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>
182 #include "mono/sgen/sgen-gc.h"
183 #include "mono/sgen/sgen-cardtable.h"
184 #include "mono/sgen/sgen-protocol.h"
185 #include "mono/sgen/sgen-memory-governor.h"
186 #include "mono/sgen/sgen-hash-table.h"
187 #include "mono/sgen/sgen-pinning.h"
188 #include "mono/sgen/sgen-workers.h"
189 #include "mono/sgen/sgen-client.h"
190 #include "mono/sgen/sgen-pointer-queue.h"
191 #include "mono/sgen/gc-internal-agnostic.h"
192 #include "mono/utils/mono-proclib.h"
193 #include "mono/utils/mono-memory-model.h"
194 #include "mono/utils/hazard-pointer.h"
196 #include <mono/utils/memcheck.h>
197 #include <mono/utils/mono-mmap-internals.h>
198 #include <mono/utils/unlocked.h>
200 #undef pthread_create
202 #undef pthread_detach
205 * ######################################################################
206 * ######## Types and constants used by the GC.
207 * ######################################################################
210 /* 0 means not initialized, 1 is initialized, -1 means in progress */
211 static int gc_initialized = 0;
212 /* If set, check if we need to do something every X allocations */
213 gboolean has_per_allocation_action;
214 /* If set, do a heap check every X allocation */
215 guint32 verify_before_allocs = 0;
216 /* If set, do a minor collection before every X allocation */
217 guint32 collect_before_allocs = 0;
218 /* If set, do a whole heap check before each collection */
219 static gboolean whole_heap_check_before_collection = FALSE;
220 /* If set, do a remset consistency check at various opportunities */
221 static gboolean remset_consistency_checks = FALSE;
222 /* If set, do a mod union consistency check before each finishing collection pause */
223 static gboolean mod_union_consistency_check = FALSE;
224 /* If set, check whether mark bits are consistent after major collections */
225 static gboolean check_mark_bits_after_major_collection = FALSE;
226 /* If set, check that all nursery objects are pinned/not pinned, depending on context */
227 static gboolean check_nursery_objects_pinned = FALSE;
228 /* If set, do a few checks when the concurrent collector is used */
229 static gboolean do_concurrent_checks = FALSE;
230 /* If set, do a plausibility check on the scan_starts before and after
232 static gboolean do_scan_starts_check = FALSE;
234 static gboolean disable_minor_collections = FALSE;
235 static gboolean disable_major_collections = FALSE;
236 static gboolean do_verify_nursery = FALSE;
237 static gboolean do_dump_nursery_content = FALSE;
238 static gboolean enable_nursery_canaries = FALSE;
240 static gboolean precleaning_enabled = TRUE;
241 static gboolean dynamic_nursery = FALSE;
242 static size_t min_nursery_size = 0;
243 static size_t max_nursery_size = 0;
245 #ifdef HEAVY_STATISTICS
246 guint64 stat_objects_alloced_degraded = 0;
247 guint64 stat_bytes_alloced_degraded = 0;
249 guint64 stat_copy_object_called_nursery = 0;
250 guint64 stat_objects_copied_nursery = 0;
251 guint64 stat_copy_object_called_major = 0;
252 guint64 stat_objects_copied_major = 0;
254 guint64 stat_scan_object_called_nursery = 0;
255 guint64 stat_scan_object_called_major = 0;
257 guint64 stat_slots_allocated_in_vain;
259 guint64 stat_nursery_copy_object_failed_from_space = 0;
260 guint64 stat_nursery_copy_object_failed_forwarded = 0;
261 guint64 stat_nursery_copy_object_failed_pinned = 0;
262 guint64 stat_nursery_copy_object_failed_to_space = 0;
264 static guint64 stat_wbarrier_add_to_global_remset = 0;
265 static guint64 stat_wbarrier_arrayref_copy = 0;
266 static guint64 stat_wbarrier_generic_store = 0;
267 static guint64 stat_wbarrier_generic_store_atomic = 0;
268 static guint64 stat_wbarrier_set_root = 0;
271 static guint64 stat_pinned_objects = 0;
273 static guint64 time_minor_pre_collection_fragment_clear = 0;
274 static guint64 time_minor_pinning = 0;
275 static guint64 time_minor_scan_remsets = 0;
276 static guint64 time_minor_scan_major_blocks = 0;
277 static guint64 time_minor_scan_los = 0;
278 static guint64 time_minor_scan_pinned = 0;
279 static guint64 time_minor_scan_roots = 0;
280 static guint64 time_minor_finish_gray_stack = 0;
281 static guint64 time_minor_fragment_creation = 0;
283 static guint64 time_major_pre_collection_fragment_clear = 0;
284 static guint64 time_major_pinning = 0;
285 static guint64 time_major_scan_pinned = 0;
286 static guint64 time_major_scan_roots = 0;
287 static guint64 time_major_scan_mod_union_blocks = 0;
288 static guint64 time_major_scan_mod_union_los = 0;
289 static guint64 time_major_finish_gray_stack = 0;
290 static guint64 time_major_free_bigobjs = 0;
291 static guint64 time_major_los_sweep = 0;
292 static guint64 time_major_sweep = 0;
293 static guint64 time_major_fragment_creation = 0;
295 static guint64 time_max = 0;
297 static int sgen_max_pause_time = SGEN_DEFAULT_MAX_PAUSE_TIME;
298 static float sgen_max_pause_margin = SGEN_DEFAULT_MAX_PAUSE_MARGIN;
300 static SGEN_TV_DECLARE (time_major_conc_collection_start);
301 static SGEN_TV_DECLARE (time_major_conc_collection_end);
303 int gc_debug_level = 0;
305 static char* gc_params_options;
306 static char* gc_debug_options;
310 mono_gc_flush_info (void)
312 fflush (gc_debug_file);
316 #define TV_DECLARE SGEN_TV_DECLARE
317 #define TV_GETTIME SGEN_TV_GETTIME
318 #define TV_ELAPSED SGEN_TV_ELAPSED
320 static SGEN_TV_DECLARE (sgen_init_timestamp);
322 NurseryClearPolicy nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
324 #define object_is_forwarded SGEN_OBJECT_IS_FORWARDED
325 #define object_is_pinned SGEN_OBJECT_IS_PINNED
326 #define pin_object SGEN_PIN_OBJECT
328 #define ptr_in_nursery sgen_ptr_in_nursery
330 #define LOAD_VTABLE SGEN_LOAD_VTABLE
333 nursery_canaries_enabled (void)
335 return enable_nursery_canaries;
338 #define safe_object_get_size sgen_safe_object_get_size
343 SGEN_MAJOR_CONCURRENT,
344 SGEN_MAJOR_CONCURRENT_PARALLEL
350 SGEN_MINOR_SIMPLE_PARALLEL,
357 SGEN_MODE_THROUGHPUT,
362 * ######################################################################
363 * ######## Global data.
364 * ######################################################################
366 MonoCoopMutex gc_mutex;
368 #define SCAN_START_SIZE SGEN_SCAN_START_SIZE
370 size_t degraded_mode = 0;
372 static mword bytes_pinned_from_failed_allocation = 0;
374 GCMemSection *nursery_section = NULL;
375 static volatile mword lowest_heap_address = ~(mword)0;
376 static volatile mword highest_heap_address = 0;
378 MonoCoopMutex sgen_interruption_mutex;
380 int current_collection_generation = -1;
381 volatile gboolean concurrent_collection_in_progress = FALSE;
383 /* objects that are ready to be finalized */
384 static SgenPointerQueue fin_ready_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
385 static SgenPointerQueue critical_fin_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
387 /* registered roots: the key to the hash is the root start address */
389 * Different kinds of roots are kept separate to speed up pin_from_roots () for example.
391 SgenHashTable roots_hash [ROOT_TYPE_NUM] = {
392 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
393 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
394 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL)
396 static mword roots_size = 0; /* amount of memory in the root set */
398 /* The size of a TLAB */
399 /* The bigger the value, the less often we have to go to the slow path to allocate a new
400 * one, but the more space is wasted by threads not allocating much memory.
402 * FIXME: Make this self-tuning for each thread.
404 guint32 tlab_size = (1024 * 4);
406 #define MAX_SMALL_OBJ_SIZE SGEN_MAX_SMALL_OBJ_SIZE
408 #define ALLOC_ALIGN SGEN_ALLOC_ALIGN
410 #define ALIGN_UP SGEN_ALIGN_UP
412 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
413 MonoNativeThreadId main_gc_thread = NULL;
416 /*Object was pinned during the current collection*/
417 static mword objects_pinned;
420 * ######################################################################
421 * ######## Macros and function declarations.
422 * ######################################################################
425 /* forward declarations */
426 static void scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx);
428 static void pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx);
429 static void finish_gray_stack (int generation, ScanCopyContext ctx);
432 SgenMajorCollector major_collector;
433 SgenMinorCollector sgen_minor_collector;
435 static SgenRememberedSet remset;
438 * The gray queue a worker job must use. If we're not parallel or
439 * concurrent, we use the main gray queue.
441 static SgenGrayQueue*
442 sgen_workers_get_job_gray_queue (WorkerData *worker_data, SgenGrayQueue *default_gray_queue)
445 return &worker_data->private_gray_queue;
446 SGEN_ASSERT (0, default_gray_queue, "Why don't we have a default gray queue when we're not running in a worker thread?");
447 return default_gray_queue;
451 gray_queue_redirect (SgenGrayQueue *queue)
453 sgen_workers_take_from_queue (current_collection_generation, queue);
457 sgen_scan_area_with_callback (char *start, char *end, IterateObjectCallbackFunc callback, void *data, gboolean allow_flags, gboolean fail_on_canaries)
459 while (start < end) {
463 if (!*(void**)start) {
464 start += sizeof (void*); /* should be ALLOC_ALIGN, really */
469 if (!(obj = (char *)SGEN_OBJECT_IS_FORWARDED (start)))
475 if (!sgen_client_object_is_array_fill ((GCObject*)obj)) {
476 CHECK_CANARY_FOR_OBJECT ((GCObject*)obj, fail_on_canaries);
477 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
478 callback ((GCObject*)obj, size, data);
479 CANARIFY_SIZE (size);
481 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
489 * sgen_add_to_global_remset:
491 * The global remset contains locations which point into newspace after
492 * a minor collection. This can happen if the objects they point to are pinned.
494 * LOCKING: If called from a parallel collector, the global remset
495 * lock must be held. For serial collectors that is not necessary.
498 sgen_add_to_global_remset (gpointer ptr, GCObject *obj)
500 SGEN_ASSERT (5, sgen_ptr_in_nursery (obj), "Target pointer of global remset must be in the nursery");
502 HEAVY_STAT (++stat_wbarrier_add_to_global_remset);
504 if (!major_collector.is_concurrent) {
505 SGEN_ASSERT (5, current_collection_generation != -1, "Global remsets can only be added during collections");
507 if (current_collection_generation == -1)
508 SGEN_ASSERT (5, sgen_concurrent_collection_in_progress (), "Global remsets outside of collection pauses can only be added by the concurrent collector");
511 if (!object_is_pinned (obj))
512 SGEN_ASSERT (5, sgen_minor_collector.is_split || sgen_concurrent_collection_in_progress (), "Non-pinned objects can only remain in nursery if it is a split nursery");
513 else if (sgen_cement_lookup_or_register (obj))
516 remset.record_pointer (ptr);
518 sgen_pin_stats_register_global_remset (obj);
520 SGEN_LOG (8, "Adding global remset for %p", ptr);
521 binary_protocol_global_remset (ptr, obj, (gpointer)SGEN_LOAD_VTABLE (obj));
525 * sgen_drain_gray_stack:
527 * Scan objects in the gray stack until the stack is empty. This should be called
528 * frequently after each object is copied, to achieve better locality and cache
533 sgen_drain_gray_stack (ScanCopyContext ctx)
535 SGEN_ASSERT (0, ctx.ops->drain_gray_stack, "Why do we have a scan/copy context with a missing drain gray stack function?");
537 return ctx.ops->drain_gray_stack (ctx.queue);
541 * Addresses in the pin queue are already sorted. This function finds
542 * the object header for each address and pins the object. The
543 * addresses must be inside the nursery section. The (start of the)
544 * address array is overwritten with the addresses of the actually
545 * pinned objects. Return the number of pinned objects.
548 pin_objects_from_nursery_pin_queue (gboolean do_scan_objects, ScanCopyContext ctx)
550 GCMemSection *section = nursery_section;
551 void **start = sgen_pinning_get_entry (section->pin_queue_first_entry);
552 void **end = sgen_pinning_get_entry (section->pin_queue_last_entry);
553 void *start_nursery = section->data;
554 void *end_nursery = section->end_data;
559 void *pinning_front = start_nursery;
561 void **definitely_pinned = start;
562 ScanObjectFunc scan_func = ctx.ops->scan_object;
563 SgenGrayQueue *queue = ctx.queue;
565 sgen_nursery_allocator_prepare_for_pinning ();
567 while (start < end) {
568 GCObject *obj_to_pin = NULL;
569 size_t obj_to_pin_size = 0;
574 SGEN_ASSERT (0, addr >= start_nursery && addr < end_nursery, "Potential pinning address out of range");
575 SGEN_ASSERT (0, addr >= last, "Pin queue not sorted");
582 SGEN_LOG (5, "Considering pinning addr %p", addr);
583 /* We've already processed everything up to pinning_front. */
584 if (addr < pinning_front) {
590 * Find the closest scan start <= addr. We might search backward in the
591 * scan_starts array because entries might be NULL. In the worst case we
592 * start at start_nursery.
594 idx = ((char*)addr - (char*)section->data) / SCAN_START_SIZE;
595 SGEN_ASSERT (0, idx < section->num_scan_start, "Scan start index out of range");
596 search_start = (void*)section->scan_starts [idx];
597 if (!search_start || search_start > addr) {
600 search_start = section->scan_starts [idx];
601 if (search_start && search_start <= addr)
604 if (!search_start || search_start > addr)
605 search_start = start_nursery;
609 * If the pinning front is closer than the scan start we found, start
610 * searching at the front.
612 if (search_start < pinning_front)
613 search_start = pinning_front;
616 * Now addr should be in an object a short distance from search_start.
618 * search_start must point to zeroed mem or point to an object.
621 size_t obj_size, canarified_obj_size;
624 if (!*(void**)search_start) {
625 search_start = (void*)ALIGN_UP ((mword)search_start + sizeof (gpointer));
626 /* The loop condition makes sure we don't overrun addr. */
630 canarified_obj_size = obj_size = ALIGN_UP (safe_object_get_size ((GCObject*)search_start));
633 * Filler arrays are marked by an invalid sync word. We don't
634 * consider them for pinning. They are not delimited by canaries,
637 if (!sgen_client_object_is_array_fill ((GCObject*)search_start)) {
638 CHECK_CANARY_FOR_OBJECT (search_start, TRUE);
639 CANARIFY_SIZE (canarified_obj_size);
641 if (addr >= search_start && (char*)addr < (char*)search_start + obj_size) {
642 /* This is the object we're looking for. */
643 obj_to_pin = (GCObject*)search_start;
644 obj_to_pin_size = canarified_obj_size;
649 /* Skip to the next object */
650 search_start = (void*)((char*)search_start + canarified_obj_size);
651 } while (search_start <= addr);
653 /* We've searched past the address we were looking for. */
655 pinning_front = search_start;
656 goto next_pin_queue_entry;
660 * We've found an object to pin. It might still be a dummy array, but we
661 * can advance the pinning front in any case.
