sync with rev40 extensions-1.0

btr/btr0cur.c

@@ -3690,6 +3690,7 @@ btr_blob_free(
 	mtr_commit(mtr);
 
 	//buf_pool_mutex_enter();
+	mutex_enter(&LRU_list_mutex);
 	mutex_enter(&block->mutex);
 
 	/* Only free the block if it is still allocated to
@@ -3700,7 +3701,7 @@ btr_blob_free(
 	    && buf_block_get_space(block) == space
 	    && buf_block_get_page_no(block) == page_no) {
 
-		if (buf_LRU_free_block(&block->page, all, NULL)
+		if (buf_LRU_free_block(&block->page, all, NULL, TRUE)
 		    != BUF_LRU_FREED
 		    && all && block->page.zip.data
 		    /* Now, buf_LRU_free_block() may release mutex temporarily */
@@ -3710,11 +3711,12 @@ btr_blob_free(
 			/* Attempt to deallocate the uncompressed page
 			if the whole block cannot be deallocted. */
 
-			buf_LRU_free_block(&block->page, FALSE, NULL);
+			buf_LRU_free_block(&block->page, FALSE, NULL, TRUE);
 		}
 	}
 
 	//buf_pool_mutex_exit();
+	mutex_exit(&LRU_list_mutex);
 	mutex_exit(&block->mutex);
 }
 

btr/btr0sea.c

@@ -800,7 +800,10 @@ btr_search_guess_on_hash(
 		rw_lock_s_lock(&btr_search_latch);
 	}
 
+#ifndef HAVE_GCC_ATOMIC_BUILTINS
+	/* It is used as lock word among x_lock */
 	ut_ad(btr_search_latch.writer != RW_LOCK_EX);
+#endif
 	ut_ad(btr_search_latch.reader_count > 0);
 
 	rec = ha_search_and_get_data(btr_search_sys->hash_index, fold);

buf/buf0buddy.c

@@ -123,7 +123,7 @@ buf_buddy_alloc_zip(
 	/* Valgrind would complain about accessing free memory. */
 	UT_LIST_VALIDATE(list, buf_page_t, buf_pool->zip_free[i]);
 #endif /* UNIV_DEBUG && !UNIV_DEBUG_VALGRIND */
-	bpage = UT_LIST_GET_FIRST(buf_pool->zip_free[i]);
+	bpage = UT_LIST_GET_LAST(buf_pool->zip_free[i]);
 
 	if (bpage) {
 		UNIV_MEM_VALID(bpage, BUF_BUDDY_LOW << i);
@@ -315,14 +315,18 @@ buf_buddy_alloc_low(
 	/* Try replacing an uncompressed page in the buffer pool. */
 	//buf_pool_mutex_exit();
 	mutex_exit(&LRU_list_mutex);
-	if (have_page_hash_mutex)
+	if (have_page_hash_mutex) {
+		mutex_exit(&flush_list_mutex);
 		mutex_exit(&page_hash_mutex);
+	}
 	block = buf_LRU_get_free_block(0);
 	*lru = TRUE;
 	//buf_pool_mutex_enter();
 	mutex_enter(&LRU_list_mutex);
-	if (have_page_hash_mutex)
+	if (have_page_hash_mutex) {
+		mutex_enter(&flush_list_mutex);
 		mutex_enter(&page_hash_mutex);
+	}
 
 alloc_big:
 	buf_buddy_block_register(block);
@@ -350,6 +354,7 @@ buf_buddy_relocate_block(
 	buf_page_t*	b;
 
 	//ut_ad(buf_pool_mutex_own());
+	ut_ad(mutex_own(&flush_list_mutex));
 
 	switch (buf_page_get_state(bpage)) {
 	case BUF_BLOCK_ZIP_FREE:
@@ -358,7 +363,7 @@ buf_buddy_relocate_block(
 	case BUF_BLOCK_FILE_PAGE:
 	case BUF_BLOCK_MEMORY:
 	case BUF_BLOCK_REMOVE_HASH:
-		ut_error;
+		/* ut_error; */ /* optimistic */
 	case BUF_BLOCK_ZIP_DIRTY:
 		/* Cannot relocate dirty pages. */
 		return(FALSE);
@@ -367,30 +372,24 @@ buf_buddy_relocate_block(
 		break;
 	}
 
-	/* optimistic */
+	mutex_enter(&buf_pool_zip_mutex);
+	mutex_enter(&zip_free_mutex);
+
 	if (!buf_page_can_relocate(bpage)) {
+		mutex_exit(&buf_pool_zip_mutex);
+		mutex_exit(&zip_free_mutex);
 		return(FALSE);
 	}
 
-	mutex_enter(&LRU_list_mutex);
-	mutex_enter(&flush_list_mutex);
-	mutex_enter(&page_hash_mutex);
-	mutex_enter(&buf_pool_zip_mutex);
-
-	if (!buf_page_can_relocate(bpage)) {
-		mutex_exit(&LRU_list_mutex);
-		mutex_exit(&flush_list_mutex);
-		mutex_exit(&page_hash_mutex);
+	if (bpage != buf_page_hash_get(bpage->space, bpage->offset)) {
 		mutex_exit(&buf_pool_zip_mutex);
+		mutex_exit(&zip_free_mutex);
 		return(FALSE);
 	}
 
 	buf_relocate(bpage, dpage);
 	ut_d(bpage->state = BUF_BLOCK_ZIP_FREE);
 
-	mutex_exit(&LRU_list_mutex);
-	mutex_exit(&page_hash_mutex);
-
 	/* relocate buf_pool->zip_clean */
 	b = UT_LIST_GET_PREV(list, dpage);
 	UT_LIST_REMOVE(list, buf_pool->zip_clean, dpage);
@@ -401,8 +400,8 @@ buf_buddy_relocate_block(
 		UT_LIST_ADD_FIRST(list, buf_pool->zip_clean, dpage);
 	}
 
