# name       : innodb_opt_lru_count.patch
# introduced : 11 or before
# maintainer : Yasufumi
#
#!!! notice !!!
# Any small change to this file in the main branch
# should be done or reviewed by the maintainer!
diff -ruN a/storage/innobase/buf/buf0buddy.c b/storage/innobase/buf/buf0buddy.c
--- a/storage/innobase/buf/buf0buddy.c	2010-11-03 07:01:13.000000000 +0900
+++ b/storage/innobase/buf/buf0buddy.c	2010-12-03 15:20:49.593024343 +0900
@@ -137,7 +137,7 @@
 			      ut_ad(buf_page_get_state(ut_list_node_313)
 				    == BUF_BLOCK_ZIP_FREE)));
 #endif /* !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);
diff -ruN a/storage/innobase/buf/buf0buf.c b/storage/innobase/buf/buf0buf.c
--- a/storage/innobase/buf/buf0buf.c	2010-12-03 15:18:48.866986963 +0900
+++ b/storage/innobase/buf/buf0buf.c	2010-12-03 15:20:49.595987311 +0900
@@ -881,9 +881,9 @@
 	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;
 #endif /* UNIV_DEBUG */
+	block->page.in_LRU_list = FALSE;
 #if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
 	block->n_pointers = 0;
 #endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
@@ -1494,7 +1494,7 @@
 
 	memcpy(dpage, bpage, sizeof *dpage);
 
-	ut_d(bpage->in_LRU_list = FALSE);
+	bpage->in_LRU_list = FALSE;
 	ut_d(bpage->in_page_hash = FALSE);
 
 	/* relocate buf_pool->LRU */
@@ -3730,8 +3730,8 @@
 		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);
 
@@ -3894,7 +3894,7 @@
 	ibuf_merge_or_delete_for_page(NULL, space, offset, zip_size, TRUE);
 
 	/* Flush pages from the end of the LRU list if necessary */
-	buf_flush_free_margin(buf_pool);
+	buf_flush_free_margin(buf_pool, FALSE);
 
 	frame = block->frame;
 
diff -ruN a/storage/innobase/buf/buf0flu.c b/storage/innobase/buf/buf0flu.c
--- a/storage/innobase/buf/buf0flu.c	2010-12-03 15:18:48.868953442 +0900
+++ b/storage/innobase/buf/buf0flu.c	2010-12-03 15:20:49.599986956 +0900
@@ -431,19 +431,21 @@
 				buf_page_in_file(bpage) and in the LRU list */
 {
 #ifdef UNIV_DEBUG
-	buf_pool_t*	buf_pool = buf_pool_from_bpage(bpage);
-	ut_ad(buf_pool_mutex_own(buf_pool));
+	//buf_pool_t*	buf_pool = buf_pool_from_bpage(bpage);
+	//ut_ad(buf_pool_mutex_own(buf_pool));
 #endif
-	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
 		       && bpage->buf_fix_count == 0);
 	}
 
+	/* permited not to own LRU_mutex..  */
+/*
 	ut_print_timestamp(stderr);
 	fprintf(stderr,
 		"  InnoDB: Error: buffer block state %lu"
@@ -451,6 +453,7 @@
 		(ulong) buf_page_get_state(bpage));
 	ut_print_buf(stderr, bpage, sizeof(buf_page_t));
 	putc('\n', stderr);
+*/
 
 	return(FALSE);
 }
@@ -1983,8 +1986,14 @@
 	buf_page_t*	bpage;
 	ulint		n_replaceable;
 	ulint		distance	= 0;
+	ibool		have_LRU_mutex = FALSE;
 
-	buf_pool_mutex_enter(buf_pool);
+	if(UT_LIST_GET_LEN(buf_pool->unzip_LRU))
+		have_LRU_mutex = TRUE;
+retry:
+	//buf_pool_mutex_enter(buf_pool);
+	if (have_LRU_mutex)
+		buf_pool_mutex_enter(buf_pool);
 
 	n_replaceable = UT_LIST_GET_LEN(buf_pool->free);
 
@@ -1995,7 +2004,13 @@
 		   + BUF_FLUSH_EXTRA_MARGIN(buf_pool))
 	       && (distance < BUF_LRU_FREE_SEARCH_LEN(buf_pool))) {
 
