2 * sgen-gc.c: Simple generational GC.
5 * Paolo Molaro (lupus@ximian.com)
6 * Rodrigo Kumpera (kumpera@gmail.com)
8 * Copyright 2005-2011 Novell, Inc (http://www.novell.com)
9 * Copyright 2011 Xamarin Inc (http://www.xamarin.com)
11 * Thread start/stop adapted from Boehm's GC:
12 * Copyright (c) 1994 by Xerox Corporation. All rights reserved.
13 * Copyright (c) 1996 by Silicon Graphics. All rights reserved.
14 * Copyright (c) 1998 by Fergus Henderson. All rights reserved.
15 * Copyright (c) 2000-2004 by Hewlett-Packard Company. All rights reserved.
16 * Copyright 2001-2003 Ximian, Inc
17 * Copyright 2003-2010 Novell, Inc.
18 * Copyright 2011 Xamarin, Inc.
19 * Copyright (C) 2012 Xamarin Inc
21 * Licensed under the MIT license. See LICENSE file in the project root for full license information.
23 * Important: allocation provides always zeroed memory, having to do
24 * a memset after allocation is deadly for performance.
25 * Memory usage at startup is currently as follows:
27 * 64 KB internal space
29 * We should provide a small memory config with half the sizes
31 * We currently try to make as few mono assumptions as possible:
32 * 1) 2-word header with no GC pointers in it (first vtable, second to store the
34 * 2) gc descriptor is the second word in the vtable (first word in the class)
35 * 3) 8 byte alignment is the minimum and enough (not true for special structures (SIMD), FIXME)
36 * 4) there is a function to get an object's size and the number of
37 * elements in an array.
38 * 5) we know the special way bounds are allocated for complex arrays
39 * 6) we know about proxies and how to treat them when domains are unloaded
41 * Always try to keep stack usage to a minimum: no recursive behaviour
42 * and no large stack allocs.
44 * General description.
45 * Objects are initially allocated in a nursery using a fast bump-pointer technique.
46 * When the nursery is full we start a nursery collection: this is performed with a
48 * When the old generation is full we start a copying GC of the old generation as well:
49 * this will be changed to mark&sweep with copying when fragmentation becomes to severe
50 * in the future. Maybe we'll even do both during the same collection like IMMIX.
52 * The things that complicate this description are:
53 * *) pinned objects: we can't move them so we need to keep track of them
54 * *) no precise info of the thread stacks and registers: we need to be able to
55 * quickly find the objects that may be referenced conservatively and pin them
56 * (this makes the first issues more important)
57 * *) large objects are too expensive to be dealt with using copying GC: we handle them
58 * with mark/sweep during major collections
59 * *) some objects need to not move even if they are small (interned strings, Type handles):
60 * we use mark/sweep for them, too: they are not allocated in the nursery, but inside
61 * PinnedChunks regions
67 *) we could have a function pointer in MonoClass to implement
68 customized write barriers for value types
70 *) investigate the stuff needed to advance a thread to a GC-safe
71 point (single-stepping, read from unmapped memory etc) and implement it.
72 This would enable us to inline allocations and write barriers, for example,
73 or at least parts of them, like the write barrier checks.
74 We may need this also for handling precise info on stacks, even simple things
75 as having uninitialized data on the stack and having to wait for the prolog
76 to zero it. Not an issue for the last frame that we scan conservatively.
77 We could always not trust the value in the slots anyway.
79 *) modify the jit to save info about references in stack locations:
80 this can be done just for locals as a start, so that at least
81 part of the stack is handled precisely.
83 *) test/fix endianess issues
85 *) Implement a card table as the write barrier instead of remembered
86 sets? Card tables are not easy to implement with our current
87 memory layout. We have several different kinds of major heap
88 objects: Small objects in regular blocks, small objects in pinned
89 chunks and LOS objects. If we just have a pointer we have no way
90 to tell which kind of object it points into, therefore we cannot
91 know where its card table is. The least we have to do to make
92 this happen is to get rid of write barriers for indirect stores.
95 *) Get rid of write barriers for indirect stores. We can do this by
96 telling the GC to wbarrier-register an object once we do an ldloca
97 or ldelema on it, and to unregister it once it's not used anymore
98 (it can only travel downwards on the stack). The problem with
99 unregistering is that it needs to happen eventually no matter
100 what, even if exceptions are thrown, the thread aborts, etc.
101 Rodrigo suggested that we could do only the registering part and
102 let the collector find out (pessimistically) when it's safe to
103 unregister, namely when the stack pointer of the thread that
104 registered the object is higher than it was when the registering
105 happened. This might make for a good first implementation to get
106 some data on performance.
108 *) Some sort of blacklist support? Blacklists is a concept from the
109 Boehm GC: if during a conservative scan we find pointers to an
110 area which we might use as heap, we mark that area as unusable, so
111 pointer retention by random pinning pointers is reduced.
113 *) experiment with max small object size (very small right now - 2kb,
114 because it's tied to the max freelist size)
116 *) add an option to mmap the whole heap in one chunk: it makes for many
117 simplifications in the checks (put the nursery at the top and just use a single
118 check for inclusion/exclusion): the issue this has is that on 32 bit systems it's
119 not flexible (too much of the address space may be used by default or we can't
120 increase the heap as needed) and we'd need a race-free mechanism to return memory
121 back to the system (mprotect(PROT_NONE) will still keep the memory allocated if it
122 was written to, munmap is needed, but the following mmap may not find the same segment
125 *) memzero the major fragments after restarting the world and optionally a smaller
128 *) investigate having fragment zeroing threads
130 *) separate locks for finalization and other minor stuff to reduce
133 *) try a different copying order to improve memory locality
135 *) a thread abort after a store but before the write barrier will
136 prevent the write barrier from executing
138 *) specialized dynamically generated markers/copiers
140 *) Dynamically adjust TLAB size to the number of threads. If we have
141 too many threads that do allocation, we might need smaller TLABs,
142 and we might get better performance with larger TLABs if we only
143 have a handful of threads. We could sum up the space left in all
144 assigned TLABs and if that's more than some percentage of the
145 nursery size, reduce the TLAB size.
147 *) Explore placing unreachable objects on unused nursery memory.
148 Instead of memset'ng a region to zero, place an int[] covering it.
149 A good place to start is add_nursery_frag. The tricky thing here is
150 placing those objects atomically outside of a collection.
152 *) Allocation should use asymmetric Dekker synchronization:
153 http://blogs.oracle.com/dave/resource/Asymmetric-Dekker-Synchronization.txt
154 This should help weak consistency archs.
161 #define _XOPEN_SOURCE
162 #define _DARWIN_C_SOURCE
168 #ifdef HAVE_PTHREAD_H
171 #ifdef HAVE_PTHREAD_NP_H
172 #include <pthread_np.h>
180 #include "mono/sgen/sgen-gc.h"
181 #include "mono/sgen/sgen-cardtable.h"
182 #include "mono/sgen/sgen-protocol.h"
183 #include "mono/sgen/sgen-memory-governor.h"
184 #include "mono/sgen/sgen-hash-table.h"
185 #include "mono/sgen/sgen-cardtable.h"
186 #include "mono/sgen/sgen-pinning.h"
187 #include "mono/sgen/sgen-workers.h"
188 #include "mono/sgen/sgen-client.h"
189 #include "mono/sgen/sgen-pointer-queue.h"
190 #include "mono/sgen/gc-internal-agnostic.h"
191 #include "mono/utils/mono-proclib.h"
192 #include "mono/utils/mono-memory-model.h"
193 #include "mono/utils/hazard-pointer.h"
195 #include <mono/utils/memcheck.h>
197 #undef pthread_create
199 #undef pthread_detach
202 * ######################################################################
203 * ######## Types and constants used by the GC.
