project-base/server-skynet-source-3rd-jemalloc
2
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
a7619b7fa5
server-skynet-source-3rd-je.../src/prof.c

1418 lines
30 KiB
C
Raw Normal View History

Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
#define JEMALLOC_PROF_C_
Restructure source tree.
2010-02-12 06:45:59 +08:00
#include "jemalloc/internal/jemalloc_internal.h"
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
/******************************************************************************/
#ifdef JEMALLOC_PROF_LIBUNWIND
#define UNW_LOCAL_ONLY
#include <libunwind.h>
#endif
Improve backtracing-related configuration. Clean up configuration for backtracing when profiling is enabled, and document the configuration logic in INSTALL. Disable libgcc-based backtracing except on x64 (where it is known to work). Add the --disable-prof-gcc option.
2011-03-16 13:23:12 +08:00
#ifdef JEMALLOC_PROF_LIBGCC
#include <unwind.h>
#endif
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
/******************************************************************************/
/* Data. */
Implement tsd. Implement tsd, which is a TLS/TSD abstraction that uses one or both internally. Modify bootstrapping such that no tsd's are utilized until allocation is safe. Remove malloc_[v]tprintf(), and use malloc_snprintf() instead. Fix %p argument size handling in malloc_vsnprintf(). Fix a long-standing statistics-related bug in the "thread.arena" mallctl that could cause crashes due to linked list corruption.
2012-03-22 09:33:03 +08:00
malloc_tsd_data(, prof_tdata, prof_tdata_t *, NULL)
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
bool opt_prof = false;
Add sampling activation/deactivation control. Add the E/e options to control whether the application starts with sampling active/inactive (secondary control to F/f). Add the prof.active mallctl so that the application can activate/deactivate sampling on the fly.
2010-04-01 09:43:24 +08:00
bool opt_prof_active = true;
Implement sampling for heap profiling.
2010-03-02 12:15:26 +08:00
size_t opt_lg_prof_sample = LG_PROF_SAMPLE_DEFAULT;
Make interval-triggered profile dumping optional. Make it possible to disable interval-triggered profile dumping, even if profiling is enabled. This is useful if the user only wants a single dump at exit, or if the application manually triggers profile dumps.
2010-04-01 08:35:51 +08:00
ssize_t opt_lg_prof_interval = LG_PROF_INTERVAL_DEFAULT;
Replace JEMALLOC_OPTIONS with MALLOC_CONF. Replace the single-character run-time flags with key/value pairs, which can be set via the malloc_conf global, /etc/malloc.conf, and the MALLOC_CONF environment variable. Replace the JEMALLOC_PROF_PREFIX environment variable with the "opt.prof_prefix" option. Replace umax2s() with u2s().
2010-10-24 09:37:06 +08:00
bool opt_prof_gdump = false;
Update prof defaults to match common usage. Change the "opt.lg_prof_sample" default from 0 to 19 (1 B to 512 KiB). Change the "opt.prof_accum" default from true to false. Add the "opt.prof_final" mallctl, so that "opt.prof_prefix" need not be abused to disable final profile dumping.
2012-04-18 07:39:33 +08:00
bool opt_prof_final = true;
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
bool opt_prof_leak = false;
Update prof defaults to match common usage. Change the "opt.lg_prof_sample" default from 0 to 19 (1 B to 512 KiB). Change the "opt.prof_accum" default from true to false. Add the "opt.prof_final" mallctl, so that "opt.prof_prefix" need not be abused to disable final profile dumping.
2012-04-18 07:39:33 +08:00
bool opt_prof_accum = false;
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
char opt_prof_prefix[
/* Minimize memory bloat for non-prof builds. */
#ifdef JEMALLOC_PROF
PATH_MAX +
#endif
Fix a variable prototype/definition mismatch.
2014-01-17 10:04:30 +08:00
1];
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
Avoid arena_prof_accum()-related locking when possible. Refactor arena_prof_accum() and its callers to avoid arena locking when prof_interval is 0 (as when profiling is disabled). Reported by Ben Maurer.
2012-11-14 04:56:27 +08:00
uint64_t prof_interval = 0;
Implement interval-based heap profile dumping. Add mallctl interfaces for profiling parameters. Fix a file descriptor leak in heap profile dumping.
2010-02-12 05:19:21 +08:00
Remove ephemeral mutexes. Remove ephemeral mutexes from the prof machinery, and remove malloc_mutex_destroy(). This simplifies mutex management on systems that call malloc()/free() inside pthread_mutex_{create,destroy}(). Add atomic_*_u() for operation on unsigned values. Fix prof_printf() to call malloc_vsnprintf() rather than malloc_snprintf().
2012-03-24 09:05:51 +08:00
/*
* Table of mutexes that are shared among ctx's. These are leaf locks, so
* there is no problem with using them for more than one ctx at the same time.
* The primary motivation for this sharing though is that ctx's are ephemeral,
* and destroying mutexes causes complications for systems that allocate when
* creating/destroying mutexes.
*/
static malloc_mutex_t *ctx_locks;
static unsigned cum_ctxs; /* Atomic counter. */
Increase default backtrace depth from 4 to 128. Increase the default backtrace depth, because shallow backtraces tend to result in confusing pprof output graphs.
2010-10-03 13:38:14 +08:00
/*
Inline the fast path for heap sampling. Inline the heap sampling code that is executed for every allocation event (regardless of whether a sample is taken). Combine all prof TLS data into a single data structure, in order to reduce the TLS lookup volume.
2010-10-21 10:05:59 +08:00
* Global hash of (prof_bt_t *)-->(prof_ctx_t *). This is the master data
* structure that knows about all backtraces currently captured.
Increase default backtrace depth from 4 to 128. Increase the default backtrace depth, because shallow backtraces tend to result in confusing pprof output graphs.
2010-10-03 13:38:14 +08:00
*/
Inline the fast path for heap sampling. Inline the heap sampling code that is executed for every allocation event (regardless of whether a sample is taken). Combine all prof TLS data into a single data structure, in order to reduce the TLS lookup volume.
2010-10-21 10:05:59 +08:00
static ckh_t bt2ctx;
static malloc_mutex_t bt2ctx_mtx;
Implement sampling for heap profiling.
2010-03-02 12:15:26 +08:00
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
static malloc_mutex_t prof_dump_seq_mtx;
static uint64_t prof_dump_seq;
static uint64_t prof_dump_iseq;
static uint64_t prof_dump_mseq;
static uint64_t prof_dump_useq;
/*
* This buffer is rather large for stack allocation, so use a single buffer for
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
* all profile dumps.
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
*/
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
static malloc_mutex_t prof_dump_mtx;
static char prof_dump_buf[
/* Minimize memory bloat for non-prof builds. */
#ifdef JEMALLOC_PROF
PROF_DUMP_BUFSIZE
#else
1
#endif
];
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
static unsigned prof_dump_buf_end;
static int prof_dump_fd;
/* Do not dump any profiles until bootstrapping is complete. */
static bool prof_booted = false;
/******************************************************************************/
Inline the fast path for heap sampling. Inline the heap sampling code that is executed for every allocation event (regardless of whether a sample is taken). Combine all prof TLS data into a single data structure, in order to reduce the TLS lookup volume.
2010-10-21 10:05:59 +08:00
void
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
bt_init(prof_bt_t *bt, void **vec)
{
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
bt->vec = vec;
bt->len = 0;
}
Make cumulative heap profile data optional. Add the R option to control whether cumulative heap profile data are maintained. Add the T option to control the size of per thread backtrace caches, primarily because when the R option is specified, backtraces that no longer have allocations associated with them are discarded as soon as no thread caches refer to them.
2010-10-03 06:18:50 +08:00
static void
bt_destroy(prof_bt_t *bt)
{
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Make cumulative heap profile data optional. Add the R option to control whether cumulative heap profile data are maintained. Add the T option to control the size of per thread backtrace caches, primarily because when the R option is specified, backtraces that no longer have allocations associated with them are discarded as soon as no thread caches refer to them.
2010-10-03 06:18:50 +08:00
idalloc(bt);
}
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
static prof_bt_t *
bt_dup(prof_bt_t *bt)
{
prof_bt_t *ret;
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
/*
* Create a single allocation that has space for vec immediately
* following the prof_bt_t structure. The backtraces that get
* stored in the backtrace caches are copied from stack-allocated
* temporary variables, so size is known at creation time. Making this
* a contiguous object improves cache locality.
*/
ret = (prof_bt_t *)imalloc(QUANTUM_CEILING(sizeof(prof_bt_t)) +
(bt->len * sizeof(void *)));
if (ret == NULL)
return (NULL);
ret->vec = (void **)((uintptr_t)ret +
QUANTUM_CEILING(sizeof(prof_bt_t)));
memcpy(ret->vec, bt->vec, bt->len * sizeof(void *));
ret->len = bt->len;
return (ret);
}
static inline void
Fix heap profiling bugs. Fix a potential deadlock that could occur during interval- and growth-triggered heap profile dumps. Fix an off-by-one heap profile statistics bug that could be observed in interval- and growth-triggered heap profiles. Fix heap profile dump filename sequence numbers (regression during conversion to malloc_snprintf()).
2012-04-23 07:00:11 +08:00
prof_enter(prof_tdata_t *prof_tdata)
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
{
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Fix heap profiling bugs. Fix a potential deadlock that could occur during interval- and growth-triggered heap profile dumps. Fix an off-by-one heap profile statistics bug that could be observed in interval- and growth-triggered heap profiles. Fix heap profile dump filename sequence numbers (regression during conversion to malloc_snprintf()).
2012-04-23 07:00:11 +08:00
assert(prof_tdata->enq == false);
prof_tdata->enq = true;
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
malloc_mutex_lock(&bt2ctx_mtx);
}
static inline void
Fix heap profiling bugs. Fix a potential deadlock that could occur during interval- and growth-triggered heap profile dumps. Fix an off-by-one heap profile statistics bug that could be observed in interval- and growth-triggered heap profiles. Fix heap profile dump filename sequence numbers (regression during conversion to malloc_snprintf()).
2012-04-23 07:00:11 +08:00
prof_leave(prof_tdata_t *prof_tdata)
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
{
Replace JEMALLOC_OPTIONS with MALLOC_CONF. Replace the single-character run-time flags with key/value pairs, which can be set via the malloc_conf global, /etc/malloc.conf, and the MALLOC_CONF environment variable. Replace the JEMALLOC_PROF_PREFIX environment variable with the "opt.prof_prefix" option. Replace umax2s() with u2s().
2010-10-24 09:37:06 +08:00
bool idump, gdump;
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
malloc_mutex_unlock(&bt2ctx_mtx);
Fix heap profiling bugs. Fix a potential deadlock that could occur during interval- and growth-triggered heap profile dumps. Fix an off-by-one heap profile statistics bug that could be observed in interval- and growth-triggered heap profiles. Fix heap profile dump filename sequence numbers (regression during conversion to malloc_snprintf()).
2012-04-23 07:00:11 +08:00
assert(prof_tdata->enq);
prof_tdata->enq = false;
idump = prof_tdata->enq_idump;
prof_tdata->enq_idump = false;
gdump = prof_tdata->enq_gdump;
prof_tdata->enq_gdump = false;
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
Implement interval-based heap profile dumping. Add mallctl interfaces for profiling parameters. Fix a file descriptor leak in heap profile dumping.
2010-02-12 05:19:21 +08:00
if (idump)
prof_idump();
Replace JEMALLOC_OPTIONS with MALLOC_CONF. Replace the single-character run-time flags with key/value pairs, which can be set via the malloc_conf global, /etc/malloc.conf, and the MALLOC_CONF environment variable. Replace the JEMALLOC_PROF_PREFIX environment variable with the "opt.prof_prefix" option. Replace umax2s() with u2s().
2010-10-24 09:37:06 +08:00
if (gdump)
prof_gdump();
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
}
Improve backtracing-related configuration. Clean up configuration for backtracing when profiling is enabled, and document the configuration logic in INSTALL. Disable libgcc-based backtracing except on x64 (where it is known to work). Add the --disable-prof-gcc option.
2011-03-16 13:23:12 +08:00
#ifdef JEMALLOC_PROF_LIBUNWIND
Inline the fast path for heap sampling. Inline the heap sampling code that is executed for every allocation event (regardless of whether a sample is taken). Combine all prof TLS data into a single data structure, in order to reduce the TLS lookup volume.
2010-10-21 10:05:59 +08:00
void
Remove the opt.lg_prof_bt_max option. Remove opt.lg_prof_bt_max, and hard code it to 7. The original intention of this option was to enable faster backtracing by limiting backtrace depth. However, this makes graphical pprof output very difficult to interpret. In practice, decreasing sampling frequency is a better mechanism for limiting profiling overhead.
2012-02-14 10:23:41 +08:00
prof_backtrace(prof_bt_t *bt, unsigned nignore)
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
{
unw_context_t uc;
unw_cursor_t cursor;
unsigned i;
int err;
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
assert(bt->len == 0);
assert(bt->vec != NULL);
unw_getcontext(&uc);
unw_init_local(&cursor, &uc);
Revert "Add support for libunwind backtrace caching." This reverts commit adc675c8ef55b59bb2facf795a3c26411cfbf3ed. The original commit added support for a non-standard libunwind API, so it was not of general utility.
2011-03-23 11:44:40 +08:00
/* Throw away (nignore+1) stack frames, if that many exist. */
for (i = 0; i < nignore + 1; i++) {
err = unw_step(&cursor);
if (err <= 0)
return;
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
}
Revert "Add support for libunwind backtrace caching." This reverts commit adc675c8ef55b59bb2facf795a3c26411cfbf3ed. The original commit added support for a non-standard libunwind API, so it was not of general utility.
2011-03-23 11:44:40 +08:00
/*
* Iterate over stack frames until there are no more, or until no space
* remains in bt.
*/
Remove the opt.lg_prof_bt_max option. Remove opt.lg_prof_bt_max, and hard code it to 7. The original intention of this option was to enable faster backtracing by limiting backtrace depth. However, this makes graphical pprof output very difficult to interpret. In practice, decreasing sampling frequency is a better mechanism for limiting profiling overhead.
