Merge lgrimmer@bk-internal.mysql.com:/home/bk/mysql-5.0

into mysql.com:/space/my/mysql-5.0

8 files changed, 170 insertions, 24 deletions

diff --git a/innobase/buf/buf0buf.c b/innobase/buf/buf0buf.c
index 3930ea93889..89f851709db 100644
--- a/innobase/buf/buf0buf.c
+++ b/innobase/buf/buf0buf.c
@@ -331,33 +331,43 @@ buf_page_is_corrupted(
 		}
 	}
 #endif
-	old_checksum = buf_calc_page_old_checksum(read_buf);
-
-	old_checksum_field = mach_read_from_4(read_buf + UNIV_PAGE_SIZE
+  
+  /* If we use checksums validation, make additional check before returning
+  TRUE to ensure that the checksum is not equal to BUF_NO_CHECKSUM_MAGIC which
+  might be stored by InnoDB with checksums disabled.
+     Otherwise, skip checksum calculation and return FALSE */
+  
+  if (srv_use_checksums) {
+    old_checksum = buf_calc_page_old_checksum(read_buf); 
+
+    old_checksum_field = mach_read_from_4(read_buf + UNIV_PAGE_SIZE
 					- FIL_PAGE_END_LSN_OLD_CHKSUM);
 
-	/* There are 2 valid formulas for old_checksum_field:
-	1. Very old versions of InnoDB only stored 8 byte lsn to the start
-	and the end of the page.
-	2. Newer InnoDB versions store the old formula checksum there. */
+    /* There are 2 valid formulas for old_checksum_field:
+	  1. Very old versions of InnoDB only stored 8 byte lsn to the start
+	  and the end of the page.
+	  2. Newer InnoDB versions store the old formula checksum there. */
 	
-	if (old_checksum_field != mach_read_from_4(read_buf + FIL_PAGE_LSN)
-	    && old_checksum_field != old_checksum) {
+    if (old_checksum_field != mach_read_from_4(read_buf + FIL_PAGE_LSN)
+        && old_checksum_field != old_checksum
+        && old_checksum_field != BUF_NO_CHECKSUM_MAGIC) {
 
-		return(TRUE);
-	}
+      return(TRUE);
+    }
 
-	checksum = buf_calc_page_new_checksum(read_buf);
-	checksum_field = mach_read_from_4(read_buf + FIL_PAGE_SPACE_OR_CHKSUM);
+    checksum = buf_calc_page_new_checksum(read_buf);
+    checksum_field = mach_read_from_4(read_buf + FIL_PAGE_SPACE_OR_CHKSUM);
 
-	/* InnoDB versions < 4.0.14 and < 4.1.1 stored the space id
-	(always equal to 0), to FIL_PAGE_SPACE_SPACE_OR_CHKSUM */
+    /* InnoDB versions < 4.0.14 and < 4.1.1 stored the space id
+	  (always equal to 0), to FIL_PAGE_SPACE_SPACE_OR_CHKSUM */
 
-	if (checksum_field != 0 && checksum_field != checksum) {
-
-	        return(TRUE);
-	}
+    if (checksum_field != 0 && checksum_field != checksum
+        && checksum_field != BUF_NO_CHECKSUM_MAGIC) {
 
+      return(TRUE);
+    }
+  }
+  
 	return(FALSE);
 }
 
@@ -379,8 +389,10 @@ buf_page_print(
 	ut_print_buf(stderr, read_buf, UNIV_PAGE_SIZE);
 	fputs("InnoDB: End of page dump\n", stderr);
 
-	checksum = buf_calc_page_new_checksum(read_buf);
-	old_checksum = buf_calc_page_old_checksum(read_buf);
+	checksum = srv_use_checksums ?
+    buf_calc_page_new_checksum(read_buf) : BUF_NO_CHECKSUM_MAGIC;
+	old_checksum = srv_use_checksums ?
+    buf_calc_page_old_checksum(read_buf) : BUF_NO_CHECKSUM_MAGIC;
 
 	ut_print_timestamp(stderr);
 	fprintf(stderr, 
@@ -548,7 +560,7 @@ buf_pool_init(
 		}
 		/*----------------------------------------*/
 	} else {
-		buf_pool->frame_mem = ut_malloc_low(
+		buf_pool->frame_mem = os_mem_alloc_large(
 					UNIV_PAGE_SIZE * (n_frames + 1),
 					TRUE, FALSE);
 	}
diff --git a/innobase/buf/buf0flu.c b/innobase/buf/buf0flu.c
index aff4fe92a71..a0ca614d9b3 100644
--- a/innobase/buf/buf0flu.c
+++ b/innobase/buf/buf0flu.c
@@ -448,7 +448,8 @@ buf_flush_init_for_writing(
 	/* Store the new formula checksum */
 
 	mach_write_to_4(page + FIL_PAGE_SPACE_OR_CHKSUM,
-					buf_calc_page_new_checksum(page));
+					srv_use_checksums ?
+                  buf_calc_page_new_checksum(page) : BUF_NO_CHECKSUM_MAGIC);
 
