nexmon – Blame information for rev 1

Subversion Repositories:
Rev:
Rev Author Line No. Line
1 office 1 /* GLIB sliced memory - fast concurrent memory chunk allocator
2 * Copyright (C) 2005 Tim Janik
3 *
4 * This library is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU Lesser General Public
6 * License as published by the Free Software Foundation; either
7 * version 2 of the License, or (at your option) any later version.
8 *
9 * This library is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * Lesser General Public License for more details.
13 *
14 * You should have received a copy of the GNU Lesser General Public
15 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
16 */
17 /* MT safe */
18  
19 #include "config.h"
20 #include "glibconfig.h"
21  
22 #if defined HAVE_POSIX_MEMALIGN && defined POSIX_MEMALIGN_WITH_COMPLIANT_ALLOCS
23 # define HAVE_COMPLIANT_POSIX_MEMALIGN 1
24 #endif
25  
26 #if defined(HAVE_COMPLIANT_POSIX_MEMALIGN) && !defined(_XOPEN_SOURCE)
27 #define _XOPEN_SOURCE 600 /* posix_memalign() */
28 #endif
29 #include <stdlib.h> /* posix_memalign() */
30 #include <string.h>
31 #include <errno.h>
32  
33 #ifdef G_OS_UNIX
34 #include <unistd.h> /* sysconf() */
35 #endif
36 #ifdef G_OS_WIN32
37 #include <windows.h>
38 #include <process.h>
39 #endif
40  
41 #include <stdio.h> /* fputs/fprintf */
42  
43 #include "gslice.h"
44  
45 #include "gmain.h"
46 #include "gmem.h" /* gslice.h */
47 #include "gstrfuncs.h"
48 #include "gutils.h"
49 #include "gtrashstack.h"
50 #include "gtestutils.h"
51 #include "gthread.h"
52 #include "glib_trace.h"
53  
54 #include "valgrind.h"
55  
56 /**
57 * SECTION:memory_slices
58 * @title: Memory Slices
59 * @short_description: efficient way to allocate groups of equal-sized
60 * chunks of memory
61 *
62 * Memory slices provide a space-efficient and multi-processing scalable
63 * way to allocate equal-sized pieces of memory, just like the original
64 * #GMemChunks (from GLib 2.8), while avoiding their excessive
65 * memory-waste, scalability and performance problems.
66 *
67 * To achieve these goals, the slice allocator uses a sophisticated,
68 * layered design that has been inspired by Bonwick's slab allocator
69 * ([Bonwick94](http://citeseer.ist.psu.edu/bonwick94slab.html)
70 * Jeff Bonwick, The slab allocator: An object-caching kernel
71 * memory allocator. USENIX 1994, and
72 * [Bonwick01](http://citeseer.ist.psu.edu/bonwick01magazines.html)
73 * Bonwick and Jonathan Adams, Magazines and vmem: Extending the
74 * slab allocator to many cpu's and arbitrary resources. USENIX 2001)
75 *
76 * It uses posix_memalign() to optimize allocations of many equally-sized
77 * chunks, and has per-thread free lists (the so-called magazine layer)
78 * to quickly satisfy allocation requests of already known structure sizes.
79 * This is accompanied by extra caching logic to keep freed memory around
80 * for some time before returning it to the system. Memory that is unused
81 * due to alignment constraints is used for cache colorization (random
82 * distribution of chunk addresses) to improve CPU cache utilization. The
83 * caching layer of the slice allocator adapts itself to high lock contention
84 * to improve scalability.
85 *
86 * The slice allocator can allocate blocks as small as two pointers, and
87 * unlike malloc(), it does not reserve extra space per block. For large block
88 * sizes, g_slice_new() and g_slice_alloc() will automatically delegate to the
89 * system malloc() implementation. For newly written code it is recommended
90 * to use the new `g_slice` API instead of g_malloc() and
91 * friends, as long as objects are not resized during their lifetime and the
92 * object size used at allocation time is still available when freeing.
93 *
94 * Here is an example for using the slice allocator:
95 * |[<!-- language="C" -->
96 * gchar *mem[10000];
97 * gint i;
98 *
99 * // Allocate 10000 blocks.
100 * for (i = 0; i < 10000; i++)
101 * {
102 * mem[i] = g_slice_alloc (50);
103 *
104 * // Fill in the memory with some junk.
105 * for (j = 0; j < 50; j++)
106 * mem[i][j] = i * j;
107 * }
108 *
109 * // Now free all of the blocks.
110 * for (i = 0; i < 10000; i++)
111 * g_slice_free1 (50, mem[i]);
112 * ]|
113 *
114 * And here is an example for using the using the slice allocator
115 * with data structures:
116 * |[<!-- language="C" -->
117 * GRealArray *array;
118 *
119 * // Allocate one block, using the g_slice_new() macro.
120 * array = g_slice_new (GRealArray);
121  
122 * // We can now use array just like a normal pointer to a structure.
123 * array->data = NULL;
124 * array->len = 0;
125 * array->alloc = 0;
126 * array->zero_terminated = (zero_terminated ? 1 : 0);
127 * array->clear = (clear ? 1 : 0);
128 * array->elt_size = elt_size;
129 *
130 * // We can free the block, so it can be reused.
131 * g_slice_free (GRealArray, array);
132 * ]|
133 */
134  
135 /* the GSlice allocator is split up into 4 layers, roughly modelled after the slab
136 * allocator and magazine extensions as outlined in:
137 * + [Bonwick94] Jeff Bonwick, The slab allocator: An object-caching kernel
138 * memory allocator. USENIX 1994, http://citeseer.ist.psu.edu/bonwick94slab.html
139 * + [Bonwick01] Bonwick and Jonathan Adams, Magazines and vmem: Extending the
140 * slab allocator to many cpu's and arbitrary resources.
141 * USENIX 2001, http://citeseer.ist.psu.edu/bonwick01magazines.html
142 * the layers are:
143 * - the thread magazines. for each (aligned) chunk size, a magazine (a list)
144 * of recently freed and soon to be allocated chunks is maintained per thread.
145 * this way, most alloc/free requests can be quickly satisfied from per-thread
146 * free lists which only require one g_private_get() call to retrive the
147 * thread handle.
148 * - the magazine cache. allocating and freeing chunks to/from threads only
149 * occours at magazine sizes from a global depot of magazines. the depot
150 * maintaines a 15 second working set of allocated magazines, so full
151 * magazines are not allocated and released too often.
152 * the chunk size dependent magazine sizes automatically adapt (within limits,
153 * see [3]) to lock contention to properly scale performance across a variety
154 * of SMP systems.
155 * - the slab allocator. this allocator allocates slabs (blocks of memory) close
156 * to the system page size or multiples thereof which have to be page aligned.
157 * the blocks are divided into smaller chunks which are used to satisfy
158 * allocations from the upper layers. the space provided by the reminder of
159 * the chunk size division is used for cache colorization (random distribution
160 * of chunk addresses) to improve processor cache utilization. multiple slabs
161 * with the same chunk size are kept in a partially sorted ring to allow O(1)
162 * freeing and allocation of chunks (as long as the allocation of an entirely
163 * new slab can be avoided).
164 * - the page allocator. on most modern systems, posix_memalign(3) or
165 * memalign(3) should be available, so this is used to allocate blocks with
166 * system page size based alignments and sizes or multiples thereof.
167 * if no memalign variant is provided, valloc() is used instead and
168 * block sizes are limited to the system page size (no multiples thereof).
169 * as a fallback, on system without even valloc(), a malloc(3)-based page
170 * allocator with alloc-only behaviour is used.
171 *
172 * NOTES:
173 * [1] some systems memalign(3) implementations may rely on boundary tagging for
174 * the handed out memory chunks. to avoid excessive page-wise fragmentation,
175 * we reserve 2 * sizeof (void*) per block size for the systems memalign(3),
176 * specified in NATIVE_MALLOC_PADDING.
177 * [2] using the slab allocator alone already provides for a fast and efficient
178 * allocator, it doesn't properly scale beyond single-threaded uses though.
179 * also, the slab allocator implements eager free(3)-ing, i.e. does not
180 * provide any form of caching or working set maintenance. so if used alone,
181 * it's vulnerable to trashing for sequences of balanced (alloc, free) pairs
182 * at certain thresholds.
183 * [3] magazine sizes are bound by an implementation specific minimum size and
184 * a chunk size specific maximum to limit magazine storage sizes to roughly
185 * 16KB.
186 * [4] allocating ca. 8 chunks per block/page keeps a good balance between
187 * external and internal fragmentation (<= 12.5%). [Bonwick94]
188 */
189  
190 /* --- macros and constants --- */
191 #define LARGEALIGNMENT (256)
192 #define P2ALIGNMENT (2 * sizeof (gsize)) /* fits 2 pointers (assumed to be 2 * GLIB_SIZEOF_SIZE_T below) */
193 #define ALIGN(size, base) ((base) * (gsize) (((size) + (base) - 1) / (base)))
194 #define NATIVE_MALLOC_PADDING P2ALIGNMENT /* per-page padding left for native malloc(3) see [1] */
195 #define SLAB_INFO_SIZE P2ALIGN (sizeof (SlabInfo) + NATIVE_MALLOC_PADDING)
196 #define MAX_MAGAZINE_SIZE (256) /* see [3] and allocator_get_magazine_threshold() for this */
197 #define MIN_MAGAZINE_SIZE (4)
198 #define MAX_STAMP_COUNTER (7) /* distributes the load of gettimeofday() */
199 #define MAX_SLAB_CHUNK_SIZE(al) (((al)->max_page_size - SLAB_INFO_SIZE) / 8) /* we want at last 8 chunks per page, see [4] */
200 #define MAX_SLAB_INDEX(al) (SLAB_INDEX (al, MAX_SLAB_CHUNK_SIZE (al)) + 1)
201 #define SLAB_INDEX(al, asize) ((asize) / P2ALIGNMENT - 1) /* asize must be P2ALIGNMENT aligned */
202 #define SLAB_CHUNK_SIZE(al, ix) (((ix) + 1) * P2ALIGNMENT)
203 #define SLAB_BPAGE_SIZE(al,csz) (8 * (csz) + SLAB_INFO_SIZE)
204  
205 /* optimized version of ALIGN (size, P2ALIGNMENT) */
206 #if GLIB_SIZEOF_SIZE_T * 2 == 8 /* P2ALIGNMENT */
207 #define P2ALIGN(size) (((size) + 0x7) & ~(gsize) 0x7)
208 #elif GLIB_SIZEOF_SIZE_T * 2 == 16 /* P2ALIGNMENT */
209 #define P2ALIGN(size) (((size) + 0xf) & ~(gsize) 0xf)
210 #else
211 #define P2ALIGN(size) ALIGN (size, P2ALIGNMENT)
212 #endif
213  
214 /* special helpers to avoid gmessage.c dependency */
215 static void mem_error (const char *format, ...) G_GNUC_PRINTF (1,2);
216 #define mem_assert(cond) do { if (G_LIKELY (cond)) ; else mem_error ("assertion failed: %s", #cond); } while (0)
217  
218 /* --- structures --- */
219 typedef struct _ChunkLink ChunkLink;
220 typedef struct _SlabInfo SlabInfo;
221 typedef struct _CachedMagazine CachedMagazine;
222 struct _ChunkLink {
223 ChunkLink *next;
224 ChunkLink *data;
225 };
226 struct _SlabInfo {
227 ChunkLink *chunks;
228 guint n_allocated;
229 SlabInfo *next, *prev;
230 };
231 typedef struct {
232 ChunkLink *chunks;
233 gsize count; /* approximative chunks list length */
234 } Magazine;
235 typedef struct {
236 Magazine *magazine1; /* array of MAX_SLAB_INDEX (allocator) */
237 Magazine *magazine2; /* array of MAX_SLAB_INDEX (allocator) */
238 } ThreadMemory;
239 typedef struct {
240 gboolean always_malloc;
241 gboolean bypass_magazines;
242 gboolean debug_blocks;
243 gsize working_set_msecs;
244 guint color_increment;
245 } SliceConfig;
246 typedef struct {
247 /* const after initialization */
248 gsize min_page_size, max_page_size;
249 SliceConfig config;
250 gsize max_slab_chunk_size_for_magazine_cache;
251 /* magazine cache */
252 GMutex magazine_mutex;
253 ChunkLink **magazines; /* array of MAX_SLAB_INDEX (allocator) */
254 guint *contention_counters; /* array of MAX_SLAB_INDEX (allocator) */
255 gint mutex_counter;
256 guint stamp_counter;
257 guint last_stamp;
258 /* slab allocator */
259 GMutex slab_mutex;
260 SlabInfo **slab_stack; /* array of MAX_SLAB_INDEX (allocator) */
261 guint color_accu;
262 } Allocator;
263  
264 /* --- g-slice prototypes --- */
265 static gpointer slab_allocator_alloc_chunk (gsize chunk_size);
266 static void slab_allocator_free_chunk (gsize chunk_size,
267 gpointer mem);
268 static void private_thread_memory_cleanup (gpointer data);
269 static gpointer allocator_memalign (gsize alignment,
270 gsize memsize);
271 static void allocator_memfree (gsize memsize,
272 gpointer mem);
273 static inline void magazine_cache_update_stamp (void);
274 static inline gsize allocator_get_magazine_threshold (Allocator *allocator,
275 guint ix);
276  
277 /* --- g-slice memory checker --- */
278 static void smc_notify_alloc (void *pointer,
279 size_t size);
280 static int smc_notify_free (void *pointer,
281 size_t size);
282  
283 /* --- variables --- */
284 static GPrivate private_thread_memory = G_PRIVATE_INIT (private_thread_memory_cleanup);
285 static gsize sys_page_size = 0;
286 static Allocator allocator[1] = { { 0, }, };
287 static SliceConfig slice_config = {
288 FALSE, /* always_malloc */
289 FALSE, /* bypass_magazines */
290 FALSE, /* debug_blocks */
291 15 * 1000, /* working_set_msecs */
292 1, /* color increment, alt: 0x7fffffff */
293 };
294 static GMutex smc_tree_mutex; /* mutex for G_SLICE=debug-blocks */
295  
296 /* --- auxiliary funcitons --- */
297 void
298 g_slice_set_config (GSliceConfig ckey,
299 gint64 value)
300 {
301 g_return_if_fail (sys_page_size == 0);
302 switch (ckey)
303 {
304 case G_SLICE_CONFIG_ALWAYS_MALLOC:
305 slice_config.always_malloc = value != 0;
306 break;
307 case G_SLICE_CONFIG_BYPASS_MAGAZINES:
308 slice_config.bypass_magazines = value != 0;
309 break;
310 case G_SLICE_CONFIG_WORKING_SET_MSECS:
311 slice_config.working_set_msecs = value;
312 break;
313 case G_SLICE_CONFIG_COLOR_INCREMENT:
314 slice_config.color_increment = value;
315 default: ;
316 }
317 }
318  
319 gint64
320 g_slice_get_config (GSliceConfig ckey)
321 {
322 switch (ckey)
323 {
324 case G_SLICE_CONFIG_ALWAYS_MALLOC:
325 return slice_config.always_malloc;
326 case G_SLICE_CONFIG_BYPASS_MAGAZINES:
327 return slice_config.bypass_magazines;
328 case G_SLICE_CONFIG_WORKING_SET_MSECS:
329 return slice_config.working_set_msecs;
330 case G_SLICE_CONFIG_CHUNK_SIZES:
331 return MAX_SLAB_INDEX (allocator);
332 case G_SLICE_CONFIG_COLOR_INCREMENT:
333 return slice_config.color_increment;
334 default:
335 return 0;
336 }
337 }
338  
339 gint64*
340 g_slice_get_config_state (GSliceConfig ckey,
341 gint64 address,
342 guint *n_values)
343 {
344 guint i = 0;
345 g_return_val_if_fail (n_values != NULL, NULL);
346 *n_values = 0;
347 switch (ckey)
348 {
349 gint64 array[64];
350 case G_SLICE_CONFIG_CONTENTION_COUNTER:
351 array[i++] = SLAB_CHUNK_SIZE (allocator, address);
352 array[i++] = allocator->contention_counters[address];
353 array[i++] = allocator_get_magazine_threshold (allocator, address);
354 *n_values = i;
355 return g_memdup (array, sizeof (array[0]) * *n_values);
356 default:
357 return NULL;
358 }
359 }
360  
361 static void
362 slice_config_init (SliceConfig *config)
363 {
364 const gchar *val;
365  
366 *config = slice_config;
367  
368 val = getenv ("G_SLICE");
369 if (val != NULL)
370 {
371 gint flags;
372 const GDebugKey keys[] = {
373 { "always-malloc", 1 << 0 },
374 { "debug-blocks", 1 << 1 },
375 };
376  
377 flags = g_parse_debug_string (val, keys, G_N_ELEMENTS (keys));
378 if (flags & (1 << 0))
379 config->always_malloc = TRUE;
380 if (flags & (1 << 1))
381 config->debug_blocks = TRUE;
382 }
383 else
384 {
385 /* G_SLICE was not specified, so check if valgrind is running and
386 * disable ourselves if it is.
387 *
388 * This way it's possible to force gslice to be enabled under
389 * valgrind just by setting G_SLICE to the empty string.
390 */
391 if (RUNNING_ON_VALGRIND)
392 config->always_malloc = TRUE;
393 }
394 }
395  
396 static void
397 g_slice_init_nomessage (void)
398 {
399 /* we may not use g_error() or friends here */
400 mem_assert (sys_page_size == 0);
401 mem_assert (MIN_MAGAZINE_SIZE >= 4);
402  
403 #ifdef G_OS_WIN32
404 {
405 SYSTEM_INFO system_info;
406 GetSystemInfo (&system_info);
407 sys_page_size = system_info.dwPageSize;
408 }
409 #else
410 sys_page_size = sysconf (_SC_PAGESIZE); /* = sysconf (_SC_PAGE_SIZE); = getpagesize(); */
411 #endif
412 mem_assert (sys_page_size >= 2 * LARGEALIGNMENT);
413 mem_assert ((sys_page_size & (sys_page_size - 1)) == 0);
414 slice_config_init (&allocator->config);
415 allocator->min_page_size = sys_page_size;
416 #if HAVE_COMPLIANT_POSIX_MEMALIGN || HAVE_MEMALIGN
417 /* allow allocation of pages up to 8KB (with 8KB alignment).
418 * this is useful because many medium to large sized structures
419 * fit less than 8 times (see [4]) into 4KB pages.
420 * we allow very small page sizes here, to reduce wastage in
421 * threads if only small allocations are required (this does
422 * bear the risk of increasing allocation times and fragmentation
423 * though).
424 */
425 allocator->min_page_size = MAX (allocator->min_page_size, 4096);
426 allocator->max_page_size = MAX (allocator->min_page_size, 8192);
427 allocator->min_page_size = MIN (allocator->min_page_size, 128);
428 #else
429 /* we can only align to system page size */
430 allocator->max_page_size = sys_page_size;
431 #endif
432 if (allocator->config.always_malloc)
433 {
434 allocator->contention_counters = NULL;
435 allocator->magazines = NULL;
436 allocator->slab_stack = NULL;
437 }
438 else
439 {
440 allocator->contention_counters = g_new0 (guint, MAX_SLAB_INDEX (allocator));
441 allocator->magazines = g_new0 (ChunkLink*, MAX_SLAB_INDEX (allocator));
442 allocator->slab_stack = g_new0 (SlabInfo*, MAX_SLAB_INDEX (allocator));
443 }
444  
445 allocator->mutex_counter = 0;
446 allocator->stamp_counter = MAX_STAMP_COUNTER; /* force initial update */
447 allocator->last_stamp = 0;
448 allocator->color_accu = 0;
449 magazine_cache_update_stamp();
450 /* values cached for performance reasons */
451 allocator->max_slab_chunk_size_for_magazine_cache = MAX_SLAB_CHUNK_SIZE (allocator);
452 if (allocator->config.always_malloc || allocator->config.bypass_magazines)
453 allocator->max_slab_chunk_size_for_magazine_cache = 0; /* non-optimized cases */
454 }
455  
456 static inline guint
457 allocator_categorize (gsize aligned_chunk_size)
458 {
459 /* speed up the likely path */
460 if (G_LIKELY (aligned_chunk_size && aligned_chunk_size <= allocator->max_slab_chunk_size_for_magazine_cache))
461 return 1; /* use magazine cache */
462  
463 if (!allocator->config.always_malloc &&
464 aligned_chunk_size &&
465 aligned_chunk_size <= MAX_SLAB_CHUNK_SIZE (allocator))
466 {
467 if (allocator->config.bypass_magazines)
468 return 2; /* use slab allocator, see [2] */
469 return 1; /* use magazine cache */
470 }
471 return 0; /* use malloc() */
472 }
473  
474 static inline void
475 g_mutex_lock_a (GMutex *mutex,
476 guint *contention_counter)
477 {
478 gboolean contention = FALSE;
479 if (!g_mutex_trylock (mutex))
480 {
481 g_mutex_lock (mutex);
482 contention = TRUE;
483 }
484 if (contention)
485 {
486 allocator->mutex_counter++;
487 if (allocator->mutex_counter >= 1) /* quickly adapt to contention */
488 {
489 allocator->mutex_counter = 0;
490 *contention_counter = MIN (*contention_counter + 1, MAX_MAGAZINE_SIZE);
491 }
492 }
493 else /* !contention */
494 {
495 allocator->mutex_counter--;
496 if (allocator->mutex_counter < -11) /* moderately recover magazine sizes */
497 {
498 allocator->mutex_counter = 0;
499 *contention_counter = MAX (*contention_counter, 1) - 1;
500 }
501 }
502 }
503  
504 static inline ThreadMemory*
505 thread_memory_from_self (void)
506 {
507 ThreadMemory *tmem = g_private_get (&private_thread_memory);
508 if (G_UNLIKELY (!tmem))
509 {
510 static GMutex init_mutex;
511 guint n_magazines;
512  
513 g_mutex_lock (&init_mutex);
514 if G_UNLIKELY (sys_page_size == 0)
515 g_slice_init_nomessage ();
516 g_mutex_unlock (&init_mutex);
517  
518 n_magazines = MAX_SLAB_INDEX (allocator);
519 tmem = g_malloc0 (sizeof (ThreadMemory) + sizeof (Magazine) * 2 * n_magazines);
520 tmem->magazine1 = (Magazine*) (tmem + 1);
521 tmem->magazine2 = &tmem->magazine1[n_magazines];
522 g_private_set (&private_thread_memory, tmem);
523 }
524 return tmem;
525 }
526  
527 static inline ChunkLink*
528 magazine_chain_pop_head (ChunkLink **magazine_chunks)
529 {
530 /* magazine chains are linked via ChunkLink->next.
531 * each ChunkLink->data of the toplevel chain may point to a subchain,
532 * linked via ChunkLink->next. ChunkLink->data of the subchains just
533 * contains uninitialized junk.
534 */
535 ChunkLink *chunk = (*magazine_chunks)->data;
536 if (G_UNLIKELY (chunk))
537 {
538 /* allocating from freed list */
539 (*magazine_chunks)->data = chunk->next;
540 }
541 else
542 {
543 chunk = *magazine_chunks;
544 *magazine_chunks = chunk->next;
545 }
546 return chunk;
547 }
548  
549 #if 0 /* useful for debugging */
550 static guint
551 magazine_count (ChunkLink *head)
552 {
553 guint count = 0;
554 if (!head)
555 return 0;
556 while (head)
557 {
558 ChunkLink *child = head->data;
559 count += 1;
560 for (child = head->data; child; child = child->next)
561 count += 1;
562 head = head->next;
563 }
564 return count;
565 }
566 #endif
567  
568 static inline gsize
569 allocator_get_magazine_threshold (Allocator *allocator,
570 guint ix)
571 {
572 /* the magazine size calculated here has a lower bound of MIN_MAGAZINE_SIZE,
573 * which is required by the implementation. also, for moderately sized chunks
574 * (say >= 64 bytes), magazine sizes shouldn't be much smaller then the number
575 * of chunks available per page/2 to avoid excessive traffic in the magazine
576 * cache for small to medium sized structures.
577 * the upper bound of the magazine size is effectively provided by
578 * MAX_MAGAZINE_SIZE. for larger chunks, this number is scaled down so that
579 * the content of a single magazine doesn't exceed ca. 16KB.
580 */
581 gsize chunk_size = SLAB_CHUNK_SIZE (allocator, ix);
582 guint threshold = MAX (MIN_MAGAZINE_SIZE, allocator->max_page_size / MAX (5 * chunk_size, 5 * 32));
583 guint contention_counter = allocator->contention_counters[ix];
584 if (G_UNLIKELY (contention_counter)) /* single CPU bias */
585 {
586 /* adapt contention counter thresholds to chunk sizes */
587 contention_counter = contention_counter * 64 / chunk_size;
588 threshold = MAX (threshold, contention_counter);
589 }
590 return threshold;
591 }
592  
593 /* --- magazine cache --- */
594 static inline void
595 magazine_cache_update_stamp (void)
596 {
597 if (allocator->stamp_counter >= MAX_STAMP_COUNTER)
598 {
599 GTimeVal tv;
600 g_get_current_time (&tv);
601 allocator->last_stamp = tv.tv_sec * 1000 + tv.tv_usec / 1000; /* milli seconds */
602 allocator->stamp_counter = 0;
603 }
604 else
605 allocator->stamp_counter++;
606 }
607  
608 static inline ChunkLink*
609 magazine_chain_prepare_fields (ChunkLink *magazine_chunks)
610 {
611 ChunkLink *chunk1;
612 ChunkLink *chunk2;
613 ChunkLink *chunk3;
614 ChunkLink *chunk4;
615 /* checked upon initialization: mem_assert (MIN_MAGAZINE_SIZE >= 4); */
616 /* ensure a magazine with at least 4 unused data pointers */
617 chunk1 = magazine_chain_pop_head (&magazine_chunks);
618 chunk2 = magazine_chain_pop_head (&magazine_chunks);
619 chunk3 = magazine_chain_pop_head (&magazine_chunks);
620 chunk4 = magazine_chain_pop_head (&magazine_chunks);
621 chunk4->next = magazine_chunks;
622 chunk3->next = chunk4;
623 chunk2->next = chunk3;
624 chunk1->next = chunk2;
625 return chunk1;
626 }
627  
628 /* access the first 3 fields of a specially prepared magazine chain */
629 #define magazine_chain_prev(mc) ((mc)->data)
630 #define magazine_chain_stamp(mc) ((mc)->next->data)
631 #define magazine_chain_uint_stamp(mc) GPOINTER_TO_UINT ((mc)->next->data)
632 #define magazine_chain_next(mc) ((mc)->next->next->data)
633 #define magazine_chain_count(mc) ((mc)->next->next->next->data)
634  
635 static void
636 magazine_cache_trim (Allocator *allocator,
637 guint ix,
638 guint stamp)
639 {
640 /* g_mutex_lock (allocator->mutex); done by caller */
641 /* trim magazine cache from tail */
642 ChunkLink *current = magazine_chain_prev (allocator->magazines[ix]);
643 ChunkLink *trash = NULL;
644 while (ABS (stamp - magazine_chain_uint_stamp (current)) >= allocator->config.working_set_msecs)
645 {
646 /* unlink */
647 ChunkLink *prev = magazine_chain_prev (current);
648 ChunkLink *next = magazine_chain_next (current);
649 magazine_chain_next (prev) = next;
650 magazine_chain_prev (next) = prev;
651 /* clear special fields, put on trash stack */
652 magazine_chain_next (current) = NULL;
653 magazine_chain_count (current) = NULL;
654 magazine_chain_stamp (current) = NULL;
655 magazine_chain_prev (current) = trash;
656 trash = current;
657 /* fixup list head if required */
658 if (current == allocator->magazines[ix])
659 {
660 allocator->magazines[ix] = NULL;
661 break;
662 }
663 current = prev;
664 }
665 g_mutex_unlock (&allocator->magazine_mutex);
666 /* free trash */
667 if (trash)
668 {
669 const gsize chunk_size = SLAB_CHUNK_SIZE (allocator, ix);
670 g_mutex_lock (&allocator->slab_mutex);
671 while (trash)
672 {
673 current = trash;
674 trash = magazine_chain_prev (current);
675 magazine_chain_prev (current) = NULL; /* clear special field */
676 while (current)
677 {
678 ChunkLink *chunk = magazine_chain_pop_head (&current);
679 slab_allocator_free_chunk (chunk_size, chunk);
680 }
681 }
682 g_mutex_unlock (&allocator->slab_mutex);
683 }
684 }
685  
686 static void
687 magazine_cache_push_magazine (guint ix,
688 ChunkLink *magazine_chunks,
689 gsize count) /* must be >= MIN_MAGAZINE_SIZE */
690 {
691 ChunkLink *current = magazine_chain_prepare_fields (magazine_chunks);
692 ChunkLink *next, *prev;
693 g_mutex_lock (&allocator->magazine_mutex);
694 /* add magazine at head */
695 next = allocator->magazines[ix];
696 if (next)
697 prev = magazine_chain_prev (next);
698 else
699 next = prev = current;
700 magazine_chain_next (prev) = current;
701 magazine_chain_prev (next) = current;
702 magazine_chain_prev (current) = prev;
703 magazine_chain_next (current) = next;
704 magazine_chain_count (current) = (gpointer) count;
705 /* stamp magazine */
706 magazine_cache_update_stamp();
707 magazine_chain_stamp (current) = GUINT_TO_POINTER (allocator->last_stamp);
708 allocator->magazines[ix] = current;
709 /* free old magazines beyond a certain threshold */
710 magazine_cache_trim (allocator, ix, allocator->last_stamp);
711 /* g_mutex_unlock (allocator->mutex); was done by magazine_cache_trim() */
712 }
713  
714 static ChunkLink*
715 magazine_cache_pop_magazine (guint ix,
716 gsize *countp)
717 {
718 g_mutex_lock_a (&allocator->magazine_mutex, &allocator->contention_counters[ix]);
719 if (!allocator->magazines[ix])
720 {
721 guint magazine_threshold = allocator_get_magazine_threshold (allocator, ix);
722 gsize i, chunk_size = SLAB_CHUNK_SIZE (allocator, ix);
723 ChunkLink *chunk, *head;
724 g_mutex_unlock (&allocator->magazine_mutex);
725 g_mutex_lock (&allocator->slab_mutex);
726 head = slab_allocator_alloc_chunk (chunk_size);
727 head->data = NULL;
728 chunk = head;
729 for (i = 1; i < magazine_threshold; i++)
730 {
731 chunk->next = slab_allocator_alloc_chunk (chunk_size);
732 chunk = chunk->next;
733 chunk->data = NULL;
734 }
735 chunk->next = NULL;
736 g_mutex_unlock (&allocator->slab_mutex);
737 *countp = i;
738 return head;
739 }
740 else
741 {
742 ChunkLink *current = allocator->magazines[ix];
743 ChunkLink *prev = magazine_chain_prev (current);
744 ChunkLink *next = magazine_chain_next (current);
745 /* unlink */
746 magazine_chain_next (prev) = next;
747 magazine_chain_prev (next) = prev;
748 allocator->magazines[ix] = next == current ? NULL : next;
749 g_mutex_unlock (&allocator->magazine_mutex);
750 /* clear special fields and hand out */
751 *countp = (gsize) magazine_chain_count (current);
752 magazine_chain_prev (current) = NULL;
753 magazine_chain_next (current) = NULL;
754 magazine_chain_count (current) = NULL;
755 magazine_chain_stamp (current) = NULL;
756 return current;
757 }
758 }
759  
760 /* --- thread magazines --- */
761 static void
762 private_thread_memory_cleanup (gpointer data)
763 {
764 ThreadMemory *tmem = data;
765 const guint n_magazines = MAX_SLAB_INDEX (allocator);
766 guint ix;
767 for (ix = 0; ix < n_magazines; ix++)
768 {
769 Magazine *mags[2];
770 guint j;
771 mags[0] = &tmem->magazine1[ix];
772 mags[1] = &tmem->magazine2[ix];
773 for (j = 0; j < 2; j++)
774 {
775 Magazine *mag = mags[j];
776 if (mag->count >= MIN_MAGAZINE_SIZE)
777 magazine_cache_push_magazine (ix, mag->chunks, mag->count);
778 else
779 {
780 const gsize chunk_size = SLAB_CHUNK_SIZE (allocator, ix);
781 g_mutex_lock (&allocator->slab_mutex);
782 while (mag->chunks)
783 {
784 ChunkLink *chunk = magazine_chain_pop_head (&mag->chunks);
785 slab_allocator_free_chunk (chunk_size, chunk);
786 }
787 g_mutex_unlock (&allocator->slab_mutex);
788 }
789 }
790 }
791 g_free (tmem);
792 }
793  
794 static void
795 thread_memory_magazine1_reload (ThreadMemory *tmem,
796 guint ix)
797 {
798 Magazine *mag = &tmem->magazine1[ix];
799 mem_assert (mag->chunks == NULL); /* ensure that we may reset mag->count */
800 mag->count = 0;
801 mag->chunks = magazine_cache_pop_magazine (ix, &mag->count);
802 }
803  
804 static void
805 thread_memory_magazine2_unload (ThreadMemory *tmem,
806 guint ix)
807 {
808 Magazine *mag = &tmem->magazine2[ix];
809 magazine_cache_push_magazine (ix, mag->chunks, mag->count);
810 mag->chunks = NULL;
811 mag->count = 0;
812 }
813  
814 static inline void
815 thread_memory_swap_magazines (ThreadMemory *tmem,
816 guint ix)
817 {
818 Magazine xmag = tmem->magazine1[ix];
819 tmem->magazine1[ix] = tmem->magazine2[ix];
820 tmem->magazine2[ix] = xmag;
821 }
822  
823 static inline gboolean
824 thread_memory_magazine1_is_empty (ThreadMemory *tmem,
825 guint ix)
826 {
827 return tmem->magazine1[ix].chunks == NULL;
828 }
829  
830 static inline gboolean
831 thread_memory_magazine2_is_full (ThreadMemory *tmem,
832 guint ix)
833 {
834 return tmem->magazine2[ix].count >= allocator_get_magazine_threshold (allocator, ix);
835 }
836  
837 static inline gpointer
838 thread_memory_magazine1_alloc (ThreadMemory *tmem,
839 guint ix)
840 {
841 Magazine *mag = &tmem->magazine1[ix];
842 ChunkLink *chunk = magazine_chain_pop_head (&mag->chunks);
843 if (G_LIKELY (mag->count > 0))
844 mag->count--;
845 return chunk;
846 }
847  
848 static inline void
849 thread_memory_magazine2_free (ThreadMemory *tmem,
850 guint ix,
851 gpointer mem)
852 {
853 Magazine *mag = &tmem->magazine2[ix];
854 ChunkLink *chunk = mem;
855 chunk->data = NULL;
856 chunk->next = mag->chunks;
857 mag->chunks = chunk;
858 mag->count++;
859 }
860  
861 /* --- API functions --- */
862  
863 /**
864 * g_slice_new:
865 * @type: the type to allocate, typically a structure name
866 *
867 * A convenience macro to allocate a block of memory from the
868 * slice allocator.
869 *
870 * It calls g_slice_alloc() with `sizeof (@type)` and casts the
871 * returned pointer to a pointer of the given type, avoiding a type
872 * cast in the source code. Note that the underlying slice allocation
873 * mechanism can be changed with the [`G_SLICE=always-malloc`][G_SLICE]
874 * environment variable.
875 *
876 * This can never return %NULL as the minimum allocation size from
877 * `sizeof (@type)` is 1 byte.
878 *
879 * Returns: (not nullable): a pointer to the allocated block, cast to a pointer
880 * to @type
881 *
882 * Since: 2.10
883 */
884  
885 /**
886 * g_slice_new0:
887 * @type: the type to allocate, typically a structure name
888 *
889 * A convenience macro to allocate a block of memory from the
890 * slice allocator and set the memory to 0.
891 *
892 * It calls g_slice_alloc0() with `sizeof (@type)`
893 * and casts the returned pointer to a pointer of the given type,
894 * avoiding a type cast in the source code.
895 * Note that the underlying slice allocation mechanism can
896 * be changed with the [`G_SLICE=always-malloc`][G_SLICE]
897 * environment variable.
898 *
899 * This can never return %NULL as the minimum allocation size from
900 * `sizeof (@type)` is 1 byte.
901 *
902 * Returns: (not nullable): a pointer to the allocated block, cast to a pointer
903 * to @type
904 *
905 * Since: 2.10
906 */
907  
908 /**
909 * g_slice_dup:
910 * @type: the type to duplicate, typically a structure name
911 * @mem: (not nullable): the memory to copy into the allocated block
912 *
913 * A convenience macro to duplicate a block of memory using
914 * the slice allocator.
915 *
916 * It calls g_slice_copy() with `sizeof (@type)`
917 * and casts the returned pointer to a pointer of the given type,
918 * avoiding a type cast in the source code.
919 * Note that the underlying slice allocation mechanism can
920 * be changed with the [`G_SLICE=always-malloc`][G_SLICE]
921 * environment variable.
922 *
923 * This can never return %NULL.
924 *
925 * Returns: (not nullable): a pointer to the allocated block, cast to a pointer
926 * to @type
927 *
928 * Since: 2.14
929 */
930  
931 /**
932 * g_slice_free:
933 * @type: the type of the block to free, typically a structure name
934 * @mem: a pointer to the block to free
935 *
936 * A convenience macro to free a block of memory that has
937 * been allocated from the slice allocator.
938 *
939 * It calls g_slice_free1() using `sizeof (type)`
940 * as the block size.
941 * Note that the exact release behaviour can be changed with the
942 * [`G_DEBUG=gc-friendly`][G_DEBUG] environment variable, also see
943 * [`G_SLICE`][G_SLICE] for related debugging options.
944 *
945 * If @mem is %NULL, this macro does nothing.
946 *
947 * Since: 2.10
948 */
949  
950 /**
951 * g_slice_free_chain:
952 * @type: the type of the @mem_chain blocks
953 * @mem_chain: a pointer to the first block of the chain
954 * @next: the field name of the next pointer in @type
955 *
956 * Frees a linked list of memory blocks of structure type @type.
957 * The memory blocks must be equal-sized, allocated via
958 * g_slice_alloc() or g_slice_alloc0() and linked together by
959 * a @next pointer (similar to #GSList). The name of the
960 * @next field in @type is passed as third argument.
961 * Note that the exact release behaviour can be changed with the
962 * [`G_DEBUG=gc-friendly`][G_DEBUG] environment variable, also see
963 * [`G_SLICE`][G_SLICE] for related debugging options.
964 *
965 * If @mem_chain is %NULL, this function does nothing.
966 *
967 * Since: 2.10
968 */
969  
970 /**
971 * g_slice_alloc:
972 * @block_size: the number of bytes to allocate
973 *
974 * Allocates a block of memory from the slice allocator.
975 * The block adress handed out can be expected to be aligned
976 * to at least 1 * sizeof (void*),
977 * though in general slices are 2 * sizeof (void*) bytes aligned,
978 * if a malloc() fallback implementation is used instead,
979 * the alignment may be reduced in a libc dependent fashion.
980 * Note that the underlying slice allocation mechanism can
981 * be changed with the [`G_SLICE=always-malloc`][G_SLICE]
982 * environment variable.
983 *
984 * Returns: a pointer to the allocated memory block, which will be %NULL if and
985 * only if @mem_size is 0
986 *
987 * Since: 2.10
988 */
989 gpointer
990 g_slice_alloc (gsize mem_size)
991 {
992 ThreadMemory *tmem;
993 gsize chunk_size;
994 gpointer mem;
995 guint acat;
996  
997 /* This gets the private structure for this thread. If the private
998 * structure does not yet exist, it is created.
999 *
1000 * This has a side effect of causing GSlice to be initialised, so it
1001 * must come first.
1002 */
1003 tmem = thread_memory_from_self ();
1004  
1005 chunk_size = P2ALIGN (mem_size);
1006 acat = allocator_categorize (chunk_size);
1007 if (G_LIKELY (acat == 1)) /* allocate through magazine layer */
1008 {
1009 guint ix = SLAB_INDEX (allocator, chunk_size);
1010 if (G_UNLIKELY (thread_memory_magazine1_is_empty (tmem, ix)))
1011 {
1012 thread_memory_swap_magazines (tmem, ix);
1013 if (G_UNLIKELY (thread_memory_magazine1_is_empty (tmem, ix)))
1014 thread_memory_magazine1_reload (tmem, ix);
1015 }
1016 mem = thread_memory_magazine1_alloc (tmem, ix);
1017 }
1018 else if (acat == 2) /* allocate through slab allocator */
1019 {
1020 g_mutex_lock (&allocator->slab_mutex);
1021 mem = slab_allocator_alloc_chunk (chunk_size);
1022 g_mutex_unlock (&allocator->slab_mutex);
1023 }
1024 else /* delegate to system malloc */
1025 mem = g_malloc (mem_size);
1026 if (G_UNLIKELY (allocator->config.debug_blocks))
1027 smc_notify_alloc (mem, mem_size);
1028  
1029 TRACE (GLIB_SLICE_ALLOC((void*)mem, mem_size));
1030  
1031 return mem;
1032 }
1033  
1034 /**
1035 * g_slice_alloc0:
1036 * @block_size: the number of bytes to allocate
1037 *
1038 * Allocates a block of memory via g_slice_alloc() and initializes
1039 * the returned memory to 0. Note that the underlying slice allocation
1040 * mechanism can be changed with the [`G_SLICE=always-malloc`][G_SLICE]
1041 * environment variable.
1042 *
1043 * Returns: a pointer to the allocated block, which will be %NULL if and only
1044 * if @mem_size is 0
1045 *
1046 * Since: 2.10
1047 */
1048 gpointer
1049 g_slice_alloc0 (gsize mem_size)
1050 {
1051 gpointer mem = g_slice_alloc (mem_size);
1052 if (mem)
1053 memset (mem, 0, mem_size);
1054 return mem;
1055 }
1056  
1057 /**
1058 * g_slice_copy:
1059 * @block_size: the number of bytes to allocate
1060 * @mem_block: the memory to copy
1061 *
1062 * Allocates a block of memory from the slice allocator
1063 * and copies @block_size bytes into it from @mem_block.
1064 *
1065 * @mem_block must be non-%NULL if @block_size is non-zero.
1066 *
1067 * Returns: a pointer to the allocated memory block, which will be %NULL if and
1068 * only if @mem_size is 0
1069 *
1070 * Since: 2.14
1071 */
1072 gpointer
1073 g_slice_copy (gsize mem_size,
1074 gconstpointer mem_block)
1075 {
1076 gpointer mem = g_slice_alloc (mem_size);
1077 if (mem)
1078 memcpy (mem, mem_block, mem_size);
1079 return mem;
1080 }
1081  
1082 /**
1083 * g_slice_free1:
1084 * @block_size: the size of the block
1085 * @mem_block: a pointer to the block to free
1086 *
1087 * Frees a block of memory.
1088 *
1089 * The memory must have been allocated via g_slice_alloc() or
1090 * g_slice_alloc0() and the @block_size has to match the size
1091 * specified upon allocation. Note that the exact release behaviour
1092 * can be changed with the [`G_DEBUG=gc-friendly`][G_DEBUG] environment
1093 * variable, also see [`G_SLICE`][G_SLICE] for related debugging options.
1094 *
1095 * If @mem_block is %NULL, this function does nothing.
1096 *
1097 * Since: 2.10
1098 */
1099 void
1100 g_slice_free1 (gsize mem_size,
1101 gpointer mem_block)
1102 {
1103 gsize chunk_size = P2ALIGN (mem_size);
1104 guint acat = allocator_categorize (chunk_size);
1105 if (G_UNLIKELY (!mem_block))
1106 return;
1107 if (G_UNLIKELY (allocator->config.debug_blocks) &&
1108 !smc_notify_free (mem_block, mem_size))
1109 abort();
1110 if (G_LIKELY (acat == 1)) /* allocate through magazine layer */
1111 {
1112 ThreadMemory *tmem = thread_memory_from_self();
1113 guint ix = SLAB_INDEX (allocator, chunk_size);
1114 if (G_UNLIKELY (thread_memory_magazine2_is_full (tmem, ix)))
1115 {
1116 thread_memory_swap_magazines (tmem, ix);
1117 if (G_UNLIKELY (thread_memory_magazine2_is_full (tmem, ix)))
1118 thread_memory_magazine2_unload (tmem, ix);
1119 }
1120 if (G_UNLIKELY (g_mem_gc_friendly))
1121 memset (mem_block, 0, chunk_size);
1122 thread_memory_magazine2_free (tmem, ix, mem_block);
1123 }
1124 else if (acat == 2) /* allocate through slab allocator */
1125 {
1126 if (G_UNLIKELY (g_mem_gc_friendly))
1127 memset (mem_block, 0, chunk_size);
1128 g_mutex_lock (&allocator->slab_mutex);
1129 slab_allocator_free_chunk (chunk_size, mem_block);
1130 g_mutex_unlock (&allocator->slab_mutex);
1131 }
1132 else /* delegate to system malloc */
1133 {
1134 if (G_UNLIKELY (g_mem_gc_friendly))
1135 memset (mem_block, 0, mem_size);
1136 g_free (mem_block);
1137 }
1138 TRACE (GLIB_SLICE_FREE((void*)mem_block, mem_size));
1139 }
1140  
1141 /**
1142 * g_slice_free_chain_with_offset:
1143 * @block_size: the size of the blocks
1144 * @mem_chain: a pointer to the first block of the chain
1145 * @next_offset: the offset of the @next field in the blocks
1146 *
1147 * Frees a linked list of memory blocks of structure type @type.
1148 *
1149 * The memory blocks must be equal-sized, allocated via
1150 * g_slice_alloc() or g_slice_alloc0() and linked together by a
1151 * @next pointer (similar to #GSList). The offset of the @next
1152 * field in each block is passed as third argument.
1153 * Note that the exact release behaviour can be changed with the
1154 * [`G_DEBUG=gc-friendly`][G_DEBUG] environment variable, also see
1155 * [`G_SLICE`][G_SLICE] for related debugging options.
1156 *
1157 * If @mem_chain is %NULL, this function does nothing.
1158 *
1159 * Since: 2.10
1160 */
1161 void
1162 g_slice_free_chain_with_offset (gsize mem_size,
1163 gpointer mem_chain,
1164 gsize next_offset)
1165 {
1166 gpointer slice = mem_chain;
1167 /* while the thread magazines and the magazine cache are implemented so that
1168 * they can easily be extended to allow for free lists containing more free
1169 * lists for the first level nodes, which would allow O(1) freeing in this
1170 * function, the benefit of such an extension is questionable, because:
1171 * - the magazine size counts will become mere lower bounds which confuses
1172 * the code adapting to lock contention;
1173 * - freeing a single node to the thread magazines is very fast, so this
1174 * O(list_length) operation is multiplied by a fairly small factor;
1175 * - memory usage histograms on larger applications seem to indicate that
1176 * the amount of released multi node lists is negligible in comparison
1177 * to single node releases.
1178 * - the major performance bottle neck, namely g_private_get() or
1179 * g_mutex_lock()/g_mutex_unlock() has already been moved out of the
1180 * inner loop for freeing chained slices.
1181 */
1182 gsize chunk_size = P2ALIGN (mem_size);
1183 guint acat = allocator_categorize (chunk_size);
1184 if (G_LIKELY (acat == 1)) /* allocate through magazine layer */
1185 {
1186 ThreadMemory *tmem = thread_memory_from_self();
1187 guint ix = SLAB_INDEX (allocator, chunk_size);
1188 while (slice)
1189 {
1190 guint8 *current = slice;
1191 slice = *(gpointer*) (current + next_offset);
1192 if (G_UNLIKELY (allocator->config.debug_blocks) &&
1193 !smc_notify_free (current, mem_size))
1194 abort();
1195 if (G_UNLIKELY (thread_memory_magazine2_is_full (tmem, ix)))
1196 {
1197 thread_memory_swap_magazines (tmem, ix);
1198 if (G_UNLIKELY (thread_memory_magazine2_is_full (tmem, ix)))
1199 thread_memory_magazine2_unload (tmem, ix);
1200 }
1201 if (G_UNLIKELY (g_mem_gc_friendly))
1202 memset (current, 0, chunk_size);
1203 thread_memory_magazine2_free (tmem, ix, current);
1204 }
1205 }
1206 else if (acat == 2) /* allocate through slab allocator */
1207 {
1208 g_mutex_lock (&allocator->slab_mutex);
1209 while (slice)
1210 {
1211 guint8 *current = slice;
1212 slice = *(gpointer*) (current + next_offset);
1213 if (G_UNLIKELY (allocator->config.debug_blocks) &&
1214 !smc_notify_free (current, mem_size))
1215 abort();
1216 if (G_UNLIKELY (g_mem_gc_friendly))
1217 memset (current, 0, chunk_size);
1218 slab_allocator_free_chunk (chunk_size, current);
1219 }
1220 g_mutex_unlock (&allocator->slab_mutex);
1221 }
1222 else /* delegate to system malloc */
1223 while (slice)
1224 {
1225 guint8 *current = slice;
1226 slice = *(gpointer*) (current + next_offset);
1227 if (G_UNLIKELY (allocator->config.debug_blocks) &&
1228 !smc_notify_free (current, mem_size))
1229 abort();
1230 if (G_UNLIKELY (g_mem_gc_friendly))
1231 memset (current, 0, mem_size);
1232 g_free (current);
1233 }
1234 }
1235  
1236 /* --- single page allocator --- */
1237 static void
1238 allocator_slab_stack_push (Allocator *allocator,
1239 guint ix,
1240 SlabInfo *sinfo)
1241 {
1242 /* insert slab at slab ring head */
1243 if (!allocator->slab_stack[ix])
1244 {
1245 sinfo->next = sinfo;
1246 sinfo->prev = sinfo;
1247 }
1248 else
1249 {
1250 SlabInfo *next = allocator->slab_stack[ix], *prev = next->prev;
1251 next->prev = sinfo;
1252 prev->next = sinfo;
1253 sinfo->next = next;
1254 sinfo->prev = prev;
1255 }
1256 allocator->slab_stack[ix] = sinfo;
1257 }
1258  
1259 static gsize
1260 allocator_aligned_page_size (Allocator *allocator,
1261 gsize n_bytes)
1262 {
1263 gsize val = 1 << g_bit_storage (n_bytes - 1);
1264 val = MAX (val, allocator->min_page_size);
1265 return val;
1266 }
1267  
1268 static void
1269 allocator_add_slab (Allocator *allocator,
1270 guint ix,
1271 gsize chunk_size)
1272 {
1273 ChunkLink *chunk;
1274 SlabInfo *sinfo;
1275 gsize addr, padding, n_chunks, color = 0;
1276 gsize page_size = allocator_aligned_page_size (allocator, SLAB_BPAGE_SIZE (allocator, chunk_size));
1277 /* allocate 1 page for the chunks and the slab */
1278 gpointer aligned_memory = allocator_memalign (page_size, page_size - NATIVE_MALLOC_PADDING);
1279 guint8 *mem = aligned_memory;
1280 guint i;
1281 if (!mem)
1282 {
1283 const gchar *syserr = strerror (errno);
1284 mem_error ("failed to allocate %u bytes (alignment: %u): %s\n",
1285 (guint) (page_size - NATIVE_MALLOC_PADDING), (guint) page_size, syserr);
1286 }
1287 /* mask page address */
1288 addr = ((gsize) mem / page_size) * page_size;
1289 /* assert alignment */
1290 mem_assert (aligned_memory == (gpointer) addr);
1291 /* basic slab info setup */
1292 sinfo = (SlabInfo*) (mem + page_size - SLAB_INFO_SIZE);
1293 sinfo->n_allocated = 0;
1294 sinfo->chunks = NULL;
1295 /* figure cache colorization */
1296 n_chunks = ((guint8*) sinfo - mem) / chunk_size;
1297 padding = ((guint8*) sinfo - mem) - n_chunks * chunk_size;
1298 if (padding)
1299 {
1300 color = (allocator->color_accu * P2ALIGNMENT) % padding;
1301 allocator->color_accu += allocator->config.color_increment;
1302 }
1303 /* add chunks to free list */
1304 chunk = (ChunkLink*) (mem + color);
1305 sinfo->chunks = chunk;
1306 for (i = 0; i < n_chunks - 1; i++)
1307 {
1308 chunk->next = (ChunkLink*) ((guint8*) chunk + chunk_size);
1309 chunk = chunk->next;
1310 }
1311 chunk->next = NULL; /* last chunk */
1312 /* add slab to slab ring */
1313 allocator_slab_stack_push (allocator, ix, sinfo);
1314 }
1315  
1316 static gpointer
1317 slab_allocator_alloc_chunk (gsize chunk_size)
1318 {
1319 ChunkLink *chunk;
1320 guint ix = SLAB_INDEX (allocator, chunk_size);
1321 /* ensure non-empty slab */
1322 if (!allocator->slab_stack[ix] || !allocator->slab_stack[ix]->chunks)
1323 allocator_add_slab (allocator, ix, chunk_size);
1324 /* allocate chunk */
1325 chunk = allocator->slab_stack[ix]->chunks;
1326 allocator->slab_stack[ix]->chunks = chunk->next;
1327 allocator->slab_stack[ix]->n_allocated++;
1328 /* rotate empty slabs */
1329 if (!allocator->slab_stack[ix]->chunks)
1330 allocator->slab_stack[ix] = allocator->slab_stack[ix]->next;
1331 return chunk;
1332 }
1333  
1334 static void
1335 slab_allocator_free_chunk (gsize chunk_size,
1336 gpointer mem)
1337 {
1338 ChunkLink *chunk;
1339 gboolean was_empty;
1340 guint ix = SLAB_INDEX (allocator, chunk_size);
1341 gsize page_size = allocator_aligned_page_size (allocator, SLAB_BPAGE_SIZE (allocator, chunk_size));
1342 gsize addr = ((gsize) mem / page_size) * page_size;
1343 /* mask page address */
1344 guint8 *page = (guint8*) addr;
1345 SlabInfo *sinfo = (SlabInfo*) (page + page_size - SLAB_INFO_SIZE);
1346 /* assert valid chunk count */
1347 mem_assert (sinfo->n_allocated > 0);
1348 /* add chunk to free list */
1349 was_empty = sinfo->chunks == NULL;
1350 chunk = (ChunkLink*) mem;
1351 chunk->next = sinfo->chunks;
1352 sinfo->chunks = chunk;
1353 sinfo->n_allocated--;
1354 /* keep slab ring partially sorted, empty slabs at end */
1355 if (was_empty)
1356 {
1357 /* unlink slab */
1358 SlabInfo *next = sinfo->next, *prev = sinfo->prev;
1359 next->prev = prev;
1360 prev->next = next;
1361 if (allocator->slab_stack[ix] == sinfo)
1362 allocator->slab_stack[ix] = next == sinfo ? NULL : next;
1363 /* insert slab at head */
1364 allocator_slab_stack_push (allocator, ix, sinfo);
1365 }
1366 /* eagerly free complete unused slabs */
1367 if (!sinfo->n_allocated)
1368 {
1369 /* unlink slab */
1370 SlabInfo *next = sinfo->next, *prev = sinfo->prev;
1371 next->prev = prev;
1372 prev->next = next;
1373 if (allocator->slab_stack[ix] == sinfo)
1374 allocator->slab_stack[ix] = next == sinfo ? NULL : next;
1375 /* free slab */
1376 allocator_memfree (page_size, page);
1377 }
1378 }
1379  
1380 /* --- memalign implementation --- */
1381 #ifdef HAVE_MALLOC_H
1382 #include <malloc.h> /* memalign() */
1383 #endif
1384  
1385 /* from config.h:
1386 * define HAVE_POSIX_MEMALIGN 1 // if free(posix_memalign(3)) works, <stdlib.h>
1387 * define HAVE_COMPLIANT_POSIX_MEMALIGN 1 // if free(posix_memalign(3)) works for sizes != 2^n, <stdlib.h>
1388 * define HAVE_MEMALIGN 1 // if free(memalign(3)) works, <malloc.h>
1389 * define HAVE_VALLOC 1 // if free(valloc(3)) works, <stdlib.h> or <malloc.h>
1390 * if none is provided, we implement malloc(3)-based alloc-only page alignment
1391 */
1392  
1393 #if !(HAVE_COMPLIANT_POSIX_MEMALIGN || HAVE_MEMALIGN || HAVE_VALLOC)
1394 static GTrashStack *compat_valloc_trash = NULL;
1395 #endif
1396  
1397 static gpointer
1398 allocator_memalign (gsize alignment,
1399 gsize memsize)
1400 {
1401 gpointer aligned_memory = NULL;
1402 gint err = ENOMEM;
1403 #if HAVE_COMPLIANT_POSIX_MEMALIGN
1404 err = posix_memalign (&aligned_memory, alignment, memsize);
1405 #elif HAVE_MEMALIGN
1406 errno = 0;
1407 aligned_memory = memalign (alignment, memsize);
1408 err = errno;
1409 #elif HAVE_VALLOC
1410 errno = 0;
1411 aligned_memory = valloc (memsize);
1412 err = errno;
1413 #else
1414 /* simplistic non-freeing page allocator */
1415 mem_assert (alignment == sys_page_size);
1416 mem_assert (memsize <= sys_page_size);
1417 if (!compat_valloc_trash)
1418 {
1419 const guint n_pages = 16;
1420 guint8 *mem = malloc (n_pages * sys_page_size);
1421 err = errno;
1422 if (mem)
1423 {
1424 gint i = n_pages;
1425 guint8 *amem = (guint8*) ALIGN ((gsize) mem, sys_page_size);
1426 if (amem != mem)
1427 i--; /* mem wasn't page aligned */
1428 while (--i >= 0)
1429 g_trash_stack_push (&compat_valloc_trash, amem + i * sys_page_size);
1430 }
1431 }
1432 aligned_memory = g_trash_stack_pop (&compat_valloc_trash);
1433 #endif
1434 if (!aligned_memory)
1435 errno = err;
1436 return aligned_memory;
1437 }
1438  
1439 static void
1440 allocator_memfree (gsize memsize,
1441 gpointer mem)
1442 {
1443 #if HAVE_COMPLIANT_POSIX_MEMALIGN || HAVE_MEMALIGN || HAVE_VALLOC
1444 free (mem);
1445 #else
1446 mem_assert (memsize <= sys_page_size);
1447 g_trash_stack_push (&compat_valloc_trash, mem);
1448 #endif
1449 }
1450  
1451 static void
1452 mem_error (const char *format,
1453 ...)
1454 {
1455 const char *pname;
1456 va_list args;
1457 /* at least, put out "MEMORY-ERROR", in case we segfault during the rest of the function */
1458 fputs ("\n***MEMORY-ERROR***: ", stderr);
1459 pname = g_get_prgname();
1460 fprintf (stderr, "%s[%ld]: GSlice: ", pname ? pname : "", (long)getpid());
1461 va_start (args, format);
1462 vfprintf (stderr, format, args);
1463 va_end (args);
1464 fputs ("\n", stderr);
1465 abort();
1466 _exit (1);
1467 }
1468  
1469 /* --- g-slice memory checker tree --- */
1470 typedef size_t SmcKType; /* key type */
1471 typedef size_t SmcVType; /* value type */
1472 typedef struct {
1473 SmcKType key;
1474 SmcVType value;
1475 } SmcEntry;
1476 static void smc_tree_insert (SmcKType key,
1477 SmcVType value);
1478 static gboolean smc_tree_lookup (SmcKType key,
1479 SmcVType *value_p);
1480 static gboolean smc_tree_remove (SmcKType key);
1481  
1482  
1483 /* --- g-slice memory checker implementation --- */
1484 static void
1485 smc_notify_alloc (void *pointer,
1486 size_t size)
1487 {
1488 size_t adress = (size_t) pointer;
1489 if (pointer)
1490 smc_tree_insert (adress, size);
1491 }
1492  
1493 #if 0
1494 static void
1495 smc_notify_ignore (void *pointer)
1496 {
1497 size_t adress = (size_t) pointer;
1498 if (pointer)
1499 smc_tree_remove (adress);
1500 }
1501 #endif
1502  
1503 static int
1504 smc_notify_free (void *pointer,
1505 size_t size)
1506 {
1507 size_t adress = (size_t) pointer;
1508 SmcVType real_size;
1509 gboolean found_one;
1510  
1511 if (!pointer)
1512 return 1; /* ignore */
1513 found_one = smc_tree_lookup (adress, &real_size);
1514 if (!found_one)
1515 {
1516 fprintf (stderr, "GSlice: MemChecker: attempt to release non-allocated block: %p size=%" G_GSIZE_FORMAT "\n", pointer, size);
1517 return 0;
1518 }
1519 if (real_size != size && (real_size || size))
1520 {
1521 fprintf (stderr, "GSlice: MemChecker: attempt to release block with invalid size: %p size=%" G_GSIZE_FORMAT " invalid-size=%" G_GSIZE_FORMAT "\n", pointer, real_size, size);
1522 return 0;
1523 }
1524 if (!smc_tree_remove (adress))
1525 {
1526 fprintf (stderr, "GSlice: MemChecker: attempt to release non-allocated block: %p size=%" G_GSIZE_FORMAT "\n", pointer, size);
1527 return 0;
1528 }
1529 return 1; /* all fine */
1530 }
1531  
1532 /* --- g-slice memory checker tree implementation --- */
1533 #define SMC_TRUNK_COUNT (4093 /* 16381 */) /* prime, to distribute trunk collisions (big, allocated just once) */
1534 #define SMC_BRANCH_COUNT (511) /* prime, to distribute branch collisions */
1535 #define SMC_TRUNK_EXTENT (SMC_BRANCH_COUNT * 2039) /* key address space per trunk, should distribute uniformly across BRANCH_COUNT */
1536 #define SMC_TRUNK_HASH(k) ((k / SMC_TRUNK_EXTENT) % SMC_TRUNK_COUNT) /* generate new trunk hash per megabyte (roughly) */
1537 #define SMC_BRANCH_HASH(k) (k % SMC_BRANCH_COUNT)
1538  
1539 typedef struct {
1540 SmcEntry *entries;
1541 unsigned int n_entries;
1542 } SmcBranch;
1543  
1544 static SmcBranch **smc_tree_root = NULL;
1545  
1546 static void
1547 smc_tree_abort (int errval)
1548 {
1549 const char *syserr = strerror (errval);
1550 mem_error ("MemChecker: failure in debugging tree: %s", syserr);
1551 }
1552  
1553 static inline SmcEntry*
1554 smc_tree_branch_grow_L (SmcBranch *branch,
1555 unsigned int index)
1556 {
1557 unsigned int old_size = branch->n_entries * sizeof (branch->entries[0]);
1558 unsigned int new_size = old_size + sizeof (branch->entries[0]);
1559 SmcEntry *entry;
1560 mem_assert (index <= branch->n_entries);
1561 branch->entries = (SmcEntry*) realloc (branch->entries, new_size);
1562 if (!branch->entries)
1563 smc_tree_abort (errno);
1564 entry = branch->entries + index;
1565 memmove (entry + 1, entry, (branch->n_entries - index) * sizeof (entry[0]));
1566 branch->n_entries += 1;
1567 return entry;
1568 }
1569  
1570 static inline SmcEntry*
1571 smc_tree_branch_lookup_nearest_L (SmcBranch *branch,
1572 SmcKType key)
1573 {
1574 unsigned int n_nodes = branch->n_entries, offs = 0;
1575 SmcEntry *check = branch->entries;
1576 int cmp = 0;
1577 while (offs < n_nodes)
1578 {
1579 unsigned int i = (offs + n_nodes) >> 1;
1580 check = branch->entries + i;
1581 cmp = key < check->key ? -1 : key != check->key;
1582 if (cmp == 0)
1583 return check; /* return exact match */
1584 else if (cmp < 0)
1585 n_nodes = i;
1586 else /* (cmp > 0) */
1587 offs = i + 1;
1588 }
1589 /* check points at last mismatch, cmp > 0 indicates greater key */
1590 return cmp > 0 ? check + 1 : check; /* return insertion position for inexact match */
1591 }
1592  
1593 static void
1594 smc_tree_insert (SmcKType key,
1595 SmcVType value)
1596 {
1597 unsigned int ix0, ix1;
1598 SmcEntry *entry;
1599  
1600 g_mutex_lock (&smc_tree_mutex);
1601 ix0 = SMC_TRUNK_HASH (key);
1602 ix1 = SMC_BRANCH_HASH (key);
1603 if (!smc_tree_root)
1604 {
1605 smc_tree_root = calloc (SMC_TRUNK_COUNT, sizeof (smc_tree_root[0]));
1606 if (!smc_tree_root)
1607 smc_tree_abort (errno);
1608 }
1609 if (!smc_tree_root[ix0])
1610 {
1611 smc_tree_root[ix0] = calloc (SMC_BRANCH_COUNT, sizeof (smc_tree_root[0][0]));
1612 if (!smc_tree_root[ix0])
1613 smc_tree_abort (errno);
1614 }
1615 entry = smc_tree_branch_lookup_nearest_L (&smc_tree_root[ix0][ix1], key);
1616 if (!entry || /* need create */
1617 entry >= smc_tree_root[ix0][ix1].entries + smc_tree_root[ix0][ix1].n_entries || /* need append */
1618 entry->key != key) /* need insert */
1619 entry = smc_tree_branch_grow_L (&smc_tree_root[ix0][ix1], entry - smc_tree_root[ix0][ix1].entries);
1620 entry->key = key;
1621 entry->value = value;
1622 g_mutex_unlock (&smc_tree_mutex);
1623 }
1624  
1625 static gboolean
1626 smc_tree_lookup (SmcKType key,
1627 SmcVType *value_p)
1628 {
1629 SmcEntry *entry = NULL;
1630 unsigned int ix0 = SMC_TRUNK_HASH (key), ix1 = SMC_BRANCH_HASH (key);
1631 gboolean found_one = FALSE;
1632 *value_p = 0;
1633 g_mutex_lock (&smc_tree_mutex);
1634 if (smc_tree_root && smc_tree_root[ix0])
1635 {
1636 entry = smc_tree_branch_lookup_nearest_L (&smc_tree_root[ix0][ix1], key);
1637 if (entry &&
1638 entry < smc_tree_root[ix0][ix1].entries + smc_tree_root[ix0][ix1].n_entries &&
1639 entry->key == key)
1640 {
1641 found_one = TRUE;
1642 *value_p = entry->value;
1643 }
1644 }
1645 g_mutex_unlock (&smc_tree_mutex);
1646 return found_one;
1647 }
1648  
1649 static gboolean
1650 smc_tree_remove (SmcKType key)
1651 {
1652 unsigned int ix0 = SMC_TRUNK_HASH (key), ix1 = SMC_BRANCH_HASH (key);
1653 gboolean found_one = FALSE;
1654 g_mutex_lock (&smc_tree_mutex);
1655 if (smc_tree_root && smc_tree_root[ix0])
1656 {
1657 SmcEntry *entry = smc_tree_branch_lookup_nearest_L (&smc_tree_root[ix0][ix1], key);
1658 if (entry &&
1659 entry < smc_tree_root[ix0][ix1].entries + smc_tree_root[ix0][ix1].n_entries &&
1660 entry->key == key)
1661 {
1662 unsigned int i = entry - smc_tree_root[ix0][ix1].entries;
1663 smc_tree_root[ix0][ix1].n_entries -= 1;
1664 memmove (entry, entry + 1, (smc_tree_root[ix0][ix1].n_entries - i) * sizeof (entry[0]));
1665 if (!smc_tree_root[ix0][ix1].n_entries)
1666 {
1667 /* avoid useless pressure on the memory system */
1668 free (smc_tree_root[ix0][ix1].entries);
1669 smc_tree_root[ix0][ix1].entries = NULL;
1670 }
1671 found_one = TRUE;
1672 }
1673 }
1674 g_mutex_unlock (&smc_tree_mutex);
1675 return found_one;
1676 }
1677  
1678 #ifdef G_ENABLE_DEBUG
1679 void
1680 g_slice_debug_tree_statistics (void)
1681 {
1682 g_mutex_lock (&smc_tree_mutex);
1683 if (smc_tree_root)
1684 {
1685 unsigned int i, j, t = 0, o = 0, b = 0, su = 0, ex = 0, en = 4294967295u;
1686 double tf, bf;
1687 for (i = 0; i < SMC_TRUNK_COUNT; i++)
1688 if (smc_tree_root[i])
1689 {
1690 t++;
1691 for (j = 0; j < SMC_BRANCH_COUNT; j++)
1692 if (smc_tree_root[i][j].n_entries)
1693 {
1694 b++;
1695 su += smc_tree_root[i][j].n_entries;
1696 en = MIN (en, smc_tree_root[i][j].n_entries);
1697 ex = MAX (ex, smc_tree_root[i][j].n_entries);
1698 }
1699 else if (smc_tree_root[i][j].entries)
1700 o++; /* formerly used, now empty */
1701 }
1702 en = b ? en : 0;
1703 tf = MAX (t, 1.0); /* max(1) to be a valid divisor */
1704 bf = MAX (b, 1.0); /* max(1) to be a valid divisor */
1705 fprintf (stderr, "GSlice: MemChecker: %u trunks, %u branches, %u old branches\n", t, b, o);
1706 fprintf (stderr, "GSlice: MemChecker: %f branches per trunk, %.2f%% utilization\n",
1707 b / tf,
1708 100.0 - (SMC_BRANCH_COUNT - b / tf) / (0.01 * SMC_BRANCH_COUNT));
1709 fprintf (stderr, "GSlice: MemChecker: %f entries per branch, %u minimum, %u maximum\n",
1710 su / bf, en, ex);
1711 }
1712 else
1713 fprintf (stderr, "GSlice: MemChecker: root=NULL\n");
1714 g_mutex_unlock (&smc_tree_mutex);
1715  
1716 /* sample statistics (beast + GSLice + 24h scripted core & GUI activity):
1717 * PID %CPU %MEM VSZ RSS COMMAND
1718 * 8887 30.3 45.8 456068 414856 beast-0.7.1 empty.bse
1719 * $ cat /proc/8887/statm # total-program-size resident-set-size shared-pages text/code data/stack library dirty-pages
1720 * 114017 103714 2354 344 0 108676 0
1721 * $ cat /proc/8887/status
1722 * Name: beast-0.7.1
1723 * VmSize: 456068 kB
1724 * VmLck: 0 kB
1725 * VmRSS: 414856 kB
1726 * VmData: 434620 kB
1727 * VmStk: 84 kB
1728 * VmExe: 1376 kB
1729 * VmLib: 13036 kB
1730 * VmPTE: 456 kB
1731 * Threads: 3
1732 * (gdb) print g_slice_debug_tree_statistics ()
1733 * GSlice: MemChecker: 422 trunks, 213068 branches, 0 old branches
1734 * GSlice: MemChecker: 504.900474 branches per trunk, 98.81% utilization
1735 * GSlice: MemChecker: 4.965039 entries per branch, 1 minimum, 37 maximum
1736 */
1737 }
1738 #endif /* G_ENABLE_DEBUG */