-	mutex_exit(&flush_list_mutex);
 	mutex_exit(&buf_pool_zip_mutex);
+	mutex_exit(&zip_free_mutex);
 	return(TRUE);
 }
 
@@ -448,8 +447,11 @@ buf_buddy_relocate(
 		/* This is a compressed page. */
 		mutex_t*	mutex;
 
-		if (!have_page_hash_mutex)
+		if (!have_page_hash_mutex) {
+			mutex_enter(&LRU_list_mutex);
+			mutex_enter(&flush_list_mutex);
 			mutex_enter(&page_hash_mutex);
+		}
 		/* The src block may be split into smaller blocks,
 		some of which may be free.  Thus, the
 		mach_read_from_4() calls below may attempt to read
@@ -471,8 +473,10 @@ buf_buddy_relocate(
 			it cannot be relocated. */
 
 			if (!have_page_hash_mutex) {
-				mutex_exit(&page_hash_mutex);
 				mutex_enter(&zip_free_mutex);
+				mutex_exit(&LRU_list_mutex);
+				mutex_exit(&flush_list_mutex);
+				mutex_exit(&page_hash_mutex);
 			}
 			return(FALSE);
 		}
@@ -484,8 +488,10 @@ buf_buddy_relocate(
 			ut_ad(page_zip_get_size(&bpage->zip) < size);
 
 			if (!have_page_hash_mutex) {
-				mutex_exit(&page_hash_mutex);
 				mutex_enter(&zip_free_mutex);
+				mutex_exit(&LRU_list_mutex);
+				mutex_exit(&flush_list_mutex);
+				mutex_exit(&page_hash_mutex);
 			}
 			return(FALSE);
 		}
@@ -501,8 +507,7 @@ buf_buddy_relocate(
 		mutex = buf_page_get_mutex(bpage);
 
 		mutex_enter(mutex);
-		if (!have_page_hash_mutex)
-			mutex_exit(&page_hash_mutex);
+		mutex_enter(&zip_free_mutex);
 
 		if (buf_page_can_relocate(bpage)) {
 			/* Relocate the compressed page. */
@@ -520,29 +525,52 @@ buf_buddy_relocate(
 					+= ut_time_us(NULL) - usec;
 			}
 
-			mutex_enter(&zip_free_mutex);
+			if (!have_page_hash_mutex) {
+				mutex_exit(&LRU_list_mutex);
+				mutex_exit(&flush_list_mutex);
+				mutex_exit(&page_hash_mutex);
+			}
 			return(TRUE);
 		}
 
+		if (!have_page_hash_mutex) {
+			mutex_exit(&LRU_list_mutex);
+			mutex_exit(&flush_list_mutex);
+			mutex_exit(&page_hash_mutex);
+		}
+
 		mutex_exit(mutex);
-		mutex_enter(&zip_free_mutex);
 	} else if (i == buf_buddy_get_slot(sizeof(buf_page_t))) {
 		/* This must be a buf_page_t object. */
 		UNIV_MEM_ASSERT_RW(src, size);
 
-		if (have_page_hash_mutex)
-			mutex_exit(&page_hash_mutex);
-
 		mutex_exit(&zip_free_mutex);
+
+		if (!have_page_hash_mutex) {
+			mutex_enter(&LRU_list_mutex);
+			mutex_enter(&flush_list_mutex);
+			mutex_enter(&page_hash_mutex);
+		}
+
 		if (buf_buddy_relocate_block(src, dst)) {
+			mutex_enter(&zip_free_mutex);
 
-			goto success;
+			if (!have_page_hash_mutex) {
+				mutex_exit(&LRU_list_mutex);
+				mutex_exit(&flush_list_mutex);
+				mutex_exit(&page_hash_mutex);
 			}
 
-		if (have_page_hash_mutex)
-			mutex_enter(&page_hash_mutex);
+			goto success;
+		}
 
 		mutex_enter(&zip_free_mutex);
+
+		if (!have_page_hash_mutex) {
+			mutex_exit(&LRU_list_mutex);
+			mutex_exit(&flush_list_mutex);
+			mutex_exit(&page_hash_mutex);
+		}
 	}
 
 	return(FALSE);
@@ -632,7 +660,7 @@ buf_buddy_free_low(
 #endif /* UNIV_DEBUG_VALGRIND */
 
 	/* The buddy is not free. Is there a free block of this size? */
-	bpage = UT_LIST_GET_FIRST(buf_pool->zip_free[i]);
+	bpage = UT_LIST_GET_LAST(buf_pool->zip_free[i]);
 
 	if (bpage) {
 		/* Remove the block from the free list, because a successful

buf/buf0buf.c

@@ -662,10 +662,10 @@ buf_block_init(
 	block->page.in_zip_hash = FALSE;
 	block->page.in_flush_list = FALSE;
 	block->page.in_free_list = FALSE;
-	block->page.in_LRU_list = FALSE;
-	block->in_unzip_LRU_list = FALSE;
 	block->n_pointers = 0;
 #endif /* UNIV_DEBUG */
+	block->page.in_LRU_list = FALSE;
+	block->in_unzip_LRU_list = FALSE;
 	page_zip_des_init(&block->page.zip);
 
 	mutex_create(&block->mutex, SYNC_BUF_BLOCK);
@@ -1191,7 +1191,7 @@ buf_pool_shrink(
 
