Merge 10.5 into 10.6

include/my_atomic_wrapper.h

@@ -43,9 +43,10 @@ template <typename Type> class Atomic_relaxed
 
   Type load(std::memory_order o= std::memory_order_relaxed) const
   { return m.load(o); }
+  void store(Type i, std::memory_order o= std::memory_order_relaxed)
+  { m.store(i, o); }
   operator Type() const { return m.load(); }
-  Type operator=(const Type val)
-  { m.store(val, std::memory_order_relaxed); return val; }
+  Type operator=(const Type i) { store(i); return i; }
   Type operator=(const Atomic_relaxed<Type> &rhs) { return *this= Type{rhs}; }
   Type operator+=(const Type i) { return fetch_add(i); }
   Type fetch_add(const Type i, std::memory_order o= std::memory_order_relaxed)

storage/innobase/buf/buf0buf.cc

@@ -1095,21 +1095,19 @@ inline const buf_block_t *buf_pool_t::chunk_t::not_freed() const
       /* Skip blocks that are not being used for file pages. */
       break;
     case BUF_BLOCK_FILE_PAGE:
+      const lsn_t lsn= block->page.oldest_modification();
+
       if (srv_read_only_mode)
       {
         /* The page cleaner is disabled in read-only mode.  No pages
         can be dirtied, so all of them must be clean. */
-        ut_d(lsn_t oldest_modification= block->page.oldest_modification());
-        ut_ad(oldest_modification == 0 ||
-              oldest_modification == recv_sys.recovered_lsn ||
+        ut_ad(lsn == 0 || lsn == recv_sys.recovered_lsn ||
               srv_force_recovery == SRV_FORCE_NO_LOG_REDO);
         ut_ad(!block->page.buf_fix_count());
         ut_ad(block->page.io_fix() == BUF_IO_NONE);
         break;
       }
 
-      const lsn_t lsn= block->page.oldest_modification();
-
       if (fsp_is_system_temporary(block->page.id().space()))
       {
         ut_ad(lsn == 0 || lsn == 2);

storage/innobase/buf/buf0flu.cc

@@ -292,7 +292,7 @@ buf_flush_relocate_on_flush_list(
 	mysql_mutex_assert_owner(&buf_pool.mutex);
 	ut_ad(!fsp_is_system_temporary(bpage->id().space()));
 
-	const lsn_t lsn = bpage->oldest_modification();
+	const lsn_t lsn = bpage->oldest_modification_acquire();
 
 	if (!lsn) {
 		return;
@@ -320,13 +320,14 @@ buf_flush_relocate_on_flush_list(
 		/* bpage was removed from buf_pool.flush_list
 		since we last checked, and before we acquired
 		buf_pool.flush_list_mutex. */
-		dpage->list.prev = nullptr;
-		dpage->list.next = nullptr;
 		goto was_clean;
 	}
 
 	if (lsn == 1) {
+		buf_pool.stat.flush_list_bytes -= bpage->physical_size();
 was_clean:
+		dpage->list.prev = nullptr;
+		dpage->list.next = nullptr;
 		dpage->clear_oldest_modification();
 	} else if (prev) {
 		ut_ad(prev->oldest_modification());
@@ -770,14 +771,15 @@ inline void buf_pool_t::release_freed_page(buf_page_t *bpage)
   bpage->set_io_fix(BUF_IO_NONE);
   bpage->status= buf_page_t::NORMAL;
   mysql_mutex_lock(&flush_list_mutex);
+  ut_d(const lsn_t oldest_modification= bpage->oldest_modification();)
   if (fsp_is_system_temporary(bpage->id().space()))
   {
     ut_ad(uncompressed);
-    ut_ad(bpage->oldest_modification() == 2);
+    ut_ad(oldest_modification == 2);
   }
   else
   {
-    ut_ad(bpage->oldest_modification() > 2);
+    ut_ad(oldest_modification > 2);
     delete_from_flush_list(bpage, false);
   }
   bpage->clear_oldest_modification();
@@ -805,6 +807,7 @@ static bool buf_flush_page(buf_page_t *bpage, bool lru, fil_space_t *space)
   ut_ad(space->purpose == FIL_TYPE_TABLESPACE ||
         space->atomic_write_supported);
   ut_ad(space->referenced());
+  ut_ad(lru || space != fil_system.temp_space);
 
   block_lock *rw_lock;
 
