bytes of uninitialized memory\&. The allocated space is suitably aligned (after possible pointer coercion) for storage of any type of object\&.
.PP
The
\fBcalloc\fR\fB\fR
calloc()
function allocates space for
\fInumber\fR
objects, each
\fIsize\fR
bytes in length\&. The result is identical to calling
\fBmalloc\fR\fB\fR
malloc()
with an argument of
\fInumber\fR
*
\fIsize\fR, with the exception that the allocated memory is explicitly initialized to zero bytes\&.
.PP
The
\fBposix_memalign\fR\fB\fR
posix_memalign()
function allocates
\fIsize\fR
bytes of memory such that the allocation\*(Aqs base address is a multiple of
...
...
@@ -116,7 +116,7 @@ must be a power of 2 at least as large as
sizeof(\fBvoid *\fR)\&.
.PP
The
\fBaligned_alloc\fR\fB\fR
aligned_alloc()
function allocates
\fIsize\fR
bytes of memory such that the allocation\*(Aqs base address is a multiple of
...
...
@@ -128,7 +128,7 @@ is not an integral multiple of
\fIalignment\fR\&.
.PP
The
\fBrealloc\fR\fB\fR
realloc()
function changes the size of the previously allocated memory referenced by
\fIptr\fR
to
...
...
@@ -136,19 +136,19 @@ to
bytes\&. The contents of the memory are unchanged up to the lesser of the new and old sizes\&. If the new size is larger, the contents of the newly allocated portion of the memory are undefined\&. Upon success, the memory referenced by
\fIptr\fR
is freed and a pointer to the newly allocated memory is returned\&. Note that
\fBrealloc\fR\fB\fR
realloc()
may move the memory allocation, resulting in a different return value than
\fIptr\fR\&. If
\fIptr\fR
is
\fBNULL\fR, the
\fBrealloc\fR\fB\fR
realloc()
function behaves identically to
\fBmalloc\fR\fB\fR
malloc()
for the specified size\&.
.PP
The
\fBfree\fR\fB\fR
free()
function causes the allocated memory referenced by
\fIptr\fR
to be made available for future allocations\&. If
...
...
@@ -158,13 +158,13 @@ is
.SS "Non\-standard API"
.PP
The
\fBmallocx\fR\fB\fR,
\fBrallocx\fR\fB\fR,
\fBxallocx\fR\fB\fR,
\fBsallocx\fR\fB\fR,
\fBdallocx\fR\fB\fR,
\fBsdallocx\fR\fB\fR, and
\fBnallocx\fR\fB\fR
mallocx(),
rallocx(),
xallocx(),
sallocx(),
dallocx(),
sdallocx(), and
nallocx()
functions all have a
\fIflags\fR
argument that can be used to specify options\&. The functions only check the options that are contextually relevant\&. Use bitwise or (|) operations to specify one or more of the following:
...
...
@@ -196,7 +196,7 @@ Initialize newly allocated memory to contain zero bytes\&. In the growing reallo
.RS 4
Use the thread\-specific cache (tcache) specified by the identifier
\fItc\fR, which must have been acquired via the
"tcache\&.create"
tcache\&.create
mallctl\&. This macro does not validate that
\fItc\fR
specifies a valid identifier\&.
...
...
@@ -223,16 +223,16 @@ specifies an arena index in the valid range\&.
.RE
.PP
The
\fBmallocx\fR\fB\fR
mallocx()
function allocates at least
\fIsize\fR
bytes of memory, and returns a pointer to the base address of the allocation\&. Behavior is undefined if
\fIsize\fR
is
\fB0\fR, or if request size overflows due to size class and/or alignment constraints\&.
\fB0\fR\&.
.PP
The
\fBrallocx\fR\fB\fR
rallocx()
function resizes the allocation at
\fIptr\fR
to be at least
...
...
@@ -240,10 +240,10 @@ to be at least
bytes, and returns a pointer to the base address of the resulting allocation, which may or may not have moved from its original location\&. Behavior is undefined if
\fIsize\fR
is
\fB0\fR, or if request size overflows due to size class and/or alignment constraints\&.
\fB0\fR\&.
.PP
The
\fBxallocx\fR\fB\fR
xallocx()
function resizes the allocation at
\fIptr\fR
in place to be at least
...
...
@@ -259,40 +259,42 @@ is
(\fIsize\fR + \fIextra\fR > \fBSIZE_T_MAX\fR)\&.
.PP
The
\fBsallocx\fR\fB\fR
sallocx()
function returns the real size of the allocation at
\fIptr\fR\&.
.PP
The
\fBdallocx\fR\fB\fR
dallocx()
function causes the memory referenced by
\fIptr\fR
to be made available for future allocations\&.
.PP
The
\fBsdallocx\fR\fB\fR
sdallocx()
function is an extension of
\fBdallocx\fR\fB\fR
dallocx()
with a
\fIsize\fR
parameter to allow the caller to pass in the allocation size as an optimization\&. The minimum valid input size is the original requested size of the allocation, and the maximum valid input size is the corresponding value returned by
\fBnallocx\fR\fB\fR
nallocx()
or
\fBsallocx\fR\fB\fR\&.
sallocx()\&.
.PP
The
\fBnallocx\fR\fB\fR
nallocx()
function allocates no memory, but it performs the same size computation as the
\fBmallocx\fR\fB\fR
mallocx()
function, and returns the real size of the allocation that would result from the equivalent
\fBmallocx\fR\fB\fR
function call\&. Behavior is undefined if
mallocx()
function call, or
\fB0\fR
if the inputs exceed the maximum supported size class and/or alignment\&. Behavior is undefined if
\fIsize\fR
is
\fB0\fR, or if request size overflows due to size class and/or alignment constraints\&.
\fB0\fR\&.
.PP
The
\fBmallctl\fR\fB\fR
mallctl()
function provides a general interface for introspecting the memory allocator, as well as setting modifiable parameters and triggering actions\&. The period\-separated
\fIname\fR
argument specifies a location in a tree\-structured namespace; see the
...
...
@@ -311,10 +313,12 @@ and
\fB0\fR\&.
.PP
The
\fBmallctlnametomib\fR\fB\fR
function provides a way to avoid repeated name lookups for applications that repeatedly query the same portion of the namespace, by translating a name to a \(lqManagement Information Base\(rq (MIB) that can be passed repeatedly to
\fBmallctlbymib\fR\fB\fR\&. Upon successful return from
\fBmallctlnametomib\fR\fB\fR,
mallctlnametomib()
function provides a way to avoid repeated name lookups for applications that repeatedly query the same portion of the namespace, by translating a name to a
\(lqManagement Information Base\(rq
(MIB) that can be passed repeatedly to
mallctlbymib()\&. Upon successful return from
mallctlnametomib(),
\fImibp\fR
contains an array of
\fI*miblenp\fR
...
...
@@ -326,7 +330,7 @@ and the input value of
\fI*miblenp\fR\&. Thus it is possible to pass a
\fI*miblenp\fR
that is smaller than the number of period\-separated name components, which results in a partial MIB that can be used as the basis for constructing a complete MIB\&. For name components that are integers (e\&.g\&. the 2 in
"arenas\&.bin\&.2\&.size"), the corresponding MIB component will always be that integer\&. Therefore, it is legitimate to construct code like the following:
arenas\&.bin\&.2\&.size), the corresponding MIB component will always be that integer\&. Therefore, it is legitimate to construct code like the following:
@@ -355,67 +359,87 @@ for (i = 0; i < nbins; i++) {
.\}
.PP
The
\fBmalloc_stats_print\fR\fB\fR
function writes human\-readable summary statistics via the
malloc_stats_print()
function writes summary statistics via the
\fIwrite_cb\fR
callback function pointer and
\fIcbopaque\fR
data passed to
\fIwrite_cb\fR, or
\fBmalloc_message\fR\fB\fR
malloc_message()
if
\fIwrite_cb\fR
is
\fBNULL\fR\&. This function can be called repeatedly\&. General information that never changes during execution can be omitted by specifying "g" as a character within the
\fBNULL\fR\&. The statistics are presented in human\-readable form unless
\(lqJ\(rq
is specified as a character within the
\fIopts\fR
string, in which case the statistics are presented in
\m[blue]\fBJSON format\fR\m[]\&\s-2\u[2]\d\s+2\&. This function can be called repeatedly\&. General information that never changes during execution can be omitted by specifying
\(lqg\(rq
as a character within the
\fIopts\fR
string\&. Note that
\fBmalloc_message\fR\fB\fR
malloc_message()
uses the
\fBmallctl*\fR\fB\fR
mallctl*()
functions internally, so inconsistent statistics can be reported if multiple threads use these functions simultaneously\&. If
\fB\-\-enable\-stats\fR
is specified during configuration, \(lqm\(rq and \(lqa\(rq can be specified to omit merged arena and per arena statistics, respectively; \(lqb\(rq, \(lql\(rq, and \(lqh\(rq can be specified to omit per size class statistics for bins, large objects, and huge objects, respectively\&. Unrecognized characters are silently ignored\&. Note that thread caching may prevent some statistics from being completely up to date, since extra locking would be required to merge counters that track thread cache operations\&.
is specified during configuration,
\(lqm\(rq
and
\(lqa\(rq
can be specified to omit merged arena and per arena statistics, respectively;
\(lqb\(rq,
\(lql\(rq, and
\(lqh\(rq
can be specified to omit per size class statistics for bins, large objects, and huge objects, respectively\&. Unrecognized characters are silently ignored\&. Note that thread caching may prevent some statistics from being completely up to date, since extra locking would be required to merge counters that track thread cache operations\&.
.PP
The
\fBmalloc_usable_size\fR\fB\fR
malloc_usable_size()
function returns the usable size of the allocation pointed to by
\fIptr\fR\&. The return value may be larger than the size that was requested during allocation\&. The
\fBmalloc_usable_size\fR\fB\fR
malloc_usable_size()
function is not a mechanism for in\-place
\fBrealloc\fR\fB\fR; rather it is provided solely as a tool for introspection purposes\&. Any discrepancy between the requested allocation size and the size reported by
\fBmalloc_usable_size\fR\fB\fR
realloc(); rather it is provided solely as a tool for introspection purposes\&. Any discrepancy between the requested allocation size and the size reported by
malloc_usable_size()
should not be depended on, since such behavior is entirely implementation\-dependent\&.
.SH "TUNING"
.PP
Once, when the first call is made to one of the memory allocation routines, the allocator initializes its internals based in part on various options that can be specified at compile\- or run\-time\&.
.PP
The string pointed to by the global variable
\fImalloc_conf\fR, the \(lqname\(rq of the file referenced by the symbolic link named
The string specified via
\fB\-\-with\-malloc\-conf\fR, the string pointed to by the global variable
\fImalloc_conf\fR, the
\(lqname\(rq
of the file referenced by the symbolic link named
/etc/malloc\&.conf, and the value of the environment variable
\fBMALLOC_CONF\fR, will be interpreted, in that order, from left to right as options\&. Note that
\fImalloc_conf\fR
may be read before
\fBmain\fR\fB\fR
main()
is entered, so the declaration of
\fImalloc_conf\fR
should specify an initializer that contains the final value to be read by jemalloc\&.
\fB\-\-with\-malloc\-conf\fR
and
\fImalloc_conf\fR
is a compile\-time setting, whereas
are compile\-time mechanisms, whereas
/etc/malloc\&.conf
and
\fBMALLOC_CONF\fR
can be safely set any time prior to program invocation\&.
.PP
An options string is a comma\-separated list of option:value pairs\&. There is one key corresponding to each
"opt\&.*"
opt\&.*
mallctl (see the
MALLCTL NAMESPACE
section for options documentation)\&. For example,
abort:true,narenas:1
sets the
"opt\&.abort"
opt\&.abort
and
"opt\&.narenas"
opt\&.narenas
options\&. Some options have boolean values (true/false), others have integer values (base 8, 10, or 16, depending on prefix), and yet others have raw string values\&.
.SH "IMPLEMENTATION NOTES"
.PP
...
...
