader by decrementing lock_word */
+	lock->lock_word--;
 
 	lock->last_s_file_name = file_name;
 	lock->last_s_line = line;
@@ -193,13 +376,11 @@ rw_lock_x_lock_direct(
 	ulint		line)		/* in: line where lock requested */
 {
 	ut_ad(rw_lock_validate(lock));
-	ut_ad(rw_lock_get_reader_count(lock) == 0);
-	ut_ad(rw_lock_get_writer(lock) == RW_LOCK_NOT_LOCKED);
+	ut_ad(lock->lock_word == X_LOCK_DECR);
 
-	rw_lock_set_writer(lock, RW_LOCK_EX);
+	lock->lock_word -= X_LOCK_DECR;
 	lock->writer_thread = os_thread_get_curr_id();
-	lock->writer_count++;
-	lock->pass = 0;
+	lock->recursive = TRUE;
 
 	lock->last_x_file_name = file_name;
 	lock->last_x_line = line;
@@ -240,15 +421,12 @@ rw_lock_s_lock_func(
 	ut_ad(!rw_lock_own(lock, RW_LOCK_SHARED)); /* see NOTE above */
 #endif /* UNIV_SYNC_DEBUG */
 
-	mutex_enter(rw_lock_get_mutex(lock));
-
-	if (UNIV_LIKELY(rw_lock_s_lock_low(lock, pass, file_name, line))) {
-		mutex_exit(rw_lock_get_mutex(lock));
+	/* TODO: study performance of UNIV_LIKELY branch prediction hints. */
+	if (rw_lock_s_lock_low(lock, pass, file_name, line)) {
 
 		return; /* Success */
 	} else {
 		/* Did not succeed, try spin wait */
-		mutex_exit(rw_lock_get_mutex(lock));
 
 		rw_lock_s_lock_spin(lock, pass, file_name, line);
 
@@ -258,86 +436,60 @@ rw_lock_s_lock_func(
 
 /**********************************************************************
 NOTE! Use the corresponding macro, not directly this function! Lock an
-rw-lock in shared mode for the current thread if the lock can be acquired
-immediately. */
+rw-lock in exclusive mode for the current thread if the lock can be
+obtained immediately. */
 UNIV_INLINE
 ibool
-rw_lock_s_lock_func_nowait(
+rw_lock_x_lock_func_nowait(
 /*=======================*/
 				/* out: TRUE if success */
 	rw_lock_t*	lock,	/* in: pointer to rw-lock */
 	const char*	file_name,/* in: file name where lock requested */
 	ulint		line)	/* in: line where requested */
 {
-	ibool	success	= FALSE;
-
-	mutex_enter(rw_lock_get_mutex(lock));
-
-	if (lock->writer == RW_LOCK_NOT_LOCKED) {
-		/* Set the shared lock by incrementing the reader count */
-		lock->reader_count++;
+	os_thread_id_t	curr_thread	= os_thread_get_curr_id();
 
-#ifdef UNIV_SYNC_DEBUG
-		rw_lock_add_debug_info(lock, 0, RW_LOCK_SHARED, file_name,
-				       line);
-#endif
+	ibool success;
 
-		lock->last_s_file_name = file_name;
-		lock->last_s_line = line;
+#ifdef INNODB_RW_LOCKS_USE_ATOMICS
+	success = os_compare_and_swap(&(lock->lock_word), X_LOCK_DECR, 0);
+#else
 
+	success = FALSE;
+	mutex_enter(&(lock->mutex));
+	if (lock->lock_word == X_LOCK_DECR) {
+		lock->lock_word = 0;
 		success = TRUE;
 	}
+	mutex_exit(&(lock->mutex));
 
-	mutex_exit(rw_lock_get_mutex(lock));
-
-	return(success);
-}
+#endif
+	if (success) {
+		rw_lock_set_writer_id_and_recursion_flag(lock, TRUE);
 
-/**********************************************************************
-NOTE! Use the corresponding macro, not directly this function! Lock an
-rw-lock in exclusive mode for the current thread if the lock can be
-obtained immediately. */
-UNIV_INLINE
-ibool
-rw_lock_x_lock_func_nowait(
-/*=======================*/
-				/* out: TRUE if success */
-	rw_lock_t*	lock,	/* in: pointer to rw-lock */
-	const char*	file_name,/* in: file name where lock requested */
-	ulint		line)	/* in: line where requested */
-{
-	ibool		success		= FALSE;
-	os_thread_id_t	curr_thread	= os_thread_get_curr_id();
-	mutex_enter(rw_lock_get_mutex(lock));
+	} else if (lock->recursive &&
+		   os_thread_eq(lock->writer_thread, curr_thread)) {
+		/* Relock: this lock_word modification is safe since no other
+		threads can modify (lock, unlock, or reserve) lock_word while
+		there is an exclusive writer and this is the writer thread. */
+		lock->lock_word -= X_LOCK_DECR;
 
-	if (UNIV_UNLIKELY(rw_lock_get_reader_count(lock) != 0)) {
-	} else if (UNIV_LIKELY(rw_lock_get_writer(lock)
-			       == RW_LOCK_NOT_LOCKED)) {
-		rw_lock_set_writer(lock, RW_LOCK_EX);
-		lock->writer_thread = curr_thread;
-		lock->pass = 0;
-relock:
-		lock->writer_count++;
+		ut_ad(((-lock->lock_word) % X_LOCK_DECR) == 0);
 
+	} else {
+		/* Failure */
+		return(FALSE);
+	}
 #ifdef UNIV_SYNC_DEBUG
-		rw_lock_add_debug_info(lock, 0, RW_LOCK_EX, file_name, line);
+	rw_lock_add_debug_info(lock, 0, RW_LOCK_EX, file_name, line);
 #endif
 
-		lock->last_x_file_name = file_name;
-		lock->last_x_line = line;
-
-		success = TRUE;
-	} else if (rw_lock_get_writer(lock) == RW_LOCK_EX
-		   && lock->pass == 0
-		   && os_thread_eq(lock->writer_thread, curr_thread)) {
-		goto relock;
-	}
-
-	mutex_exit(rw_lock_get_mutex(lock));
+	lock->last_x_file_name = file_name;
+	lock->last_x_line = line;
 
 	ut_ad(rw_lock_validate(lock));
 
-	return(success);
+	return(TRUE);
 }
 
 /**********************************************************************
@@ -353,39 +505,21 @@ rw_lock_s_unlock_func(
 #endif
 	)
 {
-	mutex_t*	mutex	= &(lock->mutex);
-	ibool		sg	= FALSE;
-
-	/* Acquire the mutex protecting the rw-lock fields */
-	mutex_enter(mutex);
-
-	/* Reset the shared lock by decrementing the reader count */
-
-	ut_a(lock->reader_count > 0);
-	lock->reader_count--;
+	ut_ad((lock->lock_word % X_LOCK_DECR) != 0);
 
 #ifdef UNIV_SYNC_DEBUG
 	rw_lock_remove_debug_info(lock, pass, RW_LOCK_SHARED);
 #endif
 
-	/* If there may be waiters and this was the last s-lock,
-	signal the object */
-
-	if (UNIV_UNLIKELY(lock->waiters)
-	    && lock->reader_count == 0) {
-		sg = TRUE;
+	/* Increment lock_word to indicate 1 less reader */
+	if (rw_lock_lock_word_incr(lock, 1) == 0) {
 
-		rw_lock_set_waiters(lock, 0);
-	}
-
-	mutex_exit(mutex);
-
-	if (UNIV_UNLIKELY(sg)) {
-#ifdef __WIN__
+		/* wait_ex waiter exists. It may not be asleep, but we signal
+                anyway. We do not wake other waiters, because they can't
+                exist without wait_ex waiter and wait_ex waiter goes first.*/
 		os_event_set(lock->wait_ex_event);
-#endif
-		os_event_set(lock->event);
 		sync_array_object_signalled(sync_primary_wait_array);
+
 	}
 
 	ut_ad(rw_lock_validate(lock));
@@ -404,16 +538,15 @@ rw_lock_s_unlock_direct(
 /*====================*/
 	rw_lock_t*	lock)	/* in: rw-lock */
 {
-	/* Reset the shared lock by decrementing the reader count */
-
-	ut_ad(lock->reader_count > 0);
-
-	lock->reader_count--;
+	ut_ad(lock->lock_word < X_LOCK_DECR);
 
 #ifdef UNIV_SYNC_DEBUG
 	rw_lock_remove_debug_info(lock, 0, RW_LOCK_SHARED);
 #endif
 
+	/* Decrease reader count by incrementing lock_word */
+	lock->lock_word++;
+
 	ut_ad(!lock->waiters);
 	ut_ad(rw_lock_validate(lock));
 #ifdef UNIV_SYNC_PERF_STAT
@@ -434,42 +567,32 @@ rw_lock_x_unlock_func(
 #endif
 	)
 {
-	ibool	sg	= FALSE;
-
-	/* Acquire the mutex protecting the rw-lock fields */
-	mutex_enter(&(lock->mutex));
-
-	/* Reset the exclusive lock if this thread no longer has an x-mode
-	lock */
-
-	ut_ad(lock->writer_count > 0);
-
-	lock->writer_count--;
-
-	if (lock->writer_count == 0) {
-		rw_lock_set_writer(lock, RW_LOCK_NOT_LOCKED);
+	ut_ad((lock->lock_word % X_LOCK_DECR) == 0);
+
+	/* lock->recursive flag also indicates if lock->writer_thread is
+	valid or stale. If we are the last of the recursive callers
+	then we must unset lock->recursive flag to indicate that the
+	lock->writer_thread is now stale.
+	Note that since we still hold the x-lock we can safely read the
+	lock_word. */
+	if (lock->lock_word == 0) {
+		/* Last caller in a possible recursive chain. */
+		lock->recursive = FALSE;
 	}
 
 #ifdef UNIV_SYNC_DEBUG
 	rw_lock_remove_debug_info(lock, pass, RW_LOCK_EX);
 #endif
 
-	/* If there may be waiters, signal the lock */
-	if (UNIV_UNLIKELY(lock->waiters)
-	    && lock->writer_count == 0) {
-
-		sg = TRUE;
-		rw_lock_set_waiters(lock, 0);
-	}
-
-	mutex_exit(&(lock->mutex));
-
-	if (UNIV_UNLIKELY(sg)) {
-#ifdef __WIN__
-		os_event_set(lock->wait_ex_event);
-#endif
-		os_event_set(lock->event);
-		sync_array_object_signalled(sync_primary_wait_array);
+	if (rw_lock_lock_word_incr(lock, X_LOCK_DECR) == X_LOCK_DECR) {
+		/* Lock is now free. May have to signal read/write waiters.
+                We do not need to signal wait_ex waiters, since they cannot
+                exist when there is a writer. */
+		if (lock->waiters) {
+			rw_lock_reset_waiter_flag(lock);
+			os_event_set(lock->event);
+			sync_array_object_signalled(sync_primary_wait_array);
+		}
 	}
 
 	ut_ad(rw_lock_validate(lock));
@@ -481,7 +604,7 @@ rw_lock_x_unlock_func(
 
 /**********************************************************************
 Releases an exclusive mode lock when we know there are no waiters, and
-none else will access the lock durint the time this function is executed. */
+none else will access the lock during the time this function is executed. */
 UNIV_INLINE
 void
 rw_lock_x_unlock_direct(
@@ -491,18 +614,18 @@ rw_lock_x_unlock_direct(
 	/* Reset the exclusive lock if this thread no longer has an x-mode
 	lock */
 
