1 /**
2 * \file
3 * Simple generational GC.
4 *
5 * Author:
6 * Paolo Molaro (lupus@ximian.com)
7 * Rodrigo Kumpera (kumpera@gmail.com)
8 *
9 * Copyright 2005-2011 Novell, Inc (http://www.novell.com)
10 * Copyright 2011 Xamarin Inc (http://www.xamarin.com)
11 *
12 * Thread start/stop adapted from Boehm's GC:
13 * Copyright (c) 1994 by Xerox Corporation. All rights reserved.
14 * Copyright (c) 1996 by Silicon Graphics. All rights reserved.
15 * Copyright (c) 1998 by Fergus Henderson. All rights reserved.
16 * Copyright (c) 2000-2004 by Hewlett-Packard Company. All rights reserved.
17 * Copyright 2001-2003 Ximian, Inc
18 * Copyright 2003-2010 Novell, Inc.
19 * Copyright 2011 Xamarin, Inc.
20 * Copyright (C) 2012 Xamarin Inc
21 *
22 * Licensed under the MIT license. See LICENSE file in the project root for full license information.
23 *
24 * Important: allocation provides always zeroed memory, having to do
25 * a memset after allocation is deadly for performance.
26 * Memory usage at startup is currently as follows:
27 * 64 KB pinned space
28 * 64 KB internal space
29 * size of nursery
30 * We should provide a small memory config with half the sizes
31 *
32 * We currently try to make as few mono assumptions as possible:
33 * 1) 2-word header with no GC pointers in it (first vtable, second to store the
34 * forwarding ptr)
35 * 2) gc descriptor is the second word in the vtable (first word in the class)
36 * 3) 8 byte alignment is the minimum and enough (not true for special structures (SIMD), FIXME)
37 * 4) there is a function to get an object's size and the number of
38 * elements in an array.
39 * 5) we know the special way bounds are allocated for complex arrays
40 * 6) we know about proxies and how to treat them when domains are unloaded
41 *
42 * Always try to keep stack usage to a minimum: no recursive behaviour
43 * and no large stack allocs.
44 *
45 * General description.
46 * Objects are initially allocated in a nursery using a fast bump-pointer technique.
47 * When the nursery is full we start a nursery collection: this is performed with a
48 * copying GC.
49 * When the old generation is full we start a copying GC of the old generation as well:
50 * this will be changed to mark&sweep with copying when fragmentation becomes to severe
51 * in the future. Maybe we'll even do both during the same collection like IMMIX.
52 *
53 * The things that complicate this description are:
54 * *) pinned objects: we can't move them so we need to keep track of them
55 * *) no precise info of the thread stacks and registers: we need to be able to
56 * quickly find the objects that may be referenced conservatively and pin them
57 * (this makes the first issues more important)
58 * *) large objects are too expensive to be dealt with using copying GC: we handle them
59 * with mark/sweep during major collections
60 * *) some objects need to not move even if they are small (interned strings, Type handles):
61 * we use mark/sweep for them, too: they are not allocated in the nursery, but inside
62 * PinnedChunks regions
63 */
64
65 /*
66 * TODO:
67
68 *) we could have a function pointer in MonoClass to implement
69 customized write barriers for value types
70
71 *) investigate the stuff needed to advance a thread to a GC-safe
72 point (single-stepping, read from unmapped memory etc) and implement it.
73 This would enable us to inline allocations and write barriers, for example,
74 or at least parts of them, like the write barrier checks.
75 We may need this also for handling precise info on stacks, even simple things
76 as having uninitialized data on the stack and having to wait for the prolog
77 to zero it. Not an issue for the last frame that we scan conservatively.
78 We could always not trust the value in the slots anyway.
79
80 *) modify the jit to save info about references in stack locations:
81 this can be done just for locals as a start, so that at least
82 part of the stack is handled precisely.
83
84 *) test/fix endianess issues
85
86 *) Implement a card table as the write barrier instead of remembered
87 sets? Card tables are not easy to implement with our current
88 memory layout. We have several different kinds of major heap
89 objects: Small objects in regular blocks, small objects in pinned
90 chunks and LOS objects. If we just have a pointer we have no way
91 to tell which kind of object it points into, therefore we cannot
92 know where its card table is. The least we have to do to make
93 this happen is to get rid of write barriers for indirect stores.
94 (See next item)
95
96 *) Get rid of write barriers for indirect stores. We can do this by
97 telling the GC to wbarrier-register an object once we do an ldloca
98 or ldelema on it, and to unregister it once it's not used anymore
99 (it can only travel downwards on the stack). The problem with
100 unregistering is that it needs to happen eventually no matter
101 what, even if exceptions are thrown, the thread aborts, etc.
102 Rodrigo suggested that we could do only the registering part and
103 let the collector find out (pessimistically) when it's safe to
104 unregister, namely when the stack pointer of the thread that
105 registered the object is higher than it was when the registering
106 happened. This might make for a good first implementation to get
107 some data on performance.
108
109 *) Some sort of blacklist support? Blacklists is a concept from the
110 Boehm GC: if during a conservative scan we find pointers to an
111 area which we might use as heap, we mark that area as unusable, so
112 pointer retention by random pinning pointers is reduced.
113
114 *) experiment with max small object size (very small right now - 2kb,
115 because it's tied to the max freelist size)
116
117 *) add an option to mmap the whole heap in one chunk: it makes for many
118 simplifications in the checks (put the nursery at the top and just use a single
119 check for inclusion/exclusion): the issue this has is that on 32 bit systems it's
120 not flexible (too much of the address space may be used by default or we can't
121 increase the heap as needed) and we'd need a race-free mechanism to return memory
122 back to the system (mprotect(PROT_NONE) will still keep the memory allocated if it
123 was written to, munmap is needed, but the following mmap may not find the same segment
124 free...)
125
126 *) memzero the major fragments after restarting the world and optionally a smaller
127 chunk at a time
128
129 *) investigate having fragment zeroing threads
130
131 *) separate locks for finalization and other minor stuff to reduce
132 lock contention
133
134 *) try a different copying order to improve memory locality
135
136 *) a thread abort after a store but before the write barrier will
137 prevent the write barrier from executing
138
139 *) specialized dynamically generated markers/copiers
140
141 *) Dynamically adjust TLAB size to the number of threads. If we have
142 too many threads that do allocation, we might need smaller TLABs,
143 and we might get better performance with larger TLABs if we only
144 have a handful of threads. We could sum up the space left in all
145 assigned TLABs and if that's more than some percentage of the
146 nursery size, reduce the TLAB size.
147
148 *) Explore placing unreachable objects on unused nursery memory.
149 Instead of memset'ng a region to zero, place an int[] covering it.
150 A good place to start is add_nursery_frag. The tricky thing here is
151 placing those objects atomically outside of a collection.
152
153 *) Allocation should use asymmetric Dekker synchronization:
154 http://blogs.oracle.com/dave/resource/Asymmetric-Dekker-Synchronization.txt
155 This should help weak consistency archs.
156 */
157 #include "config.h"
158 #ifdef HAVE_SGEN_GC
159
160 #ifdef __MACH__
161 #undef _XOPEN_SOURCE
162 #define _XOPEN_SOURCE
163 #define _DARWIN_C_SOURCE
164 #endif
165
166 #ifdef HAVE_UNISTD_H
167 #include <unistd.h>
168 #endif
169 #ifdef HAVE_PTHREAD_H
170 #include <pthread.h>
171 #endif
172 #ifdef HAVE_PTHREAD_NP_H
173 #include <pthread_np.h>
174 #endif
175 #include <stdio.h>
176 #include <string.h>
177 #include <errno.h>
178 #include <assert.h>
179 #include <stdlib.h>
180 #include <glib.h>
181
182 #include "mono/sgen/sgen-gc.h"
183 #include "mono/sgen/sgen-cardtable.h"
184 #include "mono/sgen/sgen-protocol.h"
185 #include "mono/sgen/sgen-memory-governor.h"
186 #include "mono/sgen/sgen-hash-table.h"
187 #include "mono/sgen/sgen-pinning.h"
188 #include "mono/sgen/sgen-workers.h"
189 #include "mono/sgen/sgen-client.h"
190 #include "mono/sgen/sgen-pointer-queue.h"
191 #include "mono/sgen/gc-internal-agnostic.h"
192 #include "mono/utils/mono-proclib.h"
193 #include "mono/utils/mono-memory-model.h"
194 #include "mono/utils/hazard-pointer.h"
195
196 #include <mono/utils/memcheck.h>
197 #include <mono/utils/mono-mmap-internals.h>
198 #include <mono/utils/unlocked.h>
199
200 #undef pthread_create
201 #undef pthread_join
202 #undef pthread_detach
203
204 /*
205 * ######################################################################
206 * ######## Types and constants used by the GC.
207 * ######################################################################
208 */
209
210 /* 0 means not initialized, 1 is initialized, -1 means in progress */
211 static int gc_initialized = 0;
212 /* If set, check if we need to do something every X allocations */
213 gboolean has_per_allocation_action;
214 /* If set, do a heap check every X allocation */
215 guint32 verify_before_allocs = 0;
216 /* If set, do a minor collection before every X allocation */
217 guint32 collect_before_allocs = 0;
218 /* If set, do a whole heap check before each collection */
219 static gboolean whole_heap_check_before_collection = FALSE;
220 /* If set, do a remset consistency check at various opportunities */
221 static gboolean remset_consistency_checks = FALSE;
222 /* If set, do a mod union consistency check before each finishing collection pause */
223 static gboolean mod_union_consistency_check = FALSE;
224 /* If set, check whether mark bits are consistent after major collections */
225 static gboolean check_mark_bits_after_major_collection = FALSE;
226 /* If set, check that all nursery objects are pinned/not pinned, depending on context */
227 static gboolean check_nursery_objects_pinned = FALSE;
228 /* If set, do a few checks when the concurrent collector is used */
229 static gboolean do_concurrent_checks = FALSE;
230 /* If set, do a plausibility check on the scan_starts before and after
231 each collection */
232 static gboolean do_scan_starts_check = FALSE;
233
234 static gboolean disable_minor_collections = FALSE;
235 static gboolean disable_major_collections = FALSE;
236 static gboolean do_verify_nursery = FALSE;
237 static gboolean do_dump_nursery_content = FALSE;
238 static gboolean enable_nursery_canaries = FALSE;
239
240 static gboolean precleaning_enabled = TRUE;
241 static gboolean dynamic_nursery = FALSE;
242 static size_t min_nursery_size = 0;
243 static size_t max_nursery_size = 0;
244
245 #ifdef HEAVY_STATISTICS
246 guint64 stat_objects_alloced_degraded = 0;
247 guint64 stat_bytes_alloced_degraded = 0;
248
249 guint64 stat_copy_object_called_nursery = 0;
250 guint64 stat_objects_copied_nursery = 0;
251 guint64 stat_copy_object_called_major = 0;
252 guint64 stat_objects_copied_major = 0;
253
254 guint64 stat_scan_object_called_nursery = 0;
255 guint64 stat_scan_object_called_major = 0;
256
257 guint64 stat_slots_allocated_in_vain;
258
259 guint64 stat_nursery_copy_object_failed_from_space = 0;
260 guint64 stat_nursery_copy_object_failed_forwarded = 0;
261 guint64 stat_nursery_copy_object_failed_pinned = 0;
262 guint64 stat_nursery_copy_object_failed_to_space = 0;
263
264 static guint64 stat_wbarrier_add_to_global_remset = 0;
265 static guint64 stat_wbarrier_arrayref_copy = 0;
266 static guint64 stat_wbarrier_generic_store = 0;
267 static guint64 stat_wbarrier_generic_store_atomic = 0;
268 static guint64 stat_wbarrier_set_root = 0;
269 #endif
270
271 static guint64 stat_pinned_objects = 0;
272
273 static guint64 time_minor_pre_collection_fragment_clear = 0;
274 static guint64 time_minor_pinning = 0;
275 static guint64 time_minor_scan_remsets = 0;
276 static guint64 time_minor_scan_major_blocks = 0;
277 static guint64 time_minor_scan_los = 0;
278 static guint64 time_minor_scan_pinned = 0;
279 static guint64 time_minor_scan_roots = 0;
280 static guint64 time_minor_finish_gray_stack = 0;
281 static guint64 time_minor_fragment_creation = 0;
282
283 static guint64 time_major_pre_collection_fragment_clear = 0;
284 static guint64 time_major_pinning = 0;
285 static guint64 time_major_scan_pinned = 0;
286 static guint64 time_major_scan_roots = 0;
287 static guint64 time_major_scan_mod_union_blocks = 0;
288 static guint64 time_major_scan_mod_union_los = 0;
289 static guint64 time_major_finish_gray_stack = 0;
290 static guint64 time_major_free_bigobjs = 0;
291 static guint64 time_major_los_sweep = 0;
292 static guint64 time_major_sweep = 0;
293 static guint64 time_major_fragment_creation = 0;
294
295 static guint64 time_max = 0;
296
297 static int sgen_max_pause_time = SGEN_DEFAULT_MAX_PAUSE_TIME;
298 static float sgen_max_pause_margin = SGEN_DEFAULT_MAX_PAUSE_MARGIN;
299
300 static SGEN_TV_DECLARE (time_major_conc_collection_start);
301 static SGEN_TV_DECLARE (time_major_conc_collection_end);
302
303 int gc_debug_level = 0;
304 FILE* gc_debug_file;
305 static char* gc_params_options;
306 static char* gc_debug_options;
307
308 /*
309 void
310 mono_gc_flush_info (void)
311 {
312 fflush (gc_debug_file);
313 }
314 */
315
316 #define TV_DECLARE SGEN_TV_DECLARE
317 #define TV_GETTIME SGEN_TV_GETTIME
318 #define TV_ELAPSED SGEN_TV_ELAPSED
319
320 static SGEN_TV_DECLARE (sgen_init_timestamp);
321
322 NurseryClearPolicy nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
323
324 #define object_is_forwarded SGEN_OBJECT_IS_FORWARDED
325 #define object_is_pinned SGEN_OBJECT_IS_PINNED
326 #define pin_object SGEN_PIN_OBJECT
327
328 #define ptr_in_nursery sgen_ptr_in_nursery
329
330 #define LOAD_VTABLE SGEN_LOAD_VTABLE
331
332 gboolean
nursery_canaries_enabled(void)333 nursery_canaries_enabled (void)
334 {
335 return enable_nursery_canaries;
336 }
337
338 #define safe_object_get_size sgen_safe_object_get_size
339
340 typedef enum {
341 SGEN_MAJOR_DEFAULT,
342 SGEN_MAJOR_SERIAL,
343 SGEN_MAJOR_CONCURRENT,
344 SGEN_MAJOR_CONCURRENT_PARALLEL
345 } SgenMajor;
346
347 typedef enum {
348 SGEN_MINOR_DEFAULT,
349 SGEN_MINOR_SIMPLE,
350 SGEN_MINOR_SIMPLE_PARALLEL,
351 SGEN_MINOR_SPLIT
352 } SgenMinor;
353
354 typedef enum {
355 SGEN_MODE_NONE,
356 SGEN_MODE_BALANCED,
357 SGEN_MODE_THROUGHPUT,
358 SGEN_MODE_PAUSE
359 } SgenMode;
360
361 /*
362 * ######################################################################
363 * ######## Global data.
364 * ######################################################################
365 */
366 MonoCoopMutex gc_mutex;
367
368 #define SCAN_START_SIZE SGEN_SCAN_START_SIZE
369
370 size_t degraded_mode = 0;
371
372 static mword bytes_pinned_from_failed_allocation = 0;
373
374 GCMemSection *nursery_section = NULL;
375 static volatile mword lowest_heap_address = ~(mword)0;
376 static volatile mword highest_heap_address = 0;
377
378 MonoCoopMutex sgen_interruption_mutex;
379
380 int current_collection_generation = -1;
381 volatile gboolean concurrent_collection_in_progress = FALSE;
382
383 /* objects that are ready to be finalized */
384 static SgenPointerQueue fin_ready_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
385 static SgenPointerQueue critical_fin_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
386
387 /* registered roots: the key to the hash is the root start address */
388 /*
389 * Different kinds of roots are kept separate to speed up pin_from_roots () for example.
390 */
391 SgenHashTable roots_hash [ROOT_TYPE_NUM] = {
392 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
393 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
394 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL)
395 };
396 static mword roots_size = 0; /* amount of memory in the root set */
397
398 /* The size of a TLAB */
399 /* The bigger the value, the less often we have to go to the slow path to allocate a new
400 * one, but the more space is wasted by threads not allocating much memory.
401 * FIXME: Tune this.
402 * FIXME: Make this self-tuning for each thread.
403 */
404 guint32 tlab_size = (1024 * 4);
405
406 #define MAX_SMALL_OBJ_SIZE SGEN_MAX_SMALL_OBJ_SIZE
407
408 #define ALLOC_ALIGN SGEN_ALLOC_ALIGN
409
410 #define ALIGN_UP SGEN_ALIGN_UP
411
412 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
413 MonoNativeThreadId main_gc_thread = NULL;
414 #endif
415
416 /*Object was pinned during the current collection*/
417 static mword objects_pinned;
418
419 /*
420 * ######################################################################
421 * ######## Macros and function declarations.
422 * ######################################################################
423 */
424
425 /* forward declarations */
426 static void scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx);
427
428 static void pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx);
429 static void finish_gray_stack (int generation, ScanCopyContext ctx);
430
431
432 SgenMajorCollector major_collector;
433 SgenMinorCollector sgen_minor_collector;
434
435 static SgenRememberedSet remset;
436
437 /*
438 * The gray queue a worker job must use. If we're not parallel or
439 * concurrent, we use the main gray queue.
440 */
441 static SgenGrayQueue*
sgen_workers_get_job_gray_queue(WorkerData * worker_data,SgenGrayQueue * default_gray_queue)442 sgen_workers_get_job_gray_queue (WorkerData *worker_data, SgenGrayQueue *default_gray_queue)
443 {
444 if (worker_data)
445 return &worker_data->private_gray_queue;
446 SGEN_ASSERT (0, default_gray_queue, "Why don't we have a default gray queue when we're not running in a worker thread?");
447 return default_gray_queue;
448 }
449
450 static void
gray_queue_redirect(SgenGrayQueue * queue)451 gray_queue_redirect (SgenGrayQueue *queue)
452 {
453 sgen_workers_take_from_queue (current_collection_generation, queue);
454 }
455
456 void
sgen_scan_area_with_callback(char * start,char * end,IterateObjectCallbackFunc callback,void * data,gboolean allow_flags,gboolean fail_on_canaries)457 sgen_scan_area_with_callback (char *start, char *end, IterateObjectCallbackFunc callback, void *data, gboolean allow_flags, gboolean fail_on_canaries)
458 {
459 while (start < end) {
460 size_t size;
461 char *obj;
462
463 if (!*(void**)start) {
464 start += sizeof (void*); /* should be ALLOC_ALIGN, really */
465 continue;
466 }
467
468 if (allow_flags) {
469 if (!(obj = (char *)SGEN_OBJECT_IS_FORWARDED (start)))
470 obj = start;
471 } else {
472 obj = start;
473 }
474
475 if (!sgen_client_object_is_array_fill ((GCObject*)obj)) {
476 CHECK_CANARY_FOR_OBJECT ((GCObject*)obj, fail_on_canaries);
477 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
478 callback ((GCObject*)obj, size, data);
479 CANARIFY_SIZE (size);
480 } else {
481 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
482 }
483
484 start += size;
485 }
486 }
487
488 /*
489 * sgen_add_to_global_remset:
490 *
491 * The global remset contains locations which point into newspace after
492 * a minor collection. This can happen if the objects they point to are pinned.
493 *
494 * LOCKING: If called from a parallel collector, the global remset
495 * lock must be held. For serial collectors that is not necessary.