@@ -436,29 +460,26 @@ In addition to multiple arenas, unless
\fB\-\-disable\-tcache\fR
is specified during configuration, this allocator supports thread\-specific caching for small and large objects, in order to make it possible to completely avoid synchronization for most allocation requests\&. Such caching allows very fast allocation in the common case, but it increases memory usage and fragmentation, since a bounded number of objects can remain allocated in each thread cache\&.
.PP
Memory is conceptually broken into equal\-sized chunks, where the chunk size is a power of two that is greater than the page size\&. Chunks are always aligned to multiples of the chunk size\&. This alignment makes it possible to find metadata for user objects very quickly\&.
.PP
User objects are broken into three categories according to size: small, large, and huge\&. Small and large objects are managed entirely by arenas; huge objects are additionally aggregated in a single data structure that is shared by all threads\&. Huge objects are typically used by applications infrequently enough that this single data structure is not a scalability issue\&.
.PP
Each chunk that is managed by an arena tracks its contents as runs of contiguous pages (unused, backing a set of small objects, or backing one large object)\&. The combination of chunk alignment and chunk page maps makes it possible to determine all metadata regarding small and large allocations in constant time\&.
Memory is conceptually broken into equal\-sized chunks, where the chunk size is a power of two that is greater than the page size\&. Chunks are always aligned to multiples of the chunk size\&. This alignment makes it possible to find metadata for user objects very quickly\&. User objects are broken into three categories according to size: small, large, and huge\&. Multiple small and large objects can reside within a single chunk, whereas huge objects each have one or more chunks backing them\&. Each chunk that contains small and/or large objects tracks its contents as runs of contiguous pages (unused, backing a set of small objects, or backing one large object)\&. The combination of chunk alignment and chunk page maps makes it possible to determine all metadata regarding small and large allocations in constant time\&.
.PP
Small objects are managed in groups by page runs\&. Each run maintains a bitmap to track which regions are in use\&. Allocation requests that are no more than half the quantum (8 or 16, depending on architecture) are rounded up to the nearest power of two that is at least
sizeof(\fBdouble\fR)\&. All other object size classes are multiples of the quantum, spaced such that there are four size classes for each doubling in size, which limits internal fragmentation to approximately 20% for all but the smallest size classes\&. Small size classes are smaller than four times the page size, large size classes are smaller than the chunk size (see the
"opt\&.lg_chunk"
option), and huge size classes extend from the chunk size up to one size class less than the full address space size\&.
opt\&.lg_chunk
option), and huge size classes extend from the chunk size up to the largest size class that does not exceed
\fBPTRDIFF_MAX\fR\&.
.PP
Allocations are packed tightly together, which can be an issue for multi\-threaded applications\&. If you need to assure that allocations do not suffer from cacheline sharing, round your allocation requests up to the nearest multiple of the cacheline size, or specify cacheline alignment when allocating\&.
.PP
The
\fBrealloc\fR\fB\fR,
\fBrallocx\fR\fB\fR, and
\fBxallocx\fR\fB\fR
realloc(),
rallocx(), and
xallocx()
functions may resize allocations without moving them under limited circumstances\&. Unlike the
\fB*allocx\fR\fB\fR
*allocx()
API, the standard API does not officially round up the usable size of an allocation to the nearest size class, so technically it is necessary to call
\fBrealloc\fR\fB\fR
realloc()
to grow e\&.g\&. a 9\-byte allocation to 16 bytes, or shrink a 16\-byte allocation to 9 bytes\&. Growth and shrinkage trivially succeeds in place as long as the pre\-size and post\-size both round up to the same size class\&. No other API guarantees are made regarding in\-place resizing, but the current implementation also tries to resize large and huge allocations in place, as long as the pre\-size and post\-size are both large or both huge\&. In such cases shrinkage always succeeds for large size classes, but for huge size classes the chunk allocator must support splitting (see
"arena\&.<i>\&.chunk_hooks")\&. Growth only succeeds if the trailing memory is currently available, and additionally for huge size classes the chunk allocator must support merging\&.
arena\&.<i>\&.chunk_hooks)\&. Growth only succeeds if the trailing memory is currently available, and additionally for huge size classes the chunk allocator must support merging\&.
.PP
Assuming 2 MiB chunks, 4 KiB pages, and a 16\-byte quantum on a 64\-bit system, the size classes in each category are as shown in
Table 1\&.
...
...
@@ -502,6 +523,8 @@ l r l
^ r l
^ r l
^ r l
^ r l
^ r l
^ r l.
T{
Small
...
...
@@ -629,12 +652,22 @@ T}
T}:T{
\&.\&.\&.
T}
:T{
512 PiB
T}:T{
[2560 PiB, 3 EiB, 3584 PiB, 4 EiB]
T}
:T{
1 EiB
T}:T{
[5 EiB, 6 EiB, 7 EiB]
T}
.TE
.sp 1
.SH "MALLCTL NAMESPACE"
.PP
The following names are defined in the namespace accessible via the
\fBmallctl*\fR\fB\fR
mallctl*()
functions\&. Value types are specified in parentheses, their readable/writable statuses are encoded as
rw,
r\-,
...
...
@@ -644,111 +677,118 @@ r\-,
or
<j>
indicates an integer component, where the integer varies from 0 to some upper value that must be determined via introspection\&. In the case of
"stats\&.arenas\&.<i>\&.*",
stats\&.arenas\&.<i>\&.*,
<i>
equal to
"arenas\&.narenas"
arenas\&.narenas
can be used to access the summation of statistics from all arenas\&. Take special note of the
"epoch"
epoch
mallctl, which controls refreshing of cached dynamic statistics\&.
.PP
"version" (\fBconst char *\fR) r\-
version (\fBconst char *\fR) r\-
.RS 4
Return the jemalloc version string\&.
.RE
.PP
"epoch" (\fBuint64_t\fR) rw
epoch (\fBuint64_t\fR) rw
.RS 4
If a value is passed in, refresh the data from which the
\fBmallctl*\fR\fB\fR
mallctl*()
functions report values, and increment the epoch\&. Return the current epoch\&. This is useful for detecting whether another thread caused a refresh\&.
Abort\-on\-warning enabled/disabled\&. If true, most warnings are fatal\&. The process will call
\fBabort\fR(3)
...
...
@@ -757,97 +797,132 @@ in these cases\&. This option is disabled by default unless
is specified during configuration, in which case it is enabled by default\&.
.RE
.PP
"opt\&.dss" (\fBconst char *\fR) r\-
opt\&.dss (\fBconst char *\fR) r\-
.RS 4
dss (\fBsbrk\fR(2)) allocation precedence as related to
\fBmmap\fR(2)
allocation\&. The following settings are supported if
\fBsbrk\fR(2)
is supported by the operating system: \(lqdisabled\(rq, \(lqprimary\(rq, and \(lqsecondary\(rq; otherwise only \(lqdisabled\(rq is supported\&. The default is \(lqsecondary\(rq if
is supported by the operating system:
\(lqdisabled\(rq,
\(lqprimary\(rq, and
\(lqsecondary\(rq; otherwise only
\(lqdisabled\(rq
is supported\&. The default is
\(lqsecondary\(rq
if
\fBsbrk\fR(2)
is supported by the operating system; \(lqdisabled\(rq otherwise\&.
is supported by the operating system;
\(lqdisabled\(rq
otherwise\&.
.RE
.PP
"opt\&.lg_chunk" (\fBsize_t\fR) r\-
opt\&.lg_chunk (\fBsize_t\fR) r\-
.RS 4
Virtual memory chunk size (log base 2)\&. If a chunk size outside the supported size range is specified, the size is silently clipped to the minimum/maximum supported size\&. The default chunk size is 2 MiB (2^21)\&.
.RE
.PP
"opt\&.narenas" (\fBsize_t\fR) r\-
opt\&.narenas (\fBunsigned\fR) r\-
.RS 4
Maximum number of arenas to use for automatic multiplexing of threads and arenas\&. The default is four times the number of CPUs, or one if there is a single CPU\&.
.RE
.PP
"opt\&.lg_dirty_mult" (\fBssize_t\fR) r\-
opt\&.purge (\fBconst char *\fR) r\-
.RS 4
Purge mode is \(lqratio\(rq (default) or \(lqdecay\(rq\&. See
opt\&.lg_dirty_mult
for details of the ratio mode\&. See
opt\&.decay_time
for details of the decay mode\&.
.RE
.PP
opt\&.lg_dirty_mult (\fBssize_t\fR) r\-
.RS 4
Per\-arena minimum ratio (log base 2) of active to dirty pages\&. Some dirty unused pages may be allowed to accumulate, within the limit set by the ratio (or one chunk worth of dirty pages, whichever is greater), before informing the kernel about some of those pages via
\fBmadvise\fR(2)
or a similar system call\&. This provides the kernel with sufficient information to recycle dirty pages if physical memory becomes scarce and the pages remain unused\&. The default minimum ratio is 8:1 (2^3:1); an option value of \-1 will disable dirty page purging\&. See
"arenas\&.lg_dirty_mult"
arenas\&.lg_dirty_mult
and
"arena\&.<i>\&.lg_dirty_mult"
arena\&.<i>\&.lg_dirty_mult
for related dynamic control options\&.
.RE
.PP
"opt\&.stats_print" (\fBbool\fR) r\-
opt\&.decay_time (\fBssize_t\fR) r\-
.RS 4
Approximate time in seconds from the creation of a set of unused dirty pages until an equivalent set of unused dirty pages is purged and/or reused\&. The pages are incrementally purged according to a sigmoidal decay curve that starts and ends with zero purge rate\&. A decay time of 0 causes all unused dirty pages to be purged immediately upon creation\&. A decay time of \-1 disables purging\&. The default decay time is 10 seconds\&. See
arenas\&.decay_time
and
arena\&.<i>\&.decay_time
for related dynamic control options\&.
.RE
.PP
opt\&.stats_print (\fBbool\fR) r\-
.RS 4
Enable/disable statistics printing at exit\&. If enabled, the
\fBmalloc_stats_print\fR\fB\fR
malloc_stats_print()
function is called at program exit via an
\fBatexit\fR(3)
function\&. If
\fB\-\-enable\-stats\fR
is specified during configuration, this has the potential to cause deadlock for a multi\-threaded process that exits while one or more threads are executing in the memory allocation functions\&. Furthermore,
\fBatexit\fR\fB\fR
atexit()
may allocate memory during application initialization and then deadlock internally when jemalloc in turn calls
\fBatexit\fR\fB\fR, so this option is not univerally usable (though the application can register its own
\fBatexit\fR\fB\fR
atexit(), so this option is not universally usable (though the application can register its own
atexit()
function with equivalent functionality)\&. Therefore, this option should only be used with care; it is primarily intended as a performance tuning aid during application development\&. This option is disabled by default\&.
Junk filling\&. If set to "alloc", each byte of uninitialized allocated memory will be initialized to
0xa5\&. If set to "free", all deallocated memory will be initialized to
0x5a\&. If set to "true", both allocated and deallocated memory will be initialized, and if set to "false", junk filling be disabled entirely\&. This is intended for debugging and will impact performance negatively\&. This option is "false" by default unless
Junk filling\&. If set to
\(lqalloc\(rq, each byte of uninitialized allocated memory will be initialized to
0xa5\&. If set to
\(lqfree\(rq, all deallocated memory will be initialized to
0x5a\&. If set to
\(lqtrue\(rq, both allocated and deallocated memory will be initialized, and if set to
\(lqfalse\(rq, junk filling be disabled entirely\&. This is intended for debugging and will impact performance negatively\&. This option is
\(lqfalse\(rq
by default unless
\fB\-\-enable\-debug\fR
is specified during configuration, in which case it is "true" by default unless running inside
\m[blue]\fBValgrind\fR\m[]\&\s-2\u[2]\d\s+2\&.
is specified during configuration, in which case it is
Per thread quarantine size in bytes\&. If non\-zero, each thread maintains a FIFO object quarantine that stores up to the specified number of bytes of memory\&. The quarantined memory is not freed until it is released from quarantine, though it is immediately junk\-filled if the
"opt\&.junk"
opt\&.junk
option is enabled\&. This feature is of particular use in combination with
\m[blue]\fBValgrind\fR\m[]\&\s-2\u[2]\d\s+2, which can detect attempts to access quarantined objects\&. This is intended for debugging and will impact performance negatively\&. The default quarantine size is 0 unless running inside Valgrind, in which case the default is 16 MiB\&.