2012-02-14 10:23:41 +08:00
for (i = 0; i < PROF_BT_MAX; i++) {
Revert "Add support for libunwind backtrace caching." This reverts commit adc675c8ef55b59bb2facf795a3c26411cfbf3ed. The original commit added support for a non-standard libunwind API, so it was not of general utility.
2011-03-23 11:44:40 +08:00
unw_get_reg(&cursor, UNW_REG_IP, (unw_word_t *)&bt->vec[i]);
bt->len++;
err = unw_step(&cursor);
if (err <= 0)
break;
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
}
}
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
#elif (defined(JEMALLOC_PROF_LIBGCC))
Improve backtracing-related configuration. Clean up configuration for backtracing when profiling is enabled, and document the configuration logic in INSTALL. Disable libgcc-based backtracing except on x64 (where it is known to work). Add the --disable-prof-gcc option.
2011-03-16 13:23:12 +08:00
static _Unwind_Reason_Code
prof_unwind_init_callback(struct _Unwind_Context *context, void *arg)
{
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Improve backtracing-related configuration. Clean up configuration for backtracing when profiling is enabled, and document the configuration logic in INSTALL. Disable libgcc-based backtracing except on x64 (where it is known to work). Add the --disable-prof-gcc option.
2011-03-16 13:23:12 +08:00
return (_URC_NO_REASON);
}
static _Unwind_Reason_Code
prof_unwind_callback(struct _Unwind_Context *context, void *arg)
{
prof_unwind_data_t *data = (prof_unwind_data_t *)arg;
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Improve backtracing-related configuration. Clean up configuration for backtracing when profiling is enabled, and document the configuration logic in INSTALL. Disable libgcc-based backtracing except on x64 (where it is known to work). Add the --disable-prof-gcc option.
2011-03-16 13:23:12 +08:00
if (data->nignore > 0)
data->nignore--;
else {
data->bt->vec[data->bt->len] = (void *)_Unwind_GetIP(context);
data->bt->len++;
if (data->bt->len == data->max)
return (_URC_END_OF_STACK);
}
return (_URC_NO_REASON);
}
void
Remove the opt.lg_prof_bt_max option. Remove opt.lg_prof_bt_max, and hard code it to 7. The original intention of this option was to enable faster backtracing by limiting backtrace depth. However, this makes graphical pprof output very difficult to interpret. In practice, decreasing sampling frequency is a better mechanism for limiting profiling overhead.
2012-02-14 10:23:41 +08:00
prof_backtrace(prof_bt_t *bt, unsigned nignore)
Improve backtracing-related configuration. Clean up configuration for backtracing when profiling is enabled, and document the configuration logic in INSTALL. Disable libgcc-based backtracing except on x64 (where it is known to work). Add the --disable-prof-gcc option.
2011-03-16 13:23:12 +08:00
{
Remove the opt.lg_prof_bt_max option. Remove opt.lg_prof_bt_max, and hard code it to 7. The original intention of this option was to enable faster backtracing by limiting backtrace depth. However, this makes graphical pprof output very difficult to interpret. In practice, decreasing sampling frequency is a better mechanism for limiting profiling overhead.
2012-02-14 10:23:41 +08:00
prof_unwind_data_t data = {bt, nignore, PROF_BT_MAX};
Improve backtracing-related configuration. Clean up configuration for backtracing when profiling is enabled, and document the configuration logic in INSTALL. Disable libgcc-based backtracing except on x64 (where it is known to work). Add the --disable-prof-gcc option.
2011-03-16 13:23:12 +08:00
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Improve backtracing-related configuration. Clean up configuration for backtracing when profiling is enabled, and document the configuration logic in INSTALL. Disable libgcc-based backtracing except on x64 (where it is known to work). Add the --disable-prof-gcc option.
2011-03-16 13:23:12 +08:00
_Unwind_Backtrace(prof_unwind_callback, &data);
}
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
#elif (defined(JEMALLOC_PROF_GCC))
Inline the fast path for heap sampling. Inline the heap sampling code that is executed for every allocation event (regardless of whether a sample is taken). Combine all prof TLS data into a single data structure, in order to reduce the TLS lookup volume.
2010-10-21 10:05:59 +08:00
void
Remove the opt.lg_prof_bt_max option. Remove opt.lg_prof_bt_max, and hard code it to 7. The original intention of this option was to enable faster backtracing by limiting backtrace depth. However, this makes graphical pprof output very difficult to interpret. In practice, decreasing sampling frequency is a better mechanism for limiting profiling overhead.
2012-02-14 10:23:41 +08:00
prof_backtrace(prof_bt_t *bt, unsigned nignore)
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
{
#define BT_FRAME(i) \
Remove the opt.lg_prof_bt_max option. Remove opt.lg_prof_bt_max, and hard code it to 7. The original intention of this option was to enable faster backtracing by limiting backtrace depth. However, this makes graphical pprof output very difficult to interpret. In practice, decreasing sampling frequency is a better mechanism for limiting profiling overhead.
2012-02-14 10:23:41 +08:00
if ((i) < nignore + PROF_BT_MAX) { \
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
void *p; \
if (__builtin_frame_address(i) == 0) \
Various heap profiling improvements. Add the --disable-prof-libgcc configure option, and add backtracing based on libgcc, which is used by default. Fix a bug in hash(). Fix various configuration-dependent compilation errors.
2010-02-11 10:15:53 +08:00
return; \
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
p = __builtin_return_address(i); \
if (p == NULL) \
Various heap profiling improvements. Add the --disable-prof-libgcc configure option, and add backtracing based on libgcc, which is used by default. Fix a bug in hash(). Fix various configuration-dependent compilation errors.
2010-02-11 10:15:53 +08:00
return; \
Fix heap profiling bugs. Fix a regression due to the recent heap profiling accuracy improvements: prof_{m,re}alloc() must set the object's profiling context regardless of whether it is sampled. Fix management of the CHUNK_MAP_CLASS chunk map bits, such that all large object (re-)allocation paths correctly initialize the bits. Prior to this fix, in-place realloc() cleared the bits, resulting in incorrect reported object size from arena_salloc_demote(). After this fix the non-demoted bit pattern is all zeros (instead of all ones), which makes it easier to assure that the bits are properly set.
2010-10-23 01:45:59 +08:00
if (i >= nignore) { \
bt->vec[(i) - nignore] = p; \
bt->len = (i) - nignore + 1; \
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
} \
} else \
Various heap profiling improvements. Add the --disable-prof-libgcc configure option, and add backtracing based on libgcc, which is used by default. Fix a bug in hash(). Fix various configuration-dependent compilation errors.
2010-02-11 10:15:53 +08:00
return;
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Fix prof bugs. Fix a race condition in ctx destruction that could cause undefined behavior (deadlock observed). Add mutex unlocks to some OOM error paths.
2010-10-28 10:47:40 +08:00
assert(nignore <= 3);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
BT_FRAME(0)
BT_FRAME(1)
BT_FRAME(2)
BT_FRAME(3)
BT_FRAME(4)
BT_FRAME(5)
BT_FRAME(6)
BT_FRAME(7)
BT_FRAME(8)
BT_FRAME(9)
BT_FRAME(10)
BT_FRAME(11)
BT_FRAME(12)
BT_FRAME(13)
BT_FRAME(14)
BT_FRAME(15)
BT_FRAME(16)
BT_FRAME(17)
BT_FRAME(18)
BT_FRAME(19)
BT_FRAME(20)
BT_FRAME(21)
BT_FRAME(22)
BT_FRAME(23)
BT_FRAME(24)
BT_FRAME(25)
BT_FRAME(26)
BT_FRAME(27)
BT_FRAME(28)
BT_FRAME(29)
BT_FRAME(30)
BT_FRAME(31)
BT_FRAME(32)
BT_FRAME(33)
BT_FRAME(34)
BT_FRAME(35)
BT_FRAME(36)
BT_FRAME(37)
BT_FRAME(38)
BT_FRAME(39)
BT_FRAME(40)
BT_FRAME(41)
BT_FRAME(42)
BT_FRAME(43)
BT_FRAME(44)
BT_FRAME(45)
BT_FRAME(46)
BT_FRAME(47)
BT_FRAME(48)
BT_FRAME(49)
BT_FRAME(50)
BT_FRAME(51)
BT_FRAME(52)
BT_FRAME(53)
BT_FRAME(54)
BT_FRAME(55)
BT_FRAME(56)
BT_FRAME(57)
BT_FRAME(58)
BT_FRAME(59)
BT_FRAME(60)
BT_FRAME(61)
BT_FRAME(62)
BT_FRAME(63)
BT_FRAME(64)
BT_FRAME(65)
BT_FRAME(66)
BT_FRAME(67)
BT_FRAME(68)
BT_FRAME(69)
BT_FRAME(70)
BT_FRAME(71)
BT_FRAME(72)
BT_FRAME(73)
BT_FRAME(74)
BT_FRAME(75)
BT_FRAME(76)
BT_FRAME(77)
BT_FRAME(78)
BT_FRAME(79)
BT_FRAME(80)
BT_FRAME(81)
BT_FRAME(82)
BT_FRAME(83)
BT_FRAME(84)
BT_FRAME(85)
BT_FRAME(86)
BT_FRAME(87)
BT_FRAME(88)
BT_FRAME(89)
BT_FRAME(90)
BT_FRAME(91)
BT_FRAME(92)
BT_FRAME(93)
BT_FRAME(94)
BT_FRAME(95)
BT_FRAME(96)
BT_FRAME(97)
BT_FRAME(98)
BT_FRAME(99)
BT_FRAME(100)
BT_FRAME(101)
BT_FRAME(102)
BT_FRAME(103)
BT_FRAME(104)
BT_FRAME(105)
BT_FRAME(106)
BT_FRAME(107)
BT_FRAME(108)
BT_FRAME(109)
BT_FRAME(110)
BT_FRAME(111)
BT_FRAME(112)
BT_FRAME(113)
BT_FRAME(114)
BT_FRAME(115)
BT_FRAME(116)
BT_FRAME(117)
BT_FRAME(118)
BT_FRAME(119)
BT_FRAME(120)
BT_FRAME(121)
BT_FRAME(122)
BT_FRAME(123)
BT_FRAME(124)
BT_FRAME(125)
BT_FRAME(126)
BT_FRAME(127)
Fix prof bugs. Fix a race condition in ctx destruction that could cause undefined behavior (deadlock observed). Add mutex unlocks to some OOM error paths.
2010-10-28 10:47:40 +08:00
/* Extras to compensate for nignore. */
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
BT_FRAME(128)
BT_FRAME(129)
BT_FRAME(130)
#undef BT_FRAME
}
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
#else
void
Remove the opt.lg_prof_bt_max option. Remove opt.lg_prof_bt_max, and hard code it to 7. The original intention of this option was to enable faster backtracing by limiting backtrace depth. However, this makes graphical pprof output very difficult to interpret. In practice, decreasing sampling frequency is a better mechanism for limiting profiling overhead.
2012-02-14 10:23:41 +08:00
prof_backtrace(prof_bt_t *bt, unsigned nignore)
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
{
cassert(config_prof);
Prefer not_reached() over assert(false) where appropriate.
2013-10-22 05:56:27 +08:00
not_reached();
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
}
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
#endif
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
static malloc_mutex_t *
prof_ctx_mutex_choose(void)
{
unsigned nctxs = atomic_add_u(&cum_ctxs, 1);
return (&ctx_locks[(nctxs - 1) % PROF_NCTX_LOCKS]);
}
static void
prof_ctx_init(prof_ctx_t *ctx, prof_bt_t *bt)
{
ctx->bt = bt;
ctx->lock = prof_ctx_mutex_choose();
/*
* Set nlimbo to 1, in order to avoid a race condition with
* prof_ctx_merge()/prof_ctx_destroy().
*/
ctx->nlimbo = 1;
ql_elm_new(ctx, dump_link);
memset(&ctx->cnt_merged, 0, sizeof(prof_cnt_t));
ql_new(&ctx->cnts_ql);
}
static void
prof_ctx_destroy(prof_ctx_t *ctx)
{
prof_tdata_t *prof_tdata;
cassert(config_prof);
/*
* Check that ctx is still unused by any thread cache before destroying
* it. prof_lookup() increments ctx->nlimbo in order to avoid a race
* condition with this function, as does prof_ctx_merge() in order to
* avoid a race between the main body of prof_ctx_merge() and entry
* into this function.
*/
prof_tdata = prof_tdata_get(false);
assert((uintptr_t)prof_tdata > (uintptr_t)PROF_TDATA_STATE_MAX);
prof_enter(prof_tdata);
malloc_mutex_lock(ctx->lock);
if (ql_first(&ctx->cnts_ql) == NULL && ctx->cnt_merged.curobjs == 0 &&
ctx->nlimbo == 1) {
assert(ctx->cnt_merged.curbytes == 0);
assert(ctx->cnt_merged.accumobjs == 0);
assert(ctx->cnt_merged.accumbytes == 0);
/* Remove ctx from bt2ctx. */
if (ckh_remove(&bt2ctx, ctx->bt, NULL, NULL))
not_reached();
prof_leave(prof_tdata);
/* Destroy ctx. */
malloc_mutex_unlock(ctx->lock);
bt_destroy(ctx->bt);
idalloc(ctx);
} else {
/*
* Compensate for increment in prof_ctx_merge() or
* prof_lookup().