496 */
497 void
sgen_add_to_global_remset(gpointer ptr,GCObject * obj)498 sgen_add_to_global_remset (gpointer ptr, GCObject *obj)
499 {
500 SGEN_ASSERT (5, sgen_ptr_in_nursery (obj), "Target pointer of global remset must be in the nursery");
501
502 HEAVY_STAT (++stat_wbarrier_add_to_global_remset);
503
504 if (!major_collector.is_concurrent) {
505 SGEN_ASSERT (5, current_collection_generation != -1, "Global remsets can only be added during collections");
506 } else {
507 if (current_collection_generation == -1)
508 SGEN_ASSERT (5, sgen_concurrent_collection_in_progress (), "Global remsets outside of collection pauses can only be added by the concurrent collector");
509 }
510
511 if (!object_is_pinned (obj))
512 SGEN_ASSERT (5, sgen_minor_collector.is_split || sgen_concurrent_collection_in_progress (), "Non-pinned objects can only remain in nursery if it is a split nursery");
513 else if (sgen_cement_lookup_or_register (obj))
514 return;
515
516 remset.record_pointer (ptr);
517
518 sgen_pin_stats_register_global_remset (obj);
519
520 SGEN_LOG (8, "Adding global remset for %p", ptr);
521 binary_protocol_global_remset (ptr, obj, (gpointer)SGEN_LOAD_VTABLE (obj));
522 }
523
524 /*
525 * sgen_drain_gray_stack:
526 *
527 * Scan objects in the gray stack until the stack is empty. This should be called
528 * frequently after each object is copied, to achieve better locality and cache
529 * usage.
530 *
531 */
532 gboolean
sgen_drain_gray_stack(ScanCopyContext ctx)533 sgen_drain_gray_stack (ScanCopyContext ctx)
534 {
535 SGEN_ASSERT (0, ctx.ops->drain_gray_stack, "Why do we have a scan/copy context with a missing drain gray stack function?");
536
537 return ctx.ops->drain_gray_stack (ctx.queue);
538 }
539
540 /*
541 * Addresses in the pin queue are already sorted. This function finds
542 * the object header for each address and pins the object. The
543 * addresses must be inside the nursery section. The (start of the)
544 * address array is overwritten with the addresses of the actually
545 * pinned objects. Return the number of pinned objects.
546 */
547 static int
pin_objects_from_nursery_pin_queue(gboolean do_scan_objects,ScanCopyContext ctx)548 pin_objects_from_nursery_pin_queue (gboolean do_scan_objects, ScanCopyContext ctx)
549 {
550 GCMemSection *section = nursery_section;
551 void **start = sgen_pinning_get_entry (section->pin_queue_first_entry);
552 void **end = sgen_pinning_get_entry (section->pin_queue_last_entry);
553 void *start_nursery = section->data;
554 void *end_nursery = section->end_data;
555 void *last = NULL;
556 int count = 0;
557 void *search_start;
558 void *addr;
559 void *pinning_front = start_nursery;
560 size_t idx;
561 void **definitely_pinned = start;
562 ScanObjectFunc scan_func = ctx.ops->scan_object;
563 SgenGrayQueue *queue = ctx.queue;
564
565 sgen_nursery_allocator_prepare_for_pinning ();
566
567 while (start < end) {
568 GCObject *obj_to_pin = NULL;
569 size_t obj_to_pin_size = 0;
570 SgenDescriptor desc;
571
572 addr = *start;
573
574 SGEN_ASSERT (0, addr >= start_nursery && addr < end_nursery, "Potential pinning address out of range");
575 SGEN_ASSERT (0, addr >= last, "Pin queue not sorted");
576
577 if (addr == last) {
578 ++start;
579 continue;
580 }
581
582 SGEN_LOG (5, "Considering pinning addr %p", addr);
583 /* We've already processed everything up to pinning_front. */
584 if (addr < pinning_front) {
585 start++;
586 continue;
587 }
588
589 /*
590 * Find the closest scan start <= addr. We might search backward in the
591 * scan_starts array because entries might be NULL. In the worst case we
592 * start at start_nursery.
593 */
594 idx = ((char*)addr - (char*)section->data) / SCAN_START_SIZE;
595 SGEN_ASSERT (0, idx < section->num_scan_start, "Scan start index out of range");
596 search_start = (void*)section->scan_starts [idx];
597 if (!search_start || search_start > addr) {
598 while (idx) {
599 --idx;
600 search_start = section->scan_starts [idx];
601 if (search_start && search_start <= addr)
602 break;
603 }
604 if (!search_start || search_start > addr)
605 search_start = start_nursery;
606 }
607
608 /*
609 * If the pinning front is closer than the scan start we found, start
610 * searching at the front.
611 */
612 if (search_start < pinning_front)
613 search_start = pinning_front;
614
615 /*
616 * Now addr should be in an object a short distance from search_start.
617 *
618 * search_start must point to zeroed mem or point to an object.
619 */
620 do {
621 size_t obj_size, canarified_obj_size;
622
623 /* Skip zeros. */
624 if (!*(void**)search_start) {
625 search_start = (void*)ALIGN_UP ((mword)search_start + sizeof (gpointer));
626 /* The loop condition makes sure we don't overrun addr. */
627 continue;
628 }
629
630 canarified_obj_size = obj_size = ALIGN_UP (safe_object_get_size ((GCObject*)search_start));
631
632 /*
633 * Filler arrays are marked by an invalid sync word. We don't
634 * consider them for pinning. They are not delimited by canaries,
635 * either.
636 */
637 if (!sgen_client_object_is_array_fill ((GCObject*)search_start)) {
638 CHECK_CANARY_FOR_OBJECT (search_start, TRUE);
639 CANARIFY_SIZE (canarified_obj_size);
640
641 if (addr >= search_start && (char*)addr < (char*)search_start + obj_size) {
642 /* This is the object we're looking for. */
643 obj_to_pin = (GCObject*)search_start;
644 obj_to_pin_size = canarified_obj_size;
645 break;
646 }
647 }
648
649 /* Skip to the next object */
650 search_start = (void*)((char*)search_start + canarified_obj_size);
651 } while (search_start <= addr);
652
653 /* We've searched past the address we were looking for. */
654 if (!obj_to_pin) {
655 pinning_front = search_start;
656 goto next_pin_queue_entry;
657 }
658
659 /*
660 * We've found an object to pin. It might still be a dummy array, but we
661 * can advance the pinning front in any case.
662 */
663 pinning_front = (char*)obj_to_pin + obj_to_pin_size;
664
665 /*
666 * If this is a dummy array marking the beginning of a nursery
667 * fragment, we don't pin it.
668 */
669 if (sgen_client_object_is_array_fill (obj_to_pin))
670 goto next_pin_queue_entry;
671
672 /*
673 * Finally - pin the object!
674 */
675 desc = sgen_obj_get_descriptor_safe (obj_to_pin);
676 if (do_scan_objects) {
677 scan_func (obj_to_pin, desc, queue);
678 } else {
679 SGEN_LOG (4, "Pinned object %p, vtable %p (%s), count %d\n",
680 obj_to_pin, *(void**)obj_to_pin, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj_to_pin)), count);
681 binary_protocol_pin (obj_to_pin,
682 (gpointer)LOAD_VTABLE (obj_to_pin),
683 safe_object_get_size (obj_to_pin));
684
685 pin_object (obj_to_pin);
686 GRAY_OBJECT_ENQUEUE_SERIAL (queue, obj_to_pin, desc);
687 sgen_pin_stats_register_object (obj_to_pin, GENERATION_NURSERY);
688 definitely_pinned [count] = obj_to_pin;
689 count++;
690 }
691 if (concurrent_collection_in_progress)
692 sgen_pinning_register_pinned_in_nursery (obj_to_pin);
693
694 next_pin_queue_entry:
695 last = addr;
696 ++start;
697 }
698 sgen_client_nursery_objects_pinned (definitely_pinned, count);
699 stat_pinned_objects += count;
700 return count;
701 }
702
703 static void
pin_objects_in_nursery(gboolean do_scan_objects,ScanCopyContext ctx)704 pin_objects_in_nursery (gboolean do_scan_objects, ScanCopyContext ctx)
705 {
706 size_t reduced_to;
707
708 if (nursery_section->pin_queue_first_entry == nursery_section->pin_queue_last_entry)
709 return;
710
711 reduced_to = pin_objects_from_nursery_pin_queue (do_scan_objects, ctx);
712 nursery_section->pin_queue_last_entry = nursery_section->pin_queue_first_entry + reduced_to;
713 }
714
715 /*
716 * This function is only ever called (via `collector_pin_object()` in `sgen-copy-object.h`)
717 * when we can't promote an object because we're out of memory.
718 */
719 void
sgen_pin_object(GCObject * object,SgenGrayQueue * queue)720 sgen_pin_object (GCObject *object, SgenGrayQueue *queue)
721 {
722 SGEN_ASSERT (0, sgen_ptr_in_nursery (object), "We're only supposed to use this for pinning nursery objects when out of memory.");
723
724 /*
725 * All pinned objects are assumed to have been staged, so we need to stage as well.
726 * Also, the count of staged objects shows that "late pinning" happened.
727 */
728 sgen_pin_stage_ptr (object);
729
730 SGEN_PIN_OBJECT (object);
731 binary_protocol_pin (object, (gpointer)LOAD_VTABLE (object), safe_object_get_size (object));
732
733 ++objects_pinned;
734 sgen_pin_stats_register_object (object, GENERATION_NURSERY);
735
736 GRAY_OBJECT_ENQUEUE_SERIAL (queue, object, sgen_obj_get_descriptor_safe (object));
737 }
738
739 /* Sort the addresses in array in increasing order.
740 * Done using a by-the book heap sort. Which has decent and stable performance, is pretty cache efficient.
741 */
742 void
sgen_sort_addresses(void ** array,size_t size)743 sgen_sort_addresses (void **array, size_t size)
744 {
745 size_t i;
746 void *tmp;
747
748 for (i = 1; i < size; ++i) {
749 size_t child = i;
750 while (child > 0) {
751 size_t parent = (child - 1) / 2;
752
753 if (array [parent] >= array [child])
754 break;
755
756 tmp = array [parent];
757 array [parent] = array [child];
758 array [child] = tmp;
759
760 child = parent;
761 }
762 }
763
764 for (i = size - 1; i > 0; --i) {
765 size_t end, root;
766 tmp = array [i];
767 array [i] = array [0];
768 array [0] = tmp;
769
770 end = i - 1;
771 root = 0;
772
773 while (root * 2 + 1 <= end) {
774 size_t child = root * 2 + 1;
775
776 if (child < end && array [child] < array [child + 1])
777 ++child;
778 if (array [root] >= array [child])
779 break;
780
781 tmp = array [root];
782 array [root] = array [child];
783 array [child] = tmp;
784
785 root = child;
786 }
787 }
788 }
789
790 /*
791 * Scan the memory between start and end and queue values which could be pointers
792 * to the area between start_nursery and end_nursery for later consideration.
793 * Typically used for thread stacks.
794 */
795 MONO_NO_SANITIZE_ADDRESS
796 void
sgen_conservatively_pin_objects_from(void ** start,void ** end,void * start_nursery,void * end_nursery,int pin_type)797 sgen_conservatively_pin_objects_from (void **start, void **end, void *start_nursery, void *end_nursery, int pin_type)
798 {
799 int count = 0;
800
801 SGEN_ASSERT (0, ((mword)start & (SIZEOF_VOID_P - 1)) == 0, "Why are we scanning for references in unaligned memory ?");
802
803 #if defined(VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE) && !defined(_WIN64)
804 VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE (start, (char*)end - (char*)start);
805 #endif
806
807 while (start < end) {
808 /*
809 * *start can point to the middle of an object
810 * note: should we handle pointing at the end of an object?
811 * pinning in C# code disallows pointing at the end of an object
812 * but there is some small chance that an optimizing C compiler
813 * may keep the only reference to an object by pointing
814 * at the end of it. We ignore this small chance for now.
815 * Pointers to the end of an object are indistinguishable
816 * from pointers to the start of the next object in memory
817 * so if we allow that we'd need to pin two objects...
818 * We queue the pointer in an array, the
819 * array will then be sorted and uniqued. This way
820 * we can coalesce several pinning pointers and it should
821 * be faster since we'd do a memory scan with increasing
822 * addresses. Note: we can align the address to the allocation
823 * alignment, so the unique process is more effective.
824 */
825 mword addr = (mword)*start;
826 addr &= ~(ALLOC_ALIGN - 1);
827 if (addr >= (mword)start_nursery && addr < (mword)end_nursery) {
828 SGEN_LOG (6, "Pinning address %p from %p", (void*)addr, start);
829 sgen_pin_stage_ptr ((void*)addr);
830 binary_protocol_pin_stage (start, (void*)addr);
831 sgen_pin_stats_register_address ((char*)addr, pin_type);
832 count++;
833 }
834 start++;
835 }
836 if (count)
837 SGEN_LOG (7, "found %d potential pinned heap pointers", count);
838 }
839
840 /*
841 * The first thing we do in a collection is to identify pinned objects.
842 * This function considers all the areas of memory that need to be
843 * conservatively scanned.
844 */
845 static void
pin_from_roots(void * start_nursery,void * end_nursery,ScanCopyContext ctx)846 pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx)
847 {
848 void **start_root;
849 RootRecord *root;
850 SGEN_LOG (2, "Scanning pinned roots (%d bytes, %d/%d entries)", (int)roots_size, roots_hash [ROOT_TYPE_NORMAL].num_entries, roots_hash [ROOT_TYPE_PINNED].num_entries);
851 /* objects pinned from the API are inside these roots */
852 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_PINNED], void **, start_root, RootRecord *, root) {
853 SGEN_LOG (6, "Pinned roots %p-%p", start_root, root->end_root);
854 sgen_conservatively_pin_objects_from (start_root, (void**)root->end_root, start_nursery, end_nursery, PIN_TYPE_OTHER);
855 } SGEN_HASH_TABLE_FOREACH_END;
856 /* now deal with the thread stacks
857 * in the future we should be able to conservatively scan only:
858 * *) the cpu registers
859 * *) the unmanaged stack frames
860 * *) the _last_ managed stack frame
861 * *) pointers slots in managed frames
862 */
863 sgen_client_scan_thread_data (start_nursery, end_nursery, FALSE, ctx);
864 }
865
866 static void
single_arg_user_copy_or_mark(GCObject ** obj,void * gc_data)867 single_arg_user_copy_or_mark (GCObject **obj, void *gc_data)
868 {
869 ScanCopyContext *ctx = (ScanCopyContext *)gc_data;
870 ctx->ops->copy_or_mark_object (obj, ctx->queue);
871 }
872
873 /*
874 * The memory area from start_root to end_root contains pointers to objects.
875 * Their position is precisely described by @desc (this means that the pointer
876 * can be either NULL or the pointer to the start of an object).
877 * This functions copies them to to_space updates them.
878 *
879 * This function is not thread-safe!
880 */
881 static void
precisely_scan_objects_from(void ** start_root,void ** end_root,char * n_start,char * n_end,SgenDescriptor desc,ScanCopyContext ctx)882 precisely_scan_objects_from (void** start_root, void** end_root, char* n_start, char *n_end, SgenDescriptor desc, ScanCopyContext ctx)
883 {
884 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
885 ScanPtrFieldFunc scan_field_func = ctx.ops->scan_ptr_field;
886 SgenGrayQueue *queue = ctx.queue;
887
888 switch (desc & ROOT_DESC_TYPE_MASK) {
889 case ROOT_DESC_BITMAP:
890 desc >>= ROOT_DESC_TYPE_SHIFT;
891 while (desc) {
892 if ((desc & 1) && *start_root) {
893 copy_func ((GCObject**)start_root, queue);
894 SGEN_LOG (9, "Overwrote root at %p with %p", start_root, *start_root);
895 }
896 desc >>= 1;
897 start_root++;
898 }
899 return;
900 case ROOT_DESC_COMPLEX: {
901 gsize *bitmap_data = (gsize *)sgen_get_complex_descriptor_bitmap (desc);
902 gsize bwords = (*bitmap_data) - 1;
903 void **start_run = start_root;
904 bitmap_data++;
905 while (bwords-- > 0) {
906 gsize bmap = *bitmap_data++;
907 void **objptr = start_run;
908 while (bmap) {
909 if ((bmap & 1) && *objptr) {
910 copy_func ((GCObject**)objptr, queue);
911 SGEN_LOG (9, "Overwrote root at %p with %p", objptr, *objptr);
912 }
913 bmap >>= 1;
914 ++objptr;
915 }
916 start_run += GC_BITS_PER_WORD;
917 }
918 break;
919 }
920 case ROOT_DESC_VECTOR: {
921 void **p;
922
923 for (p = start_root; p < end_root; p++) {
924 if (*p)
925 scan_field_func (NULL, (GCObject**)p, queue);
926 }
927 break;
928 }
929 case ROOT_DESC_USER: {
930 SgenUserRootMarkFunc marker = sgen_get_user_descriptor_func (desc);
931 marker (start_root, single_arg_user_copy_or_mark, &ctx);
932 break;
933 }
934 case ROOT_DESC_RUN_LEN:
935 g_assert_not_reached ();
936 default:
937 g_assert_not_reached ();
938 }
939 }
940
941 static void
reset_heap_boundaries(void)942 reset_heap_boundaries (void)
943 {
944 lowest_heap_address = ~(mword)0;
945 highest_heap_address = 0;
946 }
947
948 void
sgen_update_heap_boundaries(mword low,mword high)949 sgen_update_heap_boundaries (mword low, mword high)
950 {
951 mword old;
952
953 do {
954 old = lowest_heap_address;
955 if (low >= old)
956 break;
957 } while (SGEN_CAS_PTR ((gpointer*)&lowest_heap_address, (gpointer)low, (gpointer)old) != (gpointer)old);
958
959 do {
960 old = highest_heap_address;
961 if (high <= old)
962 break;
963 } while (SGEN_CAS_PTR ((gpointer*)&highest_heap_address, (gpointer)high, (gpointer)old) != (gpointer)old);
964 }
965
966 /*
967 * Allocate and setup the data structures needed to be able to allocate objects
968 * in the nursery. The nursery is stored in nursery_section.
969 */
970 static void
alloc_nursery(gboolean dynamic,size_t min_size,size_t max_size)971 alloc_nursery (gboolean dynamic, size_t min_size, size_t max_size)
972 {
973 char *data;
974 size_t scan_starts;
975
976 if (dynamic) {
977 if (!min_size)
978 min_size = SGEN_DEFAULT_NURSERY_MIN_SIZE;
979 if (!max_size)
980 max_size = SGEN_DEFAULT_NURSERY_MAX_SIZE;
981 } else {
982 SGEN_ASSERT (0, min_size == max_size, "We can't have nursery ranges for static configuration.");
983 if (!min_size)
984 min_size = max_size = SGEN_DEFAULT_NURSERY_SIZE;
985 }
986
987 SGEN_ASSERT (0, !nursery_section, "Why are we allocating the nursery twice?");
988 SGEN_LOG (2, "Allocating nursery size: %zu, initial %zu", max_size, min_size);
989
990 /* FIXME: handle OOM */
991 nursery_section = (GCMemSection *)sgen_alloc_internal (INTERNAL_MEM_SECTION);
992
993 /* If there isn't enough space even for the nursery we should simply abort. */
994 g_assert (sgen_memgov_try_alloc_space (max_size, SPACE_NURSERY));
995
996 /*
997 * The nursery section range represents the memory section where objects
998 * can be found. This is used when iterating for objects in the nursery,
999 * pinning etc. sgen_nursery_max_size represents the total allocated space
1000 * for the nursery. sgen_nursery_size represents the current size of the
1001 * nursery and it is used for allocation limits, heuristics etc. The
1002 * nursery section is not always identical to the current nursery size
1003 * because it can contain pinned objects from when the nursery was larger.