\m[blue]\fBValgrind\fR\m[]\&\s-2\u[3]\d\s+2, which can detect attempts to access quarantined objects\&. This is intended for debugging and will impact performance negatively\&. The default quarantine size is 0 unless running inside Valgrind, in which case the default is 16 MiB\&.
Redzones enabled/disabled\&. If enabled, small allocations have redzones before and after them\&. Furthermore, if the
"opt\&.junk"
opt\&.junk
option is enabled, the redzones are checked for corruption during deallocation\&. However, the primary intended purpose of this feature is to be used in combination with
\m[blue]\fBValgrind\fR\m[]\&\s-2\u[2]\d\s+2, which needs redzones in order to do effective buffer overflow/underflow detection\&. This option is intended for debugging and will impact performance negatively\&. This option is disabled by default unless running inside Valgrind\&.
\m[blue]\fBValgrind\fR\m[]\&\s-2\u[3]\d\s+2, which needs redzones in order to do effective buffer overflow/underflow detection\&. This option is intended for debugging and will impact performance negatively\&. This option is disabled by default unless running inside Valgrind\&.
Zero filling enabled/disabled\&. If enabled, each byte of uninitialized allocated memory will be initialized to 0\&. Note that this initialization only happens once for each byte, so
\fBrealloc\fR\fB\fR
realloc()
and
\fBrallocx\fR\fB\fR
rallocx()
calls do not zero memory that was previously allocated\&. This is intended for debugging and will impact performance negatively\&. This option is disabled by default\&.
Thread\-specific caching (tcache) enabled/disabled\&. When there are multiple threads, each thread uses a tcache for objects up to a certain size\&. Thread\-specific caching allows many allocations to be satisfied without performing any thread synchronization, at the cost of increased memory use\&. See the
"opt\&.lg_tcache_max"
opt\&.lg_tcache_max
option for related tuning information\&. This option is enabled by default unless running inside
\m[blue]\fBValgrind\fR\m[]\&\s-2\u[2]\d\s+2, in which case it is forcefully disabled\&.
\m[blue]\fBValgrind\fR\m[]\&\s-2\u[3]\d\s+2, in which case it is forcefully disabled\&.
Maximum size class (log base 2) to cache in the thread\-specific cache (tcache)\&. At a minimum, all small size classes are cached, and at a maximum all large size classes are cached\&. The default maximum is 32 KiB (2^15)\&.
Filename prefix for profile dumps\&. If the prefix is set to the empty string, no automatic dumps will occur; this is primarily useful for disabling the automatic final heap dump (which also disables leak reporting, if enabled)\&. The default prefix is
Profiling activated/deactivated\&. This is a secondary control mechanism that makes it possible to start the application with profiling enabled (see the
"opt\&.prof"
opt\&.prof
option) but inactive, then toggle profiling at any time during program execution with the
Average interval (log base 2) between allocation samples, as measured in bytes of allocation activity\&. Increasing the sampling interval decreases profile fidelity, but also decreases the computational overhead\&. The default sample interval is 512 KiB (2^19 B)\&.
Reporting of cumulative object/byte counts in profile dumps enabled/disabled\&. If this option is enabled, every unique backtrace must be stored for the duration of execution\&. Depending on the application, this can impose a large memory overhead, and the cumulative counts are not always of interest\&. This option is disabled by default\&.
Average interval (log base 2) between memory profile dumps, as measured in bytes of allocation activity\&. The actual interval between dumps may be sporadic because decentralized allocation counters are used to avoid synchronization bottlenecks\&. Profiles are dumped to files named according to the pattern
<prefix>\&.<pid>\&.<seq>\&.i<iseq>\&.heap, where
<prefix>
is controlled by the
"opt\&.prof_prefix"
opt\&.prof_prefix
option\&. By default, interval\-triggered profile dumping is disabled (encoded as \-1)\&.
"prof\&.gdump", which when enabled triggers a memory profile dump every time the total virtual memory exceeds the previous maximum\&. This option is disabled by default\&.
prof\&.gdump, which when enabled triggers a memory profile dump every time the total virtual memory exceeds the previous maximum\&. This option is disabled by default\&.
Get the total number of bytes ever allocated by the calling thread\&. This counter has the potential to wrap around; it is up to the application to appropriately interpret the counter in such cases\&.
Get the total number of bytes ever deallocated by the calling thread\&. This counter has the potential to wrap around; it is up to the application to appropriately interpret the counter in such cases\&.
Flush calling thread\*(Aqs thread\-specific cache (tcache)\&. This interface releases all cached objects and internal data structures associated with the calling thread\*(Aqs tcache\&. Ordinarily, this interface need not be called, since automatic periodic incremental garbage collection occurs, and the thread cache is automatically discarded when a thread exits\&. However, garbage collection is triggered by allocation activity, so it is possible for a thread that stops allocating/deallocating to retain its cache indefinitely, in which case the developer may find manual flushing useful\&.
.RE
.PP
"thread\&.prof\&.name" (\fBconst char *\fR) r\- or \-w [\fB\-\-enable\-prof\fR]
thread\&.prof\&.name (\fBconst char *\fR) r\- or \-w [\fB\-\-enable\-prof\fR]
.RS 4
Get/set the descriptive name associated with the calling thread in memory profile dumps\&. An internal copy of the name string is created, so the input string need not be maintained after this interface completes execution\&. The output string of this interface should be copied for non\-ephemeral uses, because multiple implementation details can cause asynchronous string deallocation\&. Furthermore, each invocation of this interface can only read or write; simultaneous read/write is not supported due to string lifetime limitations\&. The name string must nil\-terminated and comprised only of characters in the sets recognized by
Get/set the descriptive name associated with the calling thread in memory profile dumps\&. An internal copy of the name string is created, so the input string need not be maintained after this interface completes execution\&. The output string of this interface should be copied for non\-ephemeral uses, because multiple implementation details can cause asynchronous string deallocation\&. Furthermore, each invocation of this interface can only read or write; simultaneous read/write is not supported due to string lifetime limitations\&. The name string must be nil\-terminated and comprised only of characters in the sets recognized by
Create an explicit thread\-specific cache (tcache) and return an identifier that can be passed to the
\fBMALLOCX_TCACHE(\fR\fB\fItc\fR\fR\fB)\fR
macro to explicitly use the specified cache rather than the automatically managed one that is used by default\&. Each explicit cache can be used by only one thread at a time; the application must assure that this constraint holds\&.
Flush the specified thread\-specific cache (tcache) and make the identifier available for use during a future tcache creation\&.
.RE
.PP
"arena\&.<i>\&.purge" (\fBvoid\fR) \-\-
arena\&.<i>\&.purge (\fBvoid\fR) \-\-
.RS 4
Purge unused dirty pages for arena <i>, or for all arenas if <i> equals
"arenas\&.narenas"\&.
Purge all unused dirty pages for arena <i>, or for all arenas if <i> equals
arenas\&.narenas\&.
.RE
.PP
arena\&.<i>\&.decay (\fBvoid\fR) \-\-
.RS 4
Trigger decay\-based purging of unused dirty pages for arena <i>, or for all arenas if <i> equals
arenas\&.narenas\&. The proportion of unused dirty pages to be purged depends on the current time; see
opt\&.decay_time
for details\&.
.RE
.PP
"arena\&.<i>\&.dss" (\fBconst char *\fR) rw
arena\&.<i>\&.reset (\fBvoid\fR) \-\-
.RS 4
Discard all of the arena\*(Aqs extant allocations\&. This interface can only be used with arenas created via
arenas\&.extend\&. None of the arena\*(Aqs discarded/cached allocations may accessed afterward\&. As part of this requirement, all thread caches which were used to allocate/deallocate in conjunction with the arena must be flushed beforehand\&. This interface cannot be used if running inside Valgrind, nor if the
quarantine
size is non\-zero\&.
.RE
.PP
arena\&.<i>\&.dss (\fBconst char *\fR) rw
.RS 4
Set the precedence of dss allocation as related to mmap allocation for arena <i>, or for all arenas if <i> equals
"arenas\&.narenas"\&. See
"opt\&.dss"
arenas\&.narenas\&. See
opt\&.dss
for supported settings\&.
.RE
.PP
"arena\&.<i>\&.lg_dirty_mult" (\fBssize_t\fR) rw
arena\&.<i>\&.lg_dirty_mult (\fBssize_t\fR) rw
.RS 4
Current per\-arena minimum ratio (log base 2) of active to dirty pages for arena <i>\&. Each time this interface is set and the ratio is increased, pages are synchronously purged as necessary to impose the new ratio\&. See
"opt\&.lg_dirty_mult"
opt\&.lg_dirty_mult
for additional information\&.
.RE
.PP
arena\&.<i>\&.decay_time (\fBssize_t\fR) rw
.RS 4
Current per\-arena approximate time in seconds from the creation of a set of unused dirty pages until an equivalent set of unused dirty pages is purged and/or reused\&. Each time this interface is set, all currently unused dirty pages are considered to have fully decayed, which causes immediate purging of all unused dirty pages unless the decay time is set to \-1 (i\&.e\&. purging disabled)\&. See
Get or set the chunk management hook functions for arena <i>\&. The functions must be capable of operating on all extant chunks associated with arena <i>, usually by passing unknown chunks to the replaced functions\&. In practice, it is feasible to control allocation for arenas created via
"arenas\&.extend"
arenas\&.extend
such that all chunks originate from an application\-supplied chunk allocator (by setting custom chunk hook functions just after arena creation), but the automatically created arenas may have already created chunks prior to the application having an opportunity to take over chunk allocation\&.
.sp
.if n \{\
...
...
@@ -1149,7 +1249,7 @@ is not
on success or
\fBNULL\fR
on error\&. Committed memory may be committed in absolute terms as on a system that does not overcommit, or in implicit terms as on a system that overcommits and satisfies physical memory needs on demand via soft page faults\&. Note that replacing the default chunk allocation function makes the arena\*(Aqs
@@ -1296,407 +1396,504 @@ into one contiguous chunk, operating on
\fIarena_ind\fR, returning false upon success\&. If the function returns true, this indicates that the chunks remain distinct mappings and therefore should continue to be operated on independently\&.
.RE
.PP
"arenas\&.narenas" (\fBunsigned\fR) r\-
arenas\&.narenas (\fBunsigned\fR) r\-
.RS 4
Current limit on number of arenas\&.
.RE
.PP
"arenas\&.initialized" (\fBbool *\fR) r\-
arenas\&.initialized (\fBbool *\fR) r\-
.RS 4
An array of
"arenas\&.narenas"
arenas\&.narenas
booleans\&. Each boolean indicates whether the corresponding arena is initialized\&.
.RE
.PP
"arenas\&.lg_dirty_mult" (\fBssize_t\fR) rw
arenas\&.lg_dirty_mult (\fBssize_t\fR) rw
.RS 4
Current default per\-arena minimum ratio (log base 2) of active to dirty pages, used to initialize
"arena\&.<i>\&.lg_dirty_mult"
arena\&.<i>\&.lg_dirty_mult
during arena creation\&. See
"opt\&.lg_dirty_mult"
opt\&.lg_dirty_mult
for additional information\&.
.RE
.PP
"arenas\&.quantum" (\fBsize_t\fR) r\-
arenas\&.decay_time (\fBssize_t\fR) rw
.RS 4
Current default per\-arena approximate time in seconds from the creation of a set of unused dirty pages until an equivalent set of unused dirty pages is purged and/or reused, used to initialize
When enabled, trigger a memory profile dump every time the total virtual memory exceeds the previous maximum\&. Profiles are dumped to files named according to the pattern
Pointer to a counter that contains an approximate count of the current number of bytes in active pages\&. The estimate may be high, but never low, because each arena rounds up when computing its contribution to the counter\&. Note that the
"epoch"
epoch
mallctl has no bearing on this counter\&. Furthermore, counter consistency is maintained via atomic operations, so it is necessary to use an atomic operation in order to guarantee a consistent read when dereferencing the pointer\&.
