1 /******************************************************************************/ 2 #ifdef JEMALLOC_H_TYPES 3 4 #define LARGE_MINCLASS (ZU(1) << LG_LARGE_MINCLASS) 5 6 /* Maximum number of regions in one run. */ 7 #define LG_RUN_MAXREGS (LG_PAGE - LG_TINY_MIN) 8 #define RUN_MAXREGS (1U << LG_RUN_MAXREGS) 9 10 /* 11 * Minimum redzone size. Redzones may be larger than this if necessary to 12 * preserve region alignment. 13 */ 14 #define REDZONE_MINSIZE 16 15 16 /* 17 * The minimum ratio of active:dirty pages per arena is computed as: 18 * 19 * (nactive >> lg_dirty_mult) >= ndirty 20 * 21 * So, supposing that lg_dirty_mult is 3, there can be no less than 8 times as 22 * many active pages as dirty pages. 23 */ 24 #define LG_DIRTY_MULT_DEFAULT 3 25 26 typedef enum { 27 purge_mode_ratio = 0, 28 purge_mode_decay = 1, 29 30 purge_mode_limit = 2 31 } purge_mode_t; 32 /* ANDROID change */ 33 /* Use the decay mode purge method. 34 * Setting this value to zero results in performance issues because it 35 * causes purges at every free. Leave the default at zero, but zygote 36 * processes will set this to one using mallopt. This allows apps which 37 * tend to be active to benefit from the extra performance, but allow system 38 * servers to free PSS while they are sitting idle. 39 */ 40 #define PURGE_DEFAULT purge_mode_decay 41 /* Default decay time in seconds. */ 42 #define DECAY_TIME_DEFAULT 0 43 /* End ANDROID change */ 44 /* Number of event ticks between time checks. */ 45 #define DECAY_NTICKS_PER_UPDATE 1000 46 47 typedef struct arena_runs_dirty_link_s arena_runs_dirty_link_t; 48 typedef struct arena_avail_links_s arena_avail_links_t; 49 typedef struct arena_run_s arena_run_t; 50 typedef struct arena_chunk_map_bits_s arena_chunk_map_bits_t; 51 typedef struct arena_chunk_map_misc_s arena_chunk_map_misc_t; 52 typedef struct arena_chunk_s arena_chunk_t; 53 typedef struct arena_bin_info_s arena_bin_info_t; 54 typedef struct arena_decay_s arena_decay_t; 55 typedef struct arena_bin_s arena_bin_t; 56 typedef struct arena_s arena_t; 57 typedef struct arena_tdata_s arena_tdata_t; 58 59 #endif /* JEMALLOC_H_TYPES */ 60 /******************************************************************************/ 61 #ifdef JEMALLOC_H_STRUCTS 62 63 #ifdef JEMALLOC_ARENA_STRUCTS_A 64 struct arena_run_s { 65 /* Index of bin this run is associated with. */ 66 szind_t binind; 67 68 /* Number of free regions in run. */ 69 unsigned nfree; 70 71 /* Per region allocated/deallocated bitmap. */ 72 bitmap_t bitmap[BITMAP_GROUPS_MAX]; 73 }; 74 75 /* Each element of the chunk map corresponds to one page within the chunk. */ 76 struct arena_chunk_map_bits_s { 77 /* 78 * Run address (or size) and various flags are stored together. The bit 79 * layout looks like (assuming 32-bit system): 80 * 81 * ???????? ???????? ???nnnnn nnndumla 82 * 83 * ? : Unallocated: Run address for first/last pages, unset for internal 84 * pages. 85 * Small: Run page offset. 86 * Large: Run page count for first page, unset for trailing pages. 87 * n : binind for small size class, BININD_INVALID for large size class. 88 * d : dirty? 89 * u : unzeroed? 90 * m : decommitted? 91 * l : large? 92 * a : allocated? 93 * 94 * Following are example bit patterns for the three types of runs. 95 * 96 * p : run page offset 97 * s : run size 98 * n : binind for size class; large objects set these to BININD_INVALID 99 * x : don't care 100 * - : 0 101 * + : 1 102 * [DUMLA] : bit set 103 * [dumla] : bit unset 104 * 105 * Unallocated (clean): 106 * ssssssss ssssssss sss+++++ +++dum-a 107 * xxxxxxxx xxxxxxxx xxxxxxxx xxx-Uxxx 108 * ssssssss ssssssss sss+++++ +++dUm-a 109 * 110 * Unallocated (dirty): 111 * ssssssss ssssssss sss+++++ +++D-m-a 112 * xxxxxxxx xxxxxxxx xxxxxxxx xxxxxxxx 113 * ssssssss ssssssss sss+++++ +++D-m-a 114 * 115 * Small: 116 * pppppppp pppppppp pppnnnnn nnnd---A 117 * pppppppp pppppppp pppnnnnn nnn----A 118 * pppppppp pppppppp pppnnnnn nnnd---A 119 * 120 * Large: 121 * ssssssss ssssssss sss+++++ +++D--LA 122 * xxxxxxxx xxxxxxxx xxxxxxxx xxxxxxxx 123 * -------- -------- ---+++++ +++D--LA 124 * 125 * Large (sampled, size <= LARGE_MINCLASS): 126 * ssssssss ssssssss sssnnnnn nnnD--LA 127 * xxxxxxxx xxxxxxxx xxxxxxxx xxxxxxxx 128 * -------- -------- ---+++++ +++D--LA 129 * 130 * Large (not sampled, size == LARGE_MINCLASS): 131 * ssssssss ssssssss sss+++++ +++D--LA 132 * xxxxxxxx xxxxxxxx xxxxxxxx xxxxxxxx 133 * -------- -------- ---+++++ +++D--LA 134 */ 135 size_t bits; 136 #define CHUNK_MAP_ALLOCATED ((size_t)0x01U) 137 #define CHUNK_MAP_LARGE ((size_t)0x02U) 138 #define CHUNK_MAP_STATE_MASK ((size_t)0x3U) 139 140 #define CHUNK_MAP_DECOMMITTED ((size_t)0x04U) 141 #define CHUNK_MAP_UNZEROED ((size_t)0x08U) 142 #define CHUNK_MAP_DIRTY ((size_t)0x10U) 143 #define CHUNK_MAP_FLAGS_MASK ((size_t)0x1cU) 144 145 #define CHUNK_MAP_BININD_SHIFT 5 146 #define BININD_INVALID ((size_t)0xffU) 147 #define CHUNK_MAP_BININD_MASK (BININD_INVALID << CHUNK_MAP_BININD_SHIFT) 148 #define CHUNK_MAP_BININD_INVALID CHUNK_MAP_BININD_MASK 149 150 #define CHUNK_MAP_RUNIND_SHIFT (CHUNK_MAP_BININD_SHIFT + 8) 151 #define CHUNK_MAP_SIZE_SHIFT (CHUNK_MAP_RUNIND_SHIFT - LG_PAGE) 152 #define CHUNK_MAP_SIZE_MASK \ 153 (~(CHUNK_MAP_BININD_MASK | CHUNK_MAP_FLAGS_MASK | CHUNK_MAP_STATE_MASK)) 154 }; 155 156 struct arena_runs_dirty_link_s { 157 qr(arena_runs_dirty_link_t) rd_link; 158 }; 159 160 /* 161 * Each arena_chunk_map_misc_t corresponds to one page within the chunk, just 162 * like arena_chunk_map_bits_t. Two separate arrays are stored within each 163 * chunk header in order to improve cache locality. 164 */ 165 struct arena_chunk_map_misc_s { 166 /* 167 * Linkage for run heaps. There are two disjoint uses: 168 * 169 * 1) arena_t's runs_avail heaps. 170 * 2) arena_run_t conceptually uses this linkage for in-use non-full 171 * runs, rather than directly embedding linkage. 172 */ 173 phn(arena_chunk_map_misc_t) ph_link; 174 175 union { 176 /* Linkage for list of dirty runs. */ 177 arena_runs_dirty_link_t rd; 178 179 /* Profile counters, used for large object runs. */ 180 union { 181 void *prof_tctx_pun; 182 prof_tctx_t *prof_tctx; 183 }; 184 185 /* Small region run metadata. */ 186 arena_run_t run; 187 }; 188 }; 189 typedef ph(arena_chunk_map_misc_t) arena_run_heap_t; 190 #endif /* JEMALLOC_ARENA_STRUCTS_A */ 191 192 #ifdef JEMALLOC_ARENA_STRUCTS_B 193 /* Arena chunk header. */ 194 struct arena_chunk_s { 195 /* 196 * A pointer to the arena that owns the chunk is stored within the node. 197 * This field as a whole is used by chunks_rtree to support both 198 * ivsalloc() and core-based debugging. 199 */ 200 extent_node_t node; 201 202 /* 203 * True if memory could be backed by transparent huge pages. This is 204 * only directly relevant to Linux, since it is the only supported 205 * platform on which jemalloc interacts with explicit transparent huge 206 * page controls. 207 */ 208 bool hugepage; 209 210 /* 211 * Map of pages within chunk that keeps track of free/large/small. The 212 * first map_bias entries are omitted, since the chunk header does not 213 * need to be tracked in the map. This omission saves a header page 214 * for common chunk sizes (e.g. 4 MiB). 215 */ 216 arena_chunk_map_bits_t map_bits[1]; /* Dynamically sized. */ 217 }; 218 219 /* 220 * Read-only information associated with each element of arena_t's bins array 221 * is stored separately, partly to reduce memory usage (only one copy, rather 222 * than one per arena), but mainly to avoid false cacheline sharing. 223 * 224 * Each run has the following layout: 225 * 226 * /--------------------\ 227 * | pad? | 228 * |--------------------| 229 * | redzone | 230 * reg0_offset | region 0 | 231 * | redzone | 232 * |--------------------| \ 233 * | redzone | | 234 * | region 1 | > reg_interval 235 * | redzone | / 236 * |--------------------| 237 * | ... | 238 * | ... | 239 * | ... | 240 * |--------------------| 241 * | redzone | 242 * | region nregs-1 | 243 * | redzone | 244 * |--------------------| 245 * | alignment pad? | 246 * \--------------------/ 247 * 248 * reg_interval has at least the same minimum alignment as reg_size; this 249 * preserves the alignment constraint that sa2u() depends on. Alignment pad is 250 * either 0 or redzone_size; it is present only if needed to align reg0_offset. 251 */ 252 struct arena_bin_info_s { 253 /* Size of regions in a run for this bin's size class. */ 254 size_t reg_size; 255 256 /* Redzone size. */ 257 size_t redzone_size; 258 259 /* Interval between regions (reg_size + (redzone_size << 1)). */ 260 size_t reg_interval; 261 262 /* Total size of a run for this bin's size class. */ 263 size_t run_size; 264 265 /* Total number of regions in a run for this bin's size class. */ 266 uint32_t nregs; 267 268 /* 269 * Metadata used to manipulate bitmaps for runs associated with this 270 * bin. 271 */ 272 bitmap_info_t bitmap_info; 273 274 /* Offset of first region in a run for this bin's size class. */ 275 uint32_t reg0_offset; 276 }; 277 278 struct arena_decay_s { 279 /* 280 * Approximate time in seconds from the creation of a set of unused 281 * dirty pages until an equivalent set of unused dirty pages is purged 282 * and/or reused. 283 */ 284 ssize_t time; 285 /* time / SMOOTHSTEP_NSTEPS. */ 286 nstime_t interval; 287 /* 288 * Time at which the current decay interval logically started. We do 289 * not actually advance to a new epoch until sometime after it starts 290 * because of scheduling and computation delays, and it is even possible 291 * to completely skip epochs. In all cases, during epoch advancement we 292 * merge all relevant activity into the most recently recorded epoch. 293 */ 294 nstime_t epoch; 295 /* Deadline randomness generator. */ 296 uint64_t jitter_state; 297 /* 298 * Deadline for current epoch. This is the sum of interval and per 299 * epoch jitter which is a uniform random variable in [0..interval). 300 * Epochs always advance by precise multiples of interval, but we 301 * randomize the deadline to reduce the likelihood of arenas purging in 302 * lockstep. 303 */ 304 nstime_t deadline; 305 /* 306 * Number of dirty pages at beginning of current epoch. During epoch 307 * advancement we use the delta between arena->decay.ndirty and 308 * arena->ndirty to determine how many dirty pages, if any, were 309 * generated. 310 */ 311 size_t ndirty; 312 /* 313 * Trailing log of how many unused dirty pages were generated during 314 * each of the past SMOOTHSTEP_NSTEPS decay epochs, where the last 315 * element is the most recent epoch. Corresponding epoch times are 316 * relative to epoch. 317 */ 318 size_t backlog[SMOOTHSTEP_NSTEPS]; 319 }; 320 321 struct arena_bin_s { 322 /* 323 * All operations on runcur, runs, and stats require that lock be 324 * locked. Run allocation/deallocation are protected by the arena lock, 325 * which may be acquired while holding one or more bin locks, but not 326 * vise versa. 327 */ 328 malloc_mutex_t lock; 329 330 /* 331 * Current run being used to service allocations of this bin's size 332 * class. 333 */ 334 arena_run_t *runcur; 335 336 /* 337 * Heap of non-full runs. This heap is used when looking for an 338 * existing run when runcur is no longer usable. We choose the 339 * non-full run that is lowest in memory; this policy tends to keep 340 * objects packed well, and it can also help reduce the number of 341 * almost-empty chunks. 342 */ 343 arena_run_heap_t runs; 344 345 /* Bin statistics. */ 346 malloc_bin_stats_t stats; 347 }; 348 349 struct arena_s { 350 /* This arena's index within the arenas array. */ 351 unsigned ind; 352 353 /* 354 * Number of threads currently assigned to this arena, synchronized via 355 * atomic operations. Each thread has two distinct assignments, one for 356 * application-serving allocation, and the other for internal metadata 357 * allocation. Internal metadata must not be allocated from arenas 358 * created via the arenas.extend mallctl, because the arena.<i>.reset 359 * mallctl indiscriminately discards all allocations for the affected 360 * arena. 361 * 362 * 0: Application allocation. 363 * 1: Internal metadata allocation. 364 */ 365 unsigned nthreads[2]; 366 367 /* 368 * There are three classes of arena operations from a locking 369 * perspective: 370 * 1) Thread assignment (modifies nthreads) is synchronized via atomics. 371 * 2) Bin-related operations are protected by bin locks. 372 * 3) Chunk- and run-related operations are protected by this mutex. 373 */ 374 malloc_mutex_t lock; 375 376 arena_stats_t stats; 377 /* 378 * List of tcaches for extant threads associated with this arena. 379 * Stats from these are merged incrementally, and at exit if 380 * opt_stats_print is enabled. 381 */ 382 ql_head(tcache_t) tcache_ql; 383 384 uint64_t prof_accumbytes; 385 386 /* 387 * PRNG state for cache index randomization of large allocation base 388 * pointers. 389 */ 390 size_t offset_state; 391 392 dss_prec_t dss_prec; 393 394 /* Extant arena chunks. */ 395 ql_head(extent_node_t) achunks; 396 397 /* Extent serial number generator state. */ 398 size_t extent_sn_next; 399 400 /* 401 * In order to avoid rapid chunk allocation/deallocation when an arena 402 * oscillates right on the cusp of needing a new chunk, cache the most 403 * recently freed chunk. The spare is left in the arena's chunk trees 404 * until it is deleted. 