Optimize Valgrind integration.

Forcefully disable tcache if running inside Valgrind, and remove Valgrind calls in tcache-specific code. Restructure Valgrind-related code to move most Valgrind calls out of the fast path functions. Take advantage of static knowledge to elide some branches in JEMALLOC_VALGRIND_REALLOC().
2014-04-15 16:35:08 -07:00 · 2014-04-15 16:35:08 -07:00 · bd87b01999
commit bd87b01999
parent ecd3e59ca3
12 changed files with 231 additions and 136 deletions
--- a/Makefile.in
+++ b/Makefile.in
@ -48,6 +48,7 @@ cfgoutputs_in := @cfgoutputs_in@
 cfgoutputs_out := @cfgoutputs_out@
 enable_autogen := @enable_autogen@
 enable_code_coverage := @enable_code_coverage@
+enable_valgrind := @enable_valgrind@
 enable_zone_allocator := @enable_zone_allocator@
 DSO_LDFLAGS = @DSO_LDFLAGS@
 SOREV = @SOREV@
@ -82,6 +83,9 @@ C_SRCS := $(srcroot)src/jemalloc.c $(srcroot)src/arena.c \
 	$(srcroot)src/mb.c $(srcroot)src/mutex.c $(srcroot)src/prof.c \
 	$(srcroot)src/quarantine.c $(srcroot)src/rtree.c $(srcroot)src/stats.c \
 	$(srcroot)src/tcache.c $(srcroot)src/util.c $(srcroot)src/tsd.c
+ifeq ($(enable_valgrind), 1)
+C_SRCS += $(srcroot)src/valgrind.c
+endif
 ifeq ($(enable_zone_allocator), 1)
 C_SRCS += $(srcroot)src/zone.c
 endif
--- a/doc/jemalloc.xml.in
+++ b/doc/jemalloc.xml.in
@ -979,7 +979,8 @@ malloc_conf = "xmalloc:true";]]></programlisting>
        linkend="opt.lg_tcache_max"><mallctl>opt.lg_tcache_max</mallctl></link>
        option for related tuning information.  This option is enabled by
        default unless running inside <ulink
-        url="http://valgrind.org/">Valgrind</ulink>.</para></listitem>
+        url="http://valgrind.org/">Valgrind</ulink>, in which case it is
+        forcefully disabled.</para></listitem>
      </varlistentry>

      <varlistentry id="opt.lg_tcache_max">
--- a/include/jemalloc/internal/jemalloc_internal.h.in
+++ b/include/jemalloc/internal/jemalloc_internal.h.in
@ -60,11 +60,6 @@ typedef intptr_t ssize_t;
 #include <sys/ktrace.h>
 #endif

-#ifdef JEMALLOC_VALGRIND
-#include <valgrind/valgrind.h>
-#include <valgrind/memcheck.h>
-#endif
-
 #define	JEMALLOC_NO_DEMANGLE
 #ifdef JEMALLOC_JET
 #  define JEMALLOC_N(n) jet_##n
@ -362,81 +357,7 @@ static const bool config_ivsalloc =
 #  define VARIABLE_ARRAY(type, name, count) type name[count]
 #endif