@@ -848,7 +851,10 @@ static bool buf_flush_page(buf_page_t *bpage, bool lru, fil_space_t *space)
                         lru ? "LRU" : "flush_list",
                         bpage->id().space(), bpage->id().page_no()));
   ut_ad(bpage->io_fix() == BUF_IO_WRITE);
-  ut_ad(bpage->oldest_modification());
+  ut_d(const lsn_t oldest_modification= bpage->oldest_modification());
+  ut_ad(space == fil_system.temp_space
+        ? oldest_modification == 2
+        : oldest_modification > 2);
   ut_ad(bpage->state() ==
         (rw_lock ? BUF_BLOCK_FILE_PAGE : BUF_BLOCK_ZIP_PAGE));
   ut_ad(ULINT_UNDEFINED >
@@ -913,6 +919,7 @@ static bool buf_flush_page(buf_page_t *bpage, bool lru, fil_space_t *space)
     }
 
     ut_ad(status == bpage->status);
+    ut_ad(oldest_modification == bpage->oldest_modification());
 
     if (status != buf_page_t::NORMAL || !space->use_doublewrite())
     {
@@ -921,8 +928,7 @@ static bool buf_flush_page(buf_page_t *bpage, bool lru, fil_space_t *space)
         const lsn_t lsn= mach_read_from_8(my_assume_aligned<8>
                                           (FIL_PAGE_LSN + (frame ? frame
                                                            : block->frame)));
-        ut_ad(lsn);
-        ut_ad(lsn >= bpage->oldest_modification());
+        ut_ad(lsn >= oldest_modification);
         if (lsn > log_sys.get_flushed_lsn())
           log_write_up_to(lsn, true);
       }

storage/innobase/buf/buf0lru.cc

@@ -810,7 +810,7 @@ bool buf_LRU_free_page(buf_page_t *bpage, bool zip)
 	const ulint fold = id.fold();
 	page_hash_latch* hash_lock = buf_pool.page_hash.lock_get(fold);
 	hash_lock->write_lock();
-	lsn_t oldest_modification = bpage->oldest_modification();
+	lsn_t oldest_modification = bpage->oldest_modification_acquire();
 
 	if (UNIV_UNLIKELY(!bpage->can_relocate())) {
 		/* Do not free buffer fixed and I/O-fixed blocks. */

storage/innobase/include/buf0buf.h

@@ -680,7 +680,7 @@ class buf_page_t
   1 if no modifications are pending, but the block is in buf_pool.flush_list;
   2 if modifications are pending, but the block is not in buf_pool.flush_list
   (because id().space() is the temporary tablespace). */
-  Atomic_counter<lsn_t> oldest_modification_;
+  Atomic_relaxed<lsn_t> oldest_modification_;
 
   /** type of pending I/O operation; protected by buf_pool.mutex
   if in_LRU_list */
@@ -846,11 +846,18 @@ class buf_page_t
   inline void set_corrupt_id();
 