663 pinning_front = (char*)obj_to_pin + obj_to_pin_size;
666 * If this is a dummy array marking the beginning of a nursery
667 * fragment, we don't pin it.
669 if (sgen_client_object_is_array_fill (obj_to_pin))
670 goto next_pin_queue_entry;
673 * Finally - pin the object!
675 desc = sgen_obj_get_descriptor_safe (obj_to_pin);
676 if (do_scan_objects) {
677 scan_func (obj_to_pin, desc, queue);
679 SGEN_LOG (4, "Pinned object %p, vtable %p (%s), count %d\n",
680 obj_to_pin, *(void**)obj_to_pin, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj_to_pin)), count);
681 binary_protocol_pin (obj_to_pin,
682 (gpointer)LOAD_VTABLE (obj_to_pin),
683 safe_object_get_size (obj_to_pin));
685 pin_object (obj_to_pin);
686 GRAY_OBJECT_ENQUEUE_SERIAL (queue, obj_to_pin, desc);
687 sgen_pin_stats_register_object (obj_to_pin, GENERATION_NURSERY);
688 definitely_pinned [count] = obj_to_pin;
691 if (concurrent_collection_in_progress)
692 sgen_pinning_register_pinned_in_nursery (obj_to_pin);
694 next_pin_queue_entry:
698 sgen_client_nursery_objects_pinned (definitely_pinned, count);
699 stat_pinned_objects += count;
704 pin_objects_in_nursery (gboolean do_scan_objects, ScanCopyContext ctx)
708 if (nursery_section->pin_queue_first_entry == nursery_section->pin_queue_last_entry)
711 reduced_to = pin_objects_from_nursery_pin_queue (do_scan_objects, ctx);
712 nursery_section->pin_queue_last_entry = nursery_section->pin_queue_first_entry + reduced_to;
716 * This function is only ever called (via `collector_pin_object()` in `sgen-copy-object.h`)
717 * when we can't promote an object because we're out of memory.
720 sgen_pin_object (GCObject *object, SgenGrayQueue *queue)
722 SGEN_ASSERT (0, sgen_ptr_in_nursery (object), "We're only supposed to use this for pinning nursery objects when out of memory.");
725 * All pinned objects are assumed to have been staged, so we need to stage as well.
726 * Also, the count of staged objects shows that "late pinning" happened.
728 sgen_pin_stage_ptr (object);
730 SGEN_PIN_OBJECT (object);
731 binary_protocol_pin (object, (gpointer)LOAD_VTABLE (object), safe_object_get_size (object));
734 sgen_pin_stats_register_object (object, GENERATION_NURSERY);
736 GRAY_OBJECT_ENQUEUE_SERIAL (queue, object, sgen_obj_get_descriptor_safe (object));
739 /* Sort the addresses in array in increasing order.
740 * Done using a by-the book heap sort. Which has decent and stable performance, is pretty cache efficient.
743 sgen_sort_addresses (void **array, size_t size)
748 for (i = 1; i < size; ++i) {
751 size_t parent = (child - 1) / 2;
753 if (array [parent] >= array [child])
756 tmp = array [parent];
757 array [parent] = array [child];
764 for (i = size - 1; i > 0; --i) {
767 array [i] = array [0];
773 while (root * 2 + 1 <= end) {
774 size_t child = root * 2 + 1;
776 if (child < end && array [child] < array [child + 1])
778 if (array [root] >= array [child])
782 array [root] = array [child];
791 * Scan the memory between start and end and queue values which could be pointers
792 * to the area between start_nursery and end_nursery for later consideration.
793 * Typically used for thread stacks.
796 sgen_conservatively_pin_objects_from (void **start, void **end, void *start_nursery, void *end_nursery, int pin_type)
800 SGEN_ASSERT (0, ((mword)start & (SIZEOF_VOID_P - 1)) == 0, "Why are we scanning for references in unaligned memory ?");
802 #if defined(VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE) && !defined(_WIN64)
803 VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE (start, (char*)end - (char*)start);
806 while (start < end) {
808 * *start can point to the middle of an object
809 * note: should we handle pointing at the end of an object?
810 * pinning in C# code disallows pointing at the end of an object
811 * but there is some small chance that an optimizing C compiler
812 * may keep the only reference to an object by pointing
813 * at the end of it. We ignore this small chance for now.
814 * Pointers to the end of an object are indistinguishable
815 * from pointers to the start of the next object in memory
816 * so if we allow that we'd need to pin two objects...
817 * We queue the pointer in an array, the
818 * array will then be sorted and uniqued. This way
819 * we can coalesce several pinning pointers and it should
820 * be faster since we'd do a memory scan with increasing
821 * addresses. Note: we can align the address to the allocation
822 * alignment, so the unique process is more effective.
824 mword addr = (mword)*start;
825 addr &= ~(ALLOC_ALIGN - 1);
826 if (addr >= (mword)start_nursery && addr < (mword)end_nursery) {
827 SGEN_LOG (6, "Pinning address %p from %p", (void*)addr, start);
828 sgen_pin_stage_ptr ((void*)addr);
829 binary_protocol_pin_stage (start, (void*)addr);
830 sgen_pin_stats_register_address ((char*)addr, pin_type);
836 SGEN_LOG (7, "found %d potential pinned heap pointers", count);
840 * The first thing we do in a collection is to identify pinned objects.
841 * This function considers all the areas of memory that need to be
842 * conservatively scanned.
845 pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx)
849 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);
850 /* objects pinned from the API are inside these roots */
851 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_PINNED], void **, start_root, RootRecord *, root) {
852 SGEN_LOG (6, "Pinned roots %p-%p", start_root, root->end_root);
853 sgen_conservatively_pin_objects_from (start_root, (void**)root->end_root, start_nursery, end_nursery, PIN_TYPE_OTHER);
854 } SGEN_HASH_TABLE_FOREACH_END;
855 /* now deal with the thread stacks
856 * in the future we should be able to conservatively scan only:
857 * *) the cpu registers
858 * *) the unmanaged stack frames
859 * *) the _last_ managed stack frame
860 * *) pointers slots in managed frames
862 sgen_client_scan_thread_data (start_nursery, end_nursery, FALSE, ctx);
866 single_arg_user_copy_or_mark (GCObject **obj, void *gc_data)
868 ScanCopyContext *ctx = (ScanCopyContext *)gc_data;
869 ctx->ops->copy_or_mark_object (obj, ctx->queue);
873 * The memory area from start_root to end_root contains pointers to objects.
874 * Their position is precisely described by @desc (this means that the pointer
875 * can be either NULL or the pointer to the start of an object).
876 * This functions copies them to to_space updates them.
878 * This function is not thread-safe!
881 precisely_scan_objects_from (void** start_root, void** end_root, char* n_start, char *n_end, SgenDescriptor desc, ScanCopyContext ctx)
883 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
884 ScanPtrFieldFunc scan_field_func = ctx.ops->scan_ptr_field;
885 SgenGrayQueue *queue = ctx.queue;
887 switch (desc & ROOT_DESC_TYPE_MASK) {
888 case ROOT_DESC_BITMAP:
889 desc >>= ROOT_DESC_TYPE_SHIFT;
891 if ((desc & 1) && *start_root) {
892 copy_func ((GCObject**)start_root, queue);
893 SGEN_LOG (9, "Overwrote root at %p with %p", start_root, *start_root);
899 case ROOT_DESC_COMPLEX: {
900 gsize *bitmap_data = (gsize *)sgen_get_complex_descriptor_bitmap (desc);
901 gsize bwords = (*bitmap_data) - 1;
902 void **start_run = start_root;
904 while (bwords-- > 0) {
905 gsize bmap = *bitmap_data++;
906 void **objptr = start_run;
908 if ((bmap & 1) && *objptr) {
909 copy_func ((GCObject**)objptr, queue);
910 SGEN_LOG (9, "Overwrote root at %p with %p", objptr, *objptr);
915 start_run += GC_BITS_PER_WORD;
919 case ROOT_DESC_VECTOR: {
922 for (p = start_root; p < end_root; p++) {
924 scan_field_func (NULL, (GCObject**)p, queue);
928 case ROOT_DESC_USER: {
929 SgenUserRootMarkFunc marker = sgen_get_user_descriptor_func (desc);
930 marker (start_root, single_arg_user_copy_or_mark, &ctx);
933 case ROOT_DESC_RUN_LEN:
934 g_assert_not_reached ();
936 g_assert_not_reached ();
941 reset_heap_boundaries (void)
943 lowest_heap_address = ~(mword)0;
944 highest_heap_address = 0;
948 sgen_update_heap_boundaries (mword low, mword high)
953 old = lowest_heap_address;
956 } while (SGEN_CAS_PTR ((gpointer*)&lowest_heap_address, (gpointer)low, (gpointer)old) != (gpointer)old);
959 old = highest_heap_address;
962 } while (SGEN_CAS_PTR ((gpointer*)&highest_heap_address, (gpointer)high, (gpointer)old) != (gpointer)old);
966 * Allocate and setup the data structures needed to be able to allocate objects
967 * in the nursery. The nursery is stored in nursery_section.
970 alloc_nursery (gboolean dynamic, size_t min_size, size_t max_size)
977 min_size = SGEN_DEFAULT_NURSERY_MIN_SIZE;
979 max_size = SGEN_DEFAULT_NURSERY_MAX_SIZE;
981 SGEN_ASSERT (0, min_size == max_size, "We can't have nursery ranges for static configuration.");
983 min_size = max_size = SGEN_DEFAULT_NURSERY_SIZE;
986 SGEN_ASSERT (0, !nursery_section, "Why are we allocating the nursery twice?");
987 SGEN_LOG (2, "Allocating nursery size: %zu, initial %zu", max_size, min_size);
989 /* FIXME: handle OOM */
990 nursery_section = (GCMemSection *)sgen_alloc_internal (INTERNAL_MEM_SECTION);
992 /* If there isn't enough space even for the nursery we should simply abort. */
993 g_assert (sgen_memgov_try_alloc_space (max_size, SPACE_NURSERY));
996 * The nursery section range represents the memory section where objects
997 * can be found. This is used when iterating for objects in the nursery,
998 * pinning etc. sgen_nursery_max_size represents the total allocated space
999 * for the nursery. sgen_nursery_size represents the current size of the
1000 * nursery and it is used for allocation limits, heuristics etc. The
1001 * nursery section is not always identical to the current nursery size
1002 * because it can contain pinned objects from when the nursery was larger.
1004 * sgen_nursery_size <= nursery_section size <= sgen_nursery_max_size
1006 data = (char *)major_collector.alloc_heap (max_size, max_size);
1007 sgen_update_heap_boundaries ((mword)data, (mword)(data + max_size));
1008 nursery_section->data = data;
1009 nursery_section->end_data = data + min_size;
1010 scan_starts = (max_size + SCAN_START_SIZE - 1) / SCAN_START_SIZE;
1011 nursery_section->scan_starts = (char **)sgen_alloc_internal_dynamic (sizeof (char*) * scan_starts, INTERNAL_MEM_SCAN_STARTS, TRUE);
1012 nursery_section->num_scan_start = scan_starts;
1014 sgen_nursery_allocator_set_nursery_bounds (data, min_size, max_size);
1018 mono_gc_get_logfile (void)
1020 return gc_debug_file;
1024 mono_gc_params_set (const char* options)
1026 if (gc_params_options)
1027 g_free (gc_params_options);
1029 gc_params_options = g_strdup (options);
1033 mono_gc_debug_set (const char* options)
1035 if (gc_debug_options)
1036 g_free (gc_debug_options);
1038 gc_debug_options = g_strdup (options);
1042 scan_finalizer_entries (SgenPointerQueue *fin_queue, ScanCopyContext ctx)
1044 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
1045 SgenGrayQueue *queue = ctx.queue;
1048 for (i = 0; i < fin_queue->next_slot; ++i) {
1049 GCObject *obj = (GCObject *)fin_queue->data [i];
1052 SGEN_LOG (5, "Scan of fin ready object: %p (%s)\n", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
1053 copy_func ((GCObject**)&fin_queue->data [i], queue);
1058 generation_name (int generation)
1060 switch (generation) {
1061 case GENERATION_NURSERY: return "nursery";
1062 case GENERATION_OLD: return "old";
1063 default: g_assert_not_reached ();
1068 sgen_generation_name (int generation)
1070 return generation_name (generation);
1074 finish_gray_stack (int generation, ScanCopyContext ctx)
1078 int done_with_ephemerons, ephemeron_rounds = 0;
1079 char *start_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_start () : NULL;
1080 char *end_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_end () : (char*)-1;
1081 SgenGrayQueue *queue = ctx.queue;
1083 binary_protocol_finish_gray_stack_start (sgen_timestamp (), generation);
1085 * We copied all the reachable objects. Now it's the time to copy
1086 * the objects that were not referenced by the roots, but by the copied objects.
1087 * we built a stack of objects pointed to by gray_start: they are
1088 * additional roots and we may add more items as we go.
1089 * We loop until gray_start == gray_objects which means no more objects have
1090 * been added. Note this is iterative: no recursion is involved.
1091 * We need to walk the LO list as well in search of marked big objects
1092 * (use a flag since this is needed only on major collections). We need to loop
1093 * here as well, so keep a counter of marked LO (increasing it in copy_object).
1094 * To achieve better cache locality and cache usage, we drain the gray stack
1095 * frequently, after each object is copied, and just finish the work here.
1097 sgen_drain_gray_stack (ctx);
1099 SGEN_LOG (2, "%s generation done", generation_name (generation));
1102 Reset bridge data, we might have lingering data from a previous collection if this is a major
1103 collection trigged by minor overflow.
1105 We must reset the gathered bridges since their original block might be evacuated due to major
1106 fragmentation in the meanwhile and the bridge code should not have to deal with that.
1108 if (sgen_client_bridge_need_processing ())
1109 sgen_client_bridge_reset_data ();
1112 * Mark all strong toggleref objects. This must be done before we walk ephemerons or finalizers
1113 * to ensure they see the full set of live objects.
1115 sgen_client_mark_togglerefs (start_addr, end_addr, ctx);
1118 * Walk the ephemeron tables marking all values with reachable keys. This must be completely done
1119 * before processing finalizable objects and non-tracking weak links to avoid finalizing/clearing
1120 * objects that are in fact reachable.
1122 done_with_ephemerons = 0;
1124 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1125 sgen_drain_gray_stack (ctx);
1127 } while (!done_with_ephemerons);
1129 if (sgen_client_bridge_need_processing ()) {
1130 /*Make sure the gray stack is empty before we process bridge objects so we get liveness right*/
1131 sgen_drain_gray_stack (ctx);
1132 sgen_collect_bridge_objects (generation, ctx);
1133 if (generation == GENERATION_OLD)
1134 sgen_collect_bridge_objects (GENERATION_NURSERY, ctx);
1137 Do the first bridge step here, as the collector liveness state will become useless after that.
1139 An important optimization is to only proccess the possibly dead part of the object graph and skip
1140 over all live objects as we transitively know everything they point must be alive too.