*/
ctx->nlimbo--;
malloc_mutex_unlock(ctx->lock);
prof_leave(prof_tdata);
}
}
static void
prof_ctx_merge(prof_ctx_t *ctx, prof_thr_cnt_t *cnt)
{
bool destroy;
cassert(config_prof);
/* Merge cnt stats and detach from ctx. */
malloc_mutex_lock(ctx->lock);
ctx->cnt_merged.curobjs += cnt->cnts.curobjs;
ctx->cnt_merged.curbytes += cnt->cnts.curbytes;
ctx->cnt_merged.accumobjs += cnt->cnts.accumobjs;
ctx->cnt_merged.accumbytes += cnt->cnts.accumbytes;
ql_remove(&ctx->cnts_ql, cnt, cnts_link);
if (opt_prof_accum == false && ql_first(&ctx->cnts_ql) == NULL &&
ctx->cnt_merged.curobjs == 0 && ctx->nlimbo == 0) {
/*
* Increment ctx->nlimbo in order to keep another thread from
* winning the race to destroy ctx while this one has ctx->lock
* dropped. Without this, it would be possible for another
* thread to:
*
* 1) Sample an allocation associated with ctx.
* 2) Deallocate the sampled object.
* 3) Successfully prof_ctx_destroy(ctx).
*
* The result would be that ctx no longer exists by the time
* this thread accesses it in prof_ctx_destroy().
*/
ctx->nlimbo++;
destroy = true;
} else
destroy = false;
malloc_mutex_unlock(ctx->lock);
if (destroy)
prof_ctx_destroy(ctx);
}
Refactor prof_lookup() by extracting prof_lookup_global().
2014-01-15 09:04:34 +08:00
static bool
prof_lookup_global(prof_bt_t *bt, prof_tdata_t *prof_tdata, void **p_btkey,
prof_ctx_t **p_ctx, bool *p_new_ctx)
{
union {
prof_ctx_t *p;
void *v;
} ctx;
union {
prof_bt_t *p;
void *v;
} btkey;
bool new_ctx;
prof_enter(prof_tdata);
if (ckh_search(&bt2ctx, bt, &btkey.v, &ctx.v)) {
/* bt has never been seen before. Insert it. */
ctx.v = imalloc(sizeof(prof_ctx_t));
if (ctx.v == NULL) {
prof_leave(prof_tdata);
return (true);
}
btkey.p = bt_dup(bt);
if (btkey.v == NULL) {
prof_leave(prof_tdata);
idalloc(ctx.v);
return (true);
}
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
prof_ctx_init(ctx.p, btkey.p);
Refactor prof_lookup() by extracting prof_lookup_global().
2014-01-15 09:04:34 +08:00
if (ckh_insert(&bt2ctx, btkey.v, ctx.v)) {
/* OOM. */
prof_leave(prof_tdata);
idalloc(btkey.v);
idalloc(ctx.v);
return (true);
}
new_ctx = true;
} else {
/*
* Increment nlimbo, in order to avoid a race condition with
* prof_ctx_merge()/prof_ctx_destroy().
*/
malloc_mutex_lock(ctx.p->lock);
ctx.p->nlimbo++;
malloc_mutex_unlock(ctx.p->lock);
new_ctx = false;
}
prof_leave(prof_tdata);
*p_btkey = btkey.v;
*p_ctx = ctx.p;
*p_new_ctx = new_ctx;
return (false);
}
Inline the fast path for heap sampling. Inline the heap sampling code that is executed for every allocation event (regardless of whether a sample is taken). Combine all prof TLS data into a single data structure, in order to reduce the TLS lookup volume.
2010-10-21 10:05:59 +08:00
prof_thr_cnt_t *
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
prof_lookup(prof_bt_t *bt)
{
Fix compiler warnings and errors. Use INT_MAX instead of MAX_INT in ALLOCM_ALIGN(), and #include <limits.h> in order to get its definition. Modify prof code related to hash tables to avoid aliasing warnings from gcc 4.1.2 (gcc 4.4.0 and 4.4.3 do not warn).
2010-09-21 10:53:25 +08:00
union {
prof_thr_cnt_t *p;
void *v;
} ret;
Inline the fast path for heap sampling. Inline the heap sampling code that is executed for every allocation event (regardless of whether a sample is taken). Combine all prof TLS data into a single data structure, in order to reduce the TLS lookup volume.
2010-10-21 10:05:59 +08:00
prof_tdata_t *prof_tdata;
Make cumulative heap profile data optional. Add the R option to control whether cumulative heap profile data are maintained. Add the T option to control the size of per thread backtrace caches, primarily because when the R option is specified, backtraces that no longer have allocations associated with them are discarded as soon as no thread caches refer to them.
2010-10-03 06:18:50 +08:00
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Fix potential TLS-related memory corruption. Avoid writing to uninitialized TLS as a side effect of deallocation. Initializing TLS during deallocation is unsafe because it is possible that a thread never did any allocation, and that TLS has already been deallocated by the threads library, resulting in write-after-free corruption. These fixes affect prof_tdata and quarantine; all other uses of TLS are already safe, whether intentionally (as for tcache) or unintentionally (as for arenas).
2013-01-31 07:03:11 +08:00
prof_tdata = prof_tdata_get(false);
Fix more prof_tdata resurrection corner cases.
2012-04-29 14:27:13 +08:00
if ((uintptr_t)prof_tdata <= (uintptr_t)PROF_TDATA_STATE_MAX)
Fix heap profiling bugs. Fix a potential deadlock that could occur during interval- and growth-triggered heap profile dumps. Fix an off-by-one heap profile statistics bug that could be observed in interval- and growth-triggered heap profiles. Fix heap profile dump filename sequence numbers (regression during conversion to malloc_snprintf()).
2012-04-23 07:00:11 +08:00
return (NULL);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
Inline the fast path for heap sampling. Inline the heap sampling code that is executed for every allocation event (regardless of whether a sample is taken). Combine all prof TLS data into a single data structure, in order to reduce the TLS lookup volume.
2010-10-21 10:05:59 +08:00
if (ckh_search(&prof_tdata->bt2cnt, bt, NULL, &ret.v)) {
Refactor prof_lookup() by extracting prof_lookup_global().
2014-01-15 09:04:34 +08:00
void *btkey;
prof_ctx_t *ctx;
Fix a heap dumping deadlock. Restructure the ctx initialization code such that the ctx isn't locked across portions of the initialization code where allocation could occur. Instead artificially inflate the cnt_merged.curobjs field, just as is done elsewhere to avoid similar races to the one that would otherwise be created by the reduction in locking scope. This bug affected interval- and growth-triggered heap dumping, but not manual heap dumping.
2011-01-15 09:27:44 +08:00
bool new_ctx;
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
/*
* This thread's cache lacks bt. Look for it in the global
* cache.
*/
Refactor prof_lookup() by extracting prof_lookup_global().
2014-01-15 09:04:34 +08:00
if (prof_lookup_global(bt, prof_tdata, &btkey, &ctx, &new_ctx))
return (NULL);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
/* Link a prof_thd_cnt_t into ctx for this thread. */
Remove the opt.lg_prof_tcmax option. Remove the opt.lg_prof_tcmax option and hard-code a cache size of 1024. This setting is something that users just shouldn't have to worry about. If lock contention actually ends up being a problem, the simple solution available to the user is to reduce sampling frequency.
2012-02-14 10:04:26 +08:00
if (ckh_count(&prof_tdata->bt2cnt) == PROF_TCMAX) {
Inline the fast path for heap sampling. Inline the heap sampling code that is executed for every allocation event (regardless of whether a sample is taken). Combine all prof TLS data into a single data structure, in order to reduce the TLS lookup volume.
2010-10-21 10:05:59 +08:00
assert(ckh_count(&prof_tdata->bt2cnt) > 0);
Make cumulative heap profile data optional. Add the R option to control whether cumulative heap profile data are maintained. Add the T option to control the size of per thread backtrace caches, primarily because when the R option is specified, backtraces that no longer have allocations associated with them are discarded as soon as no thread caches refer to them.
2010-10-03 06:18:50 +08:00
/*
Fix heap profiling bugs. Fix a regression due to the recent heap profiling accuracy improvements: prof_{m,re}alloc() must set the object's profiling context regardless of whether it is sampled. Fix management of the CHUNK_MAP_CLASS chunk map bits, such that all large object (re-)allocation paths correctly initialize the bits. Prior to this fix, in-place realloc() cleared the bits, resulting in incorrect reported object size from arena_salloc_demote(). After this fix the non-demoted bit pattern is all zeros (instead of all ones), which makes it easier to assure that the bits are properly set.
2010-10-23 01:45:59 +08:00
* Flush the least recently used cnt in order to keep
* bt2cnt from becoming too large.
Make cumulative heap profile data optional. Add the R option to control whether cumulative heap profile data are maintained. Add the T option to control the size of per thread backtrace caches, primarily because when the R option is specified, backtraces that no longer have allocations associated with them are discarded as soon as no thread caches refer to them.
2010-10-03 06:18:50 +08:00
*/
Inline the fast path for heap sampling. Inline the heap sampling code that is executed for every allocation event (regardless of whether a sample is taken). Combine all prof TLS data into a single data structure, in order to reduce the TLS lookup volume.
2010-10-21 10:05:59 +08:00
ret.p = ql_last(&prof_tdata->lru_ql, lru_link);
Make cumulative heap profile data optional. Add the R option to control whether cumulative heap profile data are maintained. Add the T option to control the size of per thread backtrace caches, primarily because when the R option is specified, backtraces that no longer have allocations associated with them are discarded as soon as no thread caches refer to them.
2010-10-03 06:18:50 +08:00
assert(ret.v != NULL);
Fix a prof-related race condition. Fix prof_lookup() to artificially raise curobjs for all paths through the code that creates a new entry in the per thread bt2cnt hash table. This fixes a race condition that could corrupt memory if prof_accum were false, and a non-default lg_prof_tcmax were used and/or threads were destroyed.
2011-08-31 14:40:11 +08:00
if (ckh_remove(&prof_tdata->bt2cnt, ret.p->ctx->bt,
NULL, NULL))
Prefer not_reached() over assert(false) where appropriate.
2013-10-22 05:56:27 +08:00
not_reached();
Inline the fast path for heap sampling. Inline the heap sampling code that is executed for every allocation event (regardless of whether a sample is taken). Combine all prof TLS data into a single data structure, in order to reduce the TLS lookup volume.
2010-10-21 10:05:59 +08:00
ql_remove(&prof_tdata->lru_ql, ret.p, lru_link);
Make cumulative heap profile data optional. Add the R option to control whether cumulative heap profile data are maintained. Add the T option to control the size of per thread backtrace caches, primarily because when the R option is specified, backtraces that no longer have allocations associated with them are discarded as soon as no thread caches refer to them.
2010-10-03 06:18:50 +08:00
prof_ctx_merge(ret.p->ctx, ret.p);
/* ret can now be re-used. */
} else {
Remove the opt.lg_prof_tcmax option. Remove the opt.lg_prof_tcmax option and hard-code a cache size of 1024. This setting is something that users just shouldn't have to worry about. If lock contention actually ends up being a problem, the simple solution available to the user is to reduce sampling frequency.
2012-02-14 10:04:26 +08:00
assert(ckh_count(&prof_tdata->bt2cnt) < PROF_TCMAX);
Make cumulative heap profile data optional. Add the R option to control whether cumulative heap profile data are maintained. Add the T option to control the size of per thread backtrace caches, primarily because when the R option is specified, backtraces that no longer have allocations associated with them are discarded as soon as no thread caches refer to them.
2010-10-03 06:18:50 +08:00
/* Allocate and partially initialize a new cnt. */
ret.v = imalloc(sizeof(prof_thr_cnt_t));
Fix prof bugs. Fix a race condition in ctx destruction that could cause undefined behavior (deadlock observed). Add mutex unlocks to some OOM error paths.
2010-10-28 10:47:40 +08:00
if (ret.p == NULL) {
Clean up prof-related comments. Clean up some prof-related comments to more accurately reflect how the code works. Simplify OOM handling code in a couple of prof-related error paths.
2011-08-10 10:06:06 +08:00
if (new_ctx)
Refactor prof_lookup() by extracting prof_lookup_global().
2014-01-15 09:04:34 +08:00
prof_ctx_destroy(ctx);
Make cumulative heap profile data optional. Add the R option to control whether cumulative heap profile data are maintained. Add the T option to control the size of per thread backtrace caches, primarily because when the R option is specified, backtraces that no longer have allocations associated with them are discarded as soon as no thread caches refer to them.
2010-10-03 06:18:50 +08:00
return (NULL);
Fix prof bugs. Fix a race condition in ctx destruction that could cause undefined behavior (deadlock observed). Add mutex unlocks to some OOM error paths.
2010-10-28 10:47:40 +08:00
}
Make cumulative heap profile data optional. Add the R option to control whether cumulative heap profile data are maintained. Add the T option to control the size of per thread backtrace caches, primarily because when the R option is specified, backtraces that no longer have allocations associated with them are discarded as soon as no thread caches refer to them.
2010-10-03 06:18:50 +08:00
ql_elm_new(ret.p, cnts_link);
ql_elm_new(ret.p, lru_link);
}
/* Finish initializing ret. */
Refactor prof_lookup() by extracting prof_lookup_global().
2014-01-15 09:04:34 +08:00
ret.p->ctx = ctx;
Fix compiler warnings and errors. Use INT_MAX instead of MAX_INT in ALLOCM_ALIGN(), and #include <limits.h> in order to get its definition. Modify prof code related to hash tables to avoid aliasing warnings from gcc 4.1.2 (gcc 4.4.0 and 4.4.3 do not warn).
2010-09-21 10:53:25 +08:00
ret.p->epoch = 0;
memset(&ret.p->cnts, 0, sizeof(prof_cnt_t));
Refactor prof_lookup() by extracting prof_lookup_global().
2014-01-15 09:04:34 +08:00
if (ckh_insert(&prof_tdata->bt2cnt, btkey, ret.v)) {
Clean up prof-related comments. Clean up some prof-related comments to more accurately reflect how the code works. Simplify OOM handling code in a couple of prof-related error paths.
2011-08-10 10:06:06 +08:00
if (new_ctx)
Refactor prof_lookup() by extracting prof_lookup_global().
2014-01-15 09:04:34 +08:00
prof_ctx_destroy(ctx);
Fix compiler warnings and errors. Use INT_MAX instead of MAX_INT in ALLOCM_ALIGN(), and #include <limits.h> in order to get its definition. Modify prof code related to hash tables to avoid aliasing warnings from gcc 4.1.2 (gcc 4.4.0 and 4.4.3 do not warn).