 				buf_LRU_make_block_old(&block->page);
 				dirty++;
-			} else if (buf_LRU_free_block(&block->page, TRUE, NULL)
+			} else if (buf_LRU_free_block(&block->page, TRUE, NULL, FALSE)
 				   != BUF_LRU_FREED) {
 				nonfree++;
 			}
@@ -1662,7 +1662,7 @@ buf_page_get_zip(
 		break;
 	case BUF_BLOCK_FILE_PAGE:
 		/* Discard the uncompressed page frame if possible. */
-		if (buf_LRU_free_block(bpage, FALSE, NULL)
+		if (buf_LRU_free_block(bpage, FALSE, NULL, FALSE)
 		    == BUF_LRU_FREED) {
 
 			mutex_exit(block_mutex);
@@ -1953,10 +1953,16 @@ buf_page_get_gen(
 		ibool		success;
 
 	case BUF_BLOCK_FILE_PAGE:
+		if (block_mutex == &buf_pool_zip_mutex) {
+			/* it is wrong mutex... */
+			mutex_exit(block_mutex);
+			goto loop;
+		}
 		break;
 
 	case BUF_BLOCK_ZIP_PAGE:
 	case BUF_BLOCK_ZIP_DIRTY:
+		ut_ad(block_mutex == &buf_pool_zip_mutex);
 		bpage = &block->page;
 
 		if (bpage->buf_fix_count
@@ -1980,7 +1986,7 @@ buf_page_get_gen(
 
 		block = buf_LRU_get_free_block(0);
 		ut_a(block);
-		block_mutex = buf_page_get_mutex((buf_page_t*)block);
+		block_mutex = &block->mutex;
 
 		//buf_pool_mutex_enter();
 		mutex_enter(&LRU_list_mutex);
@@ -2001,9 +2007,10 @@ buf_page_get_gen(
 				mutex_exit(block_mutex);
 
 				block = (buf_block_t*) hash_bpage;
+				if (block) {
 					block_mutex = buf_page_get_mutex((buf_page_t*)block);
-
 					mutex_enter(block_mutex);
+				}
 				mutex_exit(&page_hash_mutex);
 				mutex_exit(&LRU_list_mutex);
 				mutex_exit(&flush_list_mutex);
@@ -2107,6 +2114,7 @@ buf_page_get_gen(
 
 		/* Unfix and unlatch the block. */
 		//buf_pool_mutex_enter();
+		block_mutex = &block->mutex;
 		mutex_enter(block_mutex);
 		mutex_enter(&buf_pool_mutex);
 		buf_pool->n_pend_unzip--;
@@ -2119,7 +2127,7 @@ buf_page_get_gen(
 		if (UNIV_UNLIKELY(!success)) {
 
 			//buf_pool_mutex_exit();
-			mutex_exit(&block->mutex);
+			mutex_exit(block_mutex);
 			return(NULL);
 		}
 
@@ -2692,6 +2700,9 @@ buf_page_init_for_read(
 
 	//buf_pool_mutex_enter();
 	mutex_enter(&LRU_list_mutex);
+	if(!block) {
+		mutex_enter(&flush_list_mutex);
+	}
 	mutex_enter(&page_hash_mutex);
 
 	if (buf_page_hash_get(space, offset)) {
@@ -2699,14 +2710,18 @@ buf_page_init_for_read(
 err_exit:
 		if (block) {
 			mutex_enter(&block->mutex);
-			buf_LRU_block_free_non_file_page(block, TRUE);
+			mutex_exit(&LRU_list_mutex);
+			mutex_exit(&page_hash_mutex);
+			buf_LRU_block_free_non_file_page(block, FALSE);
 			mutex_exit(&block->mutex);
 		}
-
+		else {
 err_exit2:
 		//buf_pool_mutex_exit();
 		mutex_exit(&LRU_list_mutex);
+		mutex_exit(&flush_list_mutex);
 		mutex_exit(&page_hash_mutex);
+		}
 
 		if (mode == BUF_READ_IBUF_PAGES_ONLY) {
 
@@ -2790,16 +2805,14 @@ buf_page_init_for_read(
 		/* If buf_buddy_alloc() allocated storage from the LRU list,
 		it released and reacquired buf_pool_mutex.  Thus, we must
 		check the page_hash again, as it may have been modified. */
-		/* !!! page_hash_mutex is not released yet
-		So always buf_page_hash_get(space, offset) == 0 !!!*/
-		//if (UNIV_UNLIKELY(lru)
-		//    && UNIV_LIKELY_NULL(buf_page_hash_get(space, offset))) {
-
-		//	/* The block was added by some other thread. */
-		//	buf_buddy_free(bpage, sizeof *bpage);
-		//	buf_buddy_free(data, zip_size);
-		//	goto err_exit2;
-		//}
+		if (UNIV_UNLIKELY(lru)
+		    && UNIV_LIKELY_NULL(buf_page_hash_get(space, offset))) {
+
+			/* The block was added by some other thread. */
+			buf_buddy_free(bpage, sizeof *bpage, TRUE);
+			buf_buddy_free(data, zip_size, TRUE);
+			goto err_exit2;
+		}
 
 		page_zip_des_init(&bpage->zip);
 		page_zip_set_size(&bpage->zip, zip_size);
@@ -2818,8 +2831,8 @@ buf_page_init_for_read(
 		bpage->in_zip_hash = FALSE;
 		bpage->in_flush_list = FALSE;
 		bpage->in_free_list = FALSE;
-		bpage->in_LRU_list = FALSE;
 #endif /* UNIV_DEBUG */
+		bpage->in_LRU_list = FALSE;
 
 		ut_d(bpage->in_page_hash = TRUE);
 		HASH_INSERT(buf_page_t, hash, buf_pool->page_hash,
@@ -2832,6 +2845,7 @@ buf_page_init_for_read(
 		buf_LRU_insert_zip_clean(bpage);
 
 		mutex_exit(&LRU_list_mutex);
+		mutex_exit(&flush_list_mutex);
 
 		buf_page_set_io_fix(bpage, BUF_IO_READ);
 
