版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
Page磁盘读取和加载的互斥
buf_page_get_gen
| | ==> rw_lock_s_lock(hash_lock);
| | ==> block = (buf_block_t*) buf_page_hash_get_low(buf_pool, space, offset, fold);
| | ==> if (block == NULL) rw_lock_s_unlock(hash_lock);
| | ==> buf_read_page(space, zip_size, offset) buf_read_page_low
| | ==> bpage = buf_page_init_for_read(err, mode, space, zip_size, unzip, tablespace_version, offset);
| | ==> block = NULL;
| | ==> buf_pool_mutex_enter(buf_pool);
| | ==> rw_lock_x_lock(hash_lock);
| | ==> watch_page = buf_page_hash_get_low(buf_pool, space, offset, fold);
| | ==> if (watch_page && !buf_pool_watch_is_sentinel(buf_pool, watch_page))
| | ==> rw_lock_x_unlock(hash_lock);
| | ==> rw_lock_x_unlock(hash_lock);
| | ==> data = buf_buddy_alloc(buf_pool, zip_size, &lru);
| | ==> buf_buddy_alloc_low(buf_pool, buf_buddy_get_slot(size), lru));
| | ==> block = buf_LRU_get_free_only(buf_pool);
| | ==> rw_lock_x_lock(hash_lock);
| | ==> /* If buf_buddy_alloc() allocated storage from the LRU list, it released and reacquired buf_pool->mutex. Thus, we must check the page_hash again, as it may have been modified. */
| | ==> if (UNIV_UNLIKELY(lru)) {
| | ==> watch_page = buf_page_hash_get_low(buf_pool, space, offset, fold);
| | ==> if (watch_page && !buf_pool_watch_is_sentinel(buf_pool, watch_page)) rw_lock_x_unlock(hash_lock); watch_page = NULL; buf_buddy_free(buf_pool, data, zip_size);
| | ==> HASH_INSERT(buf_page_t, hash, buf_pool->page_hash, fold, bpage);
| | ==> rw_lock_x_unlock(hash_lock);
| | ==> buf_LRU_add_block(bpage, TRUE/* to old blocks */);
| | ==> buf_pool_mutex_exit(buf_pool);
| | ==> buf_block_fix(fix_block);
| | ==> #ifdef PAGE_ATOMIC_REF_COUNT
| | ==> os_atomic_increment_uint32(&block->page.buf_fix_count, 1);
| | ==> #else
| | ==> ib_mutex_t* block_mutex = buf_page_get_mutex(&block->page);
| | ==> mutex_enter(block_mutex);
| | ==> ++block->page.buf_fix_count;
| | ==> mutex_exit(block_mutex);
| | ==> rw_lock_s_unlock(hash_lock);
| | ==> buf_wait_for_read(fix_block);它的过程是：
- 对Page_Hash加s锁，判断是否存在。
- 如果不存在，释放s锁，加bp的互斥锁、加x锁。再次判断是否存在。
- 如果不存在，释放x锁，申请空闲Block。
- 然后加x锁，判断是否存在，如果不存在，插入空闲BLock。释放x锁
- 把block加入lru链表。释放bp的互斥锁。
步骤3释放x锁的原因是，线程持有bp的互斥锁，对lru\freelist的操作都依靠互斥锁。因此可以通过bp的互斥锁，阻塞其他线程申请空闲Block。但是在从Lru申请BLock过程中，可能出现由于休眠等待需要释放互斥锁，而导致其他线程进入，因此第4步根据空闲BLock的来源判断是否需要重新Check。
可以看到，从磁盘读取过程中，对hash的互斥成本是比较高的，加了2次x锁。这个地方后续版本中有优化。
Page内存读取和Page淘汰的互斥
buf_flush_batch
| | ==> //加LRU LOCK
| | ==> buf_pool_mutex_enter(buf_pool);
| | ==> count = buf_do_LRU_batch(buf_pool, min_n);
| | ==> buf_flush_LRU_list_batch(buf_pool, max - count);
| | ==> bpage = UT_LIST_GET_LAST(buf_pool->LRU);
| | ==> mutex_enter(block_mutex);
| | ==> evict = buf_flush_ready_for_replace(bpage);
| | ==> bpage->buf_fix_count == 0
| | ==> mutex_exit(block_mutex);
| | ==> if (evict) buf_LRU_free_page
| | ==> rw_lock_x_lock(hash_lock);
| | ==> mutex_enter(block_mutex);
| | ==> //重新判断引用计数
| | ==> buf_page_can_relocate
| | ==> //从LRU和HASH表删除
| | ==> buf_LRU_block_remove_hashed
| | ==> buf_LRU_remove_block(bpage);
| | ==> HASH_DELETE(buf_page_t, hash, buf_pool->page_hash, fold, bpage);
| | ==> rw_lock_x_unlock(hash_lock);
| | ==> buf_pool_mutex_exit(buf_pool);
| | ==> buf_pool_mutex_enter(buf_pool);
| | ==> buf_LRU_block_free_hashed_page((buf_block_t*) bpage);
| | ==> buf_LRU_block_free_non_file_page(block);
| | ==> UT_LIST_ADD_FIRST(list, buf_pool->free, (&block->page));
| | ==> buf_pool_mutex_exit(buf_pool);
Page淘汰过程如下：
- 加BP的互斥锁
- 从LRU链表的获取最老的数据页
- 确认Page是否可以被淘汰
- 如果可以淘汰，加hash_page的x锁，此时重新判断，因为前述判断无法拦截对Page的访问，因此此时对Page_hash加x锁，进行准确的判断。
- 如果确实可以淘汰，从Page_Hash中删除，释放hash的x锁。
- 将Page加入空闲链表