5.6.27

storage/innobase/CMakeLists.txt

@@ -1,4 +1,4 @@
-# Copyright (c) 2006, 2011, Oracle and/or its affiliates. All rights reserved.
+# Copyright (c) 2006, 2015, Oracle and/or its affiliates. All rights reserved.
 #
 # This program is free software; you can redistribute it and/or modify
 # it under the terms of the GNU General Public License as published by
@@ -29,6 +29,9 @@ IF(UNIX)
       ADD_DEFINITIONS(-DLINUX_NATIVE_AIO=1)
       LINK_LIBRARIES(aio)
     ENDIF()
+    IF(HAVE_LIBNUMA)
+      LINK_LIBRARIES(numa)
+    ENDIF()
   ELSEIF(CMAKE_SYSTEM_NAME MATCHES "HP*")
     ADD_DEFINITIONS("-DUNIV_HPUX")
   ELSEIF(CMAKE_SYSTEM_NAME STREQUAL "AIX")
@@ -145,6 +148,18 @@ IF(NOT CMAKE_CROSSCOMPILING)
   }"
   HAVE_IB_GCC_ATOMIC_THREAD_FENCE
   )
+  CHECK_C_SOURCE_RUNS(
+  "#include<stdint.h>
+  int main()
+  {
+    unsigned char	c;
+
+    __atomic_test_and_set(&c, __ATOMIC_ACQUIRE);
+    __atomic_clear(&c, __ATOMIC_RELEASE);
+    return(0);
+  }"
+  HAVE_IB_GCC_ATOMIC_TEST_AND_SET
+  )
 ENDIF()
 
 IF(HAVE_IB_GCC_ATOMIC_BUILTINS)
@@ -167,6 +182,10 @@ IF(HAVE_IB_GCC_ATOMIC_THREAD_FENCE)
  ADD_DEFINITIONS(-DHAVE_IB_GCC_ATOMIC_THREAD_FENCE=1)
 ENDIF()
 
+IF(HAVE_IB_GCC_ATOMIC_TEST_AND_SET)
+ ADD_DEFINITIONS(-DHAVE_IB_GCC_ATOMIC_TEST_AND_SET=1)
+ENDIF()
+
  # either define HAVE_IB_ATOMIC_PTHREAD_T_GCC or not
 IF(NOT CMAKE_CROSSCOMPILING)
   CHECK_C_SOURCE_RUNS(

storage/innobase/btr/btr0cur.cc

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1994, 2014, Oracle and/or its affiliates. All Rights Reserved.
+Copyright (c) 1994, 2015, Oracle and/or its affiliates. All Rights Reserved.
 Copyright (c) 2008, Google Inc.
 Copyright (c) 2012, Facebook Inc.
 
@@ -2117,6 +2117,7 @@ btr_cur_optimistic_update(
 	ulint		max_size;
 	ulint		new_rec_size;
 	ulint		old_rec_size;
+	ulint		max_ins_size = 0;
 	dtuple_t*	new_entry;
 	roll_ptr_t	roll_ptr;
 	ulint		i;
@@ -2245,6 +2246,10 @@ any_extern:
 		: (old_rec_size
 		   + page_get_max_insert_size_after_reorganize(page, 1));
 
+	if (!page_zip) {
+		max_ins_size = page_get_max_insert_size_after_reorganize(page, 1);
+	}
+
 	if (!(((max_size >= BTR_CUR_PAGE_REORGANIZE_LIMIT)
 	       && (max_size >= new_rec_size))
 	      || (page_get_n_recs(page) <= 1))) {
@@ -2304,12 +2309,15 @@ any_extern:
 	ut_ad(err == DB_SUCCESS);
 
 func_exit:
-	if (page_zip
-	    && !(flags & BTR_KEEP_IBUF_BITMAP)
+	if (!(flags & BTR_KEEP_IBUF_BITMAP)
 	    && !dict_index_is_clust(index)
 	    && page_is_leaf(page)) {
-		/* Update the free bits in the insert buffer. */
-		ibuf_update_free_bits_zip(block, mtr);
+
+		if (page_zip) {
+			ibuf_update_free_bits_zip(block, mtr);
+		} else {
+			ibuf_update_free_bits_low(block, max_ins_size, mtr);
+		}
 	}
 
 	return(err);
@@ -2444,6 +2452,7 @@ btr_cur_pessimistic_update(
 	ibool		was_first;
 	ulint		n_reserved	= 0;
 	ulint		n_ext;
+	ulint		max_ins_size	= 0;
 
 	*offsets = NULL;
 	*big_rec = NULL;
@@ -2622,6 +2631,10 @@ make_external:
 		}
 	}
 