@@ -3465,12 +3479,15 @@ buf_validate(void)
 	}
 
 	ut_a(UT_LIST_GET_LEN(buf_pool->LRU) == n_lru);
+	/* because of latching order, we cannot get free_list_mutex here. */
+/*
 	if (UT_LIST_GET_LEN(buf_pool->free) != n_free) {
 		fprintf(stderr, "Free list len %lu, free blocks %lu\n",
 			(ulong) UT_LIST_GET_LEN(buf_pool->free),
 			(ulong) n_free);
 		ut_error;
 	}
+*/
 	ut_a(UT_LIST_GET_LEN(buf_pool->flush_list) == n_flush);
 
 	ut_a(buf_pool->n_flush[BUF_FLUSH_SINGLE_PAGE] == n_single_flush);

buf/buf0flu.c

@@ -154,10 +154,10 @@ buf_flush_ready_for_replace(
 				buf_page_in_file(bpage) and in the LRU list */
 {
 	//ut_ad(buf_pool_mutex_own());
-	ut_ad(mutex_own(buf_page_get_mutex(bpage)));
-	ut_ad(bpage->in_LRU_list);
+	//ut_ad(mutex_own(buf_page_get_mutex(bpage)));
+	//ut_ad(bpage->in_LRU_list);
 
-	if (UNIV_LIKELY(buf_page_in_file(bpage))) {
+	if (UNIV_LIKELY(bpage->in_LRU_list && buf_page_in_file(bpage))) {
 
 		return(bpage->oldest_modification == 0
 		       && buf_page_get_io_fix(bpage) == BUF_IO_NONE
@@ -1237,8 +1237,14 @@ buf_flush_LRU_recommendation(void)
 	buf_page_t*	bpage;
 	ulint		n_replaceable;
 	ulint		distance	= 0;
+	ibool		have_LRU_mutex = FALSE;
+
+	if(UT_LIST_GET_LEN(buf_pool->unzip_LRU))
+		have_LRU_mutex = TRUE;
 
 	//buf_pool_mutex_enter();
+	if (have_LRU_mutex)
+		mutex_enter(&LRU_list_mutex);
 
 	n_replaceable = UT_LIST_GET_LEN(buf_pool->free);
 
@@ -1249,6 +1255,12 @@ buf_flush_LRU_recommendation(void)
 		   + BUF_FLUSH_EXTRA_MARGIN)
 	       && (distance < BUF_LRU_FREE_SEARCH_LEN)) {
 
+		if (!bpage->in_LRU_list) {
+			/* reatart. but it is very optimistic */
+			bpage = UT_LIST_GET_LAST(buf_pool->LRU);
+			continue;
+		}
+
 		mutex_t* block_mutex = buf_page_get_mutex(bpage);
 
 		mutex_enter(block_mutex);
@@ -1265,6 +1277,8 @@ buf_flush_LRU_recommendation(void)
 	}
 
 	//buf_pool_mutex_exit();
+	if (have_LRU_mutex)
+		mutex_exit(&LRU_list_mutex);
 
 	if (n_replaceable >= BUF_FLUSH_FREE_BLOCK_MARGIN) {
 

handler/ha_innodb.cc

@@ -8842,6 +8842,10 @@ innobase_xa_prepare(
 	if (thd_sql_command(thd) != SQLCOM_XA_PREPARE &&
 	    (all || !thd_test_options(thd, OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN)))
 	{
+		if (srv_enable_unsafe_group_commit && !THDVAR(thd, support_xa)) {
+			/* choose group commit rather than binlog order */
+			return(0);
+		}
 
 		/* For ibbackup to work the order of transactions in binlog
 		and InnoDB must be the same. Consider the situation
@@ -9646,6 +9650,11 @@ static MYSQL_SYSVAR_ULONG(adaptive_checkpoint, srv_adaptive_checkpoint,
   "Enable/Disable flushing along modified age. 0:disable 1:enable",
   NULL, NULL, 0, 0, 1, 0);
 
+static MYSQL_SYSVAR_ULONG(enable_unsafe_group_commit, srv_enable_unsafe_group_commit,
+  PLUGIN_VAR_RQCMDARG,
+  "Enable/Disable unsafe group commit when support_xa=OFF and use with binlog or other XA storage engine.",
+  NULL, NULL, 0, 0, 1, 0);
+
 static MYSQL_SYSVAR_ULONG(read_io_threads, innobase_read_io_threads,
   PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
   "Number of background read I/O threads in InnoDB.",
@@ -9656,6 +9665,11 @@ static MYSQL_SYSVAR_ULONG(write_io_threads, innobase_write_io_threads,
   "Number of background write I/O threads in InnoDB.",
   NULL, NULL, 1, 1, 64, 0);
 
+static MYSQL_SYSVAR_ULONG(extra_rsegments, srv_extra_rsegments,
+  PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
+  "Number of extra user rollback segments when create new database.",
+  NULL, NULL, 0, 0, 127, 0);
+
 static struct st_mysql_sys_var* innobase_system_variables[]= {
   MYSQL_SYSVAR(additional_mem_pool_size),
   MYSQL_SYSVAR(autoextend_increment),
@@ -9712,8 +9726,10 @@ static struct st_mysql_sys_var* innobase_system_variables[]= {
   MYSQL_SYSVAR(flush_neighbor_pages),
   MYSQL_SYSVAR(read_ahead),
   MYSQL_SYSVAR(adaptive_checkpoint),
+  MYSQL_SYSVAR(enable_unsafe_group_commit),
   MYSQL_SYSVAR(read_io_threads),
   MYSQL_SYSVAR(write_io_threads),
+  MYSQL_SYSVAR(extra_rsegments),
   NULL
 };
 
