3 * Simple generational GC.
6 * Paolo Molaro (lupus@ximian.com)
7 * Rodrigo Kumpera (kumpera@gmail.com)
9 * Copyright 2005-2011 Novell, Inc (http://www.novell.com)
10 * Copyright 2011 Xamarin Inc (http://www.xamarin.com)
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
22 * Licensed under the MIT license. See LICENSE file in the project root for full license information.
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:
28 * 64 KB internal space
30 * We should provide a small memory config with half the sizes
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
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
42 * Always try to keep stack usage to a minimum: no recursive behaviour
43 * and no large stack allocs.
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
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.
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
68 *) we could have a function pointer in MonoClass to implement
69 customized write barriers for value types
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.
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.
84 *) test/fix endianess issues
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.
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.
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.
114 *) experiment with max small object size (very small right now - 2kb,
115 because it's tied to the max freelist size)
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
126 *) memzero the major fragments after restarting the world and optionally a smaller
129 *) investigate having fragment zeroing threads
131 *) separate locks for finalization and other minor stuff to reduce
134 *) try a different copying order to improve memory locality
136 *) a thread abort after a store but before the write barrier will
137 prevent the write barrier from executing
139 *) specialized dynamically generated markers/copiers
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.
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.
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.
162 #define _XOPEN_SOURCE
163 #define _DARWIN_C_SOURCE
169 #ifdef HAVE_PTHREAD_H
172 #ifdef HAVE_PTHREAD_NP_H
173 #include <pthread_np.h>
181 #include "mono/sgen/sgen-gc.h"
182 #include "mono/sgen/sgen-cardtable.h"
183 #include "mono/sgen/sgen-protocol.h"
184 #include "mono/sgen/sgen-memory-governor.h"
185 #include "mono/sgen/sgen-hash-table.h"
186 #include "mono/sgen/sgen-pinning.h"
187 #include "mono/sgen/sgen-workers.h"
188 #include "mono/sgen/sgen-client.h"
189 #include "mono/sgen/sgen-pointer-queue.h"
190 #include "mono/sgen/gc-internal-agnostic.h"
191 #include "mono/utils/mono-proclib.h"
192 #include "mono/utils/mono-memory-model.h"
193 #include "mono/utils/hazard-pointer.h"
195 #include <mono/utils/memcheck.h>
196 #include <mono/utils/mono-mmap-internals.h>
198 #undef pthread_create
200 #undef pthread_detach
203 * ######################################################################
204 * ######## Types and constants used by the GC.
205 * ######################################################################
208 /* 0 means not initialized, 1 is initialized, -1 means in progress */
209 static int gc_initialized = 0;
210 /* If set, check if we need to do something every X allocations */
211 gboolean has_per_allocation_action;
212 /* If set, do a heap check every X allocation */
213 guint32 verify_before_allocs = 0;
214 /* If set, do a minor collection before every X allocation */
215 guint32 collect_before_allocs = 0;
216 /* If set, do a whole heap check before each collection */
217 static gboolean whole_heap_check_before_collection = FALSE;
218 /* If set, do a remset consistency check at various opportunities */
219 static gboolean remset_consistency_checks = FALSE;
220 /* If set, do a mod union consistency check before each finishing collection pause */
221 static gboolean mod_union_consistency_check = FALSE;
222 /* If set, check whether mark bits are consistent after major collections */
223 static gboolean check_mark_bits_after_major_collection = FALSE;
224 /* If set, check that all nursery objects are pinned/not pinned, depending on context */
225 static gboolean check_nursery_objects_pinned = FALSE;
226 /* If set, do a few checks when the concurrent collector is used */
227 static gboolean do_concurrent_checks = FALSE;
228 /* If set, do a plausibility check on the scan_starts before and after
230 static gboolean do_scan_starts_check = FALSE;
232 static gboolean disable_minor_collections = FALSE;
233 static gboolean disable_major_collections = FALSE;
234 static gboolean do_verify_nursery = FALSE;
235 static gboolean do_dump_nursery_content = FALSE;
236 static gboolean enable_nursery_canaries = FALSE;
238 static gboolean precleaning_enabled = TRUE;
240 #ifdef HEAVY_STATISTICS
241 guint64 stat_objects_alloced_degraded = 0;
242 guint64 stat_bytes_alloced_degraded = 0;
244 guint64 stat_copy_object_called_nursery = 0;
245 guint64 stat_objects_copied_nursery = 0;
246 guint64 stat_copy_object_called_major = 0;
247 guint64 stat_objects_copied_major = 0;
249 guint64 stat_scan_object_called_nursery = 0;
250 guint64 stat_scan_object_called_major = 0;
252 guint64 stat_slots_allocated_in_vain;
254 guint64 stat_nursery_copy_object_failed_from_space = 0;
255 guint64 stat_nursery_copy_object_failed_forwarded = 0;
256 guint64 stat_nursery_copy_object_failed_pinned = 0;
257 guint64 stat_nursery_copy_object_failed_to_space = 0;
259 static guint64 stat_wbarrier_add_to_global_remset = 0;
260 static guint64 stat_wbarrier_arrayref_copy = 0;
261 static guint64 stat_wbarrier_generic_store = 0;
262 static guint64 stat_wbarrier_generic_store_atomic = 0;
263 static guint64 stat_wbarrier_set_root = 0;
266 static guint64 stat_pinned_objects = 0;
268 static guint64 time_minor_pre_collection_fragment_clear = 0;
269 static guint64 time_minor_pinning = 0;
270 static guint64 time_minor_scan_remsets = 0;
271 static guint64 time_minor_scan_major_blocks = 0;
272 static guint64 time_minor_scan_los = 0;
273 static guint64 time_minor_scan_pinned = 0;
274 static guint64 time_minor_scan_roots = 0;
275 static guint64 time_minor_finish_gray_stack = 0;
276 static guint64 time_minor_fragment_creation = 0;
278 static guint64 time_major_pre_collection_fragment_clear = 0;
279 static guint64 time_major_pinning = 0;
280 static guint64 time_major_scan_pinned = 0;
281 static guint64 time_major_scan_roots = 0;
282 static guint64 time_major_scan_mod_union = 0;
283 static guint64 time_major_finish_gray_stack = 0;
284 static guint64 time_major_free_bigobjs = 0;
285 static guint64 time_major_los_sweep = 0;
286 static guint64 time_major_sweep = 0;
287 static guint64 time_major_fragment_creation = 0;
289 static guint64 time_max = 0;
291 static int sgen_max_pause_time = SGEN_MAX_PAUSE_TIME;
292 static float sgen_max_pause_margin = SGEN_MAX_PAUSE_MARGIN;
294 static SGEN_TV_DECLARE (time_major_conc_collection_start);
295 static SGEN_TV_DECLARE (time_major_conc_collection_end);
297 int gc_debug_level = 0;
299 static char* gc_params_options;
300 static char* gc_debug_options;
304 mono_gc_flush_info (void)
306 fflush (gc_debug_file);
310 #define TV_DECLARE SGEN_TV_DECLARE
311 #define TV_GETTIME SGEN_TV_GETTIME
312 #define TV_ELAPSED SGEN_TV_ELAPSED
314 static SGEN_TV_DECLARE (sgen_init_timestamp);
316 NurseryClearPolicy nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
318 #define object_is_forwarded SGEN_OBJECT_IS_FORWARDED
319 #define object_is_pinned SGEN_OBJECT_IS_PINNED
320 #define pin_object SGEN_PIN_OBJECT
322 #define ptr_in_nursery sgen_ptr_in_nursery
324 #define LOAD_VTABLE SGEN_LOAD_VTABLE
327 nursery_canaries_enabled (void)
329 return enable_nursery_canaries;
332 #define safe_object_get_size sgen_safe_object_get_size
334 #if defined(HAVE_CONC_GC_AS_DEFAULT)
335 /* Use concurrent major on deskstop platforms */
336 #define DEFAULT_MAJOR_INIT sgen_marksweep_conc_init
337 #define DEFAULT_MAJOR_NAME "marksweep-conc"
339 #define DEFAULT_MAJOR_INIT sgen_marksweep_init
340 #define DEFAULT_MAJOR_NAME "marksweep"
344 * ######################################################################
345 * ######## Global data.
346 * ######################################################################
348 MonoCoopMutex gc_mutex;
350 #define SCAN_START_SIZE SGEN_SCAN_START_SIZE
352 size_t degraded_mode = 0;
354 static mword bytes_pinned_from_failed_allocation = 0;
356 GCMemSection *nursery_section = NULL;
357 static volatile mword lowest_heap_address = ~(mword)0;
358 static volatile mword highest_heap_address = 0;
360 MonoCoopMutex sgen_interruption_mutex;
362 int current_collection_generation = -1;
363 volatile gboolean concurrent_collection_in_progress = FALSE;
365 /* objects that are ready to be finalized */
366 static SgenPointerQueue fin_ready_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
367 static SgenPointerQueue critical_fin_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
369 /* registered roots: the key to the hash is the root start address */
371 * Different kinds of roots are kept separate to speed up pin_from_roots () for example.
373 SgenHashTable roots_hash [ROOT_TYPE_NUM] = {
374 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
375 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
376 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL)
378 static mword roots_size = 0; /* amount of memory in the root set */
380 /* The size of a TLAB */
381 /* The bigger the value, the less often we have to go to the slow path to allocate a new
382 * one, but the more space is wasted by threads not allocating much memory.
384 * FIXME: Make this self-tuning for each thread.
386 guint32 tlab_size = (1024 * 4);
388 #define MAX_SMALL_OBJ_SIZE SGEN_MAX_SMALL_OBJ_SIZE
390 #define ALLOC_ALIGN SGEN_ALLOC_ALIGN
392 #define ALIGN_UP SGEN_ALIGN_UP
394 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
395 MonoNativeThreadId main_gc_thread = NULL;
398 /*Object was pinned during the current collection*/
399 static mword objects_pinned;
402 * ######################################################################
403 * ######## Macros and function declarations.
404 * ######################################################################
407 /* forward declarations */
408 static void scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx);
410 static void pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx);
411 static void finish_gray_stack (int generation, ScanCopyContext ctx);
414 SgenMajorCollector major_collector;
415 SgenMinorCollector sgen_minor_collector;
417 static SgenRememberedSet remset;
420 * The gray queue a worker job must use. If we're not parallel or
421 * concurrent, we use the main gray queue.
423 static SgenGrayQueue*
424 sgen_workers_get_job_gray_queue (WorkerData *worker_data, SgenGrayQueue *default_gray_queue)
427 return &worker_data->private_gray_queue;
428 SGEN_ASSERT (0, default_gray_queue, "Why don't we have a default gray queue when we're not running in a worker thread?");
429 return default_gray_queue;
433 gray_queue_redirect (SgenGrayQueue *queue)
435 sgen_workers_take_from_queue (queue);
439 sgen_scan_area_with_callback (char *start, char *end, IterateObjectCallbackFunc callback, void *data, gboolean allow_flags, gboolean fail_on_canaries)
441 while (start < end) {
445 if (!*(void**)start) {
446 start += sizeof (void*); /* should be ALLOC_ALIGN, really */
451 if (!(obj = (char *)SGEN_OBJECT_IS_FORWARDED (start)))
457 if (!sgen_client_object_is_array_fill ((GCObject*)obj)) {
458 CHECK_CANARY_FOR_OBJECT ((GCObject*)obj, fail_on_canaries);
459 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
460 callback ((GCObject*)obj, size, data);
461 CANARIFY_SIZE (size);
463 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
471 * sgen_add_to_global_remset:
473 * The global remset contains locations which point into newspace after
474 * a minor collection. This can happen if the objects they point to are pinned.
476 * LOCKING: If called from a parallel collector, the global remset
477 * lock must be held. For serial collectors that is not necessary.
480 sgen_add_to_global_remset (gpointer ptr, GCObject *obj)
482 SGEN_ASSERT (5, sgen_ptr_in_nursery (obj), "Target pointer of global remset must be in the nursery");
484 HEAVY_STAT (++stat_wbarrier_add_to_global_remset);
486 if (!major_collector.is_concurrent) {
487 SGEN_ASSERT (5, current_collection_generation != -1, "Global remsets can only be added during collections");
489 if (current_collection_generation == -1)
490 SGEN_ASSERT (5, sgen_concurrent_collection_in_progress (), "Global remsets outside of collection pauses can only be added by the concurrent collector");
493 if (!object_is_pinned (obj))
494 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");
495 else if (sgen_cement_lookup_or_register (obj))
498 remset.record_pointer (ptr);
500 sgen_pin_stats_register_global_remset (obj);
502 SGEN_LOG (8, "Adding global remset for %p", ptr);
503 binary_protocol_global_remset (ptr, obj, (gpointer)SGEN_LOAD_VTABLE (obj));
507 * sgen_drain_gray_stack:
509 * Scan objects in the gray stack until the stack is empty. This should be called
510 * frequently after each object is copied, to achieve better locality and cache
515 sgen_drain_gray_stack (ScanCopyContext ctx)
517 SGEN_ASSERT (0, ctx.ops->drain_gray_stack, "Why do we have a scan/copy context with a missing drain gray stack function?");
519 return ctx.ops->drain_gray_stack (ctx.queue);
523 * Addresses in the pin queue are already sorted. This function finds
524 * the object header for each address and pins the object. The
525 * addresses must be inside the nursery section. The (start of the)
526 * address array is overwritten with the addresses of the actually
527 * pinned objects. Return the number of pinned objects.