-	ut_ad(lock->writer_count > 0);
-
-	lock->writer_count--;
-
-	if (lock->writer_count == 0) {
-		rw_lock_set_writer(lock, RW_LOCK_NOT_LOCKED);
-	}
+	ut_ad((lock->lock_word % X_LOCK_DECR) == 0);
 
 #ifdef UNIV_SYNC_DEBUG
 	rw_lock_remove_debug_info(lock, 0, RW_LOCK_EX);
 #endif
 
+	if (lock->lock_word == 0) {
+		lock->recursive = FALSE;
+	}
+
+	lock->lock_word += X_LOCK_DECR;
+
 	ut_ad(!lock->waiters);
 	ut_ad(rw_lock_validate(lock));
 

include/sync0sync.h

@@ -5,6 +5,38 @@ Mutex, the basic synchronization primitive
 
 Created 9/5/1995 Heikki Tuuri
 *******************************************************/
+/***********************************************************************
+# Copyright (c) 2008, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions
+# are met:
+#	* Redistributions of source code must retain the above copyright
+#	  notice, this list of conditions and the following disclaimer.
+#	* Redistributions in binary form must reproduce the above
+#	  copyright notice, this list of conditions and the following
+#	  disclaimer in the documentation and/or other materials
+#	  provided with the distribution.
+#	* Neither the name of the Google Inc. nor the names of its
+#	  contributors may be used to endorse or promote products
+#	  derived from this software without specific prior written
+#	  permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Note, the BSD license applies to the new code. The old code is GPL.
+***********************************************************************/
 
 #ifndef sync0sync_h
 #define sync0sync_h
@@ -237,7 +269,7 @@ mutex_n_reserved(void);
 NOT to be used outside this module except in debugging! Gets the value
 of the lock word. */
 UNIV_INLINE
-ulint
+byte
 mutex_get_lock_word(
 /*================*/
 	const mutex_t*	mutex);	/* in: mutex */
@@ -463,9 +495,11 @@ implementation of a mutual exclusion semaphore. */
 
 struct mutex_struct {
 	os_event_t	event;	/* Used by sync0arr.c for the wait queue */
-	ulint	lock_word;	/* This ulint is the target of the atomic
-				test-and-set instruction in Win32 */
-#if defined WIN32 && defined UNIV_CAN_USE_X86_ASSEMBLER
+	byte	lock_word;	/* This byte is the target of the atomic
+				test-and-set instruction in Win32 and
+				x86 32/64 with GCC 4.1.0 or later version */
+#if defined(_WIN32) && defined(UNIV_CAN_USE_X86_ASSEMBLER)
+#elif defined(HAVE_GCC_ATOMIC_BUILTINS)
 #else
 	os_fast_mutex_t
 		os_fast_mutex;	/* In other systems we use this OS mutex
@@ -519,8 +553,7 @@ to 20 microseconds. */
 /* The number of system calls made in this module. Intended for performance
 monitoring. */
 
-extern	ulint	mutex_system_call_count;
-extern	ulint	mutex_exit_count;
+extern	ib_int64_t	mutex_exit_count;
 
 #ifdef UNIV_SYNC_DEBUG
 /* Latching order checks start when this is set TRUE */

include/sync0sync.ic

@@ -5,16 +5,38 @@ Mutex, the basic synchronization primitive
 
 Created 9/5/1995 Heikki Tuuri
 *******************************************************/
-
-#if defined(not_defined) && defined(__GNUC__) && defined(UNIV_INTEL_X86)
-/* %z0: Use the size of operand %0 which in our case is *m to determine
-instruction size, it should end up as xchgl. "1" in the input constraint,
-says that "in" has to go in the same place as "out".*/
-#define TAS(m, in, out) \
-	asm volatile ("xchg%z0 %2, %0" \
-	: "=g" (*(m)), "=r" (out) \
-	: "1" (in))	/* Note: "1" here refers to "=r" (out) */
-#endif
+/***********************************************************************
+# Copyright (c) 2008, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions
+# are met:
+#	* Redistributions of source code must retain the above copyright
+#	  notice, this list of conditions and the following disclaimer.
+#	* Redistributions in binary form must reproduce the above
+#	  copyright notice, this list of conditions and the following
+#	  disclaimer in the documentation and/or other materials
+#	  provided with the distribution.
+#	* Neither the name of the Google Inc. nor the names of its
+#	  contributors may be used to endorse or promote products
+#	  derived from this software without specific prior written
+#	  permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Note, the BSD license applies to the new code. The old code is GPL.
+***********************************************************************/
 
 /**********************************************************************
 Sets the waiters field in a mutex. */
@@ -59,7 +81,7 @@ mutex_signal_object(
 Performs an atomic test-and-set instruction to the lock_word field of a
 mutex. */
 UNIV_INLINE
-ulint
+byte
 mutex_test_and_set(
 /*===============*/
 				/* out: the previous value of lock_word: 0 or
@@ -67,18 +89,18 @@ mutex_test_and_set(
 	mutex_t*	mutex)	/* in: mutex */
 {
 #if defined(_WIN32) && defined(UNIV_CAN_USE_X86_ASSEMBLER)
-	ulint	res;
-	ulint*	lw;		/* assembler code is used to ensure that
+	byte	res;
+	byte*	lw;		/* assembler code is used to ensure that
 				lock_word is loaded from memory */
 	ut_ad(mutex);
-	ut_ad(sizeof(ulint) == 4);
+	ut_ad(sizeof(byte) == 1);
 
 	lw = &(mutex->lock_word);
 
 	__asm	MOV	ECX, lw
 		__asm	MOV	EDX, 1
-		__asm	XCHG	EDX, DWORD PTR [ECX]
-		__asm	MOV	res, EDX
+		__asm	XCHG	DL, BYTE PTR [ECX]
+		__asm	MOV	res, DL
 
 		/* The fence below would prevent this thread from
 		reading the data structure protected by the mutex
@@ -98,12 +120,8 @@ mutex_test_and_set(
 		/* mutex_fence(); */
 
 		return(res);
-#elif defined(not_defined) && defined(__GNUC__) && defined(UNIV_INTEL_X86)
-	ulint	res;
-
-	TAS(&mutex->lock_word, 1, res);
-
-	return(res);
+#elif defined(HAVE_GCC_ATOMIC_BUILTINS)
+	return __sync_lock_test_and_set(&(mutex->lock_word), 1);
 #else
 	ibool	ret;
 
@@ -117,7 +135,7 @@ mutex_test_and_set(
 		mutex->lock_word = 1;
 	}
 
-	return(ret);
+	return((byte)ret);
 #endif
 }
 
@@ -131,7 +149,7 @@ mutex_reset_lock_word(
 	mutex_t*	mutex)	/* in: mutex */
 {
 #if defined(_WIN32) && defined(UNIV_CAN_USE_X86_ASSEMBLER)
-	ulint*	lw;		/* assembler code is used to ensure that
+	byte*	lw;		/* assembler code is used to ensure that
 				lock_word is loaded from memory */
 	ut_ad(mutex);
 
@@ -139,11 +157,12 @@ mutex_reset_lock_word(
 
 	__asm	MOV	EDX, 0
 		__asm	MOV	ECX, lw
-		__asm	XCHG	EDX, DWORD PTR [ECX]
-#elif defined(not_defined) && defined(__GNUC__) && defined(UNIV_INTEL_X86)
-	ulint	res;
-
-	TAS(&mutex->lock_word, 0, res);
+		__asm	XCHG	DL, BYTE PTR [ECX]
+#elif defined(HAVE_GCC_ATOMIC_BUILTINS)
+	/* In theory __sync_lock_release should be used to release the lock.
+	Unfortunately, it does not work properly alone. The workaround is
+	that more conservative __sync_lock_test_and_set is used instead. */
+	__sync_lock_test_and_set(&(mutex->lock_word), 0);
 #else
 	mutex->lock_word = 0;
 
@@ -154,12 +173,12 @@ mutex_reset_lock_word(
 /**********************************************************************
 Gets the value of the lock word. */
 UNIV_INLINE
-ulint
+byte
 mutex_get_lock_word(
 /*================*/
 	const mutex_t*	mutex)	/* in: mutex */
 {
-	const volatile ulint*	ptr;	/* declared volatile to ensure that
+	const volatile byte*	ptr;	/* declared volatile to ensure that
 					lock_word is loaded from memory */
 	ut_ad(mutex);
 

include/univ.i

@@ -5,6 +5,38 @@ Version control for database, common definitions, and include files
 
 Created 1/20/1994 Heikki Tuuri
 ****************************************************************************/
+/***********************************************************************
+# Copyright (c) 2008, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions
+# are met:
+#	* Redistributions of source code must retain the above copyright
+#	  notice, this list of conditions and the following disclaimer.
+#	* Redistributions in binary form must reproduce the above
+#	  copyright notice, this list of conditions and the following
+#	  disclaimer in the documentation and/or other materials
+#	  provided with the distribution.
+#	* Neither the name of the Google Inc. nor the names of its
+#	  contributors may be used to endorse or promote products
+#	  derived from this software without specific prior written
+#	  permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Note, the BSD license applies to the new code. The old code is GPL.
+***********************************************************************/
 
 #ifndef univ_i
 #define univ_i
@@ -90,6 +122,13 @@ of the 32-bit x86 assembler in mutex operations. */
 #  define UNIV_CAN_USE_X86_ASSEMBLER
 # endif
 
+/* For InnoDB rw_locks to work with atomics we need the thread_id
+to be no more than machine word wide. The following enables using
+atomics for InnoDB rw_locks where these conditions are met. */
+# if defined(HAVE_GCC_ATOMIC_BUILTINS) && defined(UNIV_LINUX)
+#  define INNODB_RW_LOCKS_USE_ATOMICS
+# endif
+
 /* We only try to do explicit inlining of functions with gcc and
 Microsoft Visual C++ */
 

log/log0log.c

@@ -739,7 +739,7 @@ log_init(void)
 	ut_a(LOG_BUFFER_SIZE >= 16 * OS_FILE_LOG_BLOCK_SIZE);
 	ut_a(LOG_BUFFER_SIZE >= 4 * UNIV_PAGE_SIZE);
 