+	if (!page_zip) {
+		max_ins_size = page_get_max_insert_size_after_reorganize(page, 1);
+	}
+
 	/* Store state of explicit locks on rec on the page infimum record,
 	before deleting rec. The page infimum acts as a dummy carrier of the
 	locks, taking care also of lock releases, before we can move the locks
@@ -2667,13 +2680,18 @@ make_external:
 				rec_offs_make_valid(
 					page_cursor->rec, index, *offsets);
 			}
-		} else if (page_zip &&
-			   !dict_index_is_clust(index)
+		} else if (!dict_index_is_clust(index)
 			   && page_is_leaf(page)) {
+
 			/* Update the free bits in the insert buffer.
 			This is the same block which was skipped by
 			BTR_KEEP_IBUF_BITMAP. */
-			ibuf_update_free_bits_zip(block, mtr);
+			if (page_zip) {
+				ibuf_update_free_bits_zip(block, mtr);
+			} else {
+				ibuf_update_free_bits_low(block, max_ins_size,
+							  mtr);
+			}
 		}
 
 		err = DB_SUCCESS;

storage/innobase/buf/buf0buf.cc

@@ -53,6 +53,10 @@ Created 11/5/1995 Heikki Tuuri
 #include "page0zip.h"
 #include "srv0mon.h"
 #include "buf0checksum.h"
+#ifdef HAVE_LIBNUMA
+#include <numa.h>
+#include <numaif.h>
+#endif // HAVE_LIBNUMA
 
 /*
 		IMPLEMENTATION OF THE BUFFER POOL
@@ -1112,6 +1116,22 @@ buf_chunk_init(
 		return(NULL);
 	}
 
+#ifdef HAVE_LIBNUMA
+	if (srv_numa_interleave) {
+		int	st = mbind(chunk->mem, chunk->mem_size,
+				   MPOL_INTERLEAVE,
+				   numa_all_nodes_ptr->maskp,
+				   numa_all_nodes_ptr->size,
+				   MPOL_MF_MOVE);
+		if (st != 0) {
+			ib_logf(IB_LOG_LEVEL_WARN,
+				"Failed to set NUMA memory policy of buffer"
+				" pool page frames to MPOL_INTERLEAVE"
+				" (error: %s).", strerror(errno));
+		}
+	}
+#endif // HAVE_LIBNUMA
+
 	/* Allocate the block descriptors from
 	the start of the memory block. */
 	chunk->blocks = (buf_block_t*) chunk->mem;
@@ -1442,6 +1462,21 @@ buf_pool_init(
 	ut_ad(n_instances <= MAX_BUFFER_POOLS);
 	ut_ad(n_instances == srv_buf_pool_instances);
 
+#ifdef HAVE_LIBNUMA
+	if (srv_numa_interleave) {
+		ib_logf(IB_LOG_LEVEL_INFO,
+			"Setting NUMA memory policy to MPOL_INTERLEAVE");
+		if (set_mempolicy(MPOL_INTERLEAVE,
+				  numa_all_nodes_ptr->maskp,
+				  numa_all_nodes_ptr->size) != 0) {
+			ib_logf(IB_LOG_LEVEL_WARN,
+				"Failed to set NUMA memory policy to"
+				" MPOL_INTERLEAVE (error: %s).",
+				strerror(errno));
+		}
+	}
+#endif // HAVE_LIBNUMA
+
 	buf_pool_ptr = (buf_pool_t*) mem_zalloc(
 		n_instances * sizeof *buf_pool_ptr);
 
@@ -1462,6 +1497,18 @@ buf_pool_init(
 
 	btr_search_sys_create(buf_pool_get_curr_size() / sizeof(void*) / 64);
 
+#ifdef HAVE_LIBNUMA
+	if (srv_numa_interleave) {
+		ib_logf(IB_LOG_LEVEL_INFO,
+			"Setting NUMA memory policy to MPOL_DEFAULT");
+		if (set_mempolicy(MPOL_DEFAULT, NULL, 0) != 0) {
+			ib_logf(IB_LOG_LEVEL_WARN,
+				"Failed to set NUMA memory policy to"
+				" MPOL_DEFAULT (error: %s).", strerror(errno));
+		}
+	}
+#endif // HAVE_LIBNUMA
+
 	return(DB_SUCCESS);
 }
 

storage/innobase/dict/dict0dict.cc

@@ -207,14 +207,6 @@ dict_index_remove_from_cache_low(
 	dict_index_t*	index,		/*!< in, own: index */
 	ibool		lru_evict);	/*!< in: TRUE if page being evicted
 					to make room in the table LRU list */
-/**********************************************************************//**
-Removes a table object from the dictionary cache. */
-static
-void
-dict_table_remove_from_cache_low(
-/*=============================*/
-	dict_table_t*	table,		/*!< in, own: table */
-	ibool		lru_evict);	/*!< in: TRUE if evicting from LRU */
 #ifdef UNIV_DEBUG
 /**********************************************************************//**
 Validate the dictionary table LRU list.
@@ -748,6 +740,45 @@ dict_table_get_all_fts_indexes(
 	return(ib_vector_size(indexes));
 }
 