204 * ######################################################################
207 /* 0 means not initialized, 1 is initialized, -1 means in progress */
208 static int gc_initialized = 0;
209 /* If set, check if we need to do something every X allocations */
210 gboolean has_per_allocation_action;
211 /* If set, do a heap check every X allocation */
212 guint32 verify_before_allocs = 0;
213 /* If set, do a minor collection before every X allocation */
214 guint32 collect_before_allocs = 0;
215 /* If set, do a whole heap check before each collection */
216 static gboolean whole_heap_check_before_collection = FALSE;
217 /* If set, do a heap consistency check before each minor collection */
218 static gboolean consistency_check_at_minor_collection = FALSE;
219 /* If set, do a mod union consistency check before each finishing collection pause */
220 static gboolean mod_union_consistency_check = FALSE;
221 /* If set, check whether mark bits are consistent after major collections */
222 static gboolean check_mark_bits_after_major_collection = FALSE;
223 /* If set, check that all nursery objects are pinned/not pinned, depending on context */
224 static gboolean check_nursery_objects_pinned = FALSE;
225 /* If set, do a few checks when the concurrent collector is used */
226 static gboolean do_concurrent_checks = FALSE;
227 /* If set, do a plausibility check on the scan_starts before and after
229 static gboolean do_scan_starts_check = FALSE;
231 static gboolean disable_minor_collections = FALSE;
232 static gboolean disable_major_collections = FALSE;
233 static gboolean do_verify_nursery = FALSE;
234 static gboolean do_dump_nursery_content = FALSE;
235 static gboolean enable_nursery_canaries = FALSE;
237 static gboolean precleaning_enabled = TRUE;
239 #ifdef HEAVY_STATISTICS
240 guint64 stat_objects_alloced_degraded = 0;
241 guint64 stat_bytes_alloced_degraded = 0;
243 guint64 stat_copy_object_called_nursery = 0;
244 guint64 stat_objects_copied_nursery = 0;
245 guint64 stat_copy_object_called_major = 0;
246 guint64 stat_objects_copied_major = 0;
248 guint64 stat_scan_object_called_nursery = 0;
249 guint64 stat_scan_object_called_major = 0;
251 guint64 stat_slots_allocated_in_vain;
253 guint64 stat_nursery_copy_object_failed_from_space = 0;
254 guint64 stat_nursery_copy_object_failed_forwarded = 0;
255 guint64 stat_nursery_copy_object_failed_pinned = 0;
256 guint64 stat_nursery_copy_object_failed_to_space = 0;
258 static guint64 stat_wbarrier_add_to_global_remset = 0;
259 static guint64 stat_wbarrier_arrayref_copy = 0;
260 static guint64 stat_wbarrier_generic_store = 0;
261 static guint64 stat_wbarrier_generic_store_atomic = 0;
262 static guint64 stat_wbarrier_set_root = 0;
265 static guint64 stat_pinned_objects = 0;
267 static guint64 time_minor_pre_collection_fragment_clear = 0;
268 static guint64 time_minor_pinning = 0;
269 static guint64 time_minor_scan_remsets = 0;
270 static guint64 time_minor_scan_pinned = 0;
271 static guint64 time_minor_scan_roots = 0;
272 static guint64 time_minor_finish_gray_stack = 0;
273 static guint64 time_minor_fragment_creation = 0;
275 static guint64 time_major_pre_collection_fragment_clear = 0;
276 static guint64 time_major_pinning = 0;
277 static guint64 time_major_scan_pinned = 0;
278 static guint64 time_major_scan_roots = 0;
279 static guint64 time_major_scan_mod_union = 0;
280 static guint64 time_major_finish_gray_stack = 0;
281 static guint64 time_major_free_bigobjs = 0;
282 static guint64 time_major_los_sweep = 0;
283 static guint64 time_major_sweep = 0;
284 static guint64 time_major_fragment_creation = 0;
286 static guint64 time_max = 0;
288 static SGEN_TV_DECLARE (time_major_conc_collection_start);
289 static SGEN_TV_DECLARE (time_major_conc_collection_end);
291 static SGEN_TV_DECLARE (last_minor_collection_start_tv);
292 static SGEN_TV_DECLARE (last_minor_collection_end_tv);
294 int gc_debug_level = 0;
299 mono_gc_flush_info (void)
301 fflush (gc_debug_file);
305 #define TV_DECLARE SGEN_TV_DECLARE
306 #define TV_GETTIME SGEN_TV_GETTIME
307 #define TV_ELAPSED SGEN_TV_ELAPSED
309 static SGEN_TV_DECLARE (sgen_init_timestamp);
311 NurseryClearPolicy nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
313 #define object_is_forwarded SGEN_OBJECT_IS_FORWARDED
314 #define object_is_pinned SGEN_OBJECT_IS_PINNED
315 #define pin_object SGEN_PIN_OBJECT
317 #define ptr_in_nursery sgen_ptr_in_nursery
319 #define LOAD_VTABLE SGEN_LOAD_VTABLE
322 nursery_canaries_enabled (void)
324 return enable_nursery_canaries;
327 #define safe_object_get_size sgen_safe_object_get_size
330 * ######################################################################
331 * ######## Global data.
332 * ######################################################################
334 MonoCoopMutex gc_mutex;
336 #define SCAN_START_SIZE SGEN_SCAN_START_SIZE
338 size_t degraded_mode = 0;
340 static mword bytes_pinned_from_failed_allocation = 0;
342 GCMemSection *nursery_section = NULL;
343 static volatile mword lowest_heap_address = ~(mword)0;
344 static volatile mword highest_heap_address = 0;
346 MonoCoopMutex sgen_interruption_mutex;
348 int current_collection_generation = -1;
349 static volatile gboolean concurrent_collection_in_progress = FALSE;
351 /* objects that are ready to be finalized */
352 static SgenPointerQueue fin_ready_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
353 static SgenPointerQueue critical_fin_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
355 /* registered roots: the key to the hash is the root start address */
357 * Different kinds of roots are kept separate to speed up pin_from_roots () for example.
359 SgenHashTable roots_hash [ROOT_TYPE_NUM] = {
360 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
361 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
362 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL)
364 static mword roots_size = 0; /* amount of memory in the root set */
366 /* The size of a TLAB */
367 /* The bigger the value, the less often we have to go to the slow path to allocate a new
368 * one, but the more space is wasted by threads not allocating much memory.
370 * FIXME: Make this self-tuning for each thread.
372 guint32 tlab_size = (1024 * 4);
374 #define MAX_SMALL_OBJ_SIZE SGEN_MAX_SMALL_OBJ_SIZE
376 #define ALLOC_ALIGN SGEN_ALLOC_ALIGN
378 #define ALIGN_UP SGEN_ALIGN_UP
380 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
381 MonoNativeThreadId main_gc_thread = NULL;
384 /*Object was pinned during the current collection*/
385 static mword objects_pinned;
388 * ######################################################################
389 * ######## Macros and function declarations.
390 * ######################################################################
393 typedef SgenGrayQueue GrayQueue;
395 /* forward declarations */
396 static void scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx);
398 static void pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx);
399 static void finish_gray_stack (int generation, ScanCopyContext ctx);
402 SgenMajorCollector major_collector;
403 SgenMinorCollector sgen_minor_collector;
404 /* FIXME: get rid of this */
405 static GrayQueue gray_queue;
407 static SgenRememberedSet remset;
409 /* The gray queue to use from the main collection thread. */
410 #define WORKERS_DISTRIBUTE_GRAY_QUEUE (&gray_queue)
413 * The gray queue a worker job must use. If we're not parallel or
414 * concurrent, we use the main gray queue.
416 static SgenGrayQueue*
417 sgen_workers_get_job_gray_queue (WorkerData *worker_data)
419 return worker_data ? &worker_data->private_gray_queue : WORKERS_DISTRIBUTE_GRAY_QUEUE;
423 gray_queue_redirect (SgenGrayQueue *queue)
425 gboolean wake = FALSE;
428 GrayQueueSection *section = sgen_gray_object_dequeue_section (queue);
431 sgen_section_gray_queue_enqueue ((SgenSectionGrayQueue *)queue->alloc_prepare_data, section);
436 g_assert (concurrent_collection_in_progress);
437 sgen_workers_ensure_awake ();
442 gray_queue_enable_redirect (SgenGrayQueue *queue)
444 if (!concurrent_collection_in_progress)
447 sgen_gray_queue_set_alloc_prepare (queue, gray_queue_redirect, sgen_workers_get_distribute_section_gray_queue ());
448 gray_queue_redirect (queue);
452 sgen_scan_area_with_callback (char *start, char *end, IterateObjectCallbackFunc callback, void *data, gboolean allow_flags, gboolean fail_on_canaries)
454 while (start < end) {
458 if (!*(void**)start) {
459 start += sizeof (void*); /* should be ALLOC_ALIGN, really */
464 if (!(obj = (char *)SGEN_OBJECT_IS_FORWARDED (start)))
470 if (!sgen_client_object_is_array_fill ((GCObject*)obj)) {
471 CHECK_CANARY_FOR_OBJECT ((GCObject*)obj, fail_on_canaries);
472 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
473 callback ((GCObject*)obj, size, data);
474 CANARIFY_SIZE (size);
476 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
484 * sgen_add_to_global_remset:
486 * The global remset contains locations which point into newspace after
487 * a minor collection. This can happen if the objects they point to are pinned.
489 * LOCKING: If called from a parallel collector, the global remset
490 * lock must be held. For serial collectors that is not necessary.
493 sgen_add_to_global_remset (gpointer ptr, GCObject *obj)
495 SGEN_ASSERT (5, sgen_ptr_in_nursery (obj), "Target pointer of global remset must be in the nursery");
497 HEAVY_STAT (++stat_wbarrier_add_to_global_remset);
499 if (!major_collector.is_concurrent) {
500 SGEN_ASSERT (5, current_collection_generation != -1, "Global remsets can only be added during collections");
502 if (current_collection_generation == -1)
503 SGEN_ASSERT (5, sgen_concurrent_collection_in_progress (), "Global remsets outside of collection pauses can only be added by the concurrent collector");
506 if (!object_is_pinned (obj))
507 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");
508 else if (sgen_cement_lookup_or_register (obj))
511 remset.record_pointer (ptr);
513 sgen_pin_stats_register_global_remset (obj);
515 SGEN_LOG (8, "Adding global remset for %p", ptr);
516 binary_protocol_global_remset (ptr, obj, (gpointer)SGEN_LOAD_VTABLE (obj));
520 * sgen_drain_gray_stack:
522 * Scan objects in the gray stack until the stack is empty. This should be called
523 * frequently after each object is copied, to achieve better locality and cache
528 sgen_drain_gray_stack (ScanCopyContext ctx)
530 ScanObjectFunc scan_func = ctx.ops->scan_object;
531 GrayQueue *queue = ctx.queue;
533 if (ctx.ops->drain_gray_stack)
534 return ctx.ops->drain_gray_stack (queue);
539 GRAY_OBJECT_DEQUEUE (queue, &obj, &desc);
542 SGEN_LOG (9, "Precise gray object scan %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
543 scan_func (obj, desc, queue);
549 * Addresses in the pin queue are already sorted. This function finds
550 * the object header for each address and pins the object. The
551 * addresses must be inside the nursery section. The (start of the)
552 * address array is overwritten with the addresses of the actually
553 * pinned objects. Return the number of pinned objects.