-#ifdef JEMALLOC_VALGRIND
-/*
- * The JEMALLOC_VALGRIND_*() macros must be macros rather than functions
- * so that when Valgrind reports errors, there are no extra stack frames
- * in the backtraces.
- *
- * The size that is reported to valgrind must be consistent through a chain of
- * malloc..realloc..realloc calls.  Request size isn't recorded anywhere in
- * jemalloc, so it is critical that all callers of these macros provide usize
- * rather than request size.  As a result, buffer overflow detection is
- * technically weakened for the standard API, though it is generally accepted
- * practice to consider any extra bytes reported by malloc_usable_size() as
- * usable space.
- */
-#define	JEMALLOC_VALGRIND_MALLOC(cond, ptr, usize, zero) do {		\
-	if (config_valgrind && in_valgrind && cond)			\
-		VALGRIND_MALLOCLIKE_BLOCK(ptr, usize, p2rz(ptr), zero);	\
-} while (0)
-#define	JEMALLOC_VALGRIND_REALLOC(ptr, usize, old_ptr, old_usize,	\
-    old_rzsize, zero)  do {						\
-	if (config_valgrind && in_valgrind) {				\
-		size_t rzsize = p2rz(ptr);				\
-									\
-		if (ptr == old_ptr) {					\
-			VALGRIND_RESIZEINPLACE_BLOCK(ptr, old_usize,	\
-			    usize, rzsize);				\
-			if (zero && old_usize < usize) {		\
-				VALGRIND_MAKE_MEM_DEFINED(		\
-				    (void *)((uintptr_t)ptr +		\
-				    old_usize), usize - old_usize);	\
-			}						\
-		} else {						\
-			if (old_ptr != NULL) {				\
-				VALGRIND_FREELIKE_BLOCK(old_ptr,	\
-				    old_rzsize);			\
-			}						\
-			if (ptr != NULL) {				\
-				size_t copy_size = (old_usize < usize)	\
-				    ?  old_usize : usize;		\
-				size_t tail_size = usize - copy_size;	\
-				VALGRIND_MALLOCLIKE_BLOCK(ptr, usize,	\
-				    rzsize, false);			\
-				if (copy_size > 0) {			\
-					VALGRIND_MAKE_MEM_DEFINED(ptr,	\
-					    copy_size);			\
-				}					\
-				if (zero && tail_size > 0) {		\
-					VALGRIND_MAKE_MEM_DEFINED(	\
-					    (void *)((uintptr_t)ptr +	\
-					    copy_size), tail_size);	\
-				}					\
-			}						\
-		}							\
-	}								\
-} while (0)
-#define	JEMALLOC_VALGRIND_FREE(ptr, rzsize) do {			\
-	if (config_valgrind && in_valgrind)				\
-		VALGRIND_FREELIKE_BLOCK(ptr, rzsize);			\
-} while (0)
-#else
-#define	RUNNING_ON_VALGRIND	((unsigned)0)
-#define	VALGRIND_MALLOCLIKE_BLOCK(addr, sizeB, rzB, is_zeroed) \
-    do {} while (0)
-#define	VALGRIND_RESIZEINPLACE_BLOCK(addr, oldSizeB, newSizeB, rzB) \
-    do {} while (0)
-#define	VALGRIND_FREELIKE_BLOCK(addr, rzB) do {} while (0)
-#define	VALGRIND_MAKE_MEM_NOACCESS(_qzz_addr, _qzz_len) do {} while (0)
-#define	VALGRIND_MAKE_MEM_UNDEFINED(_qzz_addr, _qzz_len) do {} while (0)
-#define	VALGRIND_MAKE_MEM_DEFINED(_qzz_addr, _qzz_len) do {} while (0)
-#define	JEMALLOC_VALGRIND_MALLOC(cond, ptr, usize, zero) do {} while (0)
-#define	JEMALLOC_VALGRIND_REALLOC(ptr, usize, old_ptr, old_usize,	\
-    old_rzsize, zero) do {} while (0)
-#define	JEMALLOC_VALGRIND_FREE(ptr, rzsize) do {} while (0)
-#endif
-
+#include "jemalloc/internal/valgrind.h"
 #include "jemalloc/internal/util.h"
 #include "jemalloc/internal/atomic.h"
 #include "jemalloc/internal/prng.h"
@ -463,6 +384,7 @@ static const bool config_ivsalloc =
 /******************************************************************************/
 #define	JEMALLOC_H_STRUCTS

+#include "jemalloc/internal/valgrind.h"
 #include "jemalloc/internal/util.h"
 #include "jemalloc/internal/atomic.h"
 #include "jemalloc/internal/prng.h"
@ -534,6 +456,7 @@ void	jemalloc_prefork(void);
 void	jemalloc_postfork_parent(void);
 void	jemalloc_postfork_child(void);

+#include "jemalloc/internal/valgrind.h"
 #include "jemalloc/internal/util.h"
 #include "jemalloc/internal/atomic.h"
 #include "jemalloc/internal/prng.h"
@ -560,6 +483,7 @@ void	jemalloc_postfork_child(void);
 /******************************************************************************/
 #define	JEMALLOC_H_INLINES