405 * 406 * There is one spare chunk per arena, rather than one spare total, in 407 * order to avoid interactions between multiple threads that could make 408 * a single spare inadequate. 409 */ 410 arena_chunk_t *spare; 411 412 /* Minimum ratio (log base 2) of nactive:ndirty. */ 413 ssize_t lg_dirty_mult; 414 415 /* True if a thread is currently executing arena_purge_to_limit(). */ 416 bool purging; 417 418 /* Number of pages in active runs and huge regions. */ 419 size_t nactive; 420 421 /* 422 * Current count of pages within unused runs that are potentially 423 * dirty, and for which madvise(... MADV_DONTNEED) has not been called. 424 * By tracking this, we can institute a limit on how much dirty unused 425 * memory is mapped for each arena. 426 */ 427 size_t ndirty; 428 429 /* 430 * Unused dirty memory this arena manages. Dirty memory is conceptually 431 * tracked as an arbitrarily interleaved LRU of dirty runs and cached 432 * chunks, but the list linkage is actually semi-duplicated in order to 433 * avoid extra arena_chunk_map_misc_t space overhead. 434 * 435 * LRU-----------------------------------------------------------MRU 436 * 437 * /-- arena ---\ 438 * | | 439 * | | 440 * |------------| /- chunk -\ 441 * ...->|chunks_cache|<--------------------------->| /----\ |<--... 442 * |------------| | |node| | 443 * | | | | | | 444 * | | /- run -\ /- run -\ | | | | 445 * | | | | | | | | | | 446 * | | | | | | | | | | 447 * |------------| |-------| |-------| | |----| | 448 * ...->|runs_dirty |<-->|rd |<-->|rd |<---->|rd |<----... 449 * |------------| |-------| |-------| | |----| | 450 * | | | | | | | | | | 451 * | | | | | | | \----/ | 452 * | | \-------/ \-------/ | | 453 * | | | | 454 * | | | | 455 * \------------/ \---------/ 456 */ 457 arena_runs_dirty_link_t runs_dirty; 458 extent_node_t chunks_cache; 459 460 /* Decay-based purging state. */ 461 arena_decay_t decay; 462 463 /* Extant huge allocations. */ 464 ql_head(extent_node_t) huge; 465 /* Synchronizes all huge allocation/update/deallocation. */ 466 malloc_mutex_t huge_mtx; 467 468 /* 469 * Trees of chunks that were previously allocated (trees differ only in 470 * node ordering). These are used when allocating chunks, in an attempt 471 * to re-use address space. Depending on function, different tree 472 * orderings are needed, which is why there are two trees with the same 473 * contents. 474 */ 475 extent_tree_t chunks_szsnad_cached; 476 extent_tree_t chunks_ad_cached; 477 extent_tree_t chunks_szsnad_retained; 478 extent_tree_t chunks_ad_retained; 479 480 malloc_mutex_t chunks_mtx; 481 /* Cache of nodes that were allocated via base_alloc(). */ 482 ql_head(extent_node_t) node_cache; 483 malloc_mutex_t node_cache_mtx; 484 485 /* User-configurable chunk hook functions. */ 486 chunk_hooks_t chunk_hooks; 487 488 /* bins is used to store trees of free regions. */ 489 arena_bin_t bins[NBINS]; 490 491 /* 492 * Size-segregated address-ordered heaps of this arena's available runs, 493 * used for first-best-fit run allocation. Runs are quantized, i.e. 494 * they reside in the last heap which corresponds to a size class less 495 * than or equal to the run size. 496 */ 497 arena_run_heap_t runs_avail[NPSIZES]; 498 }; 499 500 /* Used in conjunction with tsd for fast arena-related context lookup. */ 501 struct arena_tdata_s { 502 ticker_t decay_ticker; 503 }; 504 #endif /* JEMALLOC_ARENA_STRUCTS_B */ 505 506 #endif /* JEMALLOC_H_STRUCTS */ 507 /******************************************************************************/ 508 #ifdef JEMALLOC_H_EXTERNS 509 510 static const size_t large_pad = 511 #ifdef JEMALLOC_CACHE_OBLIVIOUS 512 PAGE 513 #else 514 0 515 #endif 516 ; 517 518 extern purge_mode_t opt_purge; 519 extern const char *purge_mode_names[]; 520 extern ssize_t opt_lg_dirty_mult; 521 extern ssize_t opt_decay_time; 522 523 extern arena_bin_info_t arena_bin_info[NBINS]; 524 525 extern size_t map_bias; /* Number of arena chunk header pages. */ 526 extern size_t map_misc_offset; 527 extern size_t arena_maxrun; /* Max run size for arenas. */ 528 extern size_t large_maxclass; /* Max large size class. */ 529 extern unsigned nlclasses; /* Number of large size classes. */ 530 extern unsigned nhclasses; /* Number of huge size classes. */ 531 532 #ifdef JEMALLOC_JET 533 typedef size_t (run_quantize_t)(size_t); 534 extern run_quantize_t *run_quantize_floor; 535 extern run_quantize_t *run_quantize_ceil; 536 #endif 537 void arena_chunk_cache_maybe_insert(arena_t *arena, extent_node_t *node, 538 bool cache); 539 void arena_chunk_cache_maybe_remove(arena_t *arena, extent_node_t *node, 540 bool cache); 541 extent_node_t *arena_node_alloc(tsdn_t *tsdn, arena_t *arena); 542 void arena_node_dalloc(tsdn_t *tsdn, arena_t *arena, extent_node_t *node); 543 void *arena_chunk_alloc_huge(tsdn_t *tsdn, arena_t *arena, size_t usize, 544 size_t alignment, size_t *sn, bool *zero); 545 void arena_chunk_dalloc_huge(tsdn_t *tsdn, arena_t *arena, void *chunk, 546 size_t usize, size_t sn); 547 void arena_chunk_ralloc_huge_similar(tsdn_t *tsdn, arena_t *arena, 548 void *chunk, size_t oldsize, size_t usize); 549 void arena_chunk_ralloc_huge_shrink(tsdn_t *tsdn, arena_t *arena, 550 void *chunk, size_t oldsize, size_t usize, size_t sn); 551 bool arena_chunk_ralloc_huge_expand(tsdn_t *tsdn, arena_t *arena, 552 void *chunk, size_t oldsize, size_t usize, bool *zero); 553 ssize_t arena_lg_dirty_mult_get(tsdn_t *tsdn, arena_t *arena); 554 bool arena_lg_dirty_mult_set(tsdn_t *tsdn, arena_t *arena, 555 ssize_t lg_dirty_mult); 556 ssize_t arena_decay_time_get(tsdn_t *tsdn, arena_t *arena); 557 bool arena_decay_time_set(tsdn_t *tsdn, arena_t *arena, ssize_t decay_time); 558 void arena_purge(tsdn_t *tsdn, arena_t *arena, bool all); 559 void arena_maybe_purge(tsdn_t *tsdn, arena_t *arena); 560 void arena_reset(tsd_t *tsd, arena_t *arena); 561 void arena_tcache_fill_small(tsdn_t *tsdn, arena_t *arena, 562 tcache_bin_t *tbin, szind_t binind, uint64_t prof_accumbytes); 563 void arena_alloc_junk_small(void *ptr, arena_bin_info_t *bin_info, 564 bool zero); 565 #ifdef JEMALLOC_JET 566 typedef void (arena_redzone_corruption_t)(void *, size_t, bool, size_t, 567 uint8_t); 568 extern arena_redzone_corruption_t *arena_redzone_corruption; 569 typedef void (arena_dalloc_junk_small_t)(void *, arena_bin_info_t *); 570 extern arena_dalloc_junk_small_t *arena_dalloc_junk_small; 571 #else 572 void arena_dalloc_junk_small(void *ptr, arena_bin_info_t *bin_info); 573 #endif 574 void arena_quarantine_junk_small(void *ptr, size_t usize); 575 void *arena_malloc_large(tsdn_t *tsdn, arena_t *arena, szind_t ind, 576 bool zero); 577 void *arena_malloc_hard(tsdn_t *tsdn, arena_t *arena, size_t size, 578 szind_t ind, bool zero); 579 void *arena_palloc(tsdn_t *tsdn, arena_t *arena, size_t usize, 580 size_t alignment, bool zero, tcache_t *tcache); 581 void arena_prof_promoted(tsdn_t *tsdn, const void *ptr, size_t size); 582 void arena_dalloc_bin_junked_locked(tsdn_t *tsdn, arena_t *arena, 583 arena_chunk_t *chunk, void *ptr, arena_chunk_map_bits_t *bitselm); 584 void arena_dalloc_bin(tsdn_t *tsdn, arena_t *arena, arena_chunk_t *chunk, 585 void *ptr, size_t pageind, arena_chunk_map_bits_t *bitselm); 586 void arena_dalloc_small(tsdn_t *tsdn, arena_t *arena, arena_chunk_t *chunk, 587 void *ptr, size_t pageind); 588 #ifdef JEMALLOC_JET 589 typedef void (arena_dalloc_junk_large_t)(void *, size_t); 590 extern arena_dalloc_junk_large_t *arena_dalloc_junk_large; 591 #else 592 void arena_dalloc_junk_large(void *ptr, size_t usize); 593 #endif 594 void arena_dalloc_large_junked_locked(tsdn_t *tsdn, arena_t *arena, 595 arena_chunk_t *chunk, void *ptr); 596 void arena_dalloc_large(tsdn_t *tsdn, arena_t *arena, arena_chunk_t *chunk, 597 void *ptr); 598 #ifdef JEMALLOC_JET 599 typedef void (arena_ralloc_junk_large_t)(void *, size_t, size_t); 600 extern arena_ralloc_junk_large_t *arena_ralloc_junk_large; 601 #endif 602 bool arena_ralloc_no_move(tsdn_t *tsdn, void *ptr, size_t oldsize, 603 size_t size, size_t extra, bool zero); 604 void *arena_ralloc(tsd_t *tsd, arena_t *arena, void *ptr, size_t oldsize, 605 size_t size, size_t alignment, bool zero, tcache_t *tcache); 606 dss_prec_t arena_dss_prec_get(tsdn_t *tsdn, arena_t *arena); 607 bool arena_dss_prec_set(tsdn_t *tsdn, arena_t *arena, dss_prec_t dss_prec); 608 ssize_t arena_lg_dirty_mult_default_get(void); 609 bool arena_lg_dirty_mult_default_set(ssize_t lg_dirty_mult); 610 ssize_t arena_decay_time_default_get(void); 611 bool arena_decay_time_default_set(ssize_t decay_time); 612 void arena_basic_stats_merge(tsdn_t *tsdn, arena_t *arena, 613 unsigned *nthreads, const char **dss, ssize_t *lg_dirty_mult, 614 ssize_t *decay_time, size_t *nactive, size_t *ndirty); 615 void arena_stats_merge(tsdn_t *tsdn, arena_t *arena, unsigned *nthreads, 616 const char **dss, ssize_t *lg_dirty_mult, ssize_t *decay_time, 617 size_t *nactive, size_t *ndirty, arena_stats_t *astats, 618 malloc_bin_stats_t *bstats, malloc_large_stats_t *lstats, 619 malloc_huge_stats_t *hstats); 620 unsigned arena_nthreads_get(arena_t *arena, bool internal); 621 void arena_nthreads_inc(arena_t *arena, bool internal); 622 void arena_nthreads_dec(arena_t *arena, bool internal); 623 size_t arena_extent_sn_next(arena_t *arena); 624 arena_t *arena_new(tsdn_t *tsdn, unsigned ind); 625 void arena_boot(void); 626 void arena_prefork0(tsdn_t *tsdn, arena_t *arena); 627 void arena_prefork1(tsdn_t *tsdn, arena_t *arena); 628 void arena_prefork2(tsdn_t *tsdn, arena_t *arena); 629 void arena_prefork3(tsdn_t *tsdn, arena_t *arena); 630 void arena_postfork_parent(tsdn_t *tsdn, arena_t *arena); 631 void arena_postfork_child(tsdn_t *tsdn, arena_t *arena); 632 633 #endif /* JEMALLOC_H_EXTERNS */ 634 /******************************************************************************/ 635 #ifdef JEMALLOC_H_INLINES 636 637 #ifndef JEMALLOC_ENABLE_INLINE 638 arena_chunk_map_bits_t *arena_bitselm_get_mutable(arena_chunk_t *chunk, 639 size_t pageind); 640 const arena_chunk_map_bits_t *arena_bitselm_get_const( 641 const arena_chunk_t *chunk, size_t pageind); 642 arena_chunk_map_misc_t *arena_miscelm_get_mutable(arena_chunk_t *chunk, 643 size_t pageind); 644 const arena_chunk_map_misc_t *arena_miscelm_get_const( 645 const arena_chunk_t *chunk, size_t pageind); 646 size_t arena_miscelm_to_pageind(const arena_chunk_map_misc_t *miscelm); 647 void *arena_miscelm_to_rpages(const arena_chunk_map_misc_t *miscelm); 648 arena_chunk_map_misc_t *arena_rd_to_miscelm(arena_runs_dirty_link_t *rd); 649 arena_chunk_map_misc_t *arena_run_to_miscelm(arena_run_t *run); 650 size_t *arena_mapbitsp_get_mutable(arena_chunk_t *chunk, size_t pageind); 651 const size_t *arena_mapbitsp_get_const(const arena_chunk_t *chunk, 652 size_t pageind); 653 size_t arena_mapbitsp_read(const size_t *mapbitsp); 654 size_t arena_mapbits_get(const arena_chunk_t *chunk, size_t pageind); 655 size_t arena_mapbits_size_decode(size_t mapbits); 656 size_t arena_mapbits_unallocated_size_get(const arena_chunk_t *chunk, 657 size_t pageind); 658 size_t arena_mapbits_large_size_get(const arena_chunk_t *chunk, 659 size_t pageind); 660 size_t arena_mapbits_small_runind_get(const arena_chunk_t *chunk, 661 size_t pageind); 662 szind_t arena_mapbits_binind_get(const arena_chunk_t *chunk, size_t pageind); 663 size_t arena_mapbits_dirty_get(const arena_chunk_t *chunk, size_t pageind); 664 size_t arena_mapbits_unzeroed_get(const arena_chunk_t *chunk, size_t pageind); 665 size_t arena_mapbits_decommitted_get(const arena_chunk_t *chunk, 666 size_t pageind); 667 size_t arena_mapbits_large_get(const arena_chunk_t *chunk, size_t pageind); 668 size_t arena_mapbits_allocated_get(const arena_chunk_t *chunk, size_t pageind); 669 void arena_mapbitsp_write(size_t *mapbitsp, size_t mapbits); 670 size_t arena_mapbits_size_encode(size_t size); 671 void arena_mapbits_unallocated_set(arena_chunk_t *chunk, size_t pageind, 672 size_t size, size_t flags); 673 void arena_mapbits_unallocated_size_set(arena_chunk_t *chunk, size_t pageind, 674 size_t size); 675 void arena_mapbits_internal_set(arena_chunk_t *chunk, size_t pageind, 676 size_t flags); 677 void arena_mapbits_large_set(arena_chunk_t *chunk, size_t pageind, 678 size_t size, size_t flags); 679 void arena_mapbits_large_binind_set(arena_chunk_t *chunk, size_t pageind, 680 szind_t binind); 681 void arena_mapbits_small_set(arena_chunk_t *chunk, size_t pageind, 682 size_t runind, szind_t binind, size_t flags); 683 void arena_metadata_allocated_add(arena_t *arena, size_t size); 684 void arena_metadata_allocated_sub(arena_t *arena, size_t size); 685 size_t arena_metadata_allocated_get(arena_t *arena); 686 bool arena_prof_accum_impl(arena_t *arena, uint64_t accumbytes); 687 bool arena_prof_accum_locked(arena_t *arena, uint64_t accumbytes); 688 bool arena_prof_accum(tsdn_t *tsdn, arena_t *arena, uint64_t accumbytes); 689 szind_t arena_ptr_small_binind_get(const void *ptr, size_t mapbits); 690 szind_t arena_bin_index(arena_t *arena, arena_bin_t *bin); 691 size_t arena_run_regind(arena_run_t *run, arena_bin_info_t *bin_info, 692 const void *ptr); 693 prof_tctx_t *arena_prof_tctx_get(tsdn_t *tsdn, const void *ptr); 694 void arena_prof_tctx_set(tsdn_t *tsdn, const void *ptr, size_t usize, 695 prof_tctx_t *tctx); 696 void arena_prof_tctx_reset(tsdn_t *tsdn, const void *ptr, size_t usize, 697 const void *old_ptr, prof_tctx_t *old_tctx); 698 void arena_decay_ticks(tsdn_t *tsdn, arena_t *arena, unsigned nticks); 699 void arena_decay_tick(tsdn_t *tsdn, arena_t *arena); 700 void *arena_malloc(tsdn_t *tsdn, arena_t *arena, size_t size, szind_t ind, 701 bool zero, tcache_t *tcache, bool slow_path); 702 arena_t *arena_aalloc(const void *ptr); 703 size_t arena_salloc(tsdn_t *tsdn, const void *ptr, bool demote); 704 void arena_dalloc(tsdn_t *tsdn, void *ptr, tcache_t *tcache, bool slow_path); 705 void arena_sdalloc(tsdn_t *tsdn, void *ptr, size_t size, tcache_t *tcache, 706 bool slow_path); 707 #endif 708 709 #if (defined(JEMALLOC_ENABLE_INLINE) || defined(JEMALLOC_ARENA_C_)) 710 # ifdef JEMALLOC_ARENA_INLINE_A 711 JEMALLOC_ALWAYS_INLINE arena_chunk_map_bits_t * 712 arena_bitselm_get_mutable(arena_chunk_t *chunk, size_t pageind) 713 { 714 715 assert(pageind >= map_bias); 716 assert(pageind < chunk_npages); 717 718 return (&chunk->map_bits[pageind-map_bias]); 719 } 720 721 JEMALLOC_ALWAYS_INLINE const arena_chunk_map_bits_t * 722 arena_bitselm_get_const(const arena_chunk_t *chunk, size_t pageind) 723 { 724 725 return (arena_bitselm_get_mutable((arena_chunk_t *)chunk, pageind)); 726 } 727 728 JEMALLOC_ALWAYS_INLINE arena_chunk_map_misc_t * 729 arena_miscelm_get_mutable(arena_chunk_t *chunk, size_t pageind) 730 { 731 732 assert(pageind >= map_bias); 733 assert(pageind < chunk_npages); 734 735 return ((arena_chunk_map_misc_t *)((uintptr_t)chunk + 736 (uintptr_t)map_misc_offset) + pageind-map_bias); 737 } 738 739 JEMALLOC_ALWAYS_INLINE const arena_chunk_map_misc_t * 740 arena_miscelm_get_const(const arena_chunk_t *chunk, size_t pageind) 741 { 742 743 return (arena_miscelm_get_mutable((arena_chunk_t *)chunk, pageind)); 744 } 745 746 JEMALLOC_ALWAYS_INLINE size_t 747 arena_miscelm_to_pageind(const arena_chunk_map_misc_t *miscelm) 748 { 749 arena_chunk_t *chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(miscelm); 750 size_t pageind = ((uintptr_t)miscelm - ((uintptr_t)chunk + 751 map_misc_offset)) / sizeof(arena_chunk_map_misc_t) + map_bias; 752 753 assert(pageind >= map_bias); 754 assert(pageind < chunk_npages); 755 756 return (pageind); 757 } 758 759 JEMALLOC_ALWAYS_INLINE void * 760 arena_miscelm_to_rpages(const arena_chunk_map_misc_t *miscelm) 761 { 762 arena_chunk_t *chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(miscelm); 763 size_t pageind = arena_miscelm_to_pageind(miscelm); 764 765 return ((void *)((uintptr_t)chunk + (pageind << LG_PAGE))); 766 } 767 768 JEMALLOC_ALWAYS_INLINE arena_chunk_map_misc_t * 769 arena_rd_to_miscelm(arena_runs_dirty_link_t *rd) 770 { 771 arena_chunk_map_misc_t *miscelm = (arena_chunk_map_misc_t 772 *)((uintptr_t)rd - offsetof(arena_chunk_map_misc_t, rd)); 773 774 assert(arena_miscelm_to_pageind(miscelm) >= map_bias); 775 assert(arena_miscelm_to_pageind(miscelm) < chunk_npages); 776 777 return (miscelm); 778 } 779 780 JEMALLOC_ALWAYS_INLINE arena_chunk_map_misc_t * 781 arena_run_to_miscelm(arena_run_t *run) 782 { 783 arena_chunk_map_misc_t *miscelm = (arena_chunk_map_misc_t 784 *)((uintptr_t)run - offsetof(arena_chunk_map_misc_t, run)); 785 786 assert(arena_miscelm_to_pageind(miscelm) >= map_bias); 787 assert(arena_miscelm_to_pageind(miscelm) < chunk_npages); 788 789 return (miscelm); 790 } 791 792 JEMALLOC_ALWAYS_INLINE size_t * 793 arena_mapbitsp_get_mutable(arena_chunk_t *chunk, size_t pageind) 794 { 795 796 return (&arena_bitselm_get_mutable(chunk, pageind)->bits); 797 } 798 799 JEMALLOC_ALWAYS_INLINE const size_t * 800 arena_mapbitsp_get_const(const arena_chunk_t *chunk, size_t pageind) 801 { 802 803 return (arena_mapbitsp_get_mutable((arena_chunk_t *)chunk, pageind)); 804 } 805 806 JEMALLOC_ALWAYS_INLINE size_t 807 arena_mapbitsp_read(const size_t *mapbitsp) 808 { 809 810 return (*mapbitsp); 811 } 812 813 JEMALLOC_ALWAYS_INLINE size_t 814 arena_mapbits_get(const arena_chunk_t *chunk, size_t pageind) 815 { 816 817 return (arena_mapbitsp_read(arena_mapbitsp_get_const(chunk, pageind))); 818 } 819 820 JEMALLOC_ALWAYS_INLINE size_t 821 arena_mapbits_size_decode(size_t mapbits) 822 { 823 size_t size; 824 825 #if CHUNK_MAP_SIZE_SHIFT > 0 826 size = (mapbits & CHUNK_MAP_SIZE_MASK) >> CHUNK_MAP_SIZE_SHIFT; 827 #elif CHUNK_MAP_SIZE_SHIFT == 0 828 size = mapbits & CHUNK_MAP_SIZE_MASK; 829 #else 830 size = (mapbits & CHUNK_MAP_SIZE_MASK) << -CHUNK_MAP_SIZE_SHIFT; 831 #endif 832 833 return (size); 834 } 835 836 JEMALLOC_ALWAYS_INLINE size_t 837 arena_mapbits_unallocated_size_get(const arena_chunk_t *chunk, size_t pageind) 838 { 839 size_t mapbits; 840 841 mapbits = arena_mapbits_get(chunk, pageind); 842 assert((mapbits & (CHUNK_MAP_LARGE|CHUNK_MAP_ALLOCATED)) == 0); 843 return (arena_mapbits_size_decode(mapbits)); 844 } 845 846 JEMALLOC_ALWAYS_INLINE size_t 847 arena_mapbits_large_size_get(const arena_chunk_t *chunk, size_t pageind) 848 { 849 size_t mapbits; 850 851 mapbits = arena_mapbits_get(chunk, pageind); 852 assert((mapbits & (CHUNK_MAP_LARGE|CHUNK_MAP_ALLOCATED)) == 853 (CHUNK_MAP_LARGE|CHUNK_MAP_ALLOCATED)); 854 return (arena_mapbits_size_decode(mapbits)); 855 } 856 857 JEMALLOC_ALWAYS_INLINE size_t 858 arena_mapbits_small_runind_get(const arena_chunk_t *chunk, size_t pageind) 859 { 860 size_t mapbits; 861 862 mapbits = arena_mapbits_get(chunk, pageind); 863 assert((mapbits & (CHUNK_MAP_LARGE|CHUNK_MAP_ALLOCATED)) == 864 CHUNK_MAP_ALLOCATED); 865 return (mapbits >> CHUNK_MAP_RUNIND_SHIFT); 866 } 867 868 JEMALLOC_ALWAYS_INLINE szind_t 869 arena_mapbits_binind_get(const arena_chunk_t *chunk, size_t pageind) 870 { 871 size_t mapbits; 872 szind_t binind; 873 874 mapbits = arena_mapbits_get(chunk, pageind); 875 binind = (mapbits & CHUNK_MAP_BININD_MASK) >> CHUNK_MAP_BININD_SHIFT; 876 assert(binind < NBINS || binind == BININD_INVALID); 877 return (binind); 878 } 879 880 JEMALLOC_ALWAYS_INLINE size_t 881 arena_mapbits_dirty_get(const arena_chunk_t *chunk, size_t pageind) 882 { 883 size_t mapbits; 884 885 mapbits = arena_mapbits_get(chunk, pageind); 886 assert((mapbits & CHUNK_MAP_DECOMMITTED) == 0 || (mapbits & 887 (CHUNK_MAP_DIRTY|CHUNK_MAP_UNZEROED)) == 0); 888 return (mapbits & CHUNK_MAP_DIRTY); 889 } 890 891 JEMALLOC_ALWAYS_INLINE size_t 892 arena_mapbits_unzeroed_get(const arena_chunk_t *chunk, size_t pageind) 893 { 894 size_t mapbits; 895 896 mapbits = arena_mapbits_get(chunk, pageind); 897 assert((mapbits & CHUNK_MAP_DECOMMITTED) == 0 || (mapbits & 898 (CHUNK_MAP_DIRTY|CHUNK_MAP_UNZEROED)) == 0); 899 return (mapbits & CHUNK_MAP_UNZEROED); 900 } 901 902 JEMALLOC_ALWAYS_INLINE size_t 903 arena_mapbits_decommitted_get(const arena_chunk_t *chunk, size_t pageind) 904 { 905 size_t mapbits; 906 907 mapbits = arena_mapbits_get(chunk, pageind); 908 assert((mapbits & CHUNK_MAP_DECOMMITTED) == 0 || (mapbits & 909 (CHUNK_MAP_DIRTY|CHUNK_MAP_UNZEROED)) == 0); 910 return (mapbits & CHUNK_MAP_DECOMMITTED); 911 } 912 913 JEMALLOC_ALWAYS_INLINE size_t 914 arena_mapbits_large_get(const arena_chunk_t *chunk, size_t pageind) 915 { 916 size_t mapbits; 917 918 mapbits = arena_mapbits_get(chunk, pageind); 919 return (mapbits & CHUNK_MAP_LARGE); 920 } 921 922 JEMALLOC_ALWAYS_INLINE size_t 923 arena_mapbits_allocated_get(const arena_chunk_t *chunk, size_t pageind) 924 { 925 size_t mapbits; 926 927 mapbits = arena_mapbits_get(chunk, pageind); 928 return (mapbits & CHUNK_MAP_ALLOCATED); 929 } 930 931 JEMALLOC_ALWAYS_INLINE void 932 arena_mapbitsp_write(size_t *mapbitsp, size_t mapbits) 933 { 934 935 *mapbitsp = mapbits; 936 } 937 938 JEMALLOC_ALWAYS_INLINE size_t 939 arena_mapbits_size_encode(size_t size) 940 { 941 size_t mapbits; 942 943 #if CHUNK_MAP_SIZE_SHIFT > 0 944 mapbits = size << CHUNK_MAP_SIZE_SHIFT; 945 #elif CHUNK_MAP_SIZE_SHIFT == 0 946 mapbits = size; 947 #else 948 mapbits = size >> -CHUNK_MAP_SIZE_SHIFT; 949 #endif 950 951 assert((mapbits & ~CHUNK_MAP_SIZE_MASK) == 0); 952 return (mapbits); 953 } 954 955 JEMALLOC_ALWAYS_INLINE void 956 arena_mapbits_unallocated_set(arena_chunk_t *chunk, size_t pageind, size_t size, 957 size_t flags) 958 { 959 size_t *mapbitsp = arena_mapbitsp_get_mutable(chunk, pageind); 960 961 assert((size & PAGE_MASK) == 0); 962 assert((flags & CHUNK_MAP_FLAGS_MASK) == flags); 963 assert((flags & CHUNK_MAP_DECOMMITTED) == 0 || (flags & 964 (CHUNK_MAP_DIRTY|CHUNK_MAP_UNZEROED)) == 0); 965 arena_mapbitsp_write(mapbitsp, arena_mapbits_size_encode(size) | 966 CHUNK_MAP_BININD_INVALID | flags); 967 } 968 969 JEMALLOC_ALWAYS_INLINE void 970 arena_mapbits_unallocated_size_set(arena_chunk_t *chunk, size_t pageind, 971 size_t size) 972 { 973 size_t *mapbitsp = arena_mapbitsp_get_mutable(chunk, pageind); 974 size_t mapbits = arena_mapbitsp_read(mapbitsp); 975 976 assert((size & PAGE_MASK) == 0); 977 assert((mapbits & (CHUNK_MAP_LARGE|CHUNK_MAP_ALLOCATED)) == 0); 978 arena_mapbitsp_write(mapbitsp, arena_mapbits_size_encode(size) | 979 (mapbits & ~CHUNK_MAP_SIZE_MASK)); 980 } 981 982 JEMALLOC_ALWAYS_INLINE void 983 arena_mapbits_internal_set(arena_chunk_t *chunk, size_t pageind, size_t flags) 984 { 985 size_t *mapbitsp = arena_mapbitsp_get_mutable(chunk, pageind); 986 987 assert((flags & CHUNK_MAP_UNZEROED) == flags); 988 arena_mapbitsp_write(mapbitsp, flags); 989 } 990 991 JEMALLOC_ALWAYS_INLINE void 992 arena_mapbits_large_set(arena_chunk_t *chunk, size_t pageind, size_t size, 993 size_t flags) 994 { 995 size_t *mapbitsp = arena_mapbitsp_get_mutable(chunk, pageind); 996 997 assert((size & PAGE_MASK) == 0); 998 assert((flags & CHUNK_MAP_FLAGS_MASK) == flags); 999 assert((flags & CHUNK_MAP_DECOMMITTED) == 0 || (flags & 1000 (CHUNK_MAP_DIRTY|CHUNK_MAP_UNZEROED)) == 0); 1001 arena_mapbitsp_write(mapbitsp, arena_mapbits_size_encode(size) | 1002 CHUNK_MAP_BININD_INVALID | flags | CHUNK_MAP_LARGE | 1003 CHUNK_MAP_ALLOCATED); 1004 } 1005 1006 JEMALLOC_ALWAYS_INLINE void 1007 arena_mapbits_large_binind_set(arena_chunk_t *chunk, size_t pageind, 1008 szind_t binind) 1009 { 1010 size_t *mapbitsp = arena_mapbitsp_get_mutable(chunk, pageind); 1011 size_t mapbits = arena_mapbitsp_read(mapbitsp); 1012 1013 assert(binind <= BININD_INVALID); 1014 assert(arena_mapbits_large_size_get(chunk, pageind) == LARGE_MINCLASS + 1015 large_pad); 1016 arena_mapbitsp_write(mapbitsp, (mapbits & ~CHUNK_MAP_BININD_MASK) | 1017 (binind << CHUNK_MAP_BININD_SHIFT)); 1018 } 1019 1020 JEMALLOC_ALWAYS_INLINE void 1021 arena_mapbits_small_set(arena_chunk_t *chunk, size_t pageind, size_t runind, 1022 szind_t binind, size_t flags) 1023 { 1024 size_t *mapbitsp = arena_mapbitsp_get_mutable(chunk, pageind); 1025 1026 assert(binind < BININD_INVALID); 1027 assert(pageind - runind >= map_bias); 1028 assert((flags & CHUNK_MAP_UNZEROED) == flags); 1029 arena_mapbitsp_write(mapbitsp, (runind << CHUNK_MAP_RUNIND_SHIFT) | 1030 (binind << CHUNK_MAP_BININD_SHIFT) | flags | CHUNK_MAP_ALLOCATED); 1031 } 1032 1033 JEMALLOC_INLINE void 1034 arena_metadata_allocated_add(arena_t *arena, size_t size) 1035 { 1036 1037 atomic_add_z(&arena->stats.metadata_allocated, size); 1038 } 1039 1040 JEMALLOC_INLINE void 1041 arena_metadata_allocated_sub(arena_t *arena, size_t size) 1042 { 1043 1044 atomic_sub_z(&arena->stats.metadata_allocated, size); 1045 } 1046 1047 JEMALLOC_INLINE size_t 1048 arena_metadata_allocated_get(arena_t *arena) 1049 { 1050 1051 return (atomic_read_z(&arena->stats.metadata_allocated)); 1052 } 1053 1054 JEMALLOC_INLINE bool 1055 arena_prof_accum_impl(arena_t *arena, uint64_t accumbytes) 1056 { 1057 1058 cassert(config_prof); 1059 assert(prof_interval != 0); 1060 1061 arena->prof_accumbytes += accumbytes; 1062 if (arena->prof_accumbytes >= prof_interval) { 1063 arena->prof_accumbytes -= prof_interval; 1064 return (true); 1065 } 1066 return (false); 1067 } 1068 1069 JEMALLOC_INLINE bool 1070 arena_prof_accum_locked(arena_t *arena, uint64_t accumbytes) 1071 { 1072 1073 cassert(config_prof); 1074 1075 if (likely(prof_interval == 0)) 1076 return (false); 1077 return (arena_prof_accum_impl(arena, accumbytes)); 1078 } 1079 1080 JEMALLOC_INLINE bool 1081 arena_prof_accum(tsdn_t *tsdn, arena_t *arena, uint64_t accumbytes) 1082 { 1083 1084 cassert(config_prof); 1085 1086 if (likely(prof_interval == 0)) 1087 return (false); 1088 1089 { 1090 bool ret; 1091 1092 malloc_mutex_lock(tsdn, &arena->lock); 1093 ret = arena_prof_accum_impl(arena, accumbytes); 1094 malloc_mutex_unlock(tsdn, &arena->lock); 1095 return (ret); 1096 } 1097 } 1098 1099 JEMALLOC_ALWAYS_INLINE szind_t 1100 arena_ptr_small_binind_get(const void *ptr, size_t mapbits) 1101 { 1102 szind_t binind; 1103 1104 binind = (mapbits & CHUNK_MAP_BININD_MASK) >> CHUNK_MAP_BININD_SHIFT; 1105 1106 if (config_debug) { 1107 arena_chunk_t *chunk; 1108 arena_t *arena; 1109 size_t pageind; 1110 size_t actual_mapbits; 1111 size_t rpages_ind; 1112 const arena_run_t *run; 1113 arena_bin_t *bin; 1114 szind_t run_binind, actual_binind; 1115 arena_bin_info_t *bin_info; 1116 const arena_chunk_map_misc_t *miscelm; 1117 const void *rpages; 1118 1119 assert(binind != BININD_INVALID); 1120 assert(binind < NBINS); 1121 chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr); 1122 arena = extent_node_arena_get(&chunk->node); 1123 pageind = ((uintptr_t)ptr - (uintptr_t)chunk) >> LG_PAGE; 1124 actual_mapbits = arena_mapbits_get(chunk, pageind); 1125 assert(mapbits == actual_mapbits); 1126 