556 pin_objects_from_nursery_pin_queue (gboolean do_scan_objects, ScanCopyContext ctx)
558 GCMemSection *section = nursery_section;
559 void **start = sgen_pinning_get_entry (section->pin_queue_first_entry);
560 void **end = sgen_pinning_get_entry (section->pin_queue_last_entry);
561 void *start_nursery = section->data;
562 void *end_nursery = section->next_data;
567 void *pinning_front = start_nursery;
569 void **definitely_pinned = start;
570 ScanObjectFunc scan_func = ctx.ops->scan_object;
571 SgenGrayQueue *queue = ctx.queue;
573 sgen_nursery_allocator_prepare_for_pinning ();
575 while (start < end) {
576 GCObject *obj_to_pin = NULL;
577 size_t obj_to_pin_size = 0;
582 SGEN_ASSERT (0, addr >= start_nursery && addr < end_nursery, "Potential pinning address out of range");
583 SGEN_ASSERT (0, addr >= last, "Pin queue not sorted");
590 SGEN_LOG (5, "Considering pinning addr %p", addr);
591 /* We've already processed everything up to pinning_front. */
592 if (addr < pinning_front) {
598 * Find the closest scan start <= addr. We might search backward in the
599 * scan_starts array because entries might be NULL. In the worst case we
600 * start at start_nursery.
602 idx = ((char*)addr - (char*)section->data) / SCAN_START_SIZE;
603 SGEN_ASSERT (0, idx < section->num_scan_start, "Scan start index out of range");
604 search_start = (void*)section->scan_starts [idx];
605 if (!search_start || search_start > addr) {
608 search_start = section->scan_starts [idx];
609 if (search_start && search_start <= addr)
612 if (!search_start || search_start > addr)
613 search_start = start_nursery;
617 * If the pinning front is closer than the scan start we found, start
618 * searching at the front.
620 if (search_start < pinning_front)
621 search_start = pinning_front;
624 * Now addr should be in an object a short distance from search_start.
626 * search_start must point to zeroed mem or point to an object.
629 size_t obj_size, canarified_obj_size;
632 if (!*(void**)search_start) {
633 search_start = (void*)ALIGN_UP ((mword)search_start + sizeof (gpointer));
634 /* The loop condition makes sure we don't overrun addr. */
638 canarified_obj_size = obj_size = ALIGN_UP (safe_object_get_size ((GCObject*)search_start));
641 * Filler arrays are marked by an invalid sync word. We don't
642 * consider them for pinning. They are not delimited by canaries,
645 if (!sgen_client_object_is_array_fill ((GCObject*)search_start)) {
646 CHECK_CANARY_FOR_OBJECT (search_start, TRUE);
647 CANARIFY_SIZE (canarified_obj_size);
649 if (addr >= search_start && (char*)addr < (char*)search_start + obj_size) {
650 /* This is the object we're looking for. */
651 obj_to_pin = (GCObject*)search_start;
652 obj_to_pin_size = canarified_obj_size;
657 /* Skip to the next object */
658 search_start = (void*)((char*)search_start + canarified_obj_size);
659 } while (search_start <= addr);
661 /* We've searched past the address we were looking for. */
663 pinning_front = search_start;
664 goto next_pin_queue_entry;
668 * We've found an object to pin. It might still be a dummy array, but we
669 * can advance the pinning front in any case.
671 pinning_front = (char*)obj_to_pin + obj_to_pin_size;
674 * If this is a dummy array marking the beginning of a nursery
675 * fragment, we don't pin it.
677 if (sgen_client_object_is_array_fill (obj_to_pin))
678 goto next_pin_queue_entry;
681 * Finally - pin the object!
683 desc = sgen_obj_get_descriptor_safe (obj_to_pin);
684 if (do_scan_objects) {
685 scan_func (obj_to_pin, desc, queue);
687 SGEN_LOG (4, "Pinned object %p, vtable %p (%s), count %d\n",
688 obj_to_pin, *(void**)obj_to_pin, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj_to_pin)), count);
689 binary_protocol_pin (obj_to_pin,
690 (gpointer)LOAD_VTABLE (obj_to_pin),
691 safe_object_get_size (obj_to_pin));
693 pin_object (obj_to_pin);
694 GRAY_OBJECT_ENQUEUE (queue, obj_to_pin, desc);
695 sgen_pin_stats_register_object (obj_to_pin, obj_to_pin_size);
696 definitely_pinned [count] = obj_to_pin;
699 if (concurrent_collection_in_progress)
700 sgen_pinning_register_pinned_in_nursery (obj_to_pin);
702 next_pin_queue_entry:
706 sgen_client_nursery_objects_pinned (definitely_pinned, count);
707 stat_pinned_objects += count;
712 pin_objects_in_nursery (gboolean do_scan_objects, ScanCopyContext ctx)
716 if (nursery_section->pin_queue_first_entry == nursery_section->pin_queue_last_entry)
719 reduced_to = pin_objects_from_nursery_pin_queue (do_scan_objects, ctx);
720 nursery_section->pin_queue_last_entry = nursery_section->pin_queue_first_entry + reduced_to;
724 * This function is only ever called (via `collector_pin_object()` in `sgen-copy-object.h`)
725 * when we can't promote an object because we're out of memory.
728 sgen_pin_object (GCObject *object, GrayQueue *queue)
731 * All pinned objects are assumed to have been staged, so we need to stage as well.
732 * Also, the count of staged objects shows that "late pinning" happened.
734 sgen_pin_stage_ptr (object);
736 SGEN_PIN_OBJECT (object);
737 binary_protocol_pin (object, (gpointer)LOAD_VTABLE (object), safe_object_get_size (object));
740 sgen_pin_stats_register_object (object, safe_object_get_size (object));
742 GRAY_OBJECT_ENQUEUE (queue, object, sgen_obj_get_descriptor_safe (object));
745 /* Sort the addresses in array in increasing order.
746 * Done using a by-the book heap sort. Which has decent and stable performance, is pretty cache efficient.
749 sgen_sort_addresses (void **array, size_t size)
754 for (i = 1; i < size; ++i) {
757 size_t parent = (child - 1) / 2;
759 if (array [parent] >= array [child])
762 tmp = array [parent];
763 array [parent] = array [child];
770 for (i = size - 1; i > 0; --i) {
773 array [i] = array [0];
779 while (root * 2 + 1 <= end) {
780 size_t child = root * 2 + 1;
782 if (child < end && array [child] < array [child + 1])
784 if (array [root] >= array [child])
788 array [root] = array [child];
797 * Scan the memory between start and end and queue values which could be pointers
798 * to the area between start_nursery and end_nursery for later consideration.
799 * Typically used for thread stacks.
802 sgen_conservatively_pin_objects_from (void **start, void **end, void *start_nursery, void *end_nursery, int pin_type)
806 SGEN_ASSERT (0, ((mword)start & (SIZEOF_VOID_P - 1)) == 0, "Why are we scanning for references in unaligned memory ?");
808 #if defined(VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE) && !defined(_WIN64)
809 VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE (start, (char*)end - (char*)start);
812 while (start < end) {
814 * *start can point to the middle of an object
815 * note: should we handle pointing at the end of an object?
816 * pinning in C# code disallows pointing at the end of an object
817 * but there is some small chance that an optimizing C compiler
818 * may keep the only reference to an object by pointing
819 * at the end of it. We ignore this small chance for now.
820 * Pointers to the end of an object are indistinguishable
821 * from pointers to the start of the next object in memory
822 * so if we allow that we'd need to pin two objects...
823 * We queue the pointer in an array, the
824 * array will then be sorted and uniqued. This way
825 * we can coalesce several pinning pointers and it should
826 * be faster since we'd do a memory scan with increasing
827 * addresses. Note: we can align the address to the allocation
828 * alignment, so the unique process is more effective.
830 mword addr = (mword)*start;
831 addr &= ~(ALLOC_ALIGN - 1);
832 if (addr >= (mword)start_nursery && addr < (mword)end_nursery) {
833 SGEN_LOG (6, "Pinning address %p from %p", (void*)addr, start);
834 sgen_pin_stage_ptr ((void*)addr);
835 binary_protocol_pin_stage (start, (void*)addr);
836 sgen_pin_stats_register_address ((char*)addr, pin_type);
842 SGEN_LOG (7, "found %d potential pinned heap pointers", count);
846 * The first thing we do in a collection is to identify pinned objects.
847 * This function considers all the areas of memory that need to be
848 * conservatively scanned.
851 pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx)
855 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);
856 /* objects pinned from the API are inside these roots */
857 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_PINNED], void **, start_root, RootRecord *, root) {
858 SGEN_LOG (6, "Pinned roots %p-%p", start_root, root->end_root);
859 sgen_conservatively_pin_objects_from (start_root, (void**)root->end_root, start_nursery, end_nursery, PIN_TYPE_OTHER);
860 } SGEN_HASH_TABLE_FOREACH_END;
861 /* now deal with the thread stacks
862 * in the future we should be able to conservatively scan only:
863 * *) the cpu registers
864 * *) the unmanaged stack frames
865 * *) the _last_ managed stack frame
866 * *) pointers slots in managed frames
868 sgen_client_scan_thread_data (start_nursery, end_nursery, FALSE, ctx);
872 single_arg_user_copy_or_mark (GCObject **obj, void *gc_data)
874 ScanCopyContext *ctx = (ScanCopyContext *)gc_data;
875 ctx->ops->copy_or_mark_object (obj, ctx->queue);
879 * The memory area from start_root to end_root contains pointers to objects.