+#include "jemalloc/internal/valgrind.h"
 #include "jemalloc/internal/util.h"
 #include "jemalloc/internal/atomic.h"
 #include "jemalloc/internal/prng.h"
--- a/include/jemalloc/internal/private_symbols.txt
+++ b/include/jemalloc/internal/private_symbols.txt
@ -409,3 +409,7 @@ thread_allocated_tsd_set
 tsd_init_check_recursion
 tsd_init_finish
 u2rz
+valgrind_freelike_block
+valgrind_make_mem_defined
+valgrind_make_mem_noaccess
+valgrind_make_mem_undefined
--- a/include/jemalloc/internal/tcache.h
+++ b/include/jemalloc/internal/tcache.h
@ -314,13 +314,11 @@ tcache_alloc_small(tcache_t *tcache, size_t size, bool zero)
 			} else if (opt_zero)
 				memset(ret, 0, size);
 		}
-		VALGRIND_MAKE_MEM_UNDEFINED(ret, size);
 	} else {
 		if (config_fill && opt_junk) {
 			arena_alloc_junk_small(ret, &arena_bin_info[binind],
 			    true);
 		}
-		VALGRIND_MAKE_MEM_UNDEFINED(ret, size);
 		memset(ret, 0, size);
 	}

@ -369,11 +367,8 @@ tcache_alloc_large(tcache_t *tcache, size_t size, bool zero)
 				else if (opt_zero)
 					memset(ret, 0, size);
 			}
-			VALGRIND_MAKE_MEM_UNDEFINED(ret, size);
-		} else {
-			VALGRIND_MAKE_MEM_UNDEFINED(ret, size);
+		} else
 			memset(ret, 0, size);
-		}