assert(arena_mapbits_large_get(chunk, pageind) == 0); 1127 assert(arena_mapbits_allocated_get(chunk, pageind) != 0); 1128 rpages_ind = pageind - arena_mapbits_small_runind_get(chunk, 1129 pageind); 1130 miscelm = arena_miscelm_get_const(chunk, rpages_ind); 1131 run = &miscelm->run; 1132 run_binind = run->binind; 1133 bin = &arena->bins[run_binind]; 1134 actual_binind = (szind_t)(bin - arena->bins); 1135 assert(run_binind == actual_binind); 1136 bin_info = &arena_bin_info[actual_binind]; 1137 rpages = arena_miscelm_to_rpages(miscelm); 1138 assert(((uintptr_t)ptr - ((uintptr_t)rpages + 1139 (uintptr_t)bin_info->reg0_offset)) % bin_info->reg_interval 1140 == 0); 1141 } 1142 1143 return (binind); 1144 } 1145 # endif /* JEMALLOC_ARENA_INLINE_A */ 1146 1147 # ifdef JEMALLOC_ARENA_INLINE_B 1148 JEMALLOC_INLINE szind_t 1149 arena_bin_index(arena_t *arena, arena_bin_t *bin) 1150 { 1151 szind_t binind = (szind_t)(bin - arena->bins); 1152 assert(binind < NBINS); 1153 return (binind); 1154 } 1155 1156 JEMALLOC_INLINE size_t 1157 arena_run_regind(arena_run_t *run, arena_bin_info_t *bin_info, const void *ptr) 1158 { 1159 size_t diff, interval, shift, regind; 1160 arena_chunk_map_misc_t *miscelm = arena_run_to_miscelm(run); 1161 void *rpages = arena_miscelm_to_rpages(miscelm); 1162 1163 /* 1164 * Freeing a pointer lower than region zero can cause assertion 1165 * failure. 1166 */ 1167 assert((uintptr_t)ptr >= (uintptr_t)rpages + 1168 (uintptr_t)bin_info->reg0_offset); 1169 1170 /* 1171 * Avoid doing division with a variable divisor if possible. Using 1172 * actual division here can reduce allocator throughput by over 20%! 1173 */ 1174 diff = (size_t)((uintptr_t)ptr - (uintptr_t)rpages - 1175 bin_info->reg0_offset); 1176 1177 /* Rescale (factor powers of 2 out of the numerator and denominator). */ 1178 interval = bin_info->reg_interval; 1179 shift = ffs_zu(interval) - 1; 1180 diff >>= shift; 1181 interval >>= shift; 1182 1183 if (interval == 1) { 1184 /* The divisor was a power of 2. */ 1185 regind = diff; 1186 } else { 1187 /* 1188 * To divide by a number D that is not a power of two we 1189 * multiply by (2^21 / D) and then right shift by 21 positions. 1190 * 1191 * X / D 1192 * 1193 * becomes 1194 * 1195 * (X * interval_invs[D - 3]) >> SIZE_INV_SHIFT 1196 * 1197 * We can omit the first three elements, because we never 1198 * divide by 0, and 1 and 2 are both powers of two, which are 1199 * handled above. 1200 */ 1201 #define SIZE_INV_SHIFT ((sizeof(size_t) << 3) - LG_RUN_MAXREGS) 1202 #define SIZE_INV(s) (((ZU(1) << SIZE_INV_SHIFT) / (s)) + 1) 1203 static const size_t interval_invs[] = { 1204 SIZE_INV(3), 1205 SIZE_INV(4), SIZE_INV(5), SIZE_INV(6), SIZE_INV(7), 1206 SIZE_INV(8), SIZE_INV(9), SIZE_INV(10), SIZE_INV(11), 1207 SIZE_INV(12), SIZE_INV(13), SIZE_INV(14), SIZE_INV(15), 1208 SIZE_INV(16), SIZE_INV(17), SIZE_INV(18), SIZE_INV(19), 1209 SIZE_INV(20), SIZE_INV(21), SIZE_INV(22), SIZE_INV(23), 1210 SIZE_INV(24), SIZE_INV(25), SIZE_INV(26), SIZE_INV(27), 1211 SIZE_INV(28), SIZE_INV(29), SIZE_INV(30), SIZE_INV(31) 1212 }; 1213 1214 if (likely(interval <= ((sizeof(interval_invs) / sizeof(size_t)) 1215 + 2))) { 1216 regind = (diff * interval_invs[interval - 3]) >> 1217 SIZE_INV_SHIFT; 1218 } else 1219 regind = diff / interval; 1220 #undef SIZE_INV 1221 #undef SIZE_INV_SHIFT 1222 } 1223 assert(diff == regind * interval); 1224 assert(regind < bin_info->nregs); 1225 1226 return (regind); 1227 } 1228 1229 JEMALLOC_INLINE prof_tctx_t * 1230 arena_prof_tctx_get(tsdn_t *tsdn, const void *ptr) 1231 { 1232 prof_tctx_t *ret; 1233 arena_chunk_t *chunk; 1234 1235 cassert(config_prof); 1236 assert(ptr != NULL); 1237 1238 chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr); 1239 if (likely(chunk != ptr)) { 1240 size_t pageind = ((uintptr_t)ptr - (uintptr_t)chunk) >> LG_PAGE; 1241 size_t mapbits = arena_mapbits_get(chunk, pageind); 1242 assert((mapbits & CHUNK_MAP_ALLOCATED) != 0); 1243 if (likely((mapbits & CHUNK_MAP_LARGE) == 0)) 1244 ret = (prof_tctx_t *)(uintptr_t)1U; 1245 else { 1246 arena_chunk_map_misc_t *elm = 1247 arena_miscelm_get_mutable(chunk, pageind); 1248 ret = atomic_read_p(&elm->prof_tctx_pun); 1249 } 1250 } else 1251 ret = huge_prof_tctx_get(tsdn, ptr); 1252 1253 return (ret); 1254 } 1255 1256 JEMALLOC_INLINE void 1257 arena_prof_tctx_set(tsdn_t *tsdn, const void *ptr, size_t usize, 1258 prof_tctx_t *tctx) 1259 { 1260 arena_chunk_t *chunk; 1261 1262 cassert(config_prof); 1263 assert(ptr != NULL); 1264 1265 chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr); 1266 if (likely(chunk != ptr)) { 1267 size_t pageind = ((uintptr_t)ptr - (uintptr_t)chunk) >> LG_PAGE; 1268 1269 assert(arena_mapbits_allocated_get(chunk, pageind) != 0); 1270 1271 if (unlikely(usize > SMALL_MAXCLASS || (uintptr_t)tctx > 1272 (uintptr_t)1U)) { 1273 arena_chunk_map_misc_t *elm; 1274 1275 assert(arena_mapbits_large_get(chunk, pageind) != 0); 1276 1277 elm = arena_miscelm_get_mutable(chunk, pageind); 1278 atomic_write_p(&elm->prof_tctx_pun, tctx); 1279 } else { 1280 /* 1281 * tctx must always be initialized for large runs. 1282 * Assert that the surrounding conditional logic is 1283 * equivalent to checking whether ptr refers to a large 1284 * run. 1285 */ 1286 assert(arena_mapbits_large_get(chunk, pageind) == 0); 1287 } 1288 } else 1289 huge_prof_tctx_set(tsdn, ptr, tctx); 1290 } 1291 1292 JEMALLOC_INLINE void 1293 arena_prof_tctx_reset(tsdn_t *tsdn, const void *ptr, size_t usize, 1294 const void *old_ptr, prof_tctx_t *old_tctx) 1295 { 1296 1297 cassert(config_prof); 1298 assert(ptr != NULL); 1299 1300 if (unlikely(usize > SMALL_MAXCLASS || (ptr == old_ptr && 1301 (uintptr_t)old_tctx > (uintptr_t)1U))) { 1302 arena_chunk_t *chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr); 1303 if (likely(chunk != ptr)) { 1304 size_t pageind; 1305 arena_chunk_map_misc_t *elm; 1306 1307 pageind = ((uintptr_t)ptr - (uintptr_t)chunk) >> 1308 LG_PAGE; 1309 assert(arena_mapbits_allocated_get(chunk, pageind) != 1310 0); 1311 assert(arena_mapbits_large_get(chunk, pageind) != 0); 1312 1313 elm = arena_miscelm_get_mutable(chunk, pageind); 1314 atomic_write_p(&elm->prof_tctx_pun, 1315 (prof_tctx_t *)(uintptr_t)1U); 1316 } else 1317 huge_prof_tctx_reset(tsdn, ptr); 1318 } 1319 } 1320 1321 JEMALLOC_ALWAYS_INLINE void 1322 arena_decay_ticks(tsdn_t *tsdn, arena_t *arena, unsigned nticks) 1323 { 1324 tsd_t *tsd; 1325 ticker_t *decay_ticker; 1326 1327 if (unlikely(tsdn_null(tsdn))) 1328 return; 1329 tsd = tsdn_tsd(tsdn); 1330 decay_ticker = decay_ticker_get(tsd, arena->ind); 1331 if (unlikely(decay_ticker == NULL)) 1332 return; 1333 if (unlikely(ticker_ticks(decay_ticker, nticks))) 