880 * Their position is precisely described by @desc (this means that the pointer
881 * can be either NULL or the pointer to the start of an object).
882 * This functions copies them to to_space updates them.
884 * This function is not thread-safe!
887 precisely_scan_objects_from (void** start_root, void** end_root, char* n_start, char *n_end, SgenDescriptor desc, ScanCopyContext ctx)
889 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
890 SgenGrayQueue *queue = ctx.queue;
892 switch (desc & ROOT_DESC_TYPE_MASK) {
893 case ROOT_DESC_BITMAP:
894 desc >>= ROOT_DESC_TYPE_SHIFT;
896 if ((desc & 1) && *start_root) {
897 copy_func ((GCObject**)start_root, queue);
898 SGEN_LOG (9, "Overwrote root at %p with %p", start_root, *start_root);
904 case ROOT_DESC_COMPLEX: {
905 gsize *bitmap_data = (gsize *)sgen_get_complex_descriptor_bitmap (desc);
906 gsize bwords = (*bitmap_data) - 1;
907 void **start_run = start_root;
909 while (bwords-- > 0) {
910 gsize bmap = *bitmap_data++;
911 void **objptr = start_run;
913 if ((bmap & 1) && *objptr) {
914 copy_func ((GCObject**)objptr, queue);
915 SGEN_LOG (9, "Overwrote root at %p with %p", objptr, *objptr);
920 start_run += GC_BITS_PER_WORD;
924 case ROOT_DESC_USER: {
925 SgenUserRootMarkFunc marker = sgen_get_user_descriptor_func (desc);
926 marker (start_root, single_arg_user_copy_or_mark, &ctx);
929 case ROOT_DESC_RUN_LEN:
930 g_assert_not_reached ();
932 g_assert_not_reached ();
937 reset_heap_boundaries (void)
939 lowest_heap_address = ~(mword)0;
940 highest_heap_address = 0;
944 sgen_update_heap_boundaries (mword low, mword high)
949 old = lowest_heap_address;
952 } while (SGEN_CAS_PTR ((gpointer*)&lowest_heap_address, (gpointer)low, (gpointer)old) != (gpointer)old);
955 old = highest_heap_address;
958 } while (SGEN_CAS_PTR ((gpointer*)&highest_heap_address, (gpointer)high, (gpointer)old) != (gpointer)old);
962 * Allocate and setup the data structures needed to be able to allocate objects
963 * in the nursery. The nursery is stored in nursery_section.
968 GCMemSection *section;
975 SGEN_LOG (2, "Allocating nursery size: %zu", (size_t)sgen_nursery_size);
976 /* later we will alloc a larger area for the nursery but only activate
977 * what we need. The rest will be used as expansion if we have too many pinned
978 * objects in the existing nursery.
980 /* FIXME: handle OOM */
981 section = (GCMemSection *)sgen_alloc_internal (INTERNAL_MEM_SECTION);
983 alloc_size = sgen_nursery_size;
985 /* If there isn't enough space even for the nursery we should simply abort. */
986 g_assert (sgen_memgov_try_alloc_space (alloc_size, SPACE_NURSERY));
988 data = (char *)major_collector.alloc_heap (alloc_size, alloc_size, DEFAULT_NURSERY_BITS);
989 sgen_update_heap_boundaries ((mword)data, (mword)(data + sgen_nursery_size));
990 SGEN_LOG (4, "Expanding nursery size (%p-%p): %lu, total: %lu", data, data + alloc_size, (unsigned long)sgen_nursery_size, (unsigned long)sgen_gc_get_total_heap_allocation ());
991 section->data = section->next_data = data;
992 section->size = alloc_size;
993 section->end_data = data + sgen_nursery_size;
994 scan_starts = (alloc_size + SCAN_START_SIZE - 1) / SCAN_START_SIZE;
995 section->scan_starts = (char **)sgen_alloc_internal_dynamic (sizeof (char*) * scan_starts, INTERNAL_MEM_SCAN_STARTS, TRUE);
996 section->num_scan_start = scan_starts;
998 nursery_section = section;
1000 sgen_nursery_allocator_set_nursery_bounds (data, data + sgen_nursery_size);
1004 mono_gc_get_logfile (void)
1006 return gc_debug_file;
1010 scan_finalizer_entries (SgenPointerQueue *fin_queue, ScanCopyContext ctx)
1012 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
1013 SgenGrayQueue *queue = ctx.queue;
1016 for (i = 0; i < fin_queue->next_slot; ++i) {
1017 GCObject *obj = (GCObject *)fin_queue->data [i];
1020 SGEN_LOG (5, "Scan of fin ready object: %p (%s)\n", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
1021 copy_func ((GCObject**)&fin_queue->data [i], queue);
1026 generation_name (int generation)
1028 switch (generation) {
1029 case GENERATION_NURSERY: return "nursery";
1030 case GENERATION_OLD: return "old";
1031 default: g_assert_not_reached ();
1036 sgen_generation_name (int generation)
1038 return generation_name (generation);
1042 finish_gray_stack (int generation, ScanCopyContext ctx)
1046 int done_with_ephemerons, ephemeron_rounds = 0;
1047 char *start_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_start () : NULL;
1048 char *end_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_end () : (char*)-1;
1049 SgenGrayQueue *queue = ctx.queue;
1051 binary_protocol_finish_gray_stack_start (sgen_timestamp (), generation);
1053 * We copied all the reachable objects. Now it's the time to copy
1054 * the objects that were not referenced by the roots, but by the copied objects.
1055 * we built a stack of objects pointed to by gray_start: they are
1056 * additional roots and we may add more items as we go.
1057 * We loop until gray_start == gray_objects which means no more objects have
1058 * been added. Note this is iterative: no recursion is involved.
1059 * We need to walk the LO list as well in search of marked big objects
1060 * (use a flag since this is needed only on major collections). We need to loop
1061 * here as well, so keep a counter of marked LO (increasing it in copy_object).
1062 * To achieve better cache locality and cache usage, we drain the gray stack
1063 * frequently, after each object is copied, and just finish the work here.
1065 sgen_drain_gray_stack (ctx);
1067 SGEN_LOG (2, "%s generation done", generation_name (generation));
1070 Reset bridge data, we might have lingering data from a previous collection if this is a major
1071 collection trigged by minor overflow.
1073 We must reset the gathered bridges since their original block might be evacuated due to major
1074 fragmentation in the meanwhile and the bridge code should not have to deal with that.
1076 if (sgen_client_bridge_need_processing ())
1077 sgen_client_bridge_reset_data ();
1080 * Mark all strong toggleref objects. This must be done before we walk ephemerons or finalizers
1081 * to ensure they see the full set of live objects.
1083 sgen_client_mark_togglerefs (start_addr, end_addr, ctx);
1086 * Walk the ephemeron tables marking all values with reachable keys. This must be completely done
1087 * before processing finalizable objects and non-tracking weak links to avoid finalizing/clearing
1088 * objects that are in fact reachable.
1090 done_with_ephemerons = 0;
1092 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1093 sgen_drain_gray_stack (ctx);
1095 } while (!done_with_ephemerons);
1097 if (sgen_client_bridge_need_processing ()) {
1098 /*Make sure the gray stack is empty before we process bridge objects so we get liveness right*/
1099 sgen_drain_gray_stack (ctx);
1100 sgen_collect_bridge_objects (generation, ctx);
1101 if (generation == GENERATION_OLD)
1102 sgen_collect_bridge_objects (GENERATION_NURSERY, ctx);
1105 Do the first bridge step here, as the collector liveness state will become useless after that.
1107 An important optimization is to only proccess the possibly dead part of the object graph and skip
1108 over all live objects as we transitively know everything they point must be alive too.
1110 The above invariant is completely wrong if we let the gray queue be drained and mark/copy everything.
1112 This has the unfortunate side effect of making overflow collections perform the first step twice, but
1113 given we now have heuristics that perform major GC in anticipation of minor overflows this should not
1116 sgen_client_bridge_processing_stw_step ();
1120 Make sure we drain the gray stack before processing disappearing links and finalizers.
1121 If we don't make sure it is empty we might wrongly see a live object as dead.
1123 sgen_drain_gray_stack (ctx);
1126 We must clear weak links that don't track resurrection before processing object ready for
1127 finalization so they can be cleared before that.
1129 sgen_null_link_in_range (generation, ctx, FALSE);
1130 if (generation == GENERATION_OLD)
1131 sgen_null_link_in_range (GENERATION_NURSERY, ctx, FALSE);
1134 /* walk the finalization queue and move also the objects that need to be
1135 * finalized: use the finalized objects as new roots so the objects they depend
1136 * on are also not reclaimed. As with the roots above, only objects in the nursery
1137 * are marked/copied.
1139 sgen_finalize_in_range (generation, ctx);
1140 if (generation == GENERATION_OLD)
1141 sgen_finalize_in_range (GENERATION_NURSERY, ctx);
1142 /* drain the new stack that might have been created */
1143 SGEN_LOG (6, "Precise scan of gray area post fin");
1144 sgen_drain_gray_stack (ctx);
1147 * This must be done again after processing finalizable objects since CWL slots are cleared only after the key is finalized.
1149 done_with_ephemerons = 0;
1151 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1152 sgen_drain_gray_stack (ctx);
1154 } while (!done_with_ephemerons);
1156 sgen_client_clear_unreachable_ephemerons (ctx);
1159 * We clear togglerefs only after all possible chances of revival are done.
1160 * This is semantically more inline with what users expect and it allows for
1161 * user finalizers to correctly interact with TR objects.