1004 *
1005 * sgen_nursery_size <= nursery_section size <= sgen_nursery_max_size
1006 */
1007 data = (char *)major_collector.alloc_heap (max_size, max_size);
1008 sgen_update_heap_boundaries ((mword)data, (mword)(data + max_size));
1009 nursery_section->data = data;
1010 nursery_section->end_data = data + min_size;
1011 scan_starts = (max_size + SCAN_START_SIZE - 1) / SCAN_START_SIZE;
1012 nursery_section->scan_starts = (char **)sgen_alloc_internal_dynamic (sizeof (char*) * scan_starts, INTERNAL_MEM_SCAN_STARTS, TRUE);
1013 nursery_section->num_scan_start = scan_starts;
1014
1015 sgen_nursery_allocator_set_nursery_bounds (data, min_size, max_size);
1016 }
1017
1018 FILE *
mono_gc_get_logfile(void)1019 mono_gc_get_logfile (void)
1020 {
1021 return gc_debug_file;
1022 }
1023
1024 void
mono_gc_params_set(const char * options)1025 mono_gc_params_set (const char* options)
1026 {
1027 if (gc_params_options)
1028 g_free (gc_params_options);
1029
1030 gc_params_options = g_strdup (options);
1031 }
1032
1033 void
mono_gc_debug_set(const char * options)1034 mono_gc_debug_set (const char* options)
1035 {
1036 if (gc_debug_options)
1037 g_free (gc_debug_options);
1038
1039 gc_debug_options = g_strdup (options);
1040 }
1041
1042 static void
scan_finalizer_entries(SgenPointerQueue * fin_queue,ScanCopyContext ctx)1043 scan_finalizer_entries (SgenPointerQueue *fin_queue, ScanCopyContext ctx)
1044 {
1045 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
1046 SgenGrayQueue *queue = ctx.queue;
1047 size_t i;
1048
1049 for (i = 0; i < fin_queue->next_slot; ++i) {
1050 GCObject *obj = (GCObject *)fin_queue->data [i];
1051 if (!obj)
1052 continue;
1053 SGEN_LOG (5, "Scan of fin ready object: %p (%s)\n", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
1054 copy_func ((GCObject**)&fin_queue->data [i], queue);
1055 }
1056 }
1057
1058 static const char*
generation_name(int generation)1059 generation_name (int generation)
1060 {
1061 switch (generation) {
1062 case GENERATION_NURSERY: return "nursery";
1063 case GENERATION_OLD: return "old";
1064 default: g_assert_not_reached ();
1065 }
1066 }
1067
1068 const char*
sgen_generation_name(int generation)1069 sgen_generation_name (int generation)
1070 {
1071 return generation_name (generation);
1072 }
1073
1074 static void
finish_gray_stack(int generation,ScanCopyContext ctx)1075 finish_gray_stack (int generation, ScanCopyContext ctx)
1076 {
1077 TV_DECLARE (atv);
1078 TV_DECLARE (btv);
1079 int done_with_ephemerons, ephemeron_rounds = 0;
1080 char *start_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_start () : NULL;
1081 char *end_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_end () : (char*)-1;
1082 SgenGrayQueue *queue = ctx.queue;
1083
1084 binary_protocol_finish_gray_stack_start (sgen_timestamp (), generation);
1085 /*
1086 * We copied all the reachable objects. Now it's the time to copy
1087 * the objects that were not referenced by the roots, but by the copied objects.
1088 * we built a stack of objects pointed to by gray_start: they are
1089 * additional roots and we may add more items as we go.
1090 * We loop until gray_start == gray_objects which means no more objects have
1091 * been added. Note this is iterative: no recursion is involved.
1092 * We need to walk the LO list as well in search of marked big objects
1093 * (use a flag since this is needed only on major collections). We need to loop
1094 * here as well, so keep a counter of marked LO (increasing it in copy_object).
1095 * To achieve better cache locality and cache usage, we drain the gray stack
1096 * frequently, after each object is copied, and just finish the work here.
1097 */
1098 sgen_drain_gray_stack (ctx);
1099 TV_GETTIME (atv);
1100 SGEN_LOG (2, "%s generation done", generation_name (generation));
1101
1102 /*
1103 Reset bridge data, we might have lingering data from a previous collection if this is a major
1104 collection trigged by minor overflow.
1105
1106 We must reset the gathered bridges since their original block might be evacuated due to major
1107 fragmentation in the meanwhile and the bridge code should not have to deal with that.
1108 */
1109 if (sgen_client_bridge_need_processing ())
1110 sgen_client_bridge_reset_data ();
1111
1112 /*
1113 * Mark all strong toggleref objects. This must be done before we walk ephemerons or finalizers
1114 * to ensure they see the full set of live objects.
1115 */
1116 sgen_client_mark_togglerefs (start_addr, end_addr, ctx);
1117
1118 /*
1119 * Walk the ephemeron tables marking all values with reachable keys. This must be completely done
1120 * before processing finalizable objects and non-tracking weak links to avoid finalizing/clearing
1121 * objects that are in fact reachable.
1122 */
1123 done_with_ephemerons = 0;
1124 do {
1125 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1126 sgen_drain_gray_stack (ctx);
1127 ++ephemeron_rounds;
1128 } while (!done_with_ephemerons);
1129
1130 if (sgen_client_bridge_need_processing ()) {
1131 /*Make sure the gray stack is empty before we process bridge objects so we get liveness right*/
1132 sgen_drain_gray_stack (ctx);
1133 sgen_collect_bridge_objects (generation, ctx);
1134 if (generation == GENERATION_OLD)
1135 sgen_collect_bridge_objects (GENERATION_NURSERY, ctx);
1136
1137 /*
1138 Do the first bridge step here, as the collector liveness state will become useless after that.
1139
1140 An important optimization is to only proccess the possibly dead part of the object graph and skip
1141 over all live objects as we transitively know everything they point must be alive too.
1142
1143 The above invariant is completely wrong if we let the gray queue be drained and mark/copy everything.
1144
1145 This has the unfortunate side effect of making overflow collections perform the first step twice, but
1146 given we now have heuristics that perform major GC in anticipation of minor overflows this should not
1147 be a big deal.
1148 */
1149 sgen_client_bridge_processing_stw_step ();
1150 }
1151
1152 /*
1153 Make sure we drain the gray stack before processing disappearing links and finalizers.
1154 If we don't make sure it is empty we might wrongly see a live object as dead.
1155 */
1156 sgen_drain_gray_stack (ctx);
1157
1158 /*
1159 We must clear weak links that don't track resurrection before processing object ready for
1160 finalization so they can be cleared before that.
1161 */
1162 sgen_null_link_in_range (generation, ctx, FALSE);
1163 if (generation == GENERATION_OLD)
1164 sgen_null_link_in_range (GENERATION_NURSERY, ctx, FALSE);
1165
1166
1167 /* walk the finalization queue and move also the objects that need to be
1168 * finalized: use the finalized objects as new roots so the objects they depend
1169 * on are also not reclaimed. As with the roots above, only objects in the nursery
1170 * are marked/copied.
1171 */
1172 sgen_finalize_in_range (generation, ctx);
1173 if (generation == GENERATION_OLD)
1174 sgen_finalize_in_range (GENERATION_NURSERY, ctx);
1175 /* drain the new stack that might have been created */
1176 SGEN_LOG (6, "Precise scan of gray area post fin");
1177 sgen_drain_gray_stack (ctx);
1178
1179 /*
1180 * This must be done again after processing finalizable objects since CWL slots are cleared only after the key is finalized.
1181 */
1182 done_with_ephemerons = 0;
1183 do {
1184 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1185 sgen_drain_gray_stack (ctx);
1186 ++ephemeron_rounds;
1187 } while (!done_with_ephemerons);
1188
1189 sgen_client_clear_unreachable_ephemerons (ctx);
1190
1191 /*
1192 * We clear togglerefs only after all possible chances of revival are done.
1193 * This is semantically more inline with what users expect and it allows for
1194 * user finalizers to correctly interact with TR objects.
1195 */
1196 sgen_client_clear_togglerefs (start_addr, end_addr, ctx);
1197
1198 TV_GETTIME (btv);
1199 SGEN_LOG (2, "Finalize queue handling scan for %s generation: %lld usecs %d ephemeron rounds", generation_name (generation), (long long)TV_ELAPSED (atv, btv), ephemeron_rounds);
1200
1201 /*
1202 * handle disappearing links
1203 * Note we do this after checking the finalization queue because if an object
1204 * survives (at least long enough to be finalized) we don't clear the link.
1205 * This also deals with a possible issue with the monitor reclamation: with the Boehm
1206 * GC a finalized object my lose the monitor because it is cleared before the finalizer is
1207 * called.
1208 */
1209 g_assert (sgen_gray_object_queue_is_empty (queue));
1210 for (;;) {
1211 sgen_null_link_in_range (generation, ctx, TRUE);
1212 if (generation == GENERATION_OLD)
1213 sgen_null_link_in_range (GENERATION_NURSERY, ctx, TRUE);
1214 if (sgen_gray_object_queue_is_empty (queue))
1215 break;
1216 sgen_drain_gray_stack (ctx);
1217 }
1218
1219 g_assert (sgen_gray_object_queue_is_empty (queue));
1220
1221 binary_protocol_finish_gray_stack_end (sgen_timestamp (), generation);
1222 }
1223
1224 void
sgen_check_section_scan_starts(GCMemSection * section)1225 sgen_check_section_scan_starts (GCMemSection *section)
1226 {
1227 size_t i;
1228 for (i = 0; i < section->num_scan_start; ++i) {
1229 if (section->scan_starts [i]) {
1230 mword size = safe_object_get_size ((GCObject*) section->scan_starts [i]);
1231 SGEN_ASSERT (0, size >= SGEN_CLIENT_MINIMUM_OBJECT_SIZE && size <= MAX_SMALL_OBJ_SIZE, "Weird object size at scan starts.");
1232 }
1233 }
1234 }
1235
1236 static void
check_scan_starts(void)1237 check_scan_starts (void)
1238 {
1239 if (!do_scan_starts_check)
1240 return;
1241 sgen_check_section_scan_starts (nursery_section);
1242 major_collector.check_scan_starts ();
1243 }
1244
1245 static void
scan_from_registered_roots(char * addr_start,char * addr_end,int root_type,ScanCopyContext ctx)1246 scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx)
1247 {
1248 void **start_root;
1249 RootRecord *root;
1250 SGEN_HASH_TABLE_FOREACH (&roots_hash [root_type], void **, start_root, RootRecord *, root) {
1251 SGEN_LOG (6, "Precise root scan %p-%p (desc: %p)", start_root, root->end_root, (void*)root->root_desc);
1252 precisely_scan_objects_from (start_root, (void**)root->end_root, addr_start, addr_end, root->root_desc, ctx);
1253 } SGEN_HASH_TABLE_FOREACH_END;
1254 }
1255
1256 static void
init_stats(void)1257 init_stats (void)
1258 {
1259 static gboolean inited = FALSE;
1260
1261 if (inited)
1262 return;
1263
1264 mono_counters_register ("Collection max time", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME | MONO_COUNTER_MONOTONIC, &time_max);
1265
1266 mono_counters_register ("Minor fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pre_collection_fragment_clear);
1267 mono_counters_register ("Minor pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pinning);
1268 mono_counters_register ("Minor scan remembered set", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_remsets);
1269 mono_counters_register ("Minor scan major blocks", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_major_blocks);
1270 mono_counters_register ("Minor scan los", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_los);
1271 mono_counters_register ("Minor scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_pinned);
1272 mono_counters_register ("Minor scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_roots);
1273 mono_counters_register ("Minor fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_fragment_creation);
1274
1275 mono_counters_register ("Major fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pre_collection_fragment_clear);
1276 mono_counters_register ("Major pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pinning);
1277 mono_counters_register ("Major scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_pinned);
1278 mono_counters_register ("Major scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_roots);
1279 mono_counters_register ("Major scan mod union blocks", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_mod_union_blocks);
1280 mono_counters_register ("Major scan mod union los", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_mod_union_los);
1281 mono_counters_register ("Major finish gray stack", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_finish_gray_stack);
1282 mono_counters_register ("Major free big objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_free_bigobjs);
1283 mono_counters_register ("Major LOS sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_los_sweep);
1284 mono_counters_register ("Major sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_sweep);
1285 mono_counters_register ("Major fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_fragment_creation);
1286
1287 mono_counters_register ("Number of pinned objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_pinned_objects);
1288
1289 #ifdef HEAVY_STATISTICS
1290 mono_counters_register ("WBarrier remember pointer", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_add_to_global_remset);
1291 mono_counters_register ("WBarrier arrayref copy", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_arrayref_copy);
1292 mono_counters_register ("WBarrier generic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store);
1293 mono_counters_register ("WBarrier generic atomic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store_atomic);
1294 mono_counters_register ("WBarrier set root", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_set_root);
1295
1296 mono_counters_register ("# objects allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_alloced_degraded);
1297 mono_counters_register ("bytes allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_bytes_alloced_degraded);
1298
1299 mono_counters_register ("# copy_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_nursery);
1300 mono_counters_register ("# objects copied (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_nursery);
1301 mono_counters_register ("# copy_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_major);
1302 mono_counters_register ("# objects copied (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_major);
1303
1304 mono_counters_register ("# scan_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_nursery);
1305 mono_counters_register ("# scan_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_major);
1306
1307 mono_counters_register ("Slots allocated in vain", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_slots_allocated_in_vain);
1308
1309 mono_counters_register ("# nursery copy_object() failed from space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_from_space);
1310 mono_counters_register ("# nursery copy_object() failed forwarded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_forwarded);
1311 mono_counters_register ("# nursery copy_object() failed pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_pinned);
1312 mono_counters_register ("# nursery copy_object() failed to space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_to_space);
1313
1314 sgen_nursery_allocator_init_heavy_stats ();
1315 #endif
1316
1317 inited = TRUE;
1318 }
1319
1320
1321 static void
reset_pinned_from_failed_allocation(void)1322 reset_pinned_from_failed_allocation (void)
1323 {
1324 bytes_pinned_from_failed_allocation = 0;
1325 }
1326
1327 void
sgen_set_pinned_from_failed_allocation(mword objsize)1328 sgen_set_pinned_from_failed_allocation (mword objsize)
1329 {
1330 bytes_pinned_from_failed_allocation += objsize;
1331 }
1332
1333 gboolean
sgen_collection_is_concurrent(void)1334 sgen_collection_is_concurrent (void)
1335 {
1336 switch (current_collection_generation) {
1337 case GENERATION_NURSERY:
1338 return FALSE;
1339 case GENERATION_OLD:
1340 return concurrent_collection_in_progress;
1341 default:
1342 g_error ("Invalid current generation %d", current_collection_generation);
1343 }
1344 return FALSE;
1345 }
1346
1347 gboolean
sgen_concurrent_collection_in_progress(void)1348 sgen_concurrent_collection_in_progress (void)
1349 {
1350 return concurrent_collection_in_progress;
1351 }
1352
1353 typedef struct {
1354 SgenThreadPoolJob job;
1355 SgenObjectOperations *ops;
1356 SgenGrayQueue *gc_thread_gray_queue;
1357 } ScanJob;
1358
1359 typedef struct {
1360 ScanJob scan_job;
1361 int job_index, job_split_count;
1362 int data;
1363 } ParallelScanJob;
1364
1365 static ScanCopyContext
scan_copy_context_for_scan_job(void * worker_data_untyped,ScanJob * job)1366 scan_copy_context_for_scan_job (void *worker_data_untyped, ScanJob *job)
1367 {
1368 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1369
1370 if (!job->ops) {
1371 /*
1372 * For jobs enqueued on workers we set the ops at job runtime in order
1373 * to be able to profit from on the fly optimized object ops or other
1374 * object ops changes, like forced concurrent finish.