Total number of bytes dedicated to metadata, which comprise base allocations used for bootstrap\-sensitive internal allocator data structures, arena chunk headers (see
"stats\&.arenas\&.<i>\&.metadata\&.mapped"), and internal allocations (see
"stats\&.arenas\&.<i>\&.metadata\&.allocated")\&.
stats\&.arenas\&.<i>\&.metadata\&.mapped), and internal allocations (see
Maximum number of bytes in physically resident data pages mapped by the allocator, comprising all pages dedicated to allocator metadata, pages backing active allocations, and unused dirty pages\&. This is a maximum rather than precise because pages may not actually be physically resident if they correspond to demand\-zeroed virtual memory that has not yet been touched\&. This is a multiple of the page size, and is larger than
Total number of bytes in active chunks mapped by the allocator\&. This is a multiple of the chunk size, and is larger than
"stats\&.active"\&. This does not include inactive chunks, even those that contain unused dirty pages, which means that there is no strict ordering between this and
"stats\&.resident"\&.
stats\&.active\&. This does not include inactive chunks, even those that contain unused dirty pages, which means that there is no strict ordering between this and
Approximate time in seconds from the creation of a set of unused dirty pages until an equivalent set of unused dirty pages is purged and/or reused\&. See
Number of bytes dedicated to internal allocations\&. Internal allocations differ from application\-originated allocations in that they are for internal use, and that they are omitted from heap profiles\&. This statistic is reported separately from
"stats\&.metadata"
stats\&.metadata
and
"stats\&.arenas\&.<i>\&.metadata\&.mapped"
stats\&.arenas\&.<i>\&.metadata\&.mapped
because it overlaps with e\&.g\&. the
"stats\&.allocated"
stats\&.allocated
and
"stats\&.active"
stats\&.active
statistics, whereas the other metadata statistics do not\&.
Current number of huge allocations for this size class\&.
.RE
.SH "HEAP PROFILE FORMAT"
.PP
Although the heap profiling functionality was originally designed to be compatible with the
\fBpprof\fR
command that is developed as part of the
\m[blue]\fBgperftools package\fR\m[]\&\s-2\u[4]\d\s+2, the addition of per thread heap profiling functionality required a different heap profile format\&. The
\fBjeprof\fR
command is derived from
\fBpprof\fR, with enhancements to support the heap profile format described here\&.
When debugging, it is a good idea to configure/build jemalloc with the
...
...
@@ -1705,14 +1902,16 @@ and
\fB\-\-enable\-fill\fR
options, and recompile the program with suitable options and symbols for debugger support\&. When so configured, jemalloc incorporates a wide variety of run\-time assertions that catch application errors such as double\-free, write\-after\-free, etc\&.
.PP
Programs often accidentally depend on \(lquninitialized\(rq memory actually being filled with zero bytes\&. Junk filling (see the
"opt\&.junk"
Programs often accidentally depend on
\(lquninitialized\(rq
memory actually being filled with zero bytes\&. Junk filling (see the
opt\&.junk
option) tends to expose such bugs in the form of obviously incorrect results and/or coredumps\&. Conversely, zero filling (see the
"opt\&.zero"
opt\&.zero
option) eliminates the symptoms of such bugs\&. Between these two options, it is usually possible to quickly detect, diagnose, and eliminate such bugs\&.
.PP
This implementation does not provide much detail about the problems it detects, because the performance impact for storing such information would be prohibitive\&. However, jemalloc does integrate with the most excellent
\m[blue]\fBValgrind\fR\m[]\&\s-2\u[2]\d\s+2
\m[blue]\fBValgrind\fR\m[]\&\s-2\u[3]\d\s+2
tool if the
\fB\-\-enable\-valgrind\fR
configuration option is enabled\&.
...
...
@@ -1720,7 +1919,7 @@ configuration option is enabled\&.
.PP
If any of the memory allocation/deallocation functions detect an error or warning condition, a message will be printed to file descriptor
\fBSTDERR_FILENO\fR\&. Errors will result in the process dumping core\&. If the
"opt\&.abort"
opt\&.abort
option is set, most warnings are treated as errors\&.
.PP
The
...
...
@@ -1728,22 +1927,23 @@ The
variable allows the programmer to override the function which emits the text strings forming the errors and warnings if for some reason the
\fBSTDERR_FILENO\fR
file descriptor is not suitable for this\&.
\fBmalloc_message\fR\fB\fR
malloc_message()
takes the
\fIcbopaque\fR
pointer argument that is
\fBNULL\fR
unless overridden by the arguments in a call to
\fBmalloc_stats_print\fR\fB\fR, followed by a string pointer\&. Please note that doing anything which tries to allocate memory in this function is likely to result in a crash or deadlock\&.
malloc_stats_print(), followed by a string pointer\&. Please note that doing anything which tries to allocate memory in this function is likely to result in a crash or deadlock\&.
.PP
All messages are prefixed by \(lq<jemalloc>:\(rq\&.
All messages are prefixed by
\(lq<jemalloc>: \(rq\&.
.SH "RETURN VALUES"
.SS "Standard API"
.PP
The
\fBmalloc\fR\fB\fR
malloc()
and
\fBcalloc\fR\fB\fR
calloc()
functions return a pointer to the allocated memory if successful; otherwise a
\fBNULL\fR
pointer is returned and
...
...
@@ -1752,9 +1952,9 @@ is set to
ENOMEM\&.
.PP
The
\fBposix_memalign\fR\fB\fR
posix_memalign()
function returns the value 0 if successful; otherwise it returns an error value\&. The
<emclass="parameter"><code>size</code></em> bytes of uninitialized memory. The allocated
space is suitably aligned (after possible pointer coercion) for storage
of any type of object.</p><p>The <codeclass="function">calloc</code>(<emclass="parameter"><code></code></em>) function allocates
of any type of object.</p><p>The <codeclass="function">calloc()</code> function allocates
space for <emclass="parameter"><code>number</code></em> objects, each
<emclass="parameter"><code>size</code></em> bytes in length. The result is identical to
calling <codeclass="function">malloc</code>(<emclass="parameter"><code></code></em>) with an argument of
calling <codeclass="function">malloc()</code> with an argument of
<emclass="parameter"><code>number</code></em> * <emclass="parameter"><code>size</code></em>, with the
exception that the allocated memory is explicitly initialized to zero
bytes.</p><p>The <codeclass="function">posix_memalign</code>(<emclass="parameter"><code></code></em>) function
bytes.</p><p>The <codeclass="function">posix_memalign()</code> function
allocates <emclass="parameter"><code>size</code></em> bytes of memory such that the
allocation's base address is a multiple of
<emclass="parameter"><code>alignment</code></em>, and returns the allocation in the value
pointed to by <emclass="parameter"><code>ptr</code></em>. The requested
<emclass="parameter"><code>alignment</code></em> must be a power of 2 at least as large as
<codeclass="code">sizeof(<spanclass="type">void *</span>)</code>.</p><p>The <codeclass="function">aligned_alloc</code>(<emclass="parameter"><code></code></em>) function
<codeclass="code">sizeof(<spanclass="type">void *</span>)</code>.</p><p>The <codeclass="function">aligned_alloc()</code> function
allocates <emclass="parameter"><code>size</code></em> bytes of memory such that the
allocation's base address is a multiple of
<emclass="parameter"><code>alignment</code></em>. The requested
<emclass="parameter"><code>alignment</code></em> must be a power of 2. Behavior is
undefined if <emclass="parameter"><code>size</code></em> is not an integral multiple of
<emclass="parameter"><code>alignment</code></em>.</p><p>The <codeclass="function">realloc</code>(<emclass="parameter"><code></code></em>) function changes the
<emclass="parameter"><code>alignment</code></em>.</p><p>The <codeclass="function">realloc()</code> function changes the
size of the previously allocated memory referenced by
<emclass="parameter"><code>ptr</code></em> to <emclass="parameter"><code>size</code></em> bytes. The
contents of the memory are unchanged up to the lesser of the new and old
...
...
@@ -30,69 +31,65 @@
portion of the memory are undefined. Upon success, the memory referenced
by <emclass="parameter"><code>ptr</code></em> is freed and a pointer to the newly
allocated memory is returned. Note that
<codeclass="function">realloc</code>(<emclass="parameter"><code></code></em>) may move the memory allocation,
<codeclass="function">realloc()</code> may move the memory allocation,
resulting in a different return value than <emclass="parameter"><code>ptr</code></em>.
If <emclass="parameter"><code>ptr</code></em> is <codeclass="constant">NULL</code>, the
<codeclass="function">realloc</code>(<emclass="parameter"><code></code></em>) function behaves identically to
<codeclass="function">malloc</code>(<emclass="parameter"><code></code></em>) for the specified size.</p><p>The <codeclass="function">free</code>(<emclass="parameter"><code></code></em>) function causes the
<codeclass="function">realloc()</code> function behaves identically to
<codeclass="function">malloc()</code> for the specified size.</p><p>The <codeclass="function">free()</code> function causes the
allocated memory referenced by <emclass="parameter"><code>ptr</code></em> to be made
available for future allocations. If <emclass="parameter"><code>ptr</code></em> is
<codeclass="constant">NULL</code>, no action occurs.</p></div><divclass="refsect2"><aname="idp46144704"></a><h3>Non-standard API</h3><p>The <codeclass="function">mallocx</code>(<emclass="parameter"><code></code></em>),
</code></span></dt><dd><p>Align the memory allocation to start at an address
that is a multiple of <emclass="parameter"><code>a</code></em>, where
<emclass="parameter"><code>a</code></em> is a power of two. This macro does not
validate that <emclass="parameter"><code>a</code></em> is a power of 2.
</p></dd><dt><aname="MALLOCX_ZERO"></a><spanclass="term"><codeclass="constant">MALLOCX_ZERO</code></span></dt><dd><p>Initialize newly allocated memory to contain zero
</p></dd><dt><aname="MALLOCX_ZERO"/><spanclass="term"><codeclass="constant">MALLOCX_ZERO</code></span></dt><dd><p>Initialize newly allocated memory to contain zero
bytes. In the growing reallocation case, the real size prior to
reallocation defines the boundary between untouched bytes and those
that are initialized to contain zero bytes. If this macro is
absent, newly allocated memory is uninitialized.</p></dd><dt><aname="MALLOCX_TCACHE"></a><spanclass="term"><codeclass="constant">MALLOCX_TCACHE(<emclass="parameter"><code>tc</code></em>)
absent, newly allocated memory is uninitialized.</p></dd><dt><aname="MALLOCX_TCACHE"/><spanclass="term"><codeclass="constant">MALLOCX_TCACHE(<emclass="parameter"><code>tc</code></em>)
</code></span></dt><dd><p>Use the thread-specific cache (tcache) specified by
the identifier <emclass="parameter"><code>tc</code></em>, which must have been
acquired via the <aclass="link"href="#tcache.create">
"<codeclass="mallctl">tcache.create</code>"
</a>
acquired via the <aclass="link"href="#tcache.create"><quote><codeclass="mallctl">tcache.create</code></quote></a>
mallctl. This macro does not validate that
<emclass="parameter"><code>tc</code></em> specifies a valid
identifier.</p></dd><dt><aname="MALLOC_TCACHE_NONE"></a><spanclass="term"><codeclass="constant">MALLOCX_TCACHE_NONE</code></span></dt><dd><p>Do not use a thread-specific cache (tcache). Unless
identifier.</p></dd><dt><aname="MALLOC_TCACHE_NONE"/><spanclass="term"><codeclass="constant">MALLOCX_TCACHE_NONE</code></span></dt><dd><p>Do not use a thread-specific cache (tcache). Unless
<codeclass="constant">MALLOCX_TCACHE(<emclass="parameter"><code>tc</code></em>)</code> or
<codeclass="constant">MALLOCX_TCACHE_NONE</code> is specified, an
automatically managed tcache will be used under many circumstances.