-	buf = ut_malloc(LOG_BUFFER_SIZE + OS_FILE_LOG_BLOCK_SIZE);
+	buf = mem_alloc(LOG_BUFFER_SIZE + OS_FILE_LOG_BLOCK_SIZE);
 	log_sys->buf = ut_align(buf, OS_FILE_LOG_BLOCK_SIZE);
 
 	log_sys->buf_size = LOG_BUFFER_SIZE;

log/log0recv.c

@@ -591,6 +591,7 @@ not_consistent:
 	return(DB_SUCCESS);
 }
 
+#ifdef UNIV_HOTBACKUP
 /***********************************************************************
 Reads the checkpoint info needed in hot backup. */
 UNIV_INTERN
@@ -661,6 +662,7 @@ recv_read_cp_info_for_backup(
 
 	return(TRUE);
 }
+#endif /* UNIV_HOTBACKUP */
 
 /**********************************************************
 Checks the 4-byte checksum to the trailer checksum field of a log block.
@@ -698,6 +700,7 @@ log_block_checksum_is_ok_or_old_format(
 	return(FALSE);
 }
 
+#ifdef UNIV_HOTBACKUP
 /***********************************************************************
 Scans the log segment and n_bytes_scanned is set to the length of valid
 log scanned. */
@@ -787,6 +790,7 @@ recv_scan_log_seg_for_backup(
 		}
 	}
 }
+#endif /* UNIV_HOTBACKUP */
 
 /***********************************************************************
 Tries to parse a single log record body and also applies it to a page if

mem/mem0pool.c

@@ -478,7 +478,9 @@ mem_area_free(
 	ulint		n;
 
 	if (srv_use_sys_malloc) {
-		return(free(ptr));
+		free(ptr);
+
+		return;
 	}
 
 	/* It may be that the area was really allocated from the OS with

mysql-test/innodb-zip.result

@@ -419,3 +419,6 @@ select @@innodb_file_format_check;
 @@innodb_file_format_check
 Barracuda
 drop table normal_table, zip_table;
+set global innodb_file_format=Antelope;
+set global innodb_file_per_table=0;
+set global innodb_file_format_check=Antelope;

mysql-test/innodb-zip.test

@@ -2,6 +2,7 @@
 
 let $per_table=`select @@innodb_file_per_table`;
 let $format=`select @@innodb_file_format`;
+let $innodb_file_format_check_orig=`select @@innodb_file_format_check`;
 set global innodb_file_per_table=off;
 set global innodb_file_format=`0`;
 
@@ -157,7 +158,8 @@ set global innodb_file_format=`1a`;
 set global innodb_file_format=``;
 
 #test strict mode.
---enable_errors
+# this does not work anymore, has been removed from mysqltest
+# -- enable_errors
 set global innodb_file_per_table = on;
 set global innodb_file_format = `1`;
 
@@ -330,3 +332,11 @@ show table status;
 select @@innodb_file_format_check;
 drop table normal_table, zip_table;
 -- disable_result_log
+
+#
+# restore environment to the state it was before this test execution
+#
+
+eval set global innodb_file_format=$format;
+eval set global innodb_file_per_table=$per_table;
+eval set global innodb_file_format_check=$innodb_file_format_check_orig;

mysql-test/patches/bug35261.diff

---- mysql-test/t/date_formats.test.orig	2007-06-15 02:53:07.000000000 +0300
-+++ mysql-test/t/date_formats.test	2008-03-19 17:25:10.000000000 +0200
-@@ -7,9 +7,15 @@
- --enable_warnings
- 
- --replace_result ROW <format> STATEMENT <format> MIXED <format>
--SHOW GLOBAL VARIABLES LIKE "%e_format";
-+SELECT variable_name, variable_value
-+FROM information_schema.global_variables
-+WHERE variable_name IN ('date_format', 'datetime_format', 'time_format')
-+ORDER BY variable_name;
- --replace_result ROW <format> STATEMENT <format> MIXED <format>
--SHOW SESSION VARIABLES LIKE "%e_format";
-+SELECT variable_name, variable_value
-+FROM information_schema.session_variables
-+WHERE variable_name IN ('date_format', 'datetime_format', 'time_format')
-+ORDER BY variable_name;
- 
- #
- # Test setting a lot of different formats to see which formats are accepted and
-@@ -37,7 +43,10 @@
- set datetime_format= '%h:%i:%s.%f %p %Y-%m-%d';
- 
- --replace_result ROW <format> STATEMENT <format> MIXED <format>
--SHOW SESSION VARIABLES LIKE "%e_format";
-+SELECT variable_name, variable_value
-+FROM information_schema.session_variables
-+WHERE variable_name IN ('date_format', 'datetime_format', 'time_format')
-+ORDER BY variable_name;
- 
- --error 1231
- SET time_format='%h:%i:%s';
---- mysql-test/r/date_formats.result.orig	2008-02-12 21:09:14.000000000 +0200
-+++ mysql-test/r/date_formats.result	2008-03-19 17:26:33.000000000 +0200
-@@ -1,14 +1,20 @@
- drop table if exists t1;
--SHOW GLOBAL VARIABLES LIKE "%e_format";
--Variable_name	Value
--date_format	%d.%m.%Y
--datetime_format	%Y-%m-%d %H:%i:%s
--time_format	%H.%i.%s
--SHOW SESSION VARIABLES LIKE "%e_format";
--Variable_name	Value
--date_format	%d.%m.%Y
--datetime_format	%Y-%m-%d %H:%i:%s
--time_format	%H.%i.%s
-+SELECT variable_name, variable_value
-+FROM information_schema.global_variables
-+WHERE variable_name IN ('date_format', 'datetime_format', 'time_format')
-+ORDER BY variable_name;
-+variable_name	variable_value
-+DATETIME_FORMAT	%Y-%m-%d %H:%i:%s
-+DATE_FORMAT	%d.%m.%Y
-+TIME_FORMAT	%H.%i.%s
-+SELECT variable_name, variable_value
-+FROM information_schema.session_variables
-+WHERE variable_name IN ('date_format', 'datetime_format', 'time_format')
-+ORDER BY variable_name;
-+variable_name	variable_value
-+DATETIME_FORMAT	%Y-%m-%d %H:%i:%s
-+DATE_FORMAT	%d.%m.%Y
-+TIME_FORMAT	%H.%i.%s
- SET time_format='%H%i%s';
- SET time_format='%H:%i:%s.%f';
- SET time_format='%h-%i-%s.%f%p';
-@@ -26,11 +32,14 @@
- set datetime_format= '%H:%i:%s.%f %m-%d-%Y';
- set datetime_format= '%h:%i:%s %p %Y-%m-%d';
- set datetime_format= '%h:%i:%s.%f %p %Y-%m-%d';
--SHOW SESSION VARIABLES LIKE "%e_format";
--Variable_name	Value
--date_format	%m-%d-%Y
--datetime_format	%h:%i:%s.%f %p %Y-%m-%d
--time_format	%h:%i:%s%p
-+SELECT variable_name, variable_value
-+FROM information_schema.session_variables
-+WHERE variable_name IN ('date_format', 'datetime_format', 'time_format')
-+ORDER BY variable_name;
-+variable_name	variable_value
-+DATETIME_FORMAT	%h:%i:%s.%f %p %Y-%m-%d
-+DATE_FORMAT	%m-%d-%Y
-+TIME_FORMAT	%h:%i:%s%p
- SET time_format='%h:%i:%s';
- ERROR 42000: Variable 'time_format' can't be set to the value of '%h:%i:%s'
- SET time_format='%H %i:%s';

mysql-test/patches/information_schema.diff

-diff mysql-test/r/information_schema.result.orig mysql-test/r/information_schema.result
---- mysql-test/r/information_schema.result.orig	2008-08-04 09:27:49.000000000 +0300
-+++ mysql-test/r/information_schema.result	2008-10-07 11:21:51.000000000 +0300
-@@ -64,6 +64,13 @@
+--- mysql-test/r/information_schema.result.orig	2009-01-31 03:38:50.000000000 +0200
++++ mysql-test/r/information_schema.result	2009-01-31 07:51:58.000000000 +0200
+@@ -71,6 +71,13 @@
  TRIGGERS
  USER_PRIVILEGES
  VIEWS
@@ -15,7 +14,7 @@ diff mysql-test/r/information_schema.result.orig mysql-test/r/information_schema
  columns_priv
  db
  event
-@@ -795,6 +802,8 @@
+@@ -799,6 +806,8 @@
  TABLES	UPDATE_TIME	datetime
  TABLES	CHECK_TIME	datetime
  TRIGGERS	CREATED	datetime
@@ -24,16 +23,16 @@ diff mysql-test/r/information_schema.result.orig mysql-test/r/information_schema
  event	execute_at	datetime
  event	last_executed	datetime
  event	starts	datetime
-@@ -848,7 +857,7 @@
+@@ -852,7 +861,7 @@
  flush privileges;
- SELECT table_schema, count(*) FROM information_schema.TABLES where table_name<>'ndb_binlog_index' AND table_name<>'ndb_apply_status' GROUP BY TABLE_SCHEMA;
+ SELECT table_schema, count(*) FROM information_schema.TABLES WHERE table_schema IN ('mysql', 'INFORMATION_SCHEMA', 'test', 'mysqltest') AND table_name<>'ndb_binlog_index' AND table_name<>'ndb_apply_status' GROUP BY TABLE_SCHEMA;
  table_schema	count(*)
 -information_schema	28
 +information_schema	35
  mysql	22
  create table t1 (i int, j int);
  create trigger trg1 before insert on t1 for each row
-@@ -1263,6 +1272,13 @@
+@@ -1267,6 +1276,13 @@
  TRIGGERS	TRIGGER_SCHEMA
  USER_PRIVILEGES	GRANTEE
  VIEWS	TABLE_SCHEMA
@@ -47,7 +46,7 @@ diff mysql-test/r/information_schema.result.orig mysql-test/r/information_schema
  SELECT t.table_name, c1.column_name
  FROM information_schema.tables t
  INNER JOIN
-@@ -1306,6 +1322,13 @@
+@@ -1310,6 +1326,13 @@
  TRIGGERS	TRIGGER_SCHEMA
  USER_PRIVILEGES	GRANTEE
  VIEWS	TABLE_SCHEMA
@@ -58,10 +57,10 @@ diff mysql-test/r/information_schema.result.orig mysql-test/r/information_schema
 +INNODB_CMPMEM	page_size
 +INNODB_CMP	page_size
 +INNODB_LOCKS	lock_id
- SELECT MAX(table_name) FROM information_schema.tables;
+ SELECT MAX(table_name) FROM information_schema.tables WHERE table_schema IN ('mysql', 'INFORMATION_SCHEMA', 'test');
  MAX(table_name)
  VIEWS
-@@ -1382,6 +1405,13 @@
+@@ -1386,6 +1409,13 @@
  FILES	information_schema.FILES	1
  GLOBAL_STATUS	information_schema.GLOBAL_STATUS	1
  GLOBAL_VARIABLES	information_schema.GLOBAL_VARIABLES	1

row/row0sel.c

@@ -5,6 +5,38 @@ Select
 
 Created 12/19/1997 Heikki Tuuri
 *******************************************************/
+/***********************************************************************
+# Copyright (c) 2008, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions
+# are met:
+#	* Redistributions of source code must retain the above copyright
+#	  notice, this list of conditions and the following disclaimer.
+#	* Redistributions in binary form must reproduce the above
+#	  copyright notice, this list of conditions and the following
+#	  disclaimer in the documentation and/or other materials
+#	  provided with the distribution.
+#	* Neither the name of the Google Inc. nor the names of its
+#	  contributors may be used to endorse or promote products
+#	  derived from this software without specific prior written
+#	  permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Note, the BSD license applies to the new code. The old code is GPL.
+***********************************************************************/
 
 #include "row0sel.h"
 
@@ -1339,7 +1371,7 @@ table_loop:
 			rw_lock_s_lock(&btr_search_latch);
 
 			search_latch_locked = TRUE;
-		} else if (btr_search_latch.writer_is_wait_ex) {
+		} else if (rw_lock_get_writer(&btr_search_latch) == RW_LOCK_WAIT_EX) {
 
 			/* There is an x-latch request waiting: release the
 			s-latch for a moment; as an s-latch here is often
@@ -3364,7 +3396,7 @@ row_search_for_mysql(
 	/* PHASE 0: Release a possible s-latch we are holding on the
 	adaptive hash index latch if there is someone waiting behind */
 
-	if (UNIV_UNLIKELY(btr_search_latch.writer != RW_LOCK_NOT_LOCKED)
+	if (UNIV_UNLIKELY(rw_lock_get_writer(&btr_search_latch) != RW_LOCK_NOT_LOCKED)
 	    && trx->has_search_latch) {
 
 		/* There is an x-latch request on the adaptive hash index:

srv/srv0srv.c

@@ -24,6 +24,38 @@ thread library. This might confuse NT though.
 
