]> git.pld-linux.org Git - packages/mysql.git/blame - innodb_split_buf_pool_mutex.patch
projects / packages / mysql.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
- rel 3
[packages/mysql.git] / innodb_split_buf_pool_mutex.patch
CommitLineData
b4e1fa2c
AM		 1# name : innodb_split_buf_pool_mutex.patch
2# introduced : 11 or before
3# maintainer : Yasufumi
4#
5#!!! notice !!!
6# Any small change to this file in the main branch
7# should be done or reviewed by the maintainer!
db82db79
AM		 8--- a/storage/innobase/btr/btr0cur.c
9+++ b/storage/innobase/btr/btr0cur.c
1bfc1981 10@@ -4070,7 +4070,8 @@
b4e1fa2c
AM		 11
12 mtr_commit(mtr);
13
14- buf_pool_mutex_enter(buf_pool);
15+ //buf_pool_mutex_enter(buf_pool);
16+ mutex_enter(&buf_pool->LRU_list_mutex);
17 mutex_enter(&block->mutex);
18
19 /* Only free the block if it is still allocated to
1bfc1981 20@@ -4081,16 +4082,21 @@
b4e1fa2c
AM		 21 && buf_block_get_space(block) == space
22 && buf_block_get_page_no(block) == page_no) {
23
db82db79 24- if (!buf_LRU_free_block(&block->page, all)
b4e1fa2c 25- && all && block->page.zip.data) {
db82db79 26+ if (!buf_LRU_free_block(&block->page, all, TRUE)
b4e1fa2c
AM		 27+ && all && block->page.zip.data
28+ /* Now, buf_LRU_free_block() may release mutex temporarily */
29+ && buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE
30+ && buf_block_get_space(block) == space
31+ && buf_block_get_page_no(block) == page_no) {
32 /* Attempt to deallocate the uncompressed page
33 if the whole block cannot be deallocted. */
34
df1b5770
AM		 35- buf_LRU_free_block(&block->page, FALSE);
36+ buf_LRU_free_block(&block->page, FALSE, TRUE);
b4e1fa2c
AM		 37 }
38 }
39
40- buf_pool_mutex_exit(buf_pool);
41+ //buf_pool_mutex_exit(buf_pool);
42+ mutex_exit(&buf_pool->LRU_list_mutex);
43 mutex_exit(&block->mutex);
44 }
45
db82db79
AM		 46--- a/storage/innobase/btr/btr0sea.c
47+++ b/storage/innobase/btr/btr0sea.c
13ceb006 48@@ -1972,7 +1972,7 @@
b4e1fa2c
AM		 49 rec_offs_init(offsets_);
50
51 rw_lock_x_lock(&btr_search_latch);
52- buf_pool_mutex_enter_all();
53+ buf_pool_page_hash_x_lock_all();
54
55 cell_count = hash_get_n_cells(btr_search_sys->hash_index);
56
13ceb006 57@@ -1980,11 +1980,11 @@
b4e1fa2c
AM		 58 /* We release btr_search_latch every once in a while to
59 give other queries a chance to run. */
60 if ((i != 0) && ((i % chunk_size) == 0)) {
61- buf_pool_mutex_exit_all();
62+ buf_pool_page_hash_x_unlock_all();
63 rw_lock_x_unlock(&btr_search_latch);
64 os_thread_yield();
65 rw_lock_x_lock(&btr_search_latch);
66- buf_pool_mutex_enter_all();
67+ buf_pool_page_hash_x_lock_all();
68 }
69
70 node = hash_get_nth_cell(btr_search_sys->hash_index, i)->node;
13ceb006 71@@ -2093,11 +2093,11 @@
b4e1fa2c
AM		 72 /* We release btr_search_latch every once in a while to
73 give other queries a chance to run. */
74 if (i != 0) {
75- buf_pool_mutex_exit_all();
76+ buf_pool_page_hash_x_unlock_all();
77 rw_lock_x_unlock(&btr_search_latch);
78 os_thread_yield();
79 rw_lock_x_lock(&btr_search_latch);
80- buf_pool_mutex_enter_all();
81+ buf_pool_page_hash_x_lock_all();
82 }
83
84 if (!ha_validate(btr_search_sys->hash_index, i, end_index)) {
13ceb006 85@@ -2105,7 +2105,7 @@
b4e1fa2c
AM		 86 }
87 }
88
89- buf_pool_mutex_exit_all();
90+ buf_pool_page_hash_x_unlock_all();
91 rw_lock_x_unlock(&btr_search_latch);
92 if (UNIV_LIKELY_NULL(heap)) {
93 mem_heap_free(heap);
db82db79
AM		 94--- a/storage/innobase/buf/buf0buddy.c
95+++ b/storage/innobase/buf/buf0buddy.c
96@@ -58,7 +58,7 @@
97
98 /** Validate a given zip_free list. */
99 #define BUF_BUDDY_LIST_VALIDATE(b, i) \
100- UT_LIST_VALIDATE(list, buf_page_t, \
101+ UT_LIST_VALIDATE(zip_list, buf_page_t, \
102 b->zip_free[i], \
103 ut_ad(buf_page_get_state( \
104 ut_list_node_313) \
105@@ -75,10 +75,11 @@
106 ulint i) /*!< in: index of
107 buf_pool->zip_free[] */
108 {
b4e1fa2c
AM		 109- ut_ad(buf_pool_mutex_own(buf_pool));
110+ //ut_ad(buf_pool_mutex_own(buf_pool));
111+ ut_ad(mutex_own(&buf_pool->zip_free_mutex));
112 ut_ad(buf_page_get_state(bpage) == BUF_BLOCK_ZIP_FREE);
113 ut_ad(buf_pool->zip_free[i].start != bpage);
114- UT_LIST_ADD_FIRST(list, buf_pool->zip_free[i], bpage);
115+ UT_LIST_ADD_FIRST(zip_list, buf_pool->zip_free[i], bpage);
db82db79 116 }
b4e1fa2c 117
db82db79
AM		 118 /**********************************************************************//**
119@@ -93,16 +94,17 @@
b4e1fa2c
AM		 120 buf_pool->zip_free[] */
121 {
db82db79 122 #ifdef UNIV_DEBUG
b4e1fa2c
AM		 123- buf_page_t* prev = UT_LIST_GET_PREV(list, bpage);
124- buf_page_t* next = UT_LIST_GET_NEXT(list, bpage);
125+ buf_page_t* prev = UT_LIST_GET_PREV(zip_list, bpage);
126+ buf_page_t* next = UT_LIST_GET_NEXT(zip_list, bpage);
127
db82db79 128 ut_ad(!prev || buf_page_get_state(prev) == BUF_BLOCK_ZIP_FREE);
b4e1fa2c 129 ut_ad(!next || buf_page_get_state(next) == BUF_BLOCK_ZIP_FREE);
db82db79 130 #endif /* UNIV_DEBUG */
b4e1fa2c
AM		 131
132- ut_ad(buf_pool_mutex_own(buf_pool));
133+ //ut_ad(buf_pool_mutex_own(buf_pool));
134+ ut_ad(mutex_own(&buf_pool->zip_free_mutex));
135 ut_ad(buf_page_get_state(bpage) == BUF_BLOCK_ZIP_FREE);
136- UT_LIST_REMOVE(list, buf_pool->zip_free[i], bpage);
137+ UT_LIST_REMOVE(zip_list, buf_pool->zip_free[i], bpage);
db82db79 138 }
b4e1fa2c 139
db82db79
AM		 140 /**********************************************************************//**
141@@ -117,7 +119,8 @@
b4e1fa2c
AM		 142 {
143 buf_page_t* bpage;
144
145- ut_ad(buf_pool_mutex_own(buf_pool));
146+ //ut_ad(buf_pool_mutex_own(buf_pool));
147+ ut_ad(mutex_own(&buf_pool->zip_free_mutex));
148 ut_a(i < BUF_BUDDY_SIZES);
db82db79 149 ut_a(i >= buf_buddy_get_slot(PAGE_ZIP_MIN_SIZE));
b4e1fa2c 150
db82db79 151@@ -159,16 +162,19 @@
b4e1fa2c
AM		 152 buf_buddy_block_free(
153 /*=================*/
154 buf_pool_t* buf_pool, /*!< in: buffer pool instance */
155- void* buf) /*!< in: buffer frame to deallocate */
156+ void* buf, /*!< in: buffer frame to deallocate */
157+ ibool have_page_hash_mutex)
158 {
159 const ulint fold = BUF_POOL_ZIP_FOLD_PTR(buf);
160 buf_page_t* bpage;
161 buf_block_t* block;
162
163- ut_ad(buf_pool_mutex_own(buf_pool));
164+ //ut_ad(buf_pool_mutex_own(buf_pool));
165 ut_ad(!mutex_own(&buf_pool->zip_mutex));
166 ut_a(!ut_align_offset(buf, UNIV_PAGE_SIZE));
167
168+ mutex_enter(&buf_pool->zip_hash_mutex);
169+
170 HASH_SEARCH(hash, buf_pool->zip_hash, fold, buf_page_t*, bpage,
171 ut_ad(buf_page_get_state(bpage) == BUF_BLOCK_MEMORY
172 && bpage->in_zip_hash && !bpage->in_page_hash),
db82db79 173@@ -180,12 +186,14 @@
b4e1fa2c
AM		 174 ut_d(bpage->in_zip_hash = FALSE);
175 HASH_DELETE(buf_page_t, hash, buf_pool->zip_hash, fold, bpage);
176
177+ mutex_exit(&buf_pool->zip_hash_mutex);
178+
179 ut_d(memset(buf, 0, UNIV_PAGE_SIZE));
180 UNIV_MEM_INVALID(buf, UNIV_PAGE_SIZE);
181
182 block = (buf_block_t*) bpage;
183 mutex_enter(&block->mutex);
184- buf_LRU_block_free_non_file_page(block);
185+ buf_LRU_block_free_non_file_page(block, have_page_hash_mutex);
186 mutex_exit(&block->mutex);
187
188 ut_ad(buf_pool->buddy_n_frames > 0);
db82db79 189@@ -202,7 +210,7 @@
b4e1fa2c
AM		 190 {
191 buf_pool_t* buf_pool = buf_pool_from_block(block);
192 const ulint fold = BUF_POOL_ZIP_FOLD(block);
193- ut_ad(buf_pool_mutex_own(buf_pool));
194+ //ut_ad(buf_pool_mutex_own(buf_pool));
195 ut_ad(!mutex_own(&buf_pool->zip_mutex));
196 ut_ad(buf_block_get_state(block) == BUF_BLOCK_READY_FOR_USE);
197
db82db79 198@@ -214,7 +222,10 @@
b4e1fa2c
AM		 199 ut_ad(!block->page.in_page_hash);
200 ut_ad(!block->page.in_zip_hash);
201 ut_d(block->page.in_zip_hash = TRUE);
202+
203+ mutex_enter(&buf_pool->zip_hash_mutex);
204 HASH_INSERT(buf_page_t, hash, buf_pool->zip_hash, fold, &block->page);
205+ mutex_exit(&buf_pool->zip_hash_mutex);
206
207 ut_d(buf_pool->buddy_n_frames++);
208 }
db82db79
AM		 209@@ -268,26 +279,30 @@
210 buf_pool_t* buf_pool, /*!< in/out: buffer pool instance */
b4e1fa2c
AM		 211 ulint i, /*!< in: index of buf_pool->zip_free[],
212 or BUF_BUDDY_SIZES */
213- ibool* lru) /*!< in: pointer to a variable that
214+ ibool* lru, /*!< in: pointer to a variable that
215 will be assigned TRUE if storage was
216 allocated from the LRU list and
217 buf_pool->mutex was temporarily
db82db79 218 released */
b4e1fa2c
AM		 219+ ibool have_page_hash_mutex)
220 {
221 buf_block_t* block;
222
db82db79 223 ut_ad(lru);
b4e1fa2c
AM		 224- ut_ad(buf_pool_mutex_own(buf_pool));
225+ //ut_ad(buf_pool_mutex_own(buf_pool));
226+ ut_ad(mutex_own(&buf_pool->LRU_list_mutex));
227 ut_ad(!mutex_own(&buf_pool->zip_mutex));
db82db79 228 ut_ad(i >= buf_buddy_get_slot(PAGE_ZIP_MIN_SIZE));
b4e1fa2c
AM		 229
230 if (i < BUF_BUDDY_SIZES) {
231 /* Try to allocate from the buddy system. */
232+ mutex_enter(&buf_pool->zip_free_mutex);
233 block = buf_buddy_alloc_zip(buf_pool, i);
234
235 if (block) {
236 goto func_exit;
237 }
238+ mutex_exit(&buf_pool->zip_free_mutex);
239 }
240
241 /* Try allocating from the buf_pool->free list. */
db82db79 242@@ -299,19 +314,30 @@
b4e1fa2c
AM		 243 }
244
245 /* Try replacing an uncompressed page in the buffer pool. */
246- buf_pool_mutex_exit(buf_pool);
247+ //buf_pool_mutex_exit(buf_pool);
248+ mutex_exit(&buf_pool->LRU_list_mutex);
249+ if (have_page_hash_mutex) {
250+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
251+ }
df1b5770 252 block = buf_LRU_get_free_block(buf_pool);
b4e1fa2c
AM		 253 *lru = TRUE;
254- buf_pool_mutex_enter(buf_pool);
255+ //buf_pool_mutex_enter(buf_pool);
256+ mutex_enter(&buf_pool->LRU_list_mutex);
257+ if (have_page_hash_mutex) {
258+ rw_lock_x_lock(&buf_pool->page_hash_latch);
259+ }
260
261 alloc_big:
262 buf_buddy_block_register(block);
263
264+ mutex_enter(&buf_pool->zip_free_mutex);
265 block = buf_buddy_alloc_from(
266 buf_pool, block->frame, i, BUF_BUDDY_SIZES);
267
268 func_exit:
269 buf_pool->buddy_stat[i].used++;
270+ mutex_exit(&buf_pool->zip_free_mutex);
271+
272 return(block);
273 }
274
db82db79 275@@ -325,8 +351,9 @@
b4e1fa2c
AM		 276 buf_pool_t* buf_pool, /*!< in: buffer pool instance */
277 void* src, /*!< in: block to relocate */
278 void* dst, /*!< in: free block to relocate to */
279- ulint i) /*!< in: index of
280+ ulint i, /*!< in: index of
281 buf_pool->zip_free[] */
282+ ibool have_page_hash_mutex)
283 {
284 buf_page_t* bpage;
285 const ulint size = BUF_BUDDY_LOW << i;
734d6226 286@@ -334,13 +361,20 @@
db82db79
AM		 287 ulint space;
288 ulint page_no;
b4e1fa2c
AM		 289
290- ut_ad(buf_pool_mutex_own(buf_pool));
291+ //ut_ad(buf_pool_mutex_own(buf_pool));
292+ ut_ad(mutex_own(&buf_pool->zip_free_mutex));
293 ut_ad(!mutex_own(&buf_pool->zip_mutex));
294 ut_ad(!ut_align_offset(src, size));
295 ut_ad(!ut_align_offset(dst, size));
db82db79
AM		 296 ut_ad(i >= buf_buddy_get_slot(PAGE_ZIP_MIN_SIZE));
297 UNIV_MEM_ASSERT_W(dst, size);
b4e1fa2c 298
db82db79
AM		 299+ if (!have_page_hash_mutex) {
300+ mutex_exit(&buf_pool->zip_free_mutex);
301+ mutex_enter(&buf_pool->LRU_list_mutex);
302+ rw_lock_x_lock(&buf_pool->page_hash_latch);
303+ }
b4e1fa2c 304+
db82db79
AM		 305 /* We assume that all memory from buf_buddy_alloc()
306 is used for compressed page frames. */
b4e1fa2c 307
734d6226 308@@ -374,6 +408,11 @@
db82db79
AM		 309 added to buf_pool->page_hash yet. Obviously,
310 it cannot be relocated. */
b4e1fa2c 311
db82db79
AM		 312+ if (!have_page_hash_mutex) {
313+ mutex_enter(&buf_pool->zip_free_mutex);
314+ mutex_exit(&buf_pool->LRU_list_mutex);
315+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
316+ }
317 return(FALSE);
318 }
b4e1fa2c 319
734d6226 320@@ -383,18 +422,27 @@
db82db79
AM		 321 For the sake of simplicity, give up. */
322 ut_ad(page_zip_get_size(&bpage->zip) < size);
b4e1fa2c 323
b4e1fa2c 324+ if (!have_page_hash_mutex) {
db82db79 325+ mutex_enter(&buf_pool->zip_free_mutex);
b4e1fa2c
AM		 326+ mutex_exit(&buf_pool->LRU_list_mutex);
327+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
328+ }
db82db79
AM		 329 return(FALSE);
330 }
331
332+ /* To keep latch order */
333+ if (have_page_hash_mutex)
b4e1fa2c
AM		 334+ mutex_exit(&buf_pool->zip_free_mutex);
335+
db82db79
AM		 336 /* The block must have been allocated, but it may
337 contain uninitialized data. */
338 UNIV_MEM_ASSERT_W(src, size);
339
340- mutex = buf_page_get_mutex(bpage);
341+ mutex = buf_page_get_mutex_enter(bpage);
342
343- mutex_enter(mutex);
344+ mutex_enter(&buf_pool->zip_free_mutex);
b4e1fa2c 345
db82db79
AM		 346- if (buf_page_can_relocate(bpage)) {
347+ if (mutex && buf_page_can_relocate(bpage)) {
348 /* Relocate the compressed page. */
734d6226 349 ullint usec = ut_time_us(NULL);
db82db79 350 ut_a(bpage->zip.data == src);
db82db79
AM		 351@@ -409,10 +457,22 @@
352 buddy_stat->relocated_usec
353 += ut_time_us(NULL) - usec;
b4e1fa2c
AM		 354 }
355+
b4e1fa2c
AM		 356+ if (!have_page_hash_mutex) {
357+ mutex_exit(&buf_pool->LRU_list_mutex);
358+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
359+ }
db82db79 360 return(TRUE);
b4e1fa2c
AM		 361 }
362
db82db79
AM		 363- mutex_exit(mutex);
364+ if (!have_page_hash_mutex) {
365+ mutex_exit(&buf_pool->LRU_list_mutex);
366+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
367+ }
368+
369+ if (mutex) {
370+ mutex_exit(mutex);
371+ }
b4e1fa2c 372 return(FALSE);
db82db79
AM		 373 }
374
375@@ -425,13 +485,15 @@
b4e1fa2c
AM		 376 buf_pool_t* buf_pool, /*!< in: buffer pool instance */
377 void* buf, /*!< in: block to be freed, must not be
378 pointed to by the buffer pool */
379- ulint i) /*!< in: index of buf_pool->zip_free[],
380+ ulint i, /*!< in: index of buf_pool->zip_free[],
381 or BUF_BUDDY_SIZES */
382+ ibool have_page_hash_mutex)
383 {
384 buf_page_t* bpage;
385 buf_page_t* buddy;
386
387- ut_ad(buf_pool_mutex_own(buf_pool));
388+ //ut_ad(buf_pool_mutex_own(buf_pool));
389+ ut_ad(mutex_own(&buf_pool->zip_free_mutex));
390 ut_ad(!mutex_own(&buf_pool->zip_mutex));
391 ut_ad(i <= BUF_BUDDY_SIZES);
db82db79
AM		 392 ut_ad(i >= buf_buddy_get_slot(PAGE_ZIP_MIN_SIZE));
393@@ -443,7 +505,9 @@
394 ((buf_page_t*) buf)->state = BUF_BLOCK_ZIP_FREE;
b4e1fa2c
AM		 395
396 if (i == BUF_BUDDY_SIZES) {
397- buf_buddy_block_free(buf_pool, buf);
398+ mutex_exit(&buf_pool->zip_free_mutex);
399+ buf_buddy_block_free(buf_pool, buf, have_page_hash_mutex);
400+ mutex_enter(&buf_pool->zip_free_mutex);
401 return;
402 }
403
db82db79
AM		 404@@ -491,7 +555,7 @@
405
b4e1fa2c 406 ut_a(bpage != buf);
db82db79
AM		 407 UNIV_MEM_ASSERT_W(bpage, BUF_BUDDY_LOW << i);
408- bpage = UT_LIST_GET_NEXT(list, bpage);
409+ bpage = UT_LIST_GET_NEXT(zip_list, bpage);
410 }
b4e1fa2c 411
b4e1fa2c 412 #ifndef UNIV_DEBUG_VALGRIND
db82db79
AM		 413@@ -501,7 +565,7 @@
414 ut_d(BUF_BUDDY_LIST_VALIDATE(buf_pool, i));
b4e1fa2c
AM		 415
416 /* The buddy is not free. Is there a free block of this size? */
417- bpage = UT_LIST_GET_FIRST(buf_pool->zip_free[i]);
418+ bpage = UT_LIST_GET_LAST(buf_pool->zip_free[i]);
419
420 if (bpage) {
db82db79
AM		 421
422@@ -510,7 +574,7 @@
b4e1fa2c
AM		 423 buf_buddy_remove_from_free(buf_pool, bpage, i);
424
425 /* Try to relocate the buddy of buf to the free block. */
426- if (buf_buddy_relocate(buf_pool, buddy, bpage, i)) {
427+ if (buf_buddy_relocate(buf_pool, buddy, bpage, i, have_page_hash_mutex)) {
428
db82db79
AM		 429 buddy->state = BUF_BLOCK_ZIP_FREE;
430 goto buddy_is_free;
431--- a/storage/innobase/buf/buf0buf.c
432+++ b/storage/innobase/buf/buf0buf.c
b4e1fa2c
AM		 433@@ -263,6 +263,7 @@
434 #ifdef UNIV_PFS_RWLOCK
435 /* Keys to register buffer block related rwlocks and mutexes with
436 performance schema */
437+UNIV_INTERN mysql_pfs_key_t buf_pool_page_hash_key;
438 UNIV_INTERN mysql_pfs_key_t buf_block_lock_key;
439 # ifdef UNIV_SYNC_DEBUG
440 UNIV_INTERN mysql_pfs_key_t buf_block_debug_latch_key;
441@@ -273,6 +274,10 @@
442 UNIV_INTERN mysql_pfs_key_t buffer_block_mutex_key;
443 UNIV_INTERN mysql_pfs_key_t buf_pool_mutex_key;
444 UNIV_INTERN mysql_pfs_key_t buf_pool_zip_mutex_key;
445+UNIV_INTERN mysql_pfs_key_t buf_pool_LRU_list_mutex_key;
446+UNIV_INTERN mysql_pfs_key_t buf_pool_free_list_mutex_key;
447+UNIV_INTERN mysql_pfs_key_t buf_pool_zip_free_mutex_key;
448+UNIV_INTERN mysql_pfs_key_t buf_pool_zip_hash_mutex_key;
449 UNIV_INTERN mysql_pfs_key_t flush_list_mutex_key;
450 #endif /* UNIV_PFS_MUTEX */
451
1bfc1981 452@@ -890,9 +895,13 @@
b4e1fa2c
AM		 453 block->page.in_zip_hash = FALSE;
454 block->page.in_flush_list = FALSE;
455 block->page.in_free_list = FALSE;
456- block->in_unzip_LRU_list = FALSE;
457 #endif /* UNIV_DEBUG */
adf0fb13
AM		 458+ block->page.flush_list.prev = NULL;
459+ block->page.flush_list.next = NULL;
460+ block->page.zip_list.prev = NULL;
461+ block->page.zip_list.next = NULL;
b4e1fa2c
AM		 462 block->page.in_LRU_list = FALSE;
463+ block->in_unzip_LRU_list = FALSE;
464 #if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
465 block->n_pointers = 0;
466 #endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