 	/* We overwrite the first 4 bytes of the end lsn field to store
 	the old formula checksum. Since it depends also on the field
@@ -456,7 +457,8 @@ buf_flush_init_for_writing(
 	new formula checksum. */
 
 	mach_write_to_4(page + UNIV_PAGE_SIZE - FIL_PAGE_END_LSN_OLD_CHKSUM,
-					buf_calc_page_old_checksum(page));
+					srv_use_checksums ?
+                  buf_calc_page_old_checksum(page) : BUF_NO_CHECKSUM_MAGIC);
 }
 
 /************************************************************************
diff --git a/innobase/include/buf0buf.h b/innobase/include/buf0buf.h
index 8df1211327f..5ee323f1b1e 100644
--- a/innobase/include/buf0buf.h
+++ b/innobase/include/buf0buf.h
@@ -52,6 +52,8 @@ Created 11/5/1995 Heikki Tuuri
 /* Modes for buf_page_get_known_nowait */
 #define BUF_MAKE_YOUNG	51
 #define BUF_KEEP_OLD	52
+/* Magic value to use instead of checksums when they are disabled */
+#define BUF_NO_CHECKSUM_MAGIC 0xDEADBEEFUL
 
 extern buf_pool_t* 	buf_pool; 	/* The buffer pool of the database */
 extern ibool		buf_debug_prints;/* If this is set TRUE, the program
diff --git a/innobase/include/os0proc.h b/innobase/include/os0proc.h
index d0d3cf82e38..b0b72e18675 100644
--- a/innobase/include/os0proc.h
+++ b/innobase/include/os0proc.h
@@ -12,6 +12,11 @@ Created 9/30/1995 Heikki Tuuri
 
 #include "univ.i"
 
+#ifdef UNIV_LINUX
+#include <sys/ipc.h>
+#include <sys/shm.h>
+#endif
+
 typedef void*			os_process_t;
 typedef unsigned long int	os_process_id_t;
 
@@ -27,6 +32,10 @@ page size of an Intel x86 processor. We cannot use AWE with 2 MB or 4 MB
 pages. */
 #define	OS_AWE_X86_PAGE_SIZE	4096
 
+extern ibool os_use_large_pages;
+/* Large page size. This may be a boot-time option on some platforms */
+extern ulint os_large_page_size;
+
 /********************************************************************
 Windows AWE support. Tries to enable the "lock pages in memory" privilege for
 the current process so that the current process can allocate memory-locked
@@ -103,6 +112,25 @@ os_mem_alloc_nocache(
 			/* out: allocated memory */
 	ulint	n);	/* in: number of bytes */
 /********************************************************************
+Allocates large pages memory. */
+
+void*
+os_mem_alloc_large(
+/*=================*/
+      /* out: allocated memory */
+	ulint   n,     /* in: number of bytes */
+  ibool   set_to_zero, /* in: TRUE if allocated memory should be set
+          to zero if UNIV_SET_MEM_TO_ZERO is defined */
+  ibool	  assert_on_error); /* in: if TRUE, we crash mysqld if the memory
+          cannot be allocated */
+/********************************************************************
+Frees large pages memory. */
+
+void
+os_mem_free_large(
+/*=================*/
+void    *ptr);  /* in: number of bytes */
+/********************************************************************
 Sets the priority boost for threads released from waiting within the current
 process. */
 
diff --git a/innobase/include/srv0srv.h b/innobase/include/srv0srv.h
index d4cc7d8222f..84b7d14ca00 100644
--- a/innobase/include/srv0srv.h
+++ b/innobase/include/srv0srv.h
@@ -107,6 +107,7 @@ extern ibool	srv_very_fast_shutdown;  /* if this TRUE, do not flush the
 extern ibool	srv_innodb_status;
 
 extern ibool	srv_use_doublewrite_buf;
+extern ibool	srv_use_checksums;
 
 extern ibool    srv_set_thread_priorities;
 extern int      srv_query_thread_priority;
diff --git a/innobase/os/os0proc.c b/innobase/os/os0proc.c
index 2f155788420..98254ae1055 100644
--- a/innobase/os/os0proc.c
+++ b/innobase/os/os0proc.c
@@ -69,6 +69,10 @@ byte*		os_awe_window;
 ulint		os_awe_window_size;
 #endif
 