 Created 10/8/1995 Heikki Tuuri
 *******************************************************/
+/***********************************************************************
+# Copyright (c) 2008, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions
+# are met:
+#	* Redistributions of source code must retain the above copyright
+#	  notice, this list of conditions and the following disclaimer.
+#	* Redistributions in binary form must reproduce the above
+#	  copyright notice, this list of conditions and the following
+#	  disclaimer in the documentation and/or other materials
+#	  provided with the distribution.
+#	* Neither the name of the Google Inc. nor the names of its
+#	  contributors may be used to endorse or promote products
+#	  derived from this software without specific prior written
+#	  permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Note, the BSD license applies to the new code. The old code is GPL.
+***********************************************************************/
 /* Dummy comment */
 #include "srv0srv.h"
 
@@ -1828,6 +1860,11 @@ srv_export_innodb_status(void)
 	export_vars.innodb_buffer_pool_pages_misc = buf_pool->curr_size
 		- UT_LIST_GET_LEN(buf_pool->LRU)
 		- UT_LIST_GET_LEN(buf_pool->free);
+#ifdef HAVE_GCC_ATOMIC_BUILTINS
+	export_vars.innodb_have_atomic_builtins = 1;
+#else
+	export_vars.innodb_have_atomic_builtins = 0;
+#endif
 	export_vars.innodb_page_size = UNIV_PAGE_SIZE;
 	export_vars.innodb_log_waits = srv_log_waits;
 	export_vars.innodb_os_log_written = srv_os_log_written;

srv/srv0start.c

@@ -5,6 +5,38 @@ Starts the InnoDB database server
 
 Created 2/16/1996 Heikki Tuuri
 *************************************************************************/
+/***********************************************************************
+# Copyright (c) 2008, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions
+# are met:
+#	* Redistributions of source code must retain the above copyright
+#	  notice, this list of conditions and the following disclaimer.
+#	* Redistributions in binary form must reproduce the above
+#	  copyright notice, this list of conditions and the following
+#	  disclaimer in the documentation and/or other materials
+#	  provided with the distribution.
+#	* Neither the name of the Google Inc. nor the names of its
+#	  contributors may be used to endorse or promote products
+#	  derived from this software without specific prior written
+#	  permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Note, the BSD license applies to the new code. The old code is GPL.
+***********************************************************************/
 
 #include "os0proc.h"
 #include "sync0sync.h"
@@ -956,8 +988,7 @@ skip_size_check:
 
 /********************************************************************
 Starts InnoDB and creates a new database if database files
-are not found and the user wants. Server parameters are
-read from a file of name "srv_init" in the ib_home directory. */
+are not found and the user wants. */
 UNIV_INTERN
 int
 innobase_start_or_create_for_mysql(void)
@@ -1053,6 +1084,17 @@ innobase_start_or_create_for_mysql(void)
 			"InnoDB: The InnoDB memory heap is disabled\n");
 	}
 
+#ifdef HAVE_GCC_ATOMIC_BUILTINS
+#ifdef INNODB_RW_LOCKS_USE_ATOMICS
+	fprintf(stderr,
+		"InnoDB: Mutex and rw_lock use GCC atomic builtins.\n");
+#else
+	fprintf(stderr,
+		"InnoDB: Mutex use GCC atomic builtins.\n");
+#endif
+
+#endif
+
 	/* Since InnoDB does not currently clean up all its internal data
 	structures in MySQL Embedded Server Library server_end(), we
 	print an error message if someone tries to start up InnoDB a

sync/sync0arr.c

@@ -5,6 +5,38 @@ The wait array used in synchronization primitives
 
 Created 9/5/1995 Heikki Tuuri
 *******************************************************/
+/***********************************************************************
+# Copyright (c) 2008, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions
+# are met:
+#	* Redistributions of source code must retain the above copyright
+#	  notice, this list of conditions and the following disclaimer.
+#	* Redistributions in binary form must reproduce the above
+#	  copyright notice, this list of conditions and the following
+#	  disclaimer in the documentation and/or other materials
+#	  provided with the distribution.
+#	* Neither the name of the Google Inc. nor the names of its
+#	  contributors may be used to endorse or promote products
+#	  derived from this software without specific prior written
+#	  permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Note, the BSD license applies to the new code. The old code is GPL.
+***********************************************************************/
 
 #include "sync0arr.h"
 #ifdef UNIV_NONINL
@@ -295,25 +327,21 @@ sync_array_validate(
 }
 
 /***********************************************************************
-Puts the cell event in reset state. */
+Returns the event that the thread owning the cell waits for. */
 static
-ib_int64_t
-sync_cell_event_reset(
-/*==================*/
-				/* out: value of signal_count
-				at the time of reset. */
-	ulint		type,	/* in: lock type mutex/rw_lock */
-	void*		object) /* in: the rw_lock/mutex object */
+os_event_t
+sync_cell_get_event(
+/*================*/
+	sync_cell_t*	cell) /* in: non-empty sync array cell */
 {
+	ulint type = cell->request_type;
+
 	if (type == SYNC_MUTEX) {
-		return(os_event_reset(((mutex_t *) object)->event));
-#ifdef __WIN__
+		return(((mutex_t *) cell->wait_object)->event);
 	} else if (type == RW_LOCK_WAIT_EX) {
-		return(os_event_reset(
-		       ((rw_lock_t *) object)->wait_ex_event));
-#endif
-	} else {
-		return(os_event_reset(((rw_lock_t *) object)->event));
+		return(((rw_lock_t *) cell->wait_object)->wait_ex_event);
+	} else { /* RW_LOCK_SHARED and RW_LOCK_EX wait on the same event */
+		return(((rw_lock_t *) cell->wait_object)->event);
 	}
 }
 
@@ -332,6 +360,7 @@ sync_array_reserve_cell(
 	ulint*		index)	/* out: index of the reserved cell */
 {
 	sync_cell_t*	cell;
+	os_event_t      event;
 	ulint		i;
 
 	ut_a(object);
@@ -370,8 +399,8 @@ sync_array_reserve_cell(
 			/* Make sure the event is reset and also store
 			the value of signal_count at which the event
 			was reset. */
-			cell->signal_count = sync_cell_event_reset(type,
-								object);
+                        event = sync_cell_get_event(cell);
+			cell->signal_count = os_event_reset(event);
 
 			cell->reservation_time = time(NULL);
 
@@ -411,19 +440,7 @@ sync_array_wait_event(
 	ut_a(!cell->waiting);
 	ut_ad(os_thread_get_curr_id() == cell->thread);
 
-	if (cell->request_type == SYNC_MUTEX) {
-		event = ((mutex_t*) cell->wait_object)->event;
-#ifdef __WIN__
-	/* On windows if the thread about to wait is the one which
-	has set the state of the rw_lock to RW_LOCK_WAIT_EX, then
-	it waits on a special event i.e.: wait_ex_event. */
-	} else if (cell->request_type == RW_LOCK_WAIT_EX) {
-		event = ((rw_lock_t*) cell->wait_object)->wait_ex_event;
-#endif
-	} else {
-		event = ((rw_lock_t*) cell->wait_object)->event;
-	}
-
+	event = sync_cell_get_event(cell);
 		cell->waiting = TRUE;
 
 #ifdef UNIV_SYNC_DEBUG
@@ -462,6 +479,7 @@ sync_array_cell_print(
 	mutex_t*	mutex;
 	rw_lock_t*	rwlock;
 	ulint		type;
+	ulint		writer;
 
 	type = cell->request_type;
 
@@ -491,9 +509,7 @@ sync_array_cell_print(
 			(ulong) mutex->waiters);
 
 	} else if (type == RW_LOCK_EX
-#ifdef __WIN__
 		   || type == RW_LOCK_WAIT_EX
-#endif
 		   || type == RW_LOCK_SHARED) {
 
 		fputs(type == RW_LOCK_EX ? "X-lock on" : "S-lock on", file);
@@ -504,22 +520,25 @@ sync_array_cell_print(
 			" RW-latch at %p created in file %s line %lu\n",
 			(void*) rwlock, rwlock->cfile_name,
 			(ulong) rwlock->cline);
-		if (rwlock->writer != RW_LOCK_NOT_LOCKED) {
+		writer = rw_lock_get_writer(rwlock);
+		if (writer != RW_LOCK_NOT_LOCKED) {
 			fprintf(file,
 				"a writer (thread id %lu) has"
 				" reserved it in mode %s",
 				(ulong) os_thread_pf(rwlock->writer_thread),
-				rwlock->writer == RW_LOCK_EX
+				writer == RW_LOCK_EX
 				? " exclusive\n"
 				: " wait exclusive\n");
 		}
 
 		fprintf(file,
-			"number of readers %lu, waiters flag %lu\n"
+			"number of readers %lu, waiters flag %lu, "
+                        "lock_word: %lx\n"
 			"Last time read locked in file %s line %lu\n"
 			"Last time write locked in file %s line %lu\n",
-			(ulong) rwlock->reader_count,
+			(ulong) rw_lock_get_reader_count(rwlock),
 			(ulong) rwlock->waiters,
+			rwlock->lock_word,
 			rwlock->last_s_file_name,
 			(ulong) rwlock->last_s_line,
 			rwlock->last_x_file_name,
@@ -778,28 +797,30 @@ sync_arr_cell_can_wake_up(
 			return(TRUE);
 		}
 
-	} else if (cell->request_type == RW_LOCK_EX
-		   || cell->request_type == RW_LOCK_WAIT_EX) {
+	} else if (cell->request_type == RW_LOCK_EX) {
 
 		lock = cell->wait_object;
 