1bfc1981 467@@ -997,9 +1006,11 @@
b4e1fa2c
AM		 468 memset(block->frame, '\0', UNIV_PAGE_SIZE);
469 #endif
470 /* Add the block to the free list */
471- UT_LIST_ADD_LAST(list, buf_pool->free, (&block->page));
472+ mutex_enter(&buf_pool->free_list_mutex);
473+ UT_LIST_ADD_LAST(free, buf_pool->free, (&block->page));
474
475 ut_d(block->page.in_free_list = TRUE);
476+ mutex_exit(&buf_pool->free_list_mutex);
477 ut_ad(buf_pool_from_block(block) == buf_pool);
478
479 block++;
1bfc1981 480@@ -1054,7 +1065,8 @@
b4e1fa2c
AM		 481 buf_chunk_t* chunk = buf_pool->chunks;
482
483 ut_ad(buf_pool);
484- ut_ad(buf_pool_mutex_own(buf_pool));
485+ //ut_ad(buf_pool_mutex_own(buf_pool));
486+ ut_ad(mutex_own(&buf_pool->zip_free_mutex));
487 for (n = buf_pool->n_chunks; n--; chunk++) {
488
489 buf_block_t* block = buf_chunk_contains_zip(chunk, data);
1bfc1981 490@@ -1160,9 +1172,21 @@
b4e1fa2c
AM		 491 ------------------------------- */
492 mutex_create(buf_pool_mutex_key,
493 &buf_pool->mutex, SYNC_BUF_POOL);
494+ mutex_create(buf_pool_LRU_list_mutex_key,
495+ &buf_pool->LRU_list_mutex, SYNC_BUF_LRU_LIST);
496+ rw_lock_create(buf_pool_page_hash_key,
497+ &buf_pool->page_hash_latch, SYNC_BUF_PAGE_HASH);
498+ mutex_create(buf_pool_free_list_mutex_key,
499+ &buf_pool->free_list_mutex, SYNC_BUF_FREE_LIST);
500+ mutex_create(buf_pool_zip_free_mutex_key,
501+ &buf_pool->zip_free_mutex, SYNC_BUF_ZIP_FREE);
502+ mutex_create(buf_pool_zip_hash_mutex_key,
503+ &buf_pool->zip_hash_mutex, SYNC_BUF_ZIP_HASH);
504 mutex_create(buf_pool_zip_mutex_key,
505 &buf_pool->zip_mutex, SYNC_BUF_BLOCK);
506
507+ mutex_enter(&buf_pool->LRU_list_mutex);
508+ rw_lock_x_lock(&buf_pool->page_hash_latch);
509 buf_pool_mutex_enter(buf_pool);
510
511 if (buf_pool_size > 0) {
1bfc1981 512@@ -1175,6 +1199,8 @@
b4e1fa2c
AM		 513 mem_free(chunk);
514 mem_free(buf_pool);
515
516+ mutex_exit(&buf_pool->LRU_list_mutex);
517+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
518 buf_pool_mutex_exit(buf_pool);
519
520 return(DB_ERROR);
1bfc1981 521@@ -1205,6 +1231,8 @@
b4e1fa2c
AM		 522
523 /* All fields are initialized by mem_zalloc(). */
524
525+ mutex_exit(&buf_pool->LRU_list_mutex);
526+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
527 buf_pool_mutex_exit(buf_pool);
528
529 return(DB_SUCCESS);
1bfc1981 530@@ -1374,7 +1402,11 @@
b4e1fa2c
AM		 531 ulint fold;
532 buf_pool_t* buf_pool = buf_pool_from_bpage(bpage);
533
534- ut_ad(buf_pool_mutex_own(buf_pool));
535+ //ut_ad(buf_pool_mutex_own(buf_pool));
536+ ut_ad(mutex_own(&buf_pool->LRU_list_mutex));
537+#ifdef UNIV_SYNC_DEBUG
538+ ut_ad(rw_lock_own(&buf_pool->page_hash_latch, RW_LOCK_EX));
539+#endif
540 ut_ad(mutex_own(buf_page_get_mutex(bpage)));
541 ut_a(buf_page_get_io_fix(bpage) == BUF_IO_NONE);
542 ut_a(bpage->buf_fix_count == 0);
1bfc1981 543@@ -1485,21 +1517,32 @@
b4e1fa2c
AM		 544 buf_page_t* bpage;
545 ulint i;
546 buf_pool_t* buf_pool = buf_pool_get(space, offset);
547+ mutex_t* block_mutex;
548
549- ut_ad(buf_pool_mutex_own(buf_pool));
550+ //ut_ad(buf_pool_mutex_own(buf_pool));
551
552+ rw_lock_x_lock(&buf_pool->page_hash_latch);
553 bpage = buf_page_hash_get_low(buf_pool, space, offset, fold);
b4e1fa2c
AM		 554
555 if (UNIV_LIKELY_NULL(bpage)) {
3d3ecf24
AM		 556+
557+ block_mutex = buf_page_get_mutex_enter(bpage);
558+ ut_a(block_mutex);
559+
b4e1fa2c
AM		 560 if (!buf_pool_watch_is_sentinel(buf_pool, bpage)) {
561 /* The page was loaded meanwhile. */
562+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
563 return(bpage);
564 }
565 /* Add to an existing watch. */
566 bpage->buf_fix_count++;
567+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
568+ mutex_exit(block_mutex);
569 return(NULL);
570 }
571
572+ /* buf_pool->watch is protected by zip_mutex for now */
573+ mutex_enter(&buf_pool->zip_mutex);
574 for (i = 0; i < BUF_POOL_WATCH_SIZE; i++) {
575 bpage = &buf_pool->watch[i];
576
1bfc1981 577@@ -1523,10 +1566,12 @@
b4e1fa2c
AM		 578 bpage->space = space;
579 bpage->offset = offset;
580 bpage->buf_fix_count = 1;
581-
582+ bpage->buf_pool_index = buf_pool_index(buf_pool);
583 ut_d(bpage->in_page_hash = TRUE);
584 HASH_INSERT(buf_page_t, hash, buf_pool->page_hash,
585 fold, bpage);
586+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
587+ mutex_exit(&buf_pool->zip_mutex);
588 return(NULL);
589 case BUF_BLOCK_ZIP_PAGE:
590 ut_ad(bpage->in_page_hash);
1bfc1981 591@@ -1544,6 +1589,8 @@
b4e1fa2c
AM		 592 ut_error;
593
594 /* Fix compiler warning */
595+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
596+ mutex_exit(&buf_pool->zip_mutex);
597 return(NULL);
598 }
599
1bfc1981 600@@ -1561,7 +1608,11 @@
b4e1fa2c
AM		 601 space, offset) */
602 buf_page_t* watch) /*!< in/out: sentinel for watch */
603 {
604- ut_ad(buf_pool_mutex_own(buf_pool));
605+ //ut_ad(buf_pool_mutex_own(buf_pool));
606+#ifdef UNIV_SYNC_DEBUG
607+ ut_ad(rw_lock_own(&buf_pool->page_hash_latch, RW_LOCK_EX));
608+#endif
609+ ut_ad(mutex_own(&buf_pool->zip_mutex)); /* for now */
610
611 HASH_DELETE(buf_page_t, hash, buf_pool->page_hash, fold, watch);
612 ut_d(watch->in_page_hash = FALSE);
1bfc1981 613@@ -1583,28 +1634,31 @@
b4e1fa2c
AM		 614 buf_pool_t* buf_pool = buf_pool_get(space, offset);
615 ulint fold = buf_page_address_fold(space, offset);
616
617- buf_pool_mutex_enter(buf_pool);
618+ //buf_pool_mutex_enter(buf_pool);
619+ rw_lock_x_lock(&buf_pool->page_hash_latch);
620 bpage = buf_page_hash_get_low(buf_pool, space, offset, fold);
621 /* The page must exist because buf_pool_watch_set()
622 increments buf_fix_count. */
623 ut_a(bpage);
624
625 if (UNIV_UNLIKELY(!buf_pool_watch_is_sentinel(buf_pool, bpage))) {
626- mutex_t* mutex = buf_page_get_mutex(bpage);
627+ mutex_t* mutex = buf_page_get_mutex_enter(bpage);
628
629- mutex_enter(mutex);
630 ut_a(bpage->buf_fix_count > 0);
631 bpage->buf_fix_count--;
632 mutex_exit(mutex);
633 } else {
634+ mutex_enter(&buf_pool->zip_mutex);
635 ut_a(bpage->buf_fix_count > 0);
636
637 if (UNIV_LIKELY(!--bpage->buf_fix_count)) {
638 buf_pool_watch_remove(buf_pool, fold, bpage);
639 }
640+ mutex_exit(&buf_pool->zip_mutex);
641 }
642
643- buf_pool_mutex_exit(buf_pool);
644+ //buf_pool_mutex_exit(buf_pool);
645+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
646 }
647
648 /****************************************************************//**
1bfc1981 649@@ -1624,14 +1678,16 @@
b4e1fa2c
AM		 650 buf_pool_t* buf_pool = buf_pool_get(space, offset);
651 ulint fold = buf_page_address_fold(space, offset);
652
653- buf_pool_mutex_enter(buf_pool);
654+ //buf_pool_mutex_enter(buf_pool);
655+ rw_lock_s_lock(&buf_pool->page_hash_latch);
656
657 bpage = buf_page_hash_get_low(buf_pool, space, offset, fold);
658 /* The page must exist because buf_pool_watch_set()
659 increments buf_fix_count. */
660 ut_a(bpage);
661 ret = !buf_pool_watch_is_sentinel(buf_pool, bpage);
662- buf_pool_mutex_exit(buf_pool);
663+ //buf_pool_mutex_exit(buf_pool);
664+ rw_lock_s_unlock(&buf_pool->page_hash_latch);
665
666 return(ret);
667 }
1bfc1981 668@@ -1648,13 +1704,15 @@
b4e1fa2c
AM		 669 {
670 buf_pool_t* buf_pool = buf_pool_from_bpage(bpage);
671
672- buf_pool_mutex_enter(buf_pool);
673+ //buf_pool_mutex_enter(buf_pool);
674+ mutex_enter(&buf_pool->LRU_list_mutex);
675
676 ut_a(buf_page_in_file(bpage));
677
678 buf_LRU_make_block_young(bpage);
679
680- buf_pool_mutex_exit(buf_pool);
681+ //buf_pool_mutex_exit(buf_pool);
682+ mutex_exit(&buf_pool->LRU_list_mutex);
683 }
684
685 /********************************************************************//**
1bfc1981 686@@ -1678,14 +1736,20 @@
b4e1fa2c
AM		 687 ut_a(buf_page_in_file(bpage));
688
689 if (buf_page_peek_if_too_old(bpage)) {
690- buf_pool_mutex_enter(buf_pool);
691+ //buf_pool_mutex_enter(buf_pool);
692+ mutex_enter(&buf_pool->LRU_list_mutex);
693 buf_LRU_make_block_young(bpage);
694- buf_pool_mutex_exit(buf_pool);
695+ //buf_pool_mutex_exit(buf_pool);
696+ mutex_exit(&buf_pool->LRU_list_mutex);
697 } else if (!access_time) {
698 ulint time_ms = ut_time_ms();
699- buf_pool_mutex_enter(buf_pool);
700+ mutex_t* block_mutex = buf_page_get_mutex_enter(bpage);
701+ //buf_pool_mutex_enter(buf_pool);
702+ if (block_mutex) {
703 buf_page_set_accessed(bpage, time_ms);
704- buf_pool_mutex_exit(buf_pool);
705+ mutex_exit(block_mutex);
706+ }
707+ //buf_pool_mutex_exit(buf_pool);
708 }
709 }
710
1bfc1981 711@@ -1702,7 +1766,8 @@
b4e1fa2c
AM		 712 buf_block_t* block;
713 buf_pool_t* buf_pool = buf_pool_get(space, offset);
714
715- buf_pool_mutex_enter(buf_pool);
716+ //buf_pool_mutex_enter(buf_pool);
717+ rw_lock_s_lock(&buf_pool->page_hash_latch);
718
719 block = (buf_block_t*) buf_page_hash_get(buf_pool, space, offset);
720
1bfc1981 721@@ -1711,7 +1776,8 @@
b4e1fa2c
AM		 722 block->check_index_page_at_flush = FALSE;
723 }
724
725- buf_pool_mutex_exit(buf_pool);
726+ //buf_pool_mutex_exit(buf_pool);
727+ rw_lock_s_unlock(&buf_pool->page_hash_latch);
728 }
729
13ceb006 730 #if defined UNIV_DEBUG_FILE_ACCESSES || defined UNIV_DEBUG
1bfc1981 731@@ -1731,7 +1797,8 @@
b4e1fa2c
AM		 732 buf_page_t* bpage;
733 buf_pool_t* buf_pool = buf_pool_get(space, offset);
734
735- buf_pool_mutex_enter(buf_pool);
736+ //buf_pool_mutex_enter(buf_pool);
737+ rw_lock_s_lock(&buf_pool->page_hash_latch);
738
739 bpage = buf_page_hash_get(buf_pool, space, offset);
740
1bfc1981 741@@ -1742,7 +1809,8 @@
b4e1fa2c
AM		 742 bpage->file_page_was_freed = TRUE;
743 }
744
745- buf_pool_mutex_exit(buf_pool);
746+ //buf_pool_mutex_exit(buf_pool);
747+ rw_lock_s_unlock(&buf_pool->page_hash_latch);
748
749 return(bpage);
750 }
1bfc1981 751@@ -1763,7 +1831,8 @@
b4e1fa2c
AM		 752 buf_page_t* bpage;
753 buf_pool_t* buf_pool = buf_pool_get(space, offset);
754
755- buf_pool_mutex_enter(buf_pool);
756+ //buf_pool_mutex_enter(buf_pool);
757+ rw_lock_s_lock(&buf_pool->page_hash_latch);
758
759 bpage = buf_page_hash_get(buf_pool, space, offset);
760
1bfc1981 761@@ -1772,7 +1841,8 @@
b4e1fa2c
AM		 762 bpage->file_page_was_freed = FALSE;
763 }
764
765- buf_pool_mutex_exit(buf_pool);
766+ //buf_pool_mutex_exit(buf_pool);
767+ rw_lock_s_unlock(&buf_pool->page_hash_latch);
768
769 return(bpage);
770 }
1bfc1981 771@@ -1804,8 +1874,9 @@
b4e1fa2c
AM		 772 buf_pool->stat.n_page_gets++;
773
774 for (;;) {
775- buf_pool_mutex_enter(buf_pool);
776+ //buf_pool_mutex_enter(buf_pool);
777 lookup:
778+ rw_lock_s_lock(&buf_pool->page_hash_latch);
779 bpage = buf_page_hash_get(buf_pool, space, offset);
780 if (bpage) {
781 ut_ad(!buf_pool_watch_is_sentinel(buf_pool, bpage));
1bfc1981 782@@ -1814,7 +1885,8 @@
b4e1fa2c
AM		 783
784 /* Page not in buf_pool: needs to be read from file */
785
786- buf_pool_mutex_exit(buf_pool);
787+ //buf_pool_mutex_exit(buf_pool);
788+ rw_lock_s_unlock(&buf_pool->page_hash_latch);
789
790 buf_read_page(space, zip_size, offset);
791
1bfc1981 792@@ -1826,10 +1898,15 @@
b4e1fa2c
AM		 793 if (UNIV_UNLIKELY(!bpage->zip.data)) {
794 /* There is no compressed page. */
795 err_exit:
796- buf_pool_mutex_exit(buf_pool);
797+ //buf_pool_mutex_exit(buf_pool);
798+ rw_lock_s_unlock(&buf_pool->page_hash_latch);
799 return(NULL);
800 }
801
802+ block_mutex = buf_page_get_mutex_enter(bpage);
803+
804+ rw_lock_s_unlock(&buf_pool->page_hash_latch);
805+
806 ut_ad(!buf_pool_watch_is_sentinel(buf_pool, bpage));
807
808 switch (buf_page_get_state(bpage)) {
1bfc1981 809@@ -1838,24 +1915,43 @@
b4e1fa2c
AM		 810 case BUF_BLOCK_MEMORY:
811 case BUF_BLOCK_REMOVE_HASH:
812 case BUF_BLOCK_ZIP_FREE:
813+ if (block_mutex)
814+ mutex_exit(block_mutex);
815 break;
816 case BUF_BLOCK_ZIP_PAGE:
817 case BUF_BLOCK_ZIP_DIRTY:
818- block_mutex = &buf_pool->zip_mutex;
819- mutex_enter(block_mutex);
820+ ut_a(block_mutex == &buf_pool->zip_mutex);
821 bpage->buf_fix_count++;
822 goto got_block;
823 case BUF_BLOCK_FILE_PAGE:
824- block_mutex = &((buf_block_t*) bpage)->mutex;
b4e1fa2c 825+ ut_a(block_mutex == &((buf_block_t*) bpage)->mutex);
f7ab7acd
AM		 826+
827+ /* release mutex to obey to latch-order */
828+ mutex_exit(block_mutex);
829+
830+ /* get LRU_list_mutex for buf_LRU_free_block() */
831+ mutex_enter(&buf_pool->LRU_list_mutex);
832 mutex_enter(block_mutex);
b4e1fa2c 833
f7ab7acd 834- /* Discard the uncompressed page frame if possible. */
db82db79 835- if (buf_LRU_free_block(bpage, FALSE)) {
f7ab7acd
AM		 836+ if (UNIV_UNLIKELY(bpage->space != space
837+ || bpage->offset != offset
838+ || !bpage->in_LRU_list
839+ || !bpage->zip.data)) {
840+ /* someone should interrupt, retry */
841+ mutex_exit(&buf_pool->LRU_list_mutex);
842+ mutex_exit(block_mutex);
843+ goto lookup;
844+ }
b4e1fa2c 845
f7ab7acd 846+ /* Discard the uncompressed page frame if possible. */
db82db79 847+ if (buf_LRU_free_block(bpage, FALSE, TRUE)) {
f7ab7acd 848+ mutex_exit(&buf_pool->LRU_list_mutex);
b4e1fa2c 849 mutex_exit(block_mutex);
df1b5770 850 goto lookup;
f7ab7acd
AM		 851 }
852
853+ mutex_exit(&buf_pool->LRU_list_mutex);
854+
855 buf_block_buf_fix_inc((buf_block_t*) bpage,
856 __FILE__, __LINE__);
857 goto got_block;
1bfc1981 858@@ -1868,7 +1964,7 @@
b4e1fa2c
AM		 859 must_read = buf_page_get_io_fix(bpage) == BUF_IO_READ;
860 access_time = buf_page_is_accessed(bpage);
861
862- buf_pool_mutex_exit(buf_pool);
863+ //buf_pool_mutex_exit(buf_pool);
864
865 mutex_exit(block_mutex);
866
1bfc1981 867@@ -2179,7 +2275,7 @@
b4e1fa2c
AM		 868 const buf_block_t* block) /*!< in: pointer to block,
869 not dereferenced */
870 {
871- ut_ad(buf_pool_mutex_own(buf_pool));
872+ //ut_ad(buf_pool_mutex_own(buf_pool));
873
874 if (UNIV_UNLIKELY((((ulint) block) % sizeof *block) != 0)) {
875 /* The pointer should be aligned. */
1bfc1981 876@@ -2215,6 +2311,7 @@
b4e1fa2c
AM		 877 ulint fix_type;
878 ibool must_read;
879 ulint retries = 0;
880+ mutex_t* block_mutex = NULL;
881 buf_pool_t* buf_pool = buf_pool_get(space, offset);
882
883 ut_ad(mtr);
1bfc1981 884@@ -2248,18 +2345,24 @@
b4e1fa2c
AM		 885 fold = buf_page_address_fold(space, offset);
886 loop:
887 block = guess;
888- buf_pool_mutex_enter(buf_pool);
889+ //buf_pool_mutex_enter(buf_pool);
890
891 if (block) {
892+ block_mutex = buf_page_get_mutex_enter((buf_page_t*)block);
893+
894 /* If the guess is a compressed page descriptor that
db82db79
AM		 895 has been allocated by buf_page_alloc_descriptor(),
896 it may have been freed by buf_relocate(). */
b4e1fa2c
AM		 897
898- if (!buf_block_is_uncompressed(buf_pool, block)
899+ if (!block_mutex) {
900+ block = guess = NULL;
901+ } else if (!buf_block_is_uncompressed(buf_pool, block)
902 || offset != block->page.offset
903 || space != block->page.space
904 || buf_block_get_state(block) != BUF_BLOCK_FILE_PAGE) {
905
906+ mutex_exit(block_mutex);
907+
908 block = guess = NULL;
909 } else {
910 ut_ad(!block->page.in_zip_hash);
1bfc1981 911@@ -2268,12 +2371,19 @@
b4e1fa2c
AM		 912 }
913
914 if (block == NULL) {
915+ rw_lock_s_lock(&buf_pool->page_hash_latch);
916 block = (buf_block_t*) buf_page_hash_get_low(
917 buf_pool, space, offset, fold);
918+ if (block) {
919+ block_mutex = buf_page_get_mutex_enter((buf_page_t*)block);
920+ ut_a(block_mutex);
921+ }
922+ rw_lock_s_unlock(&buf_pool->page_hash_latch);
923 }
924
925 loop2:
926 if (block && buf_pool_watch_is_sentinel(buf_pool, &block->page)) {
927+ mutex_exit(block_mutex);
928 block = NULL;
929 }
930
1bfc1981 931@@ -2285,12 +2395,14 @@
b4e1fa2c
AM		 932 space, offset, fold);
933
934 if (UNIV_LIKELY_NULL(block)) {
935-
936+ block_mutex = buf_page_get_mutex((buf_page_t*)block);
937+ ut_a(block_mutex);
938+ ut_ad(mutex_own(block_mutex));
939 goto got_block;
940 }
941 }
942
943- buf_pool_mutex_exit(buf_pool);
944+ //buf_pool_mutex_exit(buf_pool);
945
946 if (mode == BUF_GET_IF_IN_POOL
adf0fb13 947 || mode == BUF_PEEK_IF_IN_POOL
1bfc1981 948@@ -2343,7 +2455,8 @@
b4e1fa2c
AM		 949 /* The page is being read to buffer pool,
950 but we cannot wait around for the read to
951 complete. */
952- buf_pool_mutex_exit(buf_pool);
953+ //buf_pool_mutex_exit(buf_pool);
954+ mutex_exit(block_mutex);
955
956 return(NULL);
957 }
1bfc1981 958@@ -2353,38 +2466,49 @@
b4e1fa2c
AM		 959 ibool success;
960
961 case BUF_BLOCK_FILE_PAGE:
962+ if (block_mutex == &buf_pool->zip_mutex) {
963+ /* it is wrong mutex... */
964+ mutex_exit(block_mutex);
965+ goto loop;
966+ }
967 break;
968
969 case BUF_BLOCK_ZIP_PAGE:
970 case BUF_BLOCK_ZIP_DIRTY:
971+ ut_ad(block_mutex == &buf_pool->zip_mutex);
972 bpage = &block->page;
973 /* Protect bpage->buf_fix_count. */
974- mutex_enter(&buf_pool->zip_mutex);
975+ //mutex_enter(&buf_pool->zip_mutex);
976
977 if (bpage->buf_fix_count
978 || buf_page_get_io_fix(bpage) != BUF_IO_NONE) {
979 /* This condition often occurs when the buffer
980 is not buffer-fixed, but I/O-fixed by
981 buf_page_init_for_read(). */
982- mutex_exit(&buf_pool->zip_mutex);
983+ //mutex_exit(&buf_pool->zip_mutex);
984 wait_until_unfixed:
985 /* The block is buffer-fixed or I/O-fixed.