1163 sgen_client_clear_togglerefs (start_addr, end_addr, ctx);
1166 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);
1169 * handle disappearing links
1170 * Note we do this after checking the finalization queue because if an object
1171 * survives (at least long enough to be finalized) we don't clear the link.
1172 * This also deals with a possible issue with the monitor reclamation: with the Boehm
1173 * GC a finalized object my lose the monitor because it is cleared before the finalizer is
1176 g_assert (sgen_gray_object_queue_is_empty (queue));
1178 sgen_null_link_in_range (generation, ctx, TRUE);
1179 if (generation == GENERATION_OLD)
1180 sgen_null_link_in_range (GENERATION_NURSERY, ctx, TRUE);
1181 if (sgen_gray_object_queue_is_empty (queue))
1183 sgen_drain_gray_stack (ctx);
1186 g_assert (sgen_gray_object_queue_is_empty (queue));
1188 sgen_gray_object_queue_trim_free_list (queue);
1189 binary_protocol_finish_gray_stack_end (sgen_timestamp (), generation);
1193 sgen_check_section_scan_starts (GCMemSection *section)
1196 for (i = 0; i < section->num_scan_start; ++i) {
1197 if (section->scan_starts [i]) {
1198 mword size = safe_object_get_size ((GCObject*) section->scan_starts [i]);
1199 SGEN_ASSERT (0, size >= SGEN_CLIENT_MINIMUM_OBJECT_SIZE && size <= MAX_SMALL_OBJ_SIZE, "Weird object size at scan starts.");
1205 check_scan_starts (void)
1207 if (!do_scan_starts_check)
1209 sgen_check_section_scan_starts (nursery_section);
1210 major_collector.check_scan_starts ();
1214 scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx)
1218 SGEN_HASH_TABLE_FOREACH (&roots_hash [root_type], void **, start_root, RootRecord *, root) {
1219 SGEN_LOG (6, "Precise root scan %p-%p (desc: %p)", start_root, root->end_root, (void*)root->root_desc);
1220 precisely_scan_objects_from (start_root, (void**)root->end_root, addr_start, addr_end, root->root_desc, ctx);
1221 } SGEN_HASH_TABLE_FOREACH_END;
1227 static gboolean inited = FALSE;
1232 mono_counters_register ("Collection max time", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME | MONO_COUNTER_MONOTONIC, &time_max);
1234 mono_counters_register ("Minor fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pre_collection_fragment_clear);
1235 mono_counters_register ("Minor pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pinning);
1236 mono_counters_register ("Minor scan remembered set", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_remsets);
1237 mono_counters_register ("Minor scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_pinned);
1238 mono_counters_register ("Minor scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_roots);
1239 mono_counters_register ("Minor fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_fragment_creation);
1241 mono_counters_register ("Major fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pre_collection_fragment_clear);
1242 mono_counters_register ("Major pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pinning);
1243 mono_counters_register ("Major scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_pinned);
1244 mono_counters_register ("Major scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_roots);
1245 mono_counters_register ("Major scan mod union", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_mod_union);
1246 mono_counters_register ("Major finish gray stack", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_finish_gray_stack);
1247 mono_counters_register ("Major free big objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_free_bigobjs);
1248 mono_counters_register ("Major LOS sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_los_sweep);
1249 mono_counters_register ("Major sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_sweep);
1250 mono_counters_register ("Major fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_fragment_creation);
1252 mono_counters_register ("Number of pinned objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_pinned_objects);
1254 #ifdef HEAVY_STATISTICS
1255 mono_counters_register ("WBarrier remember pointer", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_add_to_global_remset);
1256 mono_counters_register ("WBarrier arrayref copy", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_arrayref_copy);
1257 mono_counters_register ("WBarrier generic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store);
1258 mono_counters_register ("WBarrier generic atomic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store_atomic);
1259 mono_counters_register ("WBarrier set root", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_set_root);
1261 mono_counters_register ("# objects allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_alloced_degraded);
1262 mono_counters_register ("bytes allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_bytes_alloced_degraded);
1264 mono_counters_register ("# copy_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_nursery);
1265 mono_counters_register ("# objects copied (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_nursery);
1266 mono_counters_register ("# copy_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_major);
1267 mono_counters_register ("# objects copied (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_major);
1269 mono_counters_register ("# scan_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_nursery);
1270 mono_counters_register ("# scan_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_major);
1272 mono_counters_register ("Slots allocated in vain", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_slots_allocated_in_vain);
1274 mono_counters_register ("# nursery copy_object() failed from space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_from_space);
1275 mono_counters_register ("# nursery copy_object() failed forwarded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_forwarded);
1276 mono_counters_register ("# nursery copy_object() failed pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_pinned);
1277 mono_counters_register ("# nursery copy_object() failed to space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_to_space);
1279 sgen_nursery_allocator_init_heavy_stats ();
1287 reset_pinned_from_failed_allocation (void)
1289 bytes_pinned_from_failed_allocation = 0;
1293 sgen_set_pinned_from_failed_allocation (mword objsize)
1295 bytes_pinned_from_failed_allocation += objsize;
1299 sgen_collection_is_concurrent (void)
1301 switch (current_collection_generation) {
1302 case GENERATION_NURSERY:
1304 case GENERATION_OLD:
1305 return concurrent_collection_in_progress;
1307 g_error ("Invalid current generation %d", current_collection_generation);
1313 sgen_concurrent_collection_in_progress (void)
1315 return concurrent_collection_in_progress;
1319 SgenThreadPoolJob job;
1320 SgenObjectOperations *ops;
1324 job_remembered_set_scan (void *worker_data_untyped, SgenThreadPoolJob *job)
1326 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1327 ScanJob *job_data = (ScanJob*)job;
1328 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1329 remset.scan_remsets (ctx);
1333 SgenThreadPoolJob job;
1334 SgenObjectOperations *ops;
1338 } ScanFromRegisteredRootsJob;
1341 job_scan_from_registered_roots (void *worker_data_untyped, SgenThreadPoolJob *job)
1343 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1344 ScanFromRegisteredRootsJob *job_data = (ScanFromRegisteredRootsJob*)job;
1345 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1347 scan_from_registered_roots (job_data->heap_start, job_data->heap_end, job_data->root_type, ctx);
1351 SgenThreadPoolJob job;
1352 SgenObjectOperations *ops;
1355 } ScanThreadDataJob;
1358 job_scan_thread_data (void *worker_data_untyped, SgenThreadPoolJob *job)
1360 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1361 ScanThreadDataJob *job_data = (ScanThreadDataJob*)job;
1362 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1364 sgen_client_scan_thread_data (job_data->heap_start, job_data->heap_end, TRUE, ctx);
1368 SgenThreadPoolJob job;
1369 SgenObjectOperations *ops;
1370 SgenPointerQueue *queue;
1371 } ScanFinalizerEntriesJob;
1374 job_scan_finalizer_entries (void *worker_data_untyped, SgenThreadPoolJob *job)
1376 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1377 ScanFinalizerEntriesJob *job_data = (ScanFinalizerEntriesJob*)job;
1378 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1380 scan_finalizer_entries (job_data->queue, ctx);
1384 job_scan_major_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1386 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1387 ScanJob *job_data = (ScanJob*)job;
1388 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1390 g_assert (concurrent_collection_in_progress);
1391 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx);
1395 job_scan_los_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1397 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1398 ScanJob *job_data = (ScanJob*)job;
1399 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1401 g_assert (concurrent_collection_in_progress);
1402 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx);
1406 job_mod_union_preclean (void *worker_data_untyped, SgenThreadPoolJob *job)
1408 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1409 ScanJob *job_data = (ScanJob*)job;
1410 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1412 g_assert (concurrent_collection_in_progress);
1414 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx);
1415 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx);
1417 sgen_scan_pin_queue_objects (ctx);
1421 init_gray_queue (gboolean use_workers)
1424 sgen_workers_init_distribute_gray_queue ();
1425 sgen_gray_object_queue_init (&gray_queue, NULL);
1429 enqueue_scan_from_roots_jobs (char *heap_start, char *heap_end, SgenObjectOperations *ops, gboolean enqueue)
1431 ScanFromRegisteredRootsJob *scrrj;
1432 ScanThreadDataJob *stdj;
1433 ScanFinalizerEntriesJob *sfej;
1435 /* registered roots, this includes static fields */
1437 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots normal", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1439 scrrj->heap_start = heap_start;
1440 scrrj->heap_end = heap_end;
1441 scrrj->root_type = ROOT_TYPE_NORMAL;
1442 sgen_workers_enqueue_job (&scrrj->job, enqueue);
1444 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots wbarrier", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1446 scrrj->heap_start = heap_start;
1447 scrrj->heap_end = heap_end;
1448 scrrj->root_type = ROOT_TYPE_WBARRIER;
1449 sgen_workers_enqueue_job (&scrrj->job, enqueue);
1453 stdj = (ScanThreadDataJob*)sgen_thread_pool_job_alloc ("scan thread data", job_scan_thread_data, sizeof (ScanThreadDataJob));
1454 stdj->heap_start = heap_start;
1455 stdj->heap_end = heap_end;
1456 sgen_workers_enqueue_job (&stdj->job, enqueue);
1458 /* Scan the list of objects ready for finalization. */
1460 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1461 sfej->queue = &fin_ready_queue;
1463 sgen_workers_enqueue_job (&sfej->job, enqueue);
1465 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan critical finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1466 sfej->queue = &critical_fin_queue;
1468 sgen_workers_enqueue_job (&sfej->job, enqueue);
1472 * Perform a nursery collection.
1474 * Return whether any objects were late-pinned due to being out of memory.