-		if (rw_lock_get_reader_count(lock) == 0
-		    && rw_lock_get_writer(lock) == RW_LOCK_NOT_LOCKED) {
+		if (lock->lock_word > 0) {
+		/* Either unlocked or only read locked. */
 
 			return(TRUE);
 		}
 
-		if (rw_lock_get_reader_count(lock) == 0
-		    && rw_lock_get_writer(lock) == RW_LOCK_WAIT_EX
-		    && os_thread_eq(lock->writer_thread, cell->thread)) {
+        } else if (cell->request_type == RW_LOCK_WAIT_EX) {
+
+		lock = cell->wait_object;
+
+                /* lock_word == 0 means all readers have left */
+		if (lock->lock_word == 0) {
 
 			return(TRUE);
 		}
-
 	} else if (cell->request_type == RW_LOCK_SHARED) {
 		lock = cell->wait_object;
 
-		if (rw_lock_get_writer(lock) == RW_LOCK_NOT_LOCKED) {
+                /* lock_word > 0 means no writer or reserved writer */
+		if (lock->lock_word > 0) {
 
 			return(TRUE);
 		}
@@ -844,11 +865,15 @@ sync_array_object_signalled(
 /*========================*/
 	sync_array_t*	arr)	/* in: wait array */
 {
+#ifdef HAVE_GCC_ATOMIC_BUILTINS
+	os_atomic_increment((lint*) &arr->sg_count, 1);
+#else
 	sync_array_enter(arr);
 
 	arr->sg_count++;
 
 	sync_array_exit(arr);
+#endif
 }
 
 /**************************************************************************
@@ -868,6 +893,7 @@ sync_arr_wake_threads_if_sema_free(void)
 	sync_cell_t*	cell;
 	ulint		count;
 	ulint		i;
+	os_event_t      event;
 
 	sync_array_enter(arr);
 
@@ -877,36 +903,20 @@ sync_arr_wake_threads_if_sema_free(void)
 	while (count < arr->n_reserved) {
 
 		cell = sync_array_get_nth_cell(arr, i);
+		i++;
 
-		if (cell->wait_object != NULL) {
-
+		if (cell->wait_object == NULL) {
+			continue;
+		}
 			count++;
 
 			if (sync_arr_cell_can_wake_up(cell)) {
 
-				if (cell->request_type == SYNC_MUTEX) {
-					mutex_t*	mutex;
+			event = sync_cell_get_event(cell);
 
-					mutex = cell->wait_object;
-					os_event_set(mutex->event);
-#ifdef __WIN__
-				} else if (cell->request_type
-					   == RW_LOCK_WAIT_EX) {
-					rw_lock_t*	lock;
-
-					lock = cell->wait_object;
-					os_event_set(lock->wait_ex_event);
-#endif
-				} else {
-					rw_lock_t*	lock;
-
-					lock = cell->wait_object;
-					os_event_set(lock->event);
-				}
-			}
+			os_event_set(event);
 		}
 
-		i++;
 	}
 
 	sync_array_exit(arr);
@@ -1026,4 +1036,3 @@ sync_array_print_info(
 
 	sync_array_exit(arr);
 }
-

sync/sync0rw.c

@@ -5,6 +5,38 @@ The read-write lock (for thread synchronization)
 
 Created 9/11/1995 Heikki Tuuri
 *******************************************************/
+/***********************************************************************
+# Copyright (c) 2008, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions
+# are met:
+#	* Redistributions of source code must retain the above copyright
+#	  notice, this list of conditions and the following disclaimer.
+#	* Redistributions in binary form must reproduce the above
+#	  copyright notice, this list of conditions and the following
+#	  disclaimer in the documentation and/or other materials
+#	  provided with the distribution.
+#	* Neither the name of the Google Inc. nor the names of its
+#	  contributors may be used to endorse or promote products
+#	  derived from this software without specific prior written
+#	  permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Note, the BSD license applies to the new code. The old code is GPL.
+***********************************************************************/
 
 #include "sync0rw.h"
 #ifdef UNIV_NONINL
@@ -15,35 +47,125 @@ Created 9/11/1995 Heikki Tuuri
 #include "mem0mem.h"
 #include "srv0srv.h"
 
-/* number of system calls made during shared latching */
-UNIV_INTERN ulint	rw_s_system_call_count	= 0;
+/*
+	IMPLEMENTATION OF THE RW_LOCK
+	=============================
+The status of a rw_lock is held in lock_word. The initial value of lock_word is
+X_LOCK_DECR. lock_word is decremented by 1 for each s-lock and by X_LOCK_DECR
+for each x-lock. This describes the lock state for each value of lock_word:
+
+lock_word == X_LOCK_DECR:      Unlocked.
+0 < lock_word < X_LOCK_DECR:   Read locked, no waiting writers.
+			       (X_LOCK_DECR - lock_word) is the
+			       number of readers that hold the lock.
+lock_word == 0:		       Write locked
+-X_LOCK_DECR < lock_word < 0:  Read locked, with a waiting writer.
+			       (-lock_word) is the number of readers
+			       that hold the lock.
+lock_word <= -X_LOCK_DECR:     Recursively write locked. lock_word has been
+			       decremented by X_LOCK_DECR once for each lock,
+			       so the number of locks is:
+			       ((-lock_word) / X_LOCK_DECR) + 1
+When lock_word <= -X_LOCK_DECR, we also know that lock_word % X_LOCK_DECR == 0:
+other values of lock_word are invalid.
+
+The lock_word is always read and updated atomically and consistently, so that
+it always represents the state of the lock, and the state of the lock changes
+with a single atomic operation. This lock_word holds all of the information
+that a thread needs in order to determine if it is eligible to gain the lock
+or if it must spin or sleep. The one exception to this is that writer_thread
+must be verified before recursive write locks: to solve this scenario, we make
+writer_thread readable by all threads, but only writeable by the x-lock holder.
+
+The other members of the lock obey the following rules to remain consistent:
+
+recursive:	This and the writer_thread field together control the
+		behaviour of recursive x-locking.
+		lock->recursive must be FALSE in following states:
+			1) The writer_thread contains garbage i.e.: the
+			lock has just been initialized.
+			2) The lock is not x-held and there is no
+			x-waiter waiting on WAIT_EX event.
+			3) The lock is x-held or there is an x-waiter
+			waiting on WAIT_EX event but the 'pass' value
+			is non-zero.
+		lock->recursive is TRUE iff:
+			1) The lock is x-held or there is an x-waiter
+			waiting on WAIT_EX event and the 'pass' value
+			is zero.
+		This flag must be set after the writer_thread field
+		has been updated with a memory ordering barrier.
+		It is unset before the lock_word has been incremented.
+writer_thread:	Is used only in recursive x-locking. Can only be safely
+		read iff lock->recursive flag is TRUE.
+		This field is uninitialized at lock creation time and
+		is updated atomically when x-lock is acquired or when
+		move_ownership is called. A thread is only allowed to
+		set the value of this field to it's thread_id i.e.: a
+		thread cannot set writer_thread to some other thread's
+		id.
+waiters:	May be set to 1 anytime, but to avoid unnecessary wake-up
+		signals, it should only be set to 1 when there are threads
+		waiting on event. Must be 1 when a writer starts waiting to
+		ensure the current x-locking thread sends a wake-up signal
+		during unlock. May only be reset to 0 immediately before a
+		a wake-up signal is sent to event. On most platforms, a
+		memory barrier is required after waiters is set, and before
+		verifying lock_word is still held, to ensure some unlocker
+		really does see the flags new value.
+event:		Threads wait on event for read or writer lock when another
+		thread has an x-lock or an x-lock reservation (wait_ex). A
+		thread may only	wait on event after performing the following
+		actions in order:
+		   (1) Record the counter value of event (with os_event_reset).
+		   (2) Set waiters to 1.
+		   (3) Verify lock_word <= 0.
+		(1) must come before (2) to ensure signal is not missed.
+		(2) must come before (3) to ensure a signal is sent.
+		These restrictions force the above ordering.
+		Immediately before sending the wake-up signal, we should:
+		   (1) Verify lock_word == X_LOCK_DECR (unlocked)
+		   (2) Reset waiters to 0.
+wait_ex_event:	A thread may only wait on the wait_ex_event after it has
+		performed the following actions in order:
+		   (1) Decrement lock_word by X_LOCK_DECR.
+		   (2) Record counter value of wait_ex_event (os_event_reset,
+                       called from sync_array_reserve_cell).
+		   (3) Verify that lock_word < 0.
+		(1) must come first to ensures no other threads become reader
+                or next writer, and notifies unlocker that signal must be sent.
+                (2) must come before (3) to ensure the signal is not missed.
+		These restrictions force the above ordering.
+		Immediately before sending the wake-up signal, we should:
+		   Verify lock_word == 0 (waiting thread holds x_lock)
+*/
+
 
 /* number of spin waits on rw-latches,
 resulted during shared (read) locks */
-UNIV_INTERN ulint	rw_s_spin_wait_count	= 0;
+UNIV_INTERN ib_int64_t	rw_s_spin_wait_count	= 0;
+UNIV_INTERN ib_int64_t	rw_s_spin_round_count	= 0;
 
 /* number of OS waits on rw-latches,
 resulted during shared (read) locks */
-UNIV_INTERN ulint	rw_s_os_wait_count	= 0;
+UNIV_INTERN ib_int64_t	rw_s_os_wait_count	= 0;
 
 /* number of unlocks (that unlock shared locks),
 set only when UNIV_SYNC_PERF_STAT is defined */
-UNIV_INTERN ulint	rw_s_exit_count		= 0;
-
-/* number of system calls made during exclusive latching */
-UNIV_INTERN ulint	rw_x_system_call_count	= 0;
+UNIV_INTERN ib_int64_t	rw_s_exit_count		= 0;
 
 /* number of spin waits on rw-latches,
 resulted during exclusive (write) locks */
-UNIV_INTERN ulint	rw_x_spin_wait_count	= 0;
+UNIV_INTERN ib_int64_t	rw_x_spin_wait_count	= 0;
+UNIV_INTERN ib_int64_t	rw_x_spin_round_count	= 0;
 
 /* number of OS waits on rw-latches,
 resulted during exclusive (write) locks */
-UNIV_INTERN ulint	rw_x_os_wait_count	= 0;
+UNIV_INTERN ib_int64_t	rw_x_os_wait_count	= 0;
 
 /* number of unlocks (that unlock exclusive locks),
 set only when UNIV_SYNC_PERF_STAT is defined */
-UNIV_INTERN ulint	rw_x_exit_count		= 0;
+UNIV_INTERN ib_int64_t	rw_x_exit_count		= 0;
 
 /* The global list of rw-locks */
 UNIV_INTERN rw_lock_list_t	rw_lock_list;
@@ -114,11 +236,12 @@ rw_lock_create_func(
 	const char*	cmutex_name, 	/* in: mutex name */
 #endif /* UNIV_DEBUG */
 	const char*	cfile_name,	/* in: file name where created */
-	ulint 		cline)		/* in: file line where created */
+	ulint		cline)		/* in: file line where created */
 {
 	/* If this is the very first time a synchronization object is
 	created, then the following call initializes the sync system. */
 