+ibool os_use_large_pages;
+/* Large page size. This may be a boot-time option on some platforms */
+ulint os_large_page_size;
+
 /********************************************************************
 Windows AWE support. Tries to enable the "lock pages in memory" privilege for
 the current process so that the current process can allocate memory-locked
@@ -516,6 +520,83 @@ os_mem_alloc_nocache(
 }
 
 /********************************************************************
+Allocates large pages memory. */
+
+void*
+os_mem_alloc_large(
+/*=================*/
+      /* out: allocated memory */
+  ulint	n, /* in: number of bytes */
+	ibool set_to_zero, /* in: TRUE if allocated memory should be set
+        to zero if UNIV_SET_MEM_TO_ZERO is defined */
+	ibool	assert_on_error) /* in: if TRUE, we crash mysqld if the memory
+				cannot be allocated */
+{
+#ifdef UNIV_LINUX
+  ulint size;
+  int shmid;
+  void *ptr = NULL;
+  struct shmid_ds buf;
+  
+  if (!os_use_large_pages || !os_large_page_size) {
+    goto skip;
+  }
+  
+  /* Align block size to os_large_page_size */
+  size = ((n - 1) & ~(os_large_page_size - 1)) + os_large_page_size;
+  
+  shmid = shmget(IPC_PRIVATE, (size_t)size, SHM_HUGETLB | SHM_R | SHM_W);
+  if (shmid < 0) {
+    fprintf(stderr, "InnoDB: HugeTLB: Warning: Failed to allocate %lu bytes. "
+            "errno %d\n", n, errno);
+  } else {
+    ptr = shmat(shmid, NULL, 0);
+    if (ptr == (void *)-1) {
+      fprintf(stderr, "InnoDB: HugeTLB: Warning: Failed to attach shared memory "
+              "segment, errno %d\n", errno);
+    }
+    /*
+      Remove the shared memory segment so that it will be automatically freed
+      after memory is detached or process exits
+    */
+    shmctl(shmid, IPC_RMID, &buf);
+  }
+  
+  if (ptr) {
+    if (set_to_zero) {
+#ifdef UNIV_SET_MEM_TO_ZERO
+      memset(ret, '\0', size);
+#endif
+    }
+
+    return(ptr);
+  }
+
+  fprintf(stderr, "InnoDB HugeTLB: Warning: Using conventional memory pool\n");
+#endif
+skip:
+  
+	return(ut_malloc_low(n, set_to_zero, assert_on_error));
+}
+
+/********************************************************************
+Frees large pages memory. */
+
+void
+os_mem_free_large(
+/*=================*/
+	void	*ptr)	/* in: number of bytes */
+{
+#ifdef UNIV_LINUX
+  if (os_use_large_pages && os_large_page_size && !shmdt(ptr)) {
+    return;
+  }
+#endif
+
+  ut_free(ptr);
+}
+
+/********************************************************************
 Sets the priority boost for threads released from waiting within the current
 process. */
 
diff --git a/innobase/srv/srv0srv.c b/innobase/srv/srv0srv.c
index 40befae424e..83d4fb4d39d 100644
--- a/innobase/srv/srv0srv.c
+++ b/innobase/srv/srv0srv.c
@@ -313,6 +313,7 @@ ibool	srv_very_fast_shutdown	= FALSE; /* if this TRUE, do not flush the
 ibool	srv_innodb_status	= FALSE;
 
 ibool	srv_use_doublewrite_buf	= TRUE;
+ibool	srv_use_checksums = TRUE;
 
 ibool   srv_set_thread_priorities = TRUE;
 int     srv_query_thread_priority = 0;
diff --git a/innobase/trx/trx0sys.c b/innobase/trx/trx0sys.c
index 35e18064329..57166e98f45 100644
--- a/innobase/trx/trx0sys.c
+++ b/innobase/trx/trx0sys.c
@@ -125,6 +125,22 @@ trx_doublewrite_init(
 }
 
 /********************************************************************
+Frees the doublewrite buffer. */
+static
+void
+trx_doublewrite_free(void)
+/*======================*/
+{
+  mutex_free(&(trx_doublewrite->mutex));
+
+  mem_free(trx_doublewrite->buf_block_arr);
+  ut_free(trx_doublewrite->write_buf_unaligned);
+
+  mem_free(trx_doublewrite);
+  trx_doublewrite = NULL;
+}
+
+/********************************************************************
 Marks the trx sys header when we have successfully upgraded to the >= 4.1.x
 multiple tablespace format. */
 
@@ -512,6 +528,9 @@ trx_sys_doublewrite_init_or_restore_pages(
 
 	fil_flush_file_spaces(FIL_TABLESPACE);
 	
+  if (!srv_use_doublewrite_buf)
+    trx_doublewrite_free();
+
 leave_func:
 	ut_free(unaligned_read_buf);
 }