1334 arena_purge(tsdn, arena, false); 1335 } 1336 1337 JEMALLOC_ALWAYS_INLINE void 1338 arena_decay_tick(tsdn_t *tsdn, arena_t *arena) 1339 { 1340 1341 arena_decay_ticks(tsdn, arena, 1); 1342 } 1343 1344 JEMALLOC_ALWAYS_INLINE void * 1345 arena_malloc(tsdn_t *tsdn, arena_t *arena, size_t size, szind_t ind, bool zero, 1346 tcache_t *tcache, bool slow_path) 1347 { 1348 1349 assert(!tsdn_null(tsdn) || tcache == NULL); 1350 assert(size != 0); 1351 1352 if (likely(tcache != NULL)) { 1353 if (likely(size <= SMALL_MAXCLASS)) { 1354 return (tcache_alloc_small(tsdn_tsd(tsdn), arena, 1355 tcache, size, ind, zero, slow_path)); 1356 } 1357 if (likely(size <= tcache_maxclass)) { 1358 return (tcache_alloc_large(tsdn_tsd(tsdn), arena, 1359 tcache, size, ind, zero, slow_path)); 1360 } 1361 /* (size > tcache_maxclass) case falls through. */ 1362 assert(size > tcache_maxclass); 1363 } 1364 1365 return (arena_malloc_hard(tsdn, arena, size, ind, zero)); 1366 } 1367 1368 JEMALLOC_ALWAYS_INLINE arena_t * 1369 arena_aalloc(const void *ptr) 1370 { 1371 arena_chunk_t *chunk; 1372 1373 chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr); 1374 if (likely(chunk != ptr)) 1375 return (extent_node_arena_get(&chunk->node)); 1376 else 1377 return (huge_aalloc(ptr)); 1378 } 1379 1380 /* Return the size of the allocation pointed to by ptr. */ 1381 JEMALLOC_ALWAYS_INLINE size_t 1382 arena_salloc(tsdn_t *tsdn, const void *ptr, bool demote) 1383 { 1384 size_t ret; 1385 arena_chunk_t *chunk; 1386 size_t pageind; 1387 szind_t binind; 1388 1389 assert(ptr != NULL); 1390 1391 chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr); 1392 if (likely(chunk != ptr)) { 1393 pageind = ((uintptr_t)ptr - (uintptr_t)chunk) >> LG_PAGE; 1394 assert(arena_mapbits_allocated_get(chunk, pageind) != 0); 1395 binind = arena_mapbits_binind_get(chunk, pageind); 1396 if (unlikely(binind == BININD_INVALID || (config_prof && !demote 1397 && arena_mapbits_large_get(chunk, pageind) != 0))) { 1398 /* 1399 * Large allocation. In the common case (demote), and 1400 * as this is an inline function, most callers will only 1401 * end up looking at binind to determine that ptr is a 1402 * small allocation. 1403 */ 1404 assert(config_cache_oblivious || ((uintptr_t)ptr & 1405 PAGE_MASK) == 0); 1406 ret = arena_mapbits_large_size_get(chunk, pageind) - 1407 large_pad; 1408 assert(ret != 0); 1409 assert(pageind + ((ret+large_pad)>>LG_PAGE) <= 1410 chunk_npages); 1411 assert(arena_mapbits_dirty_get(chunk, pageind) == 1412 arena_mapbits_dirty_get(chunk, 1413 pageind+((ret+large_pad)>>LG_PAGE)-1)); 1414 } else { 1415 /* 1416 * Small allocation (possibly promoted to a large 1417 * object). 1418 */ 1419 assert(arena_mapbits_large_get(chunk, pageind) != 0 || 1420 arena_ptr_small_binind_get(ptr, 1421 arena_mapbits_get(chunk, pageind)) == binind); 1422 ret = index2size(binind); 1423 } 1424 } else 1425 ret = huge_salloc(tsdn, ptr); 1426 1427 return (ret); 1428 } 1429 1430 JEMALLOC_ALWAYS_INLINE void 1431 arena_dalloc(tsdn_t *tsdn, void *ptr, tcache_t *tcache, bool slow_path) 1432 { 1433 arena_chunk_t *chunk; 1434 size_t pageind, mapbits; 1435 1436 assert(!tsdn_null(tsdn) || tcache == NULL); 1437 assert(ptr != NULL); 1438 1439 chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr); 1440 if (likely(chunk != ptr)) { 1441 pageind = ((uintptr_t)ptr - (uintptr_t)chunk) >> LG_PAGE; 1442 #if defined(__ANDROID__) 1443 /* Verify the ptr is actually in the chunk. */ 1444 if (unlikely(pageind < map_bias || pageind >= chunk_npages)) { 1445 async_safe_fatal("Invalid address %p passed to free: invalid page index", ptr); 1446 } 1447 #endif 1448 mapbits = arena_mapbits_get(chunk, pageind); 1449 assert(arena_mapbits_allocated_get(chunk, pageind) != 0); 1450 #if defined(__ANDROID__) 1451 /* Verify the ptr has been allocated. */ 1452 if (unlikely((mapbits & CHUNK_MAP_ALLOCATED) == 0)) { 1453 async_safe_fatal("Invalid address %p passed to free: value not allocated", ptr); 1454 } 1455 #endif 1456 if (likely((mapbits & CHUNK_MAP_LARGE) == 0)) { 1457 /* Small allocation. */ 1458 if (likely(tcache != NULL)) { 1459 szind_t binind = arena_ptr_small_binind_get(ptr, 1460 mapbits); 1461 tcache_dalloc_small(tsdn_tsd(tsdn), tcache, ptr, 1462 binind, slow_path); 1463 } else { 1464 arena_dalloc_small(tsdn, 1465 extent_node_arena_get(&chunk->node), chunk, 1466 ptr, pageind); 1467 } 1468 } else { 1469 size_t size = arena_mapbits_large_size_get(chunk, 1470 pageind); 1471 1472 assert(config_cache_oblivious || ((uintptr_t)ptr & 1473 PAGE_MASK) == 0); 1474 1475 if (likely(tcache != NULL) && size - large_pad <= 1476 tcache_maxclass) { 1477 tcache_dalloc_large(tsdn_tsd(tsdn), tcache, ptr, 1478 size - large_pad, slow_path); 1479 } else { 1480 arena_dalloc_large(tsdn, 1481 extent_node_arena_get(&chunk->node), chunk, 1482 ptr); 1483 } 1484 } 1485 } else 1486 huge_dalloc(tsdn, ptr); 1487 } 1488 1489 JEMALLOC_ALWAYS_INLINE void 1490 arena_sdalloc(tsdn_t *tsdn, void *ptr, size_t size, tcache_t *tcache, 1491 bool slow_path) 1492 { 1493 arena_chunk_t *chunk; 1494 1495 assert(!tsdn_null(tsdn) || tcache == NULL); 1496 1497 chunk = (arena_chunk_t *)CHUNK_ADDR2BASE(ptr); 1498 if (likely(chunk != ptr)) { 1499 if (config_prof && opt_prof) { 1500 size_t pageind = ((uintptr_t)ptr - (uintptr_t)chunk) >> 1501 LG_PAGE; 1502 assert(arena_mapbits_allocated_get(chunk, pageind) != 1503 0); 1504 if (arena_mapbits_large_get(chunk, pageind) != 0) { 1505 /* 1506 * Make sure to use promoted size, not request 1507 * size. 1508 */ 1509 size = arena_mapbits_large_size_get(chunk, 1510 pageind) - large_pad; 1511 } 1512 } 1513 assert(s2u(size) == s2u(arena_salloc(tsdn, ptr, false))); 1514 1515 if (likely(size <= SMALL_MAXCLASS)) { 1516 /* Small allocation. */ 1517 if (likely(tcache != NULL)) { 1518 szind_t binind = size2index(size); 1519 tcache_dalloc_small(tsdn_tsd(tsdn), tcache, ptr, 1520 binind, slow_path); 1521 } else { 1522 size_t pageind = ((uintptr_t)ptr - 1523 (uintptr_t)chunk) >> LG_PAGE; 1524 arena_dalloc_small(tsdn, 1525 extent_node_arena_get(&chunk->node), chunk, 1526 ptr, pageind); 1527 } 1528 } else { 1529 assert(config_cache_oblivious || ((uintptr_t)ptr & 1530 PAGE_MASK) == 0); 1531 1532 if (likely(tcache != NULL) && size <= tcache_maxclass) { 1533 tcache_dalloc_large(tsdn_tsd(tsdn), tcache, ptr, 1534 size, slow_path); 1535 } else { 1536 arena_dalloc_large(tsdn, 1537 extent_node_arena_get(&chunk->node), chunk, 1538 ptr); 1539 } 1540 } 1541 } else 1542 huge_dalloc(tsdn, ptr); 1543 } 1544 # endif /* JEMALLOC_ARENA_INLINE_B */ 1545 #endif 1546 1547 #endif /* JEMALLOC_H_INLINES */ 1548 /******************************************************************************/ 1549