2010-09-21 10:53:25 +08:00
idalloc(ret.v);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
return (NULL);
}
Inline the fast path for heap sampling. Inline the heap sampling code that is executed for every allocation event (regardless of whether a sample is taken). Combine all prof TLS data into a single data structure, in order to reduce the TLS lookup volume.
2010-10-21 10:05:59 +08:00
ql_head_insert(&prof_tdata->lru_ql, ret.p, lru_link);
Refactor prof_lookup() by extracting prof_lookup_global().
2014-01-15 09:04:34 +08:00
malloc_mutex_lock(ctx->lock);
ql_tail_insert(&ctx->cnts_ql, ret.p, cnts_link);
ctx->nlimbo--;
malloc_mutex_unlock(ctx->lock);
Make cumulative heap profile data optional. Add the R option to control whether cumulative heap profile data are maintained. Add the T option to control the size of per thread backtrace caches, primarily because when the R option is specified, backtraces that no longer have allocations associated with them are discarded as soon as no thread caches refer to them.
2010-10-03 06:18:50 +08:00
} else {
/* Move ret to the front of the LRU. */
Inline the fast path for heap sampling. Inline the heap sampling code that is executed for every allocation event (regardless of whether a sample is taken). Combine all prof TLS data into a single data structure, in order to reduce the TLS lookup volume.
2010-10-21 10:05:59 +08:00
ql_remove(&prof_tdata->lru_ql, ret.p, lru_link);
ql_head_insert(&prof_tdata->lru_ql, ret.p, lru_link);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
}
Fix compiler warnings and errors. Use INT_MAX instead of MAX_INT in ALLOCM_ALIGN(), and #include <limits.h> in order to get its definition. Modify prof code related to hash tables to avoid aliasing warnings from gcc 4.1.2 (gcc 4.4.0 and 4.4.3 do not warn).
2010-09-21 10:53:25 +08:00
return (ret.p);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
}
Add heap profiling tests. Fix a regression in prof_dump_ctx() due to an uninitized variable. This was caused by revision 4f37ef693e3d5903ce07dc0b61c0da320b35e3d9, so no releases are affected.
2014-01-18 07:40:52 +08:00
#ifdef JEMALLOC_JET
size_t
prof_bt_count(void)
{
size_t bt_count;
prof_tdata_t *prof_tdata;
prof_tdata = prof_tdata_get(false);
if ((uintptr_t)prof_tdata <= (uintptr_t)PROF_TDATA_STATE_MAX)
return (0);
prof_enter(prof_tdata);
bt_count = ckh_count(&bt2ctx);
prof_leave(prof_tdata);
return (bt_count);
}
#endif
#ifdef JEMALLOC_JET
#undef prof_dump_open
#define prof_dump_open JEMALLOC_N(prof_dump_open_impl)
#endif
static int
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
prof_dump_open(bool propagate_err, const char *filename)
{
Add heap profiling tests. Fix a regression in prof_dump_ctx() due to an uninitized variable. This was caused by revision 4f37ef693e3d5903ce07dc0b61c0da320b35e3d9, so no releases are affected.
2014-01-18 07:40:52 +08:00
int fd;
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
Add heap profiling tests. Fix a regression in prof_dump_ctx() due to an uninitized variable. This was caused by revision 4f37ef693e3d5903ce07dc0b61c0da320b35e3d9, so no releases are affected.
2014-01-18 07:40:52 +08:00
fd = creat(filename, 0644);
if (fd == -1 && propagate_err == false) {
malloc_printf("<jemalloc>: creat(\"%s\"), 0644) failed\n",
filename);
if (opt_abort)
abort();
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
}
Add heap profiling tests. Fix a regression in prof_dump_ctx() due to an uninitized variable. This was caused by revision 4f37ef693e3d5903ce07dc0b61c0da320b35e3d9, so no releases are affected.
2014-01-18 07:40:52 +08:00
return (fd);
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
}
Add heap profiling tests. Fix a regression in prof_dump_ctx() due to an uninitized variable. This was caused by revision 4f37ef693e3d5903ce07dc0b61c0da320b35e3d9, so no releases are affected.
2014-01-18 07:40:52 +08:00
#ifdef JEMALLOC_JET
#undef prof_dump_open
#define prof_dump_open JEMALLOC_N(prof_dump_open)
prof_dump_open_t *prof_dump_open = JEMALLOC_N(prof_dump_open_impl);
#endif
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
static bool
prof_dump_flush(bool propagate_err)
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
{
Allow prof.dump mallctl to specify filename.
2010-03-03 03:57:30 +08:00
bool ret = false;
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
ssize_t err;
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
err = write(prof_dump_fd, prof_dump_buf, prof_dump_buf_end);
if (err == -1) {
Allow prof.dump mallctl to specify filename.
2010-03-03 03:57:30 +08:00
if (propagate_err == false) {
Simplify malloc_message(). Rather than passing four strings to malloc_message(), malloc_write4(), and all the functions that use them, only pass one string.
2010-03-04 09:45:38 +08:00
malloc_write("<jemalloc>: write() failed during heap "
"profile flush\n");
Allow prof.dump mallctl to specify filename.
2010-03-03 03:57:30 +08:00
if (opt_abort)
abort();
}
ret = true;
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
}
prof_dump_buf_end = 0;
Allow prof.dump mallctl to specify filename.
2010-03-03 03:57:30 +08:00
return (ret);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
}
Allow prof.dump mallctl to specify filename.
2010-03-03 03:57:30 +08:00
static bool
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
prof_dump_close(bool propagate_err)
{
bool ret;
assert(prof_dump_fd != -1);
ret = prof_dump_flush(propagate_err);
close(prof_dump_fd);
prof_dump_fd = -1;
return (ret);
}
static bool
prof_dump_write(bool propagate_err, const char *s)
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
{
unsigned i, slen, n;
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
i = 0;
slen = strlen(s);
while (i < slen) {
/* Flush the buffer if it is full. */
Implement tsd. Implement tsd, which is a TLS/TSD abstraction that uses one or both internally. Modify bootstrapping such that no tsd's are utilized until allocation is safe. Remove malloc_[v]tprintf(), and use malloc_snprintf() instead. Fix %p argument size handling in malloc_vsnprintf(). Fix a long-standing statistics-related bug in the "thread.arena" mallctl that could cause crashes due to linked list corruption.
2012-03-22 09:33:03 +08:00
if (prof_dump_buf_end == PROF_DUMP_BUFSIZE)
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
if (prof_dump_flush(propagate_err) && propagate_err)
Allow prof.dump mallctl to specify filename.
2010-03-03 03:57:30 +08:00
return (true);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
Implement tsd. Implement tsd, which is a TLS/TSD abstraction that uses one or both internally. Modify bootstrapping such that no tsd's are utilized until allocation is safe. Remove malloc_[v]tprintf(), and use malloc_snprintf() instead. Fix %p argument size handling in malloc_vsnprintf(). Fix a long-standing statistics-related bug in the "thread.arena" mallctl that could cause crashes due to linked list corruption.
2012-03-22 09:33:03 +08:00
if (prof_dump_buf_end + slen <= PROF_DUMP_BUFSIZE) {
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
/* Finish writing. */
n = slen - i;
} else {
/* Write as much of s as will fit. */
Implement tsd. Implement tsd, which is a TLS/TSD abstraction that uses one or both internally. Modify bootstrapping such that no tsd's are utilized until allocation is safe. Remove malloc_[v]tprintf(), and use malloc_snprintf() instead. Fix %p argument size handling in malloc_vsnprintf(). Fix a long-standing statistics-related bug in the "thread.arena" mallctl that could cause crashes due to linked list corruption.
2012-03-22 09:33:03 +08:00
n = PROF_DUMP_BUFSIZE - prof_dump_buf_end;
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
}
memcpy(&prof_dump_buf[prof_dump_buf_end], &s[i], n);
prof_dump_buf_end += n;
i += n;
}
Allow prof.dump mallctl to specify filename.
2010-03-03 03:57:30 +08:00
return (false);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
}
Implement malloc_vsnprintf(). Implement malloc_vsnprintf() (a subset of vsnprintf(3)) as well as several other printing functions based on it, so that formatted printing can be relied upon without concern for inducing a dependency on floating point runtime support. Replace malloc_write() calls with malloc_*printf() where doing so simplifies the code. Add name mangling for library-private symbols in the data and BSS sections. Adjust CONF_HANDLE_*() macros in malloc_conf_init() to expose all opt_* variable use to cpp so that proper mangling occurs.
2012-03-07 06:57:45 +08:00
JEMALLOC_ATTR(format(printf, 2, 3))
static bool
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
prof_dump_printf(bool propagate_err, const char *format, ...)
Implement malloc_vsnprintf(). Implement malloc_vsnprintf() (a subset of vsnprintf(3)) as well as several other printing functions based on it, so that formatted printing can be relied upon without concern for inducing a dependency on floating point runtime support. Replace malloc_write() calls with malloc_*printf() where doing so simplifies the code. Add name mangling for library-private symbols in the data and BSS sections. Adjust CONF_HANDLE_*() macros in malloc_conf_init() to expose all opt_* variable use to cpp so that proper mangling occurs.
2012-03-07 06:57:45 +08:00
{
bool ret;
va_list ap;
Implement tsd. Implement tsd, which is a TLS/TSD abstraction that uses one or both internally. Modify bootstrapping such that no tsd's are utilized until allocation is safe. Remove malloc_[v]tprintf(), and use malloc_snprintf() instead. Fix %p argument size handling in malloc_vsnprintf(). Fix a long-standing statistics-related bug in the "thread.arena" mallctl that could cause crashes due to linked list corruption.
2012-03-22 09:33:03 +08:00
char buf[PROF_PRINTF_BUFSIZE];
Implement malloc_vsnprintf(). Implement malloc_vsnprintf() (a subset of vsnprintf(3)) as well as several other printing functions based on it, so that formatted printing can be relied upon without concern for inducing a dependency on floating point runtime support. Replace malloc_write() calls with malloc_*printf() where doing so simplifies the code. Add name mangling for library-private symbols in the data and BSS sections. Adjust CONF_HANDLE_*() macros in malloc_conf_init() to expose all opt_* variable use to cpp so that proper mangling occurs.
2012-03-07 06:57:45 +08:00
va_start(ap, format);
Remove ephemeral mutexes. Remove ephemeral mutexes from the prof machinery, and remove malloc_mutex_destroy(). This simplifies mutex management on systems that call malloc()/free() inside pthread_mutex_{create,destroy}(). Add atomic_*_u() for operation on unsigned values. Fix prof_printf() to call malloc_vsnprintf() rather than malloc_snprintf().
2012-03-24 09:05:51 +08:00
malloc_vsnprintf(buf, sizeof(buf), format, ap);
Implement malloc_vsnprintf(). Implement malloc_vsnprintf() (a subset of vsnprintf(3)) as well as several other printing functions based on it, so that formatted printing can be relied upon without concern for inducing a dependency on floating point runtime support. Replace malloc_write() calls with malloc_*printf() where doing so simplifies the code. Add name mangling for library-private symbols in the data and BSS sections. Adjust CONF_HANDLE_*() macros in malloc_conf_init() to expose all opt_* variable use to cpp so that proper mangling occurs.
2012-03-07 06:57:45 +08:00
va_end(ap);
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
ret = prof_dump_write(propagate_err, buf);
Implement malloc_vsnprintf(). Implement malloc_vsnprintf() (a subset of vsnprintf(3)) as well as several other printing functions based on it, so that formatted printing can be relied upon without concern for inducing a dependency on floating point runtime support. Replace malloc_write() calls with malloc_*printf() where doing so simplifies the code. Add name mangling for library-private symbols in the data and BSS sections. Adjust CONF_HANDLE_*() macros in malloc_conf_init() to expose all opt_* variable use to cpp so that proper mangling occurs.
2012-03-07 06:57:45 +08:00
return (ret);
}
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
static void
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
prof_dump_ctx_prep(prof_ctx_t *ctx, prof_cnt_t *cnt_all, size_t *leak_nctx,
prof_ctx_list_t *ctx_ql)
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
{
prof_thr_cnt_t *thr_cnt;
prof_cnt_t tcnt;
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Remove ephemeral mutexes. Remove ephemeral mutexes from the prof machinery, and remove malloc_mutex_destroy(). This simplifies mutex management on systems that call malloc()/free() inside pthread_mutex_{create,destroy}(). Add atomic_*_u() for operation on unsigned values. Fix prof_printf() to call malloc_vsnprintf() rather than malloc_snprintf().
2012-03-24 09:05:51 +08:00
malloc_mutex_lock(ctx->lock);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
/*
* Increment nlimbo so that ctx won't go away before dump.
* Additionally, link ctx into the dump list so that it is included in
* prof_dump()'s second pass.
*/
ctx->nlimbo++;
ql_tail_insert(ctx_ql, ctx, dump_link);
Make cumulative heap profile data optional. Add the R option to control whether cumulative heap profile data are maintained. Add the T option to control the size of per thread backtrace caches, primarily because when the R option is specified, backtraces that no longer have allocations associated with them are discarded as soon as no thread caches refer to them.
2010-10-03 06:18:50 +08:00
memcpy(&ctx->cnt_summed, &ctx->cnt_merged, sizeof(prof_cnt_t));
ql_foreach(thr_cnt, &ctx->cnts_ql, cnts_link) {
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
volatile unsigned *epoch = &thr_cnt->epoch;
while (true) {
unsigned epoch0 = *epoch;
/* Make sure epoch is even. */
if (epoch0 & 1U)
continue;
memcpy(&tcnt, &thr_cnt->cnts, sizeof(prof_cnt_t));
/* Terminate if epoch didn't change while reading. */
if (*epoch == epoch0)
break;
}
Make cumulative heap profile data optional. Add the R option to control whether cumulative heap profile data are maintained. Add the T option to control the size of per thread backtrace caches, primarily because when the R option is specified, backtraces that no longer have allocations associated with them are discarded as soon as no thread caches refer to them.