986 Try again later. */
987- buf_pool_mutex_exit(buf_pool);
988+ //buf_pool_mutex_exit(buf_pool);
989+ mutex_exit(block_mutex);
990 os_thread_sleep(WAIT_FOR_READ);
adf0fb13 991
b4e1fa2c
AM		 992 goto loop;
993 }
994
995 /* Allocate an uncompressed page. */
996- buf_pool_mutex_exit(buf_pool);
997- mutex_exit(&buf_pool->zip_mutex);
998+ //buf_pool_mutex_exit(buf_pool);
999+ //mutex_exit(&buf_pool->zip_mutex);
1000+ mutex_exit(block_mutex);
1001
df1b5770 1002 block = buf_LRU_get_free_block(buf_pool);
b4e1fa2c
AM		 1003 ut_a(block);
1004+ block_mutex = &block->mutex;
1005
1006- buf_pool_mutex_enter(buf_pool);
1007- mutex_enter(&block->mutex);
1008+ //buf_pool_mutex_enter(buf_pool);
1009+ mutex_enter(&buf_pool->LRU_list_mutex);
1010+ rw_lock_x_lock(&buf_pool->page_hash_latch);
1011+ mutex_enter(block_mutex);
1012
1013 {
1014 buf_page_t* hash_bpage;
1bfc1981 1015@@ -2397,35 +2521,47 @@
b4e1fa2c
AM		 1016 while buf_pool->mutex was released.
1017 Free the block that was allocated. */
1018
1019- buf_LRU_block_free_non_file_page(block);
1020- mutex_exit(&block->mutex);
1021+ buf_LRU_block_free_non_file_page(block, TRUE);
1022+ mutex_exit(block_mutex);
1023
1024 block = (buf_block_t*) hash_bpage;
1025+ if (block) {
1026+ block_mutex = buf_page_get_mutex_enter((buf_page_t*)block);
1027+ ut_a(block_mutex);
1028+ }
1029+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
1030+ mutex_exit(&buf_pool->LRU_list_mutex);
1031 goto loop2;
1032 }
1033 }
1034
1035+ mutex_enter(&buf_pool->zip_mutex);
1036+
1037 if (UNIV_UNLIKELY
1038 (bpage->buf_fix_count
1039 || buf_page_get_io_fix(bpage) != BUF_IO_NONE)) {
1040
1041+ mutex_exit(&buf_pool->zip_mutex);
1042 /* The block was buffer-fixed or I/O-fixed
1043 while buf_pool->mutex was not held by this thread.
1044 Free the block that was allocated and try again.
1045 This should be extremely unlikely. */
1046
1047- buf_LRU_block_free_non_file_page(block);
1048- mutex_exit(&block->mutex);
1049+ buf_LRU_block_free_non_file_page(block, TRUE);
1050+ //mutex_exit(&block->mutex);
1051
1052+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
1053+ mutex_exit(&buf_pool->LRU_list_mutex);
1054 goto wait_until_unfixed;
1055 }
1056
1057 /* Move the compressed page from bpage to block,
1058 and uncompress it. */
1059
1060- mutex_enter(&buf_pool->zip_mutex);
1061-
1062 buf_relocate(bpage, &block->page);
1063+
1064+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
1065+
1066 buf_block_init_low(block);
1067 block->lock_hash_val = lock_rec_hash(space, offset);
1068
1bfc1981 1069@@ -2435,7 +2571,7 @@
b4e1fa2c
AM		 1070 if (buf_page_get_state(&block->page)
1071 == BUF_BLOCK_ZIP_PAGE) {
db82db79 1072 #if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
b4e1fa2c
AM		 1073- UT_LIST_REMOVE(list, buf_pool->zip_clean,
1074+ UT_LIST_REMOVE(zip_list, buf_pool->zip_clean,
1075 &block->page);
db82db79 1076 #endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
b4e1fa2c 1077 ut_ad(!block->page.in_flush_list);
1bfc1981 1078@@ -2453,18 +2589,23 @@
b4e1fa2c
AM		 1079 /* Insert at the front of unzip_LRU list */
1080 buf_unzip_LRU_add_block(block, FALSE);
1081
1082+ mutex_exit(&buf_pool->LRU_list_mutex);
1083+
1084 block->page.buf_fix_count = 1;
1085 buf_block_set_io_fix(block, BUF_IO_READ);
1086 rw_lock_x_lock_func(&block->lock, 0, file, line);
1087
1088 UNIV_MEM_INVALID(bpage, sizeof *bpage);
1089
1090- mutex_exit(&block->mutex);
1091+ mutex_exit(block_mutex);
1092 mutex_exit(&buf_pool->zip_mutex);
db82db79 1093- buf_pool->n_pend_unzip++;
b4e1fa2c 1094
db82db79
AM		 1095+ buf_pool_mutex_enter(buf_pool);
1096+ buf_pool->n_pend_unzip++;
1097 buf_pool_mutex_exit(buf_pool);
b4e1fa2c 1098
b4e1fa2c 1099+ //buf_pool_mutex_exit(buf_pool);
db82db79
AM		 1100+
1101 buf_page_free_descriptor(bpage);
b4e1fa2c
AM		 1102
1103 /* Decompress the page and apply buffered operations
1bfc1981 1104@@ -2478,12 +2619,15 @@
b4e1fa2c
AM		 1105 }
1106
1107 /* Unfix and unlatch the block. */
1108- buf_pool_mutex_enter(buf_pool);
1109- mutex_enter(&block->mutex);
1110+ //buf_pool_mutex_enter(buf_pool);
1111+ block_mutex = &block->mutex;
1112+ mutex_enter(block_mutex);
1113 block->page.buf_fix_count--;
1114 buf_block_set_io_fix(block, BUF_IO_NONE);
1115- mutex_exit(&block->mutex);
1116+
1117+ buf_pool_mutex_enter(buf_pool);
1118 buf_pool->n_pend_unzip--;
1119+ buf_pool_mutex_exit(buf_pool);
1120 rw_lock_x_unlock(&block->lock);
1121
1122 break;
1bfc1981 1123@@ -2499,7 +2643,7 @@
b4e1fa2c
AM		 1124
1125 ut_ad(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE);
1126
1127- mutex_enter(&block->mutex);
1128+ //mutex_enter(&block->mutex);
1129 #if UNIV_WORD_SIZE == 4
1130 /* On 32-bit systems, there is no padding in buf_page_t. On
1131 other systems, Valgrind could complain about uninitialized pad
1bfc1981 1132@@ -2512,8 +2656,8 @@
b4e1fa2c
AM		 1133 /* Try to evict the block from the buffer pool, to use the
1134 insert buffer (change buffer) as much as possible. */
1135
db82db79 1136- if (buf_LRU_free_block(&block->page, TRUE)) {
11822e22 1137- mutex_exit(&block->mutex);
db82db79 1138+ if (buf_LRU_free_block(&block->page, TRUE, FALSE)) {
11822e22 1139+ mutex_exit(block_mutex);
b4e1fa2c 1140 if (mode == BUF_GET_IF_IN_POOL_OR_WATCH) {
df1b5770 1141 /* Set the watch, as it would have
11822e22 1142 been set if the page were not in the
1bfc1981 1143@@ -2522,6 +2666,9 @@
11822e22
AM		 1144 space, offset, fold);
1145
1146 if (UNIV_LIKELY_NULL(block)) {
1147+ block_mutex = buf_page_get_mutex((buf_page_t*)block);
1148+ ut_a(block_mutex);
1149+ ut_ad(mutex_own(block_mutex));
1150
1151 /* The page entered the buffer
1152 pool for some reason. Try to
1bfc1981 1153@@ -2529,7 +2676,7 @@
11822e22
AM		 1154 goto got_block;
1155 }
1156 }
1157- buf_pool_mutex_exit(buf_pool);
1158+ //buf_pool_mutex_exit(buf_pool);
1159 fprintf(stderr,
1160 "innodb_change_buffering_debug evict %u %u\n",
1161 (unsigned) space, (unsigned) offset);
1bfc1981 1162@@ -2551,13 +2698,14 @@
db82db79
AM		 1163 ut_a(mode == BUF_GET_POSSIBLY_FREED
1164 || !block->page.file_page_was_freed);
1165 #endif
b4e1fa2c
AM		 1166- mutex_exit(&block->mutex);
1167+ //mutex_exit(&block->mutex);
1168
1169 /* Check if this is the first access to the page */
1170
1171 access_time = buf_page_is_accessed(&block->page);
1172
1173- buf_pool_mutex_exit(buf_pool);
1174+ //buf_pool_mutex_exit(buf_pool);
1175+ mutex_exit(block_mutex);
1176
adf0fb13
AM		 1177 if (UNIV_LIKELY(mode != BUF_PEEK_IF_IN_POOL)) {
1178 buf_page_set_accessed_make_young(&block->page, access_time);
1bfc1981 1179@@ -2790,9 +2938,11 @@
b4e1fa2c
AM		 1180 buf_pool = buf_pool_from_block(block);
1181
1182 if (mode == BUF_MAKE_YOUNG && buf_page_peek_if_too_old(&block->page)) {
1183- buf_pool_mutex_enter(buf_pool);
1184+ //buf_pool_mutex_enter(buf_pool);
1185+ mutex_enter(&buf_pool->LRU_list_mutex);
1186 buf_LRU_make_block_young(&block->page);
1187- buf_pool_mutex_exit(buf_pool);
1188+ //buf_pool_mutex_exit(buf_pool);
1189+ mutex_exit(&buf_pool->LRU_list_mutex);
1190 } else if (!buf_page_is_accessed(&block->page)) {
1191 /* Above, we do a dirty read on purpose, to avoid
1192 mutex contention. The field buf_page_t::access_time
1bfc1981 1193@@ -2800,9 +2950,11 @@
b4e1fa2c
AM		 1194 field must be protected by mutex, however. */
1195 ulint time_ms = ut_time_ms();
1196
1197- buf_pool_mutex_enter(buf_pool);
1198+ //buf_pool_mutex_enter(buf_pool);
1199+ mutex_enter(&block->mutex);
1200 buf_page_set_accessed(&block->page, time_ms);
1201- buf_pool_mutex_exit(buf_pool);
1202+ //buf_pool_mutex_exit(buf_pool);
1203+ mutex_exit(&block->mutex);
1204 }
1205
adf0fb13 1206 ut_ad(!ibuf_inside(mtr) || mode == BUF_KEEP_OLD);
1bfc1981 1207@@ -2869,18 +3021,21 @@
b4e1fa2c
AM		 1208 ut_ad(mtr);
1209 ut_ad(mtr->state == MTR_ACTIVE);
1210
1211- buf_pool_mutex_enter(buf_pool);
1212+ //buf_pool_mutex_enter(buf_pool);
1213+ rw_lock_s_lock(&buf_pool->page_hash_latch);
1214 block = buf_block_hash_get(buf_pool, space_id, page_no);
1215
1216 if (!block || buf_block_get_state(block) != BUF_BLOCK_FILE_PAGE) {
1217- buf_pool_mutex_exit(buf_pool);
1218+ //buf_pool_mutex_exit(buf_pool);
1219+ rw_lock_s_unlock(&buf_pool->page_hash_latch);
1220 return(NULL);
1221 }
1222
1223 ut_ad(!buf_pool_watch_is_sentinel(buf_pool, &block->page));
1224
1225 mutex_enter(&block->mutex);
1226- buf_pool_mutex_exit(buf_pool);
1227+ //buf_pool_mutex_exit(buf_pool);
1228+ rw_lock_s_unlock(&buf_pool->page_hash_latch);
1229
1230 #if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
1231 ut_a(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE);
1bfc1981 1232@@ -2970,7 +3125,10 @@
b4e1fa2c 1233 buf_page_t* hash_page;
b4e1fa2c 1234
734d6226 1235 ut_ad(buf_pool == buf_pool_get(space, offset));
b4e1fa2c
AM		 1236- ut_ad(buf_pool_mutex_own(buf_pool));
1237+ //ut_ad(buf_pool_mutex_own(buf_pool));
1238+#ifdef UNIV_SYNC_DEBUG
1239+ ut_ad(rw_lock_own(&buf_pool->page_hash_latch, RW_LOCK_EX));
1240+#endif
1241 ut_ad(mutex_own(&(block->mutex)));
1242 ut_a(buf_block_get_state(block) != BUF_BLOCK_FILE_PAGE);
1243
1bfc1981 1244@@ -2999,11 +3157,14 @@
b4e1fa2c
AM		 1245 if (UNIV_LIKELY(!hash_page)) {
1246 } else if (buf_pool_watch_is_sentinel(buf_pool, hash_page)) {
1247 /* Preserve the reference count. */
1248- ulint buf_fix_count = hash_page->buf_fix_count;
1249+ ulint buf_fix_count;
1250
1251+ mutex_enter(&buf_pool->zip_mutex);
1252+ buf_fix_count = hash_page->buf_fix_count;
1253 ut_a(buf_fix_count > 0);
1254 block->page.buf_fix_count += buf_fix_count;
1255 buf_pool_watch_remove(buf_pool, fold, hash_page);
1256+ mutex_exit(&buf_pool->zip_mutex);
1257 } else {
1258 fprintf(stderr,
1259 "InnoDB: Error: page %lu %lu already found"
1bfc1981 1260@@ -3013,7 +3174,8 @@
b4e1fa2c
AM		 1261 (const void*) hash_page, (const void*) block);
1262 #if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
1263 mutex_exit(&block->mutex);
1264- buf_pool_mutex_exit(buf_pool);
1265+ //buf_pool_mutex_exit(buf_pool);
1266+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
1267 buf_print();
1268 buf_LRU_print();
1269 buf_validate();
1bfc1981 1270@@ -3096,7 +3258,9 @@
b4e1fa2c
AM		 1271
1272 fold = buf_page_address_fold(space, offset);
1273
1274- buf_pool_mutex_enter(buf_pool);
1275+ //buf_pool_mutex_enter(buf_pool);
1276+ mutex_enter(&buf_pool->LRU_list_mutex);
1277+ rw_lock_x_lock(&buf_pool->page_hash_latch);
1278
1279 watch_page = buf_page_hash_get_low(buf_pool, space, offset, fold);
1280 if (watch_page && !buf_pool_watch_is_sentinel(buf_pool, watch_page)) {
1bfc1981 1281@@ -3105,9 +3269,15 @@
b4e1fa2c
AM		 1282 err_exit:
1283 if (block) {
1284 mutex_enter(&block->mutex);
1285- buf_LRU_block_free_non_file_page(block);
1286+ mutex_exit(&buf_pool->LRU_list_mutex);
1287+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
1288+ buf_LRU_block_free_non_file_page(block, FALSE);
1289 mutex_exit(&block->mutex);
1290 }
1291+ else {
1292+ mutex_exit(&buf_pool->LRU_list_mutex);
1293+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
1294+ }
1295
1296 bpage = NULL;
1297 goto func_exit;
1bfc1981 1298@@ -3130,6 +3300,8 @@
b4e1fa2c 1299
734d6226 1300 buf_page_init(buf_pool, space, offset, fold, block);
b4e1fa2c
AM		 1301
1302+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
1303+
1304 /* The block must be put to the LRU list, to the old blocks */
1305 buf_LRU_add_block(bpage, TRUE/* to old blocks */);
1306
1bfc1981 1307@@ -3157,7 +3329,7 @@
b4e1fa2c
AM		 1308 been added to buf_pool->LRU and
1309 buf_pool->page_hash. */
1310 mutex_exit(&block->mutex);
1311- data = buf_buddy_alloc(buf_pool, zip_size, &lru);
1312+ data = buf_buddy_alloc(buf_pool, zip_size, &lru, FALSE);
1313 mutex_enter(&block->mutex);
1314 block->page.zip.data = data;
1315
1bfc1981 1316@@ -3170,13 +3342,14 @@
b4e1fa2c
AM		 1317 buf_unzip_LRU_add_block(block, TRUE);
1318 }
1319
1320+ mutex_exit(&buf_pool->LRU_list_mutex);
1321 mutex_exit(&block->mutex);
1322 } else {
db82db79 1323 /* The compressed page must be allocated before the
b4e1fa2c
AM		 1324 control block (bpage), in order to avoid the
1325 invocation of buf_buddy_relocate_block() on
1326 uninitialized data. */
1327- data = buf_buddy_alloc(buf_pool, zip_size, &lru);
b4e1fa2c 1328+ data = buf_buddy_alloc(buf_pool, zip_size, &lru, TRUE);
b4e1fa2c 1329
db82db79
AM		 1330 /* If buf_buddy_alloc() allocated storage from the LRU list,
1331 it released and reacquired buf_pool->mutex. Thus, we must
1bfc1981 1332@@ -3192,7 +3365,10 @@
db82db79 1333
b4e1fa2c
AM		 1334 /* The block was added by some other thread. */
1335 watch_page = NULL;
b4e1fa2c 1336- buf_buddy_free(buf_pool, data, zip_size);
b4e1fa2c
AM		 1337+ buf_buddy_free(buf_pool, data, zip_size, TRUE);
1338+
1339+ mutex_exit(&buf_pool->LRU_list_mutex);
1340+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
1341
1342 bpage = NULL;
1343 goto func_exit;
1bfc1981 1344@@ -3240,20 +3416,26 @@
b4e1fa2c
AM		 1345 HASH_INSERT(buf_page_t, hash, buf_pool->page_hash, fold,
1346 bpage);
1347
1348+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
1349+
1350 /* The block must be put to the LRU list, to the old blocks */
1351 buf_LRU_add_block(bpage, TRUE/* to old blocks */);
db82db79 1352 #if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
b4e1fa2c 1353 buf_LRU_insert_zip_clean(bpage);
db82db79 1354 #endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
b4e1fa2c
AM		 1355
1356+ mutex_exit(&buf_pool->LRU_list_mutex);
1357+
1358 buf_page_set_io_fix(bpage, BUF_IO_READ);
1359
1360 mutex_exit(&buf_pool->zip_mutex);
1361 }
1362
1363+ buf_pool_mutex_enter(buf_pool);
1364 buf_pool->n_pend_reads++;
1365-func_exit:
1366 buf_pool_mutex_exit(buf_pool);
1367+func_exit:
1368+ //buf_pool_mutex_exit(buf_pool);
1369
1370 if (mode == BUF_READ_IBUF_PAGES_ONLY) {
1371
1bfc1981 1372@@ -3295,7 +3477,9 @@
b4e1fa2c
AM		 1373
1374 fold = buf_page_address_fold(space, offset);
1375
1376- buf_pool_mutex_enter(buf_pool);
1377+ //buf_pool_mutex_enter(buf_pool);
1378+ mutex_enter(&buf_pool->LRU_list_mutex);
1379+ rw_lock_x_lock(&buf_pool->page_hash_latch);
1380
1381 block = (buf_block_t*) buf_page_hash_get_low(
1382 buf_pool, space, offset, fold);
1bfc1981 1383@@ -3311,7 +3495,9 @@
df1b5770 1384 #endif /* UNIV_DEBUG_FILE_ACCESSES || UNIV_DEBUG */
b4e1fa2c
AM		 1385
1386 /* Page can be found in buf_pool */
1387- buf_pool_mutex_exit(buf_pool);
1388+ //buf_pool_mutex_exit(buf_pool);
1389+ mutex_exit(&buf_pool->LRU_list_mutex);
1390+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
1391
1392 buf_block_free(free_block);
1393
1bfc1981 1394@@ -3333,6 +3519,7 @@
b4e1fa2c
AM		 1395 mutex_enter(&block->mutex);
1396
734d6226 1397 buf_page_init(buf_pool, space, offset, fold, block);
b4e1fa2c
AM		 1398+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
1399
1400 /* The block must be put to the LRU list */
1401 buf_LRU_add_block(&block->page, FALSE);
1bfc1981 1402@@ -3359,7 +3546,7 @@
b4e1fa2c
AM		 1403 the reacquisition of buf_pool->mutex. We also must
1404 defer this operation until after the block descriptor