This macro cannot be used in the same <emclass="parameter"><code>flags</code></em>
</code></span></dt><dd><p>Use the arena specified by the index
<emclass="parameter"><code>a</code></em>. This macro has no effect for regions that
were allocated via an arena other than the one specified. This
macro does not validate that <emclass="parameter"><code>a</code></em> specifies an
arena index in the valid range.</p></dd></dl></div><p>
</p><p>The <codeclass="function">mallocx</code>(<emclass="parameter"><code></code></em>) function allocates at
</p><p>The <codeclass="function">mallocx()</code> function allocates at
least <emclass="parameter"><code>size</code></em> bytes of memory, and returns a pointer
to the base address of the allocation. Behavior is undefined if
<emclass="parameter"><code>size</code></em> is <codeclass="constant">0</code>, or if request size
overflows due to size class and/or alignment constraints.</p><p>The <codeclass="function">rallocx</code>(<emclass="parameter"><code></code></em>) function resizes the
<emclass="parameter"><code>size</code></em> is <codeclass="constant">0</code>.</p><p>The <codeclass="function">rallocx()</code> function resizes the
allocation at <emclass="parameter"><code>ptr</code></em> to be at least
<emclass="parameter"><code>size</code></em> bytes, and returns a pointer to the base
address of the resulting allocation, which may or may not have moved from
its original location. Behavior is undefined if
<emclass="parameter"><code>size</code></em> is <codeclass="constant">0</code>, or if request size
overflows due to size class and/or alignment constraints.</p><p>The <codeclass="function">xallocx</code>(<emclass="parameter"><code></code></em>) function resizes the
<emclass="parameter"><code>size</code></em> is <codeclass="constant">0</code>.</p><p>The <codeclass="function">xallocx()</code> function resizes the
allocation at <emclass="parameter"><code>ptr</code></em> in place to be at least
<emclass="parameter"><code>size</code></em> bytes, and returns the real size of the
allocation. If <emclass="parameter"><code>extra</code></em> is non-zero, an attempt is
...
...
@@ -101,24 +98,24 @@
the extra byte(s) will not by itself result in failure to resize.
Behavior is undefined if <emclass="parameter"><code>size</code></em> is
<codeclass="constant">0</code>, or if <codeclass="code">(<emclass="parameter"><code>size</code></em> + <emclass="parameter"><code>extra</code></em>
><codeclass="constant">SIZE_T_MAX</code>)</code>.</p><p>The <codeclass="function">sallocx</code>(<emclass="parameter"><code></code></em>) function returns the
real size of the allocation at <emclass="parameter"><code>ptr</code></em>.</p><p>The <codeclass="function">dallocx</code>(<emclass="parameter"><code></code></em>) function causes the
><codeclass="constant">SIZE_T_MAX</code>)</code>.</p><p>The <codeclass="function">sallocx()</code> function returns the
real size of the allocation at <emclass="parameter"><code>ptr</code></em>.</p><p>The <codeclass="function">dallocx()</code> function causes the
memory referenced by <emclass="parameter"><code>ptr</code></em> to be made available for
future allocations.</p><p>The <codeclass="function">sdallocx</code>(<emclass="parameter"><code></code></em>) function is an
extension of <codeclass="function">dallocx</code>(<emclass="parameter"><code></code></em>) with a
future allocations.</p><p>The <codeclass="function">sdallocx()</code> function is an
extension of <codeclass="function">dallocx()</code> with a
<emclass="parameter"><code>size</code></em> parameter to allow the caller to pass in the
allocation size as an optimization. The minimum valid input size is the
original requested size of the allocation, and the maximum valid input
size is the corresponding value returned by
<codeclass="function">nallocx</code>(<emclass="parameter"><code></code></em>) or
<codeclass="function">sallocx</code>(<emclass="parameter"><code></code></em>).</p><p>The <codeclass="function">nallocx</code>(<emclass="parameter"><code></code></em>) function allocates no
<codeclass="function">nallocx()</code> or
<codeclass="function">sallocx()</code>.</p><p>The <codeclass="function">nallocx()</code> function allocates no
memory, but it performs the same size computation as the
<codeclass="function">mallocx</code>(<emclass="parameter"><code></code></em>) function, and returns the real
<codeclass="function">mallocx()</code> function, and returns the real
size of the allocation that would result from the equivalent
<codeclass="function">mallocx</code>(<emclass="parameter"><code></code></em>) function call. Behavior is
undefined if <emclass="parameter"><code>size</code></em> is <codeclass="constant">0</code>, or if
request size overflows due to size class and/or alignment
constraints.</p><p>The <codeclass="function">mallctl</code>(<emclass="parameter"><code></code></em>) function provides a
<codeclass="function">mallocx()</code> function call, or
<codeclass="constant">0</code> if the inputs exceed the maximum supported size
class and/or alignment. Behavior is undefined if
<emclass="parameter"><code>size</code></em> is <codeclass="constant">0</code>.</p><p>The <codeclass="function">mallctl()</code> function provides a
general interface for introspecting the memory allocator, as well as
setting modifiable parameters and triggering actions. The
period-separated <emclass="parameter"><code>name</code></em> argument specifies a
...
...
@@ -130,12 +127,12 @@
write a value, pass a pointer to the value via
<emclass="parameter"><code>newp</code></em>, and its length via
}</pre><p>The <codeclass="function">malloc_stats_print</code>(<emclass="parameter"><code></code></em>) function
writes human-readable summary statistics via the
<emclass="parameter"><code>write_cb</code></em> callback function pointer and
<emclass="parameter"><code>cbopaque</code></em> data passed to
<emclass="parameter"><code>write_cb</code></em>, or
<codeclass="function">malloc_message</code>(<emclass="parameter"><code></code></em>) if
<emclass="parameter"><code>write_cb</code></em> is <codeclass="constant">NULL</code>. This
function can be called repeatedly. General information that never
changes during execution can be omitted by specifying "g" as a character
}</pre><p>The <codeclass="function">malloc_stats_print()</code> function writes
summary statistics via the <emclass="parameter"><code>write_cb</code></em> callback
function pointer and <emclass="parameter"><code>cbopaque</code></em> data passed to
<emclass="parameter"><code>write_cb</code></em>, or <codeclass="function">malloc_message()</code>
if <emclass="parameter"><code>write_cb</code></em> is <codeclass="constant">NULL</code>. The
statistics are presented in human-readable form unless <spanclass="quote">“<spanclass="quote">J</span>”</span> is
specified as a character within the <emclass="parameter"><code>opts</code></em> string, in
which case the statistics are presented in <aclass="ulink"href="http://www.json.org/"target="_top">JSON format</a>. This function can be
called repeatedly. General information that never changes during
execution can be omitted by specifying <spanclass="quote">“<spanclass="quote">g</span>”</span> as a character
within the <emclass="parameter"><code>opts</code></em> string. Note that
<codeclass="function">malloc_message</code>(<emclass="parameter"><code></code></em>) uses the
<codeclass="function">mallctl*</code>(<emclass="parameter"><code></code></em>) functions internally, so
inconsistent statistics can be reported if multiple threads use these
functions simultaneously. If <codeclass="option">--enable-stats</code> is
specified during configuration, “m” and “a” can
be specified to omit merged arena and per arena statistics, respectively;
“b”, “l”, and “h” can be specified to
omit per size class statistics for bins, large objects, and huge objects,
respectively. Unrecognized characters are silently ignored. Note that
thread caching may prevent some statistics from being completely up to
date, since extra locking would be required to merge counters that track
thread cache operations.
</p><p>The <codeclass="function">malloc_usable_size</code>(<emclass="parameter"><code></code></em>) function
<codeclass="function">malloc_message()</code> uses the
<codeclass="function">mallctl*()</code> functions internally, so inconsistent
statistics can be reported if multiple threads use these functions
simultaneously. If <codeclass="option">--enable-stats</code> is specified during
configuration, <spanclass="quote">“<spanclass="quote">m</span>”</span> and <spanclass="quote">“<spanclass="quote">a</span>”</span> can be specified to
omit merged arena and per arena statistics, respectively;
<spanclass="quote">“<spanclass="quote">b</span>”</span>, <spanclass="quote">“<spanclass="quote">l</span>”</span>, and <spanclass="quote">“<spanclass="quote">h</span>”</span> can be specified
to omit per size class statistics for bins, large objects, and huge
objects, respectively. Unrecognized characters are silently ignored.
Note that thread caching may prevent some statistics from being completely
up to date, since extra locking would be required to merge counters that
track thread cache operations.</p><p>The <codeclass="function">malloc_usable_size()</code> function
returns the usable size of the allocation pointed to by
<emclass="parameter"><code>ptr</code></em>. The return value may be larger than the size
that was requested during allocation. The
<codeclass="function">malloc_usable_size</code>(<emclass="parameter"><code></code></em>) function is not a
mechanism for in-place <codeclass="function">realloc</code>(<emclass="parameter"><code></code></em>); rather
<codeclass="function">malloc_usable_size()</code> function is not a
mechanism for in-place <codeclass="function">realloc()</code>; rather
it is provided solely as a tool for introspection purposes. Any
discrepancy between the requested allocation size and the size reported
by <codeclass="function">malloc_usable_size</code>(<emclass="parameter"><code></code></em>) should not be
by <codeclass="function">malloc_usable_size()</code> should not be
depended on, since such behavior is entirely implementation-dependent.
</p></div></div><divclass="refsect1"><aname="tuning"></a><h2>TUNING</h2><p>Once, when the first call is made to one of the memory allocation
</p></div></div><divclass="refsect1"><aname="tuning"/><h2>TUNING</h2><p>Once, when the first call is made to one of the memory allocation
routines, the allocator initializes its internals based in part on various
options that can be specified at compile- or run-time.</p><p>The string pointed to by the global variable
<codeclass="varname">malloc_conf</code>, the “name” of the file
referenced by the symbolic link named <codeclass="filename">/etc/malloc.conf</code>, and the value of the
options that can be specified at compile- or run-time.</p><p>The string specified via <codeclass="option">--with-malloc-conf</code>, the
string pointed to by the global variable <codeclass="varname">malloc_conf</code>, the
<spanclass="quote">“<spanclass="quote">name</span>”</span> of the file referenced by the symbolic link named
<codeclass="filename">/etc/malloc.conf</code>, and the value of the
environment variable <codeclass="envar">MALLOC_CONF</code>, will be interpreted, in
that order, from left to right as options. Note that
<codeclass="varname">malloc_conf</code> may be read before
<codeclass="function">main</code>(<emclass="parameter"><code></code></em>) is entered, so the declaration of
<codeclass="function">main()</code> is entered, so the declaration of
<codeclass="varname">malloc_conf</code> should specify an initializer that contains
the final value to be read by jemalloc. <codeclass="varname">malloc_conf</code> is
a compile-time setting, whereas<codeclass="filename">/etc/malloc.conf</code> and <codeclass="envar">MALLOC_CONF</code>
can be safely set any time prior to program invocation.</p><p>An options string is a comma-separated list of option:value pairs.
There is one key corresponding to each <aclass="link"href="#opt.abort">
"<codeclass="mallctl">opt.*</code>"
</a> mallctl (see the <aclass="xref"href="#mallctl_namespace"title="MALLCTL NAMESPACE">MALLCTL NAMESPACE</a> section for options
the final value to be read by jemalloc. <codeclass="option">--with-malloc-conf</code>
and<codeclass="varname">malloc_conf</code> are compile-time mechanisms, whereas
<codeclass="filename">/etc/malloc.conf</code> and
<codeclass="envar">MALLOC_CONF</code> can be safely set any time prior to program
invocation.</p><p>An options string is a comma-separated list of option:value pairs.
There is one key corresponding to each <aclass="link"href="#opt.abort"><quote><codeclass="mallctl">opt.*</code></quote></a> mallctl (see the <aclass="xref"href="#mallctl_namespace"title="MALLCTL NAMESPACE">MALLCTL NAMESPACE</a> section for options
documentation). For example, <codeclass="literal">abort:true,narenas:1</code> sets
the <aclass="link"href="#opt.abort">
"<codeclass="mallctl">opt.abort</code>"
</a> and <aclass="link"href="#opt.narenas">
"<codeclass="mallctl">opt.narenas</code>"
</a> options. Some
the <aclass="link"href="#opt.abort"><quote><codeclass="mallctl">opt.abort</code></quote></a> and <aclass="link"href="#opt.narenas"><quote><codeclass="mallctl">opt.narenas</code></quote></a> options. Some
options have boolean values (true/false), others have integer values (base
8, 10, or 16, depending on prefix), and yet others have raw string
values.</p></div><divclass="refsect1"><aname="implementation_notes"></a><h2>IMPLEMENTATION NOTES</h2><p>Traditionally, allocators have used
values.</p></div><divclass="refsect1"><aname="implementation_notes"/><h2>IMPLEMENTATION NOTES</h2><p>Traditionally, allocators have used
<spanclass="citerefentry"><spanclass="refentrytitle">sbrk</span>(2)</span> to obtain memory, which is
suboptimal for several reasons, including race conditions, increased
fragmentation, and artificial limitations on maximum usable memory. If
...