1142 The above invariant is completely wrong if we let the gray queue be drained and mark/copy everything.
1144 This has the unfortunate side effect of making overflow collections perform the first step twice, but
1145 given we now have heuristics that perform major GC in anticipation of minor overflows this should not
1148 sgen_client_bridge_processing_stw_step ();
1152 Make sure we drain the gray stack before processing disappearing links and finalizers.
1153 If we don't make sure it is empty we might wrongly see a live object as dead.
1155 sgen_drain_gray_stack (ctx);
1158 We must clear weak links that don't track resurrection before processing object ready for
1159 finalization so they can be cleared before that.
1161 sgen_null_link_in_range (generation, ctx, FALSE);
1162 if (generation == GENERATION_OLD)
1163 sgen_null_link_in_range (GENERATION_NURSERY, ctx, FALSE);
1166 /* walk the finalization queue and move also the objects that need to be
1167 * finalized: use the finalized objects as new roots so the objects they depend
1168 * on are also not reclaimed. As with the roots above, only objects in the nursery
1169 * are marked/copied.
1171 sgen_finalize_in_range (generation, ctx);
1172 if (generation == GENERATION_OLD)
1173 sgen_finalize_in_range (GENERATION_NURSERY, ctx);
1174 /* drain the new stack that might have been created */
1175 SGEN_LOG (6, "Precise scan of gray area post fin");
1176 sgen_drain_gray_stack (ctx);
1179 * This must be done again after processing finalizable objects since CWL slots are cleared only after the key is finalized.
1181 done_with_ephemerons = 0;
1183 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1184 sgen_drain_gray_stack (ctx);
1186 } while (!done_with_ephemerons);
1188 sgen_client_clear_unreachable_ephemerons (ctx);
1191 * We clear togglerefs only after all possible chances of revival are done.
1192 * This is semantically more inline with what users expect and it allows for
1193 * user finalizers to correctly interact with TR objects.
1195 sgen_client_clear_togglerefs (start_addr, end_addr, ctx);
1198 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);
1201 * handle disappearing links
1202 * Note we do this after checking the finalization queue because if an object
1203 * survives (at least long enough to be finalized) we don't clear the link.
1204 * This also deals with a possible issue with the monitor reclamation: with the Boehm
1205 * GC a finalized object my lose the monitor because it is cleared before the finalizer is
1208 g_assert (sgen_gray_object_queue_is_empty (queue));
1210 sgen_null_link_in_range (generation, ctx, TRUE);
1211 if (generation == GENERATION_OLD)
1212 sgen_null_link_in_range (GENERATION_NURSERY, ctx, TRUE);
1213 if (sgen_gray_object_queue_is_empty (queue))
1215 sgen_drain_gray_stack (ctx);
1218 g_assert (sgen_gray_object_queue_is_empty (queue));
1220 binary_protocol_finish_gray_stack_end (sgen_timestamp (), generation);
1224 sgen_check_section_scan_starts (GCMemSection *section)
1227 for (i = 0; i < section->num_scan_start; ++i) {
1228 if (section->scan_starts [i]) {
1229 mword size = safe_object_get_size ((GCObject*) section->scan_starts [i]);
1230 SGEN_ASSERT (0, size >= SGEN_CLIENT_MINIMUM_OBJECT_SIZE && size <= MAX_SMALL_OBJ_SIZE, "Weird object size at scan starts.");
1236 check_scan_starts (void)
1238 if (!do_scan_starts_check)
1240 sgen_check_section_scan_starts (nursery_section);
1241 major_collector.check_scan_starts ();
1245 scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx)
1249 SGEN_HASH_TABLE_FOREACH (&roots_hash [root_type], void **, start_root, RootRecord *, root) {
1250 SGEN_LOG (6, "Precise root scan %p-%p (desc: %p)", start_root, root->end_root, (void*)root->root_desc);
1251 precisely_scan_objects_from (start_root, (void**)root->end_root, addr_start, addr_end, root->root_desc, ctx);
1252 } SGEN_HASH_TABLE_FOREACH_END;
1258 static gboolean inited = FALSE;
1263 mono_counters_register ("Collection max time", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME | MONO_COUNTER_MONOTONIC, &time_max);
1265 mono_counters_register ("Minor fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pre_collection_fragment_clear);
1266 mono_counters_register ("Minor pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pinning);
1267 mono_counters_register ("Minor scan remembered set", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_remsets);
1268 mono_counters_register ("Minor scan major blocks", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_major_blocks);
1269 mono_counters_register ("Minor scan los", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_los);
1270 mono_counters_register ("Minor scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_pinned);
1271 mono_counters_register ("Minor scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_roots);
1272 mono_counters_register ("Minor fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_fragment_creation);
1274 mono_counters_register ("Major fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pre_collection_fragment_clear);
1275 mono_counters_register ("Major pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pinning);
1276 mono_counters_register ("Major scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_pinned);
1277 mono_counters_register ("Major scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_roots);
1278 mono_counters_register ("Major scan mod union blocks", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_mod_union_blocks);
1279 mono_counters_register ("Major scan mod union los", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_mod_union_los);
1280 mono_counters_register ("Major finish gray stack", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_finish_gray_stack);
1281 mono_counters_register ("Major free big objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_free_bigobjs);
1282 mono_counters_register ("Major LOS sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_los_sweep);
1283 mono_counters_register ("Major sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_sweep);
1284 mono_counters_register ("Major fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_fragment_creation);
1286 mono_counters_register ("Number of pinned objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_pinned_objects);
1288 #ifdef HEAVY_STATISTICS
1289 mono_counters_register ("WBarrier remember pointer", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_add_to_global_remset);
1290 mono_counters_register ("WBarrier arrayref copy", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_arrayref_copy);
1291 mono_counters_register ("WBarrier generic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store);
1292 mono_counters_register ("WBarrier generic atomic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store_atomic);
1293 mono_counters_register ("WBarrier set root", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_set_root);
1295 mono_counters_register ("# objects allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_alloced_degraded);
1296 mono_counters_register ("bytes allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_bytes_alloced_degraded);
1298 mono_counters_register ("# copy_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_nursery);
1299 mono_counters_register ("# objects copied (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_nursery);
1300 mono_counters_register ("# copy_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_major);
1301 mono_counters_register ("# objects copied (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_major);
1303 mono_counters_register ("# scan_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_nursery);
1304 mono_counters_register ("# scan_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_major);
1306 mono_counters_register ("Slots allocated in vain", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_slots_allocated_in_vain);
1308 mono_counters_register ("# nursery copy_object() failed from space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_from_space);
1309 mono_counters_register ("# nursery copy_object() failed forwarded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_forwarded);
1310 mono_counters_register ("# nursery copy_object() failed pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_pinned);
1311 mono_counters_register ("# nursery copy_object() failed to space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_to_space);
1313 sgen_nursery_allocator_init_heavy_stats ();
1321 reset_pinned_from_failed_allocation (void)
1323 bytes_pinned_from_failed_allocation = 0;
1327 sgen_set_pinned_from_failed_allocation (mword objsize)
1329 bytes_pinned_from_failed_allocation += objsize;
1333 sgen_collection_is_concurrent (void)
1335 switch (current_collection_generation) {
1336 case GENERATION_NURSERY:
1338 case GENERATION_OLD:
1339 return concurrent_collection_in_progress;
1341 g_error ("Invalid current generation %d", current_collection_generation);
1347 sgen_concurrent_collection_in_progress (void)
1349 return concurrent_collection_in_progress;
1353 SgenThreadPoolJob job;
1354 SgenObjectOperations *ops;
1355 SgenGrayQueue *gc_thread_gray_queue;
1360 int job_index, job_split_count;
1364 static ScanCopyContext
1365 scan_copy_context_for_scan_job (void *worker_data_untyped, ScanJob *job)
1367 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1371 * For jobs enqueued on workers we set the ops at job runtime in order
1372 * to be able to profit from on the fly optimized object ops or other
1373 * object ops changes, like forced concurrent finish.
1375 SGEN_ASSERT (0, sgen_workers_is_worker_thread (mono_native_thread_id_get ()), "We need a context for the scan job");
1376 job->ops = sgen_workers_get_idle_func_object_ops (worker_data);
1379 return CONTEXT_FROM_OBJECT_OPERATIONS (job->ops, sgen_workers_get_job_gray_queue (worker_data, job->gc_thread_gray_queue));
1387 } ScanFromRegisteredRootsJob;
1390 job_scan_from_registered_roots (void *worker_data_untyped, SgenThreadPoolJob *job)
1392 ScanFromRegisteredRootsJob *job_data = (ScanFromRegisteredRootsJob*)job;
1393 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1395 scan_from_registered_roots (job_data->heap_start, job_data->heap_end, job_data->root_type, ctx);
1402 } ScanThreadDataJob;
1405 job_scan_thread_data (void *worker_data_untyped, SgenThreadPoolJob *job)
1407 ScanThreadDataJob *job_data = (ScanThreadDataJob*)job;
1408 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1410 sgen_client_scan_thread_data (job_data->heap_start, job_data->heap_end, TRUE, ctx);
1415 SgenPointerQueue *queue;
1416 } ScanFinalizerEntriesJob;
1419 job_scan_finalizer_entries (void *worker_data_untyped, SgenThreadPoolJob *job)
1421 ScanFinalizerEntriesJob *job_data = (ScanFinalizerEntriesJob*)job;
1422 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1424 scan_finalizer_entries (job_data->queue, ctx);
1428 job_scan_wbroots (void *worker_data_untyped, SgenThreadPoolJob *job)
1430 ScanJob *job_data = (ScanJob*)job;
1431 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, job_data);
1433 sgen_wbroots_scan_card_table (ctx);
1437 job_scan_major_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1439 SGEN_TV_DECLARE (atv);
1440 SGEN_TV_DECLARE (btv);
1441 ParallelScanJob *job_data = (ParallelScanJob*)job;
1442 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1444 SGEN_TV_GETTIME (atv);
1445 major_collector.scan_card_table (CARDTABLE_SCAN_GLOBAL, ctx, job_data->job_index, job_data->job_split_count, job_data->data);
1446 SGEN_TV_GETTIME (btv);
1447 time_minor_scan_major_blocks += SGEN_TV_ELAPSED (atv, btv);
1449 if (worker_data_untyped)
1450 ((WorkerData*)worker_data_untyped)->major_scan_time += SGEN_TV_ELAPSED (atv, btv);
1454 job_scan_los_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1456 SGEN_TV_DECLARE (atv);
1457 SGEN_TV_DECLARE (btv);
1458 ParallelScanJob *job_data = (ParallelScanJob*)job;
1459 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1461 SGEN_TV_GETTIME (atv);
1462 sgen_los_scan_card_table (CARDTABLE_SCAN_GLOBAL, ctx, job_data->job_index, job_data->job_split_count);
1463 SGEN_TV_GETTIME (btv);
1464 time_minor_scan_los += SGEN_TV_ELAPSED (atv, btv);
1466 if (worker_data_untyped)
1467 ((WorkerData*)worker_data_untyped)->los_scan_time += SGEN_TV_ELAPSED (atv, btv);
1471 job_scan_major_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1473 SGEN_TV_DECLARE (atv);
1474 SGEN_TV_DECLARE (btv);
1475 ParallelScanJob *job_data = (ParallelScanJob*)job;
1476 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1478 g_assert (concurrent_collection_in_progress);
1479 SGEN_TV_GETTIME (atv);
1480 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx, job_data->job_index, job_data->job_split_count, job_data->data);
1481 SGEN_TV_GETTIME (btv);
1482 time_major_scan_mod_union_blocks += SGEN_TV_ELAPSED (atv, btv);
1484 if (worker_data_untyped)
1485 ((WorkerData*)worker_data_untyped)->major_scan_time += SGEN_TV_ELAPSED (atv, btv);
1489 job_scan_los_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1491 SGEN_TV_DECLARE (atv);
1492 SGEN_TV_DECLARE (btv);
1493 ParallelScanJob *job_data = (ParallelScanJob*)job;
1494 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1496 g_assert (concurrent_collection_in_progress);
1497 SGEN_TV_GETTIME (atv);
1498 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx, job_data->job_index, job_data->job_split_count);
1499 SGEN_TV_GETTIME (btv);
1500 time_major_scan_mod_union_los += SGEN_TV_ELAPSED (atv, btv);
1502 if (worker_data_untyped)
1503 ((WorkerData*)worker_data_untyped)->los_scan_time += SGEN_TV_ELAPSED (atv, btv);
1507 job_major_mod_union_preclean (void *worker_data_untyped, SgenThreadPoolJob *job)
1509 SGEN_TV_DECLARE (atv);
1510 SGEN_TV_DECLARE (btv);
1511 ParallelScanJob *job_data = (ParallelScanJob*)job;
1512 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1514 g_assert (concurrent_collection_in_progress);
1515 SGEN_TV_GETTIME (atv);
1516 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx, job_data->job_index, job_data->job_split_count, job_data->data);
1517 SGEN_TV_GETTIME (btv);
1519 g_assert (worker_data_untyped);
1520 ((WorkerData*)worker_data_untyped)->major_scan_time += SGEN_TV_ELAPSED (atv, btv);
1524 job_los_mod_union_preclean (void *worker_data_untyped, SgenThreadPoolJob *job)
1526 SGEN_TV_DECLARE (atv);
1527 SGEN_TV_DECLARE (btv);
1528 ParallelScanJob *job_data = (ParallelScanJob*)job;
1529 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1531 g_assert (concurrent_collection_in_progress);
1532 SGEN_TV_GETTIME (atv);
1533 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx, job_data->job_index, job_data->job_split_count);
1534 SGEN_TV_GETTIME (btv);
1536 g_assert (worker_data_untyped);
1537 ((WorkerData*)worker_data_untyped)->los_scan_time += SGEN_TV_ELAPSED (atv, btv);
1541 job_scan_last_pinned (void *worker_data_untyped, SgenThreadPoolJob *job)
1543 ScanJob *job_data = (ScanJob*)job;
1544 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, job_data);
1546 g_assert (concurrent_collection_in_progress);
1548 sgen_scan_pin_queue_objects (ctx);
1552 workers_finish_callback (void)
1554 ParallelScanJob *psj;
1556 size_t num_major_sections = major_collector.get_num_major_sections ();
1557 int split_count = sgen_workers_get_job_split_count (GENERATION_OLD);
1559 /* Mod union preclean jobs */
1560 for (i = 0; i < split_count; i++) {
1561 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("preclean major mod union cardtable", job_major_mod_union_preclean, sizeof (ParallelScanJob));
1562 psj->scan_job.gc_thread_gray_queue = NULL;
1564 psj->job_split_count = split_count;
1565 psj->data = num_major_sections / split_count;
1566 sgen_workers_enqueue_job (GENERATION_OLD, &psj->scan_job.job, TRUE);
1569 for (i = 0; i < split_count; i++) {
1570 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("preclean los mod union cardtable", job_los_mod_union_preclean, sizeof (ParallelScanJob));
1571 psj->scan_job.gc_thread_gray_queue = NULL;
1573 psj->job_split_count = split_count;
1574 sgen_workers_enqueue_job (GENERATION_OLD, &psj->scan_job.job, TRUE);
1577 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan last pinned", job_scan_last_pinned, sizeof (ScanJob));
1578 sj->gc_thread_gray_queue = NULL;
1579 sgen_workers_enqueue_job (GENERATION_OLD, &sj->job, TRUE);
1583 init_gray_queue (SgenGrayQueue *gc_thread_gray_queue)
1585 sgen_gray_object_queue_init (gc_thread_gray_queue, NULL, TRUE);
1589 enqueue_scan_remembered_set_jobs (SgenGrayQueue *gc_thread_gray_queue, SgenObjectOperations *ops, gboolean enqueue)
1591 int i, split_count = sgen_workers_get_job_split_count (GENERATION_NURSERY);
1592 size_t num_major_sections = major_collector.get_num_major_sections ();
1595 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan wbroots", job_scan_wbroots, sizeof (ScanJob));
1597 sj->gc_thread_gray_queue = gc_thread_gray_queue;
1598 sgen_workers_enqueue_job (GENERATION_NURSERY, &sj->job, enqueue);
1600 for (i = 0; i < split_count; i++) {
1601 ParallelScanJob *psj;
1603 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("scan major remsets", job_scan_major_card_table, sizeof (ParallelScanJob));
1604 psj->scan_job.ops = ops;
1605 psj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1607 psj->job_split_count = split_count;
1608 psj->data = num_major_sections / split_count;
1609 sgen_workers_enqueue_job (GENERATION_NURSERY, &psj->scan_job.job, enqueue);
1611 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("scan LOS remsets", job_scan_los_card_table, sizeof (ParallelScanJob));
1612 psj->scan_job.ops = ops;
1613 psj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1615 psj->job_split_count = split_count;
1616 sgen_workers_enqueue_job (GENERATION_NURSERY, &psj->scan_job.job, enqueue);
1621 enqueue_scan_from_roots_jobs (SgenGrayQueue *gc_thread_gray_queue, char *heap_start, char *heap_end, SgenObjectOperations *ops, gboolean enqueue)
1623 ScanFromRegisteredRootsJob *scrrj;
1624 ScanThreadDataJob *stdj;
1625 ScanFinalizerEntriesJob *sfej;
1627 /* registered roots, this includes static fields */
1629 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots normal", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1630 scrrj->scan_job.ops = ops;
1631 scrrj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1632 scrrj->heap_start = heap_start;
1633 scrrj->heap_end = heap_end;
1634 scrrj->root_type = ROOT_TYPE_NORMAL;
1635 sgen_workers_enqueue_job (current_collection_generation, &scrrj->scan_job.job, enqueue);
1637 if (current_collection_generation == GENERATION_OLD) {
1638 /* During minors we scan the cardtable for these roots instead */
1639 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots wbarrier", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1640 scrrj->scan_job.ops = ops;
1641 scrrj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1642 scrrj->heap_start = heap_start;
1643 scrrj->heap_end = heap_end;
1644 scrrj->root_type = ROOT_TYPE_WBARRIER;
1645 sgen_workers_enqueue_job (current_collection_generation, &scrrj->scan_job.job, enqueue);
1650 stdj = (ScanThreadDataJob*)sgen_thread_pool_job_alloc ("scan thread data", job_scan_thread_data, sizeof (ScanThreadDataJob));
1651 stdj->scan_job.ops = ops;
1652 stdj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1653 stdj->heap_start = heap_start;
1654 stdj->heap_end = heap_end;
1655 sgen_workers_enqueue_job (current_collection_generation, &stdj->scan_job.job, enqueue);
1657 /* Scan the list of objects ready for finalization. */
1659 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1660 sfej->scan_job.ops = ops;
1661 sfej->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1662 sfej->queue = &fin_ready_queue;
1663 sgen_workers_enqueue_job (current_collection_generation, &sfej->scan_job.job, enqueue);
1665 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan critical finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1666 sfej->scan_job.ops = ops;
1667 sfej->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1668 sfej->queue = &critical_fin_queue;
1669 sgen_workers_enqueue_job (current_collection_generation, &sfej->scan_job.job, enqueue);
1673 * Perform a nursery collection.
1675 * Return whether any objects were late-pinned due to being out of memory.
1678 collect_nursery (const char *reason, gboolean is_overflow, SgenGrayQueue *unpin_queue)
1680 gboolean needs_major, is_parallel = FALSE;
1681 mword fragment_total;
1682 SgenGrayQueue gc_thread_gray_queue;
1683 SgenObjectOperations *object_ops_nopar, *object_ops_par = NULL;
1684 ScanCopyContext ctx;
1687 SGEN_TV_DECLARE (last_minor_collection_start_tv);
1688 SGEN_TV_DECLARE (last_minor_collection_end_tv);
1689 guint64 major_scan_start = time_minor_scan_major_blocks;
1690 guint64 los_scan_start = time_minor_scan_los;
1691 guint64 finish_gray_start = time_minor_finish_gray_stack;
1693 if (disable_minor_collections)
1696 TV_GETTIME (last_minor_collection_start_tv);
1697 atv = last_minor_collection_start_tv;
1699 binary_protocol_collection_begin (InterlockedRead (&gc_stats.minor_gc_count), GENERATION_NURSERY);
1701 object_ops_nopar = sgen_concurrent_collection_in_progress ()
1702 ? &sgen_minor_collector.serial_ops_with_concurrent_major
1703 : &sgen_minor_collector.serial_ops;
1704 if (sgen_minor_collector.is_parallel && sgen_nursery_size >= SGEN_PARALLEL_MINOR_MIN_NURSERY_SIZE) {
1705 object_ops_par = sgen_concurrent_collection_in_progress ()
1706 ? &sgen_minor_collector.parallel_ops_with_concurrent_major
1707 : &sgen_minor_collector.parallel_ops;
1711 if (do_verify_nursery || do_dump_nursery_content)
1712 sgen_debug_verify_nursery (do_dump_nursery_content);
1714 current_collection_generation = GENERATION_NURSERY;
1716 SGEN_ASSERT (0, !sgen_collection_is_concurrent (), "Why is the nursery collection concurrent?");
1718 reset_pinned_from_failed_allocation ();
1720 check_scan_starts ();
1722 sgen_nursery_alloc_prepare_for_minor ();
1727 SGEN_LOG (1, "Start nursery collection %" G_GINT32_FORMAT " %p-%p, size: %d", InterlockedRead (&gc_stats.minor_gc_count), nursery_section->data, nursery_section->end_data, (int)(nursery_section->end_data - nursery_section->data));
1729 /* world must be stopped already */
1731 time_minor_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1733 sgen_client_pre_collection_checks ();
1735 major_collector.start_nursery_collection ();
1737 sgen_memgov_minor_collection_start ();
1739 init_gray_queue (&gc_thread_gray_queue);
1740 ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops_nopar, &gc_thread_gray_queue);
1742 InterlockedIncrement (&gc_stats.minor_gc_count);
1744 sgen_process_fin_stage_entries ();
1746 /* pin from pinned handles */
1747 sgen_init_pinning ();
1748 if (concurrent_collection_in_progress)
1749 sgen_init_pinning_for_conc ();
1750 sgen_client_binary_protocol_mark_start (GENERATION_NURSERY);
1751 pin_from_roots (nursery_section->data, nursery_section->end_data, ctx);
1752 /* pin cemented objects */
1753 sgen_pin_cemented_objects ();
1754 /* identify pinned objects */
1755 sgen_optimize_pin_queue ();
1756 sgen_pinning_setup_section (nursery_section);
1758 pin_objects_in_nursery (FALSE, ctx);
1759 sgen_pinning_trim_queue_to_section (nursery_section);
1760 if (concurrent_collection_in_progress)
1761 sgen_finish_pinning_for_conc ();
1763 if (remset_consistency_checks)
1764 sgen_check_remset_consistency ();
1766 if (whole_heap_check_before_collection) {
1767 sgen_clear_nursery_fragments ();
1768 sgen_check_whole_heap (FALSE);
1772 time_minor_pinning += TV_ELAPSED (btv, atv);
1773 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), (long long)TV_ELAPSED (btv, atv));
1774 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1776 remset.start_scan_remsets ();
1778 enqueue_scan_remembered_set_jobs (&gc_thread_gray_queue, is_parallel ? NULL : object_ops_nopar, is_parallel);
1780 /* we don't have complete write barrier yet, so we scan all the old generation sections */
1782 time_minor_scan_remsets += TV_ELAPSED (atv, btv);
1783 SGEN_LOG (2, "Old generation scan: %lld usecs", (long long)TV_ELAPSED (atv, btv));
1785 sgen_pin_stats_report ();
1787 /* FIXME: Why do we do this at this specific, seemingly random, point? */
1788 sgen_client_collecting_minor (&fin_ready_queue, &critical_fin_queue);
1791 time_minor_scan_pinned += TV_ELAPSED (btv, atv);
1793 enqueue_scan_from_roots_jobs (&gc_thread_gray_queue, nursery_section->data, nursery_section->end_data, is_parallel ? NULL : object_ops_nopar, is_parallel);
1796 gray_queue_redirect (&gc_thread_gray_queue);
1797 sgen_workers_start_all_workers (GENERATION_NURSERY, object_ops_nopar, object_ops_par, NULL);
1798 sgen_workers_join (GENERATION_NURSERY);
1802 time_minor_scan_roots += TV_ELAPSED (atv, btv);
1804 finish_gray_stack (GENERATION_NURSERY, ctx);
1807 time_minor_finish_gray_stack += TV_ELAPSED (btv, atv);
1808 sgen_client_binary_protocol_mark_end (GENERATION_NURSERY);
1810 if (objects_pinned) {
1811 sgen_optimize_pin_queue ();
1812 sgen_pinning_setup_section (nursery_section);
1816 * This is the latest point at which we can do this check, because
1817 * sgen_build_nursery_fragments() unpins nursery objects again.
1819 if (remset_consistency_checks)
1820 sgen_check_remset_consistency ();
1823 if (sgen_max_pause_time) {
1827 duration = (int)(TV_ELAPSED (last_minor_collection_start_tv, btv) / 10000);
1828 if (duration > (sgen_max_pause_time * sgen_max_pause_margin))
1829 sgen_resize_nursery (TRUE);
1831 sgen_resize_nursery (FALSE);
1833 sgen_resize_nursery (FALSE);
1836 /* walk the pin_queue, build up the fragment list of free memory, unmark
1837 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1840 sgen_client_binary_protocol_reclaim_start (GENERATION_NURSERY);
1841 fragment_total = sgen_build_nursery_fragments (nursery_section, unpin_queue);
1842 if (!fragment_total)
1845 /* Clear TLABs for all threads */
1846 sgen_clear_tlabs ();
1848 sgen_client_binary_protocol_reclaim_end (GENERATION_NURSERY);
1850 time_minor_fragment_creation += TV_ELAPSED (atv, btv);
1851 SGEN_LOG (2, "Fragment creation: %lld usecs, %lu bytes available", (long long)TV_ELAPSED (atv, btv), (unsigned long)fragment_total);
1853 if (remset_consistency_checks)
1854 sgen_check_major_refs ();
1856 major_collector.finish_nursery_collection ();
1858 TV_GETTIME (last_minor_collection_end_tv);
1859 UnlockedAdd64 (&gc_stats.minor_gc_time, TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv));
1861 sgen_debug_dump_heap ("minor", InterlockedRead (&gc_stats.minor_gc_count) - 1, NULL);
1863 /* prepare the pin queue for the next collection */
1864 sgen_finish_pinning ();
1865 if (sgen_have_pending_finalizers ()) {
1866 SGEN_LOG (4, "Finalizer-thread wakeup");
1867 sgen_client_finalize_notify ();
1869 sgen_pin_stats_reset ();
1870 /* clear cemented hash */
1871 sgen_cement_clear_below_threshold ();
1873 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
1875 check_scan_starts ();
1877 binary_protocol_flush_buffers (FALSE);
1879 sgen_memgov_minor_collection_end (reason, is_overflow);
1881 /*objects are late pinned because of lack of memory, so a major is a good call*/
1882 needs_major = objects_pinned > 0;
1883 current_collection_generation = -1;
1887 binary_protocol_collection_end_stats (0, 0, time_minor_finish_gray_stack - finish_gray_start);
1889 binary_protocol_collection_end_stats (
1890 time_minor_scan_major_blocks - major_scan_start,
1891 time_minor_scan_los - los_scan_start,
1892 time_minor_finish_gray_stack - finish_gray_start);
1894 binary_protocol_collection_end (InterlockedRead (&gc_stats.minor_gc_count) - 1, GENERATION_NURSERY, 0, 0);
1896 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1897 sgen_check_nursery_objects_pinned (unpin_queue != NULL);
1903 COPY_OR_MARK_FROM_ROOTS_SERIAL,
1904 COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT,
1905 COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT
1906 } CopyOrMarkFromRootsMode;
1909 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)
1914 /* FIXME: only use these values for the precise scan
1915 * note that to_space pointers should be excluded anyway...
1917 char *heap_start = NULL;
1918 char *heap_end = (char*)-1;
1919 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops_nopar, gc_thread_gray_queue);
1920 gboolean concurrent = mode != COPY_OR_MARK_FROM_ROOTS_SERIAL;
1922 SGEN_ASSERT (0, !!concurrent == !!concurrent_collection_in_progress, "We've been called with the wrong mode.");
1924 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1925 /*This cleans up unused fragments */
1926 sgen_nursery_allocator_prepare_for_pinning ();
1928 if (do_concurrent_checks)
1929 sgen_debug_check_nursery_is_clean ();
1931 /* The concurrent collector doesn't touch the nursery. */
1932 sgen_nursery_alloc_prepare_for_major ();
1937 /* Pinning depends on this */
1938 sgen_clear_nursery_fragments ();
1940 if (whole_heap_check_before_collection)
1941 sgen_check_whole_heap (TRUE);
1944 time_major_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1948 sgen_client_pre_collection_checks ();
1950 if (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1951 /* Remsets are not useful for a major collection */
1952 remset.clear_cards ();
1955 sgen_process_fin_stage_entries ();
1958 sgen_init_pinning ();
1959 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT)
1960 sgen_init_pinning_for_conc ();
1961 SGEN_LOG (6, "Collecting pinned addresses");
1962 pin_from_roots ((void*)lowest_heap_address, (void*)highest_heap_address, ctx);
1963 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1964 /* Pin cemented objects that were forced */
1965 sgen_pin_cemented_objects ();
1967 sgen_optimize_pin_queue ();
1968 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1970 * Cemented objects that are in the pinned list will be marked. When
1971 * marking concurrently we won't mark mod-union cards for these objects.
1972 * Instead they will remain cemented until the next major collection,
1973 * when we will recheck if they are still pinned in the roots.