530 pin_objects_from_nursery_pin_queue (gboolean do_scan_objects, ScanCopyContext ctx)
532 GCMemSection *section = nursery_section;
533 void **start = sgen_pinning_get_entry (section->pin_queue_first_entry);
534 void **end = sgen_pinning_get_entry (section->pin_queue_last_entry);
535 void *start_nursery = section->data;
536 void *end_nursery = section->end_data;
541 void *pinning_front = start_nursery;
543 void **definitely_pinned = start;
544 ScanObjectFunc scan_func = ctx.ops->scan_object;
545 SgenGrayQueue *queue = ctx.queue;
547 sgen_nursery_allocator_prepare_for_pinning ();
549 while (start < end) {
550 GCObject *obj_to_pin = NULL;
551 size_t obj_to_pin_size = 0;
556 SGEN_ASSERT (0, addr >= start_nursery && addr < end_nursery, "Potential pinning address out of range");
557 SGEN_ASSERT (0, addr >= last, "Pin queue not sorted");
564 SGEN_LOG (5, "Considering pinning addr %p", addr);
565 /* We've already processed everything up to pinning_front. */
566 if (addr < pinning_front) {
572 * Find the closest scan start <= addr. We might search backward in the
573 * scan_starts array because entries might be NULL. In the worst case we
574 * start at start_nursery.
576 idx = ((char*)addr - (char*)section->data) / SCAN_START_SIZE;
577 SGEN_ASSERT (0, idx < section->num_scan_start, "Scan start index out of range");
578 search_start = (void*)section->scan_starts [idx];
579 if (!search_start || search_start > addr) {
582 search_start = section->scan_starts [idx];
583 if (search_start && search_start <= addr)
586 if (!search_start || search_start > addr)
587 search_start = start_nursery;
591 * If the pinning front is closer than the scan start we found, start
592 * searching at the front.
594 if (search_start < pinning_front)
595 search_start = pinning_front;
598 * Now addr should be in an object a short distance from search_start.
600 * search_start must point to zeroed mem or point to an object.
603 size_t obj_size, canarified_obj_size;
606 if (!*(void**)search_start) {
607 search_start = (void*)ALIGN_UP ((mword)search_start + sizeof (gpointer));
608 /* The loop condition makes sure we don't overrun addr. */
612 canarified_obj_size = obj_size = ALIGN_UP (safe_object_get_size ((GCObject*)search_start));
615 * Filler arrays are marked by an invalid sync word. We don't
616 * consider them for pinning. They are not delimited by canaries,
619 if (!sgen_client_object_is_array_fill ((GCObject*)search_start)) {
620 CHECK_CANARY_FOR_OBJECT (search_start, TRUE);
621 CANARIFY_SIZE (canarified_obj_size);
623 if (addr >= search_start && (char*)addr < (char*)search_start + obj_size) {
624 /* This is the object we're looking for. */
625 obj_to_pin = (GCObject*)search_start;
626 obj_to_pin_size = canarified_obj_size;
631 /* Skip to the next object */
632 search_start = (void*)((char*)search_start + canarified_obj_size);
633 } while (search_start <= addr);
635 /* We've searched past the address we were looking for. */
637 pinning_front = search_start;
638 goto next_pin_queue_entry;
642 * We've found an object to pin. It might still be a dummy array, but we
643 * can advance the pinning front in any case.
645 pinning_front = (char*)obj_to_pin + obj_to_pin_size;
648 * If this is a dummy array marking the beginning of a nursery
649 * fragment, we don't pin it.
651 if (sgen_client_object_is_array_fill (obj_to_pin))
652 goto next_pin_queue_entry;
655 * Finally - pin the object!
657 desc = sgen_obj_get_descriptor_safe (obj_to_pin);
658 if (do_scan_objects) {
659 scan_func (obj_to_pin, desc, queue);
661 SGEN_LOG (4, "Pinned object %p, vtable %p (%s), count %d\n",
662 obj_to_pin, *(void**)obj_to_pin, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj_to_pin)), count);
663 binary_protocol_pin (obj_to_pin,
664 (gpointer)LOAD_VTABLE (obj_to_pin),
665 safe_object_get_size (obj_to_pin));
667 pin_object (obj_to_pin);
668 GRAY_OBJECT_ENQUEUE_SERIAL (queue, obj_to_pin, desc);
669 sgen_pin_stats_register_object (obj_to_pin, GENERATION_NURSERY);
670 definitely_pinned [count] = obj_to_pin;
673 if (concurrent_collection_in_progress)
674 sgen_pinning_register_pinned_in_nursery (obj_to_pin);
676 next_pin_queue_entry:
680 sgen_client_nursery_objects_pinned (definitely_pinned, count);
681 stat_pinned_objects += count;
686 pin_objects_in_nursery (gboolean do_scan_objects, ScanCopyContext ctx)
690 if (nursery_section->pin_queue_first_entry == nursery_section->pin_queue_last_entry)
693 reduced_to = pin_objects_from_nursery_pin_queue (do_scan_objects, ctx);
694 nursery_section->pin_queue_last_entry = nursery_section->pin_queue_first_entry + reduced_to;
698 * This function is only ever called (via `collector_pin_object()` in `sgen-copy-object.h`)
699 * when we can't promote an object because we're out of memory.
702 sgen_pin_object (GCObject *object, SgenGrayQueue *queue)
704 SGEN_ASSERT (0, sgen_ptr_in_nursery (object), "We're only supposed to use this for pinning nursery objects when out of memory.");
707 * All pinned objects are assumed to have been staged, so we need to stage as well.
708 * Also, the count of staged objects shows that "late pinning" happened.
710 sgen_pin_stage_ptr (object);
712 SGEN_PIN_OBJECT (object);
713 binary_protocol_pin (object, (gpointer)LOAD_VTABLE (object), safe_object_get_size (object));
716 sgen_pin_stats_register_object (object, GENERATION_NURSERY);
718 GRAY_OBJECT_ENQUEUE_SERIAL (queue, object, sgen_obj_get_descriptor_safe (object));
721 /* Sort the addresses in array in increasing order.
722 * Done using a by-the book heap sort. Which has decent and stable performance, is pretty cache efficient.
725 sgen_sort_addresses (void **array, size_t size)
730 for (i = 1; i < size; ++i) {
733 size_t parent = (child - 1) / 2;
735 if (array [parent] >= array [child])
738 tmp = array [parent];
739 array [parent] = array [child];
746 for (i = size - 1; i > 0; --i) {
749 array [i] = array [0];
755 while (root * 2 + 1 <= end) {
756 size_t child = root * 2 + 1;
758 if (child < end && array [child] < array [child + 1])
760 if (array [root] >= array [child])
764 array [root] = array [child];
773 * Scan the memory between start and end and queue values which could be pointers
774 * to the area between start_nursery and end_nursery for later consideration.
775 * Typically used for thread stacks.
778 sgen_conservatively_pin_objects_from (void **start, void **end, void *start_nursery, void *end_nursery, int pin_type)
782 SGEN_ASSERT (0, ((mword)start & (SIZEOF_VOID_P - 1)) == 0, "Why are we scanning for references in unaligned memory ?");
784 #if defined(VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE) && !defined(_WIN64)
785 VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE (start, (char*)end - (char*)start);
788 while (start < end) {
790 * *start can point to the middle of an object
791 * note: should we handle pointing at the end of an object?
792 * pinning in C# code disallows pointing at the end of an object
793 * but there is some small chance that an optimizing C compiler
794 * may keep the only reference to an object by pointing
795 * at the end of it. We ignore this small chance for now.
796 * Pointers to the end of an object are indistinguishable
797 * from pointers to the start of the next object in memory
798 * so if we allow that we'd need to pin two objects...
799 * We queue the pointer in an array, the
800 * array will then be sorted and uniqued. This way
801 * we can coalesce several pinning pointers and it should
802 * be faster since we'd do a memory scan with increasing
803 * addresses. Note: we can align the address to the allocation
804 * alignment, so the unique process is more effective.
806 mword addr = (mword)*start;
807 addr &= ~(ALLOC_ALIGN - 1);
808 if (addr >= (mword)start_nursery && addr < (mword)end_nursery) {
809 SGEN_LOG (6, "Pinning address %p from %p", (void*)addr, start);
810 sgen_pin_stage_ptr ((void*)addr);
811 binary_protocol_pin_stage (start, (void*)addr);
812 sgen_pin_stats_register_address ((char*)addr, pin_type);
818 SGEN_LOG (7, "found %d potential pinned heap pointers", count);
822 * The first thing we do in a collection is to identify pinned objects.
823 * This function considers all the areas of memory that need to be
824 * conservatively scanned.
827 pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx)
831 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);
832 /* objects pinned from the API are inside these roots */
833 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_PINNED], void **, start_root, RootRecord *, root) {
834 SGEN_LOG (6, "Pinned roots %p-%p", start_root, root->end_root);
835 sgen_conservatively_pin_objects_from (start_root, (void**)root->end_root, start_nursery, end_nursery, PIN_TYPE_OTHER);
836 } SGEN_HASH_TABLE_FOREACH_END;
837 /* now deal with the thread stacks
838 * in the future we should be able to conservatively scan only:
839 * *) the cpu registers
840 * *) the unmanaged stack frames
841 * *) the _last_ managed stack frame
842 * *) pointers slots in managed frames
844 sgen_client_scan_thread_data (start_nursery, end_nursery, FALSE, ctx);
848 single_arg_user_copy_or_mark (GCObject **obj, void *gc_data)
850 ScanCopyContext *ctx = (ScanCopyContext *)gc_data;
851 ctx->ops->copy_or_mark_object (obj, ctx->queue);
855 * The memory area from start_root to end_root contains pointers to objects.
856 * Their position is precisely described by @desc (this means that the pointer
857 * can be either NULL or the pointer to the start of an object).
858 * This functions copies them to to_space updates them.
860 * This function is not thread-safe!
863 precisely_scan_objects_from (void** start_root, void** end_root, char* n_start, char *n_end, SgenDescriptor desc, ScanCopyContext ctx)
865 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
866 ScanPtrFieldFunc scan_field_func = ctx.ops->scan_ptr_field;
867 SgenGrayQueue *queue = ctx.queue;
869 switch (desc & ROOT_DESC_TYPE_MASK) {
870 case ROOT_DESC_BITMAP:
871 desc >>= ROOT_DESC_TYPE_SHIFT;
873 if ((desc & 1) && *start_root) {
874 copy_func ((GCObject**)start_root, queue);
875 SGEN_LOG (9, "Overwrote root at %p with %p", start_root, *start_root);
881 case ROOT_DESC_COMPLEX: {
882 gsize *bitmap_data = (gsize *)sgen_get_complex_descriptor_bitmap (desc);
883 gsize bwords = (*bitmap_data) - 1;
884 void **start_run = start_root;
886 while (bwords-- > 0) {
887 gsize bmap = *bitmap_data++;
888 void **objptr = start_run;
890 if ((bmap & 1) && *objptr) {
891 copy_func ((GCObject**)objptr, queue);
892 SGEN_LOG (9, "Overwrote root at %p with %p", objptr, *objptr);
897 start_run += GC_BITS_PER_WORD;
901 case ROOT_DESC_VECTOR: {
904 for (p = start_root; p < end_root; p++) {
906 scan_field_func (NULL, (GCObject**)p, queue);
910 case ROOT_DESC_USER: {
911 SgenUserRootMarkFunc marker = sgen_get_user_descriptor_func (desc);
912 marker (start_root, single_arg_user_copy_or_mark, &ctx);
915 case ROOT_DESC_RUN_LEN:
916 g_assert_not_reached ();
918 g_assert_not_reached ();
923 reset_heap_boundaries (void)
925 lowest_heap_address = ~(mword)0;
926 highest_heap_address = 0;
930 sgen_update_heap_boundaries (mword low, mword high)
935 old = lowest_heap_address;
938 } while (SGEN_CAS_PTR ((gpointer*)&lowest_heap_address, (gpointer)low, (gpointer)old) != (gpointer)old);
941 old = highest_heap_address;
944 } while (SGEN_CAS_PTR ((gpointer*)&highest_heap_address, (gpointer)high, (gpointer)old) != (gpointer)old);
948 * Allocate and setup the data structures needed to be able to allocate objects
949 * in the nursery. The nursery is stored in nursery_section.
952 alloc_nursery (gboolean dynamic, size_t min_size, size_t max_size)
959 min_size = SGEN_DEFAULT_NURSERY_MIN_SIZE;
961 max_size = SGEN_DEFAULT_NURSERY_MAX_SIZE;
963 SGEN_ASSERT (0, min_size == max_size, "We can't have nursery ranges for static configuration.");
965 min_size = max_size = SGEN_DEFAULT_NURSERY_SIZE;
968 SGEN_ASSERT (0, !nursery_section, "Why are we allocating the nursery twice?");
969 SGEN_LOG (2, "Allocating nursery size: %zu, initial %zu", max_size, min_size);
971 /* FIXME: handle OOM */
972 nursery_section = (GCMemSection *)sgen_alloc_internal (INTERNAL_MEM_SECTION);
974 /* If there isn't enough space even for the nursery we should simply abort. */
975 g_assert (sgen_memgov_try_alloc_space (max_size, SPACE_NURSERY));
978 * The nursery section range represents the memory section where objects
979 * can be found. This is used when iterating for objects in the nursery,
980 * pinning etc. sgen_nursery_max_size represents the total allocated space
981 * for the nursery. sgen_nursery_size represents the current size of the
982 * nursery and it is used for allocation limits, heuristics etc. The
983 * nursery section is not always identical to the current nursery size
984 * because it can contain pinned objects from when the nursery was larger.