...
@@ -246,19 +238,18 @@ for (i = 0; i < nbins; i++) {
order to make it possible to completely avoid synchronization for most
allocation requests. Such caching allows very fast allocation in the
common case, but it increases memory usage and fragmentation, since a
bounded number of objects can remain allocated in each thread cache.</p><p>Memory is conceptually broken into equal-sized chunks, where the
chunk size is a power of two that is greater than the page size. Chunks
are always aligned to multiples of the chunk size. This alignment makes it
possible to find metadata for user objects very quickly.</p><p>User objects are broken into three categories according to size:
small, large, and huge. Small and large objects are managed entirely by
arenas; huge objects are additionally aggregated in a single data structure
that is shared by all threads. Huge objects are typically used by
applications infrequently enough that this single data structure is not a
scalability issue.</p><p>Each chunk that is managed by an arena tracks its contents as runs of
bounded number of objects can remain allocated in each thread cache.</p><p>Memory is conceptually broken into equal-sized chunks, where the chunk
size is a power of two that is greater than the page size. Chunks are
always aligned to multiples of the chunk size. This alignment makes it
possible to find metadata for user objects very quickly. User objects are
broken into three categories according to size: small, large, and huge.
Multiple small and large objects can reside within a single chunk, whereas
huge objects each have one or more chunks backing them. Each chunk that
contains small and/or large objects tracks its contents as runs of
contiguous pages (unused, backing a set of small objects, or backing one
large object). The combination of chunk alignment and chunk page maps
makes it possible to determine all metadata regarding small and large
allocations in constant time.</p><p>Small objects are managed in groups by page runs. Each run maintains
large object). The combination of chunk alignment and chunk page maps makes
it possible to determine all metadata regarding small and large allocations
in constant time.</p><p>Small objects are managed in groups by page runs. Each run maintains
a bitmap to track which regions are in use. Allocation requests that are no
more than half the quantum (8 or 16, depending on architecture) are rounded
up to the nearest power of two that is at least <codeclass="code">sizeof(<spanclass="type">double</span>)</code>. All other object size
...
...
@@ -266,22 +257,20 @@ for (i = 0; i < nbins; i++) {
classes for each doubling in size, which limits internal fragmentation to
approximately 20% for all but the smallest size classes. Small size classes
are smaller than four times the page size, large size classes are smaller
than the chunk size (see the <aclass="link"href="#opt.lg_chunk">
"<codeclass="mallctl">opt.lg_chunk</code>"
</a> option), and
huge size classes extend from the chunk size up to one size class less than
the full address space size.</p><p>Allocations are packed tightly together, which can be an issue for
than the chunk size (see the <aclass="link"href="#opt.lg_chunk"><quote><codeclass="mallctl">opt.lg_chunk</code></quote></a> option), and
huge size classes extend from the chunk size up to the largest size class
that does not exceed <codeclass="constant">PTRDIFF_MAX</code>.</p><p>Allocations are packed tightly together, which can be an issue for
multi-threaded applications. If you need to assure that allocations do not
suffer from cacheline sharing, round your allocation requests up to the
nearest multiple of the cacheline size, or specify cacheline alignment when
<codeclass="literal"><i></code> equal to <aclass="link"href="#arenas.narenas">
"<codeclass="mallctl">arenas.narenas</code>"
</a> can be
introspection. In the case of <quote><codeclass="mallctl">stats.arenas.<i>.*</code></quote>,
<codeclass="literal"><i></code> equal to <aclass="link"href="#arenas.narenas"><quote><codeclass="mallctl">arenas.narenas</code></quote></a> can be
used to access the summation of statistics from all arenas. Take special
note of the <aclass="link"href="#epoch">
"<codeclass="mallctl">epoch</code>"
</a> mallctl,
which controls refreshing of cached dynamic statistics.</p><divclass="variablelist"><dlclass="variablelist"><dt><aname="version"></a><spanclass="term">
"<codeclass="mallctl">version</code>"
note of the <aclass="link"href="#epoch"><quote><codeclass="mallctl">epoch</code></quote></a> mallctl,
which controls refreshing of cached dynamic statistics.</p><divclass="variablelist"><dlclass="variablelist"><dt><aname="version"/><spanclass="term">
@@ -538,58 +494,46 @@ for (i = 0; i < nbins; i++) {
thread maintains a FIFO object quarantine that stores up to the
specified number of bytes of memory. The quarantined memory is not
freed until it is released from quarantine, though it is immediately
junk-filled if the <aclass="link"href="#opt.junk">
"<codeclass="mallctl">opt.junk</code>"
</a> option is
junk-filled if the <aclass="link"href="#opt.junk"><quote><codeclass="mallctl">opt.junk</code></quote></a> option is
enabled. This feature is of particular use in combination with <aclass="ulink"href="http://valgrind.org/"target="_top">Valgrind</a>, which can detect attempts
to access quarantined objects. This is intended for debugging and will
impact performance negatively. The default quarantine size is 0 unless
running inside Valgrind, in which case the default is 16
option for information on interval-triggered profile dumping, the <aclass="link"href="#opt.prof_gdump">
"<codeclass="mallctl">opt.prof_gdump</code>"
</a>
memory allocation activity. See the <aclass="link"href="#opt.prof_active"><quote><codeclass="mallctl">opt.prof_active</code></quote></a>
option for on-the-fly activation/deactivation. See the <aclass="link"href="#opt.lg_prof_sample"><quote><codeclass="mallctl">opt.lg_prof_sample</code></quote></a>
option for probabilistic sampling control. See the <aclass="link"href="#opt.prof_accum"><quote><codeclass="mallctl">opt.prof_accum</code></quote></a>
option for control of cumulative sample reporting. See the <aclass="link"href="#opt.lg_prof_interval"><quote><codeclass="mallctl">opt.lg_prof_interval</code></quote></a>
option for information on interval-triggered profile dumping, the <aclass="link"href="#opt.prof_gdump"><quote><codeclass="mallctl">opt.prof_gdump</code></quote></a>
option for information on high-water-triggered profile dumping, and the
option for final profile dumping. Profile output is compatible with
the <spanclass="command"><strong>jeprof</strong></span> command, which is based on the
<spanclass="command"><strong>pprof</strong></span> that is developed as part of the <aclass="ulink"href="http://code.google.com/p/gperftools/"target="_top">gperftools
mallctl. This option is enabled by default.</p></dd><dt><aname="opt.lg_prof_sample"></a><spanclass="term">
"<codeclass="mallctl">opt.lg_prof_sample</code>"
can also be changed during execution via the <aclass="link"href="#prof.thread_active_init"><quote><codeclass="mallctl">prof.thread_active_init</code></quote></a>
mallctl. This option is enabled by default.</p></dd><dt><aname="opt.lg_prof_sample"/><spanclass="term">
where <codeclass="literal"><prefix></code> is controlled by the <aclass="link"href="#opt.prof_prefix">
"<codeclass="mallctl">opt.prof_prefix</code>"
</a>
option. Note that <codeclass="function">atexit</code>(<emclass="parameter"><code></code></em>) may allocate
where <codeclass="literal"><prefix></code> is controlled by the <aclass="link"href="#opt.prof_prefix"><quote><codeclass="mallctl">opt.prof_prefix</code></quote></a>
option. Note that <codeclass="function">atexit()</code> may allocate
memory during application initialization and then deadlock internally
when jemalloc in turn calls <codeclass="function">atexit</code>(<emclass="parameter"><code></code></em>), so
this option is not univerally usable (though the application can
register its own <codeclass="function">atexit</code>(<emclass="parameter"><code></code></em>) function with
when jemalloc in turn calls <codeclass="function">atexit()</code>, so
this option is not universally usable (though the application can
register its own <codeclass="function">atexit()</code> function with
equivalent functionality). This option is disabled by
</span></dt><dd><p>Control whether sampling is currently active for the
calling thread. This is an activation mechanism in addition to <aclass="link"href="#prof.active">
"<codeclass="mallctl">prof.active</code>"
</a>; both must
calling thread. This is an activation mechanism in addition to <aclass="link"href="#prof.active"><quote><codeclass="mallctl">prof.active</code></quote></a>; both must
be active for the calling thread to sample. This flag is enabled by
same considerations apply to this interface as to <aclass="link"href="#thread.tcache.flush"><quote><codeclass="mallctl">thread.tcache.flush</code></quote></a>,
except that the tcache will never be automatically discarded.
</span></dt><dd><p>Trigger decay-based purging of unused dirty pages for
arena <i>, or for all arenas if <i> equals <aclass="link"href="#arenas.narenas"><quote><codeclass="mallctl">arenas.narenas</code></quote></a>.
The proportion of unused dirty pages to be purged depends on the current
time; see <aclass="link"href="#opt.decay_time"><quote><codeclass="mallctl">opt.decay_time</code></quote></a> for
</span></dt><dd><p>Discard all of the arena's extant allocations. This
interface can only be used with arenas created via <aclass="link"href="#arenas.extend"><quote><codeclass="mallctl">arenas.extend</code></quote></a>. None
of the arena's discarded/cached allocations may accessed afterward. As
part of this requirement, all thread caches which were used to
allocate/deallocate in conjunction with the arena must be flushed
beforehand. This interface cannot be used if running inside Valgrind,
nor if the <aclass="link"href="#opt.quarantine">quarantine</a> size is
</span></dt><dd><p>Current per-arena approximate time in seconds from the
creation of a set of unused dirty pages until an equivalent set of
unused dirty pages is purged and/or reused. Each time this interface is
set, all currently unused dirty pages are considered to have fully
decayed, which causes immediate purging of all unused dirty pages unless
the decay time is set to -1 (i.e. purging disabled). See <aclass="link"href="#opt.decay_time"><quote><codeclass="mallctl">opt.decay_time</code></quote></a> for
default chunk allocation function makes the arena's <aclass="link"href="#arena.i.dss"><quote><codeclass="mallctl">arena.<i>.dss</code></quote></a>
<spanclass="type">chunk_dalloc_t</span> type and deallocates a
<emclass="parameter"><code>chunk</code></em> of given <emclass="parameter"><code>size</code></em> with
...
...
@@ -1042,7 +906,7 @@ typedef struct {
success. If the function returns true, this indicates opt-out from
deallocation; the virtual memory mapping associated with the chunk
remains mapped, in the same commit state, and available for future use,
in which case it will be automatically retained for later reuse.</p><divclass="funcsynopsis"><tableborder="0"class="funcprototype-table"summary="Function synopsis"style="cellspacing: 0; cellpadding: 0;"><tr><td><codeclass="funcdef">typedef bool <bclass="fsfunc">(chunk_commit_t)</b>(</code></td><td>void *<varclass="pdparam">chunk</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">size</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">offset</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">length</var>, </td></tr><tr><td></td><td>unsigned <varclass="pdparam">arena_ind</var><code>)</code>;</td></tr></table><divclass="funcprototype-spacer"></div></div><divclass="literallayout"><p></p></div><p>A chunk commit function conforms to the
in which case it will be automatically retained for later reuse.</p><divclass="funcsynopsis"><tableborder="0"class="funcprototype-table"summary="Function synopsis"style="cellspacing: 0; cellpadding: 0;"><tr><td><codeclass="funcdef">typedef bool <bclass="fsfunc">(chunk_commit_t)</b>(</code></td><td>void *<varclass="pdparam">chunk</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">size</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">offset</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">length</var>, </td></tr><tr><td></td><td>unsigned <varclass="pdparam">arena_ind</var><code>)</code>;</td></tr></table><divclass="funcprototype-spacer"></div></div><divclass="literallayout"><p/></div><p>A chunk commit function conforms to the
<spanclass="type">chunk_commit_t</span> type and commits zeroed physical memory to
back pages within a <emclass="parameter"><code>chunk</code></em> of given
<emclass="parameter"><code>size</code></em> at <emclass="parameter"><code>offset</code></em> bytes,
...