1405 has been added to buf_pool->LRU and buf_pool->page_hash. */
1406- data = buf_buddy_alloc(buf_pool, zip_size, &lru);
1407+ data = buf_buddy_alloc(buf_pool, zip_size, &lru, FALSE);
1408 mutex_enter(&block->mutex);
1409 block->page.zip.data = data;
1410
1bfc1981 1411@@ -3377,7 +3564,8 @@
b4e1fa2c
AM		 1412
1413 buf_page_set_accessed(&block->page, time_ms);
1414
1415- buf_pool_mutex_exit(buf_pool);
1416+ //buf_pool_mutex_exit(buf_pool);
1417+ mutex_exit(&buf_pool->LRU_list_mutex);
1418
1419 mtr_memo_push(mtr, block, MTR_MEMO_BUF_FIX);
1420
1bfc1981 1421@@ -3432,7 +3620,9 @@
734d6226
AM		 1422 ibool ret = TRUE;
1423
1424 /* First unfix and release lock on the bpage */
1425- buf_pool_mutex_enter(buf_pool);
1426+ //buf_pool_mutex_enter(buf_pool);
1427+ mutex_enter(&buf_pool->LRU_list_mutex);
1428+ rw_lock_x_lock(&buf_pool->page_hash_latch);
1429 mutex_enter(buf_page_get_mutex(bpage));
1430 ut_ad(buf_page_get_io_fix(bpage) == BUF_IO_READ);
1431 ut_ad(bpage->buf_fix_count == 0);
1bfc1981 1432@@ -3453,11 +3643,15 @@
734d6226
AM		 1433 ret = FALSE;
1434 }
1435
1436+ buf_pool_mutex_enter(buf_pool);
1437 ut_ad(buf_pool->n_pend_reads > 0);
1438 buf_pool->n_pend_reads--;
1439+ buf_pool_mutex_exit(buf_pool);
1440
1441 mutex_exit(buf_page_get_mutex(bpage));
1442- buf_pool_mutex_exit(buf_pool);
1443+ //buf_pool_mutex_exit(buf_pool);
1444+ mutex_exit(&buf_pool->LRU_list_mutex);
1445+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
1446
1447 return(ret);
1448 }
1bfc1981 1449@@ -3475,6 +3669,8 @@
b4e1fa2c
AM		 1450 buf_pool_t* buf_pool = buf_pool_from_bpage(bpage);
1451 const ibool uncompressed = (buf_page_get_state(bpage)
1452 == BUF_BLOCK_FILE_PAGE);
1453+ ibool have_LRU_mutex = FALSE;
1454+ mutex_t* block_mutex;
1455
1456 ut_a(buf_page_in_file(bpage));
1457
1bfc1981 1458@@ -3617,8 +3813,26 @@
b4e1fa2c
AM		 1459 }
1460 }
1461
1462+ if (io_type == BUF_IO_WRITE
1463+ && (buf_page_get_state(bpage) == BUF_BLOCK_ZIP_DIRTY
1464+ || buf_page_get_flush_type(bpage) == BUF_FLUSH_LRU)) {
1465+ /* to keep consistency at buf_LRU_insert_zip_clean() */
1466+ have_LRU_mutex = TRUE; /* optimistic */
1467+ }
1468+retry_mutex:
1469+ if (have_LRU_mutex)
1470+ mutex_enter(&buf_pool->LRU_list_mutex);
1471+ block_mutex = buf_page_get_mutex_enter(bpage);
1472+ ut_a(block_mutex);
1473+ if (io_type == BUF_IO_WRITE
1474+ && (buf_page_get_state(bpage) == BUF_BLOCK_ZIP_DIRTY
1475+ || buf_page_get_flush_type(bpage) == BUF_FLUSH_LRU)
1476+ && !have_LRU_mutex) {
1477+ mutex_exit(block_mutex);
1478+ have_LRU_mutex = TRUE;
1479+ goto retry_mutex;
1480+ }
1481 buf_pool_mutex_enter(buf_pool);
1482- mutex_enter(buf_page_get_mutex(bpage));
1483
1484 #ifdef UNIV_IBUF_COUNT_DEBUG
1485 if (io_type == BUF_IO_WRITE || uncompressed) {
1bfc1981 1486@@ -3641,6 +3855,7 @@
b4e1fa2c
AM		 1487 the x-latch to this OS thread: do not let this confuse you in
1488 debugging! */
1489
1490+ ut_a(!have_LRU_mutex);
1491 ut_ad(buf_pool->n_pend_reads > 0);
1492 buf_pool->n_pend_reads--;
1493 buf_pool->stat.n_pages_read++;
1bfc1981 1494@@ -3658,6 +3873,9 @@
b4e1fa2c
AM		 1495
1496 buf_flush_write_complete(bpage);
1497
1498+ if (have_LRU_mutex)
1499+ mutex_exit(&buf_pool->LRU_list_mutex);
1500+
1501 if (uncompressed) {
1502 rw_lock_s_unlock_gen(&((buf_block_t*) bpage)->lock,
1503 BUF_IO_WRITE);
1bfc1981 1504@@ -3680,8 +3898,8 @@
b4e1fa2c
AM		 1505 }
1506 #endif /* UNIV_DEBUG */
1507
1508- mutex_exit(buf_page_get_mutex(bpage));
1509 buf_pool_mutex_exit(buf_pool);
1510+ mutex_exit(block_mutex);
1511 }
1512
1513 /*********************************************************************//**
1bfc1981 1514@@ -3698,7 +3916,9 @@
b4e1fa2c
AM		 1515
1516 ut_ad(buf_pool);
1517
1518- buf_pool_mutex_enter(buf_pool);
1519+ //buf_pool_mutex_enter(buf_pool);
1520+ mutex_enter(&buf_pool->LRU_list_mutex);
1521+ rw_lock_x_lock(&buf_pool->page_hash_latch);
1522
1523 chunk = buf_pool->chunks;
1524
1bfc1981 1525@@ -3715,7 +3935,9 @@
b4e1fa2c
AM		 1526 }
1527 }
1528
1529- buf_pool_mutex_exit(buf_pool);
1530+ //buf_pool_mutex_exit(buf_pool);
1531+ mutex_exit(&buf_pool->LRU_list_mutex);
1532+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
1533
1534 return(TRUE);
1535 }
1bfc1981 1536@@ -3763,7 +3985,8 @@
b4e1fa2c
AM		 1537 freed = buf_LRU_search_and_free_block(buf_pool, 100);
1538 }
1539
1540- buf_pool_mutex_enter(buf_pool);
1541+ //buf_pool_mutex_enter(buf_pool);
1542+ mutex_enter(&buf_pool->LRU_list_mutex);
1543
1544 ut_ad(UT_LIST_GET_LEN(buf_pool->LRU) == 0);
1545 ut_ad(UT_LIST_GET_LEN(buf_pool->unzip_LRU) == 0);
1bfc1981 1546@@ -3776,7 +3999,8 @@
b4e1fa2c
AM		 1547 memset(&buf_pool->stat, 0x00, sizeof(buf_pool->stat));
1548 buf_refresh_io_stats(buf_pool);
1549
1550- buf_pool_mutex_exit(buf_pool);
1551+ //buf_pool_mutex_exit(buf_pool);
1552+ mutex_exit(&buf_pool->LRU_list_mutex);
1553 }
1554
1555 /*********************************************************************//**
1bfc1981 1556@@ -3818,7 +4042,10 @@
b4e1fa2c
AM		 1557
1558 ut_ad(buf_pool);
1559
1560- buf_pool_mutex_enter(buf_pool);
1561+ //buf_pool_mutex_enter(buf_pool);
1562+ mutex_enter(&buf_pool->LRU_list_mutex);
1563+ rw_lock_x_lock(&buf_pool->page_hash_latch);
1564+ /* for keep the new latch order, it cannot validate correctly... */
1565
1566 chunk = buf_pool->chunks;
1567
1bfc1981 1568@@ -3913,7 +4140,7 @@
b4e1fa2c
AM		 1569 /* Check clean compressed-only blocks. */
1570
1571 for (b = UT_LIST_GET_FIRST(buf_pool->zip_clean); b;
1572- b = UT_LIST_GET_NEXT(list, b)) {
1573+ b = UT_LIST_GET_NEXT(zip_list, b)) {
1574 ut_a(buf_page_get_state(b) == BUF_BLOCK_ZIP_PAGE);
1575 switch (buf_page_get_io_fix(b)) {
1576 case BUF_IO_NONE:
1bfc1981 1577@@ -3944,7 +4171,7 @@
b4e1fa2c
AM		 1578
1579 buf_flush_list_mutex_enter(buf_pool);
1580 for (b = UT_LIST_GET_FIRST(buf_pool->flush_list); b;
1581- b = UT_LIST_GET_NEXT(list, b)) {
1582+ b = UT_LIST_GET_NEXT(flush_list, b)) {
1583 ut_ad(b->in_flush_list);
1584 ut_a(b->oldest_modification);
1585 n_flush++;
1bfc1981 1586@@ -4003,6 +4230,8 @@
b4e1fa2c
AM		 1587 }
1588
1589 ut_a(UT_LIST_GET_LEN(buf_pool->LRU) == n_lru);
1590+ /* because of latching order with block->mutex, we cannot get needed mutexes before that */
1591+/*
1592 if (UT_LIST_GET_LEN(buf_pool->free) != n_free) {
1593 fprintf(stderr, "Free list len %lu, free blocks %lu\n",
1594 (ulong) UT_LIST_GET_LEN(buf_pool->free),
1bfc1981 1595@@ -4013,8 +4242,11 @@
b4e1fa2c
AM		 1596 ut_a(buf_pool->n_flush[BUF_FLUSH_SINGLE_PAGE] == n_single_flush);
1597 ut_a(buf_pool->n_flush[BUF_FLUSH_LIST] == n_list_flush);
1598 ut_a(buf_pool->n_flush[BUF_FLUSH_LRU] == n_lru_flush);
1599+*/
1600
1601- buf_pool_mutex_exit(buf_pool);
1602+ //buf_pool_mutex_exit(buf_pool);
1603+ mutex_exit(&buf_pool->LRU_list_mutex);
1604+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
1605
1606 ut_a(buf_LRU_validate());
1607 ut_a(buf_flush_validate(buf_pool));
1bfc1981 1608@@ -4070,7 +4302,9 @@
b4e1fa2c
AM		 1609 index_ids = mem_alloc(size * sizeof *index_ids);
1610 counts = mem_alloc(sizeof(ulint) * size);
1611
1612- buf_pool_mutex_enter(buf_pool);
1613+ //buf_pool_mutex_enter(buf_pool);
1614+ mutex_enter(&buf_pool->LRU_list_mutex);
1615+ mutex_enter(&buf_pool->free_list_mutex);
1616 buf_flush_list_mutex_enter(buf_pool);
1617
1618 fprintf(stderr,
1bfc1981 1619@@ -4139,7 +4373,9 @@
b4e1fa2c
AM		 1620 }
1621 }
1622
1623- buf_pool_mutex_exit(buf_pool);
1624+ //buf_pool_mutex_exit(buf_pool);
1625+ mutex_exit(&buf_pool->LRU_list_mutex);
1626+ mutex_exit(&buf_pool->free_list_mutex);
1627
1628 for (i = 0; i < n_found; i++) {
1629 index = dict_index_get_if_in_cache(index_ids[i]);
1bfc1981 1630@@ -4196,7 +4432,7 @@
b4e1fa2c
AM		 1631 buf_chunk_t* chunk;
1632 ulint fixed_pages_number = 0;
1633
1634- buf_pool_mutex_enter(buf_pool);
1635+ //buf_pool_mutex_enter(buf_pool);
1636
1637 chunk = buf_pool->chunks;
1638
1bfc1981 1639@@ -4230,7 +4466,7 @@
b4e1fa2c
AM		 1640 /* Traverse the lists of clean and dirty compressed-only blocks. */
1641
1642 for (b = UT_LIST_GET_FIRST(buf_pool->zip_clean); b;
1643- b = UT_LIST_GET_NEXT(list, b)) {
1644+ b = UT_LIST_GET_NEXT(zip_list, b)) {
1645 ut_a(buf_page_get_state(b) == BUF_BLOCK_ZIP_PAGE);
1646 ut_a(buf_page_get_io_fix(b) != BUF_IO_WRITE);
1647
1bfc1981 1648@@ -4242,7 +4478,7 @@
b4e1fa2c
AM		 1649
1650 buf_flush_list_mutex_enter(buf_pool);
1651 for (b = UT_LIST_GET_FIRST(buf_pool->flush_list); b;
1652- b = UT_LIST_GET_NEXT(list, b)) {
1653+ b = UT_LIST_GET_NEXT(flush_list, b)) {
1654 ut_ad(b->in_flush_list);
1655
1656 switch (buf_page_get_state(b)) {
1bfc1981 1657@@ -4268,7 +4504,7 @@
b4e1fa2c
AM		 1658
1659 buf_flush_list_mutex_exit(buf_pool);
1660 mutex_exit(&buf_pool->zip_mutex);
1661- buf_pool_mutex_exit(buf_pool);
1662+ //buf_pool_mutex_exit(buf_pool);
1663
1664 return(fixed_pages_number);
1665 }
1bfc1981 1666@@ -4426,6 +4662,8 @@
d8778560
AM		 1667 /* Find appropriate pool_info to store stats for this buffer pool */
1668 pool_info = &all_pool_info[pool_id];
b4e1fa2c
AM		 1669
1670+ mutex_enter(&buf_pool->LRU_list_mutex);
1671+ mutex_enter(&buf_pool->free_list_mutex);
1672 buf_pool_mutex_enter(buf_pool);
1673 buf_flush_list_mutex_enter(buf_pool);
1674
1bfc1981 1675@@ -4541,6 +4779,8 @@
d8778560 1676 pool_info->unzip_cur = buf_LRU_stat_cur.unzip;
b4e1fa2c
AM		 1677
1678 buf_refresh_io_stats(buf_pool);
1679+ mutex_exit(&buf_pool->LRU_list_mutex);
1680+ mutex_exit(&buf_pool->free_list_mutex);
1681 buf_pool_mutex_exit(buf_pool);
1682 }
1683
1bfc1981 1684@@ -4785,11 +5025,13 @@
b4e1fa2c
AM		 1685 {
1686 ulint len;
1687
1688- buf_pool_mutex_enter(buf_pool);
1689+ //buf_pool_mutex_enter(buf_pool);
1690+ mutex_enter(&buf_pool->free_list_mutex);
1691
1692 len = UT_LIST_GET_LEN(buf_pool->free);
1693
1694- buf_pool_mutex_exit(buf_pool);
1695+ //buf_pool_mutex_exit(buf_pool);
1696+ mutex_exit(&buf_pool->free_list_mutex);
1697
1698 return(len);
1699 }
db82db79
AM		 1700--- a/storage/innobase/buf/buf0flu.c
1701+++ b/storage/innobase/buf/buf0flu.c
d8778560 1702@@ -307,7 +307,7 @@
b4e1fa2c
AM		 1703
1704 ut_d(block->page.in_flush_list = TRUE);
1705 block->page.oldest_modification = lsn;
1706- UT_LIST_ADD_FIRST(list, buf_pool->flush_list, &block->page);
1707+ UT_LIST_ADD_FIRST(flush_list, buf_pool->flush_list, &block->page);
1708
1709 #ifdef UNIV_DEBUG_VALGRIND
1710 {
d8778560 1711@@ -401,14 +401,14 @@
b4e1fa2c
AM		 1712 > block->page.oldest_modification) {
1713 ut_ad(b->in_flush_list);
1714 prev_b = b;
1715- b = UT_LIST_GET_NEXT(list, b);
1716+ b = UT_LIST_GET_NEXT(flush_list, b);
1717 }
1718 }
1719
1720 if (prev_b == NULL) {
1721- UT_LIST_ADD_FIRST(list, buf_pool->flush_list, &block->page);
1722+ UT_LIST_ADD_FIRST(flush_list, buf_pool->flush_list, &block->page);
1723 } else {
1724- UT_LIST_INSERT_AFTER(list, buf_pool->flush_list,
1725+ UT_LIST_INSERT_AFTER(flush_list, buf_pool->flush_list,
1726 prev_b, &block->page);
1727 }
1728
d8778560 1729@@ -434,7 +434,7 @@
b4e1fa2c
AM		 1730 //buf_pool_t* buf_pool = buf_pool_from_bpage(bpage);
1731 //ut_ad(buf_pool_mutex_own(buf_pool));
1732 #endif
1733- //ut_ad(mutex_own(buf_page_get_mutex(bpage)));
1734+ ut_ad(mutex_own(buf_page_get_mutex(bpage)));
1735 //ut_ad(bpage->in_LRU_list);
1736
1737 if (UNIV_LIKELY(bpage->in_LRU_list && buf_page_in_file(bpage))) {
d8778560 1738@@ -470,14 +470,14 @@
b4e1fa2c
AM		 1739 enum buf_flush flush_type)/*!< in: BUF_FLUSH_LRU or BUF_FLUSH_LIST */
1740 {
1741 #ifdef UNIV_DEBUG
1742- buf_pool_t* buf_pool = buf_pool_from_bpage(bpage);
1743- ut_ad(buf_pool_mutex_own(buf_pool));
1744+ //buf_pool_t* buf_pool = buf_pool_from_bpage(bpage);
1745+ //ut_ad(buf_pool_mutex_own(buf_pool));
1746 #endif
1747- ut_a(buf_page_in_file(bpage));
1748+ //ut_a(buf_page_in_file(bpage));
1749 ut_ad(mutex_own(buf_page_get_mutex(bpage)));
1750 ut_ad(flush_type == BUF_FLUSH_LRU || BUF_FLUSH_LIST);
1751
1752- if (bpage->oldest_modification != 0
1753+ if (buf_page_in_file(bpage) && bpage->oldest_modification != 0
1754 && buf_page_get_io_fix(bpage) == BUF_IO_NONE) {
1755 ut_ad(bpage->in_flush_list);
1756
d8778560 1757@@ -508,7 +508,7 @@
b4e1fa2c
AM		 1758 {
1759 buf_pool_t* buf_pool = buf_pool_from_bpage(bpage);
1760
1761- ut_ad(buf_pool_mutex_own(buf_pool));
1762+ //ut_ad(buf_pool_mutex_own(buf_pool));
1763 ut_ad(mutex_own(buf_page_get_mutex(bpage)));
1764 ut_ad(bpage->in_flush_list);
1765
db82db79 1766@@ -526,13 +526,13 @@
b4e1fa2c
AM		 1767 return;
1768 case BUF_BLOCK_ZIP_DIRTY:
1769 buf_page_set_state(bpage, BUF_BLOCK_ZIP_PAGE);
1770- UT_LIST_REMOVE(list, buf_pool->flush_list, bpage);
1771+ UT_LIST_REMOVE(flush_list, buf_pool->flush_list, bpage);
db82db79 1772 #if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
b4e1fa2c 1773 buf_LRU_insert_zip_clean(bpage);
db82db79 1774 #endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
b4e1fa2c
AM		 1775 break;
1776 case BUF_BLOCK_FILE_PAGE:
1777- UT_LIST_REMOVE(list, buf_pool->flush_list, bpage);
1778+ UT_LIST_REMOVE(flush_list, buf_pool->flush_list, bpage);
1779 break;
1780 }
1781
db82db79 1782@@ -576,7 +576,7 @@
b4e1fa2c
AM		 1783 buf_page_t* prev_b = NULL;
1784 buf_pool_t* buf_pool = buf_pool_from_bpage(bpage);
1785
1786- ut_ad(buf_pool_mutex_own(buf_pool));
1787+ //ut_ad(buf_pool_mutex_own(buf_pool));
1788 /* Must reside in the same buffer pool. */
1789 ut_ad(buf_pool == buf_pool_from_bpage(dpage));
1790
db82db79 1791@@ -605,18 +605,18 @@
b4e1fa2c
AM		 1792 because we assert on in_flush_list in comparison function. */
1793 ut_d(bpage->in_flush_list = FALSE);
1794
1795- prev = UT_LIST_GET_PREV(list, bpage);
1796- UT_LIST_REMOVE(list, buf_pool->flush_list, bpage);
1797+ prev = UT_LIST_GET_PREV(flush_list, bpage);
1798+ UT_LIST_REMOVE(flush_list, buf_pool->flush_list, bpage);
1799
1800 if (prev) {
1801 ut_ad(prev->in_flush_list);
1802 UT_LIST_INSERT_AFTER(
1803- list,
1804+ flush_list,
1805 buf_pool->flush_list,
1806 prev, dpage);
1807 } else {
1808 UT_LIST_ADD_FIRST(
1809- list,
1810+ flush_list,
1811 buf_pool->flush_list,
1812 dpage);
1813 }
db82db79 1814@@ -1085,7 +1085,7 @@
b4e1fa2c
AM		 1815
1816 #ifdef UNIV_DEBUG
1817 buf_pool_t* buf_pool = buf_pool_from_bpage(bpage);
1818- ut_ad(!buf_pool_mutex_own(buf_pool));
1819+ //ut_ad(!buf_pool_mutex_own(buf_pool));
1820 #endif
1821
1822 #ifdef UNIV_LOG_DEBUG
db82db79 1823@@ -1099,7 +1099,8 @@
b4e1fa2c
AM		 1824 io_fixed and oldest_modification != 0. Thus, it cannot be
1825 relocated in the buffer pool or removed from flush_list or
1826 LRU_list. */
1827- ut_ad(!buf_pool_mutex_own(buf_pool));
1828+ //ut_ad(!buf_pool_mutex_own(buf_pool));
1829+ ut_ad(!mutex_own(&buf_pool->LRU_list_mutex));
1830 ut_ad(!buf_flush_list_mutex_own(buf_pool));
1831 ut_ad(!mutex_own(buf_page_get_mutex(bpage)));
1832 ut_ad(buf_page_get_io_fix(bpage) == BUF_IO_WRITE);
db82db79 1833@@ -1179,7 +1180,7 @@
11822e22
AM		 1834 buf_pool_t* buf_pool, /*!< in/out: buffer pool instance */
1835 buf_block_t* block) /*!< in/out: buffer control block */
1836 {
1837- ut_ad(buf_pool_mutex_own(buf_pool));
1838+ //ut_ad(buf_pool_mutex_own(buf_pool));