+/** Store autoinc value when the table is evicted.
+@param[in]	table	table evicted */
+UNIV_INTERN
+void
+dict_table_autoinc_store(
+	const dict_table_t*	table)
+{
+	ut_ad(mutex_own(&dict_sys->mutex));
+
+	if (table->autoinc != 0) {
+		ut_ad(dict_sys->autoinc_map->find(table->id)
+		      == dict_sys->autoinc_map->end());
+
+		dict_sys->autoinc_map->insert(
+			std::pair<table_id_t, ib_uint64_t>(
+			table->id, table->autoinc));
+	}
+}
+
+/** Restore autoinc value when the table is loaded.
+@param[in]	table	table loaded */
+UNIV_INTERN
+void
+dict_table_autoinc_restore(
+	dict_table_t*	table)
+{
+	ut_ad(mutex_own(&dict_sys->mutex));
+
+	autoinc_map_t::iterator	it;
+	it = dict_sys->autoinc_map->find(table->id);
+
+	if (it != dict_sys->autoinc_map->end()) {
+		table->autoinc = it->second;
+		ut_ad(table->autoinc != 0);
+
+		dict_sys->autoinc_map->erase(it);
+	}
+}
+
 /********************************************************************//**
 Reads the next autoinc value (== autoinc counter value), 0 if not yet
 initialized.
@@ -1041,6 +1072,8 @@ dict_init(void)
 		mutex_create(dict_foreign_err_mutex_key,
 			     &dict_foreign_err_mutex, SYNC_NO_ORDER_CHECK);
 	}
+
+	dict_sys->autoinc_map = new autoinc_map_t();
 }
 
 /**********************************************************************//**
@@ -1288,6 +1321,8 @@ dict_table_add_to_cache(
 		UT_LIST_ADD_FIRST(table_LRU, dict_sys->table_non_LRU, table);
 	}
 
+	dict_table_autoinc_restore(table);
+
 	ut_ad(dict_lru_validate());
 
 	dict_sys->size += mem_heap_get_size(table->heap)
@@ -1978,7 +2013,6 @@ dict_table_change_id_in_cache(
 
 /**********************************************************************//**
 Removes a table object from the dictionary cache. */
-static
 void
 dict_table_remove_from_cache_low(
 /*=============================*/
@@ -2040,6 +2074,10 @@ dict_table_remove_from_cache_low(
 
 	ut_ad(dict_lru_validate());
 
+	if (lru_evict) {
+		dict_table_autoinc_store(table);
+	}
+
 	if (lru_evict && table->drop_aborted) {
 		/* Do as dict_table_try_drop_aborted() does. */
 
@@ -6330,6 +6368,8 @@ dict_close(void)
 		mutex_free(&dict_foreign_err_mutex);
 	}
 
+	delete dict_sys->autoinc_map;
+
 	mem_free(dict_sys);
 	dict_sys = NULL;
 }

storage/innobase/handler/ha_innodb.cc

@@ -6495,7 +6495,7 @@ ha_innobase::write_row(
 
 	DBUG_ENTER("ha_innobase::write_row");
 
-	if (srv_read_only_mode) {
+	if (high_level_read_only) {
 		ib_senderrf(ha_thd(), IB_LOG_LEVEL_WARN, ER_READ_ONLY_MODE);
 		DBUG_RETURN(HA_ERR_TABLE_READONLY);
 	} else if (prebuilt->trx != trx) {
@@ -7039,7 +7039,7 @@ ha_innobase::update_row(
 
 	ut_a(prebuilt->trx == trx);
 
-	if (srv_read_only_mode) {
+	if (high_level_read_only) {
 		ib_senderrf(ha_thd(), IB_LOG_LEVEL_WARN, ER_READ_ONLY_MODE);
 		DBUG_RETURN(HA_ERR_TABLE_READONLY);
 	} else if (!trx_is_started(trx)) {
@@ -7171,7 +7171,7 @@ ha_innobase::delete_row(
 
 	ut_a(prebuilt->trx == trx);
 
-	if (srv_read_only_mode) {
+	if (high_level_read_only) {
 		ib_senderrf(ha_thd(), IB_LOG_LEVEL_WARN, ER_READ_ONLY_MODE);
 		DBUG_RETURN(HA_ERR_TABLE_READONLY);
 	} else if (!trx_is_started(trx)) {
@@ -9499,7 +9499,7 @@ ha_innobase::create(
 
 	if (form->s->fields > REC_MAX_N_USER_FIELDS) {
 		DBUG_RETURN(HA_ERR_TOO_MANY_FIELDS);
-	} else if (srv_read_only_mode) {
+	} else if (high_level_read_only) {
 		DBUG_RETURN(HA_ERR_INNODB_READ_ONLY);
 	}
 
@@ -9829,7 +9829,7 @@ ha_innobase::discard_or_import_tablespace(
 	ut_a(prebuilt->trx->magic_n == TRX_MAGIC_N);
 	ut_a(prebuilt->trx == thd_to_trx(ha_thd()));
 
-	if (srv_read_only_mode) {
+	if (high_level_read_only) {
 		DBUG_RETURN(HA_ERR_TABLE_READONLY);
 	}
 
@@ -9923,7 +9923,7 @@ ha_innobase::truncate()
 
 	DBUG_ENTER("ha_innobase::truncate");
 
-	if (srv_read_only_mode) {
+	if (high_level_read_only) {
 		DBUG_RETURN(HA_ERR_TABLE_READONLY);
 	}
 
@@ -10274,7 +10274,7 @@ ha_innobase::rename_table(
 
 	DBUG_ENTER("ha_innobase::rename_table");
 
