project-base/server-skynet-source-3rd-jemalloc
2
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Optimize Valgrind integration.

Browse Source 
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().
This commit is contained in:
Jason Evans
2014-04-15 16:35:08 -07:00
parent ecd3e59ca3
commit bd87b01999
12 changed files with 231 additions and 136 deletions
Download Patch File Download Diff File Expand all files Collapse all files

84
include/jemalloc/internal/jemalloc_internal.h.in
View File

@@ -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"

4
include/jemalloc/internal/private_symbols.txt
View File

@@ -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

7
include/jemalloc/internal/tcache.h
View File

@@ -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++;

112
include/jemalloc/internal/valgrind.h Normal file
View File

@@ -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 */
/******************************************************************************/