1839 ut_ad(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE);
1840 ut_ad(mutex_own(&block->mutex));
1841
db82db79 1842@@ -1187,8 +1188,11 @@
11822e22
AM		 1843 return(FALSE);
1844 }
1845
1846+ buf_pool_mutex_enter(buf_pool);
1847+
1848 if (buf_pool->n_flush[BUF_FLUSH_LRU] > 0
1849 || buf_pool->init_flush[BUF_FLUSH_LRU]) {
1850+ buf_pool_mutex_exit(buf_pool);
1851 /* There is already a flush batch of the same type running */
1852 return(FALSE);
1853 }
db82db79 1854@@ -1262,12 +1266,18 @@
b4e1fa2c
AM		 1855 ibool is_uncompressed;
1856
1857 ut_ad(flush_type == BUF_FLUSH_LRU || flush_type == BUF_FLUSH_LIST);
1858- ut_ad(buf_pool_mutex_own(buf_pool));
1859+ //ut_ad(buf_pool_mutex_own(buf_pool));
1860+#ifdef UNIV_SYNC_DEBUG
1861+ ut_ad(rw_lock_own(&buf_pool->page_hash_latch, RW_LOCK_SHARED));
1862+#endif
1863 ut_ad(buf_page_in_file(bpage));
1864
1865 block_mutex = buf_page_get_mutex(bpage);
1866 ut_ad(mutex_own(block_mutex));
1867
1868+ buf_pool_mutex_enter(buf_pool);
1869+ rw_lock_s_unlock(&buf_pool->page_hash_latch);
1870+
1871 ut_ad(buf_flush_ready_for_flush(bpage, flush_type));
1872
1873 buf_page_set_io_fix(bpage, BUF_IO_WRITE);
3d3ecf24 1874@@ -1455,14 +1465,16 @@
b4e1fa2c
AM		 1875
1876 buf_pool = buf_pool_get(space, i);
1877
1878- buf_pool_mutex_enter(buf_pool);
1879+ //buf_pool_mutex_enter(buf_pool);
1880+ rw_lock_s_lock(&buf_pool->page_hash_latch);
1881
1882 /* We only want to flush pages from this buffer pool. */
1883 bpage = buf_page_hash_get(buf_pool, space, i);
1884
1885 if (!bpage) {
1886
1887- buf_pool_mutex_exit(buf_pool);
1888+ //buf_pool_mutex_exit(buf_pool);
1889+ rw_lock_s_unlock(&buf_pool->page_hash_latch);
1bfc1981 1890 if (srv_flush_neighbor_pages == 2) {
b4e1fa2c 1891
1bfc1981 1892 /* This is contiguous neighbor page flush and
3d3ecf24 1893@@ -1480,11 +1492,9 @@
b4e1fa2c
AM		 1894 if (flush_type != BUF_FLUSH_LRU
1895 || i == offset
1896 || buf_page_is_old(bpage)) {
1897- mutex_t* block_mutex = buf_page_get_mutex(bpage);
b4e1fa2c
AM		 1898+ mutex_t* block_mutex = buf_page_get_mutex_enter(bpage);
1899
11822e22
AM		 1900- mutex_enter(block_mutex);
1901-
b4e1fa2c
AM		 1902- if (buf_flush_ready_for_flush(bpage, flush_type)
1903+ if (block_mutex && buf_flush_ready_for_flush(bpage, flush_type)
1904 && (i == offset || !bpage->buf_fix_count)) {
1905 /* We only try to flush those
1906 neighbors != offset where the buf fix
3d3ecf24 1907@@ -1500,11 +1510,12 @@
b4e1fa2c
AM		 1908 ut_ad(!buf_pool_mutex_own(buf_pool));
1909 count++;
1910 continue;
1911- } else {
1912+ } else if (block_mutex) {
1913 mutex_exit(block_mutex);
1914 }
1915 }
1916- buf_pool_mutex_exit(buf_pool);
1917+ //buf_pool_mutex_exit(buf_pool);
1918+ rw_lock_s_unlock(&buf_pool->page_hash_latch);
b4e1fa2c 1919
3d3ecf24 1920 if (srv_flush_neighbor_pages == 2) {
1bfc1981 1921
3d3ecf24 1922@@ -1553,21 +1564,25 @@
b4e1fa2c
AM		 1923 buf_pool_t* buf_pool = buf_pool_from_bpage(bpage);
1924 #endif /* UNIV_DEBUG */
1925
1926- ut_ad(buf_pool_mutex_own(buf_pool));
1927+ //ut_ad(buf_pool_mutex_own(buf_pool));
1928+ ut_ad(flush_type != BUF_FLUSH_LRU
1929+ || mutex_own(&buf_pool->LRU_list_mutex));
1930
1931- block_mutex = buf_page_get_mutex(bpage);
1932- mutex_enter(block_mutex);
1933+ block_mutex = buf_page_get_mutex_enter(bpage);
1934
1935- ut_a(buf_page_in_file(bpage));
1936+ //ut_a(buf_page_in_file(bpage));
1937
1938- if (buf_flush_ready_for_flush(bpage, flush_type)) {
1939+ if (block_mutex && buf_flush_ready_for_flush(bpage, flush_type)) {
1940 ulint space;
1941 ulint offset;
1942 buf_pool_t* buf_pool;
1943
1944 buf_pool = buf_pool_from_bpage(bpage);
1945
1946- buf_pool_mutex_exit(buf_pool);
1947+ //buf_pool_mutex_exit(buf_pool);
1948+ if (flush_type == BUF_FLUSH_LRU) {
1949+ mutex_exit(&buf_pool->LRU_list_mutex);
1950+ }
1951
1952 /* These fields are protected by both the
1953 buffer pool mutex and block mutex. */
3d3ecf24 1954@@ -1583,13 +1598,18 @@
b4e1fa2c
AM		 1955 *count,
1956 n_to_flush);
1957
1958- buf_pool_mutex_enter(buf_pool);
1959+ //buf_pool_mutex_enter(buf_pool);
1960+ if (flush_type == BUF_FLUSH_LRU) {
1961+ mutex_enter(&buf_pool->LRU_list_mutex);
1962+ }
1963 flushed = TRUE;
1964- } else {
1965+ } else if (block_mutex) {
1966 mutex_exit(block_mutex);
1967 }
1968
1969- ut_ad(buf_pool_mutex_own(buf_pool));
1970+ //ut_ad(buf_pool_mutex_own(buf_pool));
1971+ ut_ad(flush_type != BUF_FLUSH_LRU
1972+ || mutex_own(&buf_pool->LRU_list_mutex));
1973
1974 return(flushed);
1975 }
3d3ecf24 1976@@ -1610,7 +1630,8 @@
b4e1fa2c
AM		 1977 buf_page_t* bpage;
1978 ulint count = 0;
1979
1980- ut_ad(buf_pool_mutex_own(buf_pool));
1981+ //ut_ad(buf_pool_mutex_own(buf_pool));
1982+ ut_ad(mutex_own(&buf_pool->LRU_list_mutex));
1983
1984 do {
1985 /* Start from the end of the list looking for a
3d3ecf24 1986@@ -1632,7 +1653,8 @@
b4e1fa2c
AM		 1987 should be flushed, we factor in this value. */
1988 buf_lru_flush_page_count += count;
1989
1990- ut_ad(buf_pool_mutex_own(buf_pool));
1991+ //ut_ad(buf_pool_mutex_own(buf_pool));
1992+ ut_ad(mutex_own(&buf_pool->LRU_list_mutex));
1993
1994 return(count);
1995 }
3d3ecf24 1996@@ -1660,9 +1682,10 @@
b4e1fa2c
AM		 1997 {
1998 ulint len;
1999 buf_page_t* bpage;
2000+ buf_page_t* prev_bpage = NULL;
2001 ulint count = 0;
2002
2003- ut_ad(buf_pool_mutex_own(buf_pool));
2004+ //ut_ad(buf_pool_mutex_own(buf_pool));
2005
2006 /* If we have flushed enough, leave the loop */
2007 do {
3d3ecf24 2008@@ -1681,6 +1704,7 @@
b4e1fa2c
AM		 2009
2010 if (bpage) {
2011 ut_a(bpage->oldest_modification > 0);
2012+ prev_bpage = UT_LIST_GET_PREV(flush_list, bpage);
2013 }
2014
2015 if (!bpage || bpage->oldest_modification >= lsn_limit) {
3d3ecf24 2016@@ -1722,9 +1746,17 @@
b4e1fa2c
AM		 2017 break;
2018 }
2019
2020- bpage = UT_LIST_GET_PREV(list, bpage);
2021+ bpage = UT_LIST_GET_PREV(flush_list, bpage);
2022
2023- ut_ad(!bpage || bpage->in_flush_list);
2024+ //ut_ad(!bpage || bpage->in_flush_list);
2025+ if (bpage != prev_bpage) {
2026+ /* the search might warp.. retrying */
2027+ buf_flush_list_mutex_exit(buf_pool);
2028+ break;
2029+ }
2030+ if (bpage) {
2031+ prev_bpage = UT_LIST_GET_PREV(flush_list, bpage);
2032+ }
2033
2034 buf_flush_list_mutex_exit(buf_pool);
2035
3d3ecf24 2036@@ -1733,7 +1765,7 @@
b4e1fa2c
AM		 2037
2038 } while (count < min_n && bpage != NULL && len > 0);
2039
2040- ut_ad(buf_pool_mutex_own(buf_pool));
2041+ //ut_ad(buf_pool_mutex_own(buf_pool));
2042
2043 return(count);
2044 }
3d3ecf24 2045@@ -1772,13 +1804,15 @@
adf0fb13 2046 || sync_thread_levels_empty_except_dict());
b4e1fa2c
AM		 2047 #endif /* UNIV_SYNC_DEBUG */
2048
2049- buf_pool_mutex_enter(buf_pool);
2050+ //buf_pool_mutex_enter(buf_pool);
2051
2052 /* Note: The buffer pool mutex is released and reacquired within
2053 the flush functions. */
2054 switch(flush_type) {
2055 case BUF_FLUSH_LRU:
2056+ mutex_enter(&buf_pool->LRU_list_mutex);
2057 count = buf_flush_LRU_list_batch(buf_pool, min_n);
2058+ mutex_exit(&buf_pool->LRU_list_mutex);
2059 break;
2060 case BUF_FLUSH_LIST:
2061 count = buf_flush_flush_list_batch(buf_pool, min_n, lsn_limit);
3d3ecf24 2062@@ -1787,7 +1821,7 @@
b4e1fa2c
AM		 2063 ut_error;
2064 }
2065
2066- buf_pool_mutex_exit(buf_pool);
2067+ //buf_pool_mutex_exit(buf_pool);
2068
2069 buf_flush_buffered_writes();
2070
3d3ecf24 2071@@ -2059,7 +2093,7 @@
b4e1fa2c
AM		 2072 retry:
2073 //buf_pool_mutex_enter(buf_pool);
2074 if (have_LRU_mutex)
2075- buf_pool_mutex_enter(buf_pool);
2076+ mutex_enter(&buf_pool->LRU_list_mutex);
2077
2078 n_replaceable = UT_LIST_GET_LEN(buf_pool->free);
2079
3d3ecf24 2080@@ -2076,15 +2110,15 @@
b4e1fa2c
AM		 2081 bpage = UT_LIST_GET_LAST(buf_pool->LRU);
2082 continue;
2083 }
2084- block_mutex = buf_page_get_mutex(bpage);
2085-
2086- mutex_enter(block_mutex);
2087+ block_mutex = buf_page_get_mutex_enter(bpage);
2088
2089- if (buf_flush_ready_for_replace(bpage)) {
2090+ if (block_mutex && buf_flush_ready_for_replace(bpage)) {
2091 n_replaceable++;
2092 }
2093
2094- mutex_exit(block_mutex);
2095+ if (block_mutex) {
2096+ mutex_exit(block_mutex);
2097+ }
2098
2099 distance++;
2100
3d3ecf24 2101@@ -2093,7 +2127,7 @@
b4e1fa2c
AM		 2102
2103 //buf_pool_mutex_exit(buf_pool);
2104 if (have_LRU_mutex)
2105- buf_pool_mutex_exit(buf_pool);
2106+ mutex_exit(&buf_pool->LRU_list_mutex);
2107
2108 if (n_replaceable >= BUF_FLUSH_FREE_BLOCK_MARGIN(buf_pool)) {
2109
3d3ecf24 2110@@ -2292,7 +2326,7 @@
b4e1fa2c
AM		 2111
2112 ut_ad(buf_flush_list_mutex_own(buf_pool));
2113
2114- UT_LIST_VALIDATE(list, buf_page_t, buf_pool->flush_list,
2115+ UT_LIST_VALIDATE(flush_list, buf_page_t, buf_pool->flush_list,
2116 ut_ad(ut_list_node_313->in_flush_list));
2117
2118 bpage = UT_LIST_GET_FIRST(buf_pool->flush_list);
3d3ecf24 2119@@ -2332,7 +2366,7 @@
b4e1fa2c
AM		 2120 rnode = rbt_next(buf_pool->flush_rbt, rnode);
2121 }
2122
2123- bpage = UT_LIST_GET_NEXT(list, bpage);
2124+ bpage = UT_LIST_GET_NEXT(flush_list, bpage);
2125
2126 ut_a(!bpage || om >= bpage->oldest_modification);
2127 }
db82db79
AM		 2128--- a/storage/innobase/buf/buf0lru.c
2129+++ b/storage/innobase/buf/buf0lru.c
b4e1fa2c
AM		 2130@@ -143,8 +143,9 @@
2131 void
2132 buf_LRU_block_free_hashed_page(
2133 /*===========================*/
2134- buf_block_t* block); /*!< in: block, must contain a file page and
2135+ buf_block_t* block, /*!< in: block, must contain a file page and
2136 be in a state where it can be freed */
2137+ ibool have_page_hash_mutex);
2138
2139 /******************************************************************//**
2140 Determines if the unzip_LRU list should be used for evicting a victim
2141@@ -154,15 +155,20 @@
2142 ibool
2143 buf_LRU_evict_from_unzip_LRU(
2144 /*=========================*/
2145- buf_pool_t* buf_pool)
2146+ buf_pool_t* buf_pool,
2147+ ibool have_LRU_mutex)
2148 {
2149 ulint io_avg;
2150 ulint unzip_avg;
2151
2152- ut_ad(buf_pool_mutex_own(buf_pool));
2153+ //ut_ad(buf_pool_mutex_own(buf_pool));
2154
2155+ if (!have_LRU_mutex)
2156+ mutex_enter(&buf_pool->LRU_list_mutex);
2157 /* If the unzip_LRU list is empty, we can only use the LRU. */
2158 if (UT_LIST_GET_LEN(buf_pool->unzip_LRU) == 0) {
2159+ if (!have_LRU_mutex)
2160+ mutex_exit(&buf_pool->LRU_list_mutex);
2161 return(FALSE);
2162 }
2163
2164@@ -171,14 +177,20 @@
2165 decompressed pages in the buffer pool. */
2166 if (UT_LIST_GET_LEN(buf_pool->unzip_LRU)
2167 <= UT_LIST_GET_LEN(buf_pool->LRU) / 10) {
2168+ if (!have_LRU_mutex)
2169+ mutex_exit(&buf_pool->LRU_list_mutex);
2170 return(FALSE);
2171 }
2172
2173 /* If eviction hasn't started yet, we assume by default
2174 that a workload is disk bound. */
2175 if (buf_pool->freed_page_clock == 0) {
2176+ if (!have_LRU_mutex)
2177+ mutex_exit(&buf_pool->LRU_list_mutex);
2178 return(TRUE);
2179 }
2180+ if (!have_LRU_mutex)
2181+ mutex_exit(&buf_pool->LRU_list_mutex);
2182
2183 /* Calculate the average over past intervals, and add the values
2184 of the current interval. */
adf0fb13 2185@@ -246,18 +258,25 @@
b4e1fa2c
AM		 2186 page_arr = ut_malloc(
2187 sizeof(ulint) * BUF_LRU_DROP_SEARCH_HASH_SIZE);
2188
2189- buf_pool_mutex_enter(buf_pool);
2190+ //buf_pool_mutex_enter(buf_pool);
2191+ mutex_enter(&buf_pool->LRU_list_mutex);
adf0fb13 2192 num_entries = 0;
b4e1fa2c
AM		 2193
2194 scan_again:
b4e1fa2c
AM		 2195 bpage = UT_LIST_GET_LAST(buf_pool->LRU);
2196
2197 while (bpage != NULL) {
adf0fb13 2198+ /* bpage->state,space,io_fix,buf_fix_count are protected by block_mutex at XtraDB */
b4e1fa2c
AM		 2199+ mutex_t* block_mutex = buf_page_get_mutex_enter(bpage);
2200 buf_page_t* prev_bpage;
adf0fb13 2201 ibool is_fixed;
b4e1fa2c 2202
b4e1fa2c
AM		 2203 prev_bpage = UT_LIST_GET_PREV(LRU, bpage);
2204
adf0fb13 2205+ if (UNIV_UNLIKELY(!block_mutex)) {
b4e1fa2c
AM		 2206+ goto next_page;
2207+ }
2208+
2209 ut_a(buf_page_in_file(bpage));
2210
2211 if (buf_page_get_state(bpage) != BUF_BLOCK_FILE_PAGE
adf0fb13
AM		 2212@@ -266,23 +285,27 @@
2213 /* Compressed pages are never hashed.
2214 Skip blocks of other tablespaces.
2215 Skip I/O-fixed blocks (to be dealt with later). */
2216+ mutex_exit(block_mutex);
2217 next_page:
2218 bpage = prev_bpage;
2219 continue;
2220 }
b4e1fa2c 2221
adf0fb13
AM		 2222- mutex_enter(&((buf_block_t*) bpage)->mutex);
2223+ //mutex_enter(&((buf_block_t*) bpage)->mutex);
2224 is_fixed = bpage->buf_fix_count > 0
13ceb006 2225 || !((buf_block_t*) bpage)->index;
adf0fb13
AM		 2226- mutex_exit(&((buf_block_t*) bpage)->mutex);
2227+ //mutex_exit(&((buf_block_t*) bpage)->mutex);
b4e1fa2c 2228
adf0fb13
AM		 2229 if (is_fixed) {
2230+ mutex_exit(block_mutex);
2231 goto next_page;
2232 }
b4e1fa2c 2233
adf0fb13
AM		 2234 /* Store the page number so that we can drop the hash
2235 index in a batch later. */
2236 page_arr[num_entries] = bpage->offset;
2237+ mutex_exit(block_mutex);
2238+
2239 ut_a(num_entries < BUF_LRU_DROP_SEARCH_HASH_SIZE);
2240 ++num_entries;
b4e1fa2c 2241
adf0fb13
AM		 2242@@ -292,14 +315,16 @@
2243
2244 /* Array full. We release the buf_pool->mutex to obey
2245 the latching order. */
2246- buf_pool_mutex_exit(buf_pool);
2247+ //buf_pool_mutex_exit(buf_pool);
2248+ mutex_exit(&buf_pool->LRU_list_mutex);
2249
2250 buf_LRU_drop_page_hash_batch(
2251 id, zip_size, page_arr, num_entries);
2252
2253 num_entries = 0;
2254
2255- buf_pool_mutex_enter(buf_pool);
2256+ //buf_pool_mutex_enter(buf_pool);
2257+ mutex_enter(&buf_pool->LRU_list_mutex);
2258
2259 /* Note that we released the buf_pool mutex above
2260 after reading the prev_bpage during processing of a
2261@@ -317,13 +342,23 @@
2262 /* If, however, bpage has been removed from LRU list
2263 to the free list then we should restart the scan.
2264 bpage->state is protected by buf_pool mutex. */
2265+
2266+ /* obtain block_mutex again to avoid race condition of bpage->state */
2267+ block_mutex = buf_page_get_mutex_enter(bpage);
2268+ if (!block_mutex) {
2269+ goto scan_again;
2270+ }
2271+
2272 if (bpage
2273 && buf_page_get_state(bpage) != BUF_BLOCK_FILE_PAGE) {
2274+ mutex_exit(block_mutex);
2275 goto scan_again;
b4e1fa2c 2276 }
adf0fb13 2277+ mutex_exit(block_mutex);
b4e1fa2c
AM		 2278 }
2279
2280- buf_pool_mutex_exit(buf_pool);
2281+ //buf_pool_mutex_exit(buf_pool);
2282+ mutex_exit(&buf_pool->LRU_list_mutex);
2283
2284 /* Drop any remaining batch of search hashed pages. */
2285 buf_LRU_drop_page_hash_batch(id, zip_size, page_arr, num_entries);
adf0fb13 2286@@ -345,7 +380,9 @@
b4e1fa2c
AM		 2287 ibool all_freed;
2288
2289 scan_again:
2290- buf_pool_mutex_enter(buf_pool);
2291+ //buf_pool_mutex_enter(buf_pool);
2292+ mutex_enter(&buf_pool->LRU_list_mutex);
2293+ rw_lock_x_lock(&buf_pool->page_hash_latch);
2294
2295 all_freed = TRUE;
2296
db82db79 2297@@ -375,8 +412,15 @@
b4e1fa2c 2298 all_freed = FALSE;
db82db79 2299 goto next_page;
b4e1fa2c 2300 } else {
db82db79 2301- block_mutex = buf_page_get_mutex(bpage);
b4e1fa2c 2302- mutex_enter(block_mutex);
db82db79 2303+ block_mutex = buf_page_get_mutex_enter(bpage);
b4e1fa2c
AM		 2304+
2305+ if (!block_mutex) {
2306+ /* It may be impossible case...