2010-10-03 06:18:50 +08:00
ctx->cnt_summed.curobjs += tcnt.curobjs;
ctx->cnt_summed.curbytes += tcnt.curbytes;
if (opt_prof_accum) {
ctx->cnt_summed.accumobjs += tcnt.accumobjs;
ctx->cnt_summed.accumbytes += tcnt.accumbytes;
}
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
}
Fix leak context count reporting. Fix a bug in leak context count reporting that tended to cause the number of contexts to be underreported. The reported number of leaked objects and bytes were not affected by this bug.
2010-10-03 13:39:59 +08:00
if (ctx->cnt_summed.curobjs != 0)
(*leak_nctx)++;
Make cumulative heap profile data optional. Add the R option to control whether cumulative heap profile data are maintained. Add the T option to control the size of per thread backtrace caches, primarily because when the R option is specified, backtraces that no longer have allocations associated with them are discarded as soon as no thread caches refer to them.
2010-10-03 06:18:50 +08:00
/* Add to cnt_all. */
cnt_all->curobjs += ctx->cnt_summed.curobjs;
cnt_all->curbytes += ctx->cnt_summed.curbytes;
if (opt_prof_accum) {
cnt_all->accumobjs += ctx->cnt_summed.accumobjs;
cnt_all->accumbytes += ctx->cnt_summed.accumbytes;
}
Remove ephemeral mutexes. Remove ephemeral mutexes from the prof machinery, and remove malloc_mutex_destroy(). This simplifies mutex management on systems that call malloc()/free() inside pthread_mutex_{create,destroy}(). Add atomic_*_u() for operation on unsigned values. Fix prof_printf() to call malloc_vsnprintf() rather than malloc_snprintf().
2012-03-24 09:05:51 +08:00
malloc_mutex_unlock(ctx->lock);
Make cumulative heap profile data optional. Add the R option to control whether cumulative heap profile data are maintained. Add the T option to control the size of per thread backtrace caches, primarily because when the R option is specified, backtraces that no longer have allocations associated with them are discarded as soon as no thread caches refer to them.
2010-10-03 06:18:50 +08:00
}
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
static bool
prof_dump_header(bool propagate_err, const prof_cnt_t *cnt_all)
Make cumulative heap profile data optional. Add the R option to control whether cumulative heap profile data are maintained. Add the T option to control the size of per thread backtrace caches, primarily because when the R option is specified, backtraces that no longer have allocations associated with them are discarded as soon as no thread caches refer to them.
2010-10-03 06:18:50 +08:00
{
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
if (opt_lg_prof_sample == 0) {
if (prof_dump_printf(propagate_err,
"heap profile: %"PRId64": %"PRId64
" [%"PRIu64": %"PRIu64"] @ heapprofile\n",
cnt_all->curobjs, cnt_all->curbytes,
cnt_all->accumobjs, cnt_all->accumbytes))
return (true);
Make cumulative heap profile data optional. Add the R option to control whether cumulative heap profile data are maintained. Add the T option to control the size of per thread backtrace caches, primarily because when the R option is specified, backtraces that no longer have allocations associated with them are discarded as soon as no thread caches refer to them.
2010-10-03 06:18:50 +08:00
} else {
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
if (prof_dump_printf(propagate_err,
"heap profile: %"PRId64": %"PRId64
" [%"PRIu64": %"PRIu64"] @ heap_v2/%"PRIu64"\n",
cnt_all->curobjs, cnt_all->curbytes,
cnt_all->accumobjs, cnt_all->accumbytes,
((uint64_t)1U << opt_lg_prof_sample)))
return (true);
Make cumulative heap profile data optional. Add the R option to control whether cumulative heap profile data are maintained. Add the T option to control the size of per thread backtrace caches, primarily because when the R option is specified, backtraces that no longer have allocations associated with them are discarded as soon as no thread caches refer to them.
2010-10-03 06:18:50 +08:00
}
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
return (false);
Make cumulative heap profile data optional. Add the R option to control whether cumulative heap profile data are maintained. Add the T option to control the size of per thread backtrace caches, primarily because when the R option is specified, backtraces that no longer have allocations associated with them are discarded as soon as no thread caches refer to them.
2010-10-03 06:18:50 +08:00
}
static void
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
prof_dump_ctx_cleanup_locked(prof_ctx_t *ctx, prof_ctx_list_t *ctx_ql)
Make cumulative heap profile data optional. Add the R option to control whether cumulative heap profile data are maintained. Add the T option to control the size of per thread backtrace caches, primarily because when the R option is specified, backtraces that no longer have allocations associated with them are discarded as soon as no thread caches refer to them.
2010-10-03 06:18:50 +08:00
{
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
ctx->nlimbo--;
ql_remove(ctx_ql, ctx, dump_link);
}
static void
prof_dump_ctx_cleanup(prof_ctx_t *ctx, prof_ctx_list_t *ctx_ql)
{
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
Remove ephemeral mutexes. Remove ephemeral mutexes from the prof machinery, and remove malloc_mutex_destroy(). This simplifies mutex management on systems that call malloc()/free() inside pthread_mutex_{create,destroy}(). Add atomic_*_u() for operation on unsigned values. Fix prof_printf() to call malloc_vsnprintf() rather than malloc_snprintf().
2012-03-24 09:05:51 +08:00
malloc_mutex_lock(ctx->lock);
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
prof_dump_ctx_cleanup_locked(ctx, ctx_ql);
Remove ephemeral mutexes. Remove ephemeral mutexes from the prof machinery, and remove malloc_mutex_destroy(). This simplifies mutex management on systems that call malloc()/free() inside pthread_mutex_{create,destroy}(). Add atomic_*_u() for operation on unsigned values. Fix prof_printf() to call malloc_vsnprintf() rather than malloc_snprintf().
2012-03-24 09:05:51 +08:00
malloc_mutex_unlock(ctx->lock);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
}
Allow prof.dump mallctl to specify filename.
2010-03-03 03:57:30 +08:00
static bool
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
prof_dump_ctx(bool propagate_err, prof_ctx_t *ctx, const prof_bt_t *bt,
prof_ctx_list_t *ctx_ql)
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
{
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
bool ret;
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
unsigned i;
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Fix heap profiling bugs. Fix a potential deadlock that could occur during interval- and growth-triggered heap profile dumps. Fix an off-by-one heap profile statistics bug that could be observed in interval- and growth-triggered heap profiles. Fix heap profile dump filename sequence numbers (regression during conversion to malloc_snprintf()).
2012-04-23 07:00:11 +08:00
/*
* Current statistics can sum to 0 as a result of unmerged per thread
* statistics. Additionally, interval- and growth-triggered dumps can
* occur between the time a ctx is created and when its statistics are
* filled in. Avoid dumping any ctx that is an artifact of either
* implementation detail.
*/
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
malloc_mutex_lock(ctx->lock);
Fix heap profiling bugs. Fix a potential deadlock that could occur during interval- and growth-triggered heap profile dumps. Fix an off-by-one heap profile statistics bug that could be observed in interval- and growth-triggered heap profiles. Fix heap profile dump filename sequence numbers (regression during conversion to malloc_snprintf()).
2012-04-23 07:00:11 +08:00
if ((opt_prof_accum == false && ctx->cnt_summed.curobjs == 0) ||
(opt_prof_accum && ctx->cnt_summed.accumobjs == 0)) {
assert(ctx->cnt_summed.curobjs == 0);
Make cumulative heap profile data optional. Add the R option to control whether cumulative heap profile data are maintained. Add the T option to control the size of per thread backtrace caches, primarily because when the R option is specified, backtraces that no longer have allocations associated with them are discarded as soon as no thread caches refer to them.
2010-10-03 06:18:50 +08:00
assert(ctx->cnt_summed.curbytes == 0);
assert(ctx->cnt_summed.accumobjs == 0);
assert(ctx->cnt_summed.accumbytes == 0);
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
ret = false;
goto label_return;
Make cumulative heap profile data optional. Add the R option to control whether cumulative heap profile data are maintained. Add the T option to control the size of per thread backtrace caches, primarily because when the R option is specified, backtraces that no longer have allocations associated with them are discarded as soon as no thread caches refer to them.
2010-10-03 06:18:50 +08:00
}
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
if (prof_dump_printf(propagate_err, "%"PRId64": %"PRId64
Implement malloc_vsnprintf(). Implement malloc_vsnprintf() (a subset of vsnprintf(3)) as well as several other printing functions based on it, so that formatted printing can be relied upon without concern for inducing a dependency on floating point runtime support. Replace malloc_write() calls with malloc_*printf() where doing so simplifies the code. Add name mangling for library-private symbols in the data and BSS sections. Adjust CONF_HANDLE_*() macros in malloc_conf_init() to expose all opt_* variable use to cpp so that proper mangling occurs.
2012-03-07 06:57:45 +08:00
" [%"PRIu64": %"PRIu64"] @",
ctx->cnt_summed.curobjs, ctx->cnt_summed.curbytes,
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
ctx->cnt_summed.accumobjs, ctx->cnt_summed.accumbytes)) {
ret = true;
goto label_return;
}
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
for (i = 0; i < bt->len; i++) {
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
if (prof_dump_printf(propagate_err, " %#"PRIxPTR,
(uintptr_t)bt->vec[i])) {
ret = true;
goto label_return;
}
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
}
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
if (prof_dump_write(propagate_err, "\n")) {
ret = true;
goto label_return;
}
Allow prof.dump mallctl to specify filename.
2010-03-03 03:57:30 +08:00
Add heap profiling tests. Fix a regression in prof_dump_ctx() due to an uninitized variable. This was caused by revision 4f37ef693e3d5903ce07dc0b61c0da320b35e3d9, so no releases are affected.
2014-01-18 07:40:52 +08:00
ret = false;
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
label_return:
prof_dump_ctx_cleanup_locked(ctx, ctx_ql);
malloc_mutex_unlock(ctx->lock);
return (ret);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
}
Allow prof.dump mallctl to specify filename.
2010-03-03 03:57:30 +08:00
static bool
prof_dump_maps(bool propagate_err)
Dump /proc/<pid>/maps in heap profiles.
2010-02-12 01:25:56 +08:00
{
Fix a file descriptor leak in a prof_dump_maps() error path. Reported by Pat Lynch.
2013-10-22 06:07:40 +08:00
bool ret;
Dump /proc/<pid>/maps in heap profiles.
2010-02-12 01:25:56 +08:00
int mfd;
Implement tsd. Implement tsd, which is a TLS/TSD abstraction that uses one or both internally. Modify bootstrapping such that no tsd's are utilized until allocation is safe. Remove malloc_[v]tprintf(), and use malloc_snprintf() instead. Fix %p argument size handling in malloc_vsnprintf(). Fix a long-standing statistics-related bug in the "thread.arena" mallctl that could cause crashes due to linked list corruption.
2012-03-22 09:33:03 +08:00
char filename[PATH_MAX + 1];
Dump /proc/<pid>/maps in heap profiles.
2010-02-12 01:25:56 +08:00
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Consistently use debug lib(s) if present Fixes a situation where nm uses the debug lib but addr2line does not, which completely messes up the symbol lookup.
2014-03-18 15:00:14 +08:00
#ifdef __FreeBSD__
malloc_snprintf(filename, sizeof(filename), "/proc/curproc/map");
#else
malloc_snprintf(filename, sizeof(filename), "/proc/%d/maps",
Implement tsd. Implement tsd, which is a TLS/TSD abstraction that uses one or both internally. Modify bootstrapping such that no tsd's are utilized until allocation is safe. Remove malloc_[v]tprintf(), and use malloc_snprintf() instead. Fix %p argument size handling in malloc_vsnprintf(). Fix a long-standing statistics-related bug in the "thread.arena" mallctl that could cause crashes due to linked list corruption.
2012-03-22 09:33:03 +08:00
(int)getpid());
Consistently use debug lib(s) if present Fixes a situation where nm uses the debug lib but addr2line does not, which completely messes up the symbol lookup.
2014-03-18 15:00:14 +08:00
#endif
Implement tsd. Implement tsd, which is a TLS/TSD abstraction that uses one or both internally. Modify bootstrapping such that no tsd's are utilized until allocation is safe. Remove malloc_[v]tprintf(), and use malloc_snprintf() instead. Fix %p argument size handling in malloc_vsnprintf(). Fix a long-standing statistics-related bug in the "thread.arena" mallctl that could cause crashes due to linked list corruption.
2012-03-22 09:33:03 +08:00
mfd = open(filename, O_RDONLY);
Dump /proc/<pid>/maps in heap profiles.
2010-02-12 01:25:56 +08:00
if (mfd != -1) {
ssize_t nread;
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
if (prof_dump_write(propagate_err, "\nMAPPED_LIBRARIES:\n") &&
Fix a file descriptor leak in a prof_dump_maps() error path. Reported by Pat Lynch.
2013-10-22 06:07:40 +08:00
propagate_err) {
ret = true;
goto label_return;
}
Dump /proc/<pid>/maps in heap profiles.
2010-02-12 01:25:56 +08:00
nread = 0;
do {
prof_dump_buf_end += nread;
Implement tsd. Implement tsd, which is a TLS/TSD abstraction that uses one or both internally. Modify bootstrapping such that no tsd's are utilized until allocation is safe. Remove malloc_[v]tprintf(), and use malloc_snprintf() instead. Fix %p argument size handling in malloc_vsnprintf(). Fix a long-standing statistics-related bug in the "thread.arena" mallctl that could cause crashes due to linked list corruption.
2012-03-22 09:33:03 +08:00
if (prof_dump_buf_end == PROF_DUMP_BUFSIZE) {
Dump /proc/<pid>/maps in heap profiles.
2010-02-12 01:25:56 +08:00
/* Make space in prof_dump_buf before read(). */
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
if (prof_dump_flush(propagate_err) &&
Fix a file descriptor leak in a prof_dump_maps() error path. Reported by Pat Lynch.
2013-10-22 06:07:40 +08:00
propagate_err) {
ret = true;
goto label_return;
}
Dump /proc/<pid>/maps in heap profiles.