@@ -9894,6 +9910,7 @@ mysql_declare_plugin(innobase)
   innobase_system_variables, /* system variables */
   NULL /* reserved */
 },
+i_s_innodb_rseg,
 i_s_innodb_buffer_pool_pages,
 i_s_innodb_buffer_pool_pages_index,
 i_s_innodb_buffer_pool_pages_blob,

handler/i_s.h

@@ -20,5 +20,6 @@ extern struct st_mysql_plugin	i_s_innodb_cmp_reset;
 extern struct st_mysql_plugin	i_s_innodb_cmpmem;
 extern struct st_mysql_plugin	i_s_innodb_cmpmem_reset;
 extern struct st_mysql_plugin	i_s_innodb_patches;
+extern struct st_mysql_plugin	i_s_innodb_rseg;
 
 #endif /* i_s_h */

handler/innodb_patch_info.h

@@ -31,5 +31,6 @@ struct innodb_enhancement {
 {"innodb_buffer_pool_pages","Information of buffer pool content","","http://www.percona.com/docs/wiki/percona-xtradb:innodb_buffer_pool_pages"},
 {"innodb_split_buf_pool_mutex","More fix of buffer_pool mutex","Spliting buf_pool_mutex and optimizing based on innodb_opt_lru_count","http://www.percona.com/docs/wiki/percona-xtradb:innodb_split_buf_pool_mutex"},
 {"innodb_expand_undo_slots","expandable maximum number of undo slots","from 1024 (default) to about 4000","http://www.percona.com/docs/wiki/percona-xtradb:innodb_expand_undo_slots"},
+{"innodb_extra_rseg","allow to create extra rollback segments","When create new db, the new parameter allows to create more rollback segments","http://www.percona.com/docs/wiki/percona-xtradb:innodb_extra_rseg"},
 {NULL, NULL, NULL, NULL}
 };

include/buf0buddy.ic

@@ -107,9 +107,21 @@ buf_buddy_free(
 {
 	//ut_ad(buf_pool_mutex_own());
 
+	if (!have_page_hash_mutex) {
+		mutex_enter(&LRU_list_mutex);
+		mutex_enter(&flush_list_mutex);
+		mutex_enter(&page_hash_mutex);
+	}
+
 	mutex_enter(&zip_free_mutex);
-	buf_buddy_free_low(buf, buf_buddy_get_slot(size), have_page_hash_mutex);
+	buf_buddy_free_low(buf, buf_buddy_get_slot(size), TRUE);
 	mutex_exit(&zip_free_mutex);
+
+	if (!have_page_hash_mutex) {
+		mutex_exit(&LRU_list_mutex);
+		mutex_exit(&flush_list_mutex);
+		mutex_exit(&page_hash_mutex);
+	}
 }
 
 #ifdef UNIV_MATERIALIZE

include/buf0buf.h

@@ -1098,11 +1098,11 @@ struct buf_block_struct{
 					a block is in the unzip_LRU list
 					if page.state == BUF_BLOCK_FILE_PAGE
 					and page.zip.data != NULL */
-#ifdef UNIV_DEBUG
+//#ifdef UNIV_DEBUG
 	ibool		in_unzip_LRU_list;/* TRUE if the page is in the
 					decompressed LRU list;
 					used in debugging */
-#endif /* UNIV_DEBUG */
+//#endif /* UNIV_DEBUG */
 	byte*		frame;		/* pointer to buffer frame which
 					is of size UNIV_PAGE_SIZE, and
 					aligned to an address divisible by

include/buf0buf.ic

@@ -222,7 +222,7 @@ buf_page_in_file(
 	case BUF_BLOCK_ZIP_FREE:
 		/* This is a free page in buf_pool->zip_free[].
 		Such pages should only be accessed by the buddy allocator. */
-		ut_error;
+		/* ut_error; */ /* optimistic */
 		break;
 	case BUF_BLOCK_ZIP_PAGE:
 	case BUF_BLOCK_ZIP_DIRTY:
@@ -280,7 +280,7 @@ buf_page_get_mutex(
 {
 	switch (buf_page_get_state(bpage)) {
 	case BUF_BLOCK_ZIP_FREE:
-		ut_error;
+		/* ut_error; */ /* optimistic */
 		return(NULL);
 	case BUF_BLOCK_ZIP_PAGE:
 	case BUF_BLOCK_ZIP_DIRTY:
@@ -414,11 +414,12 @@ buf_page_can_relocate(
 	const buf_page_t*	bpage)	/* control block being relocated */
 {
 	//ut_ad(buf_pool_mutex_own());
-	ut_ad(mutex_own(buf_page_get_mutex(bpage)));
-	ut_ad(buf_page_in_file(bpage));
-	ut_ad(bpage->in_LRU_list);
+	/* optimistic */
+	//ut_ad(mutex_own(buf_page_get_mutex(bpage)));
+	//ut_ad(buf_page_in_file(bpage));
+	//ut_ad(bpage->in_LRU_list);
 
-	return(buf_page_get_io_fix(bpage) == BUF_IO_NONE
+	return(bpage->in_LRU_list && bpage->io_fix == BUF_IO_NONE
 	       && bpage->buf_fix_count == 0);
 }
 
@@ -1045,12 +1046,15 @@ buf_page_release(
 	if (rw_latch == RW_X_LATCH && mtr->modifications) {
 		//buf_pool_mutex_enter();
 		mutex_enter(&flush_list_mutex);
+		mutex_enter(&block->mutex);
+		ut_ad(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE);
 		buf_flush_note_modification(block, mtr);
 		//buf_pool_mutex_exit();
 		mutex_exit(&flush_list_mutex);
 	}
-
+	else {
 	mutex_enter(&block->mutex);
+	}
 