2307+ Something wrong, so will be scan_again */
2308+
2309+ all_freed = FALSE;
db82db79 2310+ goto next_page;
b4e1fa2c
AM		 2311+ }
2312
2313 if (bpage->buf_fix_count > 0) {
2314
db82db79
AM		 2315@@ -409,7 +453,9 @@
2316 ulint page_no;
2317 ulint zip_size;
b4e1fa2c 2318
db82db79
AM		 2319- buf_pool_mutex_exit(buf_pool);
2320+ //buf_pool_mutex_exit(buf_pool);
2321+ mutex_exit(&buf_pool->LRU_list_mutex);
2322+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
2323
2324 zip_size = buf_page_get_zip_size(bpage);
2325 page_no = buf_page_get_page_no(bpage);
2326@@ -433,7 +479,7 @@
b4e1fa2c 2327
db82db79
AM		 2328 if (buf_LRU_block_remove_hashed_page(bpage, TRUE)
2329 != BUF_BLOCK_ZIP_FREE) {
2330- buf_LRU_block_free_hashed_page((buf_block_t*) bpage);
2331+ buf_LRU_block_free_hashed_page((buf_block_t*) bpage, TRUE);
2332 mutex_exit(block_mutex);
2333 } else {
2334 /* The block_mutex should have been released
2335@@ -446,7 +492,9 @@
b4e1fa2c
AM		 2336 bpage = prev_bpage;
2337 }
2338
2339- buf_pool_mutex_exit(buf_pool);
2340+ //buf_pool_mutex_exit(buf_pool);
2341+ mutex_exit(&buf_pool->LRU_list_mutex);
2342+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
2343
2344 if (!all_freed) {
2345 os_thread_sleep(20000);
db82db79 2346@@ -493,7 +541,9 @@
b4e1fa2c
AM		 2347 buf_page_t* b;
2348 buf_pool_t* buf_pool = buf_pool_from_bpage(bpage);
2349
2350- ut_ad(buf_pool_mutex_own(buf_pool));
2351+ //ut_ad(buf_pool_mutex_own(buf_pool));
2352+ ut_ad(mutex_own(&buf_pool->LRU_list_mutex));
11822e22 2353+ ut_ad(mutex_own(&buf_pool->zip_mutex));
b4e1fa2c
AM		 2354 ut_ad(buf_page_get_state(bpage) == BUF_BLOCK_ZIP_PAGE);
2355
2356 /* Find the first successor of bpage in the LRU list
db82db79 2357@@ -501,17 +551,17 @@
b4e1fa2c
AM		 2358 b = bpage;
2359 do {
2360 b = UT_LIST_GET_NEXT(LRU, b);
2361- } while (b && buf_page_get_state(b) != BUF_BLOCK_ZIP_PAGE);
2362+ } while (b && (buf_page_get_state(b) != BUF_BLOCK_ZIP_PAGE || !b->in_LRU_list));
2363
2364 /* Insert bpage before b, i.e., after the predecessor of b. */
2365 if (b) {
2366- b = UT_LIST_GET_PREV(list, b);
2367+ b = UT_LIST_GET_PREV(zip_list, b);
2368 }
2369
2370 if (b) {
2371- UT_LIST_INSERT_AFTER(list, buf_pool->zip_clean, b, bpage);
2372+ UT_LIST_INSERT_AFTER(zip_list, buf_pool->zip_clean, b, bpage);
2373 } else {
2374- UT_LIST_ADD_FIRST(list, buf_pool->zip_clean, bpage);
2375+ UT_LIST_ADD_FIRST(zip_list, buf_pool->zip_clean, bpage);
2376 }
2377 }
db82db79
AM		 2378 #endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
2379@@ -525,18 +575,19 @@
b4e1fa2c
AM		 2380 buf_LRU_free_from_unzip_LRU_list(
2381 /*=============================*/
2382 buf_pool_t* buf_pool, /*!< in: buffer pool instance */
2383- ulint n_iterations) /*!< in: how many times this has
2384+ ulint n_iterations, /*!< in: how many times this has
2385 been called repeatedly without
2386 result: a high value means that
2387 we should search farther; we will
2388 search n_iterations / 5 of the
2389 unzip_LRU list, or nothing if
2390 n_iterations >= 5 */
2391+ ibool have_LRU_mutex)
2392 {
2393 buf_block_t* block;
2394 ulint distance;
2395
2396- ut_ad(buf_pool_mutex_own(buf_pool));
2397+ //ut_ad(buf_pool_mutex_own(buf_pool));
2398
2399 /* Theoratically it should be much easier to find a victim
2400 from unzip_LRU as we can choose even a dirty block (as we'll
db82db79 2401@@ -546,7 +597,7 @@
b4e1fa2c
AM		 2402 if we have done five iterations so far. */
2403
2404 if (UNIV_UNLIKELY(n_iterations >= 5)
2405- || !buf_LRU_evict_from_unzip_LRU(buf_pool)) {
2406+ || !buf_LRU_evict_from_unzip_LRU(buf_pool, have_LRU_mutex)) {
2407
2408 return(FALSE);
2409 }
db82db79 2410@@ -554,18 +605,25 @@
b4e1fa2c
AM		 2411 distance = 100 + (n_iterations
2412 * UT_LIST_GET_LEN(buf_pool->unzip_LRU)) / 5;
2413
2414+restart:
2415 for (block = UT_LIST_GET_LAST(buf_pool->unzip_LRU);
2416 UNIV_LIKELY(block != NULL) && UNIV_LIKELY(distance > 0);
2417 block = UT_LIST_GET_PREV(unzip_LRU, block), distance--) {
2418
db82db79 2419 ibool freed;
b4e1fa2c
AM		 2420
2421+ mutex_enter(&block->mutex);
2422+ if (!block->in_unzip_LRU_list || !block->page.in_LRU_list
2423+ || buf_block_get_state(block) != BUF_BLOCK_FILE_PAGE) {
2424+ mutex_exit(&block->mutex);
2425+ goto restart;
2426+ }
2427+
2428 ut_ad(buf_block_get_state(block) == BUF_BLOCK_FILE_PAGE);
2429 ut_ad(block->in_unzip_LRU_list);
2430 ut_ad(block->page.in_LRU_list);
2431
2432- mutex_enter(&block->mutex);
df1b5770
AM		 2433- freed = buf_LRU_free_block(&block->page, FALSE);
2434+ freed = buf_LRU_free_block(&block->page, FALSE, have_LRU_mutex);
b4e1fa2c
AM		 2435 mutex_exit(&block->mutex);
2436
db82db79
AM		 2437 if (freed) {
2438@@ -584,35 +642,46 @@
b4e1fa2c
AM		 2439 buf_LRU_free_from_common_LRU_list(
2440 /*==============================*/
2441 buf_pool_t* buf_pool,
2442- ulint n_iterations)
2443+ ulint n_iterations,
2444 /*!< in: how many times this has been called
2445 repeatedly without result: a high value means
2446 that we should search farther; if
2447 n_iterations < 10, then we search
2448 n_iterations / 10 * buf_pool->curr_size
2449 pages from the end of the LRU list */
2450+ ibool have_LRU_mutex)
2451 {
2452 buf_page_t* bpage;
2453 ulint distance;
2454
2455- ut_ad(buf_pool_mutex_own(buf_pool));
2456+ //ut_ad(buf_pool_mutex_own(buf_pool));
2457
2458 distance = 100 + (n_iterations * buf_pool->curr_size) / 10;
2459
2460+restart:
2461 for (bpage = UT_LIST_GET_LAST(buf_pool->LRU);
2462 UNIV_LIKELY(bpage != NULL) && UNIV_LIKELY(distance > 0);
2463 bpage = UT_LIST_GET_PREV(LRU, bpage), distance--) {
db82db79
AM		 2464
2465 ibool freed;
2466 unsigned accessed;
2467- mutex_t* block_mutex = buf_page_get_mutex(bpage);
2468+ mutex_t* block_mutex = buf_page_get_mutex_enter(bpage);
b4e1fa2c
AM		 2469+
2470+ if (!block_mutex) {
2471+ goto restart;
2472+ }
2473+
2474+ if (!bpage->in_LRU_list
2475+ || !buf_page_in_file(bpage)) {
2476+ mutex_exit(block_mutex);
2477+ goto restart;
2478+ }
2479
2480 ut_ad(buf_page_in_file(bpage));
2481 ut_ad(bpage->in_LRU_list);
2482
2483- mutex_enter(block_mutex);
2484 accessed = buf_page_is_accessed(bpage);
df1b5770
AM		 2485- freed = buf_LRU_free_block(bpage, TRUE);
2486+ freed = buf_LRU_free_block(bpage, TRUE, have_LRU_mutex);
b4e1fa2c
AM		 2487 mutex_exit(block_mutex);
2488
db82db79
AM		 2489 if (freed) {
2490@@ -649,16 +718,23 @@
b4e1fa2c
AM		 2491 n_iterations / 5 of the unzip_LRU list. */
2492 {
2493 ibool freed = FALSE;
2494+ ibool have_LRU_mutex = FALSE;
2495
2496- buf_pool_mutex_enter(buf_pool);
2497+ if (UT_LIST_GET_LEN(buf_pool->unzip_LRU))
2498+ have_LRU_mutex = TRUE;
db82db79
AM		 2499
2500- freed = buf_LRU_free_from_unzip_LRU_list(buf_pool, n_iterations);
b4e1fa2c
AM		 2501+ //buf_pool_mutex_enter(buf_pool);
2502+ if (have_LRU_mutex)
2503+ mutex_enter(&buf_pool->LRU_list_mutex);
db82db79 2504+
b4e1fa2c
AM		 2505+ freed = buf_LRU_free_from_unzip_LRU_list(buf_pool, n_iterations, have_LRU_mutex);
2506
2507 if (!freed) {
2508 freed = buf_LRU_free_from_common_LRU_list(
2509- buf_pool, n_iterations);
2510+ buf_pool, n_iterations, have_LRU_mutex);
2511 }
2512
2513+ buf_pool_mutex_enter(buf_pool);
2514 if (!freed) {
2515 buf_pool->LRU_flush_ended = 0;
2516 } else if (buf_pool->LRU_flush_ended > 0) {
db82db79 2517@@ -666,6 +742,8 @@
b4e1fa2c
AM		 2518 }
2519
2520 buf_pool_mutex_exit(buf_pool);
2521+ if (have_LRU_mutex)
2522+ mutex_exit(&buf_pool->LRU_list_mutex);
2523
2524 return(freed);
2525 }
db82db79 2526@@ -726,7 +804,9 @@
b4e1fa2c
AM		 2527
2528 buf_pool = buf_pool_from_array(i);
2529
2530- buf_pool_mutex_enter(buf_pool);
2531+ //buf_pool_mutex_enter(buf_pool);
2532+ mutex_enter(&buf_pool->LRU_list_mutex);
2533+ mutex_enter(&buf_pool->free_list_mutex);
2534
2535 if (!recv_recovery_on
2536 && UT_LIST_GET_LEN(buf_pool->free)
db82db79 2537@@ -736,7 +816,9 @@
b4e1fa2c
AM		 2538 ret = TRUE;
2539 }
2540
2541- buf_pool_mutex_exit(buf_pool);
2542+ //buf_pool_mutex_exit(buf_pool);
2543+ mutex_exit(&buf_pool->LRU_list_mutex);
2544+ mutex_exit(&buf_pool->free_list_mutex);
2545 }
2546
2547 return(ret);
db82db79 2548@@ -754,9 +836,10 @@
b4e1fa2c
AM		 2549 {
2550 buf_block_t* block;
2551
2552- ut_ad(buf_pool_mutex_own(buf_pool));
2553+ //ut_ad(buf_pool_mutex_own(buf_pool));
2554
2555- block = (buf_block_t*) UT_LIST_GET_FIRST(buf_pool->free);
2556+ mutex_enter(&buf_pool->free_list_mutex);
2557+ block = (buf_block_t*) UT_LIST_GET_LAST(buf_pool->free);
2558
2559 if (block) {
2560
db82db79 2561@@ -765,7 +848,9 @@
b4e1fa2c
AM		 2562 ut_ad(!block->page.in_flush_list);
2563 ut_ad(!block->page.in_LRU_list);
2564 ut_a(!buf_page_in_file(&block->page));
2565- UT_LIST_REMOVE(list, buf_pool->free, (&block->page));
2566+ UT_LIST_REMOVE(free, buf_pool->free, (&block->page));
2567+
2568+ mutex_exit(&buf_pool->free_list_mutex);
2569
2570 mutex_enter(&block->mutex);
2571
db82db79 2572@@ -775,6 +860,8 @@
b4e1fa2c
AM		 2573 ut_ad(buf_pool_from_block(block) == buf_pool);
2574
2575 mutex_exit(&block->mutex);
2576+ } else {
2577+ mutex_exit(&buf_pool->free_list_mutex);
2578 }
2579
2580 return(block);
db82db79 2581@@ -797,7 +884,7 @@
b4e1fa2c
AM		 2582 ibool mon_value_was = FALSE;
2583 ibool started_monitor = FALSE;
2584 loop:
2585- buf_pool_mutex_enter(buf_pool);
2586+ //buf_pool_mutex_enter(buf_pool);
2587
2588 if (!recv_recovery_on && UT_LIST_GET_LEN(buf_pool->free)
2589 + UT_LIST_GET_LEN(buf_pool->LRU) < buf_pool->curr_size / 20) {
db82db79 2590@@ -865,7 +952,7 @@
b4e1fa2c 2591
df1b5770
AM		 2592 /* If there is a block in the free list, take it */
2593 block = buf_LRU_get_free_only(buf_pool);
b4e1fa2c
AM		 2594- buf_pool_mutex_exit(buf_pool);
2595+ //buf_pool_mutex_exit(buf_pool);
2596
df1b5770
AM		 2597 if (block) {
2598 ut_ad(buf_pool_from_block(block) == buf_pool);
db82db79 2599@@ -965,7 +1052,8 @@
b4e1fa2c
AM		 2600 ulint new_len;
2601
2602 ut_a(buf_pool->LRU_old);
2603- ut_ad(buf_pool_mutex_own(buf_pool));
2604+ //ut_ad(buf_pool_mutex_own(buf_pool));
2605+ ut_ad(mutex_own(&buf_pool->LRU_list_mutex));
2606 ut_ad(buf_pool->LRU_old_ratio >= BUF_LRU_OLD_RATIO_MIN);
2607 ut_ad(buf_pool->LRU_old_ratio <= BUF_LRU_OLD_RATIO_MAX);
2608 #if BUF_LRU_OLD_RATIO_MIN * BUF_LRU_OLD_MIN_LEN <= BUF_LRU_OLD_RATIO_DIV * (BUF_LRU_OLD_TOLERANCE + 5)
db82db79 2609@@ -1031,7 +1119,8 @@
b4e1fa2c
AM		 2610 {
2611 buf_page_t* bpage;
2612
2613- ut_ad(buf_pool_mutex_own(buf_pool));
2614+ //ut_ad(buf_pool_mutex_own(buf_pool));
2615+ ut_ad(mutex_own(&buf_pool->LRU_list_mutex));
2616 ut_a(UT_LIST_GET_LEN(buf_pool->LRU) == BUF_LRU_OLD_MIN_LEN);
2617
2618 /* We first initialize all blocks in the LRU list as old and then use
db82db79 2619@@ -1066,13 +1155,14 @@
b4e1fa2c
AM		 2620 ut_ad(buf_pool);
2621 ut_ad(bpage);
2622 ut_ad(buf_page_in_file(bpage));
2623- ut_ad(buf_pool_mutex_own(buf_pool));
2624+ //ut_ad(buf_pool_mutex_own(buf_pool));
2625+ ut_ad(mutex_own(&buf_pool->LRU_list_mutex));
2626
2627 if (buf_page_belongs_to_unzip_LRU(bpage)) {
2628 buf_block_t* block = (buf_block_t*) bpage;
2629
2630 ut_ad(block->in_unzip_LRU_list);
2631- ut_d(block->in_unzip_LRU_list = FALSE);
2632+ block->in_unzip_LRU_list = FALSE;
2633
2634 UT_LIST_REMOVE(unzip_LRU, buf_pool->unzip_LRU, block);
2635 }
db82db79 2636@@ -1090,7 +1180,8 @@
b4e1fa2c
AM		 2637
2638 ut_ad(buf_pool);
2639 ut_ad(bpage);
2640- ut_ad(buf_pool_mutex_own(buf_pool));
2641+ //ut_ad(buf_pool_mutex_own(buf_pool));
2642+ ut_ad(mutex_own(&buf_pool->LRU_list_mutex));
2643
2644 ut_a(buf_page_in_file(bpage));
2645
db82db79 2646@@ -1167,12 +1258,13 @@
b4e1fa2c
AM		 2647
2648 ut_ad(buf_pool);
2649 ut_ad(block);
2650- ut_ad(buf_pool_mutex_own(buf_pool));
2651+ //ut_ad(buf_pool_mutex_own(buf_pool));
2652+ ut_ad(mutex_own(&buf_pool->LRU_list_mutex));
2653
2654 ut_a(buf_page_belongs_to_unzip_LRU(&block->page));
2655
2656 ut_ad(!block->in_unzip_LRU_list);
2657- ut_d(block->in_unzip_LRU_list = TRUE);
2658+ block->in_unzip_LRU_list = TRUE;
2659
2660 if (old) {
2661 UT_LIST_ADD_LAST(unzip_LRU, buf_pool->unzip_LRU, block);
db82db79 2662@@ -1193,7 +1285,8 @@
b4e1fa2c
AM		 2663
2664 ut_ad(buf_pool);
2665 ut_ad(bpage);
2666- ut_ad(buf_pool_mutex_own(buf_pool));
2667+ //ut_ad(buf_pool_mutex_own(buf_pool));
2668+ ut_ad(mutex_own(&buf_pool->LRU_list_mutex));
2669
2670 ut_a(buf_page_in_file(bpage));
2671
db82db79 2672@@ -1244,7 +1337,8 @@
b4e1fa2c
AM		 2673
2674 ut_ad(buf_pool);
2675 ut_ad(bpage);
2676- ut_ad(buf_pool_mutex_own(buf_pool));
2677+ //ut_ad(buf_pool_mutex_own(buf_pool));
2678+ ut_ad(mutex_own(&buf_pool->LRU_list_mutex));
2679
2680 ut_a(buf_page_in_file(bpage));
2681 ut_ad(!bpage->in_LRU_list);
db82db79 2682@@ -1323,7 +1417,8 @@
b4e1fa2c
AM		 2683 {
2684 buf_pool_t* buf_pool = buf_pool_from_bpage(bpage);
2685
2686- ut_ad(buf_pool_mutex_own(buf_pool));
2687+ //ut_ad(buf_pool_mutex_own(buf_pool));
2688+ ut_ad(mutex_own(&buf_pool->LRU_list_mutex));
2689
2690 if (bpage->old) {
2691 buf_pool->stat.n_pages_made_young++;
db82db79 2692@@ -1362,17 +1457,18 @@
df1b5770
AM		 2693 buf_LRU_free_block(
2694 /*===============*/
b4e1fa2c 2695 buf_page_t* bpage, /*!< in: block to be freed */
df1b5770
AM		 2696- ibool zip) /*!< in: TRUE if should remove also the
2697+ ibool zip, /*!< in: TRUE if should remove also the
b4e1fa2c 2698 compressed page of an uncompressed page */
b4e1fa2c
AM		 2699+ ibool have_LRU_mutex)
2700 {
2701 buf_page_t* b = NULL;
2702 buf_pool_t* buf_pool = buf_pool_from_bpage(bpage);
2703 mutex_t* block_mutex = buf_page_get_mutex(bpage);
2704
2705- ut_ad(buf_pool_mutex_own(buf_pool));
2706+ //ut_ad(buf_pool_mutex_own(buf_pool));
2707 ut_ad(mutex_own(block_mutex));
2708 ut_ad(buf_page_in_file(bpage));
2709- ut_ad(bpage->in_LRU_list);
2710+ //ut_ad(bpage->in_LRU_list);
2711 ut_ad(!bpage->in_flush_list == !bpage->oldest_modification);
2712 #if UNIV_WORD_SIZE == 4
2713 /* On 32-bit systems, there is no padding in buf_page_t. On
db82db79 2714@@ -1381,7 +1477,7 @@
b4e1fa2c
AM		 2715 UNIV_MEM_ASSERT_RW(bpage, sizeof *bpage);
2716 #endif
2717
2718- if (!buf_page_can_relocate(bpage)) {
2719+ if (!bpage->in_LRU_list || !block_mutex || !buf_page_can_relocate(bpage)) {
2720
2721 /* Do not free buffer-fixed or I/O-fixed blocks. */
db82db79
AM		 2722 return(FALSE);
2723@@ -1415,7 +1511,7 @@
b4e1fa2c 2724 alloc:
db82db79
AM		 2725 b = buf_page_alloc_descriptor();
2726 ut_a(b);
b4e1fa2c
AM		 2727- memcpy(b, bpage, sizeof *b);
2728+ //memcpy(b, bpage, sizeof *b);
2729 }
2730
2731 #ifdef UNIV_DEBUG
db82db79 2732@@ -1426,6 +1522,39 @@
b4e1fa2c
AM		 2733 }
2734 #endif /* UNIV_DEBUG */
2735
2736+ /* not to break latch order, must re-enter block_mutex */
2737+ mutex_exit(block_mutex);
2738+
2739+ if (!have_LRU_mutex)
2740+ mutex_enter(&buf_pool->LRU_list_mutex); /* optimistic */
2741+ rw_lock_x_lock(&buf_pool->page_hash_latch);
2742+ mutex_enter(block_mutex);
2743+
2744+ /* recheck states of block */
2745+ if (!bpage->in_LRU_list || block_mutex != buf_page_get_mutex(bpage)
2746+ || !buf_page_can_relocate(bpage)) {
2747+not_freed:
2748+ if (b) {
2749+ buf_buddy_free(buf_pool, b, sizeof *b, TRUE);
2750+ }
2751+ if (!have_LRU_mutex)
2752+ mutex_exit(&buf_pool->LRU_list_mutex);
2753+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
db82db79 2754+ return(FALSE);
b4e1fa2c
AM		 2755+ } else if (zip || !bpage->zip.data) {
2756+ if (bpage->oldest_modification)
2757+ goto not_freed;
2758+ } else if (bpage->oldest_modification) {
2759+ if (buf_page_get_state(bpage) != BUF_BLOCK_FILE_PAGE) {
2760+ ut_ad(buf_page_get_state(bpage) == BUF_BLOCK_ZIP_DIRTY);
2761+ goto not_freed;
2762+ }
2763+ }
2764+
2765+ if (b) {
2766+ memcpy(b, bpage, sizeof *b);
2767+ }
2768+
2769 if (buf_LRU_block_remove_hashed_page(bpage, zip)
2770 != BUF_BLOCK_ZIP_FREE) {
2771 ut_a(bpage->buf_fix_count == 0);
db82db79 2772@@ -1442,6 +1571,10 @@
b4e1fa2c
AM		 2773
2774 ut_a(!hash_b);
2775
2776+ while (prev_b && !prev_b->in_LRU_list) {
2777+ prev_b = UT_LIST_GET_PREV(LRU, prev_b);
2778+ }
2779+
2780 b->state = b->oldest_modification
2781 ? BUF_BLOCK_ZIP_DIRTY
2782 : BUF_BLOCK_ZIP_PAGE;
db82db79 2783@@ -1517,6 +1650,7 @@
adf0fb13
AM		 2784 buf_LRU_add_block_low(b, buf_page_is_old(b));
2785 }
2786
2787+ mutex_enter(&buf_pool->zip_mutex);
2788 if (b->state == BUF_BLOCK_ZIP_PAGE) {
db82db79 2789 #if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
adf0fb13 2790 buf_LRU_insert_zip_clean(b);
db82db79 2791@@ -1534,9 +1668,12 @@
adf0fb13
AM		 2792 buf_pool->mutex and block_mutex. */
2793 b->buf_fix_count++;
df1b5770 2794 b->io_fix = BUF_IO_READ;
adf0fb13 2795+ mutex_exit(&buf_pool->zip_mutex);
b4e1fa2c
AM		 2796 }
2797
2798- buf_pool_mutex_exit(buf_pool);
2799+ //buf_pool_mutex_exit(buf_pool);
2800+ mutex_exit(&buf_pool->LRU_list_mutex);
2801+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
2802 mutex_exit(block_mutex);
2803
2804 /* Remove possible adaptive hash index on the page.