2010-02-12 01:25:56 +08:00
}
nread = read(mfd, &prof_dump_buf[prof_dump_buf_end],
Implement tsd. Implement tsd, which is a TLS/TSD abstraction that uses one or both internally. Modify bootstrapping such that no tsd's are utilized until allocation is safe. Remove malloc_[v]tprintf(), and use malloc_snprintf() instead. Fix %p argument size handling in malloc_vsnprintf(). Fix a long-standing statistics-related bug in the "thread.arena" mallctl that could cause crashes due to linked list corruption.
2012-03-22 09:33:03 +08:00
PROF_DUMP_BUFSIZE - prof_dump_buf_end);
Dump /proc/<pid>/maps in heap profiles.
2010-02-12 01:25:56 +08:00
} while (nread > 0);
Fix a file descriptor leak in a prof_dump_maps() error path. Reported by Pat Lynch.
2013-10-22 06:07:40 +08:00
} else {
ret = true;
goto label_return;
}
Allow prof.dump mallctl to specify filename.
2010-03-03 03:57:30 +08:00
Fix a file descriptor leak in a prof_dump_maps() error path. Reported by Pat Lynch.
2013-10-22 06:07:40 +08:00
ret = false;
label_return:
if (mfd != -1)
close(mfd);
return (ret);
Dump /proc/<pid>/maps in heap profiles.
2010-02-12 01:25:56 +08:00
}
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
static void
prof_leakcheck(const prof_cnt_t *cnt_all, size_t leak_nctx,
const char *filename)
{
if (cnt_all->curbytes != 0) {
malloc_printf("<jemalloc>: Leak summary: %"PRId64" byte%s, %"
PRId64" object%s, %zu context%s\n",
cnt_all->curbytes, (cnt_all->curbytes != 1) ? "s" : "",
cnt_all->curobjs, (cnt_all->curobjs != 1) ? "s" : "",
leak_nctx, (leak_nctx != 1) ? "s" : "");
malloc_printf(
"<jemalloc>: Run pprof on \"%s\" for leak detail\n",
filename);
}
}
Allow prof.dump mallctl to specify filename.
2010-03-03 03:57:30 +08:00
static bool
Implement malloc_vsnprintf(). Implement malloc_vsnprintf() (a subset of vsnprintf(3)) as well as several other printing functions based on it, so that formatted printing can be relied upon without concern for inducing a dependency on floating point runtime support. Replace malloc_write() calls with malloc_*printf() where doing so simplifies the code. Add name mangling for library-private symbols in the data and BSS sections. Adjust CONF_HANDLE_*() macros in malloc_conf_init() to expose all opt_* variable use to cpp so that proper mangling occurs.
2012-03-07 06:57:45 +08:00
prof_dump(bool propagate_err, const char *filename, bool leakcheck)
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
{
Fix heap profiling bugs. Fix a potential deadlock that could occur during interval- and growth-triggered heap profile dumps. Fix an off-by-one heap profile statistics bug that could be observed in interval- and growth-triggered heap profiles. Fix heap profile dump filename sequence numbers (regression during conversion to malloc_snprintf()).
2012-04-23 07:00:11 +08:00
prof_tdata_t *prof_tdata;
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
prof_cnt_t cnt_all;
size_t tabind;
Fix compiler warnings and errors. Use INT_MAX instead of MAX_INT in ALLOCM_ALIGN(), and #include <limits.h> in order to get its definition. Modify prof code related to hash tables to avoid aliasing warnings from gcc 4.1.2 (gcc 4.4.0 and 4.4.3 do not warn).
2010-09-21 10:53:25 +08:00
union {
prof_ctx_t *p;
void *v;
} ctx;
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
size_t leak_nctx;
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
prof_ctx_list_t ctx_ql;
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Fix potential TLS-related memory corruption. Avoid writing to uninitialized TLS as a side effect of deallocation. Initializing TLS during deallocation is unsafe because it is possible that a thread never did any allocation, and that TLS has already been deallocated by the threads library, resulting in write-after-free corruption. These fixes affect prof_tdata and quarantine; all other uses of TLS are already safe, whether intentionally (as for tcache) or unintentionally (as for arenas).
2013-01-31 07:03:11 +08:00
prof_tdata = prof_tdata_get(false);
Fix more prof_tdata resurrection corner cases.
2012-04-29 14:27:13 +08:00
if ((uintptr_t)prof_tdata <= (uintptr_t)PROF_TDATA_STATE_MAX)
Fix heap profiling bugs. Fix a potential deadlock that could occur during interval- and growth-triggered heap profile dumps. Fix an off-by-one heap profile statistics bug that could be observed in interval- and growth-triggered heap profiles. Fix heap profile dump filename sequence numbers (regression during conversion to malloc_snprintf()).
2012-04-23 07:00:11 +08:00
return (true);
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
malloc_mutex_lock(&prof_dump_mtx);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
/* Merge per thread profile stats, and sum them in cnt_all. */
memset(&cnt_all, 0, sizeof(prof_cnt_t));
leak_nctx = 0;
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
ql_new(&ctx_ql);
prof_enter(prof_tdata);
Increase default backtrace depth from 4 to 128. Increase the default backtrace depth, because shallow backtraces tend to result in confusing pprof output graphs.
2010-10-03 13:38:14 +08:00
for (tabind = 0; ckh_iter(&bt2ctx, &tabind, NULL, &ctx.v) == false;)
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
prof_dump_ctx_prep(ctx.p, &cnt_all, &leak_nctx, &ctx_ql);
prof_leave(prof_tdata);
/* Create dump file. */
Add heap profiling tests. Fix a regression in prof_dump_ctx() due to an uninitized variable. This was caused by revision 4f37ef693e3d5903ce07dc0b61c0da320b35e3d9, so no releases are affected.
2014-01-18 07:40:52 +08:00
if ((prof_dump_fd = prof_dump_open(propagate_err, filename)) == -1)
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
goto label_open_close_error;
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
/* Dump profile header. */
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
if (prof_dump_header(propagate_err, &cnt_all))
goto label_write_error;
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
Fix whitespace.
2013-10-29 03:41:37 +08:00
/* Dump per ctx profile stats. */
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
while ((ctx.p = ql_first(&ctx_ql)) != NULL) {
if (prof_dump_ctx(propagate_err, ctx.p, ctx.p->bt, &ctx_ql))
goto label_write_error;
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
}
Dump /proc/<pid>/maps in heap profiles.
2010-02-12 01:25:56 +08:00
/* Dump /proc/<pid>/maps if possible. */
Allow prof.dump mallctl to specify filename.
2010-03-03 03:57:30 +08:00
if (prof_dump_maps(propagate_err))
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
goto label_write_error;
Dump /proc/<pid>/maps in heap profiles.
2010-02-12 01:25:56 +08:00
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
if (prof_dump_close(propagate_err))
goto label_open_close_error;
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
malloc_mutex_unlock(&prof_dump_mtx);
if (leakcheck)
prof_leakcheck(&cnt_all, leak_nctx, filename);
Allow prof.dump mallctl to specify filename.
2010-03-03 03:57:30 +08:00
return (false);
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
label_write_error:
prof_dump_close(propagate_err);
label_open_close_error:
while ((ctx.p = ql_first(&ctx_ql)) != NULL)
prof_dump_ctx_cleanup(ctx.p, &ctx_ql);
malloc_mutex_unlock(&prof_dump_mtx);
Allow prof.dump mallctl to specify filename.
2010-03-03 03:57:30 +08:00
return (true);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
}
Implement malloc_vsnprintf(). Implement malloc_vsnprintf() (a subset of vsnprintf(3)) as well as several other printing functions based on it, so that formatted printing can be relied upon without concern for inducing a dependency on floating point runtime support. Replace malloc_write() calls with malloc_*printf() where doing so simplifies the code. Add name mangling for library-private symbols in the data and BSS sections. Adjust CONF_HANDLE_*() macros in malloc_conf_init() to expose all opt_* variable use to cpp so that proper mangling occurs.
2012-03-07 06:57:45 +08:00
#define DUMP_FILENAME_BUFSIZE (PATH_MAX + 1)
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
#define VSEQ_INVALID UINT64_C(0xffffffffffffffff)
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
static void
prof_dump_filename(char *filename, char v, int64_t vseq)
{
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
if (vseq != VSEQ_INVALID) {
Implement malloc_vsnprintf(). Implement malloc_vsnprintf() (a subset of vsnprintf(3)) as well as several other printing functions based on it, so that formatted printing can be relied upon without concern for inducing a dependency on floating point runtime support. Replace malloc_write() calls with malloc_*printf() where doing so simplifies the code. Add name mangling for library-private symbols in the data and BSS sections. Adjust CONF_HANDLE_*() macros in malloc_conf_init() to expose all opt_* variable use to cpp so that proper mangling occurs.
2012-03-07 06:57:45 +08:00
/* "<prefix>.<pid>.<seq>.v<vseq>.heap" */
malloc_snprintf(filename, DUMP_FILENAME_BUFSIZE,
"%s.%d.%"PRIu64".%c%"PRId64".heap",
opt_prof_prefix, (int)getpid(), prof_dump_seq, v, vseq);
} else {
/* "<prefix>.<pid>.<seq>.<v>.heap" */
malloc_snprintf(filename, DUMP_FILENAME_BUFSIZE,
"%s.%d.%"PRIu64".%c.heap",
opt_prof_prefix, (int)getpid(), prof_dump_seq, v);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
}
Fix heap profiling bugs. Fix a potential deadlock that could occur during interval- and growth-triggered heap profile dumps. Fix an off-by-one heap profile statistics bug that could be observed in interval- and growth-triggered heap profiles. Fix heap profile dump filename sequence numbers (regression during conversion to malloc_snprintf()).
2012-04-23 07:00:11 +08:00
prof_dump_seq++;
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
}
static void
prof_fdump(void)
{
char filename[DUMP_FILENAME_BUFSIZE];
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
if (prof_booted == false)
return;
Update prof defaults to match common usage. Change the "opt.lg_prof_sample" default from 0 to 19 (1 B to 512 KiB). Change the "opt.prof_accum" default from true to false. Add the "opt.prof_final" mallctl, so that "opt.prof_prefix" need not be abused to disable final profile dumping.
2012-04-18 07:39:33 +08:00
if (opt_prof_final && opt_prof_prefix[0] != '\0') {
Replace JEMALLOC_OPTIONS with MALLOC_CONF. Replace the single-character run-time flags with key/value pairs, which can be set via the malloc_conf global, /etc/malloc.conf, and the MALLOC_CONF environment variable. Replace the JEMALLOC_PROF_PREFIX environment variable with the "opt.prof_prefix" option. Replace umax2s() with u2s().
2010-10-24 09:37:06 +08:00
malloc_mutex_lock(&prof_dump_seq_mtx);
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
prof_dump_filename(filename, 'f', VSEQ_INVALID);
Replace JEMALLOC_OPTIONS with MALLOC_CONF. Replace the single-character run-time flags with key/value pairs, which can be set via the malloc_conf global, /etc/malloc.conf, and the MALLOC_CONF environment variable. Replace the JEMALLOC_PROF_PREFIX environment variable with the "opt.prof_prefix" option. Replace umax2s() with u2s().
2010-10-24 09:37:06 +08:00
malloc_mutex_unlock(&prof_dump_seq_mtx);
Implement malloc_vsnprintf(). Implement malloc_vsnprintf() (a subset of vsnprintf(3)) as well as several other printing functions based on it, so that formatted printing can be relied upon without concern for inducing a dependency on floating point runtime support. Replace malloc_write() calls with malloc_*printf() where doing so simplifies the code. Add name mangling for library-private symbols in the data and BSS sections. Adjust CONF_HANDLE_*() macros in malloc_conf_init() to expose all opt_* variable use to cpp so that proper mangling occurs.
2012-03-07 06:57:45 +08:00
prof_dump(false, filename, opt_prof_leak);
Replace JEMALLOC_OPTIONS with MALLOC_CONF. Replace the single-character run-time flags with key/value pairs, which can be set via the malloc_conf global, /etc/malloc.conf, and the MALLOC_CONF environment variable. Replace the JEMALLOC_PROF_PREFIX environment variable with the "opt.prof_prefix" option. Replace umax2s() with u2s().
2010-10-24 09:37:06 +08:00
}
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
}
void
prof_idump(void)
{
Fix heap profiling bugs. Fix a potential deadlock that could occur during interval- and growth-triggered heap profile dumps. Fix an off-by-one heap profile statistics bug that could be observed in interval- and growth-triggered heap profiles. Fix heap profile dump filename sequence numbers (regression during conversion to malloc_snprintf()).
2012-04-23 07:00:11 +08:00
prof_tdata_t *prof_tdata;
Implement malloc_vsnprintf(). Implement malloc_vsnprintf() (a subset of vsnprintf(3)) as well as several other printing functions based on it, so that formatted printing can be relied upon without concern for inducing a dependency on floating point runtime support. Replace malloc_write() calls with malloc_*printf() where doing so simplifies the code. Add name mangling for library-private symbols in the data and BSS sections. Adjust CONF_HANDLE_*() macros in malloc_conf_init() to expose all opt_* variable use to cpp so that proper mangling occurs.
2012-03-07 06:57:45 +08:00
char filename[PATH_MAX + 1];
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
if (prof_booted == false)
return;
Fix potential TLS-related memory corruption. Avoid writing to uninitialized TLS as a side effect of deallocation. Initializing TLS during deallocation is unsafe because it is possible that a thread never did any allocation, and that TLS has already been deallocated by the threads library, resulting in write-after-free corruption. These fixes affect prof_tdata and quarantine; all other uses of TLS are already safe, whether intentionally (as for tcache) or unintentionally (as for arenas).
2013-01-31 07:03:11 +08:00
prof_tdata = prof_tdata_get(false);
Handle prof_tdata resurrection. Handle prof_tdata resurrection during thread shutdown, similarly to how tcache and quarantine handle resurrection.