 #ifdef UNIV_SYNC_DEBUG
 	rw_lock_s_unlock(&(block->debug_latch));

include/buf0lru.h

@@ -100,10 +100,11 @@ buf_LRU_free_block(
 	buf_page_t*	bpage,	/* in: block to be freed */
 	ibool		zip,	/* in: TRUE if should remove also the
 				compressed page of an uncompressed page */
-	ibool*		buf_pool_mutex_released);
+	ibool*		buf_pool_mutex_released,
 				/* in: pointer to a variable that will
 				be assigned TRUE if buf_pool_mutex
 				was temporarily released, or NULL */
+	ibool		have_LRU_mutex);
 /**********************************************************************
 Try to free a replaceable block. */
 UNIV_INTERN

include/srv0srv.h

@@ -157,9 +157,12 @@ extern long long	srv_ibuf_max_size;
 extern ulint	srv_ibuf_active_contract;
 extern ulint	srv_ibuf_accel_rate;
 extern ulint	srv_flush_neighbor_pages;
+extern ulint	srv_enable_unsafe_group_commit;
 extern ulint	srv_read_ahead;
 extern ulint	srv_adaptive_checkpoint;
 
+extern ulint	srv_extra_rsegments;
+
 /*-------------------------------------------*/
 
 extern ulint	srv_n_rows_inserted;

include/trx0sys.h

@@ -106,6 +106,13 @@ UNIV_INTERN
 void
 trx_sys_create(void);
 /*================*/
+/*********************************************************************
+Create extra rollback segments when create_new_db */
+UNIV_INTERN
+void
+trx_sys_create_extra_rseg(
+/*======================*/
+	ulint	num);	/* in: number of extra user rollback segments */
 /********************************************************************
 Looks for a free slot for a rollback segment in the trx system file copy. */
 UNIV_INTERN

include/univ.i

@@ -12,7 +12,7 @@ Created 1/20/1994 Heikki Tuuri
 #define INNODB_VERSION_MAJOR	1
 #define INNODB_VERSION_MINOR	0
 #define INNODB_VERSION_BUGFIX	2
-#define PERCONA_INNODB_VERSION	1
+#define PERCONA_INNODB_VERSION	3
 
 /* The following is the InnoDB version as shown in
 SELECT plugin_version FROM information_schema.plugins;