 		if (config_stats)
 			tbin->tstats.nrequests++;
--- a/include/jemalloc/internal/valgrind.h
+++ b/include/jemalloc/internal/valgrind.h
@ -0,0 +1,112 @@
+/******************************************************************************/
+#ifdef JEMALLOC_H_TYPES
+
+#ifdef JEMALLOC_VALGRIND
+#include <valgrind/valgrind.h>
+
+/*
+ * The size that is reported to Valgrind must be consistent through a chain of
+ * malloc..realloc..realloc calls.  Request size isn't recorded anywhere in
+ * jemalloc, so it is critical that all callers of these macros provide usize
+ * rather than request size.  As a result, buffer overflow detection is
+ * technically weakened for the standard API, though it is generally accepted
+ * practice to consider any extra bytes reported by malloc_usable_size() as
+ * usable space.
+ */
+#define	JEMALLOC_VALGRIND_MAKE_MEM_NOACCESS(ptr, usize) do {		\
+	if (in_valgrind)						\
+		valgrind_make_mem_noaccess(ptr, usize);			\
+} while (0)
+#define	JEMALLOC_VALGRIND_MAKE_MEM_UNDEFINED(ptr, usize) do {		\
+	if (in_valgrind)						\
+		valgrind_make_mem_undefined(ptr, usize);		\
+} while (0)
+#define	JEMALLOC_VALGRIND_MAKE_MEM_DEFINED(ptr, usize) do {		\
+	if (in_valgrind)						\
+		valgrind_make_mem_defined(ptr, usize);			\
+} while (0)
+/*
+ * The VALGRIND_MALLOCLIKE_BLOCK() and VALGRIND_RESIZEINPLACE_BLOCK() macro
+ * calls must be embedded in macros rather than in functions so that when
+ * Valgrind reports errors, there are no extra stack frames in the backtraces.
+ */
+#define	JEMALLOC_VALGRIND_MALLOC(cond, ptr, usize, zero) do {		\
+	if (in_valgrind && cond)					\
+		VALGRIND_MALLOCLIKE_BLOCK(ptr, usize, p2rz(ptr), zero);	\
+} while (0)
+#define	JEMALLOC_VALGRIND_REALLOC(maybe_moved, ptr, usize,		\
+    ptr_maybe_null, old_ptr, old_usize, old_rzsize, old_ptr_maybe_null,	\
+    zero) do {								\
+	if (in_valgrind) {						\
+		size_t rzsize = p2rz(ptr);				\
+									\
+		if (!maybe_moved || ptr == old_ptr) {			\
+			VALGRIND_RESIZEINPLACE_BLOCK(ptr, old_usize,	\
+			    usize, rzsize);				\
+			if (zero && old_usize < usize) {		\
+				valgrind_make_mem_defined(		\
+				    (void *)((uintptr_t)ptr +		\
+				    old_usize), usize - old_usize);	\
+			}						\
+		} else {						\
+			if (!old_ptr_maybe_null || old_ptr != NULL) {	\
+				valgrind_freelike_block(old_ptr,	\
+				    old_rzsize);			\
+			}						\
+			if (!ptr_maybe_null || ptr != NULL) {		\
+				size_t copy_size = (old_usize < usize)	\
+				    ?  old_usize : usize;		\
+				size_t tail_size = usize - copy_size;	\
+				VALGRIND_MALLOCLIKE_BLOCK(ptr, usize,	\
+				    rzsize, false);			\
+				if (copy_size > 0) {			\
+					valgrind_make_mem_defined(ptr,	\
+					copy_size);			\
+				}					\
+				if (zero && tail_size > 0) {		\
+					valgrind_make_mem_defined(	\
+					    (void *)((uintptr_t)ptr +	\
+					    copy_size), tail_size);	\
+				}					\
+			}						\
+		}							\
+	}								\
+} while (0)
+#define	JEMALLOC_VALGRIND_FREE(ptr, rzsize) do {			\
+	if (in_valgrind)						\
+		valgrind_freelike_block(ptr, rzsize);			\
+} while (0)
+#else
+#define	RUNNING_ON_VALGRIND	((unsigned)0)
+#define	JEMALLOC_VALGRIND_MAKE_MEM_NOACCESS(ptr, usize) do {} while (0)
+#define	JEMALLOC_VALGRIND_MAKE_MEM_UNDEFINED(ptr, usize) do {} while (0)
+#define	JEMALLOC_VALGRIND_MAKE_MEM_DEFINED(ptr, usize) do {} while (0)
+#define	JEMALLOC_VALGRIND_MALLOC(cond, ptr, usize, zero) do {} while (0)
+#define	JEMALLOC_VALGRIND_REALLOC(maybe_moved, ptr, usize,		\
+    ptr_maybe_null, old_ptr, old_usize, old_rzsize, old_ptr_maybe_null,	\
+    zero) do {} while (0)
+#define	JEMALLOC_VALGRIND_FREE(ptr, rzsize) do {} while (0)
+#endif
+
+#endif /* JEMALLOC_H_TYPES */
+/******************************************************************************/
+#ifdef JEMALLOC_H_STRUCTS
+
+#endif /* JEMALLOC_H_STRUCTS */
+/******************************************************************************/
+#ifdef JEMALLOC_H_EXTERNS
+
+#ifdef JEMALLOC_VALGRIND
+void	valgrind_make_mem_noaccess(void *ptr, size_t usize);
+void	valgrind_make_mem_undefined(void *ptr, size_t usize);
+void	valgrind_make_mem_defined(void *ptr, size_t usize);
+void	valgrind_freelike_block(void *ptr, size_t usize);
+#endif
+
+#endif /* JEMALLOC_H_EXTERNS */
+/******************************************************************************/
+#ifdef JEMALLOC_H_INLINES
+
+#endif /* JEMALLOC_H_INLINES */
+/******************************************************************************/
+
--- a/src/arena.c
+++ b/src/arena.c
@ -337,8 +337,8 @@ static inline void
 arena_run_zero(arena_chunk_t *chunk, size_t run_ind, size_t npages)
 {

-	VALGRIND_MAKE_MEM_UNDEFINED((void *)((uintptr_t)chunk + (run_ind <<
-	    LG_PAGE)), (npages << LG_PAGE));
+	JEMALLOC_VALGRIND_MAKE_MEM_UNDEFINED((void *)((uintptr_t)chunk +
+	    (run_ind << LG_PAGE)), (npages << LG_PAGE));
 	memset((void *)((uintptr_t)chunk + (run_ind << LG_PAGE)), 0,
 	    (npages << LG_PAGE));
 }
@ -347,8 +347,8 @@ static inline void
 arena_run_page_mark_zeroed(arena_chunk_t *chunk, size_t run_ind)
 {