...
@@ -1052,7 +916,7 @@ typedef struct {
does not overcommit, or in implicit terms as on a system that
overcommits and satisfies physical memory needs on demand via soft page
faults. If the function returns true, this indicates insufficient
physical memory to satisfy the request.</p><divclass="funcsynopsis"><tableborder="0"class="funcprototype-table"summary="Function synopsis"style="cellspacing: 0; cellpadding: 0;"><tr><td><codeclass="funcdef">typedef bool <bclass="fsfunc">(chunk_decommit_t)</b>(</code></td><td>void *<varclass="pdparam">chunk</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">size</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">offset</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">length</var>, </td></tr><tr><td></td><td>unsigned <varclass="pdparam">arena_ind</var><code>)</code>;</td></tr></table><divclass="funcprototype-spacer"></div></div><divclass="literallayout"><p></p></div><p>A chunk decommit function conforms to the
physical memory to satisfy the request.</p><divclass="funcsynopsis"><tableborder="0"class="funcprototype-table"summary="Function synopsis"style="cellspacing: 0; cellpadding: 0;"><tr><td><codeclass="funcdef">typedef bool <bclass="fsfunc">(chunk_decommit_t)</b>(</code></td><td>void *<varclass="pdparam">chunk</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">size</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">offset</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">length</var>, </td></tr><tr><td></td><td>unsigned <varclass="pdparam">arena_ind</var><code>)</code>;</td></tr></table><divclass="funcprototype-spacer"></div></div><divclass="literallayout"><p/></div><p>A chunk decommit function conforms to the
<spanclass="type">chunk_decommit_t</span> type and decommits any physical memory
that is backing pages within a <emclass="parameter"><code>chunk</code></em> of given
<emclass="parameter"><code>size</code></em> at <emclass="parameter"><code>offset</code></em> bytes,
...
...
@@ -1061,14 +925,14 @@ typedef struct {
case the pages will be committed via the chunk commit function before
being reused. If the function returns true, this indicates opt-out from
decommit; the memory remains committed and available for future use, in
which case it will be automatically retained for later reuse.</p><divclass="funcsynopsis"><tableborder="0"class="funcprototype-table"summary="Function synopsis"style="cellspacing: 0; cellpadding: 0;"><tr><td><codeclass="funcdef">typedef bool <bclass="fsfunc">(chunk_purge_t)</b>(</code></td><td>void *<varclass="pdparam">chunk</var>, </td></tr><tr><td></td><td>size_t<varclass="pdparam">size</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">offset</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">length</var>, </td></tr><tr><td></td><td>unsigned <varclass="pdparam">arena_ind</var><code>)</code>;</td></tr></table><divclass="funcprototype-spacer"></div></div><divclass="literallayout"><p></p></div><p>A chunk purge function conforms to the <spanclass="type">chunk_purge_t</span>
which case it will be automatically retained for later reuse.</p><divclass="funcsynopsis"><tableborder="0"class="funcprototype-table"summary="Function synopsis"style="cellspacing: 0; cellpadding: 0;"><tr><td><codeclass="funcdef">typedef bool <bclass="fsfunc">(chunk_purge_t)</b>(</code></td><td>void *<varclass="pdparam">chunk</var>, </td></tr><tr><td></td><td>size_t<varclass="pdparam">size</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">offset</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">length</var>, </td></tr><tr><td></td><td>unsigned <varclass="pdparam">arena_ind</var><code>)</code>;</td></tr></table><divclass="funcprototype-spacer"></div></div><divclass="literallayout"><p/></div><p>A chunk purge function conforms to the <spanclass="type">chunk_purge_t</span>
type and optionally discards physical pages within the virtual memory
mapping associated with <emclass="parameter"><code>chunk</code></em> of given
<emclass="parameter"><code>size</code></em> at <emclass="parameter"><code>offset</code></em> bytes,
extending for <emclass="parameter"><code>length</code></em> on behalf of arena
<emclass="parameter"><code>arena_ind</code></em>, returning false if pages within the
purged virtual memory range will be zero-filled the next time they are
accessed.</p><divclass="funcsynopsis"><tableborder="0"class="funcprototype-table"summary="Function synopsis"style="cellspacing: 0; cellpadding: 0;"><tr><td><codeclass="funcdef">typedef bool <bclass="fsfunc">(chunk_split_t)</b>(</code></td><td>void *<varclass="pdparam">chunk</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">size</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">size_a</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">size_b</var>, </td></tr><tr><td></td><td>bool <varclass="pdparam">committed</var>, </td></tr><tr><td></td><td>unsigned <varclass="pdparam">arena_ind</var><code>)</code>;</td></tr></table><divclass="funcprototype-spacer"></div></div><divclass="literallayout"><p></p></div><p>A chunk split function conforms to the <spanclass="type">chunk_split_t</span>
accessed.</p><divclass="funcsynopsis"><tableborder="0"class="funcprototype-table"summary="Function synopsis"style="cellspacing: 0; cellpadding: 0;"><tr><td><codeclass="funcdef">typedef bool <bclass="fsfunc">(chunk_split_t)</b>(</code></td><td>void *<varclass="pdparam">chunk</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">size</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">size_a</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">size_b</var>, </td></tr><tr><td></td><td>bool <varclass="pdparam">committed</var>, </td></tr><tr><td></td><td>unsigned <varclass="pdparam">arena_ind</var><code>)</code>;</td></tr></table><divclass="funcprototype-spacer"></div></div><divclass="literallayout"><p/></div><p>A chunk split function conforms to the <spanclass="type">chunk_split_t</span>
type and optionally splits <emclass="parameter"><code>chunk</code></em> of given
<emclass="parameter"><code>size</code></em> into two adjacent chunks, the first of
<emclass="parameter"><code>size_a</code></em> bytes, and the second of
...
...
@@ -1077,7 +941,7 @@ typedef struct {
behalf of arena <emclass="parameter"><code>arena_ind</code></em>, returning false upon
success. If the function returns true, this indicates that the chunk
remains unsplit and therefore should continue to be operated on as a
whole.</p><divclass="funcsynopsis"><tableborder="0"class="funcprototype-table"summary="Function synopsis"style="cellspacing: 0; cellpadding: 0;"><tr><td><codeclass="funcdef">typedef bool <bclass="fsfunc">(chunk_merge_t)</b>(</code></td><td>void *<varclass="pdparam">chunk_a</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">size_a</var>, </td></tr><tr><td></td><td>void *<varclass="pdparam">chunk_b</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">size_b</var>, </td></tr><tr><td></td><td>bool <varclass="pdparam">committed</var>, </td></tr><tr><td></td><td>unsigned <varclass="pdparam">arena_ind</var><code>)</code>;</td></tr></table><divclass="funcprototype-spacer"></div></div><divclass="literallayout"><p></p></div><p>A chunk merge function conforms to the <spanclass="type">chunk_merge_t</span>
whole.</p><divclass="funcsynopsis"><tableborder="0"class="funcprototype-table"summary="Function synopsis"style="cellspacing: 0; cellpadding: 0;"><tr><td><codeclass="funcdef">typedef bool <bclass="fsfunc">(chunk_merge_t)</b>(</code></td><td>void *<varclass="pdparam">chunk_a</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">size_a</var>, </td></tr><tr><td></td><td>void *<varclass="pdparam">chunk_b</var>, </td></tr><tr><td></td><td>size_t <varclass="pdparam">size_b</var>, </td></tr><tr><td></td><td>bool <varclass="pdparam">committed</var>, </td></tr><tr><td></td><td>unsigned <varclass="pdparam">arena_ind</var><code>)</code>;</td></tr></table><divclass="funcprototype-spacer"></div></div><divclass="literallayout"><p/></div><p>A chunk merge function conforms to the <spanclass="type">chunk_merge_t</span>
type and optionally merges adjacent chunks,
<emclass="parameter"><code>chunk_a</code></em> of given <emclass="parameter"><code>size_a</code></em>
and <emclass="parameter"><code>chunk_b</code></em> of given
...
...
@@ -1086,150 +950,106 @@ typedef struct {
behalf of arena <emclass="parameter"><code>arena_ind</code></em>, returning false upon
success. If the function returns true, this indicates that the chunks
remain distinct mappings and therefore should continue to be operated on
during arena creation. See <aclass="link"href="#opt.lg_dirty_mult">
"<codeclass="mallctl">opt.lg_dirty_mult</code>"
</a>
for additional information.</p></dd><dt><aname="arenas.quantum"></a><spanclass="term">
"<codeclass="mallctl">arenas.quantum</code>"
active to dirty pages, used to initialize <aclass="link"href="#arena.i.lg_dirty_mult"><quote><codeclass="mallctl">arena.<i>.lg_dirty_mult</code></quote></a>
during arena creation. See <aclass="link"href="#opt.lg_dirty_mult"><quote><codeclass="mallctl">opt.lg_dirty_mult</code></quote></a>
for additional information.</p></dd><dt><aname="arenas.decay_time"/><spanclass="term">
</span></dt><dd><p>Current default per-arena approximate time in seconds
from the creation of a set of unused dirty pages until an equivalent set
of unused dirty pages is purged and/or reused, used to initialize <aclass="link"href="#arena.i.decay_time"><quote><codeclass="mallctl">arena.<i>.decay_time</code></quote></a>
during arena creation. See <aclass="link"href="#opt.decay_time"><quote><codeclass="mallctl">opt.decay_time</code></quote></a> for
option for additional information.</p></dd><dt><aname="prof.active"></a><spanclass="term">
"<codeclass="mallctl">prof.active</code>"
</span></dt><dd><p>Control the initial setting for <aclass="link"href="#thread.prof.active"><quote><codeclass="mallctl">thread.prof.active</code></quote></a>
in newly created threads. See the <aclass="link"href="#opt.prof_thread_active_init"><quote><codeclass="mallctl">opt.prof_thread_active_init</code></quote></a>
option for additional information.</p></dd><dt><aname="prof.active"/><spanclass="term">
option for additional information, as well as the interrelated <aclass="link"href="#thread.prof.active"><quote><codeclass="mallctl">thread.prof.active</code></quote></a>
where <codeclass="literal"><prefix></code> is controlled by the <aclass="link"href="#opt.prof_prefix"><quote><codeclass="mallctl">opt.prof_prefix</code></quote></a>
</span></dt><dd><p>Get the current sample rate (see <aclass="link"href="#opt.lg_prof_sample"><quote><codeclass="mallctl">opt.lg_prof_sample</code></quote></a>).
allocator data structures, arena chunk headers (see <aclass="link"href="#stats.arenas.i.metadata.mapped"><quote><codeclass="mallctl">stats.arenas.<i>.metadata.mapped</code></quote></a>),
and internal allocations (see <aclass="link"href="#stats.arenas.i.metadata.allocated"><quote><codeclass="mallctl">stats.arenas.<i>.metadata.allocated</code></quote></a>).</p></dd><dt><aname="stats.resident"/><spanclass="term">
page size, and is larger than <aclass="link"href="#stats.active"><quote><codeclass="mallctl">stats.active</code></quote></a>.</p></dd><dt><aname="stats.mapped"/><spanclass="term">
and <aclass="link"href="#stats.resident"><quote><codeclass="mallctl">stats.resident</code></quote></a>.</p></dd><dt><aname="stats.retained"/><spanclass="term">
related to <spanclass="citerefentry"><spanclass="refentrytitle">mmap</span>(2)</span> allocation. See <aclass="link"href="#opt.dss"><quote><codeclass="mallctl">opt.dss</code></quote></a> for details.
</span></dt><dd><p>Number of retained bytes. See <aclass="link"href="#stats.retained"><quote><codeclass="mallctl">stats.retained</code></quote></a> for
because it overlaps with e.g. the <aclass="link"href="#stats.allocated">
"<codeclass="mallctl">stats.allocated</code>"
</a> and
<aclass="link"href="#stats.active">
"<codeclass="mallctl">stats.active</code>"
</a>
profiles. This statistic is reported separately from <aclass="link"href="#stats.metadata"><quote><codeclass="mallctl">stats.metadata</code></quote></a> and
</span></dt><dd><p>Current number of huge allocations for this size class.