1375 */
1376 SGEN_ASSERT (0, sgen_workers_is_worker_thread (mono_native_thread_id_get ()), "We need a context for the scan job");
1377 job->ops = sgen_workers_get_idle_func_object_ops (worker_data);
1378 }
1379
1380 return CONTEXT_FROM_OBJECT_OPERATIONS (job->ops, sgen_workers_get_job_gray_queue (worker_data, job->gc_thread_gray_queue));
1381 }
1382
1383 typedef struct {
1384 ScanJob scan_job;
1385 char *heap_start;
1386 char *heap_end;
1387 int root_type;
1388 } ScanFromRegisteredRootsJob;
1389
1390 static void
job_scan_from_registered_roots(void * worker_data_untyped,SgenThreadPoolJob * job)1391 job_scan_from_registered_roots (void *worker_data_untyped, SgenThreadPoolJob *job)
1392 {
1393 ScanFromRegisteredRootsJob *job_data = (ScanFromRegisteredRootsJob*)job;
1394 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1395
1396 scan_from_registered_roots (job_data->heap_start, job_data->heap_end, job_data->root_type, ctx);
1397 }
1398
1399 typedef struct {
1400 ScanJob scan_job;
1401 char *heap_start;
1402 char *heap_end;
1403 } ScanThreadDataJob;
1404
1405 static void
job_scan_thread_data(void * worker_data_untyped,SgenThreadPoolJob * job)1406 job_scan_thread_data (void *worker_data_untyped, SgenThreadPoolJob *job)
1407 {
1408 ScanThreadDataJob *job_data = (ScanThreadDataJob*)job;
1409 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1410
1411 sgen_client_scan_thread_data (job_data->heap_start, job_data->heap_end, TRUE, ctx);
1412 }
1413
1414 typedef struct {
1415 ScanJob scan_job;
1416 SgenPointerQueue *queue;
1417 } ScanFinalizerEntriesJob;
1418
1419 static void
job_scan_finalizer_entries(void * worker_data_untyped,SgenThreadPoolJob * job)1420 job_scan_finalizer_entries (void *worker_data_untyped, SgenThreadPoolJob *job)
1421 {
1422 ScanFinalizerEntriesJob *job_data = (ScanFinalizerEntriesJob*)job;
1423 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1424
1425 scan_finalizer_entries (job_data->queue, ctx);
1426 }
1427
1428 static void
job_scan_wbroots(void * worker_data_untyped,SgenThreadPoolJob * job)1429 job_scan_wbroots (void *worker_data_untyped, SgenThreadPoolJob *job)
1430 {
1431 ScanJob *job_data = (ScanJob*)job;
1432 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, job_data);
1433
1434 sgen_wbroots_scan_card_table (ctx);
1435 }
1436
1437 static void
job_scan_major_card_table(void * worker_data_untyped,SgenThreadPoolJob * job)1438 job_scan_major_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1439 {
1440 SGEN_TV_DECLARE (atv);
1441 SGEN_TV_DECLARE (btv);
1442 ParallelScanJob *job_data = (ParallelScanJob*)job;
1443 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1444
1445 SGEN_TV_GETTIME (atv);
1446 major_collector.scan_card_table (CARDTABLE_SCAN_GLOBAL, ctx, job_data->job_index, job_data->job_split_count, job_data->data);
1447 SGEN_TV_GETTIME (btv);
1448 time_minor_scan_major_blocks += SGEN_TV_ELAPSED (atv, btv);
1449
1450 if (worker_data_untyped)
1451 ((WorkerData*)worker_data_untyped)->major_scan_time += SGEN_TV_ELAPSED (atv, btv);
1452 }
1453
1454 static void
job_scan_los_card_table(void * worker_data_untyped,SgenThreadPoolJob * job)1455 job_scan_los_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1456 {
1457 SGEN_TV_DECLARE (atv);
1458 SGEN_TV_DECLARE (btv);
1459 ParallelScanJob *job_data = (ParallelScanJob*)job;
1460 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1461
1462 SGEN_TV_GETTIME (atv);
1463 sgen_los_scan_card_table (CARDTABLE_SCAN_GLOBAL, ctx, job_data->job_index, job_data->job_split_count);
1464 SGEN_TV_GETTIME (btv);
1465 time_minor_scan_los += SGEN_TV_ELAPSED (atv, btv);
1466
1467 if (worker_data_untyped)
1468 ((WorkerData*)worker_data_untyped)->los_scan_time += SGEN_TV_ELAPSED (atv, btv);
1469 }
1470
1471 static void
job_scan_major_mod_union_card_table(void * worker_data_untyped,SgenThreadPoolJob * job)1472 job_scan_major_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1473 {
1474 SGEN_TV_DECLARE (atv);
1475 SGEN_TV_DECLARE (btv);
1476 ParallelScanJob *job_data = (ParallelScanJob*)job;
1477 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1478
1479 g_assert (concurrent_collection_in_progress);
1480 SGEN_TV_GETTIME (atv);
1481 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx, job_data->job_index, job_data->job_split_count, job_data->data);
1482 SGEN_TV_GETTIME (btv);
1483 time_major_scan_mod_union_blocks += SGEN_TV_ELAPSED (atv, btv);
1484
1485 if (worker_data_untyped)
1486 ((WorkerData*)worker_data_untyped)->major_scan_time += SGEN_TV_ELAPSED (atv, btv);
1487 }
1488
1489 static void
job_scan_los_mod_union_card_table(void * worker_data_untyped,SgenThreadPoolJob * job)1490 job_scan_los_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1491 {
1492 SGEN_TV_DECLARE (atv);
1493 SGEN_TV_DECLARE (btv);
1494 ParallelScanJob *job_data = (ParallelScanJob*)job;
1495 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1496
1497 g_assert (concurrent_collection_in_progress);
1498 SGEN_TV_GETTIME (atv);
1499 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx, job_data->job_index, job_data->job_split_count);
1500 SGEN_TV_GETTIME (btv);
1501 time_major_scan_mod_union_los += SGEN_TV_ELAPSED (atv, btv);
1502
1503 if (worker_data_untyped)
1504 ((WorkerData*)worker_data_untyped)->los_scan_time += SGEN_TV_ELAPSED (atv, btv);
1505 }
1506
1507 static void
job_major_mod_union_preclean(void * worker_data_untyped,SgenThreadPoolJob * job)1508 job_major_mod_union_preclean (void *worker_data_untyped, SgenThreadPoolJob *job)
1509 {
1510 SGEN_TV_DECLARE (atv);
1511 SGEN_TV_DECLARE (btv);
1512 ParallelScanJob *job_data = (ParallelScanJob*)job;
1513 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1514
1515 g_assert (concurrent_collection_in_progress);
1516 SGEN_TV_GETTIME (atv);
1517 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx, job_data->job_index, job_data->job_split_count, job_data->data);
1518 SGEN_TV_GETTIME (btv);
1519
1520 g_assert (worker_data_untyped);
1521 ((WorkerData*)worker_data_untyped)->major_scan_time += SGEN_TV_ELAPSED (atv, btv);
1522 }
1523
1524 static void
job_los_mod_union_preclean(void * worker_data_untyped,SgenThreadPoolJob * job)1525 job_los_mod_union_preclean (void *worker_data_untyped, SgenThreadPoolJob *job)
1526 {
1527 SGEN_TV_DECLARE (atv);
1528 SGEN_TV_DECLARE (btv);
1529 ParallelScanJob *job_data = (ParallelScanJob*)job;
1530 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1531
1532 g_assert (concurrent_collection_in_progress);
1533 SGEN_TV_GETTIME (atv);
1534 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx, job_data->job_index, job_data->job_split_count);
1535 SGEN_TV_GETTIME (btv);
1536
1537 g_assert (worker_data_untyped);
1538 ((WorkerData*)worker_data_untyped)->los_scan_time += SGEN_TV_ELAPSED (atv, btv);
1539 }
1540
1541 static void
job_scan_last_pinned(void * worker_data_untyped,SgenThreadPoolJob * job)1542 job_scan_last_pinned (void *worker_data_untyped, SgenThreadPoolJob *job)
1543 {
1544 ScanJob *job_data = (ScanJob*)job;
1545 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, job_data);
1546
1547 g_assert (concurrent_collection_in_progress);
1548
1549 sgen_scan_pin_queue_objects (ctx);
1550 }
1551
1552 static void
workers_finish_callback(void)1553 workers_finish_callback (void)
1554 {
1555 ParallelScanJob *psj;
1556 ScanJob *sj;
1557 size_t num_major_sections = major_collector.get_num_major_sections ();
1558 int split_count = sgen_workers_get_job_split_count (GENERATION_OLD);
1559 int i;
1560 /* Mod union preclean jobs */
1561 for (i = 0; i < split_count; i++) {
1562 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("preclean major mod union cardtable", job_major_mod_union_preclean, sizeof (ParallelScanJob));
1563 psj->scan_job.gc_thread_gray_queue = NULL;
1564 psj->job_index = i;
1565 psj->job_split_count = split_count;
1566 psj->data = num_major_sections / split_count;
1567 sgen_workers_enqueue_job (GENERATION_OLD, &psj->scan_job.job, TRUE);
1568 }
1569
1570 for (i = 0; i < split_count; i++) {
1571 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("preclean los mod union cardtable", job_los_mod_union_preclean, sizeof (ParallelScanJob));
1572 psj->scan_job.gc_thread_gray_queue = NULL;
1573 psj->job_index = i;
1574 psj->job_split_count = split_count;
1575 sgen_workers_enqueue_job (GENERATION_OLD, &psj->scan_job.job, TRUE);
1576 }
1577
1578 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan last pinned", job_scan_last_pinned, sizeof (ScanJob));
1579 sj->gc_thread_gray_queue = NULL;
1580 sgen_workers_enqueue_job (GENERATION_OLD, &sj->job, TRUE);
1581 }
1582
1583 static void
init_gray_queue(SgenGrayQueue * gc_thread_gray_queue)1584 init_gray_queue (SgenGrayQueue *gc_thread_gray_queue)
1585 {
1586 sgen_gray_object_queue_init (gc_thread_gray_queue, NULL, TRUE);
1587 }
1588
1589 static void
enqueue_scan_remembered_set_jobs(SgenGrayQueue * gc_thread_gray_queue,SgenObjectOperations * ops,gboolean enqueue)1590 enqueue_scan_remembered_set_jobs (SgenGrayQueue *gc_thread_gray_queue, SgenObjectOperations *ops, gboolean enqueue)
1591 {
1592 int i, split_count = sgen_workers_get_job_split_count (GENERATION_NURSERY);
1593 size_t num_major_sections = major_collector.get_num_major_sections ();
1594 ScanJob *sj;
1595
1596 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan wbroots", job_scan_wbroots, sizeof (ScanJob));
1597 sj->ops = ops;
1598 sj->gc_thread_gray_queue = gc_thread_gray_queue;
1599 sgen_workers_enqueue_job (GENERATION_NURSERY, &sj->job, enqueue);
1600
1601 for (i = 0; i < split_count; i++) {
1602 ParallelScanJob *psj;
1603
1604 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("scan major remsets", job_scan_major_card_table, sizeof (ParallelScanJob));
1605 psj->scan_job.ops = ops;
1606 psj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1607 psj->job_index = i;
1608 psj->job_split_count = split_count;
1609 psj->data = num_major_sections / split_count;
1610 sgen_workers_enqueue_job (GENERATION_NURSERY, &psj->scan_job.job, enqueue);
1611
1612 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("scan LOS remsets", job_scan_los_card_table, sizeof (ParallelScanJob));
1613 psj->scan_job.ops = ops;
1614 psj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1615 psj->job_index = i;
1616 psj->job_split_count = split_count;
1617 sgen_workers_enqueue_job (GENERATION_NURSERY, &psj->scan_job.job, enqueue);
1618 }
1619 }
1620
1621 static void
enqueue_scan_from_roots_jobs(SgenGrayQueue * gc_thread_gray_queue,char * heap_start,char * heap_end,SgenObjectOperations * ops,gboolean enqueue)1622 enqueue_scan_from_roots_jobs (SgenGrayQueue *gc_thread_gray_queue, char *heap_start, char *heap_end, SgenObjectOperations *ops, gboolean enqueue)
1623 {
1624 ScanFromRegisteredRootsJob *scrrj;
1625 ScanThreadDataJob *stdj;
1626 ScanFinalizerEntriesJob *sfej;
1627
1628 /* registered roots, this includes static fields */
1629
1630 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots normal", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1631 scrrj->scan_job.ops = ops;
1632 scrrj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1633 scrrj->heap_start = heap_start;
1634 scrrj->heap_end = heap_end;
1635 scrrj->root_type = ROOT_TYPE_NORMAL;
1636 sgen_workers_enqueue_job (current_collection_generation, &scrrj->scan_job.job, enqueue);
1637
1638 if (current_collection_generation == GENERATION_OLD) {
1639 /* During minors we scan the cardtable for these roots instead */
1640 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots wbarrier", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1641 scrrj->scan_job.ops = ops;
1642 scrrj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1643 scrrj->heap_start = heap_start;
1644 scrrj->heap_end = heap_end;
1645 scrrj->root_type = ROOT_TYPE_WBARRIER;
1646 sgen_workers_enqueue_job (current_collection_generation, &scrrj->scan_job.job, enqueue);
1647 }
1648
1649 /* Threads */
1650
1651 stdj = (ScanThreadDataJob*)sgen_thread_pool_job_alloc ("scan thread data", job_scan_thread_data, sizeof (ScanThreadDataJob));
1652 stdj->scan_job.ops = ops;
1653 stdj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1654 stdj->heap_start = heap_start;
1655 stdj->heap_end = heap_end;
1656 sgen_workers_enqueue_job (current_collection_generation, &stdj->scan_job.job, enqueue);
1657
1658 /* Scan the list of objects ready for finalization. */
1659
1660 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1661 sfej->scan_job.ops = ops;
1662 sfej->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1663 sfej->queue = &fin_ready_queue;
1664 sgen_workers_enqueue_job (current_collection_generation, &sfej->scan_job.job, enqueue);
1665
1666 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan critical finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1667 sfej->scan_job.ops = ops;
1668 sfej->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1669 sfej->queue = &critical_fin_queue;
1670 sgen_workers_enqueue_job (current_collection_generation, &sfej->scan_job.job, enqueue);
1671 }
1672
1673 /*
1674 * Perform a nursery collection.
1675 *
1676 * Return whether any objects were late-pinned due to being out of memory.
1677 */
1678 static gboolean
collect_nursery(const char * reason,gboolean is_overflow,SgenGrayQueue * unpin_queue)1679 collect_nursery (const char *reason, gboolean is_overflow, SgenGrayQueue *unpin_queue)
1680 {
1681 gboolean needs_major, is_parallel = FALSE;
1682 mword fragment_total;
1683 SgenGrayQueue gc_thread_gray_queue;
1684 SgenObjectOperations *object_ops_nopar, *object_ops_par = NULL;
1685 ScanCopyContext ctx;
1686 TV_DECLARE (atv);
1687 TV_DECLARE (btv);
1688 SGEN_TV_DECLARE (last_minor_collection_start_tv);
1689 SGEN_TV_DECLARE (last_minor_collection_end_tv);
1690 guint64 major_scan_start = time_minor_scan_major_blocks;
1691 guint64 los_scan_start = time_minor_scan_los;
1692 guint64 finish_gray_start = time_minor_finish_gray_stack;
1693
1694 if (disable_minor_collections)
1695 return TRUE;
1696
1697 TV_GETTIME (last_minor_collection_start_tv);
1698 atv = last_minor_collection_start_tv;
1699
1700 binary_protocol_collection_begin (mono_atomic_load_i32 (&gc_stats.minor_gc_count), GENERATION_NURSERY);
1701
1702 object_ops_nopar = sgen_concurrent_collection_in_progress ()
1703 ? &sgen_minor_collector.serial_ops_with_concurrent_major
1704 : &sgen_minor_collector.serial_ops;
1705 if (sgen_minor_collector.is_parallel && sgen_nursery_size >= SGEN_PARALLEL_MINOR_MIN_NURSERY_SIZE) {
1706 object_ops_par = sgen_concurrent_collection_in_progress ()
1707 ? &sgen_minor_collector.parallel_ops_with_concurrent_major
1708 : &sgen_minor_collector.parallel_ops;
1709 is_parallel = TRUE;
1710 }
1711
1712 if (do_verify_nursery || do_dump_nursery_content)
1713 sgen_debug_verify_nursery (do_dump_nursery_content);
1714
1715 current_collection_generation = GENERATION_NURSERY;
1716
1717 SGEN_ASSERT (0, !sgen_collection_is_concurrent (), "Why is the nursery collection concurrent?");
1718
1719 reset_pinned_from_failed_allocation ();
1720
1721 check_scan_starts ();
1722
1723 sgen_nursery_alloc_prepare_for_minor ();
1724
1725 degraded_mode = 0;
1726 objects_pinned = 0;
1727
1728 SGEN_LOG (1, "Start nursery collection %" G_GINT32_FORMAT " %p-%p, size: %d", mono_atomic_load_i32 (&gc_stats.minor_gc_count), nursery_section->data, nursery_section->end_data, (int)(nursery_section->end_data - nursery_section->data));
1729
1730 /* world must be stopped already */
1731 TV_GETTIME (btv);
1732 time_minor_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1733
1734 sgen_client_pre_collection_checks ();
1735
1736 major_collector.start_nursery_collection ();
1737
1738 sgen_memgov_minor_collection_start ();
1739
1740 init_gray_queue (&gc_thread_gray_queue);
1741 ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops_nopar, &gc_thread_gray_queue);
1742
1743 mono_atomic_inc_i32 (&gc_stats.minor_gc_count);
1744
1745 sgen_process_fin_stage_entries ();
1746
1747 /* pin from pinned handles */
1748 sgen_init_pinning ();
1749 if (concurrent_collection_in_progress)
1750 sgen_init_pinning_for_conc ();
1751 sgen_client_binary_protocol_mark_start (GENERATION_NURSERY);
1752 pin_from_roots (nursery_section->data, nursery_section->end_data, ctx);
1753 /* pin cemented objects */
1754 sgen_pin_cemented_objects ();
1755 /* identify pinned objects */
1756 sgen_optimize_pin_queue ();
1757 sgen_pinning_setup_section (nursery_section);
1758
1759 pin_objects_in_nursery (FALSE, ctx);
1760 sgen_pinning_trim_queue_to_section (nursery_section);
1761 if (concurrent_collection_in_progress)
1762 sgen_finish_pinning_for_conc ();
1763
1764 if (remset_consistency_checks)
1765 sgen_check_remset_consistency ();
1766
1767 if (whole_heap_check_before_collection) {
1768 sgen_clear_nursery_fragments ();
1769 sgen_check_whole_heap (FALSE);
1770 }
1771
1772 TV_GETTIME (atv);
1773 time_minor_pinning += TV_ELAPSED (btv, atv);
1774 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), (long long)TV_ELAPSED (btv, atv));
1775 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1776
1777 remset.start_scan_remsets ();
1778
1779 enqueue_scan_remembered_set_jobs (&gc_thread_gray_queue, is_parallel ? NULL : object_ops_nopar, is_parallel);
1780
1781 /* we don't have complete write barrier yet, so we scan all the old generation sections */
1782 TV_GETTIME (btv);
1783 time_minor_scan_remsets += TV_ELAPSED (atv, btv);
1784 SGEN_LOG (2, "Old generation scan: %lld usecs", (long long)TV_ELAPSED (atv, btv));
1785
1786 sgen_pin_stats_report ();
1787
1788 /* FIXME: Why do we do this at this specific, seemingly random, point? */
1789 sgen_client_collecting_minor (&fin_ready_queue, &critical_fin_queue);
1790
1791 TV_GETTIME (atv);
1792 time_minor_scan_pinned += TV_ELAPSED (btv, atv);
1793
1794 enqueue_scan_from_roots_jobs (&gc_thread_gray_queue, nursery_section->data, nursery_section->end_data, is_parallel ? NULL : object_ops_nopar, is_parallel);
1795
1796 if (is_parallel) {
1797 gray_queue_redirect (&gc_thread_gray_queue);
1798 sgen_workers_start_all_workers (GENERATION_NURSERY, object_ops_nopar, object_ops_par, NULL);
1799 sgen_workers_join (GENERATION_NURSERY);
1800 }
1801
1802 TV_GETTIME (btv);
1803 time_minor_scan_roots += TV_ELAPSED (atv, btv);
1804
1805 finish_gray_stack (GENERATION_NURSERY, ctx);
1806
1807 TV_GETTIME (atv);
1808 time_minor_finish_gray_stack += TV_ELAPSED (btv, atv);
1809 sgen_client_binary_protocol_mark_end (GENERATION_NURSERY);
1810
1811 if (objects_pinned) {
1812 sgen_optimize_pin_queue ();
1813 sgen_pinning_setup_section (nursery_section);
1814 }
1815
1816 /*
1817 * This is the latest point at which we can do this check, because
1818 * sgen_build_nursery_fragments() unpins nursery objects again.
1819 */
1820 if (remset_consistency_checks)
1821 sgen_check_remset_consistency ();
1822
1823
1824 if (sgen_max_pause_time) {
1825 int duration;
1826
1827 TV_GETTIME (btv);
1828 duration = (int)(TV_ELAPSED (last_minor_collection_start_tv, btv) / 10000);
1829 if (duration > (sgen_max_pause_time * sgen_max_pause_margin))
1830 sgen_resize_nursery (TRUE);
1831 else
1832 sgen_resize_nursery (FALSE);
1833 } else {
1834 sgen_resize_nursery (FALSE);
1835 }
1836
1837 /* walk the pin_queue, build up the fragment list of free memory, unmark
1838 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1839 * next allocations.
1840 */
1841 sgen_client_binary_protocol_reclaim_start (GENERATION_NURSERY);
1842 fragment_total = sgen_build_nursery_fragments (nursery_section, unpin_queue);
1843 if (!fragment_total)
1844 degraded_mode = 1;
1845
1846 /* Clear TLABs for all threads */
1847 sgen_clear_tlabs ();
1848
1849 sgen_client_binary_protocol_reclaim_end (GENERATION_NURSERY);
1850 TV_GETTIME (btv);
1851 time_minor_fragment_creation += TV_ELAPSED (atv, btv);
1852 SGEN_LOG (2, "Fragment creation: %lld usecs, %lu bytes available", (long long)TV_ELAPSED (atv, btv), (unsigned long)fragment_total);
1853
1854 if (remset_consistency_checks)
1855 sgen_check_major_refs ();
1856
1857 major_collector.finish_nursery_collection ();
1858
1859 TV_GETTIME (last_minor_collection_end_tv);
1860 UnlockedAdd64 (&gc_stats.minor_gc_time, TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv));
1861
1862 sgen_debug_dump_heap ("minor", mono_atomic_load_i32 (&gc_stats.minor_gc_count) - 1, NULL);
1863
1864 /* prepare the pin queue for the next collection */
1865 sgen_finish_pinning ();
1866 if (sgen_have_pending_finalizers ()) {
1867 SGEN_LOG (4, "Finalizer-thread wakeup");
1868 sgen_client_finalize_notify ();
1869 }
1870 sgen_pin_stats_reset ();
1871 /* clear cemented hash */
1872 sgen_cement_clear_below_threshold ();
1873
1874 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
1875
1876 check_scan_starts ();
1877
1878 binary_protocol_flush_buffers (FALSE);
1879
1880 sgen_memgov_minor_collection_end (reason, is_overflow);
1881
1882 /*objects are late pinned because of lack of memory, so a major is a good call*/
1883 needs_major = objects_pinned > 0;
1884 current_collection_generation = -1;
1885 objects_pinned = 0;
1886
1887 if (is_parallel)
1888 binary_protocol_collection_end_stats (0, 0, time_minor_finish_gray_stack - finish_gray_start);
1889 else
1890 binary_protocol_collection_end_stats (
1891 time_minor_scan_major_blocks - major_scan_start,
1892 time_minor_scan_los - los_scan_start,
1893 time_minor_finish_gray_stack - finish_gray_start);
1894
1895 binary_protocol_collection_end (mono_atomic_load_i32 (&gc_stats.minor_gc_count) - 1, GENERATION_NURSERY, 0, 0);
1896
1897 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1898 sgen_check_nursery_objects_pinned (unpin_queue != NULL);
1899
1900 return needs_major;
1901 }
1902
1903 typedef enum {
1904 COPY_OR_MARK_FROM_ROOTS_SERIAL,
1905 COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT,
1906 COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT
1907 } CopyOrMarkFromRootsMode;
1908
1909 static void
major_copy_or_mark_from_roots(SgenGrayQueue * gc_thread_gray_queue,size_t * old_next_pin_slot,CopyOrMarkFromRootsMode mode,SgenObjectOperations * object_ops_nopar,SgenObjectOperations * object_ops_par)1910 major_copy_or_mark_from_roots (SgenGrayQueue *gc_thread_gray_queue, size_t *old_next_pin_slot, CopyOrMarkFromRootsMode mode, SgenObjectOperations *object_ops_nopar, SgenObjectOperations *object_ops_par)
1911 {
1912 LOSObject *bigobj;
1913 TV_DECLARE (atv);
1914 TV_DECLARE (btv);
1915 /* FIXME: only use these values for the precise scan
1916 * note that to_space pointers should be excluded anyway...
1917 */
1918 char *heap_start = NULL;
1919 char *heap_end = (char*)-1;
1920 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops_nopar, gc_thread_gray_queue);
1921 gboolean concurrent = mode != COPY_OR_MARK_FROM_ROOTS_SERIAL;
1922
1923 SGEN_ASSERT (0, !!concurrent == !!concurrent_collection_in_progress, "We've been called with the wrong mode.");
1924
1925 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1926 /*This cleans up unused fragments */
1927 sgen_nursery_allocator_prepare_for_pinning ();
1928
1929 if (do_concurrent_checks)
1930 sgen_debug_check_nursery_is_clean ();
1931 } else {
1932 /* The concurrent collector doesn't touch the nursery. */
1933 sgen_nursery_alloc_prepare_for_major ();
1934 }
1935
1936 TV_GETTIME (atv);
1937
1938 /* Pinning depends on this */
1939 sgen_clear_nursery_fragments ();
1940
1941 if (whole_heap_check_before_collection)
1942 sgen_check_whole_heap (TRUE);
1943
1944 TV_GETTIME (btv);
1945 time_major_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1946
1947 objects_pinned = 0;
1948
1949 sgen_client_pre_collection_checks ();
1950
1951 if (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1952 /* Remsets are not useful for a major collection */
1953 remset.clear_cards ();
1954 }
1955
1956 sgen_process_fin_stage_entries ();
1957
1958 TV_GETTIME (atv);
1959 sgen_init_pinning ();
1960 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT)
1961 sgen_init_pinning_for_conc ();
1962 SGEN_LOG (6, "Collecting pinned addresses");
1963 pin_from_roots ((void*)lowest_heap_address, (void*)highest_heap_address, ctx);
1964 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1965 /* Pin cemented objects that were forced */
1966 sgen_pin_cemented_objects ();
1967 }
1968 sgen_optimize_pin_queue ();
1969 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1970 /*
1971 * Cemented objects that are in the pinned list will be marked. When
1972 * marking concurrently we won't mark mod-union cards for these objects.
1973 * Instead they will remain cemented until the next major collection,
1974 * when we will recheck if they are still pinned in the roots.
1975 */
1976 sgen_cement_force_pinned ();
1977 }
1978
1979 sgen_client_collecting_major_1 ();
1980
1981 /*
1982 * pin_queue now contains all candidate pointers, sorted and
1983 * uniqued. We must do two passes now to figure out which
1984 * objects are pinned.
1985 *
1986 * The first is to find within the pin_queue the area for each
1987 * section. This requires that the pin_queue be sorted. We
1988 * also process the LOS objects and pinned chunks here.
1989 *
1990 * The second, destructive, pass is to reduce the section
1991 * areas to pointers to the actually pinned objects.
1992 */
1993 SGEN_LOG (6, "Pinning from sections");
1994 /* first pass for the sections */
1995 sgen_find_section_pin_queue_start_end (nursery_section);
1996 /* identify possible pointers to the insize of large objects */
1997 SGEN_LOG (6, "Pinning from large objects");
1998 for (bigobj = los_object_list; bigobj; bigobj = bigobj->next) {
1999 size_t dummy;
2000 if (sgen_find_optimized_pin_queue_area ((char*)bigobj->data, (char*)bigobj->data + sgen_los_object_size (bigobj), &dummy, &dummy)) {
2001 binary_protocol_pin (bigobj->data, (gpointer)LOAD_VTABLE (bigobj->data), safe_object_get_size (bigobj->data));
2002
2003 if (sgen_los_object_is_pinned (bigobj->data)) {
2004 SGEN_ASSERT (0, mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, "LOS objects can only be pinned here after concurrent marking.");
2005 continue;
2006 }
2007 sgen_los_pin_object (bigobj->data);
2008 if (SGEN_OBJECT_HAS_REFERENCES (bigobj->data))
2009 GRAY_OBJECT_ENQUEUE_SERIAL (gc_thread_gray_queue, bigobj->data, sgen_obj_get_descriptor ((GCObject*)bigobj->data));
2010 sgen_pin_stats_register_object (bigobj->data, GENERATION_OLD);
2011 SGEN_LOG (6, "Marked large object %p (%s) size: %lu from roots", bigobj->data,
2012 sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (bigobj->data)),
2013 (unsigned long)sgen_los_object_size (bigobj));
2014
2015 sgen_client_pinned_los_object (bigobj->data);
2016 }
2017 }
2018
2019 pin_objects_in_nursery (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT, ctx);
2020 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
2021 sgen_check_nursery_objects_pinned (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
2022
2023 major_collector.pin_objects (gc_thread_gray_queue);
2024 if (old_next_pin_slot)
2025 *old_next_pin_slot = sgen_get_pinned_count ();
2026
2027 TV_GETTIME (btv);
2028 time_major_pinning += TV_ELAPSED (atv, btv);
2029 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), (long long)TV_ELAPSED (atv, btv));
2030 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
2031
2032 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT)
2033 sgen_finish_pinning_for_conc ();
2034
2035 major_collector.init_to_space ();
2036
2037 SGEN_ASSERT (0, sgen_workers_all_done (), "Why are the workers not done when we start or finish a major collection?");
2038 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
2039 if (object_ops_par != NULL)
2040 sgen_workers_set_num_active_workers (GENERATION_OLD, 0);
2041 if (object_ops_par == NULL && sgen_workers_have_idle_work (GENERATION_OLD)) {
2042 /*
2043 * We force the finish of the worker with the new object ops context
2044 * which can also do copying. We need to have finished pinning. On the
2045 * parallel collector, there is no need to drain the private queues
2046 * here, since we can do it as part of the finishing work, achieving
2047 * better work distribution.
2048 */
2049 sgen_workers_start_all_workers (GENERATION_OLD, object_ops_nopar, object_ops_par, NULL);
2050
2051 sgen_workers_join (GENERATION_OLD);
2052 }
2053 }
2054
2055 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
2056 main_gc_thread = mono_native_thread_self ();
2057 #endif
2058
2059 sgen_client_collecting_major_2 ();
2060
2061 TV_GETTIME (atv);
2062 time_major_scan_pinned += TV_ELAPSED (btv, atv);
2063
2064 sgen_client_collecting_major_3 (&fin_ready_queue, &critical_fin_queue);
2065
2066 enqueue_scan_from_roots_jobs (gc_thread_gray_queue, heap_start, heap_end, object_ops_nopar, FALSE);
2067
2068 TV_GETTIME (btv);
2069 time_major_scan_roots += TV_ELAPSED (atv, btv);
2070
2071 /*
2072 * We start the concurrent worker after pinning and after we scanned the roots
2073 * in order to make sure that the worker does not finish before handling all
2074 * the roots.
2075 */
2076 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
2077 sgen_workers_set_num_active_workers (GENERATION_OLD, 1);
2078 gray_queue_redirect (gc_thread_gray_queue);
2079 if (precleaning_enabled) {
2080 sgen_workers_start_all_workers (GENERATION_OLD, object_ops_nopar, object_ops_par, workers_finish_callback);
2081 } else {
2082 sgen_workers_start_all_workers (GENERATION_OLD, object_ops_nopar, object_ops_par, NULL);
2083 }
2084 }
2085
2086 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
2087 int i, split_count = sgen_workers_get_job_split_count (GENERATION_OLD);
2088 size_t num_major_sections = major_collector.get_num_major_sections ();
2089 gboolean parallel = object_ops_par != NULL;
2090
2091 /* If we're not parallel we finish the collection on the gc thread */
2092 if (parallel)
2093 gray_queue_redirect (gc_thread_gray_queue);
2094
2095 /* Mod union card table */
2096 for (i = 0; i < split_count; i++) {
2097 ParallelScanJob *psj;
2098
2099 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("scan mod union cardtable", job_scan_major_mod_union_card_table, sizeof (ParallelScanJob));
2100 psj->scan_job.ops = parallel ? NULL : object_ops_nopar;
2101 psj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
2102 psj->job_index = i;
2103 psj->job_split_count = split_count;
2104 psj->data = num_major_sections / split_count;
2105 sgen_workers_enqueue_job (GENERATION_OLD, &psj->scan_job.job, parallel);
2106
2107 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("scan LOS mod union cardtable", job_scan_los_mod_union_card_table, sizeof (ParallelScanJob));
2108 psj->scan_job.ops = parallel ? NULL : object_ops_nopar;
2109 psj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
2110 psj->job_index = i;
2111 psj->job_split_count = split_count;
2112 sgen_workers_enqueue_job (GENERATION_OLD, &psj->scan_job.job, parallel);
2113 }
2114
2115 if (parallel) {
2116 /*
2117 * If we enqueue a job while workers are running we need to sgen_workers_ensure_awake
2118 * in order to make sure that we are running the idle func and draining all worker
2119 * gray queues. The operation of starting workers implies this, so we start them after
2120 * in order to avoid doing this operation twice. The workers will drain the main gray
2121 * stack that contained roots and pinned objects and also scan the mod union card
2122 * table.
2123 */
2124 sgen_workers_start_all_workers (GENERATION_OLD, object_ops_nopar, object_ops_par, NULL);
2125 sgen_workers_join (GENERATION_OLD);
2126 }
2127 }
2128
2129 sgen_pin_stats_report ();
2130
2131 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
2132 sgen_finish_pinning ();
2133
2134 sgen_pin_stats_reset ();
2135
2136 if (do_concurrent_checks)
2137 sgen_debug_check_nursery_is_clean ();
2138 }
2139 }
2140
2141 static void
major_start_collection(SgenGrayQueue * gc_thread_gray_queue,const char * reason,gboolean concurrent,size_t * old_next_pin_slot)2142 major_start_collection (SgenGrayQueue *gc_thread_gray_queue, const char *reason, gboolean concurrent, size_t *old_next_pin_slot)
2143 {
2144 SgenObjectOperations *object_ops_nopar, *object_ops_par = NULL;
2145
2146 binary_protocol_collection_begin (mono_atomic_load_i32 (&gc_stats.major_gc_count), GENERATION_OLD);
2147
2148 current_collection_generation = GENERATION_OLD;
2149
2150 sgen_workers_assert_gray_queue_is_empty (GENERATION_OLD);
2151
2152 if (!concurrent)
2153 sgen_cement_reset ();
2154
2155 if (concurrent) {
2156 g_assert (major_collector.is_concurrent);
2157 concurrent_collection_in_progress = TRUE;
2158
2159 object_ops_nopar = &major_collector.major_ops_concurrent_start;
2160 if (major_collector.is_parallel)
2161 object_ops_par = &major_collector.major_ops_conc_par_start;
2162
2163 } else {
2164 object_ops_nopar = &major_collector.major_ops_serial;
2165 }
2166
2167 reset_pinned_from_failed_allocation ();
2168
2169 sgen_memgov_major_collection_start (concurrent, reason);
2170
2171 //count_ref_nonref_objs ();
2172 //consistency_check ();
2173
2174 check_scan_starts ();
2175
2176 degraded_mode = 0;
2177 SGEN_LOG (1, "Start major collection %" G_GINT32_FORMAT, mono_atomic_load_i32 (&gc_stats.major_gc_count));
2178 mono_atomic_inc_i32 (&gc_stats.major_gc_count);
2179
2180 if (major_collector.start_major_collection)
2181 major_collector.start_major_collection ();
2182
2183 major_copy_or_mark_from_roots (gc_thread_gray_queue, old_next_pin_slot, concurrent ? COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT : COPY_OR_MARK_FROM_ROOTS_SERIAL, object_ops_nopar, object_ops_par);
2184 }
2185
2186 static void
major_finish_collection(SgenGrayQueue * gc_thread_gray_queue,const char * reason,gboolean is_overflow,size_t old_next_pin_slot,gboolean forced)2187 major_finish_collection (SgenGrayQueue *gc_thread_gray_queue, const char *reason, gboolean is_overflow, size_t old_next_pin_slot, gboolean forced)
2188 {
2189 ScannedObjectCounts counts;
2190 SgenObjectOperations *object_ops_nopar;
2191 mword fragment_total;
2192 TV_DECLARE (atv);
2193 TV_DECLARE (btv);
2194 guint64 major_scan_start = time_major_scan_mod_union_blocks;
2195 guint64 los_scan_start = time_major_scan_mod_union_los;
2196 guint64 finish_gray_start = time_major_finish_gray_stack;
2197
2198 if (concurrent_collection_in_progress) {
2199 SgenObjectOperations *object_ops_par = NULL;
2200
2201 object_ops_nopar = &major_collector.major_ops_concurrent_finish;
2202 if (major_collector.is_parallel)
2203 object_ops_par = &major_collector.major_ops_conc_par_finish;
2204
2205 major_copy_or_mark_from_roots (gc_thread_gray_queue, NULL, COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, object_ops_nopar, object_ops_par);
2206
2207 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
2208 main_gc_thread = NULL;
2209 #endif
2210 } else {
2211 object_ops_nopar = &major_collector.major_ops_serial;
2212 }
2213
2214 sgen_workers_assert_gray_queue_is_empty (GENERATION_OLD);
2215
2216 TV_GETTIME (btv);
2217 finish_gray_stack (GENERATION_OLD, CONTEXT_FROM_OBJECT_OPERATIONS (object_ops_nopar, gc_thread_gray_queue));
2218 TV_GETTIME (atv);
2219 time_major_finish_gray_stack += TV_ELAPSED (btv, atv);
2220
2221 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after joining");
2222
2223 if (objects_pinned) {
2224 g_assert (!concurrent_collection_in_progress);
2225
2226 /*
2227 * This is slow, but we just OOM'd.
2228 *
2229 * See comment at `sgen_pin_queue_clear_discarded_entries` for how the pin
2230 * queue is laid out at this point.
2231 */
2232 sgen_pin_queue_clear_discarded_entries (nursery_section, old_next_pin_slot);
2233 /*
2234 * We need to reestablish all pinned nursery objects in the pin queue
2235 * because they're needed for fragment creation. Unpinning happens by
2236 * walking the whole queue, so it's not necessary to reestablish where major
2237 * heap block pins are - all we care is that they're still in there
2238 * somewhere.
2239 */
2240 sgen_optimize_pin_queue ();
2241 sgen_find_section_pin_queue_start_end (nursery_section);
2242 objects_pinned = 0;
2243 }
2244
2245 reset_heap_boundaries ();
2246 sgen_update_heap_boundaries ((mword)sgen_get_nursery_start (), (mword)sgen_get_nursery_end ());
2247
2248 /* walk the pin_queue, build up the fragment list of free memory, unmark
2249 * pinned objects as we go, memzero() the empty fragments so they are ready for the
2250 * next allocations.
2251 */
2252 fragment_total = sgen_build_nursery_fragments (nursery_section, NULL);
2253 if (!fragment_total)
2254 degraded_mode = 1;
2255 SGEN_LOG (4, "Free space in nursery after major %ld", (long)fragment_total);
2256
2257 if (do_concurrent_checks && concurrent_collection_in_progress)
2258 sgen_debug_check_nursery_is_clean ();
2259
2260 /* prepare the pin queue for the next collection */
2261 sgen_finish_pinning ();
2262
2263 /* Clear TLABs for all threads */
2264 sgen_clear_tlabs ();
2265
2266 sgen_pin_stats_reset ();
2267
2268 sgen_cement_clear_below_threshold ();
2269
2270 if (check_mark_bits_after_major_collection)
2271 sgen_check_heap_marked (concurrent_collection_in_progress);
2272
2273 TV_GETTIME (btv);
2274 time_major_fragment_creation += TV_ELAPSED (atv, btv);
2275
2276 binary_protocol_sweep_begin (GENERATION_OLD, !major_collector.sweeps_lazily);
2277 sgen_memgov_major_pre_sweep ();
2278
2279 TV_GETTIME (atv);
2280 time_major_free_bigobjs += TV_ELAPSED (btv, atv);
2281
2282 sgen_los_sweep ();
2283
2284 TV_GETTIME (btv);
2285 time_major_los_sweep += TV_ELAPSED (atv, btv);
2286
2287 major_collector.sweep ();
2288
2289 binary_protocol_sweep_end (GENERATION_OLD, !major_collector.sweeps_lazily);
2290
2291 TV_GETTIME (atv);
2292 time_major_sweep += TV_ELAPSED (btv, atv);
2293
2294 sgen_debug_dump_heap ("major", mono_atomic_load_i32 (&gc_stats.major_gc_count) - 1, reason);
2295
2296 if (sgen_have_pending_finalizers ()) {
2297 SGEN_LOG (4, "Finalizer-thread wakeup");
2298 sgen_client_finalize_notify ();
2299 }
2300
2301 sgen_memgov_major_collection_end (forced, concurrent_collection_in_progress, reason, is_overflow);
2302 current_collection_generation = -1;
2303
2304 memset (&counts, 0, sizeof (ScannedObjectCounts));
2305 major_collector.finish_major_collection (&counts);
2306
2307 sgen_workers_assert_gray_queue_is_empty (GENERATION_OLD);
2308
2309 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after major collection has finished");
2310 if (concurrent_collection_in_progress)
2311 concurrent_collection_in_progress = FALSE;
2312
2313 check_scan_starts ();
2314
2315 binary_protocol_flush_buffers (FALSE);
2316
2317 //consistency_check ();
2318 if (major_collector.is_parallel)
2319 binary_protocol_collection_end_stats (0, 0, time_major_finish_gray_stack - finish_gray_start);
2320 else
2321 binary_protocol_collection_end_stats (
2322 time_major_scan_mod_union_blocks - major_scan_start,
2323 time_major_scan_mod_union_los - los_scan_start,
2324 time_major_finish_gray_stack - finish_gray_start);
2325
2326 binary_protocol_collection_end (mono_atomic_load_i32 (&gc_stats.major_gc_count) - 1, GENERATION_OLD, counts.num_scanned_objects, counts.num_unique_scanned_objects);
2327 }
2328
2329 static gboolean
major_do_collection(const char * reason,gboolean is_overflow,gboolean forced)2330 major_do_collection (const char *reason, gboolean is_overflow, gboolean forced)
2331 {
2332 TV_DECLARE (time_start);
2333 TV_DECLARE (time_end);
2334 size_t old_next_pin_slot;
2335 SgenGrayQueue gc_thread_gray_queue;
2336
2337 if (disable_major_collections)
2338 return FALSE;
2339
2340 if (major_collector.get_and_reset_num_major_objects_marked) {
2341 long long num_marked = major_collector.get_and_reset_num_major_objects_marked ();
2342 g_assert (!num_marked);
2343 }
2344
2345 /* world must be stopped already */
2346 TV_GETTIME (time_start);
2347
2348 init_gray_queue (&gc_thread_gray_queue);
2349 major_start_collection (&gc_thread_gray_queue, reason, FALSE, &old_next_pin_slot);
2350 major_finish_collection (&gc_thread_gray_queue, reason, is_overflow, old_next_pin_slot, forced);
2351 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2352
2353 TV_GETTIME (time_end);
2354 UnlockedAdd64 (&gc_stats.major_gc_time, TV_ELAPSED (time_start, time_end));
2355
2356 /* FIXME: also report this to the user, preferably in gc-end. */
2357 if (major_collector.get_and_reset_num_major_objects_marked)
2358 major_collector.get_and_reset_num_major_objects_marked ();
2359
2360 return bytes_pinned_from_failed_allocation > 0;
2361 }
2362
2363 static void
major_start_concurrent_collection(const char * reason)2364 major_start_concurrent_collection (const char *reason)
2365 {
2366 TV_DECLARE (time_start);
2367 TV_DECLARE (time_end);
2368 long long num_objects_marked;
2369 SgenGrayQueue gc_thread_gray_queue;
2370
2371 if (disable_major_collections)
2372 return;
2373
2374 TV_GETTIME (time_start);
2375 SGEN_TV_GETTIME (time_major_conc_collection_start);
2376
2377 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2378 g_assert (num_objects_marked == 0);
2379
2380 binary_protocol_concurrent_start ();
2381
2382 init_gray_queue (&gc_thread_gray_queue);
2383 // FIXME: store reason and pass it when finishing
2384 major_start_collection (&gc_thread_gray_queue, reason, TRUE, NULL);
2385 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2386
2387 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2388
2389 TV_GETTIME (time_end);
2390 UnlockedAdd64 (&gc_stats.major_gc_time, TV_ELAPSED (time_start, time_end));
2391
2392 current_collection_generation = -1;
2393 }
2394
2395 /*
2396 * Returns whether the major collection has finished.