   /** @return the log sequence number of the oldest pending modification
-  @retval 0 if the block is not in buf_pool.flush_list
+  @retval 0 if the block is being removed from (or not in) buf_pool.flush_list
   @retval 1 if the block is in buf_pool.flush_list but not modified
   @retval 2 if the block belongs to the temporary tablespace and
   has unwritten changes */
   lsn_t oldest_modification() const { return oldest_modification_; }
+  /** @return the log sequence number of the oldest pending modification,
+  @retval 0 if the block is definitely not in buf_pool.flush_list
+  @retval 1 if the block is in buf_pool.flush_list but not modified
+  @retval 2 if the block belongs to the temporary tablespace and
+  has unwritten changes */
+  lsn_t oldest_modification_acquire() const
+  { return oldest_modification_.load(std::memory_order_acquire); }
   /** Set oldest_modification when adding to buf_pool.flush_list */
   inline void set_oldest_modification(lsn_t lsn);
   /** Clear oldest_modification after removing from buf_pool.flush_list */
@@ -872,7 +879,8 @@ class buf_page_t
   void free_file_page()
   {
     ut_ad(state() == BUF_BLOCK_REMOVE_HASH);
-    ut_d(oldest_modification_= 0); /* for buf_LRU_block_free_non_file_page() */
+    /* buf_LRU_block_free_non_file_page() asserts !oldest_modification() */
+    ut_d(oldest_modification_= 0;)
     set_corrupt_id();
     ut_d(set_state(BUF_BLOCK_MEMORY));
   }
@@ -2121,7 +2129,8 @@ inline void buf_page_t::set_corrupt_id()
     break;
   case 2:
     ut_ad(fsp_is_system_temporary(id().space()));
-    ut_d(oldest_modification_= 0); /* for buf_LRU_block_free_non_file_page() */
+    /* buf_LRU_block_free_non_file_page() asserts !oldest_modification() */
+    ut_d(oldest_modification_= 0;)
     break;
   default:
     ut_ad("block is dirty" == 0);
@@ -2158,7 +2167,12 @@ inline void buf_page_t::clear_oldest_modification()
   ut_ad(state == BUF_BLOCK_FILE_PAGE || state == BUF_BLOCK_ZIP_PAGE ||
         state == BUF_BLOCK_REMOVE_HASH);
   ut_ad(oldest_modification());
-  oldest_modification_= 0;
+  ut_ad(!list.prev);
+  ut_ad(!list.next);
+  /* We must use release memory order to guarantee that callers of
+  oldest_modification_acquire() will observe the block as
+  being detached from buf_pool.flush_list, after reading the value 0. */
+  oldest_modification_.store(0, std::memory_order_release);
 }
 
 /** Note that a block is no longer dirty, while not removing
@@ -2168,8 +2182,19 @@ inline void buf_page_t::clear_oldest_modification(bool temporary)
   mysql_mutex_assert_not_owner(&buf_pool.flush_list_mutex);
   ut_ad(temporary == fsp_is_system_temporary(id().space()));
   ut_ad(io_fix_ == BUF_IO_WRITE);
-  ut_ad(temporary ? oldest_modification() == 2 : oldest_modification() > 2);
-  oldest_modification_= !temporary;
+  if (temporary)
+  {
+    ut_ad(oldest_modification() == 2);
+    oldest_modification_= 0;
+  }
+  else
+  {
+    /* We use release memory order to guarantee that callers of
+    oldest_modification_acquire() will observe the block as
+    being detached from buf_pool.flush_list, after reading the value 0. */
+    ut_ad(oldest_modification() > 2);
+    oldest_modification_.store(1, std::memory_order_release);
+  }
 }
 
 /** @return whether the block is modified and ready for flushing */

storage/innobase/include/mtr0mtr.h

@@ -84,11 +84,6 @@ savepoint. */
 #define mtr_block_x_latch_at_savepoint(m, s, b)				\
 				(m)->x_latch_at_savepoint((s), (b))
 
-/** Check if a mini-transaction is dirtying a clean page.
-@param b	block being x-fixed
-@return true if the mtr is dirtying a clean page. */
-#define mtr_block_dirtied(b)	mtr_t::is_block_dirtied((b))
-
 /** Mini-transaction memo stack slot. */
 struct mtr_memo_slot_t {
 	/** pointer to the object */