</p></dd></dl></div></div><divclass="refsect1"><aname="debugging_malloc_problems"></a><h2>DEBUGGING MALLOC PROBLEMS</h2><p>When debugging, it is a good idea to configure/build jemalloc with
</p></dd></dl></div></div><divclass="refsect1"><aname="heap_profile_format"/><h2>HEAP PROFILE FORMAT</h2><p>Although the heap profiling functionality was originally designed to
be compatible with the
<spanclass="command"><strong>pprof</strong></span> command that is developed as part of the <aclass="ulink"href="http://code.google.com/p/gperftools/"target="_top">gperftools
package</a>, the addition of per thread heap profiling functionality
required a different heap profile format. The <spanclass="command"><strong>jeprof</strong></span>
command is derived from <spanclass="command"><strong>pprof</strong></span>, with enhancements to
support the heap profile format described here.</p><p>In the following hypothetical heap profile, <codeclass="constant">[...]</code>
indicates elision for the sake of compactness. </p><preclass="programlisting">
</proc/<pid>/maps></pre></div><divclass="refsect1"><aname="debugging_malloc_problems"/><h2>DEBUGGING MALLOC PROBLEMS</h2><p>When debugging, it is a good idea to configure/build jemalloc with
the <codeclass="option">--enable-debug</code> and <codeclass="option">--enable-fill</code>
options, and recompile the program with suitable options and symbols for
debugger support. When so configured, jemalloc incorporates a wide variety
of run-time assertions that catch application errors such as double-free,
write-after-free, etc.</p><p>Programs often accidentally depend on “uninitialized”
write-after-free, etc.</p><p>Programs often accidentally depend on <spanclass="quote">“<spanclass="quote">uninitialized</span>”</span>
memory actually being filled with zero bytes. Junk filling
(see the <aclass="link"href="#opt.junk">
"<codeclass="mallctl">opt.junk</code>"
</a>
(see the <aclass="link"href="#opt.junk"><quote><codeclass="mallctl">opt.junk</code></quote></a>
option) tends to expose such bugs in the form of obviously incorrect
results and/or coredumps. Conversely, zero
filling (see the <aclass="link"href="#opt.zero">
"<codeclass="mallctl">opt.zero</code>"
</a> option) eliminates
filling (see the <aclass="link"href="#opt.zero"><quote><codeclass="mallctl">opt.zero</code></quote></a> option) eliminates
the symptoms of such bugs. Between these two options, it is usually
possible to quickly detect, diagnose, and eliminate such bugs.</p><p>This implementation does not provide much detail about the problems
it detects, because the performance impact for storing such information
would be prohibitive. However, jemalloc does integrate with the most
excellent <aclass="ulink"href="http://valgrind.org/"target="_top">Valgrind</a> tool if the
<codeclass="option">--enable-valgrind</code> configuration option is enabled.</p></div><divclass="refsect1"><aname="diagnostic_messages"></a><h2>DIAGNOSTIC MESSAGES</h2><p>If any of the memory allocation/deallocation functions detect an
<codeclass="option">--enable-valgrind</code> configuration option is enabled.</p></div><divclass="refsect1"><aname="diagnostic_messages"/><h2>DIAGNOSTIC MESSAGES</h2><p>If any of the memory allocation/deallocation functions detect an
error or warning condition, a message will be printed to file descriptor
<codeclass="constant">STDERR_FILENO</code>. Errors will result in the process
dumping core. If the <aclass="link"href="#opt.abort">
"<codeclass="mallctl">opt.abort</code>"
</a> option is set, most
dumping core. If the <aclass="link"href="#opt.abort"><quote><codeclass="mallctl">opt.abort</code></quote></a> option is set, most
warnings are treated as errors.</p><p>The <codeclass="varname">malloc_message</code> variable allows the programmer
to override the function which emits the text strings forming the errors
and warnings if for some reason the <codeclass="constant">STDERR_FILENO</code> file
descriptor is not suitable for this.
<codeclass="function">malloc_message</code>(<emclass="parameter"><code></code></em>) takes the
<codeclass="function">malloc_message()</code> takes the
<emclass="parameter"><code>cbopaque</code></em> pointer argument that is
<codeclass="constant">NULL</code> unless overridden by the arguments in a call to
<codeclass="function">malloc_stats_print</code>(<emclass="parameter"><code></code></em>), followed by a string
<codeclass="function">malloc_stats_print()</code>, followed by a string
pointer. Please note that doing anything which tries to allocate memory in
this function is likely to result in a crash or deadlock.</p><p>All messages are prefixed by
“<codeclass="computeroutput"><jemalloc>: </code>”.</p></div><divclass="refsect1"><aname="return_values"></a><h2>RETURN VALUES</h2><divclass="refsect2"><aname="idp46949776"></a><h3>Standard API</h3><p>The <codeclass="function">malloc</code>(<emclass="parameter"><code></code></em>) and
<codeclass="function">calloc</code>(<emclass="parameter"><code></code></em>) functions return a pointer to the
<spanclass="quote">“<spanclass="quote"><codeclass="computeroutput"><jemalloc>: </code></span>”</span>.</p></div><divclass="refsect1"><aname="return_values"/><h2>RETURN VALUES</h2><divclass="refsect2"><aname="idm45291897515152"/><h3>Standard API</h3><p>The <codeclass="function">malloc()</code> and
<codeclass="function">calloc()</code> functions return a pointer to the
allocated memory if successful; otherwise a <codeclass="constant">NULL</code>
pointer is returned and <codeclass="varname">errno</code> is set to
<spanclass="errorname">ENOMEM</span>.</p><p>The <codeclass="function">posix_memalign</code>(<emclass="parameter"><code></code></em>) function
<spanclass="errorname">ENOMEM</span>.</p><p>The <codeclass="function">posix_memalign()</code> function
returns the value 0 if successful; otherwise it returns an error value.
The <codeclass="function">posix_memalign</code>(<emclass="parameter"><code></code></em>) function will fail
The <codeclass="function">posix_memalign()</code> function will fail
if:
</p><divclass="variablelist"><dlclass="variablelist"><dt><spanclass="term"><spanclass="errorname">EINVAL</span></span></dt><dd><p>The <emclass="parameter"><code>alignment</code></em> parameter is
</p><p>The <codeclass="function">aligned_alloc</code>(<emclass="parameter"><code></code></em>) function returns
</p><p>The <codeclass="function">aligned_alloc()</code> function returns
a pointer to the allocated memory if successful; otherwise a
<codeclass="constant">NULL</code> pointer is returned and
<codeclass="varname">errno</code> is set. The
<codeclass="function">aligned_alloc</code>(<emclass="parameter"><code></code></em>) function will fail if:
<codeclass="function">aligned_alloc()</code> function will fail if:
</p><divclass="variablelist"><dlclass="variablelist"><dt><spanclass="term"><spanclass="errorname">EINVAL</span></span></dt><dd><p>The <emclass="parameter"><code>alignment</code></em> parameter is
</p><p>The <codeclass="function">realloc</code>(<emclass="parameter"><code></code></em>) function returns a
</p><p>The <codeclass="function">realloc()</code> function returns a
pointer, possibly identical to <emclass="parameter"><code>ptr</code></em>, to the
allocated memory if successful; otherwise a <codeclass="constant">NULL</code>
pointer is returned, and <codeclass="varname">errno</code> is set to
<spanclass="errorname">ENOMEM</span> if the error was the result of an
allocation failure. The <codeclass="function">realloc</code>(<emclass="parameter"><code></code></em>)
allocation failure. The <codeclass="function">realloc()</code>
function always leaves the original buffer intact when an error occurs.
</p><p>The <codeclass="function">free</code>(<emclass="parameter"><code></code></em>) function returns no
value.</p></div><divclass="refsect2"><aname="idp46974576"></a><h3>Non-standard API</h3><p>The <codeclass="function">mallocx</code>(<emclass="parameter"><code></code></em>) and
<codeclass="function">rallocx</code>(<emclass="parameter"><code></code></em>) functions return a pointer to
</p><p>The <codeclass="function">free()</code> function returns no
value.</p></div><divclass="refsect2"><aname="idm45291897493664"/><h3>Non-standard API</h3><p>The <codeclass="function">mallocx()</code> and
<codeclass="function">rallocx()</code> functions return a pointer to
the allocated memory if successful; otherwise a <codeclass="constant">NULL</code>
pointer is returned to indicate insufficient contiguous memory was
available to service the allocation request. </p><p>The <codeclass="function">xallocx</code>(<emclass="parameter"><code></code></em>) function returns the
available to service the allocation request. </p><p>The <codeclass="function">xallocx()</code> function returns the
real size of the resulting resized allocation pointed to by
<emclass="parameter"><code>ptr</code></em>, which is a value less than
<emclass="parameter"><code>size</code></em> if the allocation could not be adequately
grown in place. </p><p>The <codeclass="function">sallocx</code>(<emclass="parameter"><code></code></em>) function returns the
grown in place. </p><p>The <codeclass="function">sallocx()</code> function returns the
real size of the allocation pointed to by <emclass="parameter"><code>ptr</code></em>.
</p><p>The <codeclass="function">nallocx</code>(<emclass="parameter"><code></code></em>) returns the real size
</p><p>The <codeclass="function">nallocx()</code> returns the real size
that would result from a successful equivalent
<codeclass="function">mallocx</code>(<emclass="parameter"><code></code></em>) function call, or zero if
insufficient memory is available to perform the size computation. </p><p>The <codeclass="function">mallctl</code>(<emclass="parameter"><code></code></em>),
<codeclass="function">mallctlnametomib</code>(<emclass="parameter"><code></code></em>), and
<codeclass="function">mallctlbymib</code>(<emclass="parameter"><code></code></em>) functions return 0 on
<codeclass="function">mallocx()</code> function call, or zero if
insufficient memory is available to perform the size computation. </p><p>The <codeclass="function">mallctl()</code>,
<codeclass="function">mallctlnametomib()</code>, and
<codeclass="function">mallctlbymib()</code> functions return 0 on
success; otherwise they return an error value. The functions will fail
if:
</p><divclass="variablelist"><dlclass="variablelist"><dt><spanclass="term"><spanclass="errorname">EINVAL</span></span></dt><dd><p><emclass="parameter"><code>newp</code></em> is not
...
...
@@ -1798,28 +1535,28 @@ typedef struct {
value.</p></dd><dt><spanclass="term"><spanclass="errorname">EPERM</span></span></dt><dd><p>Attempt to read or write void value, or attempt to
occurred.</p></dd><dt><spanclass="term"><spanclass="errorname">EFAULT</span></span></dt><dd><p>An interface with side effects failed in some way
not directly related to <codeclass="function">mallctl*</code>(<emclass="parameter"><code></code></em>)
not directly related to <codeclass="function">mallctl*()</code>
read/write processing.</p></dd></dl></div><p>
</p><p>The <codeclass="function">malloc_usable_size</code>(<emclass="parameter"><code></code></em>) function
</p><p>The <codeclass="function">malloc_usable_size()</code> function
returns the usable size of the allocation pointed to by
<emclass="parameter"><code>ptr</code></em>. </p></div></div><divclass="refsect1"><aname="environment"></a><h2>ENVIRONMENT</h2><p>The following environment variable affects the execution of the
<emclass="parameter"><code>ptr</code></em>. </p></div></div><divclass="refsect1"><aname="environment"/><h2>ENVIRONMENT</h2><p>The following environment variable affects the execution of the
allocation functions:
</p><divclass="variablelist"><dlclass="variablelist"><dt><spanclass="term"><codeclass="envar">MALLOC_CONF</code></span></dt><dd><p>If the environment variable
<codeclass="envar">MALLOC_CONF</code> is set, the characters it contains
will be interpreted as options.</p></dd></dl></div><p>
</p></div><divclass="refsect1"><aname="examples"></a><h2>EXAMPLES</h2><p>To dump core whenever a problem occurs:
</p></div><divclass="refsect1"><aname="examples"/><h2>EXAMPLES</h2><p>To dump core whenever a problem occurs: