1
Copyright (c) 1988, 1989 Hans-J. Boehm, Alan J. Demers
2
Copyright (c) 1991-1996 by Xerox Corporation. All rights reserved.
3
Copyright (c) 1996-1999 by Silicon Graphics. All rights reserved.
4
Copyright (c) 1999-2004 Hewlett-Packard Development Company, L.P.
6
The file linux_threads.c is also
7
Copyright (c) 1998 by Fergus Henderson. All rights reserved.
9
The files Makefile.am, and configure.in are
10
Copyright (c) 2001 by Red Hat Inc. All rights reserved.
12
Several files supporting GNU-style builds are copyrighted by the Free
13
Software Foundation, and carry a different license from that given
16
THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY EXPRESSED
17
OR IMPLIED. ANY USE IS AT YOUR OWN RISK.
19
Permission is hereby granted to use or copy this program
20
for any purpose, provided the above notices are retained on all copies.
21
Permission to modify the code and to distribute modified code is granted,
22
provided the above notices are retained, and a notice that the code was
23
modified is included with the above copyright notice.
25
A few of the files needed to use the GNU-style build procedure come with
26
slightly different licenses, though they are all similar in spirit. A few
27
are GPL'ed, but with an exception that should cover all uses in the
28
collector. (If you are concerned about such things, I recommend you look
29
at the notice in config.guess or ltmain.sh.)
31
This is version 6.5 of a conservative garbage collector for C and C++.
33
You might find a more recent version of this at
35
http://www.hpl.hp.com/personal/Hans_Boehm/gc
39
This is intended to be a general purpose, garbage collecting storage
40
allocator. The algorithms used are described in:
42
Boehm, H., and M. Weiser, "Garbage Collection in an Uncooperative Environment",
43
Software Practice & Experience, September 1988, pp. 807-820.
45
Boehm, H., A. Demers, and S. Shenker, "Mostly Parallel Garbage Collection",
46
Proceedings of the ACM SIGPLAN '91 Conference on Programming Language Design
47
and Implementation, SIGPLAN Notices 26, 6 (June 1991), pp. 157-164.
49
Boehm, H., "Space Efficient Conservative Garbage Collection", Proceedings
50
of the ACM SIGPLAN '91 Conference on Programming Language Design and
51
Implementation, SIGPLAN Notices 28, 6 (June 1993), pp. 197-206.
53
Boehm H., "Reducing Garbage Collector Cache Misses", Proceedings of the
54
2000 International Symposium on Memory Management.
56
Possible interactions between the collector and optimizing compilers are
59
Boehm, H., and D. Chase, "A Proposal for GC-safe C Compilation",
60
The Journal of C Language Translation 4, 2 (December 1992).
64
Boehm H., "Simple GC-safe Compilation", Proceedings
65
of the ACM SIGPLAN '96 Conference on Programming Language Design and
68
(Some of these are also available from
69
http://www.hpl.hp.com/personal/Hans_Boehm/papers/, among other places.)
71
Unlike the collector described in the second reference, this collector
72
operates either with the mutator stopped during the entire collection
73
(default) or incrementally during allocations. (The latter is supported
74
on only a few machines.) On the most common platforms, it can be built
75
with or without thread support. On a few platforms, it can take advantage
76
of a multiprocessor to speed up garbage collection.
78
Many of the ideas underlying the collector have previously been explored
79
by others. Notably, some of the run-time systems developed at Xerox PARC
80
in the early 1980s conservatively scanned thread stacks to locate possible
81
pointers (cf. Paul Rovner, "On Adding Garbage Collection and Runtime Types
82
to a Strongly-Typed Statically Checked, Concurrent Language" Xerox PARC
83
CSL 84-7). Doug McIlroy wrote a simpler fully conservative collector that
84
was part of version 8 UNIX (tm), but appears to not have received
87
Rudimentary tools for use of the collector as a leak detector are included
88
(see http://www.hpl.hp.com/personal/Hans_Boehm/gc/leak.html),
89
as is a fairly sophisticated string package "cord" that makes use of the
90
collector. (See doc/README.cords and H.-J. Boehm, R. Atkinson, and M. Plass,
91
"Ropes: An Alternative to Strings", Software Practice and Experience 25, 12
92
(December 1995), pp. 1315-1330. This is very similar to the "rope" package
93
in Xerox Cedar, or the "rope" package in the SGI STL or the g++ distribution.)
95
Further collector documantation can be found at
97
http://www.hpl.hp.com/personal/Hans_Boehm/gc
102
This is a garbage collecting storage allocator that is intended to be
103
used as a plug-in replacement for C's malloc.
105
Since the collector does not require pointers to be tagged, it does not
106
attempt to ensure that all inaccessible storage is reclaimed. However,
107
in our experience, it is typically more successful at reclaiming unused
108
memory than most C programs using explicit deallocation. Unlike manually
109
introduced leaks, the amount of unreclaimed memory typically stays
112
In the following, an "object" is defined to be a region of memory allocated
113
by the routines described below.
115
Any objects not intended to be collected must be pointed to either
116
from other such accessible objects, or from the registers,
117
stack, data, or statically allocated bss segments. Pointers from
118
the stack or registers may point to anywhere inside an object.
119
The same is true for heap pointers if the collector is compiled with
120
ALL_INTERIOR_POINTERS defined, as is now the default.
122
Compiling without ALL_INTERIOR_POINTERS may reduce accidental retention
123
of garbage objects, by requiring pointers from the heap to to the beginning
124
of an object. But this no longer appears to be a significant
125
issue for most programs.
127
There are a number of routines which modify the pointer recognition
128
algorithm. GC_register_displacement allows certain interior pointers
129
to be recognized even if ALL_INTERIOR_POINTERS is nor defined.
130
GC_malloc_ignore_off_page allows some pointers into the middle of large objects
131
to be disregarded, greatly reducing the probablility of accidental
132
retention of large objects. For most purposes it seems best to compile
133
with ALL_INTERIOR_POINTERS and to use GC_malloc_ignore_off_page if
134
you get collector warnings from allocations of very large objects.
135
See README.debugging for details.
137
WARNING: pointers inside memory allocated by the standard "malloc" are not
138
seen by the garbage collector. Thus objects pointed to only from such a
139
region may be prematurely deallocated. It is thus suggested that the
140
standard "malloc" be used only for memory regions, such as I/O buffers, that
141
are guaranteed not to contain pointers to garbage collectable memory.
142
Pointers in C language automatic, static, or register variables,
143
are correctly recognized. (Note that GC_malloc_uncollectable has semantics
144
similar to standard malloc, but allocates objects that are traced by the
147
WARNING: the collector does not always know how to find pointers in data
148
areas that are associated with dynamic libraries. This is easy to
149
remedy IF you know how to find those data areas on your operating
150
system (see GC_add_roots). Code for doing this under SunOS, IRIX 5.X and 6.X,
151
HP/UX, Alpha OSF/1, Linux, and win32 is included and used by default. (See
152
README.win32 for win32 details.) On other systems pointers from dynamic
153
library data areas may not be considered by the collector.
154
If you're writing a program that depends on the collector scanning
155
dynamic library data areas, it may be a good idea to include at least
156
one call to GC_is_visible() to ensure that those areas are visible
159
Note that the garbage collector does not need to be informed of shared
160
read-only data. However if the shared library mechanism can introduce
161
discontiguous data areas that may contain pointers, then the collector does
164
Signal processing for most signals may be deferred during collection,
165
and during uninterruptible parts of the allocation process.
166
Like standard ANSI C mallocs, by default it is unsafe to invoke
167
malloc (and other GC routines) from a signal handler while another
168
malloc call may be in progress. Removing -DNO_SIGNALS from Makefile
169
attempts to remedy that. But that may not be reliable with a compiler that
170
substantially reorders memory operations inside GC_malloc.
172
The allocator/collector can also be configured for thread-safe operation.
173
(Full signal safety can also be achieved, but only at the cost of two system
174
calls per malloc, which is usually unacceptable.)
175
WARNING: the collector does not guarantee to scan thread-local storage
176
(e.g. of the kind accessed with pthread_getspecific()). The collector
177
does scan thread stacks, though, so generally the best solution is to
178
ensure that any pointers stored in thread-local storage are also
179
stored on the thread's stack for the duration of their lifetime.
180
(This is arguably a longstanding bug, but it hasn't been fixed yet.)
182
INSTALLATION AND PORTABILITY
184
As distributed, the macro SILENT is defined in Makefile.
185
In the event of problems, this can be removed to obtain a moderate
186
amount of descriptive output for each collection.
187
(The given statistics exhibit a few peculiarities.
188
Things don't appear to add up for a variety of reasons, most notably
189
fragmentation losses. These are probably much more significant for the
190
contrived program "test.c" than for your application.)
192
Note that typing "make test" will automatically build the collector
193
and then run setjmp_test and gctest. Setjmp_test will give you information
194
about configuring the collector, which is useful primarily if you have
195
a machine that's not already supported. Gctest is a somewhat superficial
196
test of collector functionality. Failure is indicated by a core dump or
197
a message to the effect that the collector is broken. Gctest takes about
198
35 seconds to run on a SPARCstation 2. It may use up to 8 MB of memory. (The
199
multi-threaded version will use more. 64-bit versions may use more.)
200
"Make test" will also, as its last step, attempt to build and test the
201
"cord" string library. This will fail without an ANSI C compiler, but
202
the garbage collector itself should still be usable.
204
The Makefile will generate a library gc.a which you should link against.
205
Typing "make cords" will add the cord library to gc.a.
206
Note that this requires an ANSI C compiler.
208
It is suggested that if you need to replace a piece of the collector
209
(e.g. GC_mark_rts.c) you simply list your version ahead of gc.a on the
210
ld command line, rather than replacing the one in gc.a. (This will
211
generate numerous warnings under some versions of AIX, but it still
214
All include files that need to be used by clients will be put in the
215
include subdirectory. (Normally this is just gc.h. "Make cords" adds
216
"cord.h" and "ec.h".)
218
The collector currently is designed to run essentially unmodified on
219
machines that use a flat 32-bit or 64-bit address space.
220
That includes the vast majority of Workstations and X86 (X >= 3) PCs.
221
(The list here was deleted because it was getting too long and constantly
223
It does NOT run under plain 16-bit DOS or Windows 3.X. There are however
224
various packages (e.g. win32s, djgpp) that allow flat 32-bit address
225
applications to run under those systemsif the have at least an 80386 processor,
226
and several of those are compatible with the collector.
228
In a few cases (Amiga, OS/2, Win32, MacOS) a separate makefile
229
or equivalent is supplied. Many of these have separate README.system
232
Dynamic libraries are completely supported only under SunOS/Solaris,
233
(and even that support is not functional on the last Sun 3 release),
234
Linux, FreeBSD, NetBSD, IRIX 5&6, HP/UX, Win32 (not Win32S) and OSF/1
235
on DEC AXP machines plus perhaps a few others listed near the top
236
of dyn_load.c. On other machines we recommend that you do one of
239
1) Add dynamic library support (and send us the code).
240
2) Use static versions of the libraries.
241
3) Arrange for dynamic libraries to use the standard malloc.
242
This is still dangerous if the library stores a pointer to a
243
garbage collected object. But nearly all standard interfaces
244
prohibit this, because they deal correctly with pointers
245
to stack allocated objects. (Strtok is an exception. Don't
248
In all cases we assume that pointer alignment is consistent with that
249
enforced by the standard C compilers. If you use a nonstandard compiler
250
you may have to adjust the alignment parameters defined in gc_priv.h.
251
Note that this may also be an issue with packed records/structs, if those
252
enforce less alignment for pointers.
254
A port to a machine that is not byte addressed, or does not use 32 bit
255
or 64 bit addresses will require a major effort. A port to plain MSDOS
258
For machines not already mentioned, or for nonstandard compilers, the
259
following are likely to require change:
261
1. The parameters in gcconfig.h.
262
The parameters that will usually require adjustment are
263
STACKBOTTOM, ALIGNMENT and DATASTART. Setjmp_test
264
prints its guesses of the first two.
265
DATASTART should be an expression for computing the
266
address of the beginning of the data segment. This can often be
267
&etext. But some memory management units require that there be
268
some unmapped space between the text and the data segment. Thus
269
it may be more complicated. On UNIX systems, this is rarely
270
documented. But the adb "$m" command may be helpful. (Note
271
that DATASTART will usually be a function of &etext. Thus a
272
single experiment is usually insufficient.)
273
STACKBOTTOM is used to initialize GC_stackbottom, which
274
should be a sufficient approximation to the coldest stack address.
275
On some machines, it is difficult to obtain such a value that is
276
valid across a variety of MMUs, OS releases, etc. A number of
277
alternatives exist for using the collector in spite of this. See the
278
discussion in gcconfig.h immediately preceding the various
279
definitions of STACKBOTTOM.
282
The most important routine here is one to mark from registers.
283
The distributed file includes a generic hack (based on setjmp) that
284
happens to work on many machines, and may work on yours. Try
285
compiling and running setjmp_t.c to see whether it has a chance of
286
working. (This is not correct C, so don't blame your compiler if it
287
doesn't work. Based on limited experience, register window machines
288
are likely to cause trouble. If your version of setjmp claims that
289
all accessible variables, including registers, have the value they
290
had at the time of the longjmp, it also will not work. Vanilla 4.2 BSD
291
on Vaxen makes such a claim. SunOS does not.)
292
If your compiler does not allow in-line assembly code, or if you prefer
293
not to use such a facility, mach_dep.c may be replaced by a .s file
294
(as we did for the MIPS machine and the PC/RT).
295
At this point enough architectures are supported by mach_dep.c
296
that you will rarely need to do more than adjust for assembler
299
3. os_dep.c (and gc_priv.h).
300
Several kinds of operating system dependent routines reside here.
301
Many are optional. Several are invoked only through corresponding
302
macros in gc_priv.h, which may also be redefined as appropriate.
303
The routine GC_register_data_segments is crucial. It registers static
304
data areas that must be traversed by the collector. (User calls to
305
GC_add_roots may sometimes be used for similar effect.)
306
Routines to obtain memory from the OS also reside here.
307
Alternatively this can be done entirely by the macro GET_MEM
308
defined in gc_priv.h. Routines to disable and reenable signals
309
also reside here if they are need by the macros DISABLE_SIGNALS
310
and ENABLE_SIGNALS defined in gc_priv.h.
311
In a multithreaded environment, the macros LOCK and UNLOCK
312
in gc_priv.h will need to be suitably redefined.
313
The incremental collector requires page dirty information, which
314
is acquired through routines defined in os_dep.c. Unless directed
315
otherwise by gcconfig.h, these are implemented as stubs that simply
316
treat all pages as dirty. (This of course makes the incremental
317
collector much less useful.)
320
This provides a routine that allows the collector to scan data
321
segments associated with dynamic libraries. Often it is not
322
necessary to provide this routine unless user-written dynamic
325
For a different version of UN*X or different machines using the
326
Motorola 68000, Vax, SPARC, 80386, NS 32000, PC/RT, or MIPS architecture,
327
it should frequently suffice to change definitions in gcconfig.h.
330
THE C INTERFACE TO THE ALLOCATOR
332
The following routines are intended to be directly called by the user.
333
Note that usually only GC_malloc is necessary. GC_clear_roots and GC_add_roots
334
calls may be required if the collector has to trace from nonstandard places
335
(e.g. from dynamic library data areas on a machine on which the
336
collector doesn't already understand them.) On some machines, it may
337
be desirable to set GC_stacktop to a good approximation of the stack base.
338
(This enhances code portability on HP PA machines, since there is no
339
good way for the collector to compute this value.) Client code may include
340
"gc.h", which defines all of the following, plus many others.
343
- allocate an object of size nbytes. Unlike malloc, the object is
344
cleared before being returned to the user. Gc_malloc will
345
invoke the garbage collector when it determines this to be appropriate.
346
GC_malloc may return 0 if it is unable to acquire sufficient
347
space from the operating system. This is the most probable
348
consequence of running out of space. Other possible consequences
349
are that a function call will fail due to lack of stack space,
350
or that the collector will fail in other ways because it cannot
351
maintain its internal data structures, or that a crucial system
352
process will fail and take down the machine. Most of these
353
possibilities are independent of the malloc implementation.
355
2) GC_malloc_atomic(nbytes)
356
- allocate an object of size nbytes that is guaranteed not to contain any
357
pointers. The returned object is not guaranteed to be cleared.
358
(Can always be replaced by GC_malloc, but results in faster collection
359
times. The collector will probably run faster if large character
360
arrays, etc. are allocated with GC_malloc_atomic than if they are
361
statically allocated.)
363
3) GC_realloc(object, new_size)
364
- change the size of object to be new_size. Returns a pointer to the
365
new object, which may, or may not, be the same as the pointer to
366
the old object. The new object is taken to be atomic iff the old one
367
was. If the new object is composite and larger than the original object,
368
then the newly added bytes are cleared (we hope). This is very likely
369
to allocate a new object, unless MERGE_SIZES is defined in gc_priv.h.
370
Even then, it is likely to recycle the old object only if the object
371
is grown in small additive increments (which, we claim, is generally bad
375
- explicitly deallocate an object returned by GC_malloc or
376
GC_malloc_atomic. Not necessary, but can be used to minimize
377
collections if performance is critical. Probably a performance
378
loss for very small objects (<= 8 bytes).
380
5) GC_expand_hp(bytes)
381
- Explicitly increase the heap size. (This is normally done automatically
382
if a garbage collection failed to GC_reclaim enough memory. Explicit
383
calls to GC_expand_hp may prevent unnecessarily frequent collections at
386
6) GC_malloc_ignore_off_page(bytes)
387
- identical to GC_malloc, but the client promises to keep a pointer to
388
the somewhere within the first 256 bytes of the object while it is
389
live. (This pointer should nortmally be declared volatile to prevent
390
interference from compiler optimizations.) This is the recommended
391
way to allocate anything that is likely to be larger than 100Kbytes
392
or so. (GC_malloc may result in failure to reclaim such objects.)
394
7) GC_set_warn_proc(proc)
395
- Can be used to redirect warnings from the collector. Such warnings
396
should be rare, and should not be ignored during code development.
398
8) GC_enable_incremental()
399
- Enables generational and incremental collection. Useful for large
400
heaps on machines that provide access to page dirty information.
401
Some dirty bit implementations may interfere with debugging
402
(by catching address faults) and place restrictions on heap arguments
403
to system calls (since write faults inside a system call may not be
406
9) Several routines to allow for registration of finalization code.
407
User supplied finalization code may be invoked when an object becomes
408
unreachable. To call (*f)(obj, x) when obj becomes inaccessible, use
409
GC_register_finalizer(obj, f, x, 0, 0);
410
For more sophisticated uses, and for finalization ordering issues,
413
The global variable GC_free_space_divisor may be adjusted up from its
414
default value of 4 to use less space and more collection time, or down for
415
the opposite effect. Setting it to 1 or 0 will effectively disable collections
416
and cause all allocations to simply grow the heap.
418
The variable GC_non_gc_bytes, which is normally 0, may be changed to reflect
419
the amount of memory allocated by the above routines that should not be
420
considered as a candidate for collection. Careless use may, of course, result
421
in excessive memory consumption.
423
Some additional tuning is possible through the parameters defined
424
near the top of gc_priv.h.
426
If only GC_malloc is intended to be used, it might be appropriate to define:
428
#define malloc(n) GC_malloc(n)
429
#define calloc(m,n) GC_malloc((m)*(n))
431
For small pieces of VERY allocation intensive code, gc_inl.h
432
includes some allocation macros that may be used in place of GC_malloc
435
All externally visible names in the garbage collector start with "GC_".
436
To avoid name conflicts, client code should avoid this prefix, except when
437
accessing garbage collector routines or variables.
439
There are provisions for allocation with explicit type information.
440
This is rarely necessary. Details can be found in gc_typed.h.
442
THE C++ INTERFACE TO THE ALLOCATOR:
444
The Ellis-Hull C++ interface to the collector is included in
445
the collector distribution. If you intend to use this, type
446
"make c++" after the initial build of the collector is complete.
447
See gc_cpp.h for the definition of the interface. This interface
448
tries to approximate the Ellis-Detlefs C++ garbage collection
449
proposal without compiler changes.
452
1. Arrays allocated without new placement syntax are
453
allocated as uncollectable objects. They are traced by the
454
collector, but will not be reclaimed.
456
2. Failure to use "make c++" in combination with (1) will
457
result in arrays allocated using the default new operator.
458
This is likely to result in disaster without linker warnings.
460
3. If your compiler supports an overloaded new[] operator,
461
then gc_cpp.cc and gc_cpp.h should be suitably modified.
463
4. Many current C++ compilers have deficiencies that
464
break some of the functionality. See the comments in gc_cpp.h
465
for suggested workarounds.
467
USE AS LEAK DETECTOR:
469
The collector may be used to track down leaks in C programs that are
470
intended to run with malloc/free (e.g. code with extreme real-time or
471
portability constraints). To do so define FIND_LEAK in Makefile
472
This will cause the collector to invoke the report_leak
473
routine defined near the top of reclaim.c whenever an inaccessible
474
object is found that has not been explicitly freed. Such objects will
475
also be automatically reclaimed.
476
Productive use of this facility normally involves redefining report_leak
477
to do something more intelligent. This typically requires annotating
478
objects with additional information (e.g. creation time stack trace) that
479
identifies their origin. Such code is typically not very portable, and is
480
not included here, except on SPARC machines.
481
If all objects are allocated with GC_DEBUG_MALLOC (see next section),
482
then the default version of report_leak will report the source file
483
and line number at which the leaked object was allocated. This may
484
sometimes be sufficient. (On SPARC/SUNOS4 machines, it will also report
485
a cryptic stack trace. This can often be turned into a sympolic stack
486
trace by invoking program "foo" with "callprocs foo". Callprocs is
487
a short shell script that invokes adb to expand program counter values
488
to symbolic addresses. It was largely supplied by Scott Schwartz.)
489
Note that the debugging facilities described in the next section can
490
sometimes be slightly LESS effective in leak finding mode, since in
491
leak finding mode, GC_debug_free actually results in reuse of the object.
492
(Otherwise the object is simply marked invalid.) Also note that the test
493
program is not designed to run meaningfully in FIND_LEAK mode.
494
Use "make gc.a" to build the collector.
496
DEBUGGING FACILITIES:
498
The routines GC_debug_malloc, GC_debug_malloc_atomic, GC_debug_realloc,
499
and GC_debug_free provide an alternate interface to the collector, which
500
provides some help with memory overwrite errors, and the like.
501
Objects allocated in this way are annotated with additional
502
information. Some of this information is checked during garbage
503
collections, and detected inconsistencies are reported to stderr.
505
Simple cases of writing past the end of an allocated object should
506
be caught if the object is explicitly deallocated, or if the
507
collector is invoked while the object is live. The first deallocation
508
of an object will clear the debugging info associated with an
509
object, so accidentally repeated calls to GC_debug_free will report the
510
deallocation of an object without debugging information. Out of
511
memory errors will be reported to stderr, in addition to returning
514
GC_debug_malloc checking during garbage collection is enabled
515
with the first call to GC_debug_malloc. This will result in some
516
slowdown during collections. If frequent heap checks are desired,
517
this can be achieved by explicitly invoking GC_gcollect, e.g. from
520
GC_debug_malloc allocated objects should not be passed to GC_realloc
521
or GC_free, and conversely. It is however acceptable to allocate only
522
some objects with GC_debug_malloc, and to use GC_malloc for other objects,
523
provided the two pools are kept distinct. In this case, there is a very
524
low probablility that GC_malloc allocated objects may be misidentified as
525
having been overwritten. This should happen with probability at most
526
one in 2**32. This probability is zero if GC_debug_malloc is never called.
528
GC_debug_malloc, GC_malloc_atomic, and GC_debug_realloc take two
529
additional trailing arguments, a string and an integer. These are not
530
interpreted by the allocator. They are stored in the object (the string is
531
not copied). If an error involving the object is detected, they are printed.
533
The macros GC_MALLOC, GC_MALLOC_ATOMIC, GC_REALLOC, GC_FREE, and
534
GC_REGISTER_FINALIZER are also provided. These require the same arguments
535
as the corresponding (nondebugging) routines. If gc.h is included
536
with GC_DEBUG defined, they call the debugging versions of these
537
functions, passing the current file name and line number as the two
538
extra arguments, where appropriate. If gc.h is included without GC_DEBUG
539
defined, then all these macros will instead be defined to their nondebugging
540
equivalents. (GC_REGISTER_FINALIZER is necessary, since pointers to
541
objects with debugging information are really pointers to a displacement
542
of 16 bytes form the object beginning, and some translation is necessary
543
when finalization routines are invoked. For details, about what's stored
544
in the header, see the definition of the type oh in debug_malloc.c)
546
INCREMENTAL/GENERATIONAL COLLECTION:
548
The collector normally interrupts client code for the duration of
549
a garbage collection mark phase. This may be unacceptable if interactive
550
response is needed for programs with large heaps. The collector
551
can also run in a "generational" mode, in which it usually attempts to
552
collect only objects allocated since the last garbage collection.
553
Furthermore, in this mode, garbage collections run mostly incrementally,
554
with a small amount of work performed in response to each of a large number of
557
This mode is enabled by a call to GC_enable_incremental().
559
Incremental and generational collection is effective in reducing
560
pause times only if the collector has some way to tell which objects
561
or pages have been recently modified. The collector uses two sources
564
1. Information provided by the VM system. This may be provided in
565
one of several forms. Under Solaris 2.X (and potentially under other
566
similar systems) information on dirty pages can be read from the
567
/proc file system. Under other systems (currently SunOS4.X) it is
568
possible to write-protect the heap, and catch the resulting faults.
569
On these systems we require that system calls writing to the heap
570
(other than read) be handled specially by client code.
571
See os_dep.c for details.
573
2. Information supplied by the programmer. We define "stubborn"
574
objects to be objects that are rarely changed. Such an object
575
can be allocated (and enabled for writing) with GC_malloc_stubborn.
576
Once it has been initialized, the collector should be informed with
577
a call to GC_end_stubborn_change. Subsequent writes that store
578
pointers into the object must be preceded by a call to
581
This mechanism performs best for objects that are written only for
582
initialization, and such that only one stubborn object is writable
583
at once. It is typically not worth using for short-lived
584
objects. Stubborn objects are treated less efficiently than pointerfree
587
A rough rule of thumb is that, in the absence of VM information, garbage
588
collection pauses are proportional to the amount of pointerful storage
589
plus the amount of modified "stubborn" storage that is reachable during
592
Initial allocation of stubborn objects takes longer than allocation
593
of other objects, since other data structures need to be maintained.
595
We recommend against random use of stubborn objects in client
596
code, since bugs caused by inappropriate writes to stubborn objects
597
are likely to be very infrequently observed and hard to trace.
598
However, their use may be appropriate in a few carefully written
599
library routines that do not make the objects themselves available
600
for writing by client code.
605
Any memory that does not have a recognizable pointer to it will be
606
reclaimed. Exclusive-or'ing forward and backward links in a list
608
Some C optimizers may lose the last undisguised pointer to a memory
609
object as a consequence of clever optimizations. This has almost
610
never been observed in practice. Send mail to boehm@acm.org
611
for suggestions on how to fix your compiler.
612
This is not a real-time collector. In the standard configuration,
613
percentage of time required for collection should be constant across
614
heap sizes. But collection pauses will increase for larger heaps.
615
(On SPARCstation 2s collection times will be on the order of 300 msecs
616
per MB of accessible memory that needs to be scanned. Your mileage
617
may vary.) The incremental/generational collection facility helps,
618
but is portable only if "stubborn" allocation is used.
619
Please address bug reports to boehm@acm.org. If you are
620
contemplating a major addition, you might also send mail to ask whether
621
it's already been done (or whether we tried and discarded it).
added
removed
src/gc/doc/README