-		mutex_t* block_mutex = buf_page_get_mutex(bpage);
+		mutex_t* block_mutex;
+		if (!bpage->in_LRU_list) {
+			/* reatart. but it is very optimistic */
+			bpage = UT_LIST_GET_LAST(buf_pool->LRU);
+			continue;
+		}
+		block_mutex = buf_page_get_mutex(bpage);
 
 		mutex_enter(block_mutex);
 
@@ -2010,11 +2025,18 @@
 		bpage = UT_LIST_GET_PREV(LRU, bpage);
 	}
 
-	buf_pool_mutex_exit(buf_pool);
+	//buf_pool_mutex_exit(buf_pool);
+	if (have_LRU_mutex)
+		buf_pool_mutex_exit(buf_pool);
 
 	if (n_replaceable >= BUF_FLUSH_FREE_BLOCK_MARGIN(buf_pool)) {
 
 		return(0);
+	} else if (!have_LRU_mutex) {
+		/* confirm it again with LRU_mutex for exactness */
+		have_LRU_mutex = TRUE;
+		distance = 0;
+		goto retry;
 	}
 
 	return(BUF_FLUSH_FREE_BLOCK_MARGIN(buf_pool)
@@ -2032,7 +2054,8 @@
 void
 buf_flush_free_margin(
 /*==================*/
-	buf_pool_t*	buf_pool)		/*!< in: Buffer pool instance */
+	buf_pool_t*	buf_pool,		/*!< in: Buffer pool instance */
+	ibool		wait)
 {
 	ulint	n_to_flush;
 
@@ -2043,7 +2066,7 @@
 
 		n_flushed = buf_flush_LRU(buf_pool, n_to_flush);
 
-		if (n_flushed == ULINT_UNDEFINED) {
+		if (wait && n_flushed == ULINT_UNDEFINED) {
 			/* There was an LRU type flush batch already running;
 			let us wait for it to end */
 
@@ -2056,8 +2079,9 @@
 Flushes pages from the end of all the LRU lists. */
 UNIV_INTERN
 void
-buf_flush_free_margins(void)
+buf_flush_free_margins(
 /*========================*/
+	ibool	wait)
 {
 	ulint	i;
 
@@ -2066,7 +2090,7 @@
 
 		buf_pool = buf_pool_from_array(i);
 
-		buf_flush_free_margin(buf_pool);
+		buf_flush_free_margin(buf_pool, wait);
 	}
 }
 
diff -ruN a/storage/innobase/buf/buf0lru.c b/storage/innobase/buf/buf0lru.c
--- a/storage/innobase/buf/buf0lru.c	2010-11-03 07:01:13.000000000 +0900
+++ b/storage/innobase/buf/buf0lru.c	2010-12-03 15:20:49.602952786 +0900
@@ -992,7 +992,7 @@
 
 	/* No free block was found: try to flush the LRU list */
 
-	buf_flush_free_margin(buf_pool);
+	buf_flush_free_margin(buf_pool, TRUE);
 	++srv_buf_pool_wait_free;
 
 	os_aio_simulated_wake_handler_threads();
@@ -1189,7 +1189,7 @@
 
 	/* Remove the block from the LRU list */
 	UT_LIST_REMOVE(LRU, buf_pool->LRU, bpage);
-	ut_d(bpage->in_LRU_list = FALSE);
+	bpage->in_LRU_list = FALSE;
 
 	buf_unzip_LRU_remove_block_if_needed(bpage);
 