db82db79 2805@@ -1568,7 +1705,9 @@
b4e1fa2c
AM		 2806 : BUF_NO_CHECKSUM_MAGIC);
2807 }
2808
2809- buf_pool_mutex_enter(buf_pool);
2810+ //buf_pool_mutex_enter(buf_pool);
2811+ if (have_LRU_mutex)
2812+ mutex_enter(&buf_pool->LRU_list_mutex);
2813 mutex_enter(block_mutex);
2814
2815 if (b) {
db82db79 2816@@ -1578,13 +1717,17 @@
b4e1fa2c
AM		 2817 mutex_exit(&buf_pool->zip_mutex);
2818 }
2819
2820- buf_LRU_block_free_hashed_page((buf_block_t*) bpage);
2821+ buf_LRU_block_free_hashed_page((buf_block_t*) bpage, FALSE);
2822 } else {
2823 /* The block_mutex should have been released by
2824 buf_LRU_block_remove_hashed_page() when it returns
2825 BUF_BLOCK_ZIP_FREE. */
2826 ut_ad(block_mutex == &buf_pool->zip_mutex);
2827 mutex_enter(block_mutex);
2828+
2829+ if (!have_LRU_mutex)
2830+ mutex_exit(&buf_pool->LRU_list_mutex);
2831+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
2832 }
2833
db82db79
AM		 2834 return(TRUE);
2835@@ -1596,13 +1739,14 @@
b4e1fa2c
AM		 2836 void
2837 buf_LRU_block_free_non_file_page(
2838 /*=============================*/
2839- buf_block_t* block) /*!< in: block, must not contain a file page */
2840+ buf_block_t* block, /*!< in: block, must not contain a file page */
2841+ ibool have_page_hash_mutex)
2842 {
2843 void* data;
2844 buf_pool_t* buf_pool = buf_pool_from_block(block);
2845
2846 ut_ad(block);
2847- ut_ad(buf_pool_mutex_own(buf_pool));
2848+ //ut_ad(buf_pool_mutex_own(buf_pool));
2849 ut_ad(mutex_own(&block->mutex));
2850
2851 switch (buf_block_get_state(block)) {
db82db79 2852@@ -1636,18 +1780,21 @@
b4e1fa2c
AM		 2853 if (data) {
2854 block->page.zip.data = NULL;
2855 mutex_exit(&block->mutex);
2856- buf_pool_mutex_exit_forbid(buf_pool);
2857+ //buf_pool_mutex_exit_forbid(buf_pool);
2858
2859 buf_buddy_free(
2860- buf_pool, data, page_zip_get_size(&block->page.zip));
2861+ buf_pool, data, page_zip_get_size(&block->page.zip),
2862+ have_page_hash_mutex);
2863
2864- buf_pool_mutex_exit_allow(buf_pool);
2865+ //buf_pool_mutex_exit_allow(buf_pool);
2866 mutex_enter(&block->mutex);
2867 page_zip_set_size(&block->page.zip, 0);
2868 }
2869
2870- UT_LIST_ADD_FIRST(list, buf_pool->free, (&block->page));
2871+ mutex_enter(&buf_pool->free_list_mutex);
2872+ UT_LIST_ADD_FIRST(free, buf_pool->free, (&block->page));
2873 ut_d(block->page.in_free_list = TRUE);
2874+ mutex_exit(&buf_pool->free_list_mutex);
2875
2876 UNIV_MEM_ASSERT_AND_FREE(block->frame, UNIV_PAGE_SIZE);
2877 }
db82db79 2878@@ -1677,7 +1824,11 @@
b4e1fa2c
AM		 2879 buf_pool_t* buf_pool = buf_pool_from_bpage(bpage);
2880
2881 ut_ad(bpage);
2882- ut_ad(buf_pool_mutex_own(buf_pool));
2883+ //ut_ad(buf_pool_mutex_own(buf_pool));
2884+ ut_ad(mutex_own(&buf_pool->LRU_list_mutex));
2885+#ifdef UNIV_SYNC_DEBUG
2886+ ut_ad(rw_lock_own(&buf_pool->page_hash_latch, RW_LOCK_EX));
2887+#endif
2888 ut_ad(mutex_own(buf_page_get_mutex(bpage)));
2889
2890 ut_a(buf_page_get_io_fix(bpage) == BUF_IO_NONE);
db82db79 2891@@ -1785,7 +1936,9 @@
b4e1fa2c
AM		 2892
2893 #if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
2894 mutex_exit(buf_page_get_mutex(bpage));
2895- buf_pool_mutex_exit(buf_pool);
2896+ //buf_pool_mutex_exit(buf_pool);
2897+ mutex_exit(&buf_pool->LRU_list_mutex);
2898+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
2899 buf_print();
2900 buf_LRU_print();
2901 buf_validate();
db82db79 2902@@ -1807,17 +1960,17 @@
b4e1fa2c
AM		 2903 ut_a(buf_page_get_zip_size(bpage));
2904
db82db79 2905 #if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
b4e1fa2c
AM		 2906- UT_LIST_REMOVE(list, buf_pool->zip_clean, bpage);
2907+ UT_LIST_REMOVE(zip_list, buf_pool->zip_clean, bpage);
db82db79 2908 #endif /* UNIV_DEBUG || UNIV_BUF_DEBUG */
b4e1fa2c
AM		 2909
2910 mutex_exit(&buf_pool->zip_mutex);
2911- buf_pool_mutex_exit_forbid(buf_pool);
2912+ //buf_pool_mutex_exit_forbid(buf_pool);
2913
2914 buf_buddy_free(
2915 buf_pool, bpage->zip.data,
2916- page_zip_get_size(&bpage->zip));
2917+ page_zip_get_size(&bpage->zip), TRUE);
2918
b4e1fa2c 2919- buf_pool_mutex_exit_allow(buf_pool);
b4e1fa2c 2920+ //buf_pool_mutex_exit_allow(buf_pool);
db82db79 2921 buf_page_free_descriptor(bpage);
b4e1fa2c 2922 return(BUF_BLOCK_ZIP_FREE);
db82db79
AM		 2923
2924@@ -1839,13 +1992,13 @@
b4e1fa2c
AM		 2925 ut_ad(!bpage->in_flush_list);
2926 ut_ad(!bpage->in_LRU_list);
2927 mutex_exit(&((buf_block_t*) bpage)->mutex);
2928- buf_pool_mutex_exit_forbid(buf_pool);
2929+ //buf_pool_mutex_exit_forbid(buf_pool);
2930
2931 buf_buddy_free(
2932 buf_pool, data,
2933- page_zip_get_size(&bpage->zip));
2934+ page_zip_get_size(&bpage->zip), TRUE);
2935
2936- buf_pool_mutex_exit_allow(buf_pool);
2937+ //buf_pool_mutex_exit_allow(buf_pool);
2938 mutex_enter(&((buf_block_t*) bpage)->mutex);
2939 page_zip_set_size(&bpage->zip, 0);
2940 }
db82db79 2941@@ -1871,18 +2024,19 @@
b4e1fa2c
AM		 2942 void
2943 buf_LRU_block_free_hashed_page(
2944 /*===========================*/
2945- buf_block_t* block) /*!< in: block, must contain a file page and
2946+ buf_block_t* block, /*!< in: block, must contain a file page and
2947 be in a state where it can be freed */
2948+ ibool have_page_hash_mutex)
2949 {
2950 #ifdef UNIV_DEBUG
2951- buf_pool_t* buf_pool = buf_pool_from_block(block);
2952- ut_ad(buf_pool_mutex_own(buf_pool));
2953+ //buf_pool_t* buf_pool = buf_pool_from_block(block);
2954+ //ut_ad(buf_pool_mutex_own(buf_pool));
2955 #endif
2956 ut_ad(mutex_own(&block->mutex));
2957
2958 buf_block_set_state(block, BUF_BLOCK_MEMORY);
2959
2960- buf_LRU_block_free_non_file_page(block);
2961+ buf_LRU_block_free_non_file_page(block, have_page_hash_mutex);
2962 }
2963
734d6226
AM		 2964 /******************************************************************//**
2965@@ -1897,7 +2051,7 @@
2966 {
2967 if (buf_LRU_block_remove_hashed_page(bpage, TRUE)
2968 != BUF_BLOCK_ZIP_FREE) {
2969- buf_LRU_block_free_hashed_page((buf_block_t*) bpage);
2970+ buf_LRU_block_free_hashed_page((buf_block_t*) bpage, TRUE);
2971 }
2972 }
2973
2974@@ -1925,7 +2079,8 @@
b4e1fa2c
AM		 2975 }
2976
2977 if (adjust) {
2978- buf_pool_mutex_enter(buf_pool);
2979+ //buf_pool_mutex_enter(buf_pool);
2980+ mutex_enter(&buf_pool->LRU_list_mutex);
2981
2982 if (ratio != buf_pool->LRU_old_ratio) {
2983 buf_pool->LRU_old_ratio = ratio;
734d6226 2984@@ -1937,7 +2092,8 @@
b4e1fa2c
AM		 2985 }
2986 }
2987
2988- buf_pool_mutex_exit(buf_pool);
2989+ //buf_pool_mutex_exit(buf_pool);
2990+ mutex_exit(&buf_pool->LRU_list_mutex);
2991 } else {
2992 buf_pool->LRU_old_ratio = ratio;
2993 }
734d6226 2994@@ -2042,7 +2198,8 @@
b4e1fa2c
AM		 2995 ulint new_len;
2996
2997 ut_ad(buf_pool);
2998- buf_pool_mutex_enter(buf_pool);
2999+ //buf_pool_mutex_enter(buf_pool);
3000+ mutex_enter(&buf_pool->LRU_list_mutex);
3001
3002 if (UT_LIST_GET_LEN(buf_pool->LRU) >= BUF_LRU_OLD_MIN_LEN) {
3003
734d6226 3004@@ -2103,16 +2260,22 @@
b4e1fa2c
AM		 3005
3006 ut_a(buf_pool->LRU_old_len == old_len);
3007
3008- UT_LIST_VALIDATE(list, buf_page_t, buf_pool->free,
3009+ mutex_exit(&buf_pool->LRU_list_mutex);
3010+ mutex_enter(&buf_pool->free_list_mutex);
3011+
3012+ UT_LIST_VALIDATE(free, buf_page_t, buf_pool->free,
3013 ut_ad(ut_list_node_313->in_free_list));
3014
3015 for (bpage = UT_LIST_GET_FIRST(buf_pool->free);
3016 bpage != NULL;
3017- bpage = UT_LIST_GET_NEXT(list, bpage)) {
3018+ bpage = UT_LIST_GET_NEXT(free, bpage)) {
3019
3020 ut_a(buf_page_get_state(bpage) == BUF_BLOCK_NOT_USED);
3021 }
3022
3023+ mutex_exit(&buf_pool->free_list_mutex);
3024+ mutex_enter(&buf_pool->LRU_list_mutex);
3025+
3026 UT_LIST_VALIDATE(unzip_LRU, buf_block_t, buf_pool->unzip_LRU,
3027 ut_ad(ut_list_node_313->in_unzip_LRU_list
3028 && ut_list_node_313->page.in_LRU_list));
734d6226 3029@@ -2126,7 +2289,8 @@
b4e1fa2c
AM		 3030 ut_a(buf_page_belongs_to_unzip_LRU(&block->page));
3031 }
3032
3033- buf_pool_mutex_exit(buf_pool);
3034+ //buf_pool_mutex_exit(buf_pool);
3035+ mutex_exit(&buf_pool->LRU_list_mutex);
3036 }
3037
3038 /**********************************************************************//**
734d6226 3039@@ -2162,7 +2326,8 @@
b4e1fa2c
AM		 3040 const buf_page_t* bpage;
3041
3042 ut_ad(buf_pool);
3043- buf_pool_mutex_enter(buf_pool);
3044+ //buf_pool_mutex_enter(buf_pool);
3045+ mutex_enter(&buf_pool->LRU_list_mutex);
3046
3047 bpage = UT_LIST_GET_FIRST(buf_pool->LRU);
3048
734d6226 3049@@ -2219,7 +2384,8 @@
b4e1fa2c
AM		 3050 bpage = UT_LIST_GET_NEXT(LRU, bpage);
3051 }
3052
3053- buf_pool_mutex_exit(buf_pool);
3054+ //buf_pool_mutex_exit(buf_pool);
3055+ mutex_exit(&buf_pool->LRU_list_mutex);
3056 }
3057
3058 /**********************************************************************//**
db82db79
AM		 3059--- a/storage/innobase/buf/buf0rea.c
3060+++ b/storage/innobase/buf/buf0rea.c
734d6226 3061@@ -478,6 +478,7 @@
b4e1fa2c
AM		 3062
3063 return(0);
3064 }
3065+ buf_pool_mutex_exit(buf_pool);
3066
3067 /* Check that almost all pages in the area have been accessed; if
3068 offset == low, the accesses must be in a descending order, otherwise,
734d6226 3069@@ -496,6 +497,7 @@
b4e1fa2c
AM		 3070
3071 fail_count = 0;
3072
3073+ rw_lock_s_lock(&buf_pool->page_hash_latch);
3074 for (i = low; i < high; i++) {
3075 bpage = buf_page_hash_get(buf_pool, space, i);
3076
734d6226 3077@@ -523,7 +525,8 @@
b4e1fa2c
AM		 3078
3079 if (fail_count > threshold) {
3080 /* Too many failures: return */
3081- buf_pool_mutex_exit(buf_pool);
3082+ //buf_pool_mutex_exit(buf_pool);
3083+ rw_lock_s_unlock(&buf_pool->page_hash_latch);
3084 return(0);
3085 }
3086
734d6226 3087@@ -538,7 +541,8 @@
b4e1fa2c
AM		 3088 bpage = buf_page_hash_get(buf_pool, space, offset);
3089
3090 if (bpage == NULL) {
3091- buf_pool_mutex_exit(buf_pool);
3092+ //buf_pool_mutex_exit(buf_pool);
3093+ rw_lock_s_unlock(&buf_pool->page_hash_latch);
3094
3095 return(0);
3096 }
734d6226 3097@@ -564,7 +568,8 @@
b4e1fa2c
AM		 3098 pred_offset = fil_page_get_prev(frame);
3099 succ_offset = fil_page_get_next(frame);
3100
3101- buf_pool_mutex_exit(buf_pool);
3102+ //buf_pool_mutex_exit(buf_pool);
3103+ rw_lock_s_unlock(&buf_pool->page_hash_latch);
3104
3105 if ((offset == low) && (succ_offset == offset + 1)) {
3106
db82db79
AM		 3107--- a/storage/innobase/handler/ha_innodb.cc
3108+++ b/storage/innobase/handler/ha_innodb.cc
3109@@ -265,6 +265,10 @@
b4e1fa2c
AM		 3110 # endif /* !PFS_SKIP_BUFFER_MUTEX_RWLOCK */
3111 {&buf_pool_mutex_key, "buf_pool_mutex", 0},
3112 {&buf_pool_zip_mutex_key, "buf_pool_zip_mutex", 0},
3113+ {&buf_pool_LRU_list_mutex_key, "buf_pool_LRU_list_mutex", 0},
3114+ {&buf_pool_free_list_mutex_key, "buf_pool_free_list_mutex", 0},
3115+ {&buf_pool_zip_free_mutex_key, "buf_pool_zip_free_mutex", 0},
3116+ {&buf_pool_zip_hash_mutex_key, "buf_pool_zip_hash_mutex", 0},
3117 {&cache_last_read_mutex_key, "cache_last_read_mutex", 0},
3118 {&dict_foreign_err_mutex_key, "dict_foreign_err_mutex", 0},
3119 {&dict_sys_mutex_key, "dict_sys_mutex", 0},
db82db79 3120@@ -314,6 +318,7 @@
b4e1fa2c
AM		 3121 {&archive_lock_key, "archive_lock", 0},
3122 # endif /* UNIV_LOG_ARCHIVE */
3123 {&btr_search_latch_key, "btr_search_latch", 0},
3124+ {&buf_pool_page_hash_key, "buf_pool_page_hash_latch", 0},
3125 # ifndef PFS_SKIP_BUFFER_MUTEX_RWLOCK
3126 {&buf_block_lock_key, "buf_block_lock", 0},
3127 # endif /* !PFS_SKIP_BUFFER_MUTEX_RWLOCK */
db82db79
AM		 3128--- a/storage/innobase/handler/i_s.cc
3129+++ b/storage/innobase/handler/i_s.cc
734d6226 3130@@ -1583,7 +1583,8 @@
b4e1fa2c
AM		 3131
3132 buf_pool = buf_pool_from_array(i);
3133
3134- buf_pool_mutex_enter(buf_pool);
3135+ //buf_pool_mutex_enter(buf_pool);
3136+ mutex_enter(&buf_pool->zip_free_mutex);
3137
3138 for (uint x = 0; x <= BUF_BUDDY_SIZES; x++) {
3139 buf_buddy_stat_t* buddy_stat;
734d6226 3140@@ -1613,7 +1614,8 @@
b4e1fa2c
AM		 3141 }
3142 }
3143
3144- buf_pool_mutex_exit(buf_pool);
3145+ //buf_pool_mutex_exit(buf_pool);
3146+ mutex_exit(&buf_pool->zip_free_mutex);
3147
3148 if (status) {
3149 break;
db82db79
AM		 3150--- a/storage/innobase/ibuf/ibuf0ibuf.c
3151+++ b/storage/innobase/ibuf/ibuf0ibuf.c
1bfc1981 3152@@ -3760,9 +3760,11 @@
b4e1fa2c
AM		 3153 ulint fold = buf_page_address_fold(space, page_no);
3154 buf_pool_t* buf_pool = buf_pool_get(space, page_no);
3155
3156- buf_pool_mutex_enter(buf_pool);
3157+ //buf_pool_mutex_enter(buf_pool);
3158+ rw_lock_s_lock(&buf_pool->page_hash_latch);
3159 bpage = buf_page_hash_get_low(buf_pool, space, page_no, fold);
3160- buf_pool_mutex_exit(buf_pool);
3161+ //buf_pool_mutex_exit(buf_pool);
3162+ rw_lock_s_unlock(&buf_pool->page_hash_latch);
3163
3164 if (UNIV_LIKELY_NULL(bpage)) {
3165 /* A buffer pool watch has been set or the
db82db79
AM		 3166--- a/storage/innobase/include/buf0buddy.h
3167+++ b/storage/innobase/include/buf0buddy.h
3168@@ -49,11 +49,12 @@
3169 ulint size, /*!< in: compressed page size
3170 (between PAGE_ZIP_MIN_SIZE and
3171 UNIV_PAGE_SIZE) */
3172- ibool* lru) /*!< in: pointer to a variable
3173+ ibool* lru, /*!< in: pointer to a variable
3174 that will be assigned TRUE if
3175 storage was allocated from the
3176 LRU list and buf_pool->mutex was
3177 temporarily released */
3178+ ibool have_page_hash_mutex)
3179 __attribute__((malloc, nonnull));
b4e1fa2c
AM		 3180
3181 /**********************************************************************//**
db82db79
AM		 3182@@ -66,8 +67,9 @@
3183 the block resides */
3184 void* buf, /*!< in: block to be freed, must not
3185 be pointed to by the buffer pool */
3186- ulint size) /*!< in: block size,
3187+ ulint size, /*!< in: block size,
3188 up to UNIV_PAGE_SIZE */
3189+ ibool have_page_hash_mutex)
b4e1fa2c
AM		 3190 __attribute__((nonnull));
3191
3192 #ifndef UNIV_NONINL
db82db79
AM		 3193--- a/storage/innobase/include/buf0buddy.ic
3194+++ b/storage/innobase/include/buf0buddy.ic
3195@@ -45,11 +45,12 @@
3196 buf_pool_t* buf_pool, /*!< in/out: buffer pool instance */
3197 ulint i, /*!< in: index of buf_pool->zip_free[],
3198 or BUF_BUDDY_SIZES */
3199- ibool* lru) /*!< in: pointer to a variable that
3200+ ibool* lru, /*!< in: pointer to a variable that
3201 will be assigned TRUE if storage was
3202 allocated from the LRU list and
3203 buf_pool->mutex was temporarily
3204 released */
3205+ ibool have_page_hash_mutex)
3206 __attribute__((malloc, nonnull));
b4e1fa2c
AM		 3207
3208 /**********************************************************************//**
3209@@ -61,8 +62,9 @@
3210 buf_pool_t* buf_pool, /*!< in: buffer pool instance */
3211 void* buf, /*!< in: block to be freed, must not be
3212 pointed to by the buffer pool */
3213- ulint i) /*!< in: index of buf_pool->zip_free[],
3214+ ulint i, /*!< in: index of buf_pool->zip_free[],
3215 or BUF_BUDDY_SIZES */
3216+ ibool have_page_hash_mutex)
3217 __attribute__((nonnull));
3218
3219 /**********************************************************************//**
db82db79
AM		 3220@@ -101,19 +103,20 @@
3221 ulint size, /*!< in: compressed page size
3222 (between PAGE_ZIP_MIN_SIZE and
3223 UNIV_PAGE_SIZE) */
b4e1fa2c
AM		 3224- ibool* lru) /*!< in: pointer to a variable
3225+ ibool* lru, /*!< in: pointer to a variable
3226 that will be assigned TRUE if
3227 storage was allocated from the
3228 LRU list and buf_pool->mutex was
db82db79 3229 temporarily released */
b4e1fa2c
AM		 3230+ ibool have_page_hash_mutex)
3231 {
3232- ut_ad(buf_pool_mutex_own(buf_pool));
3233+ //ut_ad(buf_pool_mutex_own(buf_pool));
db82db79
AM		 3234 ut_ad(ut_is_2pow(size));
3235 ut_ad(size >= PAGE_ZIP_MIN_SIZE);
3236 ut_ad(size <= UNIV_PAGE_SIZE);
b4e1fa2c 3237
db82db79
AM		 3238 return((byte*) buf_buddy_alloc_low(buf_pool, buf_buddy_get_slot(size),
3239- lru));
3240+ lru, have_page_hash_mutex));
b4e1fa2c
AM		 3241 }
3242
3243 /**********************************************************************//**
db82db79
AM		 3244@@ -126,15 +129,28 @@
3245 the block resides */
3246 void* buf, /*!< in: block to be freed, must not
3247 be pointed to by the buffer pool */
3248- ulint size) /*!< in: block size,
3249+ ulint size, /*!< in: block size,
3250 up to UNIV_PAGE_SIZE */
b4e1fa2c
AM		 3251+ ibool have_page_hash_mutex)
3252 {
3253- ut_ad(buf_pool_mutex_own(buf_pool));
3254+ //ut_ad(buf_pool_mutex_own(buf_pool));
db82db79
AM		 3255 ut_ad(ut_is_2pow(size));
3256 ut_ad(size >= PAGE_ZIP_MIN_SIZE);
3257 ut_ad(size <= UNIV_PAGE_SIZE);
3258
3259- buf_buddy_free_low(buf_pool, buf, buf_buddy_get_slot(size));
b4e1fa2c
AM		 3260+ if (!have_page_hash_mutex) {
3261+ mutex_enter(&buf_pool->LRU_list_mutex);
3262+ rw_lock_x_lock(&buf_pool->page_hash_latch);
3263+ }
db82db79 3264+
b4e1fa2c
AM		 3265+ mutex_enter(&buf_pool->zip_free_mutex);
3266+ buf_buddy_free_low(buf_pool, buf, buf_buddy_get_slot(size), TRUE);
3267+ mutex_exit(&buf_pool->zip_free_mutex);
3268+
3269+ if (!have_page_hash_mutex) {
3270+ mutex_exit(&buf_pool->LRU_list_mutex);
3271+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
3272+ }
3273 }
3274
3275 #ifdef UNIV_MATERIALIZE
db82db79
AM		 3276--- a/storage/innobase/include/buf0buf.h
3277+++ b/storage/innobase/include/buf0buf.h
734d6226 3278@@ -212,6 +212,20 @@
b4e1fa2c
AM		 3279 /*==========================*/
3280
3281 /********************************************************************//**
3282+*/
3283+UNIV_INLINE
3284+void
3285+buf_pool_page_hash_x_lock_all(void);
3286+/*================================*/
3287+
3288+/********************************************************************//**
3289+*/
3290+UNIV_INLINE
3291+void
3292+buf_pool_page_hash_x_unlock_all(void);
3293+/*==================================*/
3294+
3295+/********************************************************************//**
3296 Creates the buffer pool.
3297 @return own: buf_pool object, NULL if not enough memory or error */
3298 UNIV_INTERN
13ceb006 3299@@ -851,6 +865,15 @@
b4e1fa2c
AM		 3300 const buf_page_t* bpage) /*!< in: pointer to control block */
3301 __attribute__((pure));
3302
3303+/*************************************************************************
3304+Gets the mutex of a block and enter the mutex with consistency. */
3305+UNIV_INLINE
3306+mutex_t*
3307+buf_page_get_mutex_enter(
3308+/*=========================*/
3309+ const buf_page_t* bpage) /*!< in: pointer to control block */
3310+ __attribute__((pure));
3311+
3312 /*********************************************************************//**
3313 Get the flush type of a page.
3314 @return flush type */
13ceb006 3315@@ -1332,7 +1355,7 @@
b4e1fa2c
AM		 3316 All these are protected by buf_pool->mutex. */
3317 /* @{ */
3318
3319- UT_LIST_NODE_T(buf_page_t) list;
3320+ /* UT_LIST_NODE_T(buf_page_t) list; */
3321 /*!< based on state, this is a
3322 list node, protected either by
3323 buf_pool->mutex or by
13ceb006 3324@@ -1360,6 +1383,10 @@
b4e1fa2c
AM		 3325 BUF_BLOCK_REMOVE_HASH or
3326 BUF_BLOCK_READY_IN_USE. */
3327
3328+ /* resplit for optimistic use */
3329+ UT_LIST_NODE_T(buf_page_t) free;
3330+ UT_LIST_NODE_T(buf_page_t) flush_list;
3331+ UT_LIST_NODE_T(buf_page_t) zip_list; /* zip_clean or zip_free[] */
3332 #ifdef UNIV_DEBUG
3333 ibool in_flush_list; /*!< TRUE if in buf_pool->flush_list;
3334 when buf_pool->flush_list_mutex is
13ceb006 3335@@ -1452,11 +1479,11 @@
b4e1fa2c
AM		 3336 a block is in the unzip_LRU list
3337 if page.state == BUF_BLOCK_FILE_PAGE
3338 and page.zip.data != NULL */
3339-#ifdef UNIV_DEBUG
3340+//#ifdef UNIV_DEBUG
3341 ibool in_unzip_LRU_list;/*!< TRUE if the page is in the
3342 decompressed LRU list;
3343 used in debugging */
3344-#endif /* UNIV_DEBUG */
3345+//#endif /* UNIV_DEBUG */
3346 mutex_t mutex; /*!< mutex protecting this block:
3347 state (also protected by the buffer
3348 pool mutex), io_fix, buf_fix_count,
13ceb006 3349@@ -1636,6 +1663,11 @@
b4e1fa2c
AM		 3350 pool instance, protects compressed
3351 only pages (of type buf_page_t, not
3352 buf_block_t */
3353+ mutex_t LRU_list_mutex;
3354+ rw_lock_t page_hash_latch;
3355+ mutex_t free_list_mutex;
3356+ mutex_t zip_free_mutex;
3357+ mutex_t zip_hash_mutex;
3358 ulint instance_no; /*!< Array index of this buffer
3359 pool instance */
3360 ulint old_pool_size; /*!< Old pool size in bytes */
13ceb006 3361@@ -1789,8 +1821,8 @@
11822e22
AM		 3362 /** Test if a buffer pool mutex is owned. */
3363 #define buf_pool_mutex_own(b) mutex_own(&b->mutex)
3364 /** Acquire a buffer pool mutex. */
3365+/* the buf_pool_mutex is changed the latch order */
3366 #define buf_pool_mutex_enter(b) do { \
3367- ut_ad(!mutex_own(&b->zip_mutex)); \
3368 mutex_enter(&b->mutex); \
3369 } while (0)
3370
db82db79
AM		 3371--- a/storage/innobase/include/buf0buf.ic
3372+++ b/storage/innobase/include/buf0buf.ic
734d6226 3373@@ -292,7 +292,7 @@
b4e1fa2c
AM		 3374 case BUF_BLOCK_ZIP_FREE:
3375 /* This is a free page in buf_pool->zip_free[].