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 +64,17 @@
+@@ -64,6 +64,18 @@
  TRIGGERS
  USER_PRIVILEGES
  VIEWS
-+INNODB_CMP
-+INNODB_BUFFER_POOL_PAGES
-+INNODB_CMPMEM
 +INNODB_BUFFER_POOL_PAGES_INDEX
++INNODB_RSEG
++INNODB_LOCKS
++INNODB_BUFFER_POOL_PAGES
 +XTRADB_ENHANCEMENTS
 +INNODB_TRX
 +INNODB_BUFFER_POOL_PAGES_BLOB
 +INNODB_LOCK_WAITS
 +INNODB_CMP_RESET
-+INNODB_LOCKS
++INNODB_CMP
 +INNODB_CMPMEM_RESET
++INNODB_CMPMEM
  columns_priv
  db
  event
-@@ -795,6 +806,8 @@
+@@ -795,6 +807,8 @@
  TABLES	UPDATE_TIME	datetime
  TABLES	CHECK_TIME	datetime
  TRIGGERS	CREATED	datetime
@@ -28,7 +29,7 @@ diff mysql-test/r/information_schema.result.orig mysql-test/r/information_schema
  event	execute_at	datetime
  event	last_executed	datetime
  event	starts	datetime
-@@ -843,12 +856,13 @@
+@@ -843,12 +857,13 @@
  TABLE_CONSTRAINTS	TABLE_NAME	select
  TABLE_PRIVILEGES	TABLE_NAME	select
  VIEWS	TABLE_NAME	select
@@ -39,43 +40,45 @@ diff mysql-test/r/information_schema.result.orig mysql-test/r/information_schema
  SELECT table_schema, count(*) FROM information_schema.TABLES where table_name<>'ndb_binlog_index' AND table_name<>'ndb_apply_status' GROUP BY TABLE_SCHEMA;
  table_schema	count(*)
 -information_schema	28
-+information_schema	39
++information_schema	40
  mysql	22
  create table t1 (i int, j int);
  create trigger trg1 before insert on t1 for each row
-@@ -1263,6 +1277,17 @@
+@@ -1263,6 +1278,18 @@
  TRIGGERS	TRIGGER_SCHEMA
  USER_PRIVILEGES	GRANTEE
  VIEWS	TABLE_SCHEMA
-+INNODB_CMP	page_size
-+INNODB_BUFFER_POOL_PAGES	page_type
-+INNODB_CMPMEM	page_size
 +INNODB_BUFFER_POOL_PAGES_INDEX	schema_name
++INNODB_RSEG	rseg_id
++INNODB_LOCKS	lock_id
++INNODB_BUFFER_POOL_PAGES	page_type
 +XTRADB_ENHANCEMENTS	name
 +INNODB_TRX	trx_id
 +INNODB_BUFFER_POOL_PAGES_BLOB	space_id
 +INNODB_LOCK_WAITS	requesting_trx_id
 +INNODB_CMP_RESET	page_size
-+INNODB_LOCKS	lock_id
++INNODB_CMP	page_size
 +INNODB_CMPMEM_RESET	page_size
++INNODB_CMPMEM	page_size
  SELECT t.table_name, c1.column_name
  FROM information_schema.tables t
  INNER JOIN
-@@ -1306,14 +1331,25 @@
+@@ -1306,14 +1333,26 @@
  TRIGGERS	TRIGGER_SCHEMA
  USER_PRIVILEGES	GRANTEE
  VIEWS	TABLE_SCHEMA
-+INNODB_CMP	page_size
-+INNODB_BUFFER_POOL_PAGES	page_type
-+INNODB_CMPMEM	page_size
 +INNODB_BUFFER_POOL_PAGES_INDEX	schema_name
++INNODB_RSEG	rseg_id
++INNODB_LOCKS	lock_id
++INNODB_BUFFER_POOL_PAGES	page_type
 +XTRADB_ENHANCEMENTS	name
 +INNODB_TRX	trx_id
 +INNODB_BUFFER_POOL_PAGES_BLOB	space_id
 +INNODB_LOCK_WAITS	requesting_trx_id
 +INNODB_CMP_RESET	page_size
-+INNODB_LOCKS	lock_id
++INNODB_CMP	page_size
 +INNODB_CMPMEM_RESET	page_size
++INNODB_CMPMEM	page_size
  SELECT MAX(table_name) FROM information_schema.tables;
  MAX(table_name)
 -VIEWS
@@ -89,7 +92,7 @@ diff mysql-test/r/information_schema.result.orig mysql-test/r/information_schema
  DROP TABLE IF EXISTS bug23037;
  DROP FUNCTION IF EXISTS get_value;
  SELECT COLUMN_NAME, MD5(COLUMN_DEFAULT), LENGTH(COLUMN_DEFAULT) FROM INFORMATION_SCHEMA.COLUMNS WHERE TABLE_NAME='bug23037';
-@@ -1382,6 +1418,16 @@
+@@ -1382,6 +1421,17 @@
  FILES	information_schema.FILES	1
  GLOBAL_STATUS	information_schema.GLOBAL_STATUS	1
  GLOBAL_VARIABLES	information_schema.GLOBAL_VARIABLES	1
@@ -102,11 +105,12 @@ diff mysql-test/r/information_schema.result.orig mysql-test/r/information_schema
 +INNODB_CMP_RESET	information_schema.INNODB_CMP_RESET	1
 +INNODB_LOCKS	information_schema.INNODB_LOCKS	1
 +INNODB_LOCK_WAITS	information_schema.INNODB_LOCK_WAITS	1
++INNODB_RSEG	information_schema.INNODB_RSEG	1
 +INNODB_TRX	information_schema.INNODB_TRX	1
  KEY_COLUMN_USAGE	information_schema.KEY_COLUMN_USAGE	1
  PARTITIONS	information_schema.PARTITIONS	1
  PLUGINS	information_schema.PLUGINS	1
-@@ -1400,6 +1446,7 @@
+@@ -1400,6 +1450,7 @@
  TRIGGERS	information_schema.TRIGGERS	1
  USER_PRIVILEGES	information_schema.USER_PRIVILEGES	1
  VIEWS	information_schema.VIEWS	1
@@ -117,60 +121,63 @@ diff mysql-test/r/information_schema.result.orig mysql-test/r/information_schema
 diff mysql-test/r/information_schema_db.result.orig mysql-test/r/information_schema_db.result
 --- mysql-test/r/information_schema_db.result.orig	2008-08-04 09:27:49.000000000 +0300
 +++ mysql-test/r/information_schema_db.result	2008-10-07 12:26:31.000000000 +0300
-@@ -33,6 +33,17 @@
+@@ -33,6 +33,18 @@
  TRIGGERS
  USER_PRIVILEGES
  VIEWS
-+INNODB_CMP
-+INNODB_BUFFER_POOL_PAGES
-+INNODB_CMPMEM
 +INNODB_BUFFER_POOL_PAGES_INDEX
++INNODB_RSEG
++INNODB_LOCKS
++INNODB_BUFFER_POOL_PAGES
 +XTRADB_ENHANCEMENTS
 +INNODB_TRX
 +INNODB_BUFFER_POOL_PAGES_BLOB
 +INNODB_LOCK_WAITS
 +INNODB_CMP_RESET
-+INNODB_LOCKS
++INNODB_CMP
 +INNODB_CMPMEM_RESET
++INNODB_CMPMEM
  show tables from INFORMATION_SCHEMA like 'T%';
  Tables_in_information_schema (T%)
  TABLES
 diff mysql-test/r/mysqlshow.result.orig mysql-test/r/mysqlshow.result
 --- mysql-test/r/mysqlshow.result.orig	2008-08-04 09:27:51.000000000 +0300
 +++ mysql-test/r/mysqlshow.result	2008-10-07 12:35:39.000000000 +0300
-@@ -107,6 +107,17 @@
+@@ -107,6 +107,18 @@
  | TRIGGERS                              |
  | USER_PRIVILEGES                       |
  | VIEWS                                 |
-+| INNODB_CMP                            |
-+| INNODB_BUFFER_POOL_PAGES              |
-+| INNODB_CMPMEM                         |
 +| INNODB_BUFFER_POOL_PAGES_INDEX        |
++| INNODB_RSEG                           |
++| INNODB_LOCKS                          |
++| INNODB_BUFFER_POOL_PAGES              |
 +| XTRADB_ENHANCEMENTS                   |
 +| INNODB_TRX                            |
 +| INNODB_BUFFER_POOL_PAGES_BLOB         |
 +| INNODB_LOCK_WAITS                     |
 +| INNODB_CMP_RESET                      |
-+| INNODB_LOCKS                          |
++| INNODB_CMP                            |
 +| INNODB_CMPMEM_RESET                   |
++| INNODB_CMPMEM                         |
  +---------------------------------------+
  Database: INFORMATION_SCHEMA
  +---------------------------------------+
-@@ -140,6 +151,17 @@
+@@ -140,6 +152,18 @@
  | TRIGGERS                              |
  | USER_PRIVILEGES                       |
  | VIEWS                                 |
-+| INNODB_CMP                            |
-+| INNODB_BUFFER_POOL_PAGES              |
-+| INNODB_CMPMEM                         |
 +| INNODB_BUFFER_POOL_PAGES_INDEX        |
++| INNODB_RSEG                           |
++| INNODB_LOCKS                          |
++| INNODB_BUFFER_POOL_PAGES              |
 +| XTRADB_ENHANCEMENTS                   |
 +| INNODB_TRX                            |
 +| INNODB_BUFFER_POOL_PAGES_BLOB         |
 +| INNODB_LOCK_WAITS                     |
 +| INNODB_CMP_RESET                      |
-+| INNODB_LOCKS                          |
++| INNODB_CMP                            |
 +| INNODB_CMPMEM_RESET                   |
++| INNODB_CMPMEM                         |
  +---------------------------------------+
  Wildcard: inf_rmation_schema
  +--------------------+