986 * sgen_nursery_size <= nursery_section size <= sgen_nursery_max_size
988 data = (char *)major_collector.alloc_heap (max_size, max_size);
989 sgen_update_heap_boundaries ((mword)data, (mword)(data + max_size));
990 nursery_section->data = data;
991 nursery_section->end_data = data + min_size;
992 scan_starts = (max_size + SCAN_START_SIZE - 1) / SCAN_START_SIZE;
993 nursery_section->scan_starts = (char **)sgen_alloc_internal_dynamic (sizeof (char*) * scan_starts, INTERNAL_MEM_SCAN_STARTS, TRUE);
994 nursery_section->num_scan_start = scan_starts;
996 sgen_nursery_allocator_set_nursery_bounds (data, min_size, max_size);
1000 mono_gc_get_logfile (void)
1002 return gc_debug_file;
1006 mono_gc_params_set (const char* options)
1008 if (gc_params_options)
1009 g_free (gc_params_options);
1011 gc_params_options = g_strdup (options);
1015 mono_gc_debug_set (const char* options)
1017 if (gc_debug_options)
1018 g_free (gc_debug_options);
1020 gc_debug_options = g_strdup (options);
1024 scan_finalizer_entries (SgenPointerQueue *fin_queue, ScanCopyContext ctx)
1026 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
1027 SgenGrayQueue *queue = ctx.queue;
1030 for (i = 0; i < fin_queue->next_slot; ++i) {
1031 GCObject *obj = (GCObject *)fin_queue->data [i];
1034 SGEN_LOG (5, "Scan of fin ready object: %p (%s)\n", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
1035 copy_func ((GCObject**)&fin_queue->data [i], queue);
1040 generation_name (int generation)
1042 switch (generation) {
1043 case GENERATION_NURSERY: return "nursery";
1044 case GENERATION_OLD: return "old";
1045 default: g_assert_not_reached ();
1050 sgen_generation_name (int generation)
1052 return generation_name (generation);
1056 finish_gray_stack (int generation, ScanCopyContext ctx)
1060 int done_with_ephemerons, ephemeron_rounds = 0;
1061 char *start_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_start () : NULL;
1062 char *end_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_end () : (char*)-1;
1063 SgenGrayQueue *queue = ctx.queue;
1065 binary_protocol_finish_gray_stack_start (sgen_timestamp (), generation);
1067 * We copied all the reachable objects. Now it's the time to copy
1068 * the objects that were not referenced by the roots, but by the copied objects.
1069 * we built a stack of objects pointed to by gray_start: they are
1070 * additional roots and we may add more items as we go.
1071 * We loop until gray_start == gray_objects which means no more objects have
1072 * been added. Note this is iterative: no recursion is involved.
1073 * We need to walk the LO list as well in search of marked big objects
1074 * (use a flag since this is needed only on major collections). We need to loop
1075 * here as well, so keep a counter of marked LO (increasing it in copy_object).
1076 * To achieve better cache locality and cache usage, we drain the gray stack
1077 * frequently, after each object is copied, and just finish the work here.
1079 sgen_drain_gray_stack (ctx);
1081 SGEN_LOG (2, "%s generation done", generation_name (generation));
1084 Reset bridge data, we might have lingering data from a previous collection if this is a major
1085 collection trigged by minor overflow.
1087 We must reset the gathered bridges since their original block might be evacuated due to major
1088 fragmentation in the meanwhile and the bridge code should not have to deal with that.
1090 if (sgen_client_bridge_need_processing ())
1091 sgen_client_bridge_reset_data ();
1094 * Mark all strong toggleref objects. This must be done before we walk ephemerons or finalizers
1095 * to ensure they see the full set of live objects.
1097 sgen_client_mark_togglerefs (start_addr, end_addr, ctx);
1100 * Walk the ephemeron tables marking all values with reachable keys. This must be completely done
1101 * before processing finalizable objects and non-tracking weak links to avoid finalizing/clearing
1102 * objects that are in fact reachable.
1104 done_with_ephemerons = 0;
1106 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1107 sgen_drain_gray_stack (ctx);
1109 } while (!done_with_ephemerons);
1111 if (sgen_client_bridge_need_processing ()) {
1112 /*Make sure the gray stack is empty before we process bridge objects so we get liveness right*/
1113 sgen_drain_gray_stack (ctx);
1114 sgen_collect_bridge_objects (generation, ctx);
1115 if (generation == GENERATION_OLD)
1116 sgen_collect_bridge_objects (GENERATION_NURSERY, ctx);
1119 Do the first bridge step here, as the collector liveness state will become useless after that.
1121 An important optimization is to only proccess the possibly dead part of the object graph and skip
1122 over all live objects as we transitively know everything they point must be alive too.
1124 The above invariant is completely wrong if we let the gray queue be drained and mark/copy everything.
1126 This has the unfortunate side effect of making overflow collections perform the first step twice, but
1127 given we now have heuristics that perform major GC in anticipation of minor overflows this should not
1130 sgen_client_bridge_processing_stw_step ();
1134 Make sure we drain the gray stack before processing disappearing links and finalizers.
1135 If we don't make sure it is empty we might wrongly see a live object as dead.
1137 sgen_drain_gray_stack (ctx);
1140 We must clear weak links that don't track resurrection before processing object ready for
1141 finalization so they can be cleared before that.
1143 sgen_null_link_in_range (generation, ctx, FALSE);
1144 if (generation == GENERATION_OLD)
1145 sgen_null_link_in_range (GENERATION_NURSERY, ctx, FALSE);
1148 /* walk the finalization queue and move also the objects that need to be
1149 * finalized: use the finalized objects as new roots so the objects they depend
1150 * on are also not reclaimed. As with the roots above, only objects in the nursery
1151 * are marked/copied.
1153 sgen_finalize_in_range (generation, ctx);
1154 if (generation == GENERATION_OLD)
1155 sgen_finalize_in_range (GENERATION_NURSERY, ctx);
1156 /* drain the new stack that might have been created */
1157 SGEN_LOG (6, "Precise scan of gray area post fin");
1158 sgen_drain_gray_stack (ctx);
1161 * This must be done again after processing finalizable objects since CWL slots are cleared only after the key is finalized.
1163 done_with_ephemerons = 0;
1165 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1166 sgen_drain_gray_stack (ctx);
1168 } while (!done_with_ephemerons);
1170 sgen_client_clear_unreachable_ephemerons (ctx);
1173 * We clear togglerefs only after all possible chances of revival are done.
1174 * This is semantically more inline with what users expect and it allows for
1175 * user finalizers to correctly interact with TR objects.
1177 sgen_client_clear_togglerefs (start_addr, end_addr, ctx);
1180 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);
1183 * handle disappearing links
1184 * Note we do this after checking the finalization queue because if an object
1185 * survives (at least long enough to be finalized) we don't clear the link.
1186 * This also deals with a possible issue with the monitor reclamation: with the Boehm
1187 * GC a finalized object my lose the monitor because it is cleared before the finalizer is
1190 g_assert (sgen_gray_object_queue_is_empty (queue));
1192 sgen_null_link_in_range (generation, ctx, TRUE);
1193 if (generation == GENERATION_OLD)
1194 sgen_null_link_in_range (GENERATION_NURSERY, ctx, TRUE);
1195 if (sgen_gray_object_queue_is_empty (queue))
1197 sgen_drain_gray_stack (ctx);
1200 g_assert (sgen_gray_object_queue_is_empty (queue));
1202 binary_protocol_finish_gray_stack_end (sgen_timestamp (), generation);
1206 sgen_check_section_scan_starts (GCMemSection *section)
1209 for (i = 0; i < section->num_scan_start; ++i) {
1210 if (section->scan_starts [i]) {
1211 mword size = safe_object_get_size ((GCObject*) section->scan_starts [i]);
1212 SGEN_ASSERT (0, size >= SGEN_CLIENT_MINIMUM_OBJECT_SIZE && size <= MAX_SMALL_OBJ_SIZE, "Weird object size at scan starts.");
1218 check_scan_starts (void)
1220 if (!do_scan_starts_check)
1222 sgen_check_section_scan_starts (nursery_section);
1223 major_collector.check_scan_starts ();
1227 scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx)
1231 SGEN_HASH_TABLE_FOREACH (&roots_hash [root_type], void **, start_root, RootRecord *, root) {
1232 SGEN_LOG (6, "Precise root scan %p-%p (desc: %p)", start_root, root->end_root, (void*)root->root_desc);
1233 precisely_scan_objects_from (start_root, (void**)root->end_root, addr_start, addr_end, root->root_desc, ctx);
1234 } SGEN_HASH_TABLE_FOREACH_END;
1240 static gboolean inited = FALSE;
1245 mono_counters_register ("Collection max time", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME | MONO_COUNTER_MONOTONIC, &time_max);
1247 mono_counters_register ("Minor fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pre_collection_fragment_clear);
1248 mono_counters_register ("Minor pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pinning);
1249 mono_counters_register ("Minor scan remembered set", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_remsets);
1250 mono_counters_register ("Minor scan major blocks", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_major_blocks);
1251 mono_counters_register ("Minor scan los", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_los);
1252 mono_counters_register ("Minor scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_pinned);
1253 mono_counters_register ("Minor scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_roots);
1254 mono_counters_register ("Minor fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_fragment_creation);
1256 mono_counters_register ("Major fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pre_collection_fragment_clear);
1257 mono_counters_register ("Major pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pinning);
1258 mono_counters_register ("Major scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_pinned);
1259 mono_counters_register ("Major scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_roots);
1260 mono_counters_register ("Major scan mod union", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_mod_union);
1261 mono_counters_register ("Major finish gray stack", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_finish_gray_stack);
1262 mono_counters_register ("Major free big objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_free_bigobjs);
1263 mono_counters_register ("Major LOS sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_los_sweep);
1264 mono_counters_register ("Major sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_sweep);
1265 mono_counters_register ("Major fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_fragment_creation);
1267 mono_counters_register ("Number of pinned objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_pinned_objects);
1269 #ifdef HEAVY_STATISTICS
1270 mono_counters_register ("WBarrier remember pointer", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_add_to_global_remset);
1271 mono_counters_register ("WBarrier arrayref copy", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_arrayref_copy);
1272 mono_counters_register ("WBarrier generic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store);
1273 mono_counters_register ("WBarrier generic atomic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store_atomic);
1274 mono_counters_register ("WBarrier set root", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_set_root);
1276 mono_counters_register ("# objects allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_alloced_degraded);
1277 mono_counters_register ("bytes allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_bytes_alloced_degraded);
1279 mono_counters_register ("# copy_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_nursery);
1280 mono_counters_register ("# objects copied (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_nursery);
1281 mono_counters_register ("# copy_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_major);
1282 mono_counters_register ("# objects copied (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_major);
1284 mono_counters_register ("# scan_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_nursery);
1285 mono_counters_register ("# scan_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_major);
1287 mono_counters_register ("Slots allocated in vain", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_slots_allocated_in_vain);
1289 mono_counters_register ("# nursery copy_object() failed from space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_from_space);
1290 mono_counters_register ("# nursery copy_object() failed forwarded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_forwarded);
1291 mono_counters_register ("# nursery copy_object() failed pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_pinned);
1292 mono_counters_register ("# nursery copy_object() failed to space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_to_space);
1294 sgen_nursery_allocator_init_heavy_stats ();
1302 reset_pinned_from_failed_allocation (void)
1304 bytes_pinned_from_failed_allocation = 0;
1308 sgen_set_pinned_from_failed_allocation (mword objsize)
1310 bytes_pinned_from_failed_allocation += objsize;
1314 sgen_collection_is_concurrent (void)
1316 switch (current_collection_generation) {
1317 case GENERATION_NURSERY:
1319 case GENERATION_OLD:
1320 return concurrent_collection_in_progress;
1322 g_error ("Invalid current generation %d", current_collection_generation);
1328 sgen_concurrent_collection_in_progress (void)
1330 return concurrent_collection_in_progress;
1334 SgenThreadPoolJob job;
1335 SgenObjectOperations *ops;
1336 SgenGrayQueue *gc_thread_gray_queue;
1344 static ScanCopyContext
1345 scan_copy_context_for_scan_job (void *worker_data_untyped, ScanJob *job)
1347 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1349 return CONTEXT_FROM_OBJECT_OPERATIONS (job->ops, sgen_workers_get_job_gray_queue (worker_data, job->gc_thread_gray_queue));
1357 } ScanFromRegisteredRootsJob;
1360 job_scan_from_registered_roots (void *worker_data_untyped, SgenThreadPoolJob *job)
1362 ScanFromRegisteredRootsJob *job_data = (ScanFromRegisteredRootsJob*)job;
1363 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1365 scan_from_registered_roots (job_data->heap_start, job_data->heap_end, job_data->root_type, ctx);
1372 } ScanThreadDataJob;
1375 job_scan_thread_data (void *worker_data_untyped, SgenThreadPoolJob *job)
1377 ScanThreadDataJob *job_data = (ScanThreadDataJob*)job;
1378 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1380 sgen_client_scan_thread_data (job_data->heap_start, job_data->heap_end, TRUE, ctx);
1385 SgenPointerQueue *queue;
1386 } ScanFinalizerEntriesJob;
1389 job_scan_finalizer_entries (void *worker_data_untyped, SgenThreadPoolJob *job)
1391 ScanFinalizerEntriesJob *job_data = (ScanFinalizerEntriesJob*)job;
1392 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1394 scan_finalizer_entries (job_data->queue, ctx);
1398 job_scan_wbroots (void *worker_data_untyped, SgenThreadPoolJob *job)
1400 ScanJob *job_data = (ScanJob*)job;
1401 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, job_data);
1403 sgen_wbroots_scan_card_table (ctx);
1407 job_scan_major_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1409 SGEN_TV_DECLARE (atv);
1410 SGEN_TV_DECLARE (btv);
1411 ParallelScanJob *job_data = (ParallelScanJob*)job;
1412 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1414 SGEN_TV_GETTIME (atv);
1415 major_collector.scan_card_table (CARDTABLE_SCAN_GLOBAL, ctx, job_data->job_index, sgen_workers_get_job_split_count ());
1416 SGEN_TV_GETTIME (btv);
1417 time_minor_scan_major_blocks += SGEN_TV_ELAPSED (atv, btv);
1421 job_scan_los_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1423 SGEN_TV_DECLARE (atv);
1424 SGEN_TV_DECLARE (btv);
1425 ParallelScanJob *job_data = (ParallelScanJob*)job;
1426 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1428 SGEN_TV_GETTIME (atv);
1429 sgen_los_scan_card_table (CARDTABLE_SCAN_GLOBAL, ctx, job_data->job_index, sgen_workers_get_job_split_count ());
1430 SGEN_TV_GETTIME (btv);
1431 time_minor_scan_los += SGEN_TV_ELAPSED (atv, btv);
1435 job_scan_major_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1437 ParallelScanJob *job_data = (ParallelScanJob*)job;
1438 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1440 g_assert (concurrent_collection_in_progress);
1441 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx, job_data->job_index, sgen_workers_get_job_split_count ());
1445 job_scan_los_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1447 ParallelScanJob *job_data = (ParallelScanJob*)job;
1448 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1450 g_assert (concurrent_collection_in_progress);
1451 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx, job_data->job_index, sgen_workers_get_job_split_count ());
1455 job_major_mod_union_preclean (void *worker_data_untyped, SgenThreadPoolJob *job)
1457 ParallelScanJob *job_data = (ParallelScanJob*)job;
1458 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1460 g_assert (concurrent_collection_in_progress);
1462 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx, job_data->job_index, sgen_workers_get_job_split_count ());
1466 job_los_mod_union_preclean (void *worker_data_untyped, SgenThreadPoolJob *job)
1468 ParallelScanJob *job_data = (ParallelScanJob*)job;
1469 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1471 g_assert (concurrent_collection_in_progress);
1473 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx, job_data->job_index, sgen_workers_get_job_split_count ());
1477 job_scan_last_pinned (void *worker_data_untyped, SgenThreadPoolJob *job)
1479 ScanJob *job_data = (ScanJob*)job;
1480 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, job_data);
1482 g_assert (concurrent_collection_in_progress);
1484 sgen_scan_pin_queue_objects (ctx);
1488 workers_finish_callback (void)
1490 ParallelScanJob *psj;
1492 int split_count = sgen_workers_get_job_split_count ();
1494 /* Mod union preclean jobs */
1495 for (i = 0; i < split_count; i++) {
1496 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("preclean major mod union cardtable", job_major_mod_union_preclean, sizeof (ParallelScanJob));
1497 psj->scan_job.ops = sgen_workers_get_idle_func_object_ops ();
1498 psj->scan_job.gc_thread_gray_queue = NULL;
1500 sgen_workers_enqueue_job (&psj->scan_job.job, TRUE);
1503 for (i = 0; i < split_count; i++) {
1504 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("preclean los mod union cardtable", job_los_mod_union_preclean, sizeof (ParallelScanJob));
1505 psj->scan_job.ops = sgen_workers_get_idle_func_object_ops ();
1506 psj->scan_job.gc_thread_gray_queue = NULL;
1508 sgen_workers_enqueue_job (&psj->scan_job.job, TRUE);
1511 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan last pinned", job_scan_last_pinned, sizeof (ScanJob));
1512 sj->ops = sgen_workers_get_idle_func_object_ops ();
1513 sj->gc_thread_gray_queue = NULL;
1514 sgen_workers_enqueue_job (&sj->job, TRUE);
1518 init_gray_queue (SgenGrayQueue *gc_thread_gray_queue, gboolean use_workers)
1521 sgen_workers_init_distribute_gray_queue ();
1522 sgen_gray_object_queue_init (gc_thread_gray_queue, NULL, TRUE);
1526 enqueue_scan_remembered_set_jobs (SgenGrayQueue *gc_thread_gray_queue, SgenObjectOperations *ops, gboolean enqueue)
1528 int i, split_count = sgen_workers_get_job_split_count ();
1531 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan wbroots", job_scan_wbroots, sizeof (ScanJob));
1533 sj->gc_thread_gray_queue = gc_thread_gray_queue;
1534 sgen_workers_enqueue_job (&sj->job, enqueue);
1536 for (i = 0; i < split_count; i++) {
1537 ParallelScanJob *psj;
1539 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("scan major remsets", job_scan_major_card_table, sizeof (ParallelScanJob));
1540 psj->scan_job.ops = ops;
1541 psj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1543 sgen_workers_enqueue_job (&psj->scan_job.job, enqueue);
1545 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("scan LOS remsets", job_scan_los_card_table, sizeof (ParallelScanJob));
1546 psj->scan_job.ops = ops;
1547 psj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1549 sgen_workers_enqueue_job (&psj->scan_job.job, enqueue);
1554 enqueue_scan_from_roots_jobs (SgenGrayQueue *gc_thread_gray_queue, char *heap_start, char *heap_end, SgenObjectOperations *ops, gboolean enqueue)
1556 ScanFromRegisteredRootsJob *scrrj;
1557 ScanThreadDataJob *stdj;
1558 ScanFinalizerEntriesJob *sfej;
1560 /* registered roots, this includes static fields */
1562 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots normal", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1563 scrrj->scan_job.ops = ops;
1564 scrrj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1565 scrrj->heap_start = heap_start;
1566 scrrj->heap_end = heap_end;
1567 scrrj->root_type = ROOT_TYPE_NORMAL;
1568 sgen_workers_enqueue_job (&scrrj->scan_job.job, enqueue);
1570 if (current_collection_generation == GENERATION_OLD) {
1571 /* During minors we scan the cardtable for these roots instead */
1572 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots wbarrier", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1573 scrrj->scan_job.ops = ops;
1574 scrrj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1575 scrrj->heap_start = heap_start;
1576 scrrj->heap_end = heap_end;
1577 scrrj->root_type = ROOT_TYPE_WBARRIER;
1578 sgen_workers_enqueue_job (&scrrj->scan_job.job, enqueue);
1583 stdj = (ScanThreadDataJob*)sgen_thread_pool_job_alloc ("scan thread data", job_scan_thread_data, sizeof (ScanThreadDataJob));
1584 stdj->scan_job.ops = ops;
1585 stdj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1586 stdj->heap_start = heap_start;
1587 stdj->heap_end = heap_end;
1588 sgen_workers_enqueue_job (&stdj->scan_job.job, enqueue);
1590 /* Scan the list of objects ready for finalization. */
1592 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1593 sfej->scan_job.ops = ops;
1594 sfej->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1595 sfej->queue = &fin_ready_queue;
1596 sgen_workers_enqueue_job (&sfej->scan_job.job, enqueue);
1598 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan critical finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1599 sfej->scan_job.ops = ops;
1600 sfej->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1601 sfej->queue = &critical_fin_queue;
1602 sgen_workers_enqueue_job (&sfej->scan_job.job, enqueue);
1606 * Perform a nursery collection.
1608 * Return whether any objects were late-pinned due to being out of memory.
1611 collect_nursery (const char *reason, gboolean is_overflow, SgenGrayQueue *unpin_queue)
1613 gboolean needs_major, is_parallel = FALSE;
1614 mword fragment_total;
1615 SgenGrayQueue gc_thread_gray_queue;
1616 SgenObjectOperations *object_ops_nopar, *object_ops_par = NULL;
1617 ScanCopyContext ctx;
1621 SGEN_TV_DECLARE (last_minor_collection_start_tv);
1622 SGEN_TV_DECLARE (last_minor_collection_end_tv);
1624 if (disable_minor_collections)
1627 TV_GETTIME (last_minor_collection_start_tv);
1628 atv = last_minor_collection_start_tv;
1630 binary_protocol_collection_begin (gc_stats.minor_gc_count, GENERATION_NURSERY);
1632 if (sgen_concurrent_collection_in_progress ()) {
1633 /* FIXME Support parallel nursery collections with concurrent major */
1634 object_ops_nopar = &sgen_minor_collector.serial_ops_with_concurrent_major;
1636 object_ops_nopar = &sgen_minor_collector.serial_ops;
1637 if (sgen_minor_collector.is_parallel) {
1638 object_ops_par = &sgen_minor_collector.parallel_ops;
1643 if (do_verify_nursery || do_dump_nursery_content)
1644 sgen_debug_verify_nursery (do_dump_nursery_content);
1646 current_collection_generation = GENERATION_NURSERY;
1648 SGEN_ASSERT (0, !sgen_collection_is_concurrent (), "Why is the nursery collection concurrent?");
1650 reset_pinned_from_failed_allocation ();
1652 check_scan_starts ();
1654 sgen_nursery_alloc_prepare_for_minor ();
1659 SGEN_LOG (1, "Start nursery collection %d %p-%p, size: %d", gc_stats.minor_gc_count, nursery_section->data, nursery_section->end_data, (int)(nursery_section->end_data - nursery_section->data));
1661 /* world must be stopped already */
1663 time_minor_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1665 sgen_client_pre_collection_checks ();
1667 major_collector.start_nursery_collection ();
1669 sgen_memgov_minor_collection_start ();
1671 init_gray_queue (&gc_thread_gray_queue, is_parallel);
1672 ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops_nopar, &gc_thread_gray_queue);
1674 gc_stats.minor_gc_count ++;
1676 sgen_process_fin_stage_entries ();
1678 /* pin from pinned handles */
1679 sgen_init_pinning ();
1680 sgen_client_binary_protocol_mark_start (GENERATION_NURSERY);
1681 pin_from_roots (nursery_section->data, nursery_section->end_data, ctx);
1682 /* pin cemented objects */
1683 sgen_pin_cemented_objects ();
1684 /* identify pinned objects */
1685 sgen_optimize_pin_queue ();
1686 sgen_pinning_setup_section (nursery_section);
1688 pin_objects_in_nursery (FALSE, ctx);
1689 sgen_pinning_trim_queue_to_section (nursery_section);
1691 if (remset_consistency_checks)
1692 sgen_check_remset_consistency ();
1694 if (whole_heap_check_before_collection) {
1695 sgen_clear_nursery_fragments ();
1696 sgen_check_whole_heap (FALSE);
1700 time_minor_pinning += TV_ELAPSED (btv, atv);
1701 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), (long long)TV_ELAPSED (btv, atv));
1702 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1704 remset.start_scan_remsets ();
1706 enqueue_scan_remembered_set_jobs (&gc_thread_gray_queue, is_parallel ? object_ops_par : object_ops_nopar, is_parallel);
1708 /* we don't have complete write barrier yet, so we scan all the old generation sections */
1710 time_minor_scan_remsets += TV_ELAPSED (atv, btv);
1711 SGEN_LOG (2, "Old generation scan: %lld usecs", (long long)TV_ELAPSED (atv, btv));
1713 sgen_pin_stats_report ();
1715 /* FIXME: Why do we do this at this specific, seemingly random, point? */
1716 sgen_client_collecting_minor (&fin_ready_queue, &critical_fin_queue);
1719 time_minor_scan_pinned += TV_ELAPSED (btv, atv);
1721 enqueue_scan_from_roots_jobs (&gc_thread_gray_queue, nursery_section->data, nursery_section->end_data, is_parallel ? object_ops_par : object_ops_nopar, is_parallel);
1724 gray_queue_redirect (&gc_thread_gray_queue);
1725 sgen_workers_start_all_workers (object_ops_nopar, object_ops_par, NULL);
1726 sgen_workers_join ();
1730 time_minor_scan_roots += TV_ELAPSED (atv, btv);
1732 finish_gray_stack (GENERATION_NURSERY, ctx);
1735 time_minor_finish_gray_stack += TV_ELAPSED (btv, atv);
1736 sgen_client_binary_protocol_mark_end (GENERATION_NURSERY);
1738 if (objects_pinned) {
1739 sgen_optimize_pin_queue ();
1740 sgen_pinning_setup_section (nursery_section);
1744 * This is the latest point at which we can do this check, because
1745 * sgen_build_nursery_fragments() unpins nursery objects again.
1747 if (remset_consistency_checks)
1748 sgen_check_remset_consistency ();
1752 duration = (int)(TV_ELAPSED (last_minor_collection_start_tv, btv) / 10000);
1753 if (duration > (sgen_max_pause_time * sgen_max_pause_margin))
1754 sgen_resize_nursery (TRUE);
1756 sgen_resize_nursery (FALSE);
1758 /* walk the pin_queue, build up the fragment list of free memory, unmark
1759 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1762 sgen_client_binary_protocol_reclaim_start (GENERATION_NURSERY);
1763 fragment_total = sgen_build_nursery_fragments (nursery_section, unpin_queue);
1764 if (!fragment_total)
1767 /* Clear TLABs for all threads */
1768 sgen_clear_tlabs ();
1770 sgen_client_binary_protocol_reclaim_end (GENERATION_NURSERY);
1772 time_minor_fragment_creation += TV_ELAPSED (atv, btv);
1773 SGEN_LOG (2, "Fragment creation: %lld usecs, %lu bytes available", (long long)TV_ELAPSED (atv, btv), (unsigned long)fragment_total);
1775 if (remset_consistency_checks)
1776 sgen_check_major_refs ();
1778 major_collector.finish_nursery_collection ();
1780 TV_GETTIME (last_minor_collection_end_tv);
1781 gc_stats.minor_gc_time += TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv);
1783 sgen_debug_dump_heap ("minor", gc_stats.minor_gc_count - 1, NULL);
1785 /* prepare the pin queue for the next collection */
1786 sgen_finish_pinning ();
1787 if (sgen_have_pending_finalizers ()) {
1788 SGEN_LOG (4, "Finalizer-thread wakeup");
1789 sgen_client_finalize_notify ();
1791 sgen_pin_stats_reset ();
1792 /* clear cemented hash */
1793 sgen_cement_clear_below_threshold ();
1795 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
1797 check_scan_starts ();
1799 binary_protocol_flush_buffers (FALSE);
1801 sgen_memgov_minor_collection_end (reason, is_overflow);
1803 /*objects are late pinned because of lack of memory, so a major is a good call*/
1804 needs_major = objects_pinned > 0;
1805 current_collection_generation = -1;
1808 binary_protocol_collection_end (gc_stats.minor_gc_count - 1, GENERATION_NURSERY, 0, 0);
1810 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1811 sgen_check_nursery_objects_pinned (unpin_queue != NULL);
1817 COPY_OR_MARK_FROM_ROOTS_SERIAL,
1818 COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT,
1819 COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT
1820 } CopyOrMarkFromRootsMode;
1823 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)
1828 /* FIXME: only use these values for the precise scan
1829 * note that to_space pointers should be excluded anyway...
1831 char *heap_start = NULL;
1832 char *heap_end = (char*)-1;
1833 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops_nopar, gc_thread_gray_queue);
1834 gboolean concurrent = mode != COPY_OR_MARK_FROM_ROOTS_SERIAL;
1836 SGEN_ASSERT (0, !!concurrent == !!concurrent_collection_in_progress, "We've been called with the wrong mode.");
1838 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1839 /*This cleans up unused fragments */
1840 sgen_nursery_allocator_prepare_for_pinning ();
1842 if (do_concurrent_checks)
1843 sgen_debug_check_nursery_is_clean ();
1845 /* The concurrent collector doesn't touch the nursery. */
1846 sgen_nursery_alloc_prepare_for_major ();
1851 /* Pinning depends on this */
1852 sgen_clear_nursery_fragments ();
1854 if (whole_heap_check_before_collection)
1855 sgen_check_whole_heap (TRUE);
1858 time_major_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1862 sgen_client_pre_collection_checks ();
1864 if (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1865 /* Remsets are not useful for a major collection */
1866 remset.clear_cards ();
1869 sgen_process_fin_stage_entries ();
1872 sgen_init_pinning ();
1873 SGEN_LOG (6, "Collecting pinned addresses");
1874 pin_from_roots ((void*)lowest_heap_address, (void*)highest_heap_address, ctx);
1875 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1876 /* Pin cemented objects that were forced */
1877 sgen_pin_cemented_objects ();
1879 sgen_optimize_pin_queue ();
1880 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1882 * Cemented objects that are in the pinned list will be marked. When
1883 * marking concurrently we won't mark mod-union cards for these objects.
1884 * Instead they will remain cemented until the next major collection,
1885 * when we will recheck if they are still pinned in the roots.
1887 sgen_cement_force_pinned ();
1890 sgen_client_collecting_major_1 ();
1893 * pin_queue now contains all candidate pointers, sorted and
1894 * uniqued. We must do two passes now to figure out which
1895 * objects are pinned.
1897 * The first is to find within the pin_queue the area for each
1898 * section. This requires that the pin_queue be sorted. We
1899 * also process the LOS objects and pinned chunks here.
1901 * The second, destructive, pass is to reduce the section
1902 * areas to pointers to the actually pinned objects.
1904 SGEN_LOG (6, "Pinning from sections");
1905 /* first pass for the sections */
1906 sgen_find_section_pin_queue_start_end (nursery_section);
1907 /* identify possible pointers to the insize of large objects */
1908 SGEN_LOG (6, "Pinning from large objects");
1909 for (bigobj = los_object_list; bigobj; bigobj = bigobj->next) {
1911 if (sgen_find_optimized_pin_queue_area ((char*)bigobj->data, (char*)bigobj->data + sgen_los_object_size (bigobj), &dummy, &dummy)) {
1912 binary_protocol_pin (bigobj->data, (gpointer)LOAD_VTABLE (bigobj->data), safe_object_get_size (bigobj->data));
1914 if (sgen_los_object_is_pinned (bigobj->data)) {
1915 SGEN_ASSERT (0, mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, "LOS objects can only be pinned here after concurrent marking.");
1918 sgen_los_pin_object (bigobj->data);
1919 if (SGEN_OBJECT_HAS_REFERENCES (bigobj->data))
1920 GRAY_OBJECT_ENQUEUE_SERIAL (gc_thread_gray_queue, bigobj->data, sgen_obj_get_descriptor ((GCObject*)bigobj->data));
1921 sgen_pin_stats_register_object (bigobj->data, GENERATION_OLD);
1922 SGEN_LOG (6, "Marked large object %p (%s) size: %lu from roots", bigobj->data,
1923 sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (bigobj->data)),
1924 (unsigned long)sgen_los_object_size (bigobj));
1926 sgen_client_pinned_los_object (bigobj->data);
1930 pin_objects_in_nursery (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT, ctx);
1931 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1932 sgen_check_nursery_objects_pinned (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1934 major_collector.pin_objects (gc_thread_gray_queue);
1935 if (old_next_pin_slot)
1936 *old_next_pin_slot = sgen_get_pinned_count ();
1939 time_major_pinning += TV_ELAPSED (atv, btv);
1940 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), (long long)TV_ELAPSED (atv, btv));
1941 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1943 major_collector.init_to_space ();
1945 SGEN_ASSERT (0, sgen_workers_all_done (), "Why are the workers not done when we start or finish a major collection?");
1946 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1947 if (object_ops_par != NULL)
1948 sgen_workers_set_num_active_workers (0);
1949 if (sgen_workers_have_idle_work ()) {
1951 * We force the finish of the worker with the new object ops context
1952 * which can also do copying. We need to have finished pinning.
1954 sgen_workers_start_all_workers (object_ops_nopar, object_ops_par, NULL);
1956 sgen_workers_join ();
1960 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1961 main_gc_thread = mono_native_thread_self ();
1964 sgen_client_collecting_major_2 ();
1967 time_major_scan_pinned += TV_ELAPSED (btv, atv);
1969 sgen_client_collecting_major_3 (&fin_ready_queue, &critical_fin_queue);
1971 enqueue_scan_from_roots_jobs (gc_thread_gray_queue, heap_start, heap_end, object_ops_nopar, FALSE);
1974 time_major_scan_roots += TV_ELAPSED (atv, btv);
1977 * We start the concurrent worker after pinning and after we scanned the roots
1978 * in order to make sure that the worker does not finish before handling all
1981 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1982 sgen_workers_set_num_active_workers (1);
1983 gray_queue_redirect (gc_thread_gray_queue);
1984 if (precleaning_enabled) {
1985 sgen_workers_start_all_workers (object_ops_nopar, object_ops_par, workers_finish_callback);
1987 sgen_workers_start_all_workers (object_ops_nopar, object_ops_par, NULL);
1991 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1992 int i, split_count = sgen_workers_get_job_split_count ();
1993 gboolean parallel = object_ops_par != NULL;
1995 /* If we're not parallel we finish the collection on the gc thread */
1997 gray_queue_redirect (gc_thread_gray_queue);
1999 /* Mod union card table */
2000 for (i = 0; i < split_count; i++) {
2001 ParallelScanJob *psj;
2003 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("scan mod union cardtable", job_scan_major_mod_union_card_table, sizeof (ParallelScanJob));
2004 psj->scan_job.ops = object_ops_par ? object_ops_par : object_ops_nopar;
2005 psj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
2007 sgen_workers_enqueue_job (&psj->scan_job.job, parallel);
2009 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("scan LOS mod union cardtable", job_scan_los_mod_union_card_table, sizeof (ParallelScanJob));
2010 psj->scan_job.ops = object_ops_par ? object_ops_par : object_ops_nopar;
2011 psj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
2013 sgen_workers_enqueue_job (&psj->scan_job.job, parallel);
2018 * If we enqueue a job while workers are running we need to sgen_workers_ensure_awake
2019 * in order to make sure that we are running the idle func and draining all worker
2020 * gray queues. The operation of starting workers implies this, so we start them after
2021 * in order to avoid doing this operation twice. The workers will drain the main gray
2022 * stack that contained roots and pinned objects and also scan the mod union card
2025 sgen_workers_start_all_workers (object_ops_nopar, object_ops_par, NULL);
2026 sgen_workers_join ();
2030 sgen_pin_stats_report ();
2032 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
2033 sgen_finish_pinning ();
2035 sgen_pin_stats_reset ();
2037 if (do_concurrent_checks)
2038 sgen_debug_check_nursery_is_clean ();
2043 major_start_collection (SgenGrayQueue *gc_thread_gray_queue, const char *reason, gboolean concurrent, size_t *old_next_pin_slot)
2045 SgenObjectOperations *object_ops_nopar, *object_ops_par = NULL;
2047 binary_protocol_collection_begin (gc_stats.major_gc_count, GENERATION_OLD);
2049 current_collection_generation = GENERATION_OLD;
2051 sgen_workers_assert_gray_queue_is_empty ();
2054 sgen_cement_reset ();
2057 g_assert (major_collector.is_concurrent);
2058 concurrent_collection_in_progress = TRUE;
2060 object_ops_nopar = &major_collector.major_ops_concurrent_start;
2061 if (major_collector.is_parallel)
2062 object_ops_par = &major_collector.major_ops_conc_par_start;
2065 object_ops_nopar = &major_collector.major_ops_serial;
2068 reset_pinned_from_failed_allocation ();
2070 sgen_memgov_major_collection_start (concurrent, reason);
2072 //count_ref_nonref_objs ();
2073 //consistency_check ();
2075 check_scan_starts ();
2078 SGEN_LOG (1, "Start major collection %d", gc_stats.major_gc_count);
2079 gc_stats.major_gc_count ++;
2081 if (major_collector.start_major_collection)
2082 major_collector.start_major_collection ();
2084 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);
2088 major_finish_collection (SgenGrayQueue *gc_thread_gray_queue, const char *reason, gboolean is_overflow, size_t old_next_pin_slot, gboolean forced)
2090 ScannedObjectCounts counts;
2091 SgenObjectOperations *object_ops_nopar;
2092 mword fragment_total;
2098 if (concurrent_collection_in_progress) {
2099 SgenObjectOperations *object_ops_par = NULL;
2101 object_ops_nopar = &major_collector.major_ops_concurrent_finish;
2102 if (major_collector.is_parallel)
2103 object_ops_par = &major_collector.major_ops_conc_par_finish;
2105 major_copy_or_mark_from_roots (gc_thread_gray_queue, NULL, COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, object_ops_nopar, object_ops_par);
2107 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
2108 main_gc_thread = NULL;
2111 object_ops_nopar = &major_collector.major_ops_serial;
2114 sgen_workers_assert_gray_queue_is_empty ();
2116 finish_gray_stack (GENERATION_OLD, CONTEXT_FROM_OBJECT_OPERATIONS (object_ops_nopar, gc_thread_gray_queue));
2118 time_major_finish_gray_stack += TV_ELAPSED (btv, atv);
2120 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after joining");
2122 if (objects_pinned) {
2123 g_assert (!concurrent_collection_in_progress);
2126 * This is slow, but we just OOM'd.
2128 * See comment at `sgen_pin_queue_clear_discarded_entries` for how the pin
2129 * queue is laid out at this point.
2131 sgen_pin_queue_clear_discarded_entries (nursery_section, old_next_pin_slot);
2133 * We need to reestablish all pinned nursery objects in the pin queue
2134 * because they're needed for fragment creation. Unpinning happens by
2135 * walking the whole queue, so it's not necessary to reestablish where major
2136 * heap block pins are - all we care is that they're still in there
2139 sgen_optimize_pin_queue ();
2140 sgen_find_section_pin_queue_start_end (nursery_section);
2144 reset_heap_boundaries ();
2145 sgen_update_heap_boundaries ((mword)sgen_get_nursery_start (), (mword)sgen_get_nursery_end ());
2147 /* walk the pin_queue, build up the fragment list of free memory, unmark
2148 * pinned objects as we go, memzero() the empty fragments so they are ready for the
2151 fragment_total = sgen_build_nursery_fragments (nursery_section, NULL);
2152 if (!fragment_total)
2154 SGEN_LOG (4, "Free space in nursery after major %ld", (long)fragment_total);
2156 if (do_concurrent_checks && concurrent_collection_in_progress)
2157 sgen_debug_check_nursery_is_clean ();
2159 /* prepare the pin queue for the next collection */
2160 sgen_finish_pinning ();
2162 /* Clear TLABs for all threads */
2163 sgen_clear_tlabs ();
2165 sgen_pin_stats_reset ();
2167 sgen_cement_clear_below_threshold ();
2169 if (check_mark_bits_after_major_collection)
2170 sgen_check_heap_marked (concurrent_collection_in_progress);
2173 time_major_fragment_creation += TV_ELAPSED (atv, btv);
2175 binary_protocol_sweep_begin (GENERATION_OLD, !major_collector.sweeps_lazily);
2176 sgen_memgov_major_pre_sweep ();
2179 time_major_free_bigobjs += TV_ELAPSED (btv, atv);
2184 time_major_los_sweep += TV_ELAPSED (atv, btv);
2186 major_collector.sweep ();
2188 binary_protocol_sweep_end (GENERATION_OLD, !major_collector.sweeps_lazily);
2191 time_major_sweep += TV_ELAPSED (btv, atv);
2193 sgen_debug_dump_heap ("major", gc_stats.major_gc_count - 1, reason);
2195 if (sgen_have_pending_finalizers ()) {
2196 SGEN_LOG (4, "Finalizer-thread wakeup");
2197 sgen_client_finalize_notify ();
2200 sgen_memgov_major_collection_end (forced, concurrent_collection_in_progress, reason, is_overflow);
2201 current_collection_generation = -1;
2203 memset (&counts, 0, sizeof (ScannedObjectCounts));
2204 major_collector.finish_major_collection (&counts);
2206 sgen_workers_assert_gray_queue_is_empty ();
2208 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after major collection has finished");
2209 if (concurrent_collection_in_progress)
2210 concurrent_collection_in_progress = FALSE;
2212 check_scan_starts ();
2214 binary_protocol_flush_buffers (FALSE);
2216 //consistency_check ();
2218 binary_protocol_collection_end (gc_stats.major_gc_count - 1, GENERATION_OLD, counts.num_scanned_objects, counts.num_unique_scanned_objects);
2222 major_do_collection (const char *reason, gboolean is_overflow, gboolean forced)
2224 TV_DECLARE (time_start);
2225 TV_DECLARE (time_end);
2226 size_t old_next_pin_slot;
2227 SgenGrayQueue gc_thread_gray_queue;
2229 if (disable_major_collections)
2232 if (major_collector.get_and_reset_num_major_objects_marked) {
2233 long long num_marked = major_collector.get_and_reset_num_major_objects_marked ();
2234 g_assert (!num_marked);
2237 /* world must be stopped already */
2238 TV_GETTIME (time_start);
2240 init_gray_queue (&gc_thread_gray_queue, FALSE);
2241 major_start_collection (&gc_thread_gray_queue, reason, FALSE, &old_next_pin_slot);
2242 major_finish_collection (&gc_thread_gray_queue, reason, is_overflow, old_next_pin_slot, forced);
2243 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2245 TV_GETTIME (time_end);
2246 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2248 /* FIXME: also report this to the user, preferably in gc-end. */
2249 if (major_collector.get_and_reset_num_major_objects_marked)
2250 major_collector.get_and_reset_num_major_objects_marked ();
2252 return bytes_pinned_from_failed_allocation > 0;
2256 major_start_concurrent_collection (const char *reason)
2258 TV_DECLARE (time_start);
2259 TV_DECLARE (time_end);
2260 long long num_objects_marked;
2261 SgenGrayQueue gc_thread_gray_queue;
2263 if (disable_major_collections)
2266 TV_GETTIME (time_start);
2267 SGEN_TV_GETTIME (time_major_conc_collection_start);
2269 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2270 g_assert (num_objects_marked == 0);
2272 binary_protocol_concurrent_start ();
2274 init_gray_queue (&gc_thread_gray_queue, TRUE);
2275 // FIXME: store reason and pass it when finishing
2276 major_start_collection (&gc_thread_gray_queue, reason, TRUE, NULL);
2277 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2279 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2281 TV_GETTIME (time_end);
2282 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2284 current_collection_generation = -1;
2288 * Returns whether the major collection has finished.
2291 major_should_finish_concurrent_collection (void)
2293 return sgen_workers_all_done ();
2297 major_update_concurrent_collection (void)
2299 TV_DECLARE (total_start);
2300 TV_DECLARE (total_end);
2302 TV_GETTIME (total_start);
2304 binary_protocol_concurrent_update ();
2306 major_collector.update_cardtable_mod_union ();
2307 sgen_los_update_cardtable_mod_union ();
2309 TV_GETTIME (total_end);
2310 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end);
2314 major_finish_concurrent_collection (gboolean forced)
2316 SgenGrayQueue gc_thread_gray_queue;
2317 TV_DECLARE (total_start);
2318 TV_DECLARE (total_end);
2320 TV_GETTIME (total_start);
2322 binary_protocol_concurrent_finish ();
2325 * We need to stop all workers since we're updating the cardtable below.
2326 * The workers will be resumed with a finishing pause context to avoid
2327 * additional cardtable and object scanning.
2329 sgen_workers_stop_all_workers ();
2331 SGEN_TV_GETTIME (time_major_conc_collection_end);
2332 gc_stats.major_gc_time_concurrent += SGEN_TV_ELAPSED (time_major_conc_collection_start, time_major_conc_collection_end);
2334 major_collector.update_cardtable_mod_union ();
2335 sgen_los_update_cardtable_mod_union ();
2337 if (mod_union_consistency_check)
2338 sgen_check_mod_union_consistency ();
2340 current_collection_generation = GENERATION_OLD;
2341 sgen_cement_reset ();
2342 init_gray_queue (&gc_thread_gray_queue, FALSE);
2343 major_finish_collection (&gc_thread_gray_queue, "finishing", FALSE, -1, forced);
2344 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2346 TV_GETTIME (total_end);
2347 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end);
2349 current_collection_generation = -1;
2353 * Ensure an allocation request for @size will succeed by freeing enough memory.
2355 * LOCKING: The GC lock MUST be held.
2358 sgen_ensure_free_space (size_t size, int generation)
2360 int generation_to_collect = -1;
2361 const char *reason = NULL;
2363 if (generation == GENERATION_OLD) {
2364 if (sgen_need_major_collection (size)) {
2365 reason = "LOS overflow";
2366 generation_to_collect = GENERATION_OLD;
2369 if (degraded_mode) {
2370 if (sgen_need_major_collection (size)) {
2371 reason = "Degraded mode overflow";
2372 generation_to_collect = GENERATION_OLD;
2374 } else if (sgen_need_major_collection (size)) {
2375 reason = concurrent_collection_in_progress ? "Forced finish concurrent collection" : "Minor allowance";
2376 generation_to_collect = GENERATION_OLD;
2378 generation_to_collect = GENERATION_NURSERY;
2379 reason = "Nursery full";
2383 if (generation_to_collect == -1) {
2384 if (concurrent_collection_in_progress && sgen_workers_all_done ()) {
2385 generation_to_collect = GENERATION_OLD;
2386 reason = "Finish concurrent collection";
2390 if (generation_to_collect == -1)
2392 sgen_perform_collection (size, generation_to_collect, reason, FALSE, TRUE);
2396 * LOCKING: Assumes the GC lock is held.
2399 sgen_perform_collection (size_t requested_size, int generation_to_collect, const char *reason, gboolean wait_to_finish, gboolean stw)
2401 TV_DECLARE (gc_total_start);
2402 TV_DECLARE (gc_total_end);
2403 int overflow_generation_to_collect = -1;
2404 int oldest_generation_collected = generation_to_collect;
2405 const char *overflow_reason = NULL;
2406 gboolean finish_concurrent = concurrent_collection_in_progress && (major_should_finish_concurrent_collection () || generation_to_collect == GENERATION_OLD);
2408 binary_protocol_collection_requested (generation_to_collect, requested_size, wait_to_finish ? 1 : 0);
2410 SGEN_ASSERT (0, generation_to_collect == GENERATION_NURSERY || generation_to_collect == GENERATION_OLD, "What generation is this?");
2413 sgen_stop_world (generation_to_collect);
2415 SGEN_ASSERT (0, sgen_is_world_stopped (), "We can only collect if the world is stopped");
2418 TV_GETTIME (gc_total_start);
2420 // FIXME: extract overflow reason
2421 // FIXME: minor overflow for concurrent case
2422 if (generation_to_collect == GENERATION_NURSERY && !finish_concurrent) {
2423 if (concurrent_collection_in_progress)
2424 major_update_concurrent_collection ();
2426 if (collect_nursery (reason, FALSE, NULL) && !concurrent_collection_in_progress) {
2427 overflow_generation_to_collect = GENERATION_OLD;
2428 overflow_reason = "Minor overflow";
2430 } else if (finish_concurrent) {
2431 major_finish_concurrent_collection (wait_to_finish);
2432 oldest_generation_collected = GENERATION_OLD;
2434 SGEN_ASSERT (0, generation_to_collect == GENERATION_OLD, "We should have handled nursery collections above");
2435 if (major_collector.is_concurrent && !wait_to_finish) {
2436 collect_nursery ("Concurrent start", FALSE, NULL);
2437 major_start_concurrent_collection (reason);
2438 oldest_generation_collected = GENERATION_NURSERY;
2439 } else if (major_do_collection (reason, FALSE, wait_to_finish)) {
2440 overflow_generation_to_collect = GENERATION_NURSERY;
2441 overflow_reason = "Excessive pinning";
2445 if (overflow_generation_to_collect != -1) {
2446 SGEN_ASSERT (0, !concurrent_collection_in_progress, "We don't yet support overflow collections with the concurrent collector");
2449 * We need to do an overflow collection, either because we ran out of memory
2450 * or the nursery is fully pinned.
2453 if (overflow_generation_to_collect == GENERATION_NURSERY)
2454 collect_nursery (overflow_reason, TRUE, NULL);
2456 major_do_collection (overflow_reason, TRUE, wait_to_finish);
2458 oldest_generation_collected = MAX (oldest_generation_collected, overflow_generation_to_collect);
2461 SGEN_LOG (2, "Heap size: %lu, LOS size: %lu", (unsigned long)sgen_gc_get_total_heap_allocation (), (unsigned long)los_memory_usage);
2463 /* this also sets the proper pointers for the next allocation */
2464 if (generation_to_collect == GENERATION_NURSERY && !sgen_can_alloc_size (requested_size)) {
2465 /* TypeBuilder and MonoMethod are killing mcs with fragmentation */
2466 SGEN_LOG (1, "nursery collection didn't find enough room for %zd alloc (%zd pinned)", requested_size, sgen_get_pinned_count ());
2467 sgen_dump_pin_queue ();
2471 TV_GETTIME (gc_total_end);
2472 time_max = MAX (time_max, TV_ELAPSED (gc_total_start, gc_total_end));
2475 sgen_restart_world (oldest_generation_collected);
2479 * ######################################################################
2480 * ######## Memory allocation from the OS
2481 * ######################################################################
2482 * This section of code deals with getting memory from the OS and
2483 * allocating memory for GC-internal data structures.
2484 * Internal memory can be handled with a freelist for small objects.
2490 G_GNUC_UNUSED static void
2491 report_internal_mem_usage (void)
2493 printf ("Internal memory usage:\n");
2494 sgen_report_internal_mem_usage ();
2495 printf ("Pinned memory usage:\n");
2496 major_collector.report_pinned_memory_usage ();
2500 * ######################################################################
2501 * ######## Finalization support
2502 * ######################################################################
2506 * If the object has been forwarded it means it's still referenced from a root.
2507 * If it is pinned it's still alive as well.
2508 * A LOS object is only alive if we have pinned it.
2509 * Return TRUE if @obj is ready to be finalized.
2511 static inline gboolean
2512 sgen_is_object_alive (GCObject *object)
2514 if (ptr_in_nursery (object))
2515 return sgen_nursery_is_object_alive (object);
2517 return sgen_major_is_object_alive (object);
2521 * This function returns true if @object is either alive and belongs to the
2522 * current collection - major collections are full heap, so old gen objects
2523 * are never alive during a minor collection.
2526 sgen_is_object_alive_and_on_current_collection (GCObject *object)
2528 if (ptr_in_nursery (object))
2529 return sgen_nursery_is_object_alive (object);
2531 if (current_collection_generation == GENERATION_NURSERY)
2534 return sgen_major_is_object_alive (object);
2539 sgen_gc_is_object_ready_for_finalization (GCObject *object)
2541 return !sgen_is_object_alive (object);
2545 sgen_queue_finalization_entry (GCObject *obj)
2547 gboolean critical = sgen_client_object_has_critical_finalizer (obj);
2549 sgen_pointer_queue_add (critical ? &critical_fin_queue : &fin_ready_queue, obj);
2551 sgen_client_object_queued_for_finalization (obj);
2555 sgen_object_is_live (GCObject *obj)
2557 return sgen_is_object_alive_and_on_current_collection (obj);
2561 * `System.GC.WaitForPendingFinalizers` first checks `sgen_have_pending_finalizers()` to
2562 * determine whether it can exit quickly. The latter must therefore only return FALSE if
2563 * all finalizers have really finished running.
2565 * `sgen_gc_invoke_finalizers()` first dequeues a finalizable object, and then finalizes it.
2566 * This means that just checking whether the queues are empty leaves the possibility that an
2567 * object might have been dequeued but not yet finalized. That's why we need the additional
2568 * flag `pending_unqueued_finalizer`.
2571 static volatile gboolean pending_unqueued_finalizer = FALSE;
2572 volatile gboolean sgen_suspend_finalizers = FALSE;
2575 sgen_set_suspend_finalizers (void)
2577 sgen_suspend_finalizers = TRUE;
2581 sgen_gc_invoke_finalizers (void)
2585 g_assert (!pending_unqueued_finalizer);
2587 /* FIXME: batch to reduce lock contention */
2588 while (sgen_have_pending_finalizers ()) {
2594 * We need to set `pending_unqueued_finalizer` before dequeing the
2595 * finalizable object.
2597 if (!sgen_pointer_queue_is_empty (&fin_ready_queue)) {
2598 pending_unqueued_finalizer = TRUE;
2599 mono_memory_write_barrier ();
2600 obj = (GCObject *)sgen_pointer_queue_pop (&fin_ready_queue);
2601 } else if (!sgen_pointer_queue_is_empty (&critical_fin_queue)) {
2602 pending_unqueued_finalizer = TRUE;
2603 mono_memory_write_barrier ();
2604 obj = (GCObject *)sgen_pointer_queue_pop (&critical_fin_queue);
2610 SGEN_LOG (7, "Finalizing object %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
2618 /* the object is on the stack so it is pinned */
2619 /*g_print ("Calling finalizer for object: %p (%s)\n", obj, sgen_client_object_safe_name (obj));*/
2620 sgen_client_run_finalize (obj);
2623 if (pending_unqueued_finalizer) {
2624 mono_memory_write_barrier ();
2625 pending_unqueued_finalizer = FALSE;
2632 sgen_have_pending_finalizers (void)
2634 if (sgen_suspend_finalizers)
2636 return pending_unqueued_finalizer || !sgen_pointer_queue_is_empty (&fin_ready_queue) || !sgen_pointer_queue_is_empty (&critical_fin_queue);
2640 * ######################################################################
2641 * ######## registered roots support
2642 * ######################################################################
2646 * We do not coalesce roots.