-	if (srv_read_only_mode) {
+	if (high_level_read_only) {
 		ib_senderrf(thd, IB_LOG_LEVEL_WARN, ER_READ_ONLY_MODE);
 		DBUG_RETURN(HA_ERR_TABLE_READONLY);
 	}
@@ -14288,6 +14288,12 @@ innodb_internal_table_validate(
 		}
 
 		dict_table_close(user_table, FALSE, TRUE);
+
+		DBUG_EXECUTE_IF("innodb_evict_autoinc_table",
+			mutex_enter(&dict_sys->mutex);
+			dict_table_remove_from_cache_low(user_table, TRUE);
+			mutex_exit(&dict_sys->mutex);
+		);
 	}
 
 	return(ret);
@@ -16301,6 +16307,13 @@ static MYSQL_SYSVAR_BOOL(use_native_aio, srv_use_native_aio,
   "Use native AIO if supported on this platform.",
   NULL, NULL, TRUE);
 
+#ifdef HAVE_LIBNUMA
+static MYSQL_SYSVAR_BOOL(numa_interleave, srv_numa_interleave,
+  PLUGIN_VAR_NOCMDARG | PLUGIN_VAR_READONLY,
+  "Use NUMA interleave memory policy to allocate InnoDB buffer pool.",
+  NULL, NULL, FALSE);
+#endif // HAVE_LIBNUMA
+
 static MYSQL_SYSVAR_BOOL(api_enable_binlog, ib_binlog_enabled,
   PLUGIN_VAR_NOCMDARG | PLUGIN_VAR_READONLY,
   "Enable binlog for applications direct access InnoDB through InnoDB APIs",
@@ -16579,6 +16592,9 @@ static struct st_mysql_sys_var* innobase_system_variables[]= {
   MYSQL_SYSVAR(version),
   MYSQL_SYSVAR(use_sys_malloc),
   MYSQL_SYSVAR(use_native_aio),
+#ifdef HAVE_LIBNUMA
+  MYSQL_SYSVAR(numa_interleave),
+#endif // HAVE_LIBNUMA
   MYSQL_SYSVAR(change_buffering),
   MYSQL_SYSVAR(change_buffer_max_size),
 #if defined UNIV_DEBUG || defined UNIV_IBUF_DEBUG

storage/innobase/handler/handler0alter.cc

@@ -234,7 +234,7 @@ ha_innobase::check_if_supported_inplace_alter(
 {
 	DBUG_ENTER("check_if_supported_inplace_alter");
 
-	if (srv_read_only_mode) {
+	if (high_level_read_only) {
 		ha_alter_info->unsupported_reason =
 			innobase_get_err_msg(ER_READ_ONLY_MODE);
 		DBUG_RETURN(HA_ALTER_INPLACE_NOT_SUPPORTED);

storage/innobase/include/dict0dict.h

@@ -309,6 +309,21 @@ dict_table_autoinc_initialize(
 	dict_table_t*	table,	/*!< in/out: table */
 	ib_uint64_t	value)	/*!< in: next value to assign to a row */
 	__attribute__((nonnull));
+
+/** Store autoinc value when the table is evicted.
+@param[in]	table	table evicted */
+UNIV_INTERN
+void
+dict_table_autoinc_store(
+	const dict_table_t*	table);
+
+/** Restore autoinc value when the table is loaded.
+@param[in]	table	table loaded */
+UNIV_INTERN
+void
+dict_table_autoinc_restore(
+	dict_table_t*	table);
+
 /********************************************************************//**
 Reads the next autoinc value (== autoinc counter value), 0 if not yet
 initialized.
@@ -368,6 +383,15 @@ dict_table_remove_from_cache(
 	dict_table_t*	table)	/*!< in, own: table */
 	__attribute__((nonnull));
 /**********************************************************************//**
+Removes a table object from the dictionary cache. */
+UNIV_INTERN
+void
+dict_table_remove_from_cache_low(
+/*=============================*/
+	dict_table_t*	table,		/*!< in, own: table */
+	ibool		lru_evict);	/*!< in: TRUE if table being evicted
+					to make room in the table LRU list */
+/**********************************************************************//**
 Renames a table object.
 @return	TRUE if success */
 UNIV_INTERN
@@ -1543,6 +1567,8 @@ extern dict_sys_t*	dict_sys;
 /** the data dictionary rw-latch protecting dict_sys */
 extern rw_lock_t	dict_operation_lock;
 
+typedef std::map<table_id_t, ib_uint64_t> autoinc_map_t;
+
 /* Dictionary system struct */
 struct dict_sys_t{
 	ib_mutex_t		mutex;		/*!< mutex protecting the data
@@ -1577,6 +1603,8 @@ struct dict_sys_t{
 	UT_LIST_BASE_NODE_T(dict_table_t)
 			table_non_LRU;	/*!< List of tables that can't be
 					evicted from the cache */
+	autoinc_map_t*	autoinc_map;	/*!< Map to store table id and autoinc
+					when table is evicted */
 };
 #endif /* !UNIV_HOTBACKUP */
 

storage/innobase/include/ibuf0ibuf.ic

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1997, 2013, Oracle and/or its affiliates. All Rights Reserved.
+Copyright (c) 1997, 2015, Oracle and/or its affiliates. All Rights Reserved.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -128,7 +128,8 @@ ibuf_should_try(
 	       && ibuf->max_size != 0
 	       && !dict_index_is_clust(index)
 	       && index->table->quiesce == QUIESCE_NONE
-	       && (ignore_sec_unique || !dict_index_is_unique(index)));
+	       && (ignore_sec_unique || !dict_index_is_unique(index))
+	       && srv_force_recovery < SRV_FORCE_NO_IBUF_MERGE);
 }
 
 /******************************************************************//**

storage/innobase/include/os0sync.h

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1995, 2014, Oracle and/or its affiliates. All Rights Reserved.
+Copyright (c) 1995, 2015, Oracle and/or its affiliates. All Rights Reserved.
 Copyright (c) 2008, Google Inc.
 