@@ -1268,7 +1268,7 @@
 
 	ut_ad(!bpage->in_LRU_list);
 	UT_LIST_ADD_LAST(LRU, buf_pool->LRU, bpage);
-	ut_d(bpage->in_LRU_list = TRUE);
+	bpage->in_LRU_list = TRUE;
 
 	if (UT_LIST_GET_LEN(buf_pool->LRU) > BUF_LRU_OLD_MIN_LEN) {
 
@@ -1338,7 +1338,7 @@
 		buf_pool->LRU_old_len++;
 	}
 
-	ut_d(bpage->in_LRU_list = TRUE);
+	bpage->in_LRU_list = TRUE;
 
 	if (UT_LIST_GET_LEN(buf_pool->LRU) > BUF_LRU_OLD_MIN_LEN) {
 
@@ -1589,7 +1589,7 @@
 				buf_page_set_old(b, buf_page_is_old(b));
 #endif /* UNIV_LRU_DEBUG */
 			} else {
-				ut_d(b->in_LRU_list = FALSE);
+				b->in_LRU_list = FALSE;
 				buf_LRU_add_block_low(b, buf_page_is_old(b));
 			}
 
diff -ruN a/storage/innobase/buf/buf0rea.c b/storage/innobase/buf/buf0rea.c
--- a/storage/innobase/buf/buf0rea.c	2010-12-03 15:18:48.870953384 +0900
+++ b/storage/innobase/buf/buf0rea.c	2010-12-03 15:20:49.604956032 +0900
@@ -200,7 +200,7 @@
 	}
 
 	/* Flush pages from the end of the LRU list if necessary */
-	buf_flush_free_margin(buf_pool);
+	buf_flush_free_margin(buf_pool, TRUE);
 
 	/* Increment number of I/O operations used for LRU policy. */
 	buf_LRU_stat_inc_io();
@@ -476,7 +476,7 @@
 	os_aio_simulated_wake_handler_threads();
 
 	/* Flush pages from the end of the LRU list if necessary */
-	buf_flush_free_margin(buf_pool);
+	buf_flush_free_margin(buf_pool, TRUE);
 
 #ifdef UNIV_DEBUG
 	if (buf_debug_prints && (count > 0)) {
@@ -565,7 +565,7 @@
 	os_aio_simulated_wake_handler_threads();
 
 	/* Flush pages from the end of all the LRU lists if necessary */
-	buf_flush_free_margins();
+	buf_flush_free_margins(FALSE);
 
 #ifdef UNIV_DEBUG
 	if (buf_debug_prints) {
@@ -659,7 +659,7 @@
 	os_aio_simulated_wake_handler_threads();
 
 	/* Flush pages from the end of all the LRU lists if necessary */
-	buf_flush_free_margins();
+	buf_flush_free_margins(FALSE);
 
 #ifdef UNIV_DEBUG
 	if (buf_debug_prints) {
diff -ruN a/storage/innobase/include/buf0buf.h b/storage/innobase/include/buf0buf.h
--- a/storage/innobase/include/buf0buf.h	2010-11-03 07:01:13.000000000 +0900
+++ b/storage/innobase/include/buf0buf.h	2010-12-03 15:20:49.608986590 +0900
@@ -1385,11 +1385,11 @@
 
 	UT_LIST_NODE_T(buf_page_t) LRU;
 					/*!< node of the LRU list */
-#ifdef UNIV_DEBUG
+//#ifdef UNIV_DEBUG
 	ibool		in_LRU_list;	/*!< TRUE if the page is in
 					the LRU list; used in
 					debugging */
-#endif /* UNIV_DEBUG */
+//#endif /* UNIV_DEBUG */
 	unsigned	old:1;		/*!< TRUE if the block is in the old
 					blocks in buf_pool->LRU_old */
 	unsigned	freed_page_clock:31;/*!< the value of
diff -ruN a/storage/innobase/include/buf0flu.h b/storage/innobase/include/buf0flu.h
--- a/storage/innobase/include/buf0flu.h	2010-11-03 07:01:13.000000000 +0900
+++ b/storage/innobase/include/buf0flu.h	2010-12-03 15:20:49.609953185 +0900
@@ -65,13 +65,15 @@
 void
 buf_flush_free_margin(
 /*==================*/
-	 buf_pool_t*	buf_pool);
+	 buf_pool_t*	buf_pool,
+	ibool		wait);
 /*********************************************************************//**
 Flushes pages from the end of all the LRU lists. */
 UNIV_INTERN
 void
-buf_flush_free_margins(void);
+buf_flush_free_margins(
 /*=========================*/
+	ibool		wait);
 #endif /* !UNIV_HOTBACKUP */
 /********************************************************************//**
 Initializes a page for writing to the tablespace. */