2649 sgen_register_root (char *start, size_t size, SgenDescriptor descr, int root_type, int source, const char *msg)
2651 RootRecord new_root;
2654 for (i = 0; i < ROOT_TYPE_NUM; ++i) {
2655 RootRecord *root = (RootRecord *)sgen_hash_table_lookup (&roots_hash [i], start);
2656 /* we allow changing the size and the descriptor (for thread statics etc) */
2658 size_t old_size = root->end_root - start;
2659 root->end_root = start + size;
2660 SGEN_ASSERT (0, !!root->root_desc == !!descr, "Can't change whether a root is precise or conservative.");
2661 SGEN_ASSERT (0, root->source == source, "Can't change a root's source identifier.");
2662 SGEN_ASSERT (0, !!root->msg == !!msg, "Can't change a root's message.");
2663 root->root_desc = descr;
2665 roots_size -= old_size;
2671 new_root.end_root = start + size;
2672 new_root.root_desc = descr;
2673 new_root.source = source;
2676 sgen_hash_table_replace (&roots_hash [root_type], start, &new_root, NULL);
2679 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);
2686 sgen_deregister_root (char* addr)
2692 for (root_type = 0; root_type < ROOT_TYPE_NUM; ++root_type) {
2693 if (sgen_hash_table_remove (&roots_hash [root_type], addr, &root))
2694 roots_size -= (root.end_root - addr);
2700 sgen_wbroots_iterate_live_block_ranges (sgen_cardtable_block_callback cb)
2704 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_WBARRIER], void **, start_root, RootRecord *, root) {
2705 cb ((mword)start_root, (mword)root->end_root - (mword)start_root);
2706 } SGEN_HASH_TABLE_FOREACH_END;
2709 /* Root equivalent of sgen_client_cardtable_scan_object */
2711 sgen_wbroot_scan_card_table (void** start_root, mword size, ScanCopyContext ctx)
2713 ScanPtrFieldFunc scan_field_func = ctx.ops->scan_ptr_field;
2714 guint8 *card_data = sgen_card_table_get_card_scan_address ((mword)start_root);
2715 guint8 *card_base = card_data;
2716 mword card_count = sgen_card_table_number_of_cards_in_range ((mword)start_root, size);
2717 guint8 *card_data_end = card_data + card_count;
2718 mword extra_idx = 0;
2719 char *obj_start = sgen_card_table_align_pointer (start_root);
2720 char *obj_end = (char*)start_root + size;
2721 #ifdef SGEN_HAVE_OVERLAPPING_CARDS
2722 guint8 *overflow_scan_end = NULL;
2725 #ifdef SGEN_HAVE_OVERLAPPING_CARDS
2726 /*Check for overflow and if so, setup to scan in two steps*/
2727 if (card_data_end >= SGEN_SHADOW_CARDTABLE_END) {
2728 overflow_scan_end = sgen_shadow_cardtable + (card_data_end - SGEN_SHADOW_CARDTABLE_END);
2729 card_data_end = SGEN_SHADOW_CARDTABLE_END;
2735 card_data = sgen_find_next_card (card_data, card_data_end);
2737 for (; card_data < card_data_end; card_data = sgen_find_next_card (card_data + 1, card_data_end)) {
2738 size_t idx = (card_data - card_base) + extra_idx;
2739 char *start = (char*)(obj_start + idx * CARD_SIZE_IN_BYTES);
2740 char *card_end = start + CARD_SIZE_IN_BYTES;
2741 char *elem = start, *first_elem = start;
2744 * Don't clean first and last card on 32bit systems since they
2745 * may also be part from other roots.
2747 if (card_data != card_base && card_data != (card_data_end - 1))
2748 sgen_card_table_prepare_card_for_scanning (card_data);
2750 card_end = MIN (card_end, obj_end);
2752 if (elem < (char*)start_root)
2753 first_elem = elem = (char*)start_root;
2755 for (; elem < card_end; elem += SIZEOF_VOID_P) {
2756 if (*(GCObject**)elem)
2757 scan_field_func (NULL, (GCObject**)elem, ctx.queue);
2760 binary_protocol_card_scan (first_elem, elem - first_elem);
2763 #ifdef SGEN_HAVE_OVERLAPPING_CARDS
2764 if (overflow_scan_end) {
2765 extra_idx = card_data - card_base;
2766 card_base = card_data = sgen_shadow_cardtable;
2767 card_data_end = overflow_scan_end;
2768 overflow_scan_end = NULL;
2775 sgen_wbroots_scan_card_table (ScanCopyContext ctx)
2780 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_WBARRIER], void **, start_root, RootRecord *, root) {
2781 SGEN_ASSERT (0, (root->root_desc & ROOT_DESC_TYPE_MASK) == ROOT_DESC_VECTOR, "Unsupported root type");
2783 sgen_wbroot_scan_card_table (start_root, (mword)root->end_root - (mword)start_root, ctx);
2784 } SGEN_HASH_TABLE_FOREACH_END;
2788 * ######################################################################
2789 * ######## Thread handling (stop/start code)
2790 * ######################################################################
2794 sgen_get_current_collection_generation (void)
2796 return current_collection_generation;
2800 sgen_thread_register (SgenThreadInfo* info, void *stack_bottom_fallback)
2802 info->tlab_start = info->tlab_next = info->tlab_temp_end = info->tlab_real_end = NULL;
2804 sgen_client_thread_register (info, stack_bottom_fallback);
2810 sgen_thread_unregister (SgenThreadInfo *p)
2812 sgen_client_thread_unregister (p);
2816 * ######################################################################
2817 * ######## Write barriers
2818 * ######################################################################
2822 * Note: the write barriers first do the needed GC work and then do the actual store:
2823 * this way the value is visible to the conservative GC scan after the write barrier
2824 * itself. If a GC interrupts the barrier in the middle, value will be kept alive by
2825 * the conservative scan, otherwise by the remembered set scan.
2829 * mono_gc_wbarrier_arrayref_copy:
2832 mono_gc_wbarrier_arrayref_copy (gpointer dest_ptr, gpointer src_ptr, int count)
2834 HEAVY_STAT (++stat_wbarrier_arrayref_copy);
2835 /*This check can be done without taking a lock since dest_ptr array is pinned*/
2836 if (ptr_in_nursery (dest_ptr) || count <= 0) {
2837 mono_gc_memmove_aligned (dest_ptr, src_ptr, count * sizeof (gpointer));
2841 #ifdef SGEN_HEAVY_BINARY_PROTOCOL
2842 if (binary_protocol_is_heavy_enabled ()) {
2844 for (i = 0; i < count; ++i) {
2845 gpointer dest = (gpointer*)dest_ptr + i;
2846 gpointer obj = *((gpointer*)src_ptr + i);
2848 binary_protocol_wbarrier (dest, obj, (gpointer)LOAD_VTABLE (obj));
2853 remset.wbarrier_arrayref_copy (dest_ptr, src_ptr, count);
2857 * mono_gc_wbarrier_generic_nostore:
2860 mono_gc_wbarrier_generic_nostore (gpointer ptr)
2864 HEAVY_STAT (++stat_wbarrier_generic_store);
2866 sgen_client_wbarrier_generic_nostore_check (ptr);
2868 obj = *(gpointer*)ptr;
2870 binary_protocol_wbarrier (ptr, obj, (gpointer)LOAD_VTABLE (obj));
2873 * We need to record old->old pointer locations for the
2874 * concurrent collector.
2876 if (!ptr_in_nursery (obj) && !concurrent_collection_in_progress) {
2877 SGEN_LOG (8, "Skipping remset at %p", ptr);
2881 SGEN_LOG (8, "Adding remset at %p", ptr);
2883 remset.wbarrier_generic_nostore (ptr);
2887 * mono_gc_wbarrier_generic_store:
2890 mono_gc_wbarrier_generic_store (gpointer ptr, GCObject* value)
2892 SGEN_LOG (8, "Wbarrier store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2893 SGEN_UPDATE_REFERENCE_ALLOW_NULL (ptr, value);
2894 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2895 mono_gc_wbarrier_generic_nostore (ptr);
2896 sgen_dummy_use (value);
2900 * mono_gc_wbarrier_generic_store_atomic:
2901 * Same as \c mono_gc_wbarrier_generic_store but performs the store
2902 * as an atomic operation with release semantics.
2905 mono_gc_wbarrier_generic_store_atomic (gpointer ptr, GCObject *value)
2907 HEAVY_STAT (++stat_wbarrier_generic_store_atomic);
2909 SGEN_LOG (8, "Wbarrier atomic store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2911 InterlockedWritePointer ((volatile gpointer *)ptr, value);
2913 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2914 mono_gc_wbarrier_generic_nostore (ptr);
2916 sgen_dummy_use (value);
2920 sgen_wbarrier_range_copy (gpointer _dest, gpointer _src, int size)
2922 remset.wbarrier_range_copy (_dest,_src, size);
2926 * ######################################################################
2927 * ######## Other mono public interface functions.
2928 * ######################################################################
2932 sgen_gc_collect (int generation)
2937 sgen_perform_collection (0, generation, "user request", TRUE, TRUE);
2942 sgen_gc_collection_count (int generation)
2944 if (generation == 0)
2945 return gc_stats.minor_gc_count;
2946 return gc_stats.major_gc_count;
2950 sgen_gc_get_used_size (void)
2954 tot = los_memory_usage;
2955 tot += nursery_section->end_data - nursery_section->data;
2956 tot += major_collector.get_used_size ();
2957 /* FIXME: account for pinned objects */
2963 sgen_env_var_error (const char *env_var, const char *fallback, const char *description_format, ...)
2967 va_start (ap, description_format);
2969 fprintf (stderr, "Warning: In environment variable `%s': ", env_var);
2970 vfprintf (stderr, description_format, ap);
2972 fprintf (stderr, " - %s", fallback);
2973 fprintf (stderr, "\n");
2979 parse_double_in_interval (const char *env_var, const char *opt_name, const char *opt, double min, double max, double *result)
2982 double val = strtod (opt, &endptr);
2983 if (endptr == opt) {
2984 sgen_env_var_error (env_var, "Using default value.", "`%s` must be a number.", opt_name);
2987 else if (val < min || val > max) {
2988 sgen_env_var_error (env_var, "Using default value.", "`%s` must be between %.2f - %.2f.", opt_name, min, max);
3000 char *major_collector_opt = NULL;
3001 char *minor_collector_opt = NULL;
3002 char *params_opts = NULL;
3003 char *debug_opts = NULL;
3004 size_t max_heap = 0;
3005 size_t soft_limit = 0;
3007 gboolean debug_print_allowance = FALSE;
3008 double allowance_ratio = 0, save_target = 0;
3009 gboolean cement_enabled = TRUE;
3010 gboolean dynamic_nursery = FALSE;
3011 size_t min_nursery_size = 0, max_nursery_size = 0;
3014 result = InterlockedCompareExchange (&gc_initialized, -1, 0);
3017 /* already inited */
3020 /* being inited by another thread */
3021 mono_thread_info_usleep (1000);
3024 /* we will init it */
3027 g_assert_not_reached ();
3029 } while (result != 0);
3031 SGEN_TV_GETTIME (sgen_init_timestamp);
3033 #ifdef SGEN_WITHOUT_MONO
3034 mono_thread_smr_init ();
3037 mono_coop_mutex_init (&gc_mutex);
3039 gc_debug_file = stderr;
3041 mono_coop_mutex_init (&sgen_interruption_mutex);
3043 if ((env = g_getenv (MONO_GC_PARAMS_NAME)) || gc_params_options) {
3044 params_opts = g_strdup_printf ("%s,%s", gc_params_options ? gc_params_options : "", env ? env : "");
3049 opts = g_strsplit (params_opts, ",", -1);
3050 for (ptr = opts; *ptr; ++ptr) {
3052 if (g_str_has_prefix (opt, "major=")) {
3053 opt = strchr (opt, '=') + 1;
3054 major_collector_opt = g_strdup (opt);
3055 } else if (g_str_has_prefix (opt, "minor=")) {
3056 opt = strchr (opt, '=') + 1;
3057 minor_collector_opt = g_strdup (opt);
3065 sgen_init_internal_allocator ();
3066 sgen_init_nursery_allocator ();
3067 sgen_init_fin_weak_hash ();
3068 sgen_init_hash_table ();
3069 sgen_init_descriptors ();
3070 sgen_init_gray_queues ();
3071 sgen_init_allocator ();
3072 sgen_init_gchandles ();
3074 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_SECTION, SGEN_SIZEOF_GC_MEM_SECTION);
3075 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_GRAY_QUEUE, sizeof (GrayQueueSection));
3077 sgen_client_init ();
3079 if (!minor_collector_opt) {
3080 sgen_simple_nursery_init (&sgen_minor_collector, FALSE);
3082 if (!strcmp (minor_collector_opt, "simple")) {
3084 sgen_simple_nursery_init (&sgen_minor_collector, FALSE);
3085 } else if (!strcmp (minor_collector_opt, "simple-par")) {
3086 sgen_simple_nursery_init (&sgen_minor_collector, TRUE);
3087 } else if (!strcmp (minor_collector_opt, "split")) {
3088 sgen_split_nursery_init (&sgen_minor_collector);
3090 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `simple` instead.", "Unknown minor collector `%s'.", minor_collector_opt);
3091 goto use_simple_nursery;
3095 if (!major_collector_opt) {
3097 DEFAULT_MAJOR_INIT (&major_collector);
3098 } else if (!strcmp (major_collector_opt, "marksweep")) {
3099 sgen_marksweep_init (&major_collector);
3100 } else if (!strcmp (major_collector_opt, "marksweep-conc")) {
3101 sgen_marksweep_conc_init (&major_collector);
3102 } else if (!strcmp (major_collector_opt, "marksweep-conc-par")) {
3103 sgen_marksweep_conc_par_init (&major_collector);
3105 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `" DEFAULT_MAJOR_NAME "` instead.", "Unknown major collector `%s'.", major_collector_opt);
3106 goto use_default_major;
3110 gboolean usage_printed = FALSE;
3112 for (ptr = opts; *ptr; ++ptr) {
3114 if (!strcmp (opt, ""))
3116 if (g_str_has_prefix (opt, "major="))
3118 if (g_str_has_prefix (opt, "minor="))
3120 if (g_str_has_prefix (opt, "max-heap-size=")) {
3121 size_t page_size = mono_pagesize ();
3122 size_t max_heap_candidate = 0;
3123 opt = strchr (opt, '=') + 1;
3124 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &max_heap_candidate)) {
3125 max_heap = (max_heap_candidate + page_size - 1) & ~(size_t)(page_size - 1);
3126 if (max_heap != max_heap_candidate)
3127 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Rounding up.", "`max-heap-size` size must be a multiple of %d.", page_size);
3129 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`max-heap-size` must be an integer.");
3133 if (g_str_has_prefix (opt, "soft-heap-limit=")) {
3134 opt = strchr (opt, '=') + 1;
3135 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &soft_limit)) {
3136 if (soft_limit <= 0) {