 Portions of this file contain modifications contributed and copyrighted by
@@ -38,6 +38,26 @@ Created 9/6/1995 Heikki Tuuri
 #include "ut0lst.h"
 #include "sync0types.h"
 
+#if defined __i386__ || defined __x86_64__ || defined _M_IX86 \
+    || defined _M_X64 || defined __WIN__
+
+#define IB_STRONG_MEMORY_MODEL
+
+#endif /* __i386__ || __x86_64__ || _M_IX86 || _M_X64 || __WIN__ */
+
+#ifdef HAVE_WINDOWS_ATOMICS
+typedef LONG lock_word_t;	/*!< On Windows, InterlockedExchange operates
+				on LONG variable */
+#elif defined(HAVE_ATOMIC_BUILTINS) && !defined(HAVE_ATOMIC_BUILTINS_BYTE)
+typedef ulint lock_word_t;
+#else
+
+#define IB_LOCK_WORD_IS_BYTE
+
+typedef byte lock_word_t;
+
+#endif /* HAVE_WINDOWS_ATOMICS */
+
 #ifdef __WIN__
 /** Native event (slow)*/
 typedef HANDLE			os_native_event_t;
@@ -429,14 +449,61 @@ amount to decrement. */
 # define os_atomic_decrement_uint64(ptr, amount) \
 	os_atomic_decrement(ptr, amount)
 
-/**********************************************************//**
-Returns the old value of *ptr, atomically sets *ptr to new_val */
+# if defined(HAVE_IB_GCC_ATOMIC_TEST_AND_SET)
 
-# define os_atomic_test_and_set_byte(ptr, new_val) \
-	__sync_lock_test_and_set(ptr, (byte) new_val)
+/** Do an atomic test-and-set.
+@param[in,out]	ptr		Memory location to set to non-zero
+@return the previous value */
+inline
+lock_word_t
+os_atomic_test_and_set(volatile lock_word_t* ptr)
+{
+       return(__atomic_test_and_set(ptr, __ATOMIC_ACQUIRE));
+}
 
-# define os_atomic_test_and_set_ulint(ptr, new_val) \
-	__sync_lock_test_and_set(ptr, new_val)
+/** Do an atomic clear.
+@param[in,out]	ptr		Memory location to set to zero */
+inline
+void
+os_atomic_clear(volatile lock_word_t* ptr)
+{
+	__atomic_clear(ptr, __ATOMIC_RELEASE);
+}
+
+# elif defined(IB_STRONG_MEMORY_MODEL)
+
+/** Do an atomic test and set.
+@param[in,out]	ptr		Memory location to set to non-zero
+@return the previous value */
+inline
+lock_word_t
+os_atomic_test_and_set(volatile lock_word_t* ptr)
+{
+	return(__sync_lock_test_and_set(ptr, 1));
+}
+
+/** Do an atomic release.
+
+In theory __sync_lock_release should be used to release the lock.
+Unfortunately, it does not work properly alone. The workaround is
+that more conservative __sync_lock_test_and_set is used instead.
+
+Performance regression was observed at some conditions for Intel
+architecture. Disable release barrier on Intel architecture for now.
+@param[in,out]	ptr		Memory location to write to
+@return the previous value */
+inline
+lock_word_t
+os_atomic_clear(volatile lock_word_t* ptr)
+{
+	return(__sync_lock_test_and_set(ptr, 0));
+}
+
+# else
+
+#  error "Unsupported platform"
+
+# endif /* HAVE_IB_GCC_ATOMIC_TEST_AND_SET */
 
 #elif defined(HAVE_IB_SOLARIS_ATOMICS)
 
@@ -511,14 +578,51 @@ amount to decrement. */
 # define os_atomic_decrement_uint64(ptr, amount) \
 	os_atomic_increment_uint64(ptr, -(amount))
 
-/**********************************************************//**
-Returns the old value of *ptr, atomically sets *ptr to new_val */
+# ifdef IB_LOCK_WORD_IS_BYTE
 
-# define os_atomic_test_and_set_byte(ptr, new_val) \
-	atomic_swap_uchar(ptr, new_val)
+/** Do an atomic xchg and set to non-zero.
+@param[in,out]	ptr		Memory location to set to non-zero
+@return the previous value */
+inline
+lock_word_t
+os_atomic_test_and_set(volatile lock_word_t* ptr)
+{
+	return(atomic_swap_uchar(ptr, 1));
+}
 