1477 collect_nursery (SgenGrayQueue *unpin_queue, gboolean finish_up_concurrent_mark)
1479 gboolean needs_major;
1480 size_t max_garbage_amount;
1482 mword fragment_total;
1484 SgenObjectOperations *object_ops = &sgen_minor_collector.serial_ops;
1485 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, &gray_queue);
1489 if (disable_minor_collections)
1492 TV_GETTIME (last_minor_collection_start_tv);
1493 atv = last_minor_collection_start_tv;
1495 binary_protocol_collection_begin (gc_stats.minor_gc_count, GENERATION_NURSERY);
1497 if (do_verify_nursery || do_dump_nursery_content)
1498 sgen_debug_verify_nursery (do_dump_nursery_content);
1500 current_collection_generation = GENERATION_NURSERY;
1502 SGEN_ASSERT (0, !sgen_collection_is_concurrent (), "Why is the nursery collection concurrent?");
1504 reset_pinned_from_failed_allocation ();
1506 check_scan_starts ();
1508 sgen_nursery_alloc_prepare_for_minor ();
1512 nursery_next = sgen_nursery_alloc_get_upper_alloc_bound ();
1513 /* FIXME: optimize later to use the higher address where an object can be present */
1514 nursery_next = MAX (nursery_next, sgen_get_nursery_end ());
1516 SGEN_LOG (1, "Start nursery collection %d %p-%p, size: %d", gc_stats.minor_gc_count, sgen_get_nursery_start (), nursery_next, (int)(nursery_next - sgen_get_nursery_start ()));
1517 max_garbage_amount = nursery_next - sgen_get_nursery_start ();
1518 g_assert (nursery_section->size >= max_garbage_amount);
1520 /* world must be stopped already */
1522 time_minor_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1524 sgen_client_pre_collection_checks ();
1526 nursery_section->next_data = nursery_next;
1528 major_collector.start_nursery_collection ();
1530 sgen_memgov_minor_collection_start ();
1532 init_gray_queue (FALSE);
1534 gc_stats.minor_gc_count ++;
1536 if (whole_heap_check_before_collection) {
1537 sgen_clear_nursery_fragments ();
1538 sgen_check_whole_heap (finish_up_concurrent_mark);
1540 if (consistency_check_at_minor_collection)
1541 sgen_check_consistency ();
1543 sgen_process_fin_stage_entries ();
1545 /* pin from pinned handles */
1546 sgen_init_pinning ();
1547 sgen_client_binary_protocol_mark_start (GENERATION_NURSERY);
1548 pin_from_roots (sgen_get_nursery_start (), nursery_next, ctx);
1549 /* pin cemented objects */
1550 sgen_pin_cemented_objects ();
1551 /* identify pinned objects */
1552 sgen_optimize_pin_queue ();
1553 sgen_pinning_setup_section (nursery_section);
1555 pin_objects_in_nursery (FALSE, ctx);
1556 sgen_pinning_trim_queue_to_section (nursery_section);
1559 time_minor_pinning += TV_ELAPSED (btv, atv);
1560 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), (long long)TV_ELAPSED (btv, atv));
1561 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1563 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan remset", job_remembered_set_scan, sizeof (ScanJob));
1564 sj->ops = object_ops;
1565 sgen_workers_enqueue_job (&sj->job, FALSE);
1567 /* we don't have complete write barrier yet, so we scan all the old generation sections */
1569 time_minor_scan_remsets += TV_ELAPSED (atv, btv);
1570 SGEN_LOG (2, "Old generation scan: %lld usecs", (long long)TV_ELAPSED (atv, btv));
1572 sgen_pin_stats_print_class_stats ();
1574 /* FIXME: Why do we do this at this specific, seemingly random, point? */
1575 sgen_client_collecting_minor (&fin_ready_queue, &critical_fin_queue);
1578 time_minor_scan_pinned += TV_ELAPSED (btv, atv);
1580 enqueue_scan_from_roots_jobs (sgen_get_nursery_start (), nursery_next, object_ops, FALSE);
1583 time_minor_scan_roots += TV_ELAPSED (atv, btv);
1585 finish_gray_stack (GENERATION_NURSERY, ctx);
1588 time_minor_finish_gray_stack += TV_ELAPSED (btv, atv);
1589 sgen_client_binary_protocol_mark_end (GENERATION_NURSERY);
1591 if (objects_pinned) {
1592 sgen_optimize_pin_queue ();
1593 sgen_pinning_setup_section (nursery_section);
1596 /* walk the pin_queue, build up the fragment list of free memory, unmark
1597 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1600 sgen_client_binary_protocol_reclaim_start (GENERATION_NURSERY);
1601 fragment_total = sgen_build_nursery_fragments (nursery_section, unpin_queue);
1602 if (!fragment_total)
1605 /* Clear TLABs for all threads */
1606 sgen_clear_tlabs ();
1608 sgen_client_binary_protocol_reclaim_end (GENERATION_NURSERY);
1610 time_minor_fragment_creation += TV_ELAPSED (atv, btv);
1611 SGEN_LOG (2, "Fragment creation: %lld usecs, %lu bytes available", (long long)TV_ELAPSED (atv, btv), (unsigned long)fragment_total);
1613 if (consistency_check_at_minor_collection)
1614 sgen_check_major_refs ();
1616 major_collector.finish_nursery_collection ();
1618 TV_GETTIME (last_minor_collection_end_tv);
1619 gc_stats.minor_gc_time += TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv);
1621 sgen_debug_dump_heap ("minor", gc_stats.minor_gc_count - 1, NULL);
1623 /* prepare the pin queue for the next collection */
1624 sgen_finish_pinning ();
1625 if (sgen_have_pending_finalizers ()) {
1626 SGEN_LOG (4, "Finalizer-thread wakeup");
1627 sgen_client_finalize_notify ();
1629 sgen_pin_stats_reset ();
1630 /* clear cemented hash */
1631 sgen_cement_clear_below_threshold ();
1633 g_assert (sgen_gray_object_queue_is_empty (&gray_queue));
1635 remset.finish_minor_collection ();
1637 check_scan_starts ();
1639 binary_protocol_flush_buffers (FALSE);
1641 sgen_memgov_minor_collection_end ();
1643 /*objects are late pinned because of lack of memory, so a major is a good call*/
1644 needs_major = objects_pinned > 0;
1645 current_collection_generation = -1;
1648 binary_protocol_collection_end (gc_stats.minor_gc_count - 1, GENERATION_NURSERY, 0, 0);
1650 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1651 sgen_check_nursery_objects_pinned (unpin_queue != NULL);
1657 COPY_OR_MARK_FROM_ROOTS_SERIAL,
1658 COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT,
1659 COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT
1660 } CopyOrMarkFromRootsMode;
1663 major_copy_or_mark_from_roots (size_t *old_next_pin_slot, CopyOrMarkFromRootsMode mode, SgenObjectOperations *object_ops)
1668 /* FIXME: only use these values for the precise scan
1669 * note that to_space pointers should be excluded anyway...
1671 char *heap_start = NULL;
1672 char *heap_end = (char*)-1;
1673 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, WORKERS_DISTRIBUTE_GRAY_QUEUE);
1674 gboolean concurrent = mode != COPY_OR_MARK_FROM_ROOTS_SERIAL;
1676 SGEN_ASSERT (0, !!concurrent == !!concurrent_collection_in_progress, "We've been called with the wrong mode.");
1678 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1679 /*This cleans up unused fragments */
1680 sgen_nursery_allocator_prepare_for_pinning ();
1682 if (do_concurrent_checks)
1683 sgen_debug_check_nursery_is_clean ();
1685 /* The concurrent collector doesn't touch the nursery. */
1686 sgen_nursery_alloc_prepare_for_major ();
1689 init_gray_queue (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1693 /* Pinning depends on this */
1694 sgen_clear_nursery_fragments ();
1696 if (whole_heap_check_before_collection)
1697 sgen_check_whole_heap (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT);
1700 time_major_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1702 if (!sgen_collection_is_concurrent ())
1703 nursery_section->next_data = sgen_get_nursery_end ();
1704 /* we should also coalesce scanning from sections close to each other
1705 * and deal with pointers outside of the sections later.
1710 sgen_client_pre_collection_checks ();
1712 if (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1713 /* Remsets are not useful for a major collection */
1714 remset.clear_cards ();
1717 sgen_process_fin_stage_entries ();
1720 sgen_init_pinning ();
1721 SGEN_LOG (6, "Collecting pinned addresses");
1722 pin_from_roots ((void*)lowest_heap_address, (void*)highest_heap_address, ctx);
1723 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1724 /* Pin cemented objects that were forced */
1725 sgen_pin_cemented_objects ();
1727 sgen_optimize_pin_queue ();
1728 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1730 * Cemented objects that are in the pinned list will be marked. When
1731 * marking concurrently we won't mark mod-union cards for these objects.
1732 * Instead they will remain cemented until the next major collection,
1733 * when we will recheck if they are still pinned in the roots.
1735 sgen_cement_force_pinned ();
1738 sgen_client_collecting_major_1 ();
1741 * pin_queue now contains all candidate pointers, sorted and
1742 * uniqued. We must do two passes now to figure out which
1743 * objects are pinned.
1745 * The first is to find within the pin_queue the area for each
1746 * section. This requires that the pin_queue be sorted. We
1747 * also process the LOS objects and pinned chunks here.
1749 * The second, destructive, pass is to reduce the section
1750 * areas to pointers to the actually pinned objects.