+#ifndef INNODB_RW_LOCKS_USE_ATOMICS
 	mutex_create(rw_lock_get_mutex(lock), SYNC_NO_ORDER_CHECK);
 
 	lock->mutex.cfile_name = cfile_name;
@@ -129,12 +252,19 @@ rw_lock_create_func(
 	lock->mutex.mutex_type = 1;
 #endif /* UNIV_DEBUG && !UNIV_HOTBACKUP */
 
-	rw_lock_set_waiters(lock, 0);
-	rw_lock_set_writer(lock, RW_LOCK_NOT_LOCKED);
-	lock->writer_count = 0;
-	rw_lock_set_reader_count(lock, 0);
+#else /* INNODB_RW_LOCKS_USE_ATOMICS */
+#ifdef UNIV_DEBUG
+	UT_NOT_USED(cmutex_name);
+#endif
+#endif /* INNODB_RW_LOCKS_USE_ATOMICS */
+
+	lock->lock_word = X_LOCK_DECR;
+	lock->waiters = 0;
 
-	lock->writer_is_wait_ex = FALSE;
+	/* We set this value to signify that lock->writer_thread
+	contains garbage at initialization and cannot be used for
+	recursive x-locking. */
+	lock->recursive = FALSE;
 
 #ifdef UNIV_SYNC_DEBUG
 	UT_LIST_INIT(lock->debug_list);
@@ -147,15 +277,13 @@ rw_lock_create_func(
 	lock->cfile_name = cfile_name;
 	lock->cline = (unsigned int) cline;
 
+	lock->count_os_wait = 0;
 	lock->last_s_file_name = "not yet reserved";
 	lock->last_x_file_name = "not yet reserved";
 	lock->last_s_line = 0;
 	lock->last_x_line = 0;
 	lock->event = os_event_create(NULL);
-
-#ifdef __WIN__
 	lock->wait_ex_event = os_event_create(NULL);
-#endif
 
 	mutex_enter(&rw_lock_list_mutex);
 
@@ -180,20 +308,18 @@ rw_lock_free(
 	rw_lock_t*	lock)	/* in: rw-lock */
 {
 	ut_ad(rw_lock_validate(lock));
-	ut_a(rw_lock_get_writer(lock) == RW_LOCK_NOT_LOCKED);
-	ut_a(rw_lock_get_waiters(lock) == 0);
-	ut_a(rw_lock_get_reader_count(lock) == 0);
+	ut_a(lock->lock_word == X_LOCK_DECR);
 
 	lock->magic_n = 0;
 
+#ifndef INNODB_RW_LOCKS_USE_ATOMICS
 	mutex_free(rw_lock_get_mutex(lock));
+#endif /* INNODB_RW_LOCKS_USE_ATOMICS */
 
 	mutex_enter(&rw_lock_list_mutex);
 	os_event_free(lock->event);
 
-#ifdef __WIN__
 	os_event_free(lock->wait_ex_event);
-#endif
 
 	if (UT_LIST_GET_PREV(list, lock)) {
 		ut_a(UT_LIST_GET_PREV(list, lock)->magic_n == RW_LOCK_MAGIC_N);
@@ -219,19 +345,12 @@ rw_lock_validate(
 {
 	ut_a(lock);
 
-	mutex_enter(rw_lock_get_mutex(lock));
+	ulint waiters = rw_lock_get_waiters(lock);
+	lint lock_word = lock->lock_word;
 
 	ut_a(lock->magic_n == RW_LOCK_MAGIC_N);
-	ut_a((rw_lock_get_reader_count(lock) == 0)
-	     || (rw_lock_get_writer(lock) != RW_LOCK_EX));
-	ut_a((rw_lock_get_writer(lock) == RW_LOCK_EX)
-	     || (rw_lock_get_writer(lock) == RW_LOCK_WAIT_EX)
-	     || (rw_lock_get_writer(lock) == RW_LOCK_NOT_LOCKED));
-	ut_a((rw_lock_get_waiters(lock) == 0)
-	     || (rw_lock_get_waiters(lock) == 1));
-	ut_a((lock->writer != RW_LOCK_EX) || (lock->writer_count > 0));
-
-	mutex_exit(rw_lock_get_mutex(lock));
+	ut_a(waiters == 0 || waiters == 1);
+	ut_a(lock_word > -X_LOCK_DECR ||(-lock_word) % X_LOCK_DECR == 0);
 
 	return(TRUE);
 }
@@ -253,18 +372,15 @@ rw_lock_s_lock_spin(
 	ulint		line)	/* in: line where requested */
 {
 	ulint	 index;	/* index of the reserved wait cell */
-	ulint	 i;	/* spin round count */
+	ulint	 i = 0;	/* spin round count */
 
 	ut_ad(rw_lock_validate(lock));
 
+	rw_s_spin_wait_count++;	/* Count calls to this function */
 lock_loop:
-	rw_s_spin_wait_count++;
 
 	/* Spin waiting for the writer field to become free */
-	i = 0;
-
-	while (rw_lock_get_writer(lock) != RW_LOCK_NOT_LOCKED
-	       && i < SYNC_SPIN_ROUNDS) {
+	while (i < SYNC_SPIN_ROUNDS && lock->lock_word <= 0) {
 		if (srv_spin_wait_delay) {
 			ut_delay(ut_rnd_interval(0, srv_spin_wait_delay));
 		}
@@ -285,28 +401,32 @@ lock_loop:
 			lock->cfile_name, (ulong) lock->cline, (ulong) i);
 	}
 
-	mutex_enter(rw_lock_get_mutex(lock));
-
 	/* We try once again to obtain the lock */
-
 	if (TRUE == rw_lock_s_lock_low(lock, pass, file_name, line)) {
-		mutex_exit(rw_lock_get_mutex(lock));
+		rw_s_spin_round_count += i;
 
 		return; /* Success */
 	} else {
-		/* If we get here, locking did not succeed, we may
-		suspend the thread to wait in the wait array */
 
-		rw_s_system_call_count++;
+		if (i < SYNC_SPIN_ROUNDS) {
+			goto lock_loop;
+		}
+
+		rw_s_spin_round_count += i;
 
 		sync_array_reserve_cell(sync_primary_wait_array,
 					lock, RW_LOCK_SHARED,
 					file_name, line,
 					&index);
 
-		rw_lock_set_waiters(lock, 1);
+		/* Set waiters before checking lock_word to ensure wake-up
+                signal is sent. This may lead to some unnecessary signals. */
+		rw_lock_set_waiter_flag(lock);
 
-		mutex_exit(rw_lock_get_mutex(lock));
+		if (TRUE == rw_lock_s_lock_low(lock, pass, file_name, line)) {
+			sync_array_free_cell(sync_primary_wait_array, index);
+			return; /* Success */
+		}
 
 		if (srv_print_latch_waits) {
 			fprintf(stderr,
@@ -317,11 +437,13 @@ lock_loop:
 				(ulong) lock->cline);
 		}
 
-		rw_s_system_call_count++;
+		/* these stats may not be accurate */
+		lock->count_os_wait++;
 		rw_s_os_wait_count++;
 
 		sync_array_wait_event(sync_primary_wait_array, index);
 
+		i = 0;
 		goto lock_loop;
 	}
 }
@@ -343,113 +465,130 @@ rw_lock_x_lock_move_ownership(
 {
 	ut_ad(rw_lock_is_locked(lock, RW_LOCK_EX));
 
-	mutex_enter(&(lock->mutex));
-
-	lock->writer_thread = os_thread_get_curr_id();
-
-	lock->pass = 0;
-
-	mutex_exit(&(lock->mutex));
+	rw_lock_set_writer_id_and_recursion_flag(lock, TRUE);
 }
 
 /**********************************************************************
-Low-level function for acquiring an exclusive lock. */
+Function for the next writer to call. Waits for readers to exit.
+The caller must have already decremented lock_word by X_LOCK_DECR.*/
 UNIV_INLINE
-ulint
-rw_lock_x_lock_low(
-/*===============*/
-				/* out: RW_LOCK_NOT_LOCKED if did
-				not succeed, RW_LOCK_EX if success,
-				RW_LOCK_WAIT_EX, if got wait reservation */
+void
+rw_lock_x_lock_wait(
+/*================*/
 	rw_lock_t*	lock,	/* in: pointer to rw-lock */
+#ifdef UNIV_SYNC_DEBUG
 	ulint		pass,	/* in: pass value; != 0, if the lock will
 				be passed to another thread to unlock */
+#endif
 	const char*	file_name,/* in: file name where lock requested */
 	ulint		line)	/* in: line where requested */
 {
-	ut_ad(mutex_own(rw_lock_get_mutex(lock)));
+	ulint index;
+	ulint i = 0;
 
-	if (rw_lock_get_writer(lock) == RW_LOCK_NOT_LOCKED) {
+	ut_ad(lock->lock_word <= 0);
 
-		if (rw_lock_get_reader_count(lock) == 0) {
-
-			rw_lock_set_writer(lock, RW_LOCK_EX);
-			lock->writer_thread = os_thread_get_curr_id();
-			lock->writer_count++;
-			lock->pass = pass;
+	while (lock->lock_word < 0) {
+		if (srv_spin_wait_delay) {
+			ut_delay(ut_rnd_interval(0, srv_spin_wait_delay));
+		}
+		if(i < SYNC_SPIN_ROUNDS) {
+			i++;
+			continue;
+		}
 
-#ifdef UNIV_SYNC_DEBUG
-			rw_lock_add_debug_info(lock, pass, RW_LOCK_EX,
-					       file_name, line);
-#endif
-			lock->last_x_file_name = file_name;
-			lock->last_x_line = (unsigned int) line;
+		/* If there is still a reader, then go to sleep.*/
+		rw_x_spin_round_count += i;
+		i = 0;
+		sync_array_reserve_cell(sync_primary_wait_array,
+					lock,
+					RW_LOCK_WAIT_EX,
+					file_name, line,
+					&index);
+		/* Check lock_word to ensure wake-up isn't missed.*/
+		if(lock->lock_word < 0) {
 
-			/* Locking succeeded, we may return */
-			return(RW_LOCK_EX);
-		} else {
-			/* There are readers, we have to wait */
-			rw_lock_set_writer(lock, RW_LOCK_WAIT_EX);
-			lock->writer_thread = os_thread_get_curr_id();
-			lock->pass = pass;
-			lock->writer_is_wait_ex = TRUE;
+			/* these stats may not be accurate */
+			lock->count_os_wait++;
+			rw_x_os_wait_count++;
 
+                        /* Add debug info as it is needed to detect possible
+                        deadlock. We must add info for WAIT_EX thread for
+                        deadlock detection to work properly. */
 #ifdef UNIV_SYNC_DEBUG
 			rw_lock_add_debug_info(lock, pass, RW_LOCK_WAIT_EX,
 					       file_name, line);
 #endif
 