-# define os_atomic_test_and_set_ulint(ptr, new_val) \
-	atomic_swap_ulong(ptr, new_val)
+/** Do an atomic xchg and set to zero.
+@param[in,out]	ptr		Memory location to set to zero
+@return the previous value */
+inline
+lock_word_t
+os_atomic_clear(volatile lock_word_t* ptr)
+{
+	return(atomic_swap_uchar(ptr, 0));
+}
+
+# else
+
+/** Do an atomic xchg and set to non-zero.
+@param[in,out]	ptr		Memory location to set to non-zero
+@return the previous value */
+inline
+lock_word_t
+os_atomic_test_and_set(volatile lock_word_t* ptr)
+{
+	return(atomic_swap_ulong(ptr, 1));
+}
+
+/** Do an atomic xchg and set to zero.
+@param[in,out]	ptr		Memory location to set to zero
+@return the previous value */
+inline
+lock_word_t
+os_atomic_clear(volatile lock_word_t* ptr)
+{
+	return(atomic_swap_ulong(ptr, 0));
+}
+
+# endif /* IB_LOCK_WORD_IS_BYTE */
 
 #elif defined(HAVE_WINDOWS_ATOMICS)
 
@@ -633,16 +737,27 @@ amount to decrement. There is no atomic substract function on Windows */
 				(ib_int64_t*) ptr,		\
 				-(ib_int64_t) amount) - amount))
 
-/**********************************************************//**
-Returns the old value of *ptr, atomically sets *ptr to new_val.
-InterlockedExchange() operates on LONG, and the LONG will be
-clobbered */
+/** Do an atomic test and set.
+InterlockedExchange() operates on LONG, and the LONG will be clobbered
+@param[in,out]	ptr		Memory location to set to non-zero
+@return the previous value */
+inline
+lock_word_t
+os_atomic_test_and_set(volatile lock_word_t* ptr)
+{
+	return(InterlockedExchange(ptr, 1));
+}
 
-# define os_atomic_test_and_set_byte(ptr, new_val) \
-	((byte) InterlockedExchange(ptr, new_val))
-
-# define os_atomic_test_and_set_ulong(ptr, new_val) \
-	InterlockedExchange(ptr, new_val)
+/** Do an atomic release.
+InterlockedExchange() operates on LONG, and the LONG will be clobbered
+@param[in,out]	ptr		Memory location to set to zero
+@return the previous value */
+inline
+lock_word_t
+os_atomic_clear(volatile lock_word_t* ptr)
+{
+	return(InterlockedExchange(ptr, 0));
+}
 
 #else
 # define IB_ATOMICS_STARTUP_MSG \
@@ -692,7 +807,7 @@ for synchronization */
 	} while (0);
 
 /** barrier definitions for memory ordering */
-#if defined __i386__ || defined __x86_64__ || defined _M_IX86 || defined _M_X64 || defined __WIN__
+#ifdef IB_STRONG_MEMORY_MODEL
 /* Performance regression was observed at some conditions for Intel
 architecture. Disable memory barrier for Intel architecture for now. */
 # define os_rmb

storage/innobase/include/srv0srv.h

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1995, 2013, Oracle and/or its affiliates. All rights reserved.
+Copyright (c) 1995, 2015, Oracle and/or its affiliates. All rights reserved.
 Copyright (c) 2008, 2009, Google Inc.
 Copyright (c) 2009, Percona Inc.
 
@@ -187,6 +187,9 @@ extern char*	srv_arch_dir;
 recovery and open all tables in RO mode instead of RW mode. We don't
 sync the max trx id to disk either. */
 extern my_bool	srv_read_only_mode;
+/** Set if InnoDB operates in read-only mode or innodb-force-recovery
+is greater than SRV_FORCE_NO_TRX_UNDO. */
+extern my_bool	high_level_read_only;
 /** store to its own file each table created by an user; data
 dictionary tables are in the system tablespace 0 */
 extern my_bool	srv_file_per_table;
@@ -217,6 +220,7 @@ OS (provided we compiled Innobase with it in), otherwise we will
 use simulated aio we build below with threads.
 Currently we support native aio on windows and linux */
 extern my_bool	srv_use_native_aio;
+extern my_bool	srv_numa_interleave;
 #ifdef __WIN__
 extern ibool	srv_use_native_conditions;
 #endif /* __WIN__ */
@@ -875,6 +879,7 @@ struct srv_slot_t{
 #else /* !UNIV_HOTBACKUP */
 # define srv_use_adaptive_hash_indexes		FALSE
 # define srv_use_native_aio			FALSE
+# define srv_numa_interleave			FALSE
 # define srv_force_recovery			0UL
 # define srv_set_io_thread_op_info(t,info)	((void) 0)
 # define srv_reset_io_thread_op_info()		((void) 0)

storage/innobase/include/sync0sync.h

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1995, 2012, Oracle and/or its affiliates. All Rights Reserved.
+Copyright (c) 1995, 2015, Oracle and/or its affiliates. All Rights Reserved.
 Copyright (c) 2008, Google Inc.
 Copyright (c) 2012, Facebook Inc.
 