-	VALGRIND_MAKE_MEM_DEFINED((void *)((uintptr_t)chunk + (run_ind <<
-	    LG_PAGE)), PAGE);
+	JEMALLOC_VALGRIND_MAKE_MEM_DEFINED((void *)((uintptr_t)chunk + (run_ind
+	    << LG_PAGE)), PAGE);
 }

 static inline void
@ -457,7 +457,7 @@ arena_run_split_large_helper(arena_t *arena, arena_run_t *run, size_t size,
 			arena_run_zero(chunk, run_ind, need_pages);
 		}
 	} else {
-		VALGRIND_MAKE_MEM_UNDEFINED((void *)((uintptr_t)chunk +
+		JEMALLOC_VALGRIND_MAKE_MEM_UNDEFINED((void *)((uintptr_t)chunk +
 		    (run_ind << LG_PAGE)), (need_pages << LG_PAGE));
 	}

@ -525,7 +525,7 @@ arena_run_split_small(arena_t *arena, arena_run_t *run, size_t size,
 	if (config_debug && flag_dirty == 0 && arena_mapbits_unzeroed_get(chunk,
 	    run_ind+need_pages-1) == 0)
 		arena_run_page_validate_zeroed(chunk, run_ind+need_pages-1);
-	VALGRIND_MAKE_MEM_UNDEFINED((void *)((uintptr_t)chunk +
+	JEMALLOC_VALGRIND_MAKE_MEM_UNDEFINED((void *)((uintptr_t)chunk +
 	    (run_ind << LG_PAGE)), (need_pages << LG_PAGE));
 }

@ -592,14 +592,14 @@ arena_chunk_init_hard(arena_t *arena)
 	 * the chunk is not zeroed.
 	 */
 	if (zero == false) {
-		VALGRIND_MAKE_MEM_UNDEFINED((void *)arena_mapp_get(chunk,
-		    map_bias+1), (size_t)((uintptr_t) arena_mapp_get(chunk,
-		    chunk_npages-1) - (uintptr_t)arena_mapp_get(chunk,
-		    map_bias+1)));
+		JEMALLOC_VALGRIND_MAKE_MEM_UNDEFINED(
+		    (void *)arena_mapp_get(chunk, map_bias+1),
+		    (size_t)((uintptr_t) arena_mapp_get(chunk, chunk_npages-1) -
+		    (uintptr_t)arena_mapp_get(chunk, map_bias+1)));
 		for (i = map_bias+1; i < chunk_npages-1; i++)
 			arena_mapbits_unzeroed_set(chunk, i, unzeroed);
 	} else {
-		VALGRIND_MAKE_MEM_DEFINED((void *)arena_mapp_get(chunk,
+		JEMALLOC_VALGRIND_MAKE_MEM_DEFINED((void *)arena_mapp_get(chunk,
 		    map_bias+1), (size_t)((uintptr_t) arena_mapp_get(chunk,
 		    chunk_npages-1) - (uintptr_t)arena_mapp_get(chunk,
 		    map_bias+1)));
@ -1645,13 +1645,13 @@ arena_malloc_small(arena_t *arena, size_t size, bool zero)
 			} else if (opt_zero)
 				memset(ret, 0, size);
 		}
-		VALGRIND_MAKE_MEM_UNDEFINED(ret, size);
+		JEMALLOC_VALGRIND_MAKE_MEM_UNDEFINED(ret, size);
 	} else {
 		if (config_fill && opt_junk) {
 			arena_alloc_junk_small(ret, &arena_bin_info[binind],
 			    true);
 		}
-		VALGRIND_MAKE_MEM_UNDEFINED(ret, size);
+		JEMALLOC_VALGRIND_MAKE_MEM_UNDEFINED(ret, size);
 		memset(ret, 0, size);
 	}

@ -2226,7 +2226,7 @@ arena_ralloc(arena_t *arena, void *ptr, size_t oldsize, size_t size,
 	 * expectation that the extra bytes will be reliably preserved.
 	 */
 	copysize = (size < oldsize) ? size : oldsize;
-	VALGRIND_MAKE_MEM_UNDEFINED(ret, copysize);
+	JEMALLOC_VALGRIND_MAKE_MEM_UNDEFINED(ret, copysize);
 	memcpy(ret, ptr, copysize);
 	iqalloct(ptr, try_tcache_dalloc);
 	return (ret);
--- a/src/base.c
+++ b/src/base.c
@ -63,7 +63,7 @@ base_alloc(size_t size)
 	ret = base_next_addr;
 	base_next_addr = (void *)((uintptr_t)base_next_addr + csize);
 	malloc_mutex_unlock(&base_mtx);
-	VALGRIND_MAKE_MEM_UNDEFINED(ret, csize);
+	JEMALLOC_VALGRIND_MAKE_MEM_UNDEFINED(ret, csize);