2012-04-29 09:14:24 +08:00
if ((uintptr_t)prof_tdata <= (uintptr_t)PROF_TDATA_STATE_MAX)
Fix heap profiling bugs. Fix a potential deadlock that could occur during interval- and growth-triggered heap profile dumps. Fix an off-by-one heap profile statistics bug that could be observed in interval- and growth-triggered heap profiles. Fix heap profile dump filename sequence numbers (regression during conversion to malloc_snprintf()).
2012-04-23 07:00:11 +08:00
return;
if (prof_tdata->enq) {
prof_tdata->enq_idump = true;
Implement interval-based heap profile dumping. Add mallctl interfaces for profiling parameters. Fix a file descriptor leak in heap profile dumping.
2010-02-12 05:19:21 +08:00
return;
}
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
Replace JEMALLOC_OPTIONS with MALLOC_CONF. Replace the single-character run-time flags with key/value pairs, which can be set via the malloc_conf global, /etc/malloc.conf, and the MALLOC_CONF environment variable. Replace the JEMALLOC_PROF_PREFIX environment variable with the "opt.prof_prefix" option. Replace umax2s() with u2s().
2010-10-24 09:37:06 +08:00
if (opt_prof_prefix[0] != '\0') {
malloc_mutex_lock(&prof_dump_seq_mtx);
prof_dump_filename(filename, 'i', prof_dump_iseq);
prof_dump_iseq++;
malloc_mutex_unlock(&prof_dump_seq_mtx);
Implement malloc_vsnprintf(). Implement malloc_vsnprintf() (a subset of vsnprintf(3)) as well as several other printing functions based on it, so that formatted printing can be relied upon without concern for inducing a dependency on floating point runtime support. Replace malloc_write() calls with malloc_*printf() where doing so simplifies the code. Add name mangling for library-private symbols in the data and BSS sections. Adjust CONF_HANDLE_*() macros in malloc_conf_init() to expose all opt_* variable use to cpp so that proper mangling occurs.
2012-03-07 06:57:45 +08:00
prof_dump(false, filename, false);
Replace JEMALLOC_OPTIONS with MALLOC_CONF. Replace the single-character run-time flags with key/value pairs, which can be set via the malloc_conf global, /etc/malloc.conf, and the MALLOC_CONF environment variable. Replace the JEMALLOC_PROF_PREFIX environment variable with the "opt.prof_prefix" option. Replace umax2s() with u2s().
2010-10-24 09:37:06 +08:00
}
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
}
Allow prof.dump mallctl to specify filename.
2010-03-03 03:57:30 +08:00
bool
prof_mdump(const char *filename)
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
{
Allow prof.dump mallctl to specify filename.
2010-03-03 03:57:30 +08:00
char filename_buf[DUMP_FILENAME_BUFSIZE];
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Allow prof.dump mallctl to specify filename.
2010-03-03 03:57:30 +08:00
if (opt_prof == false || prof_booted == false)
return (true);
if (filename == NULL) {
/* No filename specified, so automatically generate one. */
Replace JEMALLOC_OPTIONS with MALLOC_CONF. Replace the single-character run-time flags with key/value pairs, which can be set via the malloc_conf global, /etc/malloc.conf, and the MALLOC_CONF environment variable. Replace the JEMALLOC_PROF_PREFIX environment variable with the "opt.prof_prefix" option. Replace umax2s() with u2s().
2010-10-24 09:37:06 +08:00
if (opt_prof_prefix[0] == '\0')
return (true);
Allow prof.dump mallctl to specify filename.
2010-03-03 03:57:30 +08:00
malloc_mutex_lock(&prof_dump_seq_mtx);
prof_dump_filename(filename_buf, 'm', prof_dump_mseq);
prof_dump_mseq++;
malloc_mutex_unlock(&prof_dump_seq_mtx);
filename = filename_buf;
}
Implement malloc_vsnprintf(). Implement malloc_vsnprintf() (a subset of vsnprintf(3)) as well as several other printing functions based on it, so that formatted printing can be relied upon without concern for inducing a dependency on floating point runtime support. Replace malloc_write() calls with malloc_*printf() where doing so simplifies the code. Add name mangling for library-private symbols in the data and BSS sections. Adjust CONF_HANDLE_*() macros in malloc_conf_init() to expose all opt_* variable use to cpp so that proper mangling occurs.
2012-03-07 06:57:45 +08:00
return (prof_dump(true, filename, false));
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
}
void
Replace JEMALLOC_OPTIONS with MALLOC_CONF. Replace the single-character run-time flags with key/value pairs, which can be set via the malloc_conf global, /etc/malloc.conf, and the MALLOC_CONF environment variable. Replace the JEMALLOC_PROF_PREFIX environment variable with the "opt.prof_prefix" option. Replace umax2s() with u2s().
2010-10-24 09:37:06 +08:00
prof_gdump(void)
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
{
Fix heap profiling bugs. Fix a potential deadlock that could occur during interval- and growth-triggered heap profile dumps. Fix an off-by-one heap profile statistics bug that could be observed in interval- and growth-triggered heap profiles. Fix heap profile dump filename sequence numbers (regression during conversion to malloc_snprintf()).
2012-04-23 07:00:11 +08:00
prof_tdata_t *prof_tdata;
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
char filename[DUMP_FILENAME_BUFSIZE];
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
if (prof_booted == false)
return;
Fix potential TLS-related memory corruption. Avoid writing to uninitialized TLS as a side effect of deallocation. Initializing TLS during deallocation is unsafe because it is possible that a thread never did any allocation, and that TLS has already been deallocated by the threads library, resulting in write-after-free corruption. These fixes affect prof_tdata and quarantine; all other uses of TLS are already safe, whether intentionally (as for tcache) or unintentionally (as for arenas).
2013-01-31 07:03:11 +08:00
prof_tdata = prof_tdata_get(false);
Handle prof_tdata resurrection. Handle prof_tdata resurrection during thread shutdown, similarly to how tcache and quarantine handle resurrection.
2012-04-29 09:14:24 +08:00
if ((uintptr_t)prof_tdata <= (uintptr_t)PROF_TDATA_STATE_MAX)
Fix heap profiling bugs. Fix a potential deadlock that could occur during interval- and growth-triggered heap profile dumps. Fix an off-by-one heap profile statistics bug that could be observed in interval- and growth-triggered heap profiles. Fix heap profile dump filename sequence numbers (regression during conversion to malloc_snprintf()).
2012-04-23 07:00:11 +08:00
return;
if (prof_tdata->enq) {
prof_tdata->enq_gdump = true;
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
return;
}
Replace JEMALLOC_OPTIONS with MALLOC_CONF. Replace the single-character run-time flags with key/value pairs, which can be set via the malloc_conf global, /etc/malloc.conf, and the MALLOC_CONF environment variable. Replace the JEMALLOC_PROF_PREFIX environment variable with the "opt.prof_prefix" option. Replace umax2s() with u2s().
2010-10-24 09:37:06 +08:00
if (opt_prof_prefix[0] != '\0') {
malloc_mutex_lock(&prof_dump_seq_mtx);
prof_dump_filename(filename, 'u', prof_dump_useq);
prof_dump_useq++;
malloc_mutex_unlock(&prof_dump_seq_mtx);
Implement malloc_vsnprintf(). Implement malloc_vsnprintf() (a subset of vsnprintf(3)) as well as several other printing functions based on it, so that formatted printing can be relied upon without concern for inducing a dependency on floating point runtime support. Replace malloc_write() calls with malloc_*printf() where doing so simplifies the code. Add name mangling for library-private symbols in the data and BSS sections. Adjust CONF_HANDLE_*() macros in malloc_conf_init() to expose all opt_* variable use to cpp so that proper mangling occurs.
2012-03-07 06:57:45 +08:00
prof_dump(false, filename, false);
Replace JEMALLOC_OPTIONS with MALLOC_CONF. Replace the single-character run-time flags with key/value pairs, which can be set via the malloc_conf global, /etc/malloc.conf, and the MALLOC_CONF environment variable. Replace the JEMALLOC_PROF_PREFIX environment variable with the "opt.prof_prefix" option. Replace umax2s() with u2s().
2010-10-24 09:37:06 +08:00
}
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
}
static void
Update hash from MurmurHash2 to MurmurHash3. Update hash from MurmurHash2 to MurmurHash3, primarily because the latter generates 128 bits in a single call for no extra cost, which simplifies integration with cuckoo hashing.
2013-01-23 04:02:08 +08:00
prof_bt_hash(const void *key, size_t r_hash[2])
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
{
prof_bt_t *bt = (prof_bt_t *)key;
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
Update hash from MurmurHash2 to MurmurHash3. Update hash from MurmurHash2 to MurmurHash3, primarily because the latter generates 128 bits in a single call for no extra cost, which simplifies integration with cuckoo hashing.
2013-01-23 04:02:08 +08:00
hash(bt->vec, bt->len * sizeof(void *), 0x94122f33U, r_hash);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
}
static bool
prof_bt_keycomp(const void *k1, const void *k2)
{
const prof_bt_t *bt1 = (prof_bt_t *)k1;
const prof_bt_t *bt2 = (prof_bt_t *)k2;
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
if (bt1->len != bt2->len)
return (false);
return (memcmp(bt1->vec, bt2->vec, bt1->len * sizeof(void *)) == 0);
}
Inline the fast path for heap sampling. Inline the heap sampling code that is executed for every allocation event (regardless of whether a sample is taken). Combine all prof TLS data into a single data structure, in order to reduce the TLS lookup volume.
2010-10-21 10:05:59 +08:00
prof_tdata_t *
prof_tdata_init(void)
{
prof_tdata_t *prof_tdata;
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Inline the fast path for heap sampling. Inline the heap sampling code that is executed for every allocation event (regardless of whether a sample is taken). Combine all prof TLS data into a single data structure, in order to reduce the TLS lookup volume.
2010-10-21 10:05:59 +08:00
/* Initialize an empty cache for this thread. */
prof_tdata = (prof_tdata_t *)imalloc(sizeof(prof_tdata_t));
if (prof_tdata == NULL)
return (NULL);
if (ckh_new(&prof_tdata->bt2cnt, PROF_CKH_MINITEMS,
prof_bt_hash, prof_bt_keycomp)) {
idalloc(prof_tdata);
return (NULL);
}
ql_new(&prof_tdata->lru_ql);
Remove the opt.lg_prof_bt_max option. Remove opt.lg_prof_bt_max, and hard code it to 7. The original intention of this option was to enable faster backtracing by limiting backtrace depth. However, this makes graphical pprof output very difficult to interpret. In practice, decreasing sampling frequency is a better mechanism for limiting profiling overhead.
2012-02-14 10:23:41 +08:00
prof_tdata->vec = imalloc(sizeof(void *) * PROF_BT_MAX);
Inline the fast path for heap sampling. Inline the heap sampling code that is executed for every allocation event (regardless of whether a sample is taken). Combine all prof TLS data into a single data structure, in order to reduce the TLS lookup volume.
2010-10-21 10:05:59 +08:00
if (prof_tdata->vec == NULL) {
ckh_delete(&prof_tdata->bt2cnt);
idalloc(prof_tdata);
return (NULL);
}
Rename prn to prng. Rename prn to prng so that Windows doesn't choke when trying to create a file named prn.h.
2012-03-03 07:59:45 +08:00
prof_tdata->prng_state = 0;
Inline the fast path for heap sampling. Inline the heap sampling code that is executed for every allocation event (regardless of whether a sample is taken). Combine all prof TLS data into a single data structure, in order to reduce the TLS lookup volume.
2010-10-21 10:05:59 +08:00
prof_tdata->threshold = 0;
prof_tdata->accum = 0;
Fix heap profiling bugs. Fix a potential deadlock that could occur during interval- and growth-triggered heap profile dumps. Fix an off-by-one heap profile statistics bug that could be observed in interval- and growth-triggered heap profiles. Fix heap profile dump filename sequence numbers (regression during conversion to malloc_snprintf()).
2012-04-23 07:00:11 +08:00
prof_tdata->enq = false;
prof_tdata->enq_idump = false;
prof_tdata->enq_gdump = false;
Implement tsd. Implement tsd, which is a TLS/TSD abstraction that uses one or both internally. Modify bootstrapping such that no tsd's are utilized until allocation is safe. Remove malloc_[v]tprintf(), and use malloc_snprintf() instead. Fix %p argument size handling in malloc_vsnprintf(). Fix a long-standing statistics-related bug in the "thread.arena" mallctl that could cause crashes due to linked list corruption.
2012-03-22 09:33:03 +08:00
prof_tdata_tsd_set(&prof_tdata);
Inline the fast path for heap sampling. Inline the heap sampling code that is executed for every allocation event (regardless of whether a sample is taken). Combine all prof TLS data into a single data structure, in order to reduce the TLS lookup volume.
2010-10-21 10:05:59 +08:00
return (prof_tdata);
}
Implement tsd. Implement tsd, which is a TLS/TSD abstraction that uses one or both internally. Modify bootstrapping such that no tsd's are utilized until allocation is safe. Remove malloc_[v]tprintf(), and use malloc_snprintf() instead. Fix %p argument size handling in malloc_vsnprintf(). Fix a long-standing statistics-related bug in the "thread.arena" mallctl that could cause crashes due to linked list corruption.
2012-03-22 09:33:03 +08:00
void
Inline the fast path for heap sampling. Inline the heap sampling code that is executed for every allocation event (regardless of whether a sample is taken). Combine all prof TLS data into a single data structure, in order to reduce the TLS lookup volume.
2010-10-21 10:05:59 +08:00
prof_tdata_cleanup(void *arg)
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
{
Use prof_tdata_cleanup() argument. Use the argument to prof_tdata_cleanup(), rather than calling PROF_TCACHE_GET(). This fixes a bug in the NO_TLS case.
2011-08-09 08:10:07 +08:00
prof_thr_cnt_t *cnt;
Implement tsd. Implement tsd, which is a TLS/TSD abstraction that uses one or both internally. Modify bootstrapping such that no tsd's are utilized until allocation is safe. Remove malloc_[v]tprintf(), and use malloc_snprintf() instead. Fix %p argument size handling in malloc_vsnprintf(). Fix a long-standing statistics-related bug in the "thread.arena" mallctl that could cause crashes due to linked list corruption.