scripts/install_innodb_plugins.sql

@@ -11,3 +11,4 @@ INSTALL PLUGIN XTRADB_ENHANCEMENTS SONAME 'ha_innodb.so';
 INSTALL PLUGIN INNODB_BUFFER_POOL_PAGES SONAME 'ha_innodb.so';
 INSTALL PLUGIN INNODB_BUFFER_POOL_PAGES_BLOB SONAME 'ha_innodb.so';
 INSTALL PLUGIN INNODB_BUFFER_POOL_PAGES_INDEX SONAME 'ha_innodb.so';
+INSTALL PLUGIN innodb_rseg SONAME 'ha_innodb.so';

srv/srv0srv.c

@@ -333,8 +333,11 @@ UNIV_INTERN ulint	srv_ibuf_accel_rate = 100;
 
 UNIV_INTERN ulint	srv_flush_neighbor_pages = 1; /* 0:disable 1:enable */
 
+UNIV_INTERN ulint	srv_enable_unsafe_group_commit = 0; /* 0:disable 1:enable */
 UNIV_INTERN ulint	srv_read_ahead = 3; /* 1: random  2: linear  3: Both */
 UNIV_INTERN ulint	srv_adaptive_checkpoint = 0; /* 0:disable 1:enable */
+
+UNIV_INTERN ulint	srv_extra_rsegments = 0; /* extra rseg for users */
 /*-------------------------------------------*/
 UNIV_INTERN ulong	srv_n_spin_wait_rounds	= 20;
 UNIV_INTERN ulong	srv_n_free_tickets_to_enter = 500;

srv/srv0start.c

@@ -1438,6 +1438,8 @@ innobase_start_or_create_for_mysql(void)
 		dict_create();
 		srv_startup_is_before_trx_rollback_phase = FALSE;
 
+		if (srv_extra_rsegments)
+			trx_sys_create_extra_rseg(srv_extra_rsegments);
 #ifdef UNIV_LOG_ARCHIVE
 	} else if (srv_archive_recovery) {
 		fprintf(stderr,

sync/sync0rw.c

@@ -226,6 +226,7 @@ rw_lock_validate(
 /*=============*/
 	rw_lock_t*	lock)
 {
+	ulint	tmp;
 	ut_a(lock);
 
 #ifndef HAVE_GCC_ATOMIC_BUILTINS
@@ -233,18 +234,25 @@ rw_lock_validate(
 #endif
 
 	ut_a(lock->magic_n == RW_LOCK_MAGIC_N);
+#ifndef HAVE_GCC_ATOMIC_BUILTINS
+	/* It is dynamic combination */
 	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_s_waiters(lock) == 0)
-	     || (rw_lock_get_s_waiters(lock) == 1));
-	ut_a((rw_lock_get_x_waiters(lock) == 0)
-	     || (rw_lock_get_x_waiters(lock) == 1));
-	ut_a((rw_lock_get_wx_waiters(lock) == 0)
-	     || (rw_lock_get_wx_waiters(lock) == 1));
+#endif
+	tmp = rw_lock_get_writer(lock);
+	ut_a((tmp == RW_LOCK_EX)
+	     || (tmp == RW_LOCK_WAIT_EX)
+	     || (tmp == RW_LOCK_NOT_LOCKED));
+	tmp = rw_lock_get_s_waiters(lock);
+	ut_a((tmp == 0) || (tmp == 1));
+	tmp = rw_lock_get_x_waiters(lock);
+	ut_a((tmp == 0) || (tmp == 1));
+	tmp = rw_lock_get_wx_waiters(lock);
+	ut_a((tmp == 0) || (tmp == 1));
+#ifndef HAVE_GCC_ATOMIC_BUILTINS
+	/* It is dynamic combination */
 	ut_a((lock->writer != RW_LOCK_EX) || (lock->writer_count > 0));
+#endif
 
 #ifndef HAVE_GCC_ATOMIC_BUILTINS
 	mutex_exit(rw_lock_get_mutex(lock));

trx/trx0sys.c

@@ -1049,6 +1049,31 @@ trx_sys_create(void)
 	trx_sys_init_at_db_start();
 }
 
+/*********************************************************************
+Create extra rollback segments when create_new_db */
+UNIV_INTERN
+void
+trx_sys_create_extra_rseg(
+/*======================*/
+	ulint	num)	/* in: number of extra user rollback segments */
+{
+	mtr_t	mtr;
+	ulint	slot_no;
+	ulint	i;
+
+	/* Craete extra rollback segments */
+	mtr_start(&mtr);
+	for (i = 1; i < num + 1; i++) {
+		if(!trx_rseg_create(TRX_SYS_SPACE, ULINT_MAX, &slot_no, &mtr)) {
+			fprintf(stderr,
+"InnoDB: Warning: Failed to create extra rollback segments.\n");
+			break;
+		}
+		ut_a(slot_no == i);
+	}
+	mtr_commit(&mtr);
+}
+
 /*********************************************************************
 Update the file format tag. */
 static