-			return(RW_LOCK_WAIT_EX);
-		}
-
-	} else if ((rw_lock_get_writer(lock) == RW_LOCK_WAIT_EX)
-		   && os_thread_eq(lock->writer_thread,
-				   os_thread_get_curr_id())) {
-
-		if (rw_lock_get_reader_count(lock) == 0) {
-
-			rw_lock_set_writer(lock, RW_LOCK_EX);
-			lock->writer_count++;
-			lock->pass = pass;
-			lock->writer_is_wait_ex = FALSE;
-
+			sync_array_wait_event(sync_primary_wait_array,
+					      index);
 #ifdef UNIV_SYNC_DEBUG
-			rw_lock_remove_debug_info(lock, pass, RW_LOCK_WAIT_EX);
-			rw_lock_add_debug_info(lock, pass, RW_LOCK_EX,
-					       file_name, line);
+			rw_lock_remove_debug_info(lock, pass,
+					       RW_LOCK_WAIT_EX);
 #endif
-
-			lock->last_x_file_name = file_name;
-			lock->last_x_line = (unsigned int) line;
-
-			/* Locking succeeded, we may return */
-			return(RW_LOCK_EX);
+                        /* It is possible to wake when lock_word < 0.
+                        We must pass the while-loop check to proceed.*/
+		} else {
+			sync_array_free_cell(sync_primary_wait_array,
+					     index);
 		}
+	}
+	rw_x_spin_round_count += i;
+}
+
+/**********************************************************************
+Low-level function for acquiring an exclusive lock. */
+UNIV_INLINE
+ibool
+rw_lock_x_lock_low(
+/*===============*/
+				/* out: RW_LOCK_NOT_LOCKED if did
+				not succeed, RW_LOCK_EX if success. */
+	rw_lock_t*	lock,	/* in: pointer to rw-lock */
+	ulint		pass,	/* in: pass value; != 0, if the lock will
+				be passed to another thread to unlock */
+	const char*	file_name,/* in: file name where lock requested */
+	ulint		line)	/* in: line where requested */
+{
+	os_thread_id_t	curr_thread	= os_thread_get_curr_id();
 
-		return(RW_LOCK_WAIT_EX);
+	if (rw_lock_lock_word_decr(lock, X_LOCK_DECR)) {
 
-	} else if ((rw_lock_get_writer(lock) == RW_LOCK_EX)
-		   && os_thread_eq(lock->writer_thread,
-				   os_thread_get_curr_id())
-		   && (lock->pass == 0)
-		   && (pass == 0)) {
+		/* lock->recursive also tells us if the writer_thread
+		field is stale or active. As we are going to write
+		our own thread id in that field it must be that the
+		current writer_thread value is not active. */
+		ut_a(!lock->recursive);
 
-		lock->writer_count++;
+		/* Decrement occurred: we are writer or next-writer. */
+		rw_lock_set_writer_id_and_recursion_flag(lock,
+						pass ? FALSE : TRUE);
 
+		rw_lock_x_lock_wait(lock,
 #ifdef UNIV_SYNC_DEBUG
-		rw_lock_add_debug_info(lock, pass, RW_LOCK_EX, file_name,
-				       line);
+				    pass,
 #endif
+                                    file_name, line);
 
-		lock->last_x_file_name = file_name;
-		lock->last_x_line = (unsigned int) line;
-
-		/* Locking succeeded, we may return */
-		return(RW_LOCK_EX);
+	} else {
+		/* Decrement failed: relock or failed lock */
+		if (!pass && lock->recursive &&
+                   os_thread_eq(lock->writer_thread, curr_thread)) {
+			/* Relock */
+                        lock->lock_word -= X_LOCK_DECR;
+		} else {
+			/* Another thread locked before us */
+			return(FALSE);
+		}
 	}
+#ifdef UNIV_SYNC_DEBUG
+	rw_lock_add_debug_info(lock, pass, RW_LOCK_EX,
+			       file_name, line);
+#endif
+	lock->last_x_file_name = file_name;
+	lock->last_x_line = (unsigned int) line;
 
-	/* Locking did not succeed */
-	return(RW_LOCK_NOT_LOCKED);
+	return(TRUE);
 }
 
 /**********************************************************************
@@ -472,47 +611,30 @@ rw_lock_x_lock_func(
 	ulint		line)	/* in: line where requested */
 {
 	ulint	index;	/* index of the reserved wait cell */
-	ulint	state;	/* lock state acquired */
 	ulint	i;	/* spin round count */
+	ibool   spinning = FALSE;
 
 	ut_ad(rw_lock_validate(lock));
 
-lock_loop:
-	/* Acquire the mutex protecting the rw-lock fields */
-	mutex_enter_fast(&(lock->mutex));
-
-	state = rw_lock_x_lock_low(lock, pass, file_name, line);
+	i = 0;
 
-	mutex_exit(&(lock->mutex));
+lock_loop:
 
-	if (state == RW_LOCK_EX) {
+	if (rw_lock_x_lock_low(lock, pass, file_name, line)) {
+		rw_x_spin_round_count += i;
 
 		return;	/* Locking succeeded */
 
-	} else if (state == RW_LOCK_NOT_LOCKED) {
-
-		/* Spin waiting for the writer field to become free */
-		i = 0;
-
-		while (rw_lock_get_writer(lock) != RW_LOCK_NOT_LOCKED
-		       && i < SYNC_SPIN_ROUNDS) {
-			if (srv_spin_wait_delay) {
-				ut_delay(ut_rnd_interval(0,
-							 srv_spin_wait_delay));
-			}
+	} else {
 
-			i++;
-		}
-		if (i == SYNC_SPIN_ROUNDS) {
-			os_thread_yield();
+                if (!spinning) {
+                        spinning = TRUE;
+                        rw_x_spin_wait_count++;
 		}
-	} else if (state == RW_LOCK_WAIT_EX) {
 
-		/* Spin waiting for the reader count field to become zero */
-		i = 0;
-
-		while (rw_lock_get_reader_count(lock) != 0
-		       && i < SYNC_SPIN_ROUNDS) {
+		/* Spin waiting for the lock_word to become free */
+		while (i < SYNC_SPIN_ROUNDS
+		       && lock->lock_word <= 0) {
 			if (srv_spin_wait_delay) {
 				ut_delay(ut_rnd_interval(0,
 							 srv_spin_wait_delay));
@@ -522,12 +644,13 @@ lock_loop:
 		}
 		if (i == SYNC_SPIN_ROUNDS) {
 			os_thread_yield();
+		} else {
+			goto lock_loop;
 		}
-	} else {
-		i = 0; /* Eliminate a compiler warning */
-		ut_error;
 	}
 
+	rw_x_spin_round_count += i;
+
 	if (srv_print_latch_waits) {
 		fprintf(stderr,
 			"Thread %lu spin wait rw-x-lock at %p"
@@ -536,39 +659,20 @@ lock_loop:
 			lock->cfile_name, (ulong) lock->cline, (ulong) i);
 	}
 
-	rw_x_spin_wait_count++;
-
-	/* We try once again to obtain the lock. Acquire the mutex protecting
-	the rw-lock fields */
-
-	mutex_enter(rw_lock_get_mutex(lock));
-
-	state = rw_lock_x_lock_low(lock, pass, file_name, line);
-
-	if (state == RW_LOCK_EX) {
-		mutex_exit(rw_lock_get_mutex(lock));
-
-		return;	/* Locking succeeded */
-	}
-
-	rw_x_system_call_count++;
-
 	sync_array_reserve_cell(sync_primary_wait_array,
 				lock,
-#ifdef __WIN__
-				/* On windows RW_LOCK_WAIT_EX signifies
-				that this thread should wait on the
-				special wait_ex_event. */
-				(state == RW_LOCK_WAIT_EX)
-				 ? RW_LOCK_WAIT_EX :
-#endif
 				RW_LOCK_EX,
 				file_name, line,
 				&index);
 
-	rw_lock_set_waiters(lock, 1);
+	/* Waiters must be set before checking lock_word, to ensure signal
+	is sent. This could lead to a few unnecessary wake-up signals. */
+	rw_lock_set_waiter_flag(lock);
 
-	mutex_exit(rw_lock_get_mutex(lock));
+	if (rw_lock_x_lock_low(lock, pass, file_name, line)) {
+		sync_array_free_cell(sync_primary_wait_array, index);
+		return; /* Locking succeeded */
+	}
 
 	if (srv_print_latch_waits) {
 		fprintf(stderr,
@@ -578,11 +682,13 @@ lock_loop:
 			lock->cfile_name, (ulong) lock->cline);
 	}
 
-	rw_x_system_call_count++;
+	/* these stats may not be accurate */
+	lock->count_os_wait++;
 	rw_x_os_wait_count++;
 
 	sync_array_wait_event(sync_primary_wait_array, index);
 
+	i = 0;
 	goto lock_loop;
 }
 
@@ -730,7 +836,7 @@ rw_lock_own(
 	ut_ad(lock);
 	ut_ad(rw_lock_validate(lock));
 
-	mutex_enter(&(lock->mutex));
+	rw_lock_debug_mutex_enter();
 
 	info = UT_LIST_GET_FIRST(lock->debug_list);
 
@@ -740,7 +846,7 @@ rw_lock_own(
 		    && (info->pass == 0)
 		    && (info->lock_type == lock_type)) {
 
-			mutex_exit(&(lock->mutex));
+			rw_lock_debug_mutex_exit();
 			/* Found! */
 
 			return(TRUE);
@@ -748,7 +854,7 @@ rw_lock_own(
 
 		info = UT_LIST_GET_NEXT(list, info);
 	}
-	mutex_exit(&(lock->mutex));
+	rw_lock_debug_mutex_exit();
 
 	return(FALSE);
 }
@@ -770,22 +876,18 @@ rw_lock_is_locked(
 	ut_ad(lock);
 	ut_ad(rw_lock_validate(lock));
 
-	mutex_enter(&(lock->mutex));
-
 	if (lock_type == RW_LOCK_SHARED) {
-		if (lock->reader_count > 0) {
+		if (rw_lock_get_reader_count(lock) > 0) {
 			ret = TRUE;
 		}
 	} else if (lock_type == RW_LOCK_EX) {
-		if (lock->writer == RW_LOCK_EX) {
+		if (rw_lock_get_writer(lock) == RW_LOCK_EX) {
 			ret = TRUE;
 		}
 	} else {
 		ut_error;
 	}
 
-	mutex_exit(&(lock->mutex));
-
 	return(ret);
 }
 
@@ -814,11 +916,10 @@ rw_lock_list_print_info(
 
 		count++;
 
+#ifndef INNODB_RW_LOCKS_USE_ATOMICS
 		mutex_enter(&(lock->mutex));
-
-		if ((rw_lock_get_writer(lock) != RW_LOCK_NOT_LOCKED)
-		    || (rw_lock_get_reader_count(lock) != 0)
-		    || (rw_lock_get_waiters(lock) != 0)) {
+#endif
+		if (lock->lock_word != X_LOCK_DECR) {
 
 			fprintf(file, "RW-LOCK: %p ", (void*) lock);
 