 	return (ret);
 }
@ -89,7 +89,8 @@ base_node_alloc(void)
 		ret = base_nodes;
 		base_nodes = *(extent_node_t **)ret;
 		malloc_mutex_unlock(&base_mtx);
-		VALGRIND_MAKE_MEM_UNDEFINED(ret, sizeof(extent_node_t));
+		JEMALLOC_VALGRIND_MAKE_MEM_UNDEFINED(ret,
+		    sizeof(extent_node_t));
 	} else {
 		malloc_mutex_unlock(&base_mtx);
 		ret = (extent_node_t *)base_alloc(sizeof(extent_node_t));
@ -102,7 +103,7 @@ void
 base_node_dealloc(extent_node_t *node)
 {

-	VALGRIND_MAKE_MEM_UNDEFINED(node, sizeof(extent_node_t));
+	JEMALLOC_VALGRIND_MAKE_MEM_UNDEFINED(node, sizeof(extent_node_t));
 	malloc_mutex_lock(&base_mtx);
 	*(extent_node_t **)node = base_nodes;
 	base_nodes = node;
--- a/src/chunk.c
+++ b/src/chunk.c
@ -127,7 +127,7 @@ chunk_recycle(extent_tree_t *chunks_szad, extent_tree_t *chunks_ad, size_t size,
 			size_t i;
 			size_t *p = (size_t *)(uintptr_t)ret;

-			VALGRIND_MAKE_MEM_DEFINED(ret, size);
+			JEMALLOC_VALGRIND_MAKE_MEM_DEFINED(ret, size);
 			for (i = 0; i < size / sizeof(size_t); i++)
 				assert(p[i] == 0);
 		}
@ -203,7 +203,7 @@ label_return:
 				prof_gdump();
 		}
 		if (config_valgrind)
-			VALGRIND_MAKE_MEM_UNDEFINED(ret, size);
+			JEMALLOC_VALGRIND_MAKE_MEM_UNDEFINED(ret, size);
 	}
 	assert(CHUNK_ADDR2BASE(ret) == ret);
 	return (ret);
@ -217,7 +217,7 @@ chunk_record(extent_tree_t *chunks_szad, extent_tree_t *chunks_ad, void *chunk,
 	extent_node_t *xnode, *node, *prev, *xprev, key;

 	unzeroed = pages_purge(chunk, size);
-	VALGRIND_MAKE_MEM_NOACCESS(chunk, size);
+	JEMALLOC_VALGRIND_MAKE_MEM_NOACCESS(chunk, size);

 	/*
 	 * Allocate a node before acquiring chunks_mtx even though it might not
--- a/src/chunk_dss.c
+++ b/src/chunk_dss.c
@ -126,7 +126,8 @@ chunk_alloc_dss(size_t size, size_t alignment, bool *zero)
 				if (cpad_size != 0)
 					chunk_unmap(cpad, cpad_size);
 				if (*zero) {
-					VALGRIND_MAKE_MEM_UNDEFINED(ret, size);
+					JEMALLOC_VALGRIND_MAKE_MEM_UNDEFINED(
+					    ret, size);
 					memset(ret, 0, size);
 				}
 				return (ret);
--- a/src/jemalloc.c
+++ b/src/jemalloc.c
@ -479,9 +479,10 @@ malloc_conf_init(void)