3376 Such pages should only be accessed by the buddy allocator. */
3377- ut_error;
3378+ /* ut_error; */ /* optimistic */
3379 break;
3380 case BUF_BLOCK_ZIP_PAGE:
3381 case BUF_BLOCK_ZIP_DIRTY:
734d6226 3382@@ -335,9 +335,16 @@
b4e1fa2c
AM		 3383 {
3384 buf_pool_t* buf_pool = buf_pool_from_bpage(bpage);
3385
11822e22
AM		 3386+ if (/*equivalent to buf_pool_watch_is_sentinel(buf_pool, bpage)*/
3387+ bpage >= &buf_pool->watch[0]
3388+ && bpage < &buf_pool->watch[BUF_POOL_WATCH_SIZE]) {
b4e1fa2c
AM		 3389+ /* TODO: this code is the interim. should be confirmed later. */
3390+ return(&buf_pool->zip_mutex);
3391+ }
3392+
3393 switch (buf_page_get_state(bpage)) {
3394 case BUF_BLOCK_ZIP_FREE:
3395- ut_error;
3396+ /* ut_error; */ /* optimistic */
3397 return(NULL);
3398 case BUF_BLOCK_ZIP_PAGE:
3399 case BUF_BLOCK_ZIP_DIRTY:
734d6226 3400@@ -347,6 +354,28 @@
b4e1fa2c
AM		 3401 }
3402 }
3403
3404+/*************************************************************************
3405+Gets the mutex of a block and enter the mutex with consistency. */
3406+UNIV_INLINE
3407+mutex_t*
3408+buf_page_get_mutex_enter(
3409+/*=========================*/
3410+ const buf_page_t* bpage) /*!< in: pointer to control block */
3411+{
3412+ mutex_t* block_mutex;
3413+
3414+ while(1) {
3415+ block_mutex = buf_page_get_mutex(bpage);
3416+ if (!block_mutex)
3417+ return block_mutex;
3418+
3419+ mutex_enter(block_mutex);
3420+ if (block_mutex == buf_page_get_mutex(bpage))
3421+ return block_mutex;
3422+ mutex_exit(block_mutex);
3423+ }
3424+}
3425+
3426 /*********************************************************************//**
3427 Get the flush type of a page.
3428 @return flush type */
734d6226 3429@@ -443,8 +472,8 @@
b4e1fa2c
AM		 3430 enum buf_io_fix io_fix) /*!< in: io_fix state */
3431 {
3432 #ifdef UNIV_DEBUG
3433- buf_pool_t* buf_pool = buf_pool_from_bpage(bpage);
3434- ut_ad(buf_pool_mutex_own(buf_pool));
3435+ //buf_pool_t* buf_pool = buf_pool_from_bpage(bpage);
3436+ //ut_ad(buf_pool_mutex_own(buf_pool));
3437 #endif
3438 ut_ad(mutex_own(buf_page_get_mutex(bpage)));
3439
734d6226 3440@@ -474,14 +503,14 @@
b4e1fa2c
AM		 3441 const buf_page_t* bpage) /*!< control block being relocated */
3442 {
3443 #ifdef UNIV_DEBUG
3444- buf_pool_t* buf_pool = buf_pool_from_bpage(bpage);
3445- ut_ad(buf_pool_mutex_own(buf_pool));
3446+ //buf_pool_t* buf_pool = buf_pool_from_bpage(bpage);
3447+ //ut_ad(buf_pool_mutex_own(buf_pool));
3448 #endif
3449 ut_ad(mutex_own(buf_page_get_mutex(bpage)));
3450 ut_ad(buf_page_in_file(bpage));
3451- ut_ad(bpage->in_LRU_list);
3452+ //ut_ad(bpage->in_LRU_list);
3453
3454- return(buf_page_get_io_fix(bpage) == BUF_IO_NONE
3455+ return(bpage->in_LRU_list && bpage->io_fix == BUF_IO_NONE
3456 && bpage->buf_fix_count == 0);
3457 }
3458
734d6226 3459@@ -495,8 +524,8 @@
b4e1fa2c
AM		 3460 const buf_page_t* bpage) /*!< in: control block */
3461 {
3462 #ifdef UNIV_DEBUG
3463- buf_pool_t* buf_pool = buf_pool_from_bpage(bpage);
3464- ut_ad(buf_pool_mutex_own(buf_pool));
3465+ //buf_pool_t* buf_pool = buf_pool_from_bpage(bpage);
3466+ //ut_ad(buf_pool_mutex_own(buf_pool));
3467 #endif
3468 ut_ad(buf_page_in_file(bpage));
3469
734d6226 3470@@ -516,7 +545,8 @@
b4e1fa2c
AM		 3471 buf_pool_t* buf_pool = buf_pool_from_bpage(bpage);
3472 #endif /* UNIV_DEBUG */
3473 ut_a(buf_page_in_file(bpage));
3474- ut_ad(buf_pool_mutex_own(buf_pool));
3475+ //ut_ad(buf_pool_mutex_own(buf_pool));
3476+ ut_ad(mutex_own(&buf_pool->LRU_list_mutex));
3477 ut_ad(bpage->in_LRU_list);
3478
3479 #ifdef UNIV_LRU_DEBUG
734d6226 3480@@ -563,9 +593,10 @@
b4e1fa2c
AM		 3481 ulint time_ms) /*!< in: ut_time_ms() */
3482 {
3483 #ifdef UNIV_DEBUG
3484- buf_pool_t* buf_pool = buf_pool_from_bpage(bpage);
3485- ut_ad(buf_pool_mutex_own(buf_pool));
3486+ //buf_pool_t* buf_pool = buf_pool_from_bpage(bpage);
3487+ //ut_ad(buf_pool_mutex_own(buf_pool));
3488 #endif
3489+ ut_ad(mutex_own(buf_page_get_mutex(bpage)));
3490 ut_a(buf_page_in_file(bpage));
3491
3492 if (!bpage->access_time) {
734d6226 3493@@ -808,19 +839,19 @@
b4e1fa2c
AM		 3494 /*===========*/
3495 buf_block_t* block) /*!< in, own: block to be freed */
3496 {
3497- buf_pool_t* buf_pool = buf_pool_from_bpage((buf_page_t*)block);
3498+ //buf_pool_t* buf_pool = buf_pool_from_bpage((buf_page_t*)block);
3499
3500- buf_pool_mutex_enter(buf_pool);
3501+ //buf_pool_mutex_enter(buf_pool);
3502
3503 mutex_enter(&block->mutex);
3504
3505 ut_a(buf_block_get_state(block) != BUF_BLOCK_FILE_PAGE);
3506
3507- buf_LRU_block_free_non_file_page(block);
3508+ buf_LRU_block_free_non_file_page(block, FALSE);
3509
3510 mutex_exit(&block->mutex);
3511
3512- buf_pool_mutex_exit(buf_pool);
3513+ //buf_pool_mutex_exit(buf_pool);
3514 }
3515 #endif /* !UNIV_HOTBACKUP */
3516
734d6226 3517@@ -868,17 +899,17 @@
b4e1fa2c
AM		 3518 page frame */
3519 {
3520 ib_uint64_t lsn;
3521- mutex_t* block_mutex = buf_page_get_mutex(bpage);
3522-
3523- mutex_enter(block_mutex);
3524+ mutex_t* block_mutex = buf_page_get_mutex_enter(bpage);
3525
3526- if (buf_page_in_file(bpage)) {
3527+ if (block_mutex && buf_page_in_file(bpage)) {
3528 lsn = bpage->newest_modification;
3529 } else {
3530 lsn = 0;
3531 }
3532
3533- mutex_exit(block_mutex);
3534+ if (block_mutex) {
3535+ mutex_exit(block_mutex);
3536+ }
3537
3538 return(lsn);
3539 }
734d6226 3540@@ -896,7 +927,7 @@
b4e1fa2c
AM		 3541 #ifdef UNIV_SYNC_DEBUG
3542 buf_pool_t* buf_pool = buf_pool_from_bpage((buf_page_t*)block);
3543
3544- ut_ad((buf_pool_mutex_own(buf_pool)
3545+ ut_ad((mutex_own(&buf_pool->LRU_list_mutex)
3546 && (block->page.buf_fix_count == 0))
3547 || rw_lock_own(&(block->lock), RW_LOCK_EXCLUSIVE));
3548 #endif /* UNIV_SYNC_DEBUG */
734d6226 3549@@ -1026,7 +1057,11 @@
b4e1fa2c
AM		 3550 buf_page_t* bpage;
3551
3552 ut_ad(buf_pool);
3553- ut_ad(buf_pool_mutex_own(buf_pool));
3554+ //ut_ad(buf_pool_mutex_own(buf_pool));
3555+#ifdef UNIV_SYNC_DEBUG
3556+ ut_ad(rw_lock_own(&buf_pool->page_hash_latch, RW_LOCK_EX)
3557+ || rw_lock_own(&buf_pool->page_hash_latch, RW_LOCK_SHARED));
3558+#endif
3559 ut_ad(fold == buf_page_address_fold(space, offset));
3560
3561 /* Look for the page in the hash table */
734d6226 3562@@ -1111,11 +1146,13 @@
b4e1fa2c
AM		 3563 const buf_page_t* bpage;
3564 buf_pool_t* buf_pool = buf_pool_get(space, offset);
3565
3566- buf_pool_mutex_enter(buf_pool);
3567+ //buf_pool_mutex_enter(buf_pool);
3568+ rw_lock_s_lock(&buf_pool->page_hash_latch);
3569
3570 bpage = buf_page_hash_get(buf_pool, space, offset);
3571
3572- buf_pool_mutex_exit(buf_pool);
3573+ //buf_pool_mutex_exit(buf_pool);
3574+ rw_lock_s_unlock(&buf_pool->page_hash_latch);
3575
3576 return(bpage != NULL);
3577 }
734d6226 3578@@ -1243,4 +1280,38 @@
b4e1fa2c
AM		 3579 buf_pool_mutex_exit(buf_pool);
3580 }
3581 }
3582+
3583+/********************************************************************//**
3584+*/
3585+UNIV_INLINE
3586+void
3587+buf_pool_page_hash_x_lock_all(void)
3588+/*===============================*/
3589+{
3590+ ulint i;
3591+
3592+ for (i = 0; i < srv_buf_pool_instances; i++) {
3593+ buf_pool_t* buf_pool;
3594+
3595+ buf_pool = buf_pool_from_array(i);
3596+ rw_lock_x_lock(&buf_pool->page_hash_latch);
3597+ }
3598+}
3599+
3600+/********************************************************************//**
3601+*/
3602+UNIV_INLINE
3603+void
3604+buf_pool_page_hash_x_unlock_all(void)
3605+/*=================================*/
3606+{
3607+ ulint i;
3608+
3609+ for (i = 0; i < srv_buf_pool_instances; i++) {
3610+ buf_pool_t* buf_pool;
3611+
3612+ buf_pool = buf_pool_from_array(i);
3613+ rw_lock_x_unlock(&buf_pool->page_hash_latch);
3614+ }
3615+}
3616 #endif /* !UNIV_HOTBACKUP */
db82db79
AM		 3617--- a/storage/innobase/include/buf0lru.h
3618+++ b/storage/innobase/include/buf0lru.h
3619@@ -100,8 +100,9 @@
df1b5770
AM		 3620 buf_LRU_free_block(
3621 /*===============*/
b4e1fa2c 3622 buf_page_t* bpage, /*!< in: block to be freed */
df1b5770
AM		 3623- ibool zip) /*!< in: TRUE if should remove also the
3624+ ibool zip, /*!< in: TRUE if should remove also the
b4e1fa2c 3625 compressed page of an uncompressed page */
df1b5770
AM		 3626+ ibool have_LRU_mutex)
3627 __attribute__((nonnull));
b4e1fa2c
AM		 3628 /******************************************************************//**
3629 Try to free a replaceable block.
db82db79 3630@@ -148,7 +149,8 @@
b4e1fa2c
AM		 3631 void
3632 buf_LRU_block_free_non_file_page(
3633 /*=============================*/
3634- buf_block_t* block); /*!< in: block, must not contain a file page */
3635+ buf_block_t* block, /*!< in: block, must not contain a file page */
3636+ ibool have_page_hash_mutex);
3637 /******************************************************************//**
3638 Adds a block to the LRU list. */
3639 UNIV_INTERN
db82db79
AM		 3640--- a/storage/innobase/include/sync0rw.h
3641+++ b/storage/innobase/include/sync0rw.h
b4e1fa2c
AM		 3642@@ -112,6 +112,7 @@
3643 extern mysql_pfs_key_t archive_lock_key;
3644 # endif /* UNIV_LOG_ARCHIVE */
3645 extern mysql_pfs_key_t btr_search_latch_key;
3646+extern mysql_pfs_key_t buf_pool_page_hash_key;
3647 extern mysql_pfs_key_t buf_block_lock_key;
3648 # ifdef UNIV_SYNC_DEBUG
3649 extern mysql_pfs_key_t buf_block_debug_latch_key;
db82db79
AM		 3650--- a/storage/innobase/include/sync0sync.h
3651+++ b/storage/innobase/include/sync0sync.h
b4e1fa2c
AM		 3652@@ -75,6 +75,10 @@
3653 extern mysql_pfs_key_t buffer_block_mutex_key;
3654 extern mysql_pfs_key_t buf_pool_mutex_key;
3655 extern mysql_pfs_key_t buf_pool_zip_mutex_key;
3656+extern mysql_pfs_key_t buf_pool_LRU_list_mutex_key;
3657+extern mysql_pfs_key_t buf_pool_free_list_mutex_key;
3658+extern mysql_pfs_key_t buf_pool_zip_free_mutex_key;
3659+extern mysql_pfs_key_t buf_pool_zip_hash_mutex_key;
3660 extern mysql_pfs_key_t cache_last_read_mutex_key;
3661 extern mysql_pfs_key_t dict_foreign_err_mutex_key;
3662 extern mysql_pfs_key_t dict_sys_mutex_key;
734d6226 3663@@ -667,7 +671,7 @@
b4e1fa2c 3664 #define SYNC_TRX_SYS_HEADER 290
11822e22 3665 #define SYNC_PURGE_QUEUE 200
b4e1fa2c
AM		 3666 #define SYNC_LOG 170
3667-#define SYNC_LOG_FLUSH_ORDER 147
3668+#define SYNC_LOG_FLUSH_ORDER 156
3669 #define SYNC_RECV 168
3670 #define SYNC_WORK_QUEUE 162
13ceb006
AM		 3671 #define SYNC_SEARCH_SYS 160 /* NOTE that if we have a memory
3672@@ -676,8 +680,13 @@
b4e1fa2c
AM		 3673 SYNC_SEARCH_SYS, as memory allocation
3674 can call routines there! Otherwise
3675 the level is SYNC_MEM_HASH. */
3676+#define SYNC_BUF_LRU_LIST 158
3677+#define SYNC_BUF_PAGE_HASH 157
3678+#define SYNC_BUF_BLOCK 155 /* Block mutex */
3679+#define SYNC_BUF_FREE_LIST 153
3680+#define SYNC_BUF_ZIP_FREE 152
3681+#define SYNC_BUF_ZIP_HASH 151
3682 #define SYNC_BUF_POOL 150 /* Buffer pool mutex */
3683-#define SYNC_BUF_BLOCK 146 /* Block mutex */
3684 #define SYNC_BUF_FLUSH_LIST 145 /* Buffer flush list mutex */
3685 #define SYNC_DOUBLEWRITE 140
3686 #define SYNC_ANY_LATCH 135
13ceb006 3687@@ -708,7 +717,7 @@
b4e1fa2c
AM		 3688 os_fast_mutex; /*!< We use this OS mutex in place of lock_word
3689 when atomic operations are not enabled */
3690 #endif
3691- ulint waiters; /*!< This ulint is set to 1 if there are (or
3692+ volatile ulint waiters; /*!< This ulint is set to 1 if there are (or
3693 may be) threads waiting in the global wait
3694 array for this mutex to be released.
3695 Otherwise, this is 0. */
db82db79
AM		 3696--- a/storage/innobase/srv/srv0srv.c
3697+++ b/storage/innobase/srv/srv0srv.c
3d3ecf24 3698@@ -3105,7 +3105,7 @@
b4e1fa2c
AM		 3699 level += log_sys->max_checkpoint_age
3700 - (lsn - oldest_modification);
3701 }
3702- bpage = UT_LIST_GET_NEXT(list, bpage);
3703+ bpage = UT_LIST_GET_NEXT(flush_list, bpage);
3704 n_blocks++;
3705 }
3706
3d3ecf24 3707@@ -3191,7 +3191,7 @@
b4e1fa2c
AM		 3708 found = TRUE;
3709 break;
3710 }
3711- bpage = UT_LIST_GET_NEXT(list, bpage);
3712+ bpage = UT_LIST_GET_NEXT(flush_list, bpage);
3713 new_blocks_num++;
3714 }
3715 if (!found) {
db82db79
AM		 3716--- a/storage/innobase/sync/sync0sync.c
3717+++ b/storage/innobase/sync/sync0sync.c
adf0fb13 3718@@ -285,7 +285,7 @@
b4e1fa2c
AM		 3719 mutex->lock_word = 0;
3720 #endif
3721 mutex->event = os_event_create(NULL);
3722- mutex_set_waiters(mutex, 0);
3723+ mutex->waiters = 0;
3724 #ifdef UNIV_DEBUG
3725 mutex->magic_n = MUTEX_MAGIC_N;
3726 #endif /* UNIV_DEBUG */
adf0fb13 3727@@ -464,6 +464,15 @@
b4e1fa2c
AM		 3728 mutex_t* mutex, /*!< in: mutex */
3729 ulint n) /*!< in: value to set */
3730 {
3731+#ifdef INNODB_RW_LOCKS_USE_ATOMICS
3732+ ut_ad(mutex);
3733+
3734+ if (n) {
3735+ os_compare_and_swap_ulint(&mutex->waiters, 0, 1);
3736+ } else {
3737+ os_compare_and_swap_ulint(&mutex->waiters, 1, 0);
3738+ }
3739+#else
3740 volatile ulint* ptr; /* declared volatile to ensure that
3741 the value is stored to memory */
3742 ut_ad(mutex);
adf0fb13 3743@@ -472,6 +481,7 @@
b4e1fa2c
AM		 3744
3745 *ptr = n; /* Here we assume that the write of a single
3746 word in memory is atomic */
3747+#endif
3748 }
3749
3750 /******************************************************************//**
13ceb006 3751@@ -1233,7 +1243,12 @@
b4e1fa2c
AM		 3752 ut_error;
3753 }
3754 break;
3755+ case SYNC_BUF_LRU_LIST:
3756 case SYNC_BUF_FLUSH_LIST:
3757+ case SYNC_BUF_PAGE_HASH:
3758+ case SYNC_BUF_FREE_LIST:
3759+ case SYNC_BUF_ZIP_FREE:
3760+ case SYNC_BUF_ZIP_HASH:
3761 case SYNC_BUF_POOL:
3762 /* We can have multiple mutexes of this type therefore we
3763 can only check whether the greater than condition holds. */
13ceb006 3764@@ -1251,7 +1266,8 @@
b4e1fa2c
AM		 3765 buffer block (block->mutex or buf_pool->zip_mutex). */
3766 if (!sync_thread_levels_g(array, level, FALSE)) {
3767 ut_a(sync_thread_levels_g(array, level - 1, TRUE));
3768- ut_a(sync_thread_levels_contain(array, SYNC_BUF_POOL));
3769+ /* the exact rule is not fixed yet, for now */
3770+ //ut_a(sync_thread_levels_contain(array, SYNC_BUF_LRU_LIST));
3771 }
3772 break;
3773 case SYNC_REC_LOCK:
MySQL: a very fast and reliable SQL database engine
This page took 0.56885 seconds and 4 git commands to generate.