2397 */
2398 static gboolean
major_should_finish_concurrent_collection(void)2399 major_should_finish_concurrent_collection (void)
2400 {
2401 return sgen_workers_all_done ();
2402 }
2403
2404 static void
major_update_concurrent_collection(void)2405 major_update_concurrent_collection (void)
2406 {
2407 TV_DECLARE (total_start);
2408 TV_DECLARE (total_end);
2409
2410 TV_GETTIME (total_start);
2411
2412 binary_protocol_concurrent_update ();
2413
2414 major_collector.update_cardtable_mod_union ();
2415 sgen_los_update_cardtable_mod_union ();
2416
2417 TV_GETTIME (total_end);
2418 UnlockedAdd64 (&gc_stats.major_gc_time, TV_ELAPSED (total_start, total_end));
2419 }
2420
2421 static void
major_finish_concurrent_collection(gboolean forced)2422 major_finish_concurrent_collection (gboolean forced)
2423 {
2424 SgenGrayQueue gc_thread_gray_queue;
2425 TV_DECLARE (total_start);
2426 TV_DECLARE (total_end);
2427
2428 TV_GETTIME (total_start);
2429
2430 binary_protocol_concurrent_finish ();
2431
2432 /*
2433 * We need to stop all workers since we're updating the cardtable below.
2434 * The workers will be resumed with a finishing pause context to avoid
2435 * additional cardtable and object scanning.
2436 */
2437 sgen_workers_stop_all_workers (GENERATION_OLD);
2438
2439 SGEN_TV_GETTIME (time_major_conc_collection_end);
2440 UnlockedAdd64 (&gc_stats.major_gc_time_concurrent, SGEN_TV_ELAPSED (time_major_conc_collection_start, time_major_conc_collection_end));
2441
2442 major_collector.update_cardtable_mod_union ();
2443 sgen_los_update_cardtable_mod_union ();
2444
2445 if (mod_union_consistency_check)
2446 sgen_check_mod_union_consistency ();
2447
2448 current_collection_generation = GENERATION_OLD;
2449 sgen_cement_reset ();
2450 init_gray_queue (&gc_thread_gray_queue);
2451 major_finish_collection (&gc_thread_gray_queue, "finishing", FALSE, -1, forced);
2452 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2453
2454 TV_GETTIME (total_end);
2455 UnlockedAdd64 (&gc_stats.major_gc_time, TV_ELAPSED (total_start, total_end));
2456
2457 current_collection_generation = -1;
2458 }
2459
2460 /*
2461 * Ensure an allocation request for @size will succeed by freeing enough memory.
2462 *
2463 * LOCKING: The GC lock MUST be held.
2464 */
2465 void
sgen_ensure_free_space(size_t size,int generation)2466 sgen_ensure_free_space (size_t size, int generation)
2467 {
2468 int generation_to_collect = -1;
2469 const char *reason = NULL;
2470
2471 if (generation == GENERATION_OLD) {
2472 if (sgen_need_major_collection (size)) {
2473 reason = "LOS overflow";
2474 generation_to_collect = GENERATION_OLD;
2475 }
2476 } else {
2477 if (degraded_mode) {
2478 if (sgen_need_major_collection (size)) {
2479 reason = "Degraded mode overflow";
2480 generation_to_collect = GENERATION_OLD;
2481 }
2482 } else if (sgen_need_major_collection (size)) {
2483 reason = concurrent_collection_in_progress ? "Forced finish concurrent collection" : "Minor allowance";
2484 generation_to_collect = GENERATION_OLD;
2485 } else {
2486 generation_to_collect = GENERATION_NURSERY;
2487 reason = "Nursery full";
2488 }
2489 }
2490
2491 if (generation_to_collect == -1) {
2492 if (concurrent_collection_in_progress && sgen_workers_all_done ()) {
2493 generation_to_collect = GENERATION_OLD;
2494 reason = "Finish concurrent collection";
2495 }
2496 }
2497
2498 if (generation_to_collect == -1)
2499 return;
2500 sgen_perform_collection (size, generation_to_collect, reason, FALSE, TRUE);
2501 }
2502
2503 /*
2504 * LOCKING: Assumes the GC lock is held.
2505 */
2506 static void
sgen_perform_collection_inner(size_t requested_size,int generation_to_collect,const char * reason,gboolean wait_to_finish,gboolean stw)2507 sgen_perform_collection_inner (size_t requested_size, int generation_to_collect, const char *reason, gboolean wait_to_finish, gboolean stw)
2508 {
2509 TV_DECLARE (gc_total_start);
2510 TV_DECLARE (gc_total_end);
2511 int overflow_generation_to_collect = -1;
2512 int oldest_generation_collected = generation_to_collect;
2513 const char *overflow_reason = NULL;
2514 gboolean finish_concurrent = concurrent_collection_in_progress && (major_should_finish_concurrent_collection () || generation_to_collect == GENERATION_OLD);
2515
2516 binary_protocol_collection_requested (generation_to_collect, requested_size, wait_to_finish ? 1 : 0);
2517
2518 SGEN_ASSERT (0, generation_to_collect == GENERATION_NURSERY || generation_to_collect == GENERATION_OLD, "What generation is this?");
2519
2520 if (stw)
2521 sgen_stop_world (generation_to_collect);
2522 else
2523 SGEN_ASSERT (0, sgen_is_world_stopped (), "We can only collect if the world is stopped");
2524
2525
2526 TV_GETTIME (gc_total_start);
2527
2528 // FIXME: extract overflow reason
2529 // FIXME: minor overflow for concurrent case
2530 if (generation_to_collect == GENERATION_NURSERY && !finish_concurrent) {
2531 if (concurrent_collection_in_progress)
2532 major_update_concurrent_collection ();
2533
2534 if (collect_nursery (reason, FALSE, NULL) && !concurrent_collection_in_progress) {
2535 overflow_generation_to_collect = GENERATION_OLD;
2536 overflow_reason = "Minor overflow";
2537 }
2538 } else if (finish_concurrent) {
2539 major_finish_concurrent_collection (wait_to_finish);
2540 oldest_generation_collected = GENERATION_OLD;
2541 } else {
2542 SGEN_ASSERT (0, generation_to_collect == GENERATION_OLD, "We should have handled nursery collections above");
2543 if (major_collector.is_concurrent && !wait_to_finish) {
2544 collect_nursery ("Concurrent start", FALSE, NULL);
2545 major_start_concurrent_collection (reason);
2546 oldest_generation_collected = GENERATION_NURSERY;
2547 } else if (major_do_collection (reason, FALSE, wait_to_finish)) {
2548 overflow_generation_to_collect = GENERATION_NURSERY;
2549 overflow_reason = "Excessive pinning";
2550 }
2551 }
2552
2553 if (overflow_generation_to_collect != -1) {
2554 SGEN_ASSERT (0, !concurrent_collection_in_progress, "We don't yet support overflow collections with the concurrent collector");
2555
2556 /*
2557 * We need to do an overflow collection, either because we ran out of memory
2558 * or the nursery is fully pinned.
2559 */
2560
2561 if (overflow_generation_to_collect == GENERATION_NURSERY)
2562 collect_nursery (overflow_reason, TRUE, NULL);
2563 else
2564 major_do_collection (overflow_reason, TRUE, wait_to_finish);
2565
2566 oldest_generation_collected = MAX (oldest_generation_collected, overflow_generation_to_collect);
2567 }
2568
2569 SGEN_LOG (2, "Heap size: %lu, LOS size: %lu", (unsigned long)sgen_gc_get_total_heap_allocation (), (unsigned long)los_memory_usage);
2570
2571 /* this also sets the proper pointers for the next allocation */
2572 if (generation_to_collect == GENERATION_NURSERY && !sgen_can_alloc_size (requested_size)) {
2573 /* TypeBuilder and MonoMethod are killing mcs with fragmentation */
2574 SGEN_LOG (1, "nursery collection didn't find enough room for %zd alloc (%zd pinned)", requested_size, sgen_get_pinned_count ());
2575 sgen_dump_pin_queue ();
2576 degraded_mode = 1;
2577 }
2578
2579 TV_GETTIME (gc_total_end);
2580 time_max = MAX (time_max, TV_ELAPSED (gc_total_start, gc_total_end));
2581
2582 if (stw)
2583 sgen_restart_world (oldest_generation_collected);
2584 }
2585
2586 #ifdef HOST_WASM
2587
2588 typedef struct {
2589 size_t requested_size;
2590 int generation_to_collect;
2591 const char *reason;
2592 } SgenGcRequest;
2593
2594 static SgenGcRequest gc_request;
2595 static gboolean pending_request;
2596
2597 extern void request_gc_cycle (void);
2598
2599 #include <emscripten.h>
2600
2601 EMSCRIPTEN_KEEPALIVE void
mono_gc_pump_callback(void)2602 mono_gc_pump_callback (void)
2603 {
2604 if (!pending_request)
2605 return;
2606 pending_request = FALSE;
2607 sgen_perform_collection_inner (gc_request.requested_size, gc_request.generation_to_collect, gc_request.reason, TRUE, TRUE);
2608 }
2609 #endif
2610
2611 void
sgen_perform_collection(size_t requested_size,int generation_to_collect,const char * reason,gboolean wait_to_finish,gboolean stw)2612 sgen_perform_collection (size_t requested_size, int generation_to_collect, const char *reason, gboolean wait_to_finish, gboolean stw)
2613 {
2614 #ifdef HOST_WASM
2615 g_assert (stw); //can't handle non-stw mode (IE, domain unload)
2616 //we ignore wait_to_finish
2617 if (!pending_request || gc_request.generation_to_collect <= generation_to_collect) { //no active request or request was for a smaller part of the heap
2618 gc_request.requested_size = requested_size;
2619 gc_request.generation_to_collect = generation_to_collect;
2620 gc_request.reason = reason;
2621 if (!pending_request) {
2622 request_gc_cycle ();
2623 pending_request = TRUE;
2624 }
2625 }
2626
2627 degraded_mode = 1; //enable degraded mode so allocation can continue
2628 #else
2629 sgen_perform_collection_inner (requested_size, generation_to_collect, reason, wait_to_finish, stw);
2630 #endif
2631 }
2632 /*
2633 * ######################################################################
2634 * ######## Memory allocation from the OS
2635 * ######################################################################
2636 * This section of code deals with getting memory from the OS and
2637 * allocating memory for GC-internal data structures.
2638 * Internal memory can be handled with a freelist for small objects.
2639 */
2640
2641 /*
2642 * Debug reporting.
2643 */
2644 G_GNUC_UNUSED static void
report_internal_mem_usage(void)2645 report_internal_mem_usage (void)
2646 {
2647 printf ("Internal memory usage:\n");
2648 sgen_report_internal_mem_usage ();
2649 printf ("Pinned memory usage:\n");
2650 major_collector.report_pinned_memory_usage ();
2651 }
2652
2653 /*
2654 * ######################################################################
2655 * ######## Finalization support
2656 * ######################################################################
2657 */
2658
2659 /*
2660 * This function returns true if @object is either alive and belongs to the
2661 * current collection - major collections are full heap, so old gen objects
2662 * are never alive during a minor collection.
2663 */
2664 static inline int
sgen_is_object_alive_and_on_current_collection(GCObject * object)2665 sgen_is_object_alive_and_on_current_collection (GCObject *object)
2666 {
2667 if (ptr_in_nursery (object))
2668 return sgen_nursery_is_object_alive (object);
2669
2670 if (current_collection_generation == GENERATION_NURSERY)
2671 return FALSE;
2672
2673 return sgen_major_is_object_alive (object);
2674 }
2675
2676
2677 gboolean
sgen_gc_is_object_ready_for_finalization(GCObject * object)2678 sgen_gc_is_object_ready_for_finalization (GCObject *object)
2679 {
2680 return !sgen_is_object_alive (object);
2681 }
2682
2683 void
sgen_queue_finalization_entry(GCObject * obj)2684 sgen_queue_finalization_entry (GCObject *obj)
2685 {
2686 gboolean critical = sgen_client_object_has_critical_finalizer (obj);
2687
2688 sgen_pointer_queue_add (critical ? &critical_fin_queue : &fin_ready_queue, obj);
2689
2690 sgen_client_object_queued_for_finalization (obj);
2691 }
2692
2693 gboolean
sgen_object_is_live(GCObject * obj)2694 sgen_object_is_live (GCObject *obj)
2695 {
2696 return sgen_is_object_alive_and_on_current_collection (obj);
2697 }
2698
2699 /*
2700 * `System.GC.WaitForPendingFinalizers` first checks `sgen_have_pending_finalizers()` to
2701 * determine whether it can exit quickly. The latter must therefore only return FALSE if
2702 * all finalizers have really finished running.
2703 *
2704 * `sgen_gc_invoke_finalizers()` first dequeues a finalizable object, and then finalizes it.
2705 * This means that just checking whether the queues are empty leaves the possibility that an
2706 * object might have been dequeued but not yet finalized. That's why we need the additional
2707 * flag `pending_unqueued_finalizer`.
2708 */
2709
2710 static volatile gboolean pending_unqueued_finalizer = FALSE;
2711 volatile gboolean sgen_suspend_finalizers = FALSE;
2712
2713 void
sgen_set_suspend_finalizers(void)2714 sgen_set_suspend_finalizers (void)
2715 {
2716 sgen_suspend_finalizers = TRUE;
2717 }
2718
2719 int
sgen_gc_invoke_finalizers(void)2720 sgen_gc_invoke_finalizers (void)
2721 {
2722 int count = 0;
2723
2724 g_assert (!pending_unqueued_finalizer);
2725
2726 /* FIXME: batch to reduce lock contention */
2727 while (sgen_have_pending_finalizers ()) {
2728 GCObject *obj;
2729
2730 LOCK_GC;
2731
2732 /*
2733 * We need to set `pending_unqueued_finalizer` before dequeing the
2734 * finalizable object.
2735 */
2736 if (!sgen_pointer_queue_is_empty (&fin_ready_queue)) {
2737 pending_unqueued_finalizer = TRUE;
2738 mono_memory_write_barrier ();
2739 obj = (GCObject *)sgen_pointer_queue_pop (&fin_ready_queue);
2740 } else if (!sgen_pointer_queue_is_empty (&critical_fin_queue)) {
2741 pending_unqueued_finalizer = TRUE;
2742 mono_memory_write_barrier ();
2743 obj = (GCObject *)sgen_pointer_queue_pop (&critical_fin_queue);
2744 } else {
2745 obj = NULL;
2746 }
2747
2748 if (obj)
2749 SGEN_LOG (7, "Finalizing object %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
2750
2751 UNLOCK_GC;
2752
2753 if (!obj)
2754 break;
2755
2756 count++;
2757 /* the object is on the stack so it is pinned */
2758 /*g_print ("Calling finalizer for object: %p (%s)\n", obj, sgen_client_object_safe_name (obj));*/
2759 sgen_client_run_finalize (obj);
2760 }
2761
2762 if (pending_unqueued_finalizer) {
2763 mono_memory_write_barrier ();
2764 pending_unqueued_finalizer = FALSE;
2765 }
2766
2767 return count;
2768 }
2769
2770 gboolean
sgen_have_pending_finalizers(void)2771 sgen_have_pending_finalizers (void)
2772 {
2773 if (sgen_suspend_finalizers)
2774 return FALSE;
2775 return pending_unqueued_finalizer || !sgen_pointer_queue_is_empty (&fin_ready_queue) || !sgen_pointer_queue_is_empty (&critical_fin_queue);
2776 }
2777
2778 /*
2779 * ######################################################################
2780 * ######## registered roots support
2781 * ######################################################################
2782 */
2783
2784 /*
2785 * We do not coalesce roots.
2786 */
2787 int
sgen_register_root(char * start,size_t size,SgenDescriptor descr,int root_type,int source,const char * msg)2788 sgen_register_root (char *start, size_t size, SgenDescriptor descr, int root_type, int source, const char *msg)
2789 {
2790 RootRecord new_root;
2791 int i;
2792 LOCK_GC;
2793 for (i = 0; i < ROOT_TYPE_NUM; ++i) {
2794 RootRecord *root = (RootRecord *)sgen_hash_table_lookup (&roots_hash [i], start);
2795 /* we allow changing the size and the descriptor (for thread statics etc) */
2796 if (root) {
2797 size_t old_size = root->end_root - start;
2798 root->end_root = start + size;
2799 SGEN_ASSERT (0, !!root->root_desc == !!descr, "Can't change whether a root is precise or conservative.");
2800 SGEN_ASSERT (0, root->source == source, "Can't change a root's source identifier.");
2801 SGEN_ASSERT (0, !!root->msg == !!msg, "Can't change a root's message.");
2802 root->root_desc = descr;
2803 roots_size += size;
2804 roots_size -= old_size;
2805 UNLOCK_GC;
2806 return TRUE;
2807 }
2808 }
2809
2810 new_root.end_root = start + size;
2811 new_root.root_desc = descr;
2812 new_root.source = source;
2813 new_root.msg = msg;
2814
2815 sgen_hash_table_replace (&roots_hash [root_type], start, &new_root, NULL);
2816 roots_size += size;
2817
2818 SGEN_LOG (3, "Added root for range: %p-%p, descr: %llx (%d/%d bytes)", start, new_root.end_root, (long long)descr, (int)size, (int)roots_size);
2819
2820 UNLOCK_GC;
2821 return TRUE;
2822 }
2823
2824 void
sgen_deregister_root(char * addr)2825 sgen_deregister_root (char* addr)
2826 {
2827 int root_type;
2828 RootRecord root;
2829
2830 LOCK_GC;
2831 for (root_type = 0; root_type < ROOT_TYPE_NUM; ++root_type) {
2832 if (sgen_hash_table_remove (&roots_hash [root_type], addr, &root))
2833 roots_size -= (root.end_root - addr);
2834 }
2835 UNLOCK_GC;
2836 }
2837
2838 void
sgen_wbroots_iterate_live_block_ranges(sgen_cardtable_block_callback cb)2839 sgen_wbroots_iterate_live_block_ranges (sgen_cardtable_block_callback cb)
2840 {
2841 void **start_root;
2842 RootRecord *root;
2843 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_WBARRIER], void **, start_root, RootRecord *, root) {
2844 cb ((mword)start_root, (mword)root->end_root - (mword)start_root);
2845 } SGEN_HASH_TABLE_FOREACH_END;
2846 }
2847
2848 /* Root equivalent of sgen_client_cardtable_scan_object */
2849 static void
sgen_wbroot_scan_card_table(void ** start_root,mword size,ScanCopyContext ctx)2850 sgen_wbroot_scan_card_table (void** start_root, mword size, ScanCopyContext ctx)
2851 {
2852 ScanPtrFieldFunc scan_field_func = ctx.ops->scan_ptr_field;
2853 guint8 *card_data = sgen_card_table_get_card_scan_address ((mword)start_root);
2854 guint8 *card_base = card_data;
2855 mword card_count = sgen_card_table_number_of_cards_in_range ((mword)start_root, size);
2856 guint8 *card_data_end = card_data + card_count;
2857 mword extra_idx = 0;
2858 char *obj_start = sgen_card_table_align_pointer (start_root);
2859 char *obj_end = (char*)start_root + size;
2860 #ifdef SGEN_HAVE_OVERLAPPING_CARDS
2861 guint8 *overflow_scan_end = NULL;
2862 #endif
2863
2864 #ifdef SGEN_HAVE_OVERLAPPING_CARDS
2865 /*Check for overflow and if so, setup to scan in two steps*/
2866 if (card_data_end >= SGEN_SHADOW_CARDTABLE_END) {
2867 overflow_scan_end = sgen_shadow_cardtable + (card_data_end - SGEN_SHADOW_CARDTABLE_END);
2868 card_data_end = SGEN_SHADOW_CARDTABLE_END;
2869 }
2870
2871 LOOP_HEAD:
2872 #endif
2873
2874 card_data = sgen_find_next_card (card_data, card_data_end);
2875
2876 for (; card_data < card_data_end; card_data = sgen_find_next_card (card_data + 1, card_data_end)) {
2877 size_t idx = (card_data - card_base) + extra_idx;
2878 char *start = (char*)(obj_start + idx * CARD_SIZE_IN_BYTES);
2879 char *card_end = start + CARD_SIZE_IN_BYTES;
2880 char *elem = start, *first_elem = start;
2881
2882 /*
2883 * Don't clean first and last card on 32bit systems since they
2884 * may also be part from other roots.