3137 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be positive.");
3141 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be an integer.");
3145 if (g_str_has_prefix (opt, "nursery-size=")) {
3147 opt = strchr (opt, '=') + 1;
3148 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &val)) {
3149 if ((val & (val - 1))) {
3150 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be a power of two.");
3154 if (val < SGEN_MAX_NURSERY_WASTE) {
3155 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.",
3156 "`nursery-size` must be at least %d bytes.", SGEN_MAX_NURSERY_WASTE);
3160 min_nursery_size = max_nursery_size = val;
3161 dynamic_nursery = FALSE;
3163 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be an integer.");
3168 if (g_str_has_prefix (opt, "save-target-ratio=")) {
3170 opt = strchr (opt, '=') + 1;
3171 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "save-target-ratio", opt,
3172 SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO, &val)) {
3177 if (g_str_has_prefix (opt, "default-allowance-ratio=")) {
3179 opt = strchr (opt, '=') + 1;
3180 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "default-allowance-ratio", opt,
3181 SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, SGEN_MAX_ALLOWANCE_NURSERY_SIZE_RATIO, &val)) {
3182 allowance_ratio = val;
3187 if (!strcmp (opt, "cementing")) {
3188 cement_enabled = TRUE;
3191 if (!strcmp (opt, "no-cementing")) {
3192 cement_enabled = FALSE;
3196 if (!strcmp (opt, "precleaning")) {
3197 precleaning_enabled = TRUE;
3200 if (!strcmp (opt, "no-precleaning")) {
3201 precleaning_enabled = FALSE;
3205 if (!strcmp (opt, "dynamic-nursery")) {
3206 if (sgen_minor_collector.is_split)
3207 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.",
3208 "dynamic-nursery not supported with split-nursery.");
3210 dynamic_nursery = TRUE;
3213 if (!strcmp (opt, "no-dynamic-nursery")) {
3214 dynamic_nursery = FALSE;
3218 if (major_collector.handle_gc_param && major_collector.handle_gc_param (opt))
3221 if (sgen_minor_collector.handle_gc_param && sgen_minor_collector.handle_gc_param (opt))
3224 if (sgen_client_handle_gc_param (opt))
3227 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Ignoring.", "Unknown option `%s`.", opt);
3232 fprintf (stderr, "\n%s must be a comma-delimited list of one or more of the following:\n", MONO_GC_PARAMS_NAME);
3233 fprintf (stderr, " max-heap-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
3234 fprintf (stderr, " soft-heap-limit=n (where N is an integer, possibly with a k, m or a g suffix)\n");
3235 fprintf (stderr, " nursery-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
3236 fprintf (stderr, " major=COLLECTOR (where COLLECTOR is `marksweep', `marksweep-conc', `marksweep-par')\n");
3237 fprintf (stderr, " minor=COLLECTOR (where COLLECTOR is `simple' or `split')\n");
3238 fprintf (stderr, " wbarrier=WBARRIER (where WBARRIER is `remset' or `cardtable')\n");
3239 fprintf (stderr, " [no-]cementing\n");
3240 fprintf (stderr, " [no-]dynamic-nursery\n");
3241 if (major_collector.print_gc_param_usage)
3242 major_collector.print_gc_param_usage ();
3243 if (sgen_minor_collector.print_gc_param_usage)
3244 sgen_minor_collector.print_gc_param_usage ();
3245 sgen_client_print_gc_params_usage ();
3246 fprintf (stderr, " Experimental options:\n");
3247 fprintf (stderr, " save-target-ratio=R (where R must be between %.2f - %.2f).\n", SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO);
3248 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);
3249 fprintf (stderr, "\n");
3251 usage_printed = TRUE;
3256 if (major_collector_opt)
3257 g_free (major_collector_opt);
3259 if (minor_collector_opt)
3260 g_free (minor_collector_opt);
3263 g_free (params_opts);
3265 alloc_nursery (dynamic_nursery, min_nursery_size, max_nursery_size);
3267 sgen_pinning_init ();
3268 sgen_cement_init (cement_enabled);
3270 if ((env = g_getenv (MONO_GC_DEBUG_NAME)) || gc_debug_options) {
3271 debug_opts = g_strdup_printf ("%s,%s", gc_debug_options ? gc_debug_options : "", env ? env : "");
3276 gboolean usage_printed = FALSE;
3278 opts = g_strsplit (debug_opts, ",", -1);
3279 for (ptr = opts; ptr && *ptr; ptr ++) {
3281 if (!strcmp (opt, ""))
3283 if (opt [0] >= '0' && opt [0] <= '9') {
3284 gc_debug_level = atoi (opt);
3289 char *rf = g_strdup_printf ("%s.%d", opt, mono_process_current_pid ());
3290 gc_debug_file = fopen (rf, "wb");
3292 gc_debug_file = stderr;
3295 } else if (!strcmp (opt, "print-allowance")) {
3296 debug_print_allowance = TRUE;
3297 } else if (!strcmp (opt, "print-pinning")) {
3298 sgen_pin_stats_enable ();
3299 } else if (!strcmp (opt, "verify-before-allocs")) {
3300 verify_before_allocs = 1;
3301 has_per_allocation_action = TRUE;
3302 } else if (g_str_has_prefix (opt, "max-valloc-size=")) {
3303 size_t max_valloc_size;
3304 char *arg = strchr (opt, '=') + 1;
3305 if (*opt && mono_gc_parse_environment_string_extract_number (arg, &max_valloc_size)) {
3306 mono_valloc_set_limit (max_valloc_size);
3308 sgen_env_var_error (MONO_GC_DEBUG_NAME, NULL, "`max-valloc-size` must be an integer.");
3311 } else if (g_str_has_prefix (opt, "verify-before-allocs=")) {
3312 char *arg = strchr (opt, '=') + 1;
3313 verify_before_allocs = atoi (arg);
3314 has_per_allocation_action = TRUE;
3315 } else if (!strcmp (opt, "collect-before-allocs")) {
3316 collect_before_allocs = 1;
3317 has_per_allocation_action = TRUE;
3318 } else if (g_str_has_prefix (opt, "collect-before-allocs=")) {
3319 char *arg = strchr (opt, '=') + 1;
3320 has_per_allocation_action = TRUE;
3321 collect_before_allocs = atoi (arg);
3322 } else if (!strcmp (opt, "verify-before-collections")) {
3323 whole_heap_check_before_collection = TRUE;
3324 } else if (!strcmp (opt, "check-remset-consistency")) {
3325 remset_consistency_checks = TRUE;
3326 nursery_clear_policy = CLEAR_AT_GC;
3327 } else if (!strcmp (opt, "mod-union-consistency-check")) {
3328 if (!major_collector.is_concurrent) {
3329 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`mod-union-consistency-check` only works with concurrent major collector.");
3332 mod_union_consistency_check = TRUE;
3333 } else if (!strcmp (opt, "check-mark-bits")) {
3334 check_mark_bits_after_major_collection = TRUE;
3335 } else if (!strcmp (opt, "check-nursery-pinned")) {
3336 check_nursery_objects_pinned = TRUE;
3337 } else if (!strcmp (opt, "clear-at-gc")) {
3338 nursery_clear_policy = CLEAR_AT_GC;
3339 } else if (!strcmp (opt, "clear-nursery-at-gc")) {
3340 nursery_clear_policy = CLEAR_AT_GC;
3341 } else if (!strcmp (opt, "clear-at-tlab-creation")) {
3342 nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
3343 } else if (!strcmp (opt, "debug-clear-at-tlab-creation")) {
3344 nursery_clear_policy = CLEAR_AT_TLAB_CREATION_DEBUG;
3345 } else if (!strcmp (opt, "check-scan-starts")) {
3346 do_scan_starts_check = TRUE;
3347 } else if (!strcmp (opt, "verify-nursery-at-minor-gc")) {
3348 do_verify_nursery = TRUE;
3349 } else if (!strcmp (opt, "check-concurrent")) {
3350 if (!major_collector.is_concurrent) {
3351 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`check-concurrent` only works with concurrent major collectors.");
3354 nursery_clear_policy = CLEAR_AT_GC;
3355 do_concurrent_checks = TRUE;
3356 } else if (!strcmp (opt, "dump-nursery-at-minor-gc")) {
3357 do_dump_nursery_content = TRUE;
3358 } else if (!strcmp (opt, "disable-minor")) {
3359 disable_minor_collections = TRUE;
3360 } else if (!strcmp (opt, "disable-major")) {
3361 disable_major_collections = TRUE;
3362 } else if (g_str_has_prefix (opt, "heap-dump=")) {
3363 char *filename = strchr (opt, '=') + 1;
3364 nursery_clear_policy = CLEAR_AT_GC;
3365 sgen_debug_enable_heap_dump (filename);
3366 } else if (g_str_has_prefix (opt, "binary-protocol=")) {
3367 char *filename = strchr (opt, '=') + 1;
3368 char *colon = strrchr (filename, ':');
3371 if (!mono_gc_parse_environment_string_extract_number (colon + 1, &limit)) {
3372 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring limit.", "Binary protocol file size limit must be an integer.");
3377 binary_protocol_init (filename, (long long)limit);
3378 } else if (!strcmp (opt, "nursery-canaries")) {
3379 do_verify_nursery = TRUE;
3380 enable_nursery_canaries = TRUE;
3381 } else if (!sgen_client_handle_gc_debug (opt)) {
3382 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "Unknown option `%s`.", opt);
3387 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);
3388 fprintf (stderr, "Valid <option>s are:\n");
3389 fprintf (stderr, " collect-before-allocs[=<n>]\n");
3390 fprintf (stderr, " verify-before-allocs[=<n>]\n");
3391 fprintf (stderr, " max-valloc-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
3392 fprintf (stderr, " check-remset-consistency\n");
3393 fprintf (stderr, " check-mark-bits\n");
3394 fprintf (stderr, " check-nursery-pinned\n");
3395 fprintf (stderr, " verify-before-collections\n");
3396 fprintf (stderr, " verify-nursery-at-minor-gc\n");
3397 fprintf (stderr, " dump-nursery-at-minor-gc\n");
3398 fprintf (stderr, " disable-minor\n");
3399 fprintf (stderr, " disable-major\n");
3400 fprintf (stderr, " check-concurrent\n");
3401 fprintf (stderr, " clear-[nursery-]at-gc\n");
3402 fprintf (stderr, " clear-at-tlab-creation\n");
3403 fprintf (stderr, " debug-clear-at-tlab-creation\n");
3404 fprintf (stderr, " check-scan-starts\n");
3405 fprintf (stderr, " print-allowance\n");
3406 fprintf (stderr, " print-pinning\n");
3407 fprintf (stderr, " heap-dump=<filename>\n");
3408 fprintf (stderr, " binary-protocol=<filename>[:<file-size-limit>]\n");
3409 fprintf (stderr, " nursery-canaries\n");
3410 sgen_client_print_gc_debug_usage ();
3411 fprintf (stderr, "\n");
3413 usage_printed = TRUE;
3420 g_free (debug_opts);
3422 if (check_mark_bits_after_major_collection)
3423 nursery_clear_policy = CLEAR_AT_GC;
3425 if (major_collector.post_param_init)
3426 major_collector.post_param_init (&major_collector);
3428 if (major_collector.is_concurrent || sgen_minor_collector.is_parallel) {
3429 int num_workers = 1;
3430 if (major_collector.is_parallel || sgen_minor_collector.is_parallel) {
3431 /* FIXME Detect the number of physical cores, instead of logical */
3432 num_workers = mono_cpu_count () / 2;
3433 if (num_workers < 1)
3436 sgen_workers_init (num_workers, (SgenWorkerCallback) major_collector.worker_init_cb);
3439 sgen_memgov_init (max_heap, soft_limit, debug_print_allowance, allowance_ratio, save_target);
3441 memset (&remset, 0, sizeof (remset));
3443 sgen_card_table_init (&remset);
3445 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");
3449 sgen_init_bridge ();
3453 sgen_gc_initialized ()
3455 return gc_initialized > 0;
3459 sgen_get_nursery_clear_policy (void)
3461 return nursery_clear_policy;
3467 mono_coop_mutex_lock (&gc_mutex);
3471 sgen_gc_unlock (void)
3473 mono_coop_mutex_unlock (&gc_mutex);
3477 sgen_major_collector_iterate_live_block_ranges (sgen_cardtable_block_callback callback)
3479 major_collector.iterate_live_block_ranges (callback);
3483 sgen_major_collector_iterate_block_ranges (sgen_cardtable_block_callback callback)
3485 major_collector.iterate_block_ranges (callback);
3489 sgen_get_major_collector (void)
3491 return &major_collector;
3495 sgen_get_minor_collector (void)
3497 return &sgen_minor_collector;
3501 sgen_get_remset (void)
3507 count_cards (long long *major_total, long long *major_marked, long long *los_total, long long *los_marked)
3509 sgen_get_major_collector ()->count_cards (major_total, major_marked);
3510 sgen_los_count_cards (los_total, los_marked);
3513 static gboolean world_is_stopped = FALSE;
3515 /* LOCKING: assumes the GC lock is held */
3517 sgen_stop_world (int generation)
3519 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3521 SGEN_ASSERT (0, !world_is_stopped, "Why are we stopping a stopped world?");
3523 binary_protocol_world_stopping (generation, sgen_timestamp (), (gpointer) (gsize) mono_native_thread_id_get ());
3525 sgen_client_stop_world (generation);
3527 world_is_stopped = TRUE;
3529 if (binary_protocol_is_heavy_enabled ())
3530 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3531 binary_protocol_world_stopped (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3534 /* LOCKING: assumes the GC lock is held */
3536 sgen_restart_world (int generation)
3538 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3541 SGEN_ASSERT (0, world_is_stopped, "Why are we restarting a running world?");
3543 if (binary_protocol_is_heavy_enabled ())
3544 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3545 binary_protocol_world_restarting (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3547 world_is_stopped = FALSE;
3549 sgen_client_restart_world (generation, &stw_time);
3551 binary_protocol_world_restarted (generation, sgen_timestamp ());
3553 if (sgen_client_bridge_need_processing ())
3554 sgen_client_bridge_processing_finish (generation);
3556 sgen_memgov_collection_end (generation, stw_time);
3560 sgen_is_world_stopped (void)
3562 return world_is_stopped;
3566 sgen_check_whole_heap_stw (void)
3568 sgen_stop_world (0);
3569 sgen_clear_nursery_fragments ();
3570 sgen_check_whole_heap (TRUE);
3571 sgen_restart_world (0);
3575 sgen_timestamp (void)
3577 SGEN_TV_DECLARE (timestamp);
3578 SGEN_TV_GETTIME (timestamp);
3579 return SGEN_TV_ELAPSED (sgen_init_timestamp, timestamp);
3582 #endif /* HAVE_SGEN_GC */