1975 sgen_cement_force_pinned ();
1978 sgen_client_collecting_major_1 ();
1981 * pin_queue now contains all candidate pointers, sorted and
1982 * uniqued. We must do two passes now to figure out which
1983 * objects are pinned.
1985 * The first is to find within the pin_queue the area for each
1986 * section. This requires that the pin_queue be sorted. We
1987 * also process the LOS objects and pinned chunks here.
1989 * The second, destructive, pass is to reduce the section
1990 * areas to pointers to the actually pinned objects.
1992 SGEN_LOG (6, "Pinning from sections");
1993 /* first pass for the sections */
1994 sgen_find_section_pin_queue_start_end (nursery_section);
1995 /* identify possible pointers to the insize of large objects */
1996 SGEN_LOG (6, "Pinning from large objects");
1997 for (bigobj = los_object_list; bigobj; bigobj = bigobj->next) {
1999 if (sgen_find_optimized_pin_queue_area ((char*)bigobj->data, (char*)bigobj->data + sgen_los_object_size (bigobj), &dummy, &dummy)) {
2000 binary_protocol_pin (bigobj->data, (gpointer)LOAD_VTABLE (bigobj->data), safe_object_get_size (bigobj->data));
2002 if (sgen_los_object_is_pinned (bigobj->data)) {
2003 SGEN_ASSERT (0, mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, "LOS objects can only be pinned here after concurrent marking.");
2006 sgen_los_pin_object (bigobj->data);
2007 if (SGEN_OBJECT_HAS_REFERENCES (bigobj->data))
2008 GRAY_OBJECT_ENQUEUE_SERIAL (gc_thread_gray_queue, bigobj->data, sgen_obj_get_descriptor ((GCObject*)bigobj->data));
2009 sgen_pin_stats_register_object (bigobj->data, GENERATION_OLD);
2010 SGEN_LOG (6, "Marked large object %p (%s) size: %lu from roots", bigobj->data,
2011 sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (bigobj->data)),
2012 (unsigned long)sgen_los_object_size (bigobj));
2014 sgen_client_pinned_los_object (bigobj->data);
2018 pin_objects_in_nursery (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT, ctx);
2019 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
2020 sgen_check_nursery_objects_pinned (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
2022 major_collector.pin_objects (gc_thread_gray_queue);
2023 if (old_next_pin_slot)
2024 *old_next_pin_slot = sgen_get_pinned_count ();
2027 time_major_pinning += TV_ELAPSED (atv, btv);
2028 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), (long long)TV_ELAPSED (atv, btv));
2029 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
2031 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT)
2032 sgen_finish_pinning_for_conc ();
2034 major_collector.init_to_space ();
2036 SGEN_ASSERT (0, sgen_workers_all_done (), "Why are the workers not done when we start or finish a major collection?");
2037 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
2038 if (object_ops_par != NULL)
2039 sgen_workers_set_num_active_workers (GENERATION_OLD, 0);
2040 if (object_ops_par == NULL && sgen_workers_have_idle_work (GENERATION_OLD)) {
2042 * We force the finish of the worker with the new object ops context
2043 * which can also do copying. We need to have finished pinning. On the
2044 * parallel collector, there is no need to drain the private queues
2045 * here, since we can do it as part of the finishing work, achieving
2046 * better work distribution.
2048 sgen_workers_start_all_workers (GENERATION_OLD, object_ops_nopar, object_ops_par, NULL);
2050 sgen_workers_join (GENERATION_OLD);
2054 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
2055 main_gc_thread = mono_native_thread_self ();
2058 sgen_client_collecting_major_2 ();
2061 time_major_scan_pinned += TV_ELAPSED (btv, atv);
2063 sgen_client_collecting_major_3 (&fin_ready_queue, &critical_fin_queue);
2065 enqueue_scan_from_roots_jobs (gc_thread_gray_queue, heap_start, heap_end, object_ops_nopar, FALSE);
2068 time_major_scan_roots += TV_ELAPSED (atv, btv);
2071 * We start the concurrent worker after pinning and after we scanned the roots
2072 * in order to make sure that the worker does not finish before handling all
2075 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
2076 sgen_workers_set_num_active_workers (GENERATION_OLD, 1);
2077 gray_queue_redirect (gc_thread_gray_queue);
2078 if (precleaning_enabled) {
2079 sgen_workers_start_all_workers (GENERATION_OLD, object_ops_nopar, object_ops_par, workers_finish_callback);
2081 sgen_workers_start_all_workers (GENERATION_OLD, object_ops_nopar, object_ops_par, NULL);
2085 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
2086 int i, split_count = sgen_workers_get_job_split_count (GENERATION_OLD);
2087 size_t num_major_sections = major_collector.get_num_major_sections ();
2088 gboolean parallel = object_ops_par != NULL;
2090 /* If we're not parallel we finish the collection on the gc thread */
2092 gray_queue_redirect (gc_thread_gray_queue);
2094 /* Mod union card table */
2095 for (i = 0; i < split_count; i++) {
2096 ParallelScanJob *psj;
2098 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("scan mod union cardtable", job_scan_major_mod_union_card_table, sizeof (ParallelScanJob));
2099 psj->scan_job.ops = parallel ? NULL : object_ops_nopar;
2100 psj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
2102 psj->job_split_count = split_count;
2103 psj->data = num_major_sections / split_count;
2104 sgen_workers_enqueue_job (GENERATION_OLD, &psj->scan_job.job, parallel);
2106 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("scan LOS mod union cardtable", job_scan_los_mod_union_card_table, sizeof (ParallelScanJob));
2107 psj->scan_job.ops = parallel ? NULL : object_ops_nopar;
2108 psj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
2110 psj->job_split_count = split_count;
2111 sgen_workers_enqueue_job (GENERATION_OLD, &psj->scan_job.job, parallel);
2116 * If we enqueue a job while workers are running we need to sgen_workers_ensure_awake
2117 * in order to make sure that we are running the idle func and draining all worker
2118 * gray queues. The operation of starting workers implies this, so we start them after
2119 * in order to avoid doing this operation twice. The workers will drain the main gray
2120 * stack that contained roots and pinned objects and also scan the mod union card
2123 sgen_workers_start_all_workers (GENERATION_OLD, object_ops_nopar, object_ops_par, NULL);
2124 sgen_workers_join (GENERATION_OLD);
2128 sgen_pin_stats_report ();
2130 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
2131 sgen_finish_pinning ();
2133 sgen_pin_stats_reset ();
2135 if (do_concurrent_checks)
2136 sgen_debug_check_nursery_is_clean ();
2141 major_start_collection (SgenGrayQueue *gc_thread_gray_queue, const char *reason, gboolean concurrent, size_t *old_next_pin_slot)
2143 SgenObjectOperations *object_ops_nopar, *object_ops_par = NULL;
2145 binary_protocol_collection_begin (InterlockedRead (&gc_stats.major_gc_count), GENERATION_OLD);
2147 current_collection_generation = GENERATION_OLD;
2149 sgen_workers_assert_gray_queue_is_empty (GENERATION_OLD);
2152 sgen_cement_reset ();
2155 g_assert (major_collector.is_concurrent);
2156 concurrent_collection_in_progress = TRUE;
2158 object_ops_nopar = &major_collector.major_ops_concurrent_start;
2159 if (major_collector.is_parallel)
2160 object_ops_par = &major_collector.major_ops_conc_par_start;
2163 object_ops_nopar = &major_collector.major_ops_serial;
2166 reset_pinned_from_failed_allocation ();
2168 sgen_memgov_major_collection_start (concurrent, reason);
2170 //count_ref_nonref_objs ();
2171 //consistency_check ();
2173 check_scan_starts ();
2176 SGEN_LOG (1, "Start major collection %" G_GINT32_FORMAT, InterlockedRead (&gc_stats.major_gc_count));
2177 InterlockedIncrement (&gc_stats.major_gc_count);
2179 if (major_collector.start_major_collection)
2180 major_collector.start_major_collection ();
2182 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);
2186 major_finish_collection (SgenGrayQueue *gc_thread_gray_queue, const char *reason, gboolean is_overflow, size_t old_next_pin_slot, gboolean forced)
2188 ScannedObjectCounts counts;
2189 SgenObjectOperations *object_ops_nopar;
2190 mword fragment_total;
2193 guint64 major_scan_start = time_major_scan_mod_union_blocks;
2194 guint64 los_scan_start = time_major_scan_mod_union_los;
2195 guint64 finish_gray_start = time_major_finish_gray_stack;
2197 if (concurrent_collection_in_progress) {
2198 SgenObjectOperations *object_ops_par = NULL;
2200 object_ops_nopar = &major_collector.major_ops_concurrent_finish;
2201 if (major_collector.is_parallel)
2202 object_ops_par = &major_collector.major_ops_conc_par_finish;
2204 major_copy_or_mark_from_roots (gc_thread_gray_queue, NULL, COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, object_ops_nopar, object_ops_par);
2206 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
2207 main_gc_thread = NULL;
2210 object_ops_nopar = &major_collector.major_ops_serial;
2213 sgen_workers_assert_gray_queue_is_empty (GENERATION_OLD);
2216 finish_gray_stack (GENERATION_OLD, CONTEXT_FROM_OBJECT_OPERATIONS (object_ops_nopar, gc_thread_gray_queue));
2218 time_major_finish_gray_stack += TV_ELAPSED (btv, atv);
2220 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after joining");
2222 if (objects_pinned) {
2223 g_assert (!concurrent_collection_in_progress);
2226 * This is slow, but we just OOM'd.
2228 * See comment at `sgen_pin_queue_clear_discarded_entries` for how the pin
2229 * queue is laid out at this point.
2231 sgen_pin_queue_clear_discarded_entries (nursery_section, old_next_pin_slot);
2233 * We need to reestablish all pinned nursery objects in the pin queue
2234 * because they're needed for fragment creation. Unpinning happens by
2235 * walking the whole queue, so it's not necessary to reestablish where major
2236 * heap block pins are - all we care is that they're still in there
2239 sgen_optimize_pin_queue ();
2240 sgen_find_section_pin_queue_start_end (nursery_section);
2244 reset_heap_boundaries ();
2245 sgen_update_heap_boundaries ((mword)sgen_get_nursery_start (), (mword)sgen_get_nursery_end ());
2247 /* walk the pin_queue, build up the fragment list of free memory, unmark
2248 * pinned objects as we go, memzero() the empty fragments so they are ready for the
2251 fragment_total = sgen_build_nursery_fragments (nursery_section, NULL);
2252 if (!fragment_total)
2254 SGEN_LOG (4, "Free space in nursery after major %ld", (long)fragment_total);
2256 if (do_concurrent_checks && concurrent_collection_in_progress)
2257 sgen_debug_check_nursery_is_clean ();
2259 /* prepare the pin queue for the next collection */
2260 sgen_finish_pinning ();
2262 /* Clear TLABs for all threads */
2263 sgen_clear_tlabs ();
2265 sgen_pin_stats_reset ();
2267 sgen_cement_clear_below_threshold ();
2269 if (check_mark_bits_after_major_collection)
2270 sgen_check_heap_marked (concurrent_collection_in_progress);
2273 time_major_fragment_creation += TV_ELAPSED (atv, btv);
2275 binary_protocol_sweep_begin (GENERATION_OLD, !major_collector.sweeps_lazily);
2276 sgen_memgov_major_pre_sweep ();
2279 time_major_free_bigobjs += TV_ELAPSED (btv, atv);
2284 time_major_los_sweep += TV_ELAPSED (atv, btv);
2286 major_collector.sweep ();
2288 binary_protocol_sweep_end (GENERATION_OLD, !major_collector.sweeps_lazily);
2291 time_major_sweep += TV_ELAPSED (btv, atv);
2293 sgen_debug_dump_heap ("major", InterlockedRead (&gc_stats.major_gc_count) - 1, reason);
2295 if (sgen_have_pending_finalizers ()) {
2296 SGEN_LOG (4, "Finalizer-thread wakeup");
2297 sgen_client_finalize_notify ();
2300 sgen_memgov_major_collection_end (forced, concurrent_collection_in_progress, reason, is_overflow);
2301 current_collection_generation = -1;
2303 memset (&counts, 0, sizeof (ScannedObjectCounts));
2304 major_collector.finish_major_collection (&counts);
2306 sgen_workers_assert_gray_queue_is_empty (GENERATION_OLD);
2308 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after major collection has finished");
2309 if (concurrent_collection_in_progress)
2310 concurrent_collection_in_progress = FALSE;
2312 check_scan_starts ();
2314 binary_protocol_flush_buffers (FALSE);
2316 //consistency_check ();
2317 if (major_collector.is_parallel)
2318 binary_protocol_collection_end_stats (0, 0, time_major_finish_gray_stack - finish_gray_start);
2320 binary_protocol_collection_end_stats (
2321 time_major_scan_mod_union_blocks - major_scan_start,
2322 time_major_scan_mod_union_los - los_scan_start,
2323 time_major_finish_gray_stack - finish_gray_start);
2325 binary_protocol_collection_end (InterlockedRead (&gc_stats.major_gc_count) - 1, GENERATION_OLD, counts.num_scanned_objects, counts.num_unique_scanned_objects);
2329 major_do_collection (const char *reason, gboolean is_overflow, gboolean forced)
2331 TV_DECLARE (time_start);
2332 TV_DECLARE (time_end);
2333 size_t old_next_pin_slot;
2334 SgenGrayQueue gc_thread_gray_queue;
2336 if (disable_major_collections)
2339 if (major_collector.get_and_reset_num_major_objects_marked) {
2340 long long num_marked = major_collector.get_and_reset_num_major_objects_marked ();
2341 g_assert (!num_marked);
2344 /* world must be stopped already */
2345 TV_GETTIME (time_start);
2347 init_gray_queue (&gc_thread_gray_queue);
2348 major_start_collection (&gc_thread_gray_queue, reason, FALSE, &old_next_pin_slot);
2349 major_finish_collection (&gc_thread_gray_queue, reason, is_overflow, old_next_pin_slot, forced);
2350 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2352 TV_GETTIME (time_end);
2353 UnlockedAdd64 (&gc_stats.major_gc_time, TV_ELAPSED (time_start, time_end));
2355 /* FIXME: also report this to the user, preferably in gc-end. */
2356 if (major_collector.get_and_reset_num_major_objects_marked)
2357 major_collector.get_and_reset_num_major_objects_marked ();
2359 return bytes_pinned_from_failed_allocation > 0;
2363 major_start_concurrent_collection (const char *reason)
2365 TV_DECLARE (time_start);
2366 TV_DECLARE (time_end);
2367 long long num_objects_marked;
2368 SgenGrayQueue gc_thread_gray_queue;
2370 if (disable_major_collections)
2373 TV_GETTIME (time_start);
2374 SGEN_TV_GETTIME (time_major_conc_collection_start);
2376 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2377 g_assert (num_objects_marked == 0);
2379 binary_protocol_concurrent_start ();
2381 init_gray_queue (&gc_thread_gray_queue);
2382 // FIXME: store reason and pass it when finishing
2383 major_start_collection (&gc_thread_gray_queue, reason, TRUE, NULL);
2384 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2386 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2388 TV_GETTIME (time_end);
2389 UnlockedAdd64 (&gc_stats.major_gc_time, TV_ELAPSED (time_start, time_end));
2391 current_collection_generation = -1;
2395 * Returns whether the major collection has finished.
2398 major_should_finish_concurrent_collection (void)
2400 return sgen_workers_all_done ();
2404 major_update_concurrent_collection (void)
2406 TV_DECLARE (total_start);
2407 TV_DECLARE (total_end);
2409 TV_GETTIME (total_start);
2411 binary_protocol_concurrent_update ();
2413 major_collector.update_cardtable_mod_union ();
2414 sgen_los_update_cardtable_mod_union ();
2416 TV_GETTIME (total_end);
2417 UnlockedAdd64 (&gc_stats.major_gc_time, TV_ELAPSED (total_start, total_end));
2421 major_finish_concurrent_collection (gboolean forced)
2423 SgenGrayQueue gc_thread_gray_queue;
2424 TV_DECLARE (total_start);
2425 TV_DECLARE (total_end);
2427 TV_GETTIME (total_start);
2429 binary_protocol_concurrent_finish ();
2432 * We need to stop all workers since we're updating the cardtable below.
2433 * The workers will be resumed with a finishing pause context to avoid
2434 * additional cardtable and object scanning.
2436 sgen_workers_stop_all_workers (GENERATION_OLD);
2438 SGEN_TV_GETTIME (time_major_conc_collection_end);
2439 UnlockedAdd64 (&gc_stats.major_gc_time_concurrent, SGEN_TV_ELAPSED (time_major_conc_collection_start, time_major_conc_collection_end));
2441 major_collector.update_cardtable_mod_union ();
2442 sgen_los_update_cardtable_mod_union ();
2444 if (mod_union_consistency_check)
2445 sgen_check_mod_union_consistency ();
2447 current_collection_generation = GENERATION_OLD;
2448 sgen_cement_reset ();
2449 init_gray_queue (&gc_thread_gray_queue);
2450 major_finish_collection (&gc_thread_gray_queue, "finishing", FALSE, -1, forced);
2451 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2453 TV_GETTIME (total_end);
2454 UnlockedAdd64 (&gc_stats.major_gc_time, TV_ELAPSED (total_start, total_end));
2456 current_collection_generation = -1;
2460 * Ensure an allocation request for @size will succeed by freeing enough memory.
2462 * LOCKING: The GC lock MUST be held.
2465 sgen_ensure_free_space (size_t size, int generation)
2467 int generation_to_collect = -1;
2468 const char *reason = NULL;
2470 if (generation == GENERATION_OLD) {
2471 if (sgen_need_major_collection (size)) {
2472 reason = "LOS overflow";
2473 generation_to_collect = GENERATION_OLD;
2476 if (degraded_mode) {
2477 if (sgen_need_major_collection (size)) {
2478 reason = "Degraded mode overflow";
2479 generation_to_collect = GENERATION_OLD;
2481 } else if (sgen_need_major_collection (size)) {
2482 reason = concurrent_collection_in_progress ? "Forced finish concurrent collection" : "Minor allowance";
2483 generation_to_collect = GENERATION_OLD;
2485 generation_to_collect = GENERATION_NURSERY;
2486 reason = "Nursery full";
2490 if (generation_to_collect == -1) {
2491 if (concurrent_collection_in_progress && sgen_workers_all_done ()) {
2492 generation_to_collect = GENERATION_OLD;
2493 reason = "Finish concurrent collection";
2497 if (generation_to_collect == -1)
2499 sgen_perform_collection (size, generation_to_collect, reason, FALSE, TRUE);
2503 * LOCKING: Assumes the GC lock is held.
2506 sgen_perform_collection (size_t requested_size, int generation_to_collect, const char *reason, gboolean wait_to_finish, gboolean stw)
2508 TV_DECLARE (gc_total_start);
2509 TV_DECLARE (gc_total_end);
2510 int overflow_generation_to_collect = -1;
2511 int oldest_generation_collected = generation_to_collect;
2512 const char *overflow_reason = NULL;
2513 gboolean finish_concurrent = concurrent_collection_in_progress && (major_should_finish_concurrent_collection () || generation_to_collect == GENERATION_OLD);
2515 binary_protocol_collection_requested (generation_to_collect, requested_size, wait_to_finish ? 1 : 0);
2517 SGEN_ASSERT (0, generation_to_collect == GENERATION_NURSERY || generation_to_collect == GENERATION_OLD, "What generation is this?");
2520 sgen_stop_world (generation_to_collect);
2522 SGEN_ASSERT (0, sgen_is_world_stopped (), "We can only collect if the world is stopped");
2525 TV_GETTIME (gc_total_start);
2527 // FIXME: extract overflow reason
2528 // FIXME: minor overflow for concurrent case
2529 if (generation_to_collect == GENERATION_NURSERY && !finish_concurrent) {
2530 if (concurrent_collection_in_progress)
2531 major_update_concurrent_collection ();
2533 if (collect_nursery (reason, FALSE, NULL) && !concurrent_collection_in_progress) {
2534 overflow_generation_to_collect = GENERATION_OLD;
2535 overflow_reason = "Minor overflow";
2537 } else if (finish_concurrent) {
2538 major_finish_concurrent_collection (wait_to_finish);
2539 oldest_generation_collected = GENERATION_OLD;
2541 SGEN_ASSERT (0, generation_to_collect == GENERATION_OLD, "We should have handled nursery collections above");
2542 if (major_collector.is_concurrent && !wait_to_finish) {
2543 collect_nursery ("Concurrent start", FALSE, NULL);
2544 major_start_concurrent_collection (reason);
2545 oldest_generation_collected = GENERATION_NURSERY;
2546 } else if (major_do_collection (reason, FALSE, wait_to_finish)) {
2547 overflow_generation_to_collect = GENERATION_NURSERY;
2548 overflow_reason = "Excessive pinning";
2552 if (overflow_generation_to_collect != -1) {
2553 SGEN_ASSERT (0, !concurrent_collection_in_progress, "We don't yet support overflow collections with the concurrent collector");
2556 * We need to do an overflow collection, either because we ran out of memory
2557 * or the nursery is fully pinned.
2560 if (overflow_generation_to_collect == GENERATION_NURSERY)
2561 collect_nursery (overflow_reason, TRUE, NULL);
2563 major_do_collection (overflow_reason, TRUE, wait_to_finish);
2565 oldest_generation_collected = MAX (oldest_generation_collected, overflow_generation_to_collect);
2568 SGEN_LOG (2, "Heap size: %lu, LOS size: %lu", (unsigned long)sgen_gc_get_total_heap_allocation (), (unsigned long)los_memory_usage);
2570 /* this also sets the proper pointers for the next allocation */
2571 if (generation_to_collect == GENERATION_NURSERY && !sgen_can_alloc_size (requested_size)) {
2572 /* TypeBuilder and MonoMethod are killing mcs with fragmentation */
2573 SGEN_LOG (1, "nursery collection didn't find enough room for %zd alloc (%zd pinned)", requested_size, sgen_get_pinned_count ());
2574 sgen_dump_pin_queue ();
2578 TV_GETTIME (gc_total_end);
2579 time_max = MAX (time_max, TV_ELAPSED (gc_total_start, gc_total_end));
2582 sgen_restart_world (oldest_generation_collected);
2586 * ######################################################################
2587 * ######## Memory allocation from the OS
2588 * ######################################################################
2589 * This section of code deals with getting memory from the OS and
2590 * allocating memory for GC-internal data structures.
2591 * Internal memory can be handled with a freelist for small objects.
2597 G_GNUC_UNUSED static void
2598 report_internal_mem_usage (void)
2600 printf ("Internal memory usage:\n");
2601 sgen_report_internal_mem_usage ();
2602 printf ("Pinned memory usage:\n");
2603 major_collector.report_pinned_memory_usage ();
2607 * ######################################################################
2608 * ######## Finalization support
2609 * ######################################################################
2613 * If the object has been forwarded it means it's still referenced from a root.
2614 * If it is pinned it's still alive as well.
2615 * A LOS object is only alive if we have pinned it.
2616 * Return TRUE if @obj is ready to be finalized.
2618 static inline gboolean
2619 sgen_is_object_alive (GCObject *object)
2621 if (ptr_in_nursery (object))
2622 return sgen_nursery_is_object_alive (object);
2624 return sgen_major_is_object_alive (object);
2628 * This function returns true if @object is either alive and belongs to the
2629 * current collection - major collections are full heap, so old gen objects
2630 * are never alive during a minor collection.
2633 sgen_is_object_alive_and_on_current_collection (GCObject *object)
2635 if (ptr_in_nursery (object))
2636 return sgen_nursery_is_object_alive (object);
2638 if (current_collection_generation == GENERATION_NURSERY)
2641 return sgen_major_is_object_alive (object);
2646 sgen_gc_is_object_ready_for_finalization (GCObject *object)
2648 return !sgen_is_object_alive (object);
2652 sgen_queue_finalization_entry (GCObject *obj)
2654 gboolean critical = sgen_client_object_has_critical_finalizer (obj);
2656 sgen_pointer_queue_add (critical ? &critical_fin_queue : &fin_ready_queue, obj);
2658 sgen_client_object_queued_for_finalization (obj);
2662 sgen_object_is_live (GCObject *obj)
2664 return sgen_is_object_alive_and_on_current_collection (obj);
2668 * `System.GC.WaitForPendingFinalizers` first checks `sgen_have_pending_finalizers()` to
2669 * determine whether it can exit quickly. The latter must therefore only return FALSE if
2670 * all finalizers have really finished running.
2672 * `sgen_gc_invoke_finalizers()` first dequeues a finalizable object, and then finalizes it.
2673 * This means that just checking whether the queues are empty leaves the possibility that an
2674 * object might have been dequeued but not yet finalized. That's why we need the additional
2675 * flag `pending_unqueued_finalizer`.
2678 static volatile gboolean pending_unqueued_finalizer = FALSE;
2679 volatile gboolean sgen_suspend_finalizers = FALSE;
2682 sgen_set_suspend_finalizers (void)
2684 sgen_suspend_finalizers = TRUE;
2688 sgen_gc_invoke_finalizers (void)
2692 g_assert (!pending_unqueued_finalizer);
2694 /* FIXME: batch to reduce lock contention */
2695 while (sgen_have_pending_finalizers ()) {
2701 * We need to set `pending_unqueued_finalizer` before dequeing the
2702 * finalizable object.
2704 if (!sgen_pointer_queue_is_empty (&fin_ready_queue)) {
2705 pending_unqueued_finalizer = TRUE;
2706 mono_memory_write_barrier ();
2707 obj = (GCObject *)sgen_pointer_queue_pop (&fin_ready_queue);
2708 } else if (!sgen_pointer_queue_is_empty (&critical_fin_queue)) {
2709 pending_unqueued_finalizer = TRUE;
2710 mono_memory_write_barrier ();
2711 obj = (GCObject *)sgen_pointer_queue_pop (&critical_fin_queue);
2717 SGEN_LOG (7, "Finalizing object %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
2725 /* the object is on the stack so it is pinned */
2726 /*g_print ("Calling finalizer for object: %p (%s)\n", obj, sgen_client_object_safe_name (obj));*/
2727 sgen_client_run_finalize (obj);
2730 if (pending_unqueued_finalizer) {
2731 mono_memory_write_barrier ();
2732 pending_unqueued_finalizer = FALSE;
2739 sgen_have_pending_finalizers (void)
2741 if (sgen_suspend_finalizers)
2743 return pending_unqueued_finalizer || !sgen_pointer_queue_is_empty (&fin_ready_queue) || !sgen_pointer_queue_is_empty (&critical_fin_queue);
2747 * ######################################################################
2748 * ######## registered roots support
2749 * ######################################################################
2753 * We do not coalesce roots.
2756 sgen_register_root (char *start, size_t size, SgenDescriptor descr, int root_type, int source, const char *msg)
2758 RootRecord new_root;
2761 for (i = 0; i < ROOT_TYPE_NUM; ++i) {
2762 RootRecord *root = (RootRecord *)sgen_hash_table_lookup (&roots_hash [i], start);
2763 /* we allow changing the size and the descriptor (for thread statics etc) */
2765 size_t old_size = root->end_root - start;
2766 root->end_root = start + size;
2767 SGEN_ASSERT (0, !!root->root_desc == !!descr, "Can't change whether a root is precise or conservative.");
2768 SGEN_ASSERT (0, root->source == source, "Can't change a root's source identifier.");
2769 SGEN_ASSERT (0, !!root->msg == !!msg, "Can't change a root's message.");
2770 root->root_desc = descr;
2772 roots_size -= old_size;
2778 new_root.end_root = start + size;
2779 new_root.root_desc = descr;
2780 new_root.source = source;
2783 sgen_hash_table_replace (&roots_hash [root_type], start, &new_root, NULL);
2786 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);
2793 sgen_deregister_root (char* addr)
2799 for (root_type = 0; root_type < ROOT_TYPE_NUM; ++root_type) {
2800 if (sgen_hash_table_remove (&roots_hash [root_type], addr, &root))
2801 roots_size -= (root.end_root - addr);
2807 sgen_wbroots_iterate_live_block_ranges (sgen_cardtable_block_callback cb)
2811 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_WBARRIER], void **, start_root, RootRecord *, root) {
2812 cb ((mword)start_root, (mword)root->end_root - (mword)start_root);
2813 } SGEN_HASH_TABLE_FOREACH_END;
2816 /* Root equivalent of sgen_client_cardtable_scan_object */
2818 sgen_wbroot_scan_card_table (void** start_root, mword size, ScanCopyContext ctx)
2820 ScanPtrFieldFunc scan_field_func = ctx.ops->scan_ptr_field;
2821 guint8 *card_data = sgen_card_table_get_card_scan_address ((mword)start_root);
2822 guint8 *card_base = card_data;
2823 mword card_count = sgen_card_table_number_of_cards_in_range ((mword)start_root, size);
2824 guint8 *card_data_end = card_data + card_count;
2825 mword extra_idx = 0;
2826 char *obj_start = sgen_card_table_align_pointer (start_root);
2827 char *obj_end = (char*)start_root + size;
2828 #ifdef SGEN_HAVE_OVERLAPPING_CARDS
2829 guint8 *overflow_scan_end = NULL;
2832 #ifdef SGEN_HAVE_OVERLAPPING_CARDS
2833 /*Check for overflow and if so, setup to scan in two steps*/
2834 if (card_data_end >= SGEN_SHADOW_CARDTABLE_END) {
2835 overflow_scan_end = sgen_shadow_cardtable + (card_data_end - SGEN_SHADOW_CARDTABLE_END);
2836 card_data_end = SGEN_SHADOW_CARDTABLE_END;
2842 card_data = sgen_find_next_card (card_data, card_data_end);
2844 for (; card_data < card_data_end; card_data = sgen_find_next_card (card_data + 1, card_data_end)) {
2845 size_t idx = (card_data - card_base) + extra_idx;
2846 char *start = (char*)(obj_start + idx * CARD_SIZE_IN_BYTES);
2847 char *card_end = start + CARD_SIZE_IN_BYTES;
2848 char *elem = start, *first_elem = start;
2851 * Don't clean first and last card on 32bit systems since they
2852 * may also be part from other roots.