1752 SGEN_LOG (6, "Pinning from sections");
1753 /* first pass for the sections */
1754 sgen_find_section_pin_queue_start_end (nursery_section);
1755 /* identify possible pointers to the insize of large objects */
1756 SGEN_LOG (6, "Pinning from large objects");
1757 for (bigobj = los_object_list; bigobj; bigobj = bigobj->next) {
1759 if (sgen_find_optimized_pin_queue_area ((char*)bigobj->data, (char*)bigobj->data + sgen_los_object_size (bigobj), &dummy, &dummy)) {
1760 binary_protocol_pin (bigobj->data, (gpointer)LOAD_VTABLE (bigobj->data), safe_object_get_size (bigobj->data));
1762 if (sgen_los_object_is_pinned (bigobj->data)) {
1763 SGEN_ASSERT (0, mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, "LOS objects can only be pinned here after concurrent marking.");
1766 sgen_los_pin_object (bigobj->data);
1767 if (SGEN_OBJECT_HAS_REFERENCES (bigobj->data))
1768 GRAY_OBJECT_ENQUEUE (WORKERS_DISTRIBUTE_GRAY_QUEUE, bigobj->data, sgen_obj_get_descriptor ((GCObject*)bigobj->data));
1769 sgen_pin_stats_register_object (bigobj->data, safe_object_get_size (bigobj->data));
1770 SGEN_LOG (6, "Marked large object %p (%s) size: %lu from roots", bigobj->data,
1771 sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (bigobj->data)),
1772 (unsigned long)sgen_los_object_size (bigobj));
1774 sgen_client_pinned_los_object (bigobj->data);
1778 pin_objects_in_nursery (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT, ctx);
1779 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1780 sgen_check_nursery_objects_pinned (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1782 major_collector.pin_objects (WORKERS_DISTRIBUTE_GRAY_QUEUE);
1783 if (old_next_pin_slot)
1784 *old_next_pin_slot = sgen_get_pinned_count ();
1787 time_major_pinning += TV_ELAPSED (atv, btv);
1788 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), (long long)TV_ELAPSED (atv, btv));
1789 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1791 major_collector.init_to_space ();
1793 SGEN_ASSERT (0, sgen_workers_all_done (), "Why are the workers not done when we start or finish a major collection?");
1794 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1795 if (sgen_workers_have_idle_work ()) {
1797 * We force the finish of the worker with the new object ops context
1798 * which can also do copying. We need to have finished pinning.
1800 sgen_workers_start_all_workers (object_ops, NULL);
1801 sgen_workers_join ();
1805 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1806 main_gc_thread = mono_native_thread_self ();
1809 sgen_client_collecting_major_2 ();
1812 time_major_scan_pinned += TV_ELAPSED (btv, atv);
1814 sgen_client_collecting_major_3 (&fin_ready_queue, &critical_fin_queue);
1816 enqueue_scan_from_roots_jobs (heap_start, heap_end, object_ops, FALSE);
1819 time_major_scan_roots += TV_ELAPSED (atv, btv);
1822 * We start the concurrent worker after pinning and after we scanned the roots
1823 * in order to make sure that the worker does not finish before handling all
1826 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1827 if (precleaning_enabled) {
1829 /* Mod union preclean job */
1830 sj = (ScanJob*)sgen_thread_pool_job_alloc ("preclean mod union cardtable", job_mod_union_preclean, sizeof (ScanJob));
1831 sj->ops = object_ops;
1832 sgen_workers_start_all_workers (object_ops, &sj->job);
1834 sgen_workers_start_all_workers (object_ops, NULL);
1836 gray_queue_enable_redirect (WORKERS_DISTRIBUTE_GRAY_QUEUE);
1839 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1842 /* Mod union card table */
1843 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan mod union cardtable", job_scan_major_mod_union_card_table, sizeof (ScanJob));
1844 sj->ops = object_ops;
1845 sgen_workers_enqueue_job (&sj->job, FALSE);
1847 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan LOS mod union cardtable", job_scan_los_mod_union_card_table, sizeof (ScanJob));
1848 sj->ops = object_ops;
1849 sgen_workers_enqueue_job (&sj->job, FALSE);
1852 time_major_scan_mod_union += TV_ELAPSED (btv, atv);
1855 sgen_pin_stats_print_class_stats ();
1859 major_finish_copy_or_mark (CopyOrMarkFromRootsMode mode)
1861 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1862 sgen_finish_pinning ();
1864 sgen_pin_stats_reset ();
1866 if (do_concurrent_checks)
1867 sgen_debug_check_nursery_is_clean ();
1872 major_start_collection (gboolean concurrent, size_t *old_next_pin_slot)
1874 SgenObjectOperations *object_ops;
1876 binary_protocol_collection_begin (gc_stats.major_gc_count, GENERATION_OLD);
1878 current_collection_generation = GENERATION_OLD;
1880 g_assert (sgen_section_gray_queue_is_empty (sgen_workers_get_distribute_section_gray_queue ()));
1883 sgen_cement_reset ();
1886 g_assert (major_collector.is_concurrent);
1887 concurrent_collection_in_progress = TRUE;
1889 object_ops = &major_collector.major_ops_concurrent_start;
1891 object_ops = &major_collector.major_ops_serial;
1894 reset_pinned_from_failed_allocation ();
1896 sgen_memgov_major_collection_start ();
1898 //count_ref_nonref_objs ();
1899 //consistency_check ();
1901 check_scan_starts ();
1904 SGEN_LOG (1, "Start major collection %d", gc_stats.major_gc_count);
1905 gc_stats.major_gc_count ++;
1907 if (major_collector.start_major_collection)
1908 major_collector.start_major_collection ();
1910 major_copy_or_mark_from_roots (old_next_pin_slot, concurrent ? COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT : COPY_OR_MARK_FROM_ROOTS_SERIAL, object_ops);
1911 major_finish_copy_or_mark (concurrent ? COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT : COPY_OR_MARK_FROM_ROOTS_SERIAL);
1915 major_finish_collection (const char *reason, size_t old_next_pin_slot, gboolean forced)
1917 ScannedObjectCounts counts;
1918 SgenObjectOperations *object_ops;
1919 mword fragment_total;
1925 if (concurrent_collection_in_progress) {
1926 object_ops = &major_collector.major_ops_concurrent_finish;
1928 major_copy_or_mark_from_roots (NULL, COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, object_ops);
1930 major_finish_copy_or_mark (COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT);
1932 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1933 main_gc_thread = NULL;
1936 object_ops = &major_collector.major_ops_serial;
1939 g_assert (sgen_section_gray_queue_is_empty (sgen_workers_get_distribute_section_gray_queue ()));
1941 /* all the objects in the heap */
1942 finish_gray_stack (GENERATION_OLD, CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, &gray_queue));
1944 time_major_finish_gray_stack += TV_ELAPSED (btv, atv);
1946 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after joining");
1948 if (objects_pinned) {
1949 g_assert (!concurrent_collection_in_progress);
1952 * This is slow, but we just OOM'd.
1954 * See comment at `sgen_pin_queue_clear_discarded_entries` for how the pin
1955 * queue is laid out at this point.
1957 sgen_pin_queue_clear_discarded_entries (nursery_section, old_next_pin_slot);
1959 * We need to reestablish all pinned nursery objects in the pin queue
1960 * because they're needed for fragment creation. Unpinning happens by
1961 * walking the whole queue, so it's not necessary to reestablish where major
1962 * heap block pins are - all we care is that they're still in there
1965 sgen_optimize_pin_queue ();
1966 sgen_find_section_pin_queue_start_end (nursery_section);
1970 reset_heap_boundaries ();
1971 sgen_update_heap_boundaries ((mword)sgen_get_nursery_start (), (mword)sgen_get_nursery_end ());
1973 /* walk the pin_queue, build up the fragment list of free memory, unmark
1974 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1977 fragment_total = sgen_build_nursery_fragments (nursery_section, NULL);
1978 if (!fragment_total)
1980 SGEN_LOG (4, "Free space in nursery after major %ld", (long)fragment_total);
1982 if (do_concurrent_checks && concurrent_collection_in_progress)
1983 sgen_debug_check_nursery_is_clean ();
1985 /* prepare the pin queue for the next collection */
1986 sgen_finish_pinning ();
1988 /* Clear TLABs for all threads */
1989 sgen_clear_tlabs ();
1991 sgen_pin_stats_reset ();
1993 sgen_cement_clear_below_threshold ();
1995 if (check_mark_bits_after_major_collection)
1996 sgen_check_heap_marked (concurrent_collection_in_progress);
1999 time_major_fragment_creation += TV_ELAPSED (atv, btv);
2001 binary_protocol_sweep_begin (GENERATION_OLD, !major_collector.sweeps_lazily);
2002 sgen_memgov_major_pre_sweep ();
2005 time_major_free_bigobjs += TV_ELAPSED (btv, atv);
2010 time_major_los_sweep += TV_ELAPSED (atv, btv);
2012 major_collector.sweep ();
2014 binary_protocol_sweep_end (GENERATION_OLD, !major_collector.sweeps_lazily);
2017 time_major_sweep += TV_ELAPSED (btv, atv);
2019 sgen_debug_dump_heap ("major", gc_stats.major_gc_count - 1, reason);
2021 if (sgen_have_pending_finalizers ()) {
2022 SGEN_LOG (4, "Finalizer-thread wakeup");
2023 sgen_client_finalize_notify ();
2026 g_assert (sgen_gray_object_queue_is_empty (&gray_queue));
2028 sgen_memgov_major_collection_end (forced);
2029 current_collection_generation = -1;
2031 memset (&counts, 0, sizeof (ScannedObjectCounts));
2032 major_collector.finish_major_collection (&counts);
2034 g_assert (sgen_section_gray_queue_is_empty (sgen_workers_get_distribute_section_gray_queue ()));
2036 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after major collection has finished");
2037 if (concurrent_collection_in_progress)
2038 concurrent_collection_in_progress = FALSE;
2040 check_scan_starts ();
2042 binary_protocol_flush_buffers (FALSE);
2044 //consistency_check ();
2046 binary_protocol_collection_end (gc_stats.major_gc_count - 1, GENERATION_OLD, counts.num_scanned_objects, counts.num_unique_scanned_objects);
2050 major_do_collection (const char *reason, gboolean forced)
2052 TV_DECLARE (time_start);
2053 TV_DECLARE (time_end);
2054 size_t old_next_pin_slot;
2056 if (disable_major_collections)
2059 if (major_collector.get_and_reset_num_major_objects_marked) {
2060 long long num_marked = major_collector.get_and_reset_num_major_objects_marked ();
2061 g_assert (!num_marked);
2064 /* world must be stopped already */
2065 TV_GETTIME (time_start);
2067 major_start_collection (FALSE, &old_next_pin_slot);
2068 major_finish_collection (reason, old_next_pin_slot, forced);
2070 TV_GETTIME (time_end);
2071 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2073 /* FIXME: also report this to the user, preferably in gc-end. */
2074 if (major_collector.get_and_reset_num_major_objects_marked)
2075 major_collector.get_and_reset_num_major_objects_marked ();
2077 return bytes_pinned_from_failed_allocation > 0;
2081 major_start_concurrent_collection (const char *reason)
2083 TV_DECLARE (time_start);
2084 TV_DECLARE (time_end);
2085 long long num_objects_marked;
2087 if (disable_major_collections)
2090 TV_GETTIME (time_start);
2091 SGEN_TV_GETTIME (time_major_conc_collection_start);
2093 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2094 g_assert (num_objects_marked == 0);
2096 binary_protocol_concurrent_start ();
2098 // FIXME: store reason and pass it when finishing
2099 major_start_collection (TRUE, NULL);
2101 gray_queue_redirect (&gray_queue);
2103 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2105 TV_GETTIME (time_end);
2106 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2108 current_collection_generation = -1;
2112 * Returns whether the major collection has finished.