2885 */
2886 if (card_data != card_base && card_data != (card_data_end - 1))
2887 sgen_card_table_prepare_card_for_scanning (card_data);
2888
2889 card_end = MIN (card_end, obj_end);
2890
2891 if (elem < (char*)start_root)
2892 first_elem = elem = (char*)start_root;
2893
2894 for (; elem < card_end; elem += SIZEOF_VOID_P) {
2895 if (*(GCObject**)elem)
2896 scan_field_func (NULL, (GCObject**)elem, ctx.queue);
2897 }
2898
2899 binary_protocol_card_scan (first_elem, elem - first_elem);
2900 }
2901
2902 #ifdef SGEN_HAVE_OVERLAPPING_CARDS
2903 if (overflow_scan_end) {
2904 extra_idx = card_data - card_base;
2905 card_base = card_data = sgen_shadow_cardtable;
2906 card_data_end = overflow_scan_end;
2907 overflow_scan_end = NULL;
2908 goto LOOP_HEAD;
2909 }
2910 #endif
2911 }
2912
2913 void
sgen_wbroots_scan_card_table(ScanCopyContext ctx)2914 sgen_wbroots_scan_card_table (ScanCopyContext ctx)
2915 {
2916 void **start_root;
2917 RootRecord *root;
2918
2919 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_WBARRIER], void **, start_root, RootRecord *, root) {
2920 SGEN_ASSERT (0, (root->root_desc & ROOT_DESC_TYPE_MASK) == ROOT_DESC_VECTOR, "Unsupported root type");
2921
2922 sgen_wbroot_scan_card_table (start_root, (mword)root->end_root - (mword)start_root, ctx);
2923 } SGEN_HASH_TABLE_FOREACH_END;
2924 }
2925
2926 /*
2927 * ######################################################################
2928 * ######## Thread handling (stop/start code)
2929 * ######################################################################
2930 */
2931
2932 int
sgen_get_current_collection_generation(void)2933 sgen_get_current_collection_generation (void)
2934 {
2935 return current_collection_generation;
2936 }
2937
2938 void*
sgen_thread_attach(SgenThreadInfo * info)2939 sgen_thread_attach (SgenThreadInfo* info)
2940 {
2941 info->tlab_start = info->tlab_next = info->tlab_temp_end = info->tlab_real_end = NULL;
2942
2943 sgen_client_thread_attach (info);
2944
2945 return info;
2946 }
2947
2948 void
sgen_thread_detach_with_lock(SgenThreadInfo * p)2949 sgen_thread_detach_with_lock (SgenThreadInfo *p)
2950 {
2951 sgen_client_thread_detach_with_lock (p);
2952 }
2953
2954 /*
2955 * ######################################################################
2956 * ######## Write barriers
2957 * ######################################################################
2958 */
2959
2960 /*
2961 * Note: the write barriers first do the needed GC work and then do the actual store:
2962 * this way the value is visible to the conservative GC scan after the write barrier
2963 * itself. If a GC interrupts the barrier in the middle, value will be kept alive by
2964 * the conservative scan, otherwise by the remembered set scan.
2965 */
2966
2967 /**
2968 * mono_gc_wbarrier_arrayref_copy:
2969 */
2970 void
mono_gc_wbarrier_arrayref_copy(gpointer dest_ptr,gpointer src_ptr,int count)2971 mono_gc_wbarrier_arrayref_copy (gpointer dest_ptr, gpointer src_ptr, int count)
2972 {
2973 HEAVY_STAT (++stat_wbarrier_arrayref_copy);
2974 /*This check can be done without taking a lock since dest_ptr array is pinned*/
2975 if (ptr_in_nursery (dest_ptr) || count <= 0) {
2976 mono_gc_memmove_aligned (dest_ptr, src_ptr, count * sizeof (gpointer));
2977 return;
2978 }
2979
2980 #ifdef SGEN_HEAVY_BINARY_PROTOCOL
2981 if (binary_protocol_is_heavy_enabled ()) {
2982 int i;
2983 for (i = 0; i < count; ++i) {
2984 gpointer dest = (gpointer*)dest_ptr + i;
2985 gpointer obj = *((gpointer*)src_ptr + i);
2986 if (obj)
2987 binary_protocol_wbarrier (dest, obj, (gpointer)LOAD_VTABLE (obj));
2988 }
2989 }
2990 #endif
2991
2992 remset.wbarrier_arrayref_copy (dest_ptr, src_ptr, count);
2993 }
2994
2995 /**
2996 * mono_gc_wbarrier_generic_nostore:
2997 */
2998 void
mono_gc_wbarrier_generic_nostore(gpointer ptr)2999 mono_gc_wbarrier_generic_nostore (gpointer ptr)
3000 {
3001 gpointer obj;
3002
3003 HEAVY_STAT (++stat_wbarrier_generic_store);
3004
3005 sgen_client_wbarrier_generic_nostore_check (ptr);
3006
3007 obj = *(gpointer*)ptr;
3008 if (obj)
3009 binary_protocol_wbarrier (ptr, obj, (gpointer)LOAD_VTABLE (obj));
3010
3011 /*
3012 * We need to record old->old pointer locations for the
3013 * concurrent collector.
3014 */
3015 if (!ptr_in_nursery (obj) && !concurrent_collection_in_progress) {
3016 SGEN_LOG (8, "Skipping remset at %p", ptr);
3017 return;
3018 }
3019
3020 SGEN_LOG (8, "Adding remset at %p", ptr);
3021
3022 remset.wbarrier_generic_nostore (ptr);
3023 }
3024
3025 /**
3026 * mono_gc_wbarrier_generic_store:
3027 */
3028 void
mono_gc_wbarrier_generic_store(gpointer ptr,GCObject * value)3029 mono_gc_wbarrier_generic_store (gpointer ptr, GCObject* value)
3030 {
3031 SGEN_LOG (8, "Wbarrier store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
3032 SGEN_UPDATE_REFERENCE_ALLOW_NULL (ptr, value);
3033 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
3034 mono_gc_wbarrier_generic_nostore (ptr);
3035 sgen_dummy_use (value);
3036 }
3037
3038 /**
3039 * mono_gc_wbarrier_generic_store_atomic:
3040 * Same as \c mono_gc_wbarrier_generic_store but performs the store
3041 * as an atomic operation with release semantics.
3042 */
3043 void
mono_gc_wbarrier_generic_store_atomic(gpointer ptr,GCObject * value)3044 mono_gc_wbarrier_generic_store_atomic (gpointer ptr, GCObject *value)
3045 {
3046 HEAVY_STAT (++stat_wbarrier_generic_store_atomic);
3047
3048 SGEN_LOG (8, "Wbarrier atomic store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
3049
3050 mono_atomic_store_ptr ((volatile gpointer *)ptr, value);
3051
3052 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
3053 mono_gc_wbarrier_generic_nostore (ptr);
3054
3055 sgen_dummy_use (value);
3056 }
3057
3058 void
sgen_wbarrier_range_copy(gpointer _dest,gpointer _src,int size)3059 sgen_wbarrier_range_copy (gpointer _dest, gpointer _src, int size)
3060 {
3061 remset.wbarrier_range_copy (_dest,_src, size);
3062 }
3063
3064 /*
3065 * ######################################################################
3066 * ######## Other mono public interface functions.
3067 * ######################################################################
3068 */
3069
3070 void
sgen_gc_collect(int generation)3071 sgen_gc_collect (int generation)
3072 {
3073 LOCK_GC;
3074 if (generation > 1)
3075 generation = 1;
3076 sgen_perform_collection (0, generation, "user request", TRUE, TRUE);
3077 /* Make sure we don't exceed heap size allowance by promoting */
3078 if (generation == GENERATION_NURSERY && sgen_need_major_collection (0))
3079 sgen_perform_collection (0, GENERATION_OLD, "Minor allowance", FALSE, TRUE);
3080 UNLOCK_GC;
3081 }
3082
3083 int
sgen_gc_collection_count(int generation)3084 sgen_gc_collection_count (int generation)
3085 {
3086 return mono_atomic_load_i32 (generation == GENERATION_NURSERY ? &gc_stats.minor_gc_count : &gc_stats.major_gc_count);
3087 }
3088
3089 size_t
sgen_gc_get_used_size(void)3090 sgen_gc_get_used_size (void)
3091 {
3092 gint64 tot = 0;
3093 LOCK_GC;
3094 tot = los_memory_usage;
3095 tot += nursery_section->end_data - nursery_section->data;
3096 tot += major_collector.get_used_size ();
3097 /* FIXME: account for pinned objects */
3098 UNLOCK_GC;
3099 return tot;
3100 }
3101
3102 void
sgen_env_var_error(const char * env_var,const char * fallback,const char * description_format,...)3103 sgen_env_var_error (const char *env_var, const char *fallback, const char *description_format, ...)
3104 {
3105 va_list ap;
3106
3107 va_start (ap, description_format);
3108
3109 fprintf (stderr, "Warning: In environment variable `%s': ", env_var);
3110 vfprintf (stderr, description_format, ap);
3111 if (fallback)
3112 fprintf (stderr, " - %s", fallback);
3113 fprintf (stderr, "\n");
3114
3115 va_end (ap);
3116 }
3117
3118 static gboolean
parse_double_in_interval(const char * env_var,const char * opt_name,const char * opt,double min,double max,double * result)3119 parse_double_in_interval (const char *env_var, const char *opt_name, const char *opt, double min, double max, double *result)
3120 {
3121 char *endptr;
3122 double val = strtod (opt, &endptr);
3123 if (endptr == opt) {
3124 sgen_env_var_error (env_var, "Using default value.", "`%s` must be a number.", opt_name);
3125 return FALSE;
3126 }
3127 else if (val < min || val > max) {
3128 sgen_env_var_error (env_var, "Using default value.", "`%s` must be between %.2f - %.2f.", opt_name, min, max);
3129 return FALSE;
3130 }
3131 *result = val;
3132 return TRUE;
3133 }
3134
3135 static SgenMinor
parse_sgen_minor(const char * opt)3136 parse_sgen_minor (const char *opt)
3137 {
3138 if (!opt)
3139 return SGEN_MINOR_DEFAULT;
3140
3141 if (!strcmp (opt, "simple")) {
3142 return SGEN_MINOR_SIMPLE;
3143 } else if (!strcmp (opt, "simple-par")) {
3144 return SGEN_MINOR_SIMPLE_PARALLEL;
3145 } else if (!strcmp (opt, "split")) {
3146 return SGEN_MINOR_SPLIT;
3147 } else {
3148 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default instead.", "Unknown minor collector `%s'.", opt);
3149 return SGEN_MINOR_DEFAULT;
3150 }
3151 }
3152
3153 static SgenMajor
parse_sgen_major(const char * opt)3154 parse_sgen_major (const char *opt)
3155 {
3156 if (!opt)
3157 return SGEN_MAJOR_DEFAULT;
3158
3159 if (!strcmp (opt, "marksweep")) {
3160 return SGEN_MAJOR_SERIAL;
3161 } else if (!strcmp (opt, "marksweep-conc")) {
3162 return SGEN_MAJOR_CONCURRENT;
3163 } else if (!strcmp (opt, "marksweep-conc-par")) {
3164 return SGEN_MAJOR_CONCURRENT_PARALLEL;
3165 } else {
3166 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default instead.", "Unknown major collector `%s'.", opt);
3167 return SGEN_MAJOR_DEFAULT;
3168 }
3169
3170 }
3171
3172 static SgenMode
parse_sgen_mode(const char * opt)3173 parse_sgen_mode (const char *opt)
3174 {
3175 if (!opt)
3176 return SGEN_MODE_NONE;
3177
3178 if (!strcmp (opt, "balanced")) {
3179 return SGEN_MODE_BALANCED;
3180 } else if (!strcmp (opt, "throughput")) {
3181 return SGEN_MODE_THROUGHPUT;
3182 } else if (!strcmp (opt, "pause") || g_str_has_prefix (opt, "pause:")) {
3183 return SGEN_MODE_PAUSE;
3184 } else {
3185 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default configurations.", "Unknown mode `%s'.", opt);
3186 return SGEN_MODE_NONE;
3187 }
3188 }
3189
3190 static void
init_sgen_minor(SgenMinor minor)3191 init_sgen_minor (SgenMinor minor)
3192 {
3193 switch (minor) {
3194 case SGEN_MINOR_DEFAULT:
3195 case SGEN_MINOR_SIMPLE:
3196 sgen_simple_nursery_init (&sgen_minor_collector, FALSE);
3197 break;
3198 case SGEN_MINOR_SIMPLE_PARALLEL:
3199 sgen_simple_nursery_init (&sgen_minor_collector, TRUE);
3200 break;
3201 case SGEN_MINOR_SPLIT:
3202 sgen_split_nursery_init (&sgen_minor_collector);
3203 break;
3204 default:
3205 g_assert_not_reached ();
3206 }
3207 }
3208
3209 static void
init_sgen_major(SgenMajor major)3210 init_sgen_major (SgenMajor major)
3211 {
3212 if (major == SGEN_MAJOR_DEFAULT)
3213 major = DEFAULT_MAJOR;
3214
3215 switch (major) {
3216 case SGEN_MAJOR_SERIAL:
3217 sgen_marksweep_init (&major_collector);
3218 break;
3219 case SGEN_MAJOR_CONCURRENT:
3220 sgen_marksweep_conc_init (&major_collector);
3221 break;
3222 case SGEN_MAJOR_CONCURRENT_PARALLEL:
3223 sgen_marksweep_conc_par_init (&major_collector);
3224 break;
3225 default:
3226 g_assert_not_reached ();
3227 }
3228 }
3229
3230 /*
3231 * If sgen mode is set, major/minor configuration is fixed. The other gc_params
3232 * are parsed and processed after major/minor initialization, so it can potentially
3233 * override some knobs set by the sgen mode. We can consider locking out additional
3234 * configurations when gc_modes are used.
3235 */
3236 static void
init_sgen_mode(SgenMode mode)3237 init_sgen_mode (SgenMode mode)
3238 {
3239 SgenMinor minor = SGEN_MINOR_DEFAULT;
3240 SgenMajor major = SGEN_MAJOR_DEFAULT;
3241
3242 switch (mode) {
3243 case SGEN_MODE_BALANCED:
3244 /*
3245 * Use a dynamic parallel nursery with a major concurrent collector.
3246 * This uses the default values for max pause time and nursery size.
3247 */
3248 minor = SGEN_MINOR_SIMPLE;
3249 major = SGEN_MAJOR_CONCURRENT;
3250 dynamic_nursery = TRUE;
3251 break;
3252 case SGEN_MODE_THROUGHPUT:
3253 /*
3254 * Use concurrent major to let the mutator do more work. Use a larger
3255 * nursery, without pause time constraints, in order to collect more
3256 * objects in parallel and avoid repetitive collection tasks (pinning,
3257 * root scanning etc)
3258 */
3259 minor = SGEN_MINOR_SIMPLE_PARALLEL;
3260 major = SGEN_MAJOR_CONCURRENT;
3261 dynamic_nursery = TRUE;
3262 sgen_max_pause_time = 0;
3263 break;
3264 case SGEN_MODE_PAUSE:
3265 /*
3266 * Use concurrent major and dynamic nursery with a more
3267 * aggressive shrinking relative to pause times.