2854 if (card_data != card_base && card_data != (card_data_end - 1))
2855 sgen_card_table_prepare_card_for_scanning (card_data);
2857 card_end = MIN (card_end, obj_end);
2859 if (elem < (char*)start_root)
2860 first_elem = elem = (char*)start_root;
2862 for (; elem < card_end; elem += SIZEOF_VOID_P) {
2863 if (*(GCObject**)elem)
2864 scan_field_func (NULL, (GCObject**)elem, ctx.queue);
2867 binary_protocol_card_scan (first_elem, elem - first_elem);
2870 #ifdef SGEN_HAVE_OVERLAPPING_CARDS
2871 if (overflow_scan_end) {
2872 extra_idx = card_data - card_base;
2873 card_base = card_data = sgen_shadow_cardtable;
2874 card_data_end = overflow_scan_end;
2875 overflow_scan_end = NULL;
2882 sgen_wbroots_scan_card_table (ScanCopyContext ctx)
2887 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_WBARRIER], void **, start_root, RootRecord *, root) {
2888 SGEN_ASSERT (0, (root->root_desc & ROOT_DESC_TYPE_MASK) == ROOT_DESC_VECTOR, "Unsupported root type");
2890 sgen_wbroot_scan_card_table (start_root, (mword)root->end_root - (mword)start_root, ctx);
2891 } SGEN_HASH_TABLE_FOREACH_END;
2895 * ######################################################################
2896 * ######## Thread handling (stop/start code)
2897 * ######################################################################
2901 sgen_get_current_collection_generation (void)
2903 return current_collection_generation;
2907 sgen_thread_attach (SgenThreadInfo* info)
2909 info->tlab_start = info->tlab_next = info->tlab_temp_end = info->tlab_real_end = NULL;
2911 sgen_client_thread_attach (info);
2917 sgen_thread_detach_with_lock (SgenThreadInfo *p)
2919 sgen_client_thread_detach_with_lock (p);
2923 * ######################################################################
2924 * ######## Write barriers
2925 * ######################################################################
2929 * Note: the write barriers first do the needed GC work and then do the actual store:
2930 * this way the value is visible to the conservative GC scan after the write barrier
2931 * itself. If a GC interrupts the barrier in the middle, value will be kept alive by
2932 * the conservative scan, otherwise by the remembered set scan.
2936 * mono_gc_wbarrier_arrayref_copy:
2939 mono_gc_wbarrier_arrayref_copy (gpointer dest_ptr, gpointer src_ptr, int count)
2941 HEAVY_STAT (++stat_wbarrier_arrayref_copy);
2942 /*This check can be done without taking a lock since dest_ptr array is pinned*/
2943 if (ptr_in_nursery (dest_ptr) || count <= 0) {
2944 mono_gc_memmove_aligned (dest_ptr, src_ptr, count * sizeof (gpointer));
2948 #ifdef SGEN_HEAVY_BINARY_PROTOCOL
2949 if (binary_protocol_is_heavy_enabled ()) {
2951 for (i = 0; i < count; ++i) {
2952 gpointer dest = (gpointer*)dest_ptr + i;
2953 gpointer obj = *((gpointer*)src_ptr + i);
2955 binary_protocol_wbarrier (dest, obj, (gpointer)LOAD_VTABLE (obj));
2960 remset.wbarrier_arrayref_copy (dest_ptr, src_ptr, count);
2964 * mono_gc_wbarrier_generic_nostore:
2967 mono_gc_wbarrier_generic_nostore (gpointer ptr)
2971 HEAVY_STAT (++stat_wbarrier_generic_store);
2973 sgen_client_wbarrier_generic_nostore_check (ptr);
2975 obj = *(gpointer*)ptr;
2977 binary_protocol_wbarrier (ptr, obj, (gpointer)LOAD_VTABLE (obj));
2980 * We need to record old->old pointer locations for the
2981 * concurrent collector.
2983 if (!ptr_in_nursery (obj) && !concurrent_collection_in_progress) {
2984 SGEN_LOG (8, "Skipping remset at %p", ptr);
2988 SGEN_LOG (8, "Adding remset at %p", ptr);
2990 remset.wbarrier_generic_nostore (ptr);
2994 * mono_gc_wbarrier_generic_store:
2997 mono_gc_wbarrier_generic_store (gpointer ptr, GCObject* value)
2999 SGEN_LOG (8, "Wbarrier store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
3000 SGEN_UPDATE_REFERENCE_ALLOW_NULL (ptr, value);
3001 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
3002 mono_gc_wbarrier_generic_nostore (ptr);
3003 sgen_dummy_use (value);
3007 * mono_gc_wbarrier_generic_store_atomic:
3008 * Same as \c mono_gc_wbarrier_generic_store but performs the store
3009 * as an atomic operation with release semantics.
3012 mono_gc_wbarrier_generic_store_atomic (gpointer ptr, GCObject *value)
3014 HEAVY_STAT (++stat_wbarrier_generic_store_atomic);
3016 SGEN_LOG (8, "Wbarrier atomic store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
3018 InterlockedWritePointer ((volatile gpointer *)ptr, value);
3020 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
3021 mono_gc_wbarrier_generic_nostore (ptr);
3023 sgen_dummy_use (value);
3027 sgen_wbarrier_range_copy (gpointer _dest, gpointer _src, int size)
3029 remset.wbarrier_range_copy (_dest,_src, size);
3033 * ######################################################################
3034 * ######## Other mono public interface functions.
3035 * ######################################################################
3039 sgen_gc_collect (int generation)
3044 sgen_perform_collection (0, generation, "user request", TRUE, TRUE);
3049 sgen_gc_collection_count (int generation)
3051 return InterlockedRead (generation == GENERATION_NURSERY ? &gc_stats.minor_gc_count : &gc_stats.major_gc_count);
3055 sgen_gc_get_used_size (void)
3059 tot = los_memory_usage;
3060 tot += nursery_section->end_data - nursery_section->data;
3061 tot += major_collector.get_used_size ();
3062 /* FIXME: account for pinned objects */
3068 sgen_env_var_error (const char *env_var, const char *fallback, const char *description_format, ...)
3072 va_start (ap, description_format);
3074 fprintf (stderr, "Warning: In environment variable `%s': ", env_var);
3075 vfprintf (stderr, description_format, ap);
3077 fprintf (stderr, " - %s", fallback);
3078 fprintf (stderr, "\n");
3084 parse_double_in_interval (const char *env_var, const char *opt_name, const char *opt, double min, double max, double *result)
3087 double val = strtod (opt, &endptr);
3088 if (endptr == opt) {
3089 sgen_env_var_error (env_var, "Using default value.", "`%s` must be a number.", opt_name);
3092 else if (val < min || val > max) {
3093 sgen_env_var_error (env_var, "Using default value.", "`%s` must be between %.2f - %.2f.", opt_name, min, max);
3101 parse_sgen_minor (const char *opt)
3104 return SGEN_MINOR_DEFAULT;
3106 if (!strcmp (opt, "simple")) {
3107 return SGEN_MINOR_SIMPLE;
3108 } else if (!strcmp (opt, "simple-par")) {
3109 return SGEN_MINOR_SIMPLE_PARALLEL;
3110 } else if (!strcmp (opt, "split")) {
3111 return SGEN_MINOR_SPLIT;
3113 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default instead.", "Unknown minor collector `%s'.", opt);
3114 return SGEN_MINOR_DEFAULT;
3119 parse_sgen_major (const char *opt)
3122 return SGEN_MAJOR_DEFAULT;
3124 if (!strcmp (opt, "marksweep")) {
3125 return SGEN_MAJOR_SERIAL;
3126 } else if (!strcmp (opt, "marksweep-conc")) {
3127 return SGEN_MAJOR_CONCURRENT;
3128 } else if (!strcmp (opt, "marksweep-conc-par")) {
3129 return SGEN_MAJOR_CONCURRENT_PARALLEL;
3131 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default instead.", "Unknown major collector `%s'.", opt);
3132 return SGEN_MAJOR_DEFAULT;
3138 parse_sgen_mode (const char *opt)
3141 return SGEN_MODE_NONE;
3143 if (!strcmp (opt, "balanced")) {
3144 return SGEN_MODE_BALANCED;
3145 } else if (!strcmp (opt, "throughput")) {
3146 return SGEN_MODE_THROUGHPUT;
3147 } else if (!strcmp (opt, "pause") || g_str_has_prefix (opt, "pause:")) {
3148 return SGEN_MODE_PAUSE;
3150 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default configurations.", "Unknown mode `%s'.", opt);
3151 return SGEN_MODE_NONE;
3156 init_sgen_minor (SgenMinor minor)
3159 case SGEN_MINOR_DEFAULT:
3160 case SGEN_MINOR_SIMPLE:
3161 sgen_simple_nursery_init (&sgen_minor_collector, FALSE);
3163 case SGEN_MINOR_SIMPLE_PARALLEL:
3164 sgen_simple_nursery_init (&sgen_minor_collector, TRUE);
3166 case SGEN_MINOR_SPLIT:
3167 sgen_split_nursery_init (&sgen_minor_collector);
3170 g_assert_not_reached ();
3175 init_sgen_major (SgenMajor major)
3177 if (major == SGEN_MAJOR_DEFAULT)
3178 major = DEFAULT_MAJOR;
3181 case SGEN_MAJOR_SERIAL:
3182 sgen_marksweep_init (&major_collector);
3184 case SGEN_MAJOR_CONCURRENT:
3185 sgen_marksweep_conc_init (&major_collector);
3187 case SGEN_MAJOR_CONCURRENT_PARALLEL:
3188 sgen_marksweep_conc_par_init (&major_collector);
3191 g_assert_not_reached ();
3196 * If sgen mode is set, major/minor configuration is fixed. The other gc_params
3197 * are parsed and processed after major/minor initialization, so it can potentially
3198 * override some knobs set by the sgen mode. We can consider locking out additional
3199 * configurations when gc_modes are used.
3202 init_sgen_mode (SgenMode mode)
3204 SgenMinor minor = SGEN_MINOR_DEFAULT;
3205 SgenMajor major = SGEN_MAJOR_DEFAULT;
3208 case SGEN_MODE_BALANCED:
3210 * Use a dynamic parallel nursery with a major concurrent collector.
3211 * This uses the default values for max pause time and nursery size.
3213 minor = SGEN_MINOR_SIMPLE;
3214 major = SGEN_MAJOR_CONCURRENT;
3215 dynamic_nursery = TRUE;
3217 case SGEN_MODE_THROUGHPUT:
3219 * Use concurrent major to let the mutator do more work. Use a larger
3220 * nursery, without pause time constraints, in order to collect more
3221 * objects in parallel and avoid repetitive collection tasks (pinning,
3222 * root scanning etc)
3224 minor = SGEN_MINOR_SIMPLE_PARALLEL;
3225 major = SGEN_MAJOR_CONCURRENT;
3226 dynamic_nursery = TRUE;
3227 sgen_max_pause_time = 0;
3229 case SGEN_MODE_PAUSE:
3231 * Use concurrent major and dynamic nursery with a more
3232 * aggressive shrinking relative to pause times.
3234 minor = SGEN_MINOR_SIMPLE_PARALLEL;
3235 major = SGEN_MAJOR_CONCURRENT;
3236 dynamic_nursery = TRUE;
3237 sgen_max_pause_margin = SGEN_PAUSE_MODE_MAX_PAUSE_MARGIN;
3240 g_assert_not_reached ();
3243 init_sgen_minor (minor);
3244 init_sgen_major (major);
3252 SgenMajor sgen_major = SGEN_MAJOR_DEFAULT;
3253 SgenMinor sgen_minor = SGEN_MINOR_DEFAULT;
3254 SgenMode sgen_mode = SGEN_MODE_NONE;
3255 char *params_opts = NULL;
3256 char *debug_opts = NULL;
3257 size_t max_heap = 0;
3258 size_t soft_limit = 0;
3260 gboolean debug_print_allowance = FALSE;
3261 double allowance_ratio = 0, save_target = 0;
3262 gboolean cement_enabled = TRUE;
3265 result = InterlockedCompareExchange (&gc_initialized, -1, 0);
3268 /* already inited */
3271 /* being inited by another thread */
3272 mono_thread_info_usleep (1000);
3275 /* we will init it */
3278 g_assert_not_reached ();
3280 } while (result != 0);
3282 SGEN_TV_GETTIME (sgen_init_timestamp);
3284 #ifdef SGEN_WITHOUT_MONO
3285 mono_thread_smr_init ();
3288 mono_coop_mutex_init (&gc_mutex);
3290 gc_debug_file = stderr;
3292 mono_coop_mutex_init (&sgen_interruption_mutex);
3294 if ((env = g_getenv (MONO_GC_PARAMS_NAME)) || gc_params_options) {
3295 params_opts = g_strdup_printf ("%s,%s", gc_params_options ? gc_params_options : "", env ? env : "");
3300 opts = g_strsplit (params_opts, ",", -1);
3301 for (ptr = opts; *ptr; ++ptr) {
3303 if (g_str_has_prefix (opt, "major=")) {
3304 opt = strchr (opt, '=') + 1;
3305 sgen_major = parse_sgen_major (opt);
3306 } else if (g_str_has_prefix (opt, "minor=")) {
3307 opt = strchr (opt, '=') + 1;
3308 sgen_minor = parse_sgen_minor (opt);
3309 } else if (g_str_has_prefix (opt, "mode=")) {
3310 opt = strchr (opt, '=') + 1;
3311 sgen_mode = parse_sgen_mode (opt);
3319 sgen_init_internal_allocator ();
3320 sgen_init_nursery_allocator ();
3321 sgen_init_fin_weak_hash ();
3322 sgen_init_hash_table ();
3323 sgen_init_descriptors ();
3324 sgen_init_gray_queues ();
3325 sgen_init_allocator ();
3326 sgen_init_gchandles ();
3328 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_SECTION, SGEN_SIZEOF_GC_MEM_SECTION);
3329 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_GRAY_QUEUE, sizeof (GrayQueueSection));
3331 sgen_client_init ();
3333 if (sgen_mode != SGEN_MODE_NONE) {
3334 if (sgen_minor != SGEN_MINOR_DEFAULT || sgen_major != SGEN_MAJOR_DEFAULT)
3335 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Ignoring major/minor configuration", "Major/minor configurations cannot be used with sgen modes");
3336 init_sgen_mode (sgen_mode);
3338 init_sgen_minor (sgen_minor);
3339 init_sgen_major (sgen_major);
3343 gboolean usage_printed = FALSE;
3345 for (ptr = opts; *ptr; ++ptr) {
3347 if (!strcmp (opt, ""))
3349 if (g_str_has_prefix (opt, "major="))
3351 if (g_str_has_prefix (opt, "minor="))
3353 if (g_str_has_prefix (opt, "mode=")) {
3354 if (g_str_has_prefix (opt, "mode=pause:")) {
3355 char *str_pause = strchr (opt, ':') + 1;
3356 int pause = atoi (str_pause);
3358 sgen_max_pause_time = pause;
3360 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default", "Invalid maximum pause time for `pause` sgen mode");
3364 if (g_str_has_prefix (opt, "max-heap-size=")) {
3365 size_t page_size = mono_pagesize ();
3366 size_t max_heap_candidate = 0;
3367 opt = strchr (opt, '=') + 1;
3368 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &max_heap_candidate)) {
3369 max_heap = (max_heap_candidate + page_size - 1) & ~(size_t)(page_size - 1);
3370 if (max_heap != max_heap_candidate)
3371 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Rounding up.", "`max-heap-size` size must be a multiple of %d.", page_size);
3373 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`max-heap-size` must be an integer.");
3377 if (g_str_has_prefix (opt, "soft-heap-limit=")) {
3378 opt = strchr (opt, '=') + 1;
3379 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &soft_limit)) {
3380 if (soft_limit <= 0) {
3381 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be positive.");
3385 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be an integer.");
3389 if (g_str_has_prefix (opt, "nursery-size=")) {
3391 opt = strchr (opt, '=') + 1;
3392 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &val)) {
3393 if ((val & (val - 1))) {
3394 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be a power of two.");
3398 if (val < SGEN_MAX_NURSERY_WASTE) {
3399 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.",
3400 "`nursery-size` must be at least %d bytes.", SGEN_MAX_NURSERY_WASTE);
3404 min_nursery_size = max_nursery_size = val;
3405 dynamic_nursery = FALSE;
3407 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be an integer.");
3412 if (g_str_has_prefix (opt, "save-target-ratio=")) {
3414 opt = strchr (opt, '=') + 1;
3415 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "save-target-ratio", opt,
3416 SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO, &val)) {
3421 if (g_str_has_prefix (opt, "default-allowance-ratio=")) {
3423 opt = strchr (opt, '=') + 1;
3424 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "default-allowance-ratio", opt,
3425 SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, SGEN_MAX_ALLOWANCE_NURSERY_SIZE_RATIO, &val)) {
3426 allowance_ratio = val;
3431 if (!strcmp (opt, "cementing")) {
3432 cement_enabled = TRUE;
3435 if (!strcmp (opt, "no-cementing")) {
3436 cement_enabled = FALSE;
3440 if (!strcmp (opt, "precleaning")) {
3441 precleaning_enabled = TRUE;
3444 if (!strcmp (opt, "no-precleaning")) {
3445 precleaning_enabled = FALSE;
3449 if (!strcmp (opt, "dynamic-nursery")) {
3450 if (sgen_minor_collector.is_split)
3451 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.",
3452 "dynamic-nursery not supported with split-nursery.");
3454 dynamic_nursery = TRUE;
3457 if (!strcmp (opt, "no-dynamic-nursery")) {
3458 dynamic_nursery = FALSE;
3462 if (major_collector.handle_gc_param && major_collector.handle_gc_param (opt))
3465 if (sgen_minor_collector.handle_gc_param && sgen_minor_collector.handle_gc_param (opt))
3468 if (sgen_client_handle_gc_param (opt))
3471 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Ignoring.", "Unknown option `%s`.", opt);
3476 fprintf (stderr, "\n%s must be a comma-delimited list of one or more of the following:\n", MONO_GC_PARAMS_NAME);
3477 fprintf (stderr, " max-heap-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
3478 fprintf (stderr, " soft-heap-limit=n (where N is an integer, possibly with a k, m or a g suffix)\n");
3479 fprintf (stderr, " mode=MODE (where MODE is 'balanced', 'throughput' or 'pause[:N]' and N is maximum pause in milliseconds)\n");
3480 fprintf (stderr, " nursery-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
3481 fprintf (stderr, " major=COLLECTOR (where COLLECTOR is `marksweep', `marksweep-conc', `marksweep-par')\n");
3482 fprintf (stderr, " minor=COLLECTOR (where COLLECTOR is `simple' or `split')\n");
3483 fprintf (stderr, " wbarrier=WBARRIER (where WBARRIER is `remset' or `cardtable')\n");
3484 fprintf (stderr, " [no-]cementing\n");
3485 fprintf (stderr, " [no-]dynamic-nursery\n");
3486 if (major_collector.print_gc_param_usage)
3487 major_collector.print_gc_param_usage ();
3488 if (sgen_minor_collector.print_gc_param_usage)
3489 sgen_minor_collector.print_gc_param_usage ();
3490 sgen_client_print_gc_params_usage ();
3491 fprintf (stderr, " Experimental options:\n");
3492 fprintf (stderr, " save-target-ratio=R (where R must be between %.2f - %.2f).\n", SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO);
3493 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);
3494 fprintf (stderr, "\n");
3496 usage_printed = TRUE;
3502 g_free (params_opts);
3504 alloc_nursery (dynamic_nursery, min_nursery_size, max_nursery_size);
3506 sgen_pinning_init ();
3507 sgen_cement_init (cement_enabled);
3509 if ((env = g_getenv (MONO_GC_DEBUG_NAME)) || gc_debug_options) {
3510 debug_opts = g_strdup_printf ("%s,%s", gc_debug_options ? gc_debug_options : "", env ? env : "");
3515 gboolean usage_printed = FALSE;
3517 opts = g_strsplit (debug_opts, ",", -1);
3518 for (ptr = opts; ptr && *ptr; ptr ++) {
3520 if (!strcmp (opt, ""))
3522 if (opt [0] >= '0' && opt [0] <= '9') {
3523 gc_debug_level = atoi (opt);
3528 char *rf = g_strdup_printf ("%s.%d", opt, mono_process_current_pid ());
3529 gc_debug_file = fopen (rf, "wb");
3531 gc_debug_file = stderr;
3534 } else if (!strcmp (opt, "print-allowance")) {
3535 debug_print_allowance = TRUE;
3536 } else if (!strcmp (opt, "print-pinning")) {
3537 sgen_pin_stats_enable ();
3538 } else if (!strcmp (opt, "verify-before-allocs")) {
3539 verify_before_allocs = 1;
3540 has_per_allocation_action = TRUE;
3541 } else if (g_str_has_prefix (opt, "max-valloc-size=")) {
3542 size_t max_valloc_size;
3543 char *arg = strchr (opt, '=') + 1;
3544 if (*opt && mono_gc_parse_environment_string_extract_number (arg, &max_valloc_size)) {
3545 mono_valloc_set_limit (max_valloc_size);
3547 sgen_env_var_error (MONO_GC_DEBUG_NAME, NULL, "`max-valloc-size` must be an integer.");
3550 } else if (g_str_has_prefix (opt, "verify-before-allocs=")) {
3551 char *arg = strchr (opt, '=') + 1;
3552 verify_before_allocs = atoi (arg);
3553 has_per_allocation_action = TRUE;
3554 } else if (!strcmp (opt, "collect-before-allocs")) {
3555 collect_before_allocs = 1;
3556 has_per_allocation_action = TRUE;
3557 } else if (g_str_has_prefix (opt, "collect-before-allocs=")) {
3558 char *arg = strchr (opt, '=') + 1;
3559 has_per_allocation_action = TRUE;
3560 collect_before_allocs = atoi (arg);
3561 } else if (!strcmp (opt, "verify-before-collections")) {
3562 whole_heap_check_before_collection = TRUE;
3563 } else if (!strcmp (opt, "check-remset-consistency")) {
3564 remset_consistency_checks = TRUE;
3565 nursery_clear_policy = CLEAR_AT_GC;
3566 } else if (!strcmp (opt, "mod-union-consistency-check")) {
3567 if (!major_collector.is_concurrent) {
3568 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`mod-union-consistency-check` only works with concurrent major collector.");
3571 mod_union_consistency_check = TRUE;
3572 } else if (!strcmp (opt, "check-mark-bits")) {
3573 check_mark_bits_after_major_collection = TRUE;
3574 } else if (!strcmp (opt, "check-nursery-pinned")) {
3575 check_nursery_objects_pinned = TRUE;
3576 } else if (!strcmp (opt, "clear-at-gc")) {
3577 nursery_clear_policy = CLEAR_AT_GC;
3578 } else if (!strcmp (opt, "clear-nursery-at-gc")) {
3579 nursery_clear_policy = CLEAR_AT_GC;
3580 } else if (!strcmp (opt, "clear-at-tlab-creation")) {
3581 nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
3582 } else if (!strcmp (opt, "debug-clear-at-tlab-creation")) {
3583 nursery_clear_policy = CLEAR_AT_TLAB_CREATION_DEBUG;
3584 } else if (!strcmp (opt, "check-scan-starts")) {
3585 do_scan_starts_check = TRUE;
3586 } else if (!strcmp (opt, "verify-nursery-at-minor-gc")) {
3587 do_verify_nursery = TRUE;
3588 } else if (!strcmp (opt, "check-concurrent")) {
3589 if (!major_collector.is_concurrent) {
3590 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`check-concurrent` only works with concurrent major collectors.");
3593 nursery_clear_policy = CLEAR_AT_GC;
3594 do_concurrent_checks = TRUE;
3595 } else if (!strcmp (opt, "dump-nursery-at-minor-gc")) {
3596 do_dump_nursery_content = TRUE;
3597 } else if (!strcmp (opt, "disable-minor")) {
3598 disable_minor_collections = TRUE;
3599 } else if (!strcmp (opt, "disable-major")) {
3600 disable_major_collections = TRUE;
3601 } else if (g_str_has_prefix (opt, "heap-dump=")) {
3602 char *filename = strchr (opt, '=') + 1;
3603 nursery_clear_policy = CLEAR_AT_GC;
3604 sgen_debug_enable_heap_dump (filename);
3605 } else if (g_str_has_prefix (opt, "binary-protocol=")) {
3606 char *filename = strchr (opt, '=') + 1;
3607 char *colon = strrchr (filename, ':');
3610 if (!mono_gc_parse_environment_string_extract_number (colon + 1, &limit)) {
3611 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring limit.", "Binary protocol file size limit must be an integer.");
3616 binary_protocol_init (filename, (long long)limit);
3617 } else if (!strcmp (opt, "nursery-canaries")) {
3618 do_verify_nursery = TRUE;
3619 enable_nursery_canaries = TRUE;
3620 } else if (!sgen_client_handle_gc_debug (opt)) {
3621 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "Unknown option `%s`.", opt);
3626 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);
3627 fprintf (stderr, "Valid <option>s are:\n");
3628 fprintf (stderr, " collect-before-allocs[=<n>]\n");
3629 fprintf (stderr, " verify-before-allocs[=<n>]\n");
3630 fprintf (stderr, " max-valloc-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
3631 fprintf (stderr, " check-remset-consistency\n");
3632 fprintf (stderr, " check-mark-bits\n");
3633 fprintf (stderr, " check-nursery-pinned\n");
3634 fprintf (stderr, " verify-before-collections\n");
3635 fprintf (stderr, " verify-nursery-at-minor-gc\n");
3636 fprintf (stderr, " dump-nursery-at-minor-gc\n");
3637 fprintf (stderr, " disable-minor\n");
3638 fprintf (stderr, " disable-major\n");
3639 fprintf (stderr, " check-concurrent\n");
3640 fprintf (stderr, " clear-[nursery-]at-gc\n");
3641 fprintf (stderr, " clear-at-tlab-creation\n");
3642 fprintf (stderr, " debug-clear-at-tlab-creation\n");
3643 fprintf (stderr, " check-scan-starts\n");
3644 fprintf (stderr, " print-allowance\n");
3645 fprintf (stderr, " print-pinning\n");
3646 fprintf (stderr, " heap-dump=<filename>\n");
3647 fprintf (stderr, " binary-protocol=<filename>[:<file-size-limit>]\n");
3648 fprintf (stderr, " nursery-canaries\n");
3649 sgen_client_print_gc_debug_usage ();
3650 fprintf (stderr, "\n");
3652 usage_printed = TRUE;
3659 g_free (debug_opts);
3661 if (check_mark_bits_after_major_collection)
3662 nursery_clear_policy = CLEAR_AT_GC;
3664 if (major_collector.post_param_init)
3665 major_collector.post_param_init (&major_collector);
3667 sgen_thread_pool_start ();
3669 sgen_memgov_init (max_heap, soft_limit, debug_print_allowance, allowance_ratio, save_target);
3671 memset (&remset, 0, sizeof (remset));
3673 sgen_card_table_init (&remset);
3675 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");
3679 sgen_init_bridge ();
3683 sgen_gc_initialized ()
3685 return gc_initialized > 0;
3689 sgen_get_nursery_clear_policy (void)
3691 return nursery_clear_policy;
3697 mono_coop_mutex_lock (&gc_mutex);
3701 sgen_gc_unlock (void)
3703 mono_coop_mutex_unlock (&gc_mutex);
3707 sgen_major_collector_iterate_live_block_ranges (sgen_cardtable_block_callback callback)
3709 major_collector.iterate_live_block_ranges (callback);
3713 sgen_major_collector_iterate_block_ranges (sgen_cardtable_block_callback callback)
3715 major_collector.iterate_block_ranges (callback);
3719 sgen_get_major_collector (void)
3721 return &major_collector;
3725 sgen_get_minor_collector (void)
3727 return &sgen_minor_collector;
3731 sgen_get_remset (void)
3737 count_cards (long long *major_total, long long *major_marked, long long *los_total, long long *los_marked)
3739 sgen_get_major_collector ()->count_cards (major_total, major_marked);
3740 sgen_los_count_cards (los_total, los_marked);
3743 static gboolean world_is_stopped = FALSE;
3745 /* LOCKING: assumes the GC lock is held */
3747 sgen_stop_world (int generation)
3749 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3751 SGEN_ASSERT (0, !world_is_stopped, "Why are we stopping a stopped world?");
3753 binary_protocol_world_stopping (generation, sgen_timestamp (), (gpointer) (gsize) mono_native_thread_id_get ());
3755 sgen_client_stop_world (generation);
3757 world_is_stopped = TRUE;
3759 if (binary_protocol_is_heavy_enabled ())
3760 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3761 binary_protocol_world_stopped (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3764 /* LOCKING: assumes the GC lock is held */
3766 sgen_restart_world (int generation)
3768 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3771 SGEN_ASSERT (0, world_is_stopped, "Why are we restarting a running world?");
3773 if (binary_protocol_is_heavy_enabled ())
3774 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3775 binary_protocol_world_restarting (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3777 world_is_stopped = FALSE;
3779 sgen_client_restart_world (generation, &stw_time);
3781 binary_protocol_world_restarted (generation, sgen_timestamp ());
3783 if (sgen_client_bridge_need_processing ())
3784 sgen_client_bridge_processing_finish (generation);
3786 sgen_memgov_collection_end (generation, stw_time);
3790 sgen_is_world_stopped (void)
3792 return world_is_stopped;
3796 sgen_check_whole_heap_stw (void)
3798 sgen_stop_world (0);
3799 sgen_clear_nursery_fragments ();
3800 sgen_check_whole_heap (TRUE);
3801 sgen_restart_world (0);
3805 sgen_timestamp (void)
3807 SGEN_TV_DECLARE (timestamp);
3808 SGEN_TV_GETTIME (timestamp);
3809 return SGEN_TV_ELAPSED (sgen_init_timestamp, timestamp);
3812 #endif /* HAVE_SGEN_GC */