@@ -46,15 +46,6 @@ Created 9/5/1995 Heikki Tuuri
 extern "C" my_bool	timed_mutexes;
 #endif /* UNIV_DEBUG && !UNIV_HOTBACKUP */
 
-#ifdef HAVE_WINDOWS_ATOMICS
-typedef LONG lock_word_t;	/*!< On Windows, InterlockedExchange operates
-				on LONG variable */
-#elif defined(HAVE_ATOMIC_BUILTINS) && !defined(HAVE_ATOMIC_BUILTINS_BYTE)
-typedef ulint lock_word_t;
-#else
-typedef byte lock_word_t;
-#endif
-
 #if defined UNIV_PFS_MUTEX || defined UNIV_PFS_RWLOCK
 
 /* By default, buffer mutexes and rwlocks will be excluded from

storage/innobase/include/sync0sync.ic

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1995, 2009, Oracle and/or its affiliates. All Rights Reserved.
+Copyright (c) 1995, 2015, Oracle and/or its affiliates. All Rights Reserved.
 Copyright (c) 2008, Google Inc.
 
 Portions of this file contain modifications contributed and copyrighted by
@@ -74,17 +74,13 @@ Performs an atomic test-and-set instruction to the lock_word field of a
 mutex.
 @return	the previous value of lock_word: 0 or 1 */
 UNIV_INLINE
-byte
+lock_word_t
 ib_mutex_test_and_set(
-/*===============*/
+/*==================*/
 	ib_mutex_t*	mutex)	/*!< in: mutex */
 {
 #if defined(HAVE_ATOMIC_BUILTINS)
-# if defined(HAVE_ATOMIC_BUILTINS_BYTE)
-	return(os_atomic_test_and_set_byte(&mutex->lock_word, 1));
-# else
-	return(os_atomic_test_and_set_ulint(&mutex->lock_word, 1));
-# endif
+	return(os_atomic_test_and_set(&mutex->lock_word));
 #else
 	ibool	ret;
 
@@ -100,7 +96,7 @@ ib_mutex_test_and_set(
 	}
 
 	return((byte) ret);
-#endif
+#endif /* HAVE_ATOMIC_BUILTINS */
 }
 
 /******************************************************************//**
@@ -113,19 +109,12 @@ mutex_reset_lock_word(
 	ib_mutex_t*	mutex)	/*!< in: mutex */
 {
 #if defined(HAVE_ATOMIC_BUILTINS)
-	/* In theory __sync_lock_release should be used to release the lock.
-	Unfortunately, it does not work properly alone. The workaround is
-	that more conservative __sync_lock_test_and_set is used instead. */
-# if defined(HAVE_ATOMIC_BUILTINS_BYTE)
-	os_atomic_test_and_set_byte(&mutex->lock_word, 0);
-# else
-	os_atomic_test_and_set_ulint(&mutex->lock_word, 0);
-# endif
+	os_atomic_clear(&mutex->lock_word);
 #else
 	mutex->lock_word = 0;
 
 	os_fast_mutex_unlock(&(mutex->os_fast_mutex));
-#endif
+#endif /* HAVE_ATOMIC_BUILTINS */
 }
 
 /******************************************************************//**

storage/innobase/lock/lock0lock.cc

@@ -2634,8 +2634,8 @@ lock_rec_inherit_to_gap(
 	/* If srv_locks_unsafe_for_binlog is TRUE or session is using
 	READ COMMITTED isolation level, we do not want locks set
 	by an UPDATE or a DELETE to be inherited as gap type locks. But we
-	DO want S-locks set by a consistency constraint to be inherited also
-	then. */
+	DO want S-locks/X-locks(taken for replace) set by a consistency
+	constraint to be inherited also then */
 
 	for (lock = lock_rec_get_first(block, heap_no);
 	     lock != NULL;
@@ -2645,7 +2645,8 @@ lock_rec_inherit_to_gap(
 		    && !((srv_locks_unsafe_for_binlog
 			  || lock->trx->isolation_level
 			  <= TRX_ISO_READ_COMMITTED)
-			 && lock_get_mode(lock) == LOCK_X)) {
+			 && lock_get_mode(lock) ==
+			 (lock->trx->duplicates ? LOCK_S : LOCK_X))) {
 
 			lock_rec_add_to_queue(
 				LOCK_REC | LOCK_GAP | lock_get_mode(lock),

storage/innobase/log/log0log.cc

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1995, 2014, Oracle and/or its affiliates. All Rights Reserved.
+Copyright (c) 1995, 2015, Oracle and/or its affiliates. All Rights Reserved.
 Copyright (c) 2009, Google Inc.
 
 Portions of this file contain modifications contributed and copyrighted by
@@ -3406,11 +3406,7 @@ loop:
 
 	lsn = log_sys->lsn;
 
-	ut_ad(srv_force_recovery != SRV_FORCE_NO_LOG_REDO
-	      || lsn == log_sys->last_checkpoint_lsn + LOG_BLOCK_HDR_SIZE);
-
-	if ((srv_force_recovery != SRV_FORCE_NO_LOG_REDO
-	     && lsn != log_sys->last_checkpoint_lsn)
+	if (lsn != log_sys->last_checkpoint_lsn
 #ifdef UNIV_LOG_ARCHIVE
 	    || (srv_log_archive_on
 		&& lsn != log_sys->archived_lsn + LOG_BLOCK_HDR_SIZE)

storage/innobase/row/row0ins.cc

@@ -2709,6 +2709,8 @@ row_ins_sec_index_entry_low(
 			goto func_exit;
 		}
 
+		DEBUG_SYNC_C("row_ins_sec_index_entry_dup_locks_created");
+
 		/* We did not find a duplicate and we have now
 		locked with s-locks the necessary records to
 		prevent any insertion of a duplicate by another

storage/innobase/srv/srv0srv.cc

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1995, 2014, Oracle and/or its affiliates. All Rights Reserved.
+Copyright (c) 1995, 2015, Oracle and/or its affiliates. All Rights Reserved.
 Copyright (c) 2008, 2009 Google Inc.
 Copyright (c) 2009, Percona Inc.
 