3268 */
3269 minor = SGEN_MINOR_SIMPLE_PARALLEL;
3270 major = SGEN_MAJOR_CONCURRENT;
3271 dynamic_nursery = TRUE;
3272 sgen_max_pause_margin = SGEN_PAUSE_MODE_MAX_PAUSE_MARGIN;
3273 break;
3274 default:
3275 g_assert_not_reached ();
3276 }
3277
3278 init_sgen_minor (minor);
3279 init_sgen_major (major);
3280 }
3281
3282 void
sgen_gc_init(void)3283 sgen_gc_init (void)
3284 {
3285 char *env;
3286 char **opts, **ptr;
3287 SgenMajor sgen_major = SGEN_MAJOR_DEFAULT;
3288 SgenMinor sgen_minor = SGEN_MINOR_DEFAULT;
3289 SgenMode sgen_mode = SGEN_MODE_NONE;
3290 char *params_opts = NULL;
3291 char *debug_opts = NULL;
3292 size_t max_heap = 0;
3293 size_t soft_limit = 0;
3294 int result;
3295 gboolean debug_print_allowance = FALSE;
3296 double allowance_ratio = 0, save_target = 0;
3297 gboolean cement_enabled = TRUE;
3298
3299 do {
3300 result = mono_atomic_cas_i32 (&gc_initialized, -1, 0);
3301 switch (result) {
3302 case 1:
3303 /* already inited */
3304 return;
3305 case -1:
3306 /* being inited by another thread */
3307 mono_thread_info_usleep (1000);
3308 break;
3309 case 0:
3310 /* we will init it */
3311 break;
3312 default:
3313 g_assert_not_reached ();
3314 }
3315 } while (result != 0);
3316
3317 SGEN_TV_GETTIME (sgen_init_timestamp);
3318
3319 #ifdef SGEN_WITHOUT_MONO
3320 mono_thread_smr_init ();
3321 #endif
3322
3323 mono_coop_mutex_init (&gc_mutex);
3324
3325 gc_debug_file = stderr;
3326
3327 mono_coop_mutex_init (&sgen_interruption_mutex);
3328
3329 if ((env = g_getenv (MONO_GC_PARAMS_NAME)) || gc_params_options) {
3330 params_opts = g_strdup_printf ("%s,%s", gc_params_options ? gc_params_options : "", env ? env : "");
3331 g_free (env);
3332 }
3333
3334 if (params_opts) {
3335 opts = g_strsplit (params_opts, ",", -1);
3336 for (ptr = opts; *ptr; ++ptr) {
3337 char *opt = *ptr;
3338 if (g_str_has_prefix (opt, "major=")) {
3339 opt = strchr (opt, '=') + 1;
3340 sgen_major = parse_sgen_major (opt);
3341 } else if (g_str_has_prefix (opt, "minor=")) {
3342 opt = strchr (opt, '=') + 1;
3343 sgen_minor = parse_sgen_minor (opt);
3344 } else if (g_str_has_prefix (opt, "mode=")) {
3345 opt = strchr (opt, '=') + 1;
3346 sgen_mode = parse_sgen_mode (opt);
3347 }
3348 }
3349 } else {
3350 opts = NULL;
3351 }
3352
3353 init_stats ();
3354 sgen_init_internal_allocator ();
3355 sgen_init_nursery_allocator ();
3356 sgen_init_fin_weak_hash ();
3357 sgen_init_hash_table ();
3358 sgen_init_descriptors ();
3359 sgen_init_gray_queues ();
3360 sgen_init_allocator ();
3361 sgen_init_gchandles ();
3362
3363 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_SECTION, SGEN_SIZEOF_GC_MEM_SECTION);
3364 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_GRAY_QUEUE, sizeof (GrayQueueSection));
3365
3366 sgen_client_init ();
3367
3368 if (sgen_mode != SGEN_MODE_NONE) {
3369 if (sgen_minor != SGEN_MINOR_DEFAULT || sgen_major != SGEN_MAJOR_DEFAULT)
3370 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Ignoring major/minor configuration", "Major/minor configurations cannot be used with sgen modes");
3371 init_sgen_mode (sgen_mode);
3372 } else {
3373 init_sgen_minor (sgen_minor);
3374 init_sgen_major (sgen_major);
3375 }
3376
3377 if (opts) {
3378 gboolean usage_printed = FALSE;
3379
3380 for (ptr = opts; *ptr; ++ptr) {
3381 char *opt = *ptr;
3382 if (!strcmp (opt, ""))
3383 continue;
3384 if (g_str_has_prefix (opt, "major="))
3385 continue;
3386 if (g_str_has_prefix (opt, "minor="))
3387 continue;
3388 if (g_str_has_prefix (opt, "mode=")) {
3389 if (g_str_has_prefix (opt, "mode=pause:")) {
3390 char *str_pause = strchr (opt, ':') + 1;
3391 int pause = atoi (str_pause);
3392 if (pause)
3393 sgen_max_pause_time = pause;
3394 else
3395 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default", "Invalid maximum pause time for `pause` sgen mode");
3396 }
3397 continue;
3398 }
3399 if (g_str_has_prefix (opt, "max-heap-size=")) {
3400 size_t page_size = mono_pagesize ();
3401 size_t max_heap_candidate = 0;
3402 opt = strchr (opt, '=') + 1;
3403 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &max_heap_candidate)) {
3404 max_heap = (max_heap_candidate + page_size - 1) & ~(size_t)(page_size - 1);
3405 if (max_heap != max_heap_candidate)
3406 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Rounding up.", "`max-heap-size` size must be a multiple of %d.", page_size);
3407 } else {
3408 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`max-heap-size` must be an integer.");
3409 }
3410 continue;
3411 }
3412 if (g_str_has_prefix (opt, "soft-heap-limit=")) {
3413 opt = strchr (opt, '=') + 1;
3414 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &soft_limit)) {
3415 if (soft_limit <= 0) {
3416 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be positive.");
3417 soft_limit = 0;
3418 }
3419 } else {
3420 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be an integer.");
3421 }
3422 continue;
3423 }
3424 if (g_str_has_prefix (opt, "nursery-size=")) {
3425 size_t val;
3426 opt = strchr (opt, '=') + 1;
3427 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &val)) {
3428 if ((val & (val - 1))) {
3429 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be a power of two.");
3430 continue;
3431 }
3432
3433 if (val < SGEN_MAX_NURSERY_WASTE) {
3434 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.",
3435 "`nursery-size` must be at least %d bytes.", SGEN_MAX_NURSERY_WASTE);
3436 continue;
3437 }
3438
3439 min_nursery_size = max_nursery_size = val;
3440 dynamic_nursery = FALSE;
3441 } else {
3442 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be an integer.");
3443 continue;
3444 }
3445 continue;
3446 }
3447 if (g_str_has_prefix (opt, "save-target-ratio=")) {
3448 double val;
3449 opt = strchr (opt, '=') + 1;
3450 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "save-target-ratio", opt,
3451 SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO, &val)) {
3452 save_target = val;
3453 }
3454 continue;
3455 }
3456 if (g_str_has_prefix (opt, "default-allowance-ratio=")) {
3457 double val;
3458 opt = strchr (opt, '=') + 1;
3459 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "default-allowance-ratio", opt,
3460 SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, SGEN_MAX_ALLOWANCE_NURSERY_SIZE_RATIO, &val)) {
3461 allowance_ratio = val;
3462 }
3463 continue;
3464 }
3465
3466 if (!strcmp (opt, "cementing")) {
3467 cement_enabled = TRUE;
3468 continue;
3469 }
3470 if (!strcmp (opt, "no-cementing")) {
3471 cement_enabled = FALSE;
3472 continue;
3473 }
3474
3475 if (!strcmp (opt, "precleaning")) {
3476 precleaning_enabled = TRUE;
3477 continue;
3478 }
3479 if (!strcmp (opt, "no-precleaning")) {
3480 precleaning_enabled = FALSE;
3481 continue;
3482 }
3483
3484 if (!strcmp (opt, "dynamic-nursery")) {
3485 if (sgen_minor_collector.is_split)
3486 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.",
3487 "dynamic-nursery not supported with split-nursery.");
3488 else
3489 dynamic_nursery = TRUE;
3490 continue;
3491 }
3492 if (!strcmp (opt, "no-dynamic-nursery")) {
3493 dynamic_nursery = FALSE;
3494 continue;
3495 }
3496
3497 if (major_collector.handle_gc_param && major_collector.handle_gc_param (opt))
3498 continue;
3499
3500 if (sgen_minor_collector.handle_gc_param && sgen_minor_collector.handle_gc_param (opt))
3501 continue;
3502
3503 if (sgen_client_handle_gc_param (opt))
3504 continue;
3505
3506 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Ignoring.", "Unknown option `%s`.", opt);
3507
3508 if (usage_printed)
3509 continue;
3510
3511 fprintf (stderr, "\n%s must be a comma-delimited list of one or more of the following:\n", MONO_GC_PARAMS_NAME);
3512 fprintf (stderr, " max-heap-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
3513 fprintf (stderr, " soft-heap-limit=n (where N is an integer, possibly with a k, m or a g suffix)\n");
3514 fprintf (stderr, " mode=MODE (where MODE is 'balanced', 'throughput' or 'pause[:N]' and N is maximum pause in milliseconds)\n");
3515 fprintf (stderr, " nursery-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
3516 fprintf (stderr, " major=COLLECTOR (where COLLECTOR is `marksweep', `marksweep-conc', `marksweep-par')\n");
3517 fprintf (stderr, " minor=COLLECTOR (where COLLECTOR is `simple' or `split')\n");
3518 fprintf (stderr, " wbarrier=WBARRIER (where WBARRIER is `remset' or `cardtable')\n");
3519 fprintf (stderr, " [no-]cementing\n");
3520 fprintf (stderr, " [no-]dynamic-nursery\n");
3521 if (major_collector.print_gc_param_usage)
3522 major_collector.print_gc_param_usage ();
3523 if (sgen_minor_collector.print_gc_param_usage)
3524 sgen_minor_collector.print_gc_param_usage ();
3525 sgen_client_print_gc_params_usage ();
3526 fprintf (stderr, " Experimental options:\n");
3527 fprintf (stderr, " save-target-ratio=R (where R must be between %.2f - %.2f).\n", SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO);
3528 fprintf (stderr, " default-allowance-ratio=R (where R must be between %.2f - %.2f).\n", SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, SGEN_MAX_ALLOWANCE_NURSERY_SIZE_RATIO);
3529 fprintf (stderr, "\n");
3530
3531 usage_printed = TRUE;
3532 }
3533 g_strfreev (opts);
3534 }
3535
3536 if (params_opts)
3537 g_free (params_opts);
3538
3539 alloc_nursery (dynamic_nursery, min_nursery_size, max_nursery_size);
3540
3541 sgen_pinning_init ();
3542 sgen_cement_init (cement_enabled);
3543
3544 if ((env = g_getenv (MONO_GC_DEBUG_NAME)) || gc_debug_options) {
3545 debug_opts = g_strdup_printf ("%s,%s", gc_debug_options ? gc_debug_options : "", env ? env : "");
3546 g_free (env);
3547 }
3548
3549 if (debug_opts) {
3550 gboolean usage_printed = FALSE;
3551
3552 opts = g_strsplit (debug_opts, ",", -1);
3553 for (ptr = opts; ptr && *ptr; ptr ++) {
3554 char *opt = *ptr;
3555 if (!strcmp (opt, ""))
3556 continue;
3557 if (opt [0] >= '0' && opt [0] <= '9') {
3558 gc_debug_level = atoi (opt);
3559 opt++;
3560 if (opt [0] == ':')
3561 opt++;
3562 if (opt [0]) {
3563 char *rf = g_strdup_printf ("%s.%d", opt, mono_process_current_pid ());
3564 gc_debug_file = fopen (rf, "wb");
3565 if (!gc_debug_file)
3566 gc_debug_file = stderr;
3567 g_free (rf);
3568 }
3569 } else if (!strcmp (opt, "print-allowance")) {
3570 debug_print_allowance = TRUE;
3571 } else if (!strcmp (opt, "print-pinning")) {
3572 sgen_pin_stats_enable ();
3573 } else if (!strcmp (opt, "verify-before-allocs")) {
3574 verify_before_allocs = 1;
3575 has_per_allocation_action = TRUE;
3576 } else if (g_str_has_prefix (opt, "max-valloc-size=")) {
3577 size_t max_valloc_size;
3578 char *arg = strchr (opt, '=') + 1;
3579 if (*opt && mono_gc_parse_environment_string_extract_number (arg, &max_valloc_size)) {
3580 mono_valloc_set_limit (max_valloc_size);
3581 } else {
3582 sgen_env_var_error (MONO_GC_DEBUG_NAME, NULL, "`max-valloc-size` must be an integer.");
3583 }
3584 continue;
3585 } else if (g_str_has_prefix (opt, "verify-before-allocs=")) {
3586 char *arg = strchr (opt, '=') + 1;
3587 verify_before_allocs = atoi (arg);
3588 has_per_allocation_action = TRUE;
3589 } else if (!strcmp (opt, "collect-before-allocs")) {
3590 collect_before_allocs = 1;
3591 has_per_allocation_action = TRUE;
3592 } else if (g_str_has_prefix (opt, "collect-before-allocs=")) {
3593 char *arg = strchr (opt, '=') + 1;
3594 has_per_allocation_action = TRUE;
3595 collect_before_allocs = atoi (arg);
3596 } else if (!strcmp (opt, "verify-before-collections")) {
3597 whole_heap_check_before_collection = TRUE;
3598 } else if (!strcmp (opt, "check-remset-consistency")) {
3599 remset_consistency_checks = TRUE;
3600 nursery_clear_policy = CLEAR_AT_GC;
3601 } else if (!strcmp (opt, "mod-union-consistency-check")) {
3602 if (!major_collector.is_concurrent) {
3603 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`mod-union-consistency-check` only works with concurrent major collector.");
3604 continue;
3605 }
3606 mod_union_consistency_check = TRUE;
3607 } else if (!strcmp (opt, "check-mark-bits")) {
3608 check_mark_bits_after_major_collection = TRUE;
3609 } else if (!strcmp (opt, "check-nursery-pinned")) {
3610 check_nursery_objects_pinned = TRUE;
3611 } else if (!strcmp (opt, "clear-at-gc")) {
3612 nursery_clear_policy = CLEAR_AT_GC;
3613 } else if (!strcmp (opt, "clear-nursery-at-gc")) {
3614 nursery_clear_policy = CLEAR_AT_GC;
3615 } else if (!strcmp (opt, "clear-at-tlab-creation")) {
3616 nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
3617 } else if (!strcmp (opt, "debug-clear-at-tlab-creation")) {
3618 nursery_clear_policy = CLEAR_AT_TLAB_CREATION_DEBUG;
3619 } else if (!strcmp (opt, "check-scan-starts")) {
3620 do_scan_starts_check = TRUE;
3621 } else if (!strcmp (opt, "verify-nursery-at-minor-gc")) {
3622 do_verify_nursery = TRUE;
3623 } else if (!strcmp (opt, "check-concurrent")) {
3624 if (!major_collector.is_concurrent) {
3625 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`check-concurrent` only works with concurrent major collectors.");
3626 continue;
3627 }
3628 nursery_clear_policy = CLEAR_AT_GC;
3629 do_concurrent_checks = TRUE;
3630 } else if (!strcmp (opt, "dump-nursery-at-minor-gc")) {
3631 do_dump_nursery_content = TRUE;
3632 } else if (!strcmp (opt, "disable-minor")) {
3633 disable_minor_collections = TRUE;
3634 } else if (!strcmp (opt, "disable-major")) {
3635 disable_major_collections = TRUE;
3636 } else if (g_str_has_prefix (opt, "heap-dump=")) {
3637 char *filename = strchr (opt, '=') + 1;
3638 nursery_clear_policy = CLEAR_AT_GC;
3639 sgen_debug_enable_heap_dump (filename);
3640 } else if (g_str_has_prefix (opt, "binary-protocol=")) {
3641 char *filename = strchr (opt, '=') + 1;
3642 char *colon = strrchr (filename, ':');
3643 size_t limit = 0;
3644 if (colon) {
3645 if (!mono_gc_parse_environment_string_extract_number (colon + 1, &limit)) {
3646 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring limit.", "Binary protocol file size limit must be an integer.");
3647 limit = -1;
3648 }
3649 *colon = '\0';
3650 }
3651 binary_protocol_init (filename, (long long)limit);
3652 } else if (!strcmp (opt, "nursery-canaries")) {
3653 do_verify_nursery = TRUE;
3654 enable_nursery_canaries = TRUE;
3655 } else if (!sgen_client_handle_gc_debug (opt)) {
3656 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "Unknown option `%s`.", opt);
3657
3658 if (usage_printed)
3659 continue;
3660
3661 fprintf (stderr, "\n%s must be of the format [<l>[:<filename>]|<option>]+ where <l> is a debug level 0-9.\n", MONO_GC_DEBUG_NAME);
3662 fprintf (stderr, "Valid <option>s are:\n");
3663 fprintf (stderr, " collect-before-allocs[=<n>]\n");
3664 fprintf (stderr, " verify-before-allocs[=<n>]\n");
3665 fprintf (stderr, " max-valloc-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
3666 fprintf (stderr, " check-remset-consistency\n");
3667 fprintf (stderr, " check-mark-bits\n");
3668 fprintf (stderr, " check-nursery-pinned\n");
3669 fprintf (stderr, " verify-before-collections\n");
3670 fprintf (stderr, " verify-nursery-at-minor-gc\n");
3671 fprintf (stderr, " dump-nursery-at-minor-gc\n");
3672 fprintf (stderr, " disable-minor\n");
3673 fprintf (stderr, " disable-major\n");
3674 fprintf (stderr, " check-concurrent\n");
3675 fprintf (stderr, " clear-[nursery-]at-gc\n");
3676 fprintf (stderr, " clear-at-tlab-creation\n");
3677 fprintf (stderr, " debug-clear-at-tlab-creation\n");
3678 fprintf (stderr, " check-scan-starts\n");
3679 fprintf (stderr, " print-allowance\n");
3680 fprintf (stderr, " print-pinning\n");
3681 fprintf (stderr, " heap-dump=<filename>\n");
3682 fprintf (stderr, " binary-protocol=<filename>[:<file-size-limit>]\n");
3683 fprintf (stderr, " nursery-canaries\n");
3684 sgen_client_print_gc_debug_usage ();
3685 fprintf (stderr, "\n");
3686
3687 usage_printed = TRUE;
3688 }
3689 }
3690 g_strfreev (opts);
3691 }
3692
3693 if (debug_opts)
3694 g_free (debug_opts);
3695
3696 if (check_mark_bits_after_major_collection)
3697 nursery_clear_policy = CLEAR_AT_GC;
3698
3699 if (major_collector.post_param_init)
3700 major_collector.post_param_init (&major_collector);
3701
3702 sgen_thread_pool_start ();
3703
3704 sgen_memgov_init (max_heap, soft_limit, debug_print_allowance, allowance_ratio, save_target);
3705
3706 memset (&remset, 0, sizeof (remset));
3707
3708 sgen_card_table_init (&remset);
3709
3710 sgen_register_root (NULL, 0, sgen_make_user_root_descriptor (sgen_mark_normal_gc_handles), ROOT_TYPE_NORMAL, MONO_ROOT_SOURCE_GC_HANDLE, "normal gc handles");
3711
3712 gc_initialized = 1;
3713
3714 sgen_init_bridge ();
3715 }
3716
3717 gboolean
sgen_gc_initialized()3718 sgen_gc_initialized ()
3719 {
3720 return gc_initialized > 0;
3721 }
3722
3723 NurseryClearPolicy
sgen_get_nursery_clear_policy(void)3724 sgen_get_nursery_clear_policy (void)
3725 {
3726 return nursery_clear_policy;
3727 }
3728
3729 void
sgen_gc_lock(void)3730 sgen_gc_lock (void)
3731 {
3732 mono_coop_mutex_lock (&gc_mutex);
3733 }
3734
3735 void
sgen_gc_unlock(void)3736 sgen_gc_unlock (void)
3737 {
3738 mono_coop_mutex_unlock (&gc_mutex);
3739 }
3740
3741 void
sgen_major_collector_iterate_live_block_ranges(sgen_cardtable_block_callback callback)3742 sgen_major_collector_iterate_live_block_ranges (sgen_cardtable_block_callback callback)
3743 {
3744 major_collector.iterate_live_block_ranges (callback);
3745 }
3746
3747 void
sgen_major_collector_iterate_block_ranges(sgen_cardtable_block_callback callback)3748 sgen_major_collector_iterate_block_ranges (sgen_cardtable_block_callback callback)
3749 {
3750 major_collector.iterate_block_ranges (callback);
3751 }
3752
3753 SgenMajorCollector*
sgen_get_major_collector(void)3754 sgen_get_major_collector (void)
3755 {
3756 return &major_collector;
3757 }
3758
3759 SgenMinorCollector*
sgen_get_minor_collector(void)3760 sgen_get_minor_collector (void)
3761 {
3762 return &sgen_minor_collector;
3763 }
3764
3765 SgenRememberedSet*
sgen_get_remset(void)3766 sgen_get_remset (void)
3767 {
3768 return &remset;
3769 }
3770
3771 static void
count_cards(long long * major_total,long long * major_marked,long long * los_total,long long * los_marked)3772 count_cards (long long *major_total, long long *major_marked, long long *los_total, long long *los_marked)
3773 {
3774 sgen_get_major_collector ()->count_cards (major_total, major_marked);
3775 sgen_los_count_cards (los_total, los_marked);
3776 }
3777
3778 static gboolean world_is_stopped = FALSE;
3779
3780 /* LOCKING: assumes the GC lock is held */
3781 void
sgen_stop_world(int generation)3782 sgen_stop_world (int generation)
3783 {
3784 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3785
3786 SGEN_ASSERT (0, !world_is_stopped, "Why are we stopping a stopped world?");
3787
3788 binary_protocol_world_stopping (generation, sgen_timestamp (), (gpointer) (gsize) mono_native_thread_id_get ());
3789
3790 sgen_client_stop_world (generation);
3791
3792 world_is_stopped = TRUE;
3793
3794 if (binary_protocol_is_heavy_enabled ())
3795 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3796 binary_protocol_world_stopped (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3797 }
3798
3799 /* LOCKING: assumes the GC lock is held */
3800 void
sgen_restart_world(int generation)3801 sgen_restart_world (int generation)
3802 {
3803 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3804 gint64 stw_time;
3805
3806 SGEN_ASSERT (0, world_is_stopped, "Why are we restarting a running world?");
3807
3808 if (binary_protocol_is_heavy_enabled ())
3809 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3810 binary_protocol_world_restarting (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3811
3812 world_is_stopped = FALSE;
3813
3814 sgen_client_restart_world (generation, &stw_time);
3815
3816 binary_protocol_world_restarted (generation, sgen_timestamp ());
3817
3818 if (sgen_client_bridge_need_processing ())
3819 sgen_client_bridge_processing_finish (generation);
3820
3821 sgen_memgov_collection_end (generation, stw_time);
3822 }
3823
3824 gboolean
sgen_is_world_stopped(void)3825 sgen_is_world_stopped (void)
3826 {
3827 return world_is_stopped;
3828 }
3829
3830 void
sgen_check_whole_heap_stw(void)3831 sgen_check_whole_heap_stw (void)
3832 {
3833 sgen_stop_world (0);
3834 sgen_clear_nursery_fragments ();
3835 sgen_check_whole_heap (TRUE);
3836 sgen_restart_world (0);
3837 }
3838
3839 gint64
sgen_timestamp(void)3840 sgen_timestamp (void)
3841 {
3842 SGEN_TV_DECLARE (timestamp);
3843 SGEN_TV_GETTIME (timestamp);
3844 return SGEN_TV_ELAPSED (sgen_init_timestamp, timestamp);
3845 }
3846
3847 #endif /* HAVE_SGEN_GC */
3848