@@ -834,8 +935,10 @@ rw_lock_list_print_info(
 				info = UT_LIST_GET_NEXT(list, info);
 			}
 		}
-
+#ifndef INNODB_RW_LOCKS_USE_ATOMICS
 		mutex_exit(&(lock->mutex));
+#endif
+
 		lock = UT_LIST_GET_NEXT(list, lock);
 	}
 
@@ -858,9 +961,10 @@ rw_lock_print(
 		"RW-LATCH INFO\n"
 		"RW-LATCH: %p ", (void*) lock);
 
-	if ((rw_lock_get_writer(lock) != RW_LOCK_NOT_LOCKED)
-	    || (rw_lock_get_reader_count(lock) != 0)
-	    || (rw_lock_get_waiters(lock) != 0)) {
+#ifndef INNODB_RW_LOCKS_USE_ATOMICS
+	mutex_enter(&(lock->mutex));
+#endif
+	if (lock->lock_word != X_LOCK_DECR) {
 
 		if (rw_lock_get_waiters(lock)) {
 			fputs(" Waiters for the lock exist\n", stderr);
@@ -874,6 +978,9 @@ rw_lock_print(
 			info = UT_LIST_GET_NEXT(list, info);
 		}
 	}
+#ifndef INNODB_RW_LOCKS_USE_ATOMICS
+	mutex_exit(&(lock->mutex));
+#endif
 }
 
 /*************************************************************************
@@ -922,14 +1029,11 @@ rw_lock_n_locked(void)
 	lock = UT_LIST_GET_FIRST(rw_lock_list);
 
 	while (lock != NULL) {
-		mutex_enter(rw_lock_get_mutex(lock));
 
-		if ((rw_lock_get_writer(lock) != RW_LOCK_NOT_LOCKED)
-		    || (rw_lock_get_reader_count(lock) != 0)) {
+		if (lock->lock_word != X_LOCK_DECR) {
 			count++;
 		}
 
-		mutex_exit(rw_lock_get_mutex(lock));
 		lock = UT_LIST_GET_NEXT(list, lock);
 	}
 

sync/sync0sync.c

@@ -5,6 +5,38 @@ Mutex, the basic synchronization primitive
 
 Created 9/5/1995 Heikki Tuuri
 *******************************************************/
+/***********************************************************************
+# Copyright (c) 2008, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions
+# are met:
+#	* Redistributions of source code must retain the above copyright
+#	  notice, this list of conditions and the following disclaimer.
+#	* Redistributions in binary form must reproduce the above
+#	  copyright notice, this list of conditions and the following
+#	  disclaimer in the documentation and/or other materials
+#	  provided with the distribution.
+#	* Neither the name of the Google Inc. nor the names of its
+#	  contributors may be used to endorse or promote products
+#	  derived from this software without specific prior written
+#	  permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Note, the BSD license applies to the new code. The old code is GPL.
+***********************************************************************/
 
 #include "sync0sync.h"
 #ifdef UNIV_NONINL
@@ -138,18 +170,13 @@ Therefore, this thread is guaranteed to catch the os_set_event()
 signalled unconditionally at the release of the lock.
 Q.E.D. */
 
-/* The number of system calls made in this module. Intended for performance
-monitoring. */
-
-UNIV_INTERN ulint	mutex_system_call_count		= 0;
-
 /* Number of spin waits on mutexes: for performance monitoring */
 
 /* round=one iteration of a spin loop */
-UNIV_INTERN ulint	mutex_spin_round_count		= 0;
-UNIV_INTERN ulint	mutex_spin_wait_count		= 0;
-UNIV_INTERN ulint	mutex_os_wait_count		= 0;
-UNIV_INTERN ulint	mutex_exit_count		= 0;
+UNIV_INTERN ib_int64_t	mutex_spin_round_count		= 0;
+UNIV_INTERN ib_int64_t	mutex_spin_wait_count		= 0;
+UNIV_INTERN ib_int64_t	mutex_os_wait_count		= 0;
+UNIV_INTERN ib_int64_t	mutex_exit_count		= 0;
 
 /* The global array of wait cells for implementation of the database's own
 mutexes and read-write locks */
@@ -219,6 +246,8 @@ mutex_create_func(
 {
 #if defined(_WIN32) && defined(UNIV_CAN_USE_X86_ASSEMBLER)
 	mutex_reset_lock_word(mutex);
+#elif defined(HAVE_GCC_ATOMIC_BUILTINS)
+	mutex_reset_lock_word(mutex);
 #else
 	os_fast_mutex_init(&(mutex->os_fast_mutex));
 	mutex->lock_word = 0;
@@ -309,7 +338,9 @@ mutex_free(
 
 	os_event_free(mutex->event);
 
-#if !defined(_WIN32) || !defined(UNIV_CAN_USE_X86_ASSEMBLER)
+#if defined(_WIN32) && defined(UNIV_CAN_USE_X86_ASSEMBLER)
+#elif defined(HAVE_GCC_ATOMIC_BUILTINS)
+#else
 	os_fast_mutex_free(&(mutex->os_fast_mutex));
 #endif
 	/* If we free the mutex protecting the mutex list (freeing is
@@ -426,6 +457,12 @@ mutex_spin_wait(
 #endif /* UNIV_DEBUG && !UNIV_HOTBACKUP */
 	ut_ad(mutex);
 
+	/* This update is not thread safe, but we don't mind if the count
+	isn't exact. Moved out of ifdef that follows because we are willing
+	to sacrifice the cost of counting this as the data is valuable.
+	Count the number of calls to mutex_spin_wait. */
+	mutex_spin_wait_count++;
+
 mutex_loop:
 
 	i = 0;
@@ -438,7 +475,6 @@ mutex_loop:
 
 spin_loop:
 #if defined UNIV_DEBUG && !defined UNIV_HOTBACKUP
-	mutex_spin_wait_count++;
 	mutex->count_spin_loop++;
 #endif /* UNIV_DEBUG && !UNIV_HOTBACKUP */
 
@@ -503,8 +539,6 @@ spin_loop:
 	sync_array_reserve_cell(sync_primary_wait_array, mutex,
 				SYNC_MUTEX, file_name, line, &index);
 
-	mutex_system_call_count++;
-
 	/* The memory order of the array reservation and the change in the
 	waiters field is important: when we suspend a thread, we first
 	reserve the cell and then set waiters field to 1. When threads are
@@ -551,7 +585,6 @@ spin_loop:
 		mutex->cfile_name, (ulong) mutex->cline, (ulong) i);
 #endif
 
-	mutex_system_call_count++;
 	mutex_os_wait_count++;
 
 #ifndef UNIV_HOTBACKUP
@@ -1340,21 +1373,31 @@ sync_print_wait_info(
 	FILE*	file)		/* in: file where to print */
 {
 #ifdef UNIV_SYNC_DEBUG
-	fprintf(file, "Mutex exits %lu, rws exits %lu, rwx exits %lu\n",
+	fprintf(file, "Mutex exits %llu, rws exits %llu, rwx exits %llu\n",
 		mutex_exit_count, rw_s_exit_count, rw_x_exit_count);
 #endif
 
 	fprintf(file,
-		"Mutex spin waits %lu, rounds %lu, OS waits %lu\n"
-		"RW-shared spins %lu, OS waits %lu;"
-		" RW-excl spins %lu, OS waits %lu\n",
-		(ulong) mutex_spin_wait_count,
-		(ulong) mutex_spin_round_count,
-		(ulong) mutex_os_wait_count,
-		(ulong) rw_s_spin_wait_count,
-		(ulong) rw_s_os_wait_count,
-		(ulong) rw_x_spin_wait_count,
-		(ulong) rw_x_os_wait_count);
+		"Mutex spin waits %llu, rounds %llu, OS waits %llu\n"
+		"RW-shared spins %llu, OS waits %llu;"
+		" RW-excl spins %llu, OS waits %llu\n",
+		mutex_spin_wait_count,
+		mutex_spin_round_count,
+		mutex_os_wait_count,
+		rw_s_spin_wait_count,
+		rw_s_os_wait_count,
+		rw_x_spin_wait_count,
+		rw_x_os_wait_count);
+
+	fprintf(file,
+		"Spin rounds per wait: %.2f mutex, %.2f RW-shared, "
+		"%.2f RW-excl\n",
+		(double) mutex_spin_round_count /
+		(mutex_spin_wait_count ? mutex_spin_wait_count : 1),
+		(double) rw_s_spin_round_count /
+		(rw_s_spin_wait_count ? rw_s_spin_wait_count : 1),
+		(double) rw_x_spin_round_count /
+		(rw_x_spin_wait_count ? rw_x_spin_wait_count : 1));
 }
 
 /***********************************************************************

ut/ut0mem.c

@@ -15,6 +15,7 @@ Created 5/11/1994 Heikki Tuuri
 #include "mem0mem.h"
 #include "os0sync.h"
 #include "os0thread.h"
+#include "srv0srv.h"
 
 /* This struct is placed first in every allocated memory block */
 typedef struct ut_mem_block_struct ut_mem_block_t;
@@ -68,14 +69,29 @@ ut_malloc_low(
 	ibool	assert_on_error)/* in: if TRUE, we crash mysqld if the
 				memory cannot be allocated */
 {
-	ulint	retry_count	= 0;
+	ulint	retry_count;
 	void*	ret;
 
+	if (srv_use_sys_malloc) {
+		ret = malloc(n);
+		ut_a(ret || !assert_on_error);
+
+#ifdef UNIV_SET_MEM_TO_ZERO
+		if (set_to_zero) {
+			memset(ret, '\0', n);
+			UNIV_MEM_ALLOC(ret, n);
+		}
+#endif
+		return(ret);
+	}
+
 	ut_ad((sizeof(ut_mem_block_t) % 8) == 0); /* check alignment ok */
 
-	if (!ut_mem_block_list_inited) {
+	if (UNIV_UNLIKELY(!ut_mem_block_list_inited)) {
 		ut_mem_block_list_init();
 	}
+
+	retry_count = 0;
 retry:
 	os_fast_mutex_lock(&ut_list_mutex);
 
@@ -239,6 +255,11 @@ ut_free(
 {
 	ut_mem_block_t* block;
 
+	if (srv_use_sys_malloc) {
+		free(ptr);
+		return;
+	}
+
 	block = (ut_mem_block_t*)((byte*)ptr - sizeof(ut_mem_block_t));
 
 	os_fast_mutex_lock(&ut_list_mutex);
@@ -291,6 +312,10 @@ ut_realloc(
 	ulint		min_size;
 	void*		new_ptr;
 
+	if (srv_use_sys_malloc) {
+		return(realloc(ptr, size));
+	}
+
 	if (ptr == NULL) {
 
 		return(ut_malloc(size));
@@ -338,6 +363,11 @@ ut_free_all_mem(void)
 {
 	ut_mem_block_t* block;
 
+	if (!ut_mem_block_list_inited) {
+		return;
+	}
+
+	ut_mem_block_list_inited = FALSE;
 	os_fast_mutex_free(&ut_list_mutex);
 
 	while ((block = UT_LIST_GET_FIRST(ut_mem_block_list))) {