2012-03-22 09:33:03 +08:00
prof_tdata_t *prof_tdata = *(prof_tdata_t **)arg;
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Handle prof_tdata resurrection. Handle prof_tdata resurrection during thread shutdown, similarly to how tcache and quarantine handle resurrection.
2012-04-29 09:14:24 +08:00
if (prof_tdata == PROF_TDATA_STATE_REINCARNATED) {
/*
* Another destructor deallocated memory after this destructor
* was called. Reset prof_tdata to PROF_TDATA_STATE_PURGATORY
* in order to receive another callback.
*/
prof_tdata = PROF_TDATA_STATE_PURGATORY;
prof_tdata_tsd_set(&prof_tdata);
} else if (prof_tdata == PROF_TDATA_STATE_PURGATORY) {
/*
* The previous time this destructor was called, we set the key
* to PROF_TDATA_STATE_PURGATORY so that other destructors
* wouldn't cause re-creation of the prof_tdata. This time, do
* nothing, so that the destructor will not be called again.
*/
} else if (prof_tdata != NULL) {
/*
* Delete the hash table. All of its contents can still be
* iterated over via the LRU.
*/
ckh_delete(&prof_tdata->bt2cnt);
/*
* Iteratively merge cnt's into the global stats and delete
* them.
*/
while ((cnt = ql_last(&prof_tdata->lru_ql, lru_link)) != NULL) {
ql_remove(&prof_tdata->lru_ql, cnt, lru_link);
prof_ctx_merge(cnt->ctx, cnt);
idalloc(cnt);
}
idalloc(prof_tdata->vec);
idalloc(prof_tdata);
prof_tdata = PROF_TDATA_STATE_PURGATORY;
prof_tdata_tsd_set(&prof_tdata);
Use prof_tdata_cleanup() argument. Use the argument to prof_tdata_cleanup(), rather than calling PROF_TCACHE_GET(). This fixes a bug in the NO_TLS case.
2011-08-09 08:10:07 +08:00
}
Increase default backtrace depth from 4 to 128. Increase the default backtrace depth, because shallow backtraces tend to result in confusing pprof output graphs.
2010-10-03 13:38:14 +08:00
}
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
void
prof_boot0(void)
Replace JEMALLOC_OPTIONS with MALLOC_CONF. Replace the single-character run-time flags with key/value pairs, which can be set via the malloc_conf global, /etc/malloc.conf, and the MALLOC_CONF environment variable. Replace the JEMALLOC_PROF_PREFIX environment variable with the "opt.prof_prefix" option. Replace umax2s() with u2s().
2010-10-24 09:37:06 +08:00
{
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Replace JEMALLOC_OPTIONS with MALLOC_CONF. Replace the single-character run-time flags with key/value pairs, which can be set via the malloc_conf global, /etc/malloc.conf, and the MALLOC_CONF environment variable. Replace the JEMALLOC_PROF_PREFIX environment variable with the "opt.prof_prefix" option. Replace umax2s() with u2s().
2010-10-24 09:37:06 +08:00
memcpy(opt_prof_prefix, PROF_PREFIX_DEFAULT,
sizeof(PROF_PREFIX_DEFAULT));
}
void
prof_boot1(void)
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
{
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
/*
Remove support for non-prof-promote heap profiling metadata. Make promotion of sampled small objects to large objects mandatory, so that profiling metadata can always be stored in the chunk map, rather than requiring one pointer per small region in each small-region page run. In practice the non-prof-promote code was only useful when using jemalloc to track all objects and report them as leaks at program exit. However, Valgrind is at least as good a tool for this particular use case. Furthermore, the non-prof-promote code is getting in the way of some optimizations that will make heap profiling much cheaper for the predominant use case (sampling a small representative proportion of all allocations).
2014-04-12 05:24:51 +08:00
* opt_prof must be in its final state before any arenas are
* initialized, so this function must be executed early.
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
*/
if (opt_prof_leak && opt_prof == false) {
/*
* Enable opt_prof, but in such a way that profiles are never
* automatically dumped.
*/
opt_prof = true;
Replace JEMALLOC_OPTIONS with MALLOC_CONF. Replace the single-character run-time flags with key/value pairs, which can be set via the malloc_conf global, /etc/malloc.conf, and the MALLOC_CONF environment variable. Replace the JEMALLOC_PROF_PREFIX environment variable with the "opt.prof_prefix" option. Replace umax2s() with u2s().
2010-10-24 09:37:06 +08:00
opt_prof_gdump = false;
Make interval-triggered profile dumping optional. Make it possible to disable interval-triggered profile dumping, even if profiling is enabled. This is useful if the user only wants a single dump at exit, or if the application manually triggers profile dumps.
2010-04-01 08:35:51 +08:00
} else if (opt_prof) {
if (opt_lg_prof_interval >= 0) {
prof_interval = (((uint64_t)1U) <<
opt_lg_prof_interval);
Avoid arena_prof_accum()-related locking when possible. Refactor arena_prof_accum() and its callers to avoid arena locking when prof_interval is 0 (as when profiling is disabled). Reported by Ben Maurer.
2012-11-14 04:56:27 +08:00
}
Make interval-triggered profile dumping optional. Make it possible to disable interval-triggered profile dumping, even if profiling is enabled. This is useful if the user only wants a single dump at exit, or if the application manually triggers profile dumps.
2010-04-01 08:35:51 +08:00
}
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
}
bool
Replace JEMALLOC_OPTIONS with MALLOC_CONF. Replace the single-character run-time flags with key/value pairs, which can be set via the malloc_conf global, /etc/malloc.conf, and the MALLOC_CONF environment variable. Replace the JEMALLOC_PROF_PREFIX environment variable with the "opt.prof_prefix" option. Replace umax2s() with u2s().
2010-10-24 09:37:06 +08:00
prof_boot2(void)
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
{
Reduce cpp conditional logic complexity. Convert configuration-related cpp conditional logic to use static constant variables, e.g.: #ifdef JEMALLOC_DEBUG [...] #endif becomes: if (config_debug) { [...] } The advantage is clearer, more concise code. The main disadvantage is that data structures no longer have conditionally defined fields, so they pay the cost of all fields regardless of whether they are used. In practice, this is only a minor concern; config_stats will go away in an upcoming change, and config_prof is the only other major feature that depends on more than a few special-purpose fields.
2012-02-11 12:22:09 +08:00
cassert(config_prof);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
if (opt_prof) {
Remove ephemeral mutexes. Remove ephemeral mutexes from the prof machinery, and remove malloc_mutex_destroy(). This simplifies mutex management on systems that call malloc()/free() inside pthread_mutex_{create,destroy}(). Add atomic_*_u() for operation on unsigned values. Fix prof_printf() to call malloc_vsnprintf() rather than malloc_snprintf().
2012-03-24 09:05:51 +08:00
unsigned i;
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
if (ckh_new(&bt2ctx, PROF_CKH_MINITEMS, prof_bt_hash,
prof_bt_keycomp))
return (true);
if (malloc_mutex_init(&bt2ctx_mtx))
return (true);
Implement tsd. Implement tsd, which is a TLS/TSD abstraction that uses one or both internally. Modify bootstrapping such that no tsd's are utilized until allocation is safe. Remove malloc_[v]tprintf(), and use malloc_snprintf() instead. Fix %p argument size handling in malloc_vsnprintf(). Fix a long-standing statistics-related bug in the "thread.arena" mallctl that could cause crashes due to linked list corruption.
2012-03-22 09:33:03 +08:00
if (prof_tdata_tsd_boot()) {
Simplify malloc_message(). Rather than passing four strings to malloc_message(), malloc_write4(), and all the functions that use them, only pass one string.
2010-03-04 09:45:38 +08:00
malloc_write(
"<jemalloc>: Error in pthread_key_create()\n");
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
abort();
}
if (malloc_mutex_init(&prof_dump_seq_mtx))
return (true);
Refactor prof_dump() to reduce contention. Refactor prof_dump() to use a two pass algorithm, and prof_leave() prior to the second pass. This avoids write(2) system calls while holding critical prof resources. Fix prof_dump() to close the dump file descriptor for all relevant error paths. Minimize the size of prof-related static buffers when prof is disabled. This saves roughly 65 KiB of application memory for non-prof builds. Refactor prof_ctx_init() out of prof_lookup_global().
2014-01-17 05:23:56 +08:00
if (malloc_mutex_init(&prof_dump_mtx))
return (true);
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
if (atexit(prof_fdump) != 0) {
Simplify malloc_message(). Rather than passing four strings to malloc_message(), malloc_write4(), and all the functions that use them, only pass one string.
2010-03-04 09:45:38 +08:00
malloc_write("<jemalloc>: Error in atexit()\n");
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
if (opt_abort)
abort();
}
Remove ephemeral mutexes. Remove ephemeral mutexes from the prof machinery, and remove malloc_mutex_destroy(). This simplifies mutex management on systems that call malloc()/free() inside pthread_mutex_{create,destroy}(). Add atomic_*_u() for operation on unsigned values. Fix prof_printf() to call malloc_vsnprintf() rather than malloc_snprintf().
2012-03-24 09:05:51 +08:00
ctx_locks = (malloc_mutex_t *)base_alloc(PROF_NCTX_LOCKS *
sizeof(malloc_mutex_t));
if (ctx_locks == NULL)
return (true);
for (i = 0; i < PROF_NCTX_LOCKS; i++) {
if (malloc_mutex_init(&ctx_locks[i]))
return (true);
}
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
}
Various heap profiling improvements. Add the --disable-prof-libgcc configure option, and add backtracing based on libgcc, which is used by default. Fix a bug in hash(). Fix various configuration-dependent compilation errors.
2010-02-11 10:15:53 +08:00
#ifdef JEMALLOC_PROF_LIBGCC
/*
* Cause the backtracing machinery to allocate its internal state
* before enabling profiling.
*/
_Unwind_Backtrace(prof_unwind_init_callback, NULL);
#endif
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
prof_booted = true;
return (false);
}
Fix fork(2)-related deadlocks. Add a library constructor for jemalloc that initializes the allocator. This fixes a race that could occur if threads were created by the main thread prior to any memory allocation, followed by fork(2), and then memory allocation in the child process. Fix the prefork/postfork functions to acquire/release the ctl, prof, and rtree mutexes. This fixes various fork() child process deadlocks, but one possible deadlock remains (intentionally) unaddressed: prof backtracing can acquire runtime library mutexes, so deadlock is still possible if heap profiling is enabled during fork(). This deadlock is known to be a real issue in at least the case of libgcc-based backtracing. Reported by tfengjun.
2012-10-10 05:46:22 +08:00
void
prof_prefork(void)
{
if (opt_prof) {
unsigned i;
Consistently use malloc_mutex_prefork(). Consistently use malloc_mutex_prefork() instead of malloc_mutex_lock() in all prefork functions.
2013-10-22 05:59:10 +08:00
malloc_mutex_prefork(&bt2ctx_mtx);
malloc_mutex_prefork(&prof_dump_seq_mtx);
Fix fork(2)-related deadlocks. Add a library constructor for jemalloc that initializes the allocator. This fixes a race that could occur if threads were created by the main thread prior to any memory allocation, followed by fork(2), and then memory allocation in the child process. Fix the prefork/postfork functions to acquire/release the ctl, prof, and rtree mutexes. This fixes various fork() child process deadlocks, but one possible deadlock remains (intentionally) unaddressed: prof backtracing can acquire runtime library mutexes, so deadlock is still possible if heap profiling is enabled during fork(). This deadlock is known to be a real issue in at least the case of libgcc-based backtracing. Reported by tfengjun.
2012-10-10 05:46:22 +08:00
for (i = 0; i < PROF_NCTX_LOCKS; i++)
Consistently use malloc_mutex_prefork(). Consistently use malloc_mutex_prefork() instead of malloc_mutex_lock() in all prefork functions.
2013-10-22 05:59:10 +08:00
malloc_mutex_prefork(&ctx_locks[i]);
Fix fork(2)-related deadlocks. Add a library constructor for jemalloc that initializes the allocator. This fixes a race that could occur if threads were created by the main thread prior to any memory allocation, followed by fork(2), and then memory allocation in the child process. Fix the prefork/postfork functions to acquire/release the ctl, prof, and rtree mutexes. This fixes various fork() child process deadlocks, but one possible deadlock remains (intentionally) unaddressed: prof backtracing can acquire runtime library mutexes, so deadlock is still possible if heap profiling is enabled during fork(). This deadlock is known to be a real issue in at least the case of libgcc-based backtracing. Reported by tfengjun.
2012-10-10 05:46:22 +08:00
}
}
void
prof_postfork_parent(void)
{
if (opt_prof) {
unsigned i;
for (i = 0; i < PROF_NCTX_LOCKS; i++)
malloc_mutex_postfork_parent(&ctx_locks[i]);
malloc_mutex_postfork_parent(&prof_dump_seq_mtx);
malloc_mutex_postfork_parent(&bt2ctx_mtx);
}
}
void
prof_postfork_child(void)
{
if (opt_prof) {
unsigned i;
for (i = 0; i < PROF_NCTX_LOCKS; i++)
malloc_mutex_postfork_child(&ctx_locks[i]);
malloc_mutex_postfork_child(&prof_dump_seq_mtx);
malloc_mutex_postfork_child(&bt2ctx_mtx);
}
}
Implement allocation profiling and leack checking. Add the --enable-prof and --enable-prof-libunwind configure options. Add the B/b, F/f, I/i, L/l, and U/u JEMALLOC_OPTIONS. Interval-based profile dump triggering is not yet implemented. Add supporting generic code: * Add memory barriers. * Add prn (LCG PRNG). * Add hash (Murmur hash function). * Add ckh (cuckoo hash tables).
2010-02-11 02:37:56 +08:00
/******************************************************************************/
Reference in New Issue Copy Permalink