 		while (*opts != '\0' && malloc_conf_next(&opts, &k, &klen, &v,
 		    &vlen) == false) {
-#define	CONF_HANDLE_BOOL(o, n)						\
-			if (sizeof(n)-1 == klen && strncmp(n, k,	\
-			    klen) == 0) {				\
+#define	CONF_MATCH(n)							\
+	(sizeof(n)-1 == klen && strncmp(n, k, klen) == 0)
+#define	CONF_HANDLE_BOOL(o, n, cont)					\
+			if (CONF_MATCH(n)) {				\
 				if (strncmp("true", v, vlen) == 0 &&	\
 				    vlen == sizeof("true")-1)		\
 					o = true;			\
@ -493,11 +494,11 @@ malloc_conf_init(void)
 					    "Invalid conf value",	\
 					    k, klen, v, vlen);		\
 				}					\
-				continue;				\
+				if (cont)				\
+					continue;			\
 			}
 #define	CONF_HANDLE_SIZE_T(o, n, min, max, clip)			\
-			if (sizeof(n)-1 == klen && strncmp(n, k,	\
-			    klen) == 0) {				\
+			if (CONF_MATCH(n)) {				\
 				uintmax_t um;				\
 				char *end;				\
 									\
@ -528,8 +529,7 @@ malloc_conf_init(void)
 				continue;				\
 			}
 #define	CONF_HANDLE_SSIZE_T(o, n, min, max)				\
-			if (sizeof(n)-1 == klen && strncmp(n, k,	\
-			    klen) == 0) {				\
+			if (CONF_MATCH(n)) {				\
 				long l;					\
 				char *end;				\
 									\
@ -550,8 +550,7 @@ malloc_conf_init(void)
 				continue;				\
 			}
 #define	CONF_HANDLE_CHAR_P(o, n, d)					\
-			if (sizeof(n)-1 == klen && strncmp(n, k,	\
-			    klen) == 0) {				\
+			if (CONF_MATCH(n)) {				\
 				size_t cpylen = (vlen <=		\
 				    sizeof(o)-1) ? vlen :		\
 				    sizeof(o)-1;			\
@ -560,7 +559,7 @@ malloc_conf_init(void)
 				continue;				\
 			}

-			CONF_HANDLE_BOOL(opt_abort, "abort")
+			CONF_HANDLE_BOOL(opt_abort, "abort", true)
 			/*
 			 * Chunks always require at least one header page, plus
 			 * one data page in the absence of redzones, or three
@ -599,44 +598,62 @@ malloc_conf_init(void)
 			    SIZE_T_MAX, false)
 			CONF_HANDLE_SSIZE_T(opt_lg_dirty_mult, "lg_dirty_mult",
 			    -1, (sizeof(size_t) << 3) - 1)
-			CONF_HANDLE_BOOL(opt_stats_print, "stats_print")
+			CONF_HANDLE_BOOL(opt_stats_print, "stats_print", true)
 			if (config_fill) {
-				CONF_HANDLE_BOOL(opt_junk, "junk")
+				CONF_HANDLE_BOOL(opt_junk, "junk", true)
 				CONF_HANDLE_SIZE_T(opt_quarantine, "quarantine",
 				    0, SIZE_T_MAX, false)
-				CONF_HANDLE_BOOL(opt_redzone, "redzone")
-				CONF_HANDLE_BOOL(opt_zero, "zero")
+				CONF_HANDLE_BOOL(opt_redzone, "redzone", true)
+				CONF_HANDLE_BOOL(opt_zero, "zero", true)
 			}
 			if (config_utrace) {
-				CONF_HANDLE_BOOL(opt_utrace, "utrace")
+				CONF_HANDLE_BOOL(opt_utrace, "utrace", true)
 			}
 			if (config_xmalloc) {
-				CONF_HANDLE_BOOL(opt_xmalloc, "xmalloc")
+				CONF_HANDLE_BOOL(opt_xmalloc, "xmalloc", true)
 			}
 			if (config_tcache) {
-				CONF_HANDLE_BOOL(opt_tcache, "tcache")
+				CONF_HANDLE_BOOL(opt_tcache, "tcache",
+				    !config_valgrind || !in_valgrind)
+				if (CONF_MATCH("tcache")) {
+					assert(config_valgrind && in_valgrind);
+					if (opt_tcache) {
+						opt_tcache = false;
+						malloc_conf_error(
+						"tcache cannot be enabled "
+						"while running inside Valgrind",
+						k, klen, v, vlen);
+					}
+					continue;
+				}
 				CONF_HANDLE_SSIZE_T(opt_lg_tcache_max,
 				    "lg_tcache_max", -1,
 				    (sizeof(size_t) << 3) - 1)
 			}
 			if (config_prof) {
-				CONF_HANDLE_BOOL(opt_prof, "prof")
+				CONF_HANDLE_BOOL(opt_prof, "prof", true)
 				CONF_HANDLE_CHAR_P(opt_prof_prefix,
 				    "prof_prefix", "jeprof")
-				CONF_HANDLE_BOOL(opt_prof_active, "prof_active")
+				CONF_HANDLE_BOOL(opt_prof_active, "prof_active",
+				    true)
 				CONF_HANDLE_SSIZE_T(opt_lg_prof_sample,
 				    "lg_prof_sample", 0,
 				    (sizeof(uint64_t) << 3) - 1)
-				CONF_HANDLE_BOOL(opt_prof_accum, "prof_accum")
+				CONF_HANDLE_BOOL(opt_prof_accum, "prof_accum",
+				    true)
 				CONF_HANDLE_SSIZE_T(opt_lg_prof_interval,
 				    "lg_prof_interval", -1,
 				    (sizeof(uint64_t) << 3) - 1)
-				CONF_HANDLE_BOOL(opt_prof_gdump, "prof_gdump")
-				CONF_HANDLE_BOOL(opt_prof_final, "prof_final")
-				CONF_HANDLE_BOOL(opt_prof_leak, "prof_leak")
+				CONF_HANDLE_BOOL(opt_prof_gdump, "prof_gdump",
+				    true)
+				CONF_HANDLE_BOOL(opt_prof_final, "prof_final",
+				    true)
+				CONF_HANDLE_BOOL(opt_prof_leak, "prof_leak",
+				    true)
 			}
 			malloc_conf_error("Invalid conf pair", k, klen, v,
 			    vlen);
+#undef CONF_MATCH
 #undef CONF_HANDLE_BOOL
 #undef CONF_HANDLE_SIZE_T
 #undef CONF_HANDLE_SSIZE_T
@ -1293,8 +1310,8 @@ je_realloc(void *ptr, size_t size)
 		ta->deallocated += old_usize;
 	}
 	UTRACE(ptr, size, ret);
-	JEMALLOC_VALGRIND_REALLOC(ret, usize, ptr, old_usize, old_rzsize,
-	    false);
+	JEMALLOC_VALGRIND_REALLOC(true, ret, usize, true, ptr, old_usize,
+	    old_rzsize, true, false);
 	return (ret);
 }