@@ -126,6 +126,9 @@ UNIV_INTERN ulint	srv_file_format = 0;
 UNIV_FORMAT_MAX + 1 means no checking ie. FALSE.  The default is to
 set it to the highest format we support. */
 UNIV_INTERN ulint	srv_max_file_format_at_startup = UNIV_FORMAT_MAX;
+/** Set if InnoDB operates in read-only mode or innodb-force-recovery
+is greater than SRV_FORCE_NO_TRX_UNDO. */
+UNIV_INTERN my_bool	high_level_read_only;
 
 #if UNIV_FORMAT_A
 # error "UNIV_FORMAT_A must be 0!"
@@ -144,6 +147,7 @@ OS (provided we compiled Innobase with it in), otherwise we will
 use simulated aio we build below with threads.
 Currently we support native aio on windows and linux */
 UNIV_INTERN my_bool	srv_use_native_aio = TRUE;
+UNIV_INTERN my_bool	srv_numa_interleave = FALSE;
 
 #ifdef __WIN__
 /* Windows native condition variables. We use runtime loading / function
@@ -2581,13 +2585,8 @@ srv_do_purge(
 		}
 
 		n_pages_purged = trx_purge(
-			n_use_threads, srv_purge_batch_size, false);
-
-		if (!(count++ % TRX_SYS_N_RSEGS)) {
-			/* Force a truncate of the history list. */
-			n_pages_purged += trx_purge(
-				1, srv_purge_batch_size, true);
-		}
+			n_use_threads, srv_purge_batch_size,
+			(++count % TRX_SYS_N_RSEGS) == 0);
 
 		*n_total_purged += n_pages_purged;
 
@@ -2780,8 +2779,17 @@ DECLARE_THREAD(srv_purge_coordinator_thread)(
 		n_pages_purged = trx_purge(1, srv_purge_batch_size, false);
 	}
 
-	/* Force a truncate of the history list. */
-	n_pages_purged = trx_purge(1, srv_purge_batch_size, true);
+	/* This trx_purge is called to remove any undo records (added by
+	background threads) after completion of the above loop. When
+	srv_fast_shutdown != 0, a large batch size can cause significant
+	delay in shutdown ,so reducing the batch size to magic number 20
+	(which was default in 5.5), which we hope will be sufficient to
+	remove all the undo records */
+	const	uint temp_batch_size = 20;
+
+	n_pages_purged = trx_purge(1, srv_purge_batch_size <= temp_batch_size
+				      ? srv_purge_batch_size : temp_batch_size,
+				   true);
 	ut_a(n_pages_purged == 0 || srv_fast_shutdown != 0);
 
 	/* The task queue should always be empty, independent of fast

storage/innobase/srv/srv0start.cc

@@ -1545,9 +1545,8 @@ innobase_start_or_create_for_mysql(void)
 	char*		logfile0	= NULL;
 	size_t		dirnamelen;
 
-	if (srv_force_recovery > SRV_FORCE_NO_TRX_UNDO) {
-		srv_read_only_mode = true;
-	}
+	high_level_read_only = srv_read_only_mode
+		|| srv_force_recovery > SRV_FORCE_NO_TRX_UNDO;
 
 	if (srv_read_only_mode) {
 		ib_logf(IB_LOG_LEVEL_INFO, "Started in read only mode");

storage/innobase/trx/trx0sys.cc

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1996, 2013, Oracle and/or its affiliates. All Rights Reserved.
+Copyright (c) 1996, 2015, Oracle and/or its affiliates. All Rights Reserved.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -29,7 +29,10 @@ Created 3/26/1996 Heikki Tuuri
 #include "trx0sys.ic"
 #endif
 
-#ifndef UNIV_HOTBACKUP
+#ifdef UNIV_HOTBACKUP
+#include "fsp0types.h"
+
+#else	/* !UNIV_HOTBACKUP */
 #include "fsp0fsp.h"
 #include "mtr0log.h"
 #include "mtr0log.h"
@@ -1115,18 +1118,15 @@ trx_sys_read_pertable_file_format_id(
 	/* get the file format from the page */
 	ptr = page + 54;
 	flags = mach_read_from_4(ptr);
-	if (flags == 0) {
-		/* file format is Antelope */
-		*format_id = 0;
-		return(TRUE);
-	} else if (flags & 1) {
-		/* tablespace flags are ok */
-		*format_id = (flags / 32) % 128;
-		return(TRUE);
-	} else {
+
+	if (!fsp_flags_is_valid(flags) {
 		/* bad tablespace flags */
 		return(FALSE);
 	}
+
+	*format_id = FSP_FLAGS_GET_POST_ANTELOPE(flags);
+
+	return(TRUE);
 }