@ -1604,7 +1621,8 @@ je_rallocx(void *ptr, size_t size, int flags)
 		ta->deallocated += old_usize;
 	}
 	UTRACE(ptr, size, p);
-	JEMALLOC_VALGRIND_REALLOC(p, usize, ptr, old_usize, old_rzsize, zero);
+	JEMALLOC_VALGRIND_REALLOC(true, p, usize, false, ptr, old_usize,
+	    old_rzsize, false, zero);
 	return (p);
 label_oom:
 	if (config_xmalloc && opt_xmalloc) {
@ -1731,7 +1749,8 @@ je_xallocx(void *ptr, size_t size, size_t extra, int flags)
 		ta->allocated += usize;
 		ta->deallocated += old_usize;
 	}
-	JEMALLOC_VALGRIND_REALLOC(ptr, usize, ptr, old_usize, old_rzsize, zero);
+	JEMALLOC_VALGRIND_REALLOC(false, ptr, usize, false, ptr, old_usize,
+	    old_rzsize, false, zero);
 label_not_resized:
 	UTRACE(ptr, size, ptr);
 	return (usize);
--- a/src/valgrind.c
+++ b/src/valgrind.c
@ -0,0 +1,34 @@
+#include "jemalloc/internal/jemalloc_internal.h"
+#ifndef JEMALLOC_VALGRIND
+#  error "This source file is for Valgrind integration."
+#endif
+
+#include <valgrind/memcheck.h>
+
+void
+valgrind_make_mem_noaccess(void *ptr, size_t usize)
+{
+
+	VALGRIND_MAKE_MEM_NOACCESS(ptr, usize);
+}
+
+void
+valgrind_make_mem_undefined(void *ptr, size_t usize)
+{
+
+	VALGRIND_MAKE_MEM_UNDEFINED(ptr, usize);
+}
+
+void
+valgrind_make_mem_defined(void *ptr, size_t usize)
+{
+
+	VALGRIND_MAKE_MEM_DEFINED(ptr, usize);
+}
+
+void
+valgrind_freelike_block(void *ptr, size_t usize)
+{
+
+	VALGRIND_FREELIKE_BLOCK(ptr, usize);
+}