2115 major_should_finish_concurrent_collection (void)
2117 SGEN_ASSERT (0, sgen_gray_object_queue_is_empty (&gray_queue), "Why is the gray queue not empty before we have started doing anything?");
2118 return sgen_workers_all_done ();
2122 major_update_concurrent_collection (void)
2124 TV_DECLARE (total_start);
2125 TV_DECLARE (total_end);
2127 TV_GETTIME (total_start);
2129 binary_protocol_concurrent_update ();
2131 major_collector.update_cardtable_mod_union ();
2132 sgen_los_update_cardtable_mod_union ();
2134 TV_GETTIME (total_end);
2135 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end);
2139 major_finish_concurrent_collection (gboolean forced)
2141 TV_DECLARE (total_start);
2142 TV_DECLARE (total_end);
2144 TV_GETTIME (total_start);
2146 binary_protocol_concurrent_finish ();
2149 * We need to stop all workers since we're updating the cardtable below.
2150 * The workers will be resumed with a finishing pause context to avoid
2151 * additional cardtable and object scanning.
2153 sgen_workers_stop_all_workers ();
2155 SGEN_TV_GETTIME (time_major_conc_collection_end);
2156 gc_stats.major_gc_time_concurrent += SGEN_TV_ELAPSED (time_major_conc_collection_start, time_major_conc_collection_end);
2158 major_collector.update_cardtable_mod_union ();
2159 sgen_los_update_cardtable_mod_union ();
2161 if (mod_union_consistency_check)
2162 sgen_check_mod_union_consistency ();
2164 current_collection_generation = GENERATION_OLD;
2165 sgen_cement_reset ();
2166 major_finish_collection ("finishing", -1, forced);
2168 if (whole_heap_check_before_collection)
2169 sgen_check_whole_heap (FALSE);
2171 TV_GETTIME (total_end);
2172 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end) - TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv);
2174 current_collection_generation = -1;
2178 * Ensure an allocation request for @size will succeed by freeing enough memory.
2180 * LOCKING: The GC lock MUST be held.
2183 sgen_ensure_free_space (size_t size, int generation)
2185 int generation_to_collect = -1;
2186 const char *reason = NULL;
2188 if (generation == GENERATION_OLD) {
2189 if (sgen_need_major_collection (size)) {
2190 reason = "LOS overflow";
2191 generation_to_collect = GENERATION_OLD;
2194 if (degraded_mode) {
2195 if (sgen_need_major_collection (size)) {
2196 reason = "Degraded mode overflow";
2197 generation_to_collect = GENERATION_OLD;
2199 } else if (sgen_need_major_collection (size)) {
2200 reason = concurrent_collection_in_progress ? "Forced finish concurrent collection" : "Minor allowance";
2201 generation_to_collect = GENERATION_OLD;
2203 generation_to_collect = GENERATION_NURSERY;
2204 reason = "Nursery full";
2208 if (generation_to_collect == -1) {
2209 if (concurrent_collection_in_progress && sgen_workers_all_done ()) {
2210 generation_to_collect = GENERATION_OLD;
2211 reason = "Finish concurrent collection";
2215 if (generation_to_collect == -1)
2217 sgen_perform_collection (size, generation_to_collect, reason, FALSE);
2221 * LOCKING: Assumes the GC lock is held.
2224 sgen_perform_collection (size_t requested_size, int generation_to_collect, const char *reason, gboolean wait_to_finish)
2226 TV_DECLARE (gc_start);
2227 TV_DECLARE (gc_end);
2228 TV_DECLARE (gc_total_start);
2229 TV_DECLARE (gc_total_end);
2230 GGTimingInfo infos [2];
2231 int overflow_generation_to_collect = -1;
2232 int oldest_generation_collected = generation_to_collect;
2233 const char *overflow_reason = NULL;
2234 gboolean finish_concurrent = concurrent_collection_in_progress && (major_should_finish_concurrent_collection () || generation_to_collect == GENERATION_OLD);
2236 binary_protocol_collection_requested (generation_to_collect, requested_size, wait_to_finish ? 1 : 0);
2238 SGEN_ASSERT (0, generation_to_collect == GENERATION_NURSERY || generation_to_collect == GENERATION_OLD, "What generation is this?");
2240 TV_GETTIME (gc_start);
2242 sgen_stop_world (generation_to_collect);
2244 TV_GETTIME (gc_total_start);
2246 // FIXME: extract overflow reason
2247 // FIXME: minor overflow for concurrent case
2248 if (generation_to_collect == GENERATION_NURSERY && !finish_concurrent) {
2249 if (concurrent_collection_in_progress)
2250 major_update_concurrent_collection ();
2252 if (collect_nursery (NULL, FALSE) && !concurrent_collection_in_progress) {
2253 overflow_generation_to_collect = GENERATION_OLD;
2254 overflow_reason = "Minor overflow";
2256 } else if (finish_concurrent) {
2257 major_finish_concurrent_collection (wait_to_finish);
2258 oldest_generation_collected = GENERATION_OLD;
2260 SGEN_ASSERT (0, generation_to_collect == GENERATION_OLD, "We should have handled nursery collections above");
2261 if (major_collector.is_concurrent && !wait_to_finish) {
2262 collect_nursery (NULL, FALSE);
2263 major_start_concurrent_collection (reason);
2264 oldest_generation_collected = GENERATION_NURSERY;
2265 } else if (major_do_collection (reason, wait_to_finish)) {
2266 overflow_generation_to_collect = GENERATION_NURSERY;
2267 overflow_reason = "Excessive pinning";
2271 TV_GETTIME (gc_end);
2273 memset (infos, 0, sizeof (infos));
2274 infos [0].generation = oldest_generation_collected;
2275 infos [0].reason = reason;
2276 infos [0].is_overflow = FALSE;
2277 infos [1].generation = -1;
2278 infos [0].total_time = SGEN_TV_ELAPSED (gc_start, gc_end);
2280 if (overflow_generation_to_collect != -1) {
2281 SGEN_ASSERT (0, !concurrent_collection_in_progress, "We don't yet support overflow collections with the concurrent collector");
2284 * We need to do an overflow collection, either because we ran out of memory
2285 * or the nursery is fully pinned.
2288 infos [1].generation = overflow_generation_to_collect;
2289 infos [1].reason = overflow_reason;
2290 infos [1].is_overflow = TRUE;
2293 if (overflow_generation_to_collect == GENERATION_NURSERY)
2294 collect_nursery (NULL, FALSE);
2296 major_do_collection (overflow_reason, wait_to_finish);
2298 TV_GETTIME (gc_end);
2299 infos [1].total_time = SGEN_TV_ELAPSED (gc_start, gc_end);
2301 oldest_generation_collected = MAX (oldest_generation_collected, overflow_generation_to_collect);
2304 SGEN_LOG (2, "Heap size: %lu, LOS size: %lu", (unsigned long)sgen_gc_get_total_heap_allocation (), (unsigned long)los_memory_usage);
2306 /* this also sets the proper pointers for the next allocation */
2307 if (generation_to_collect == GENERATION_NURSERY && !sgen_can_alloc_size (requested_size)) {
2308 /* TypeBuilder and MonoMethod are killing mcs with fragmentation */
2309 SGEN_LOG (1, "nursery collection didn't find enough room for %zd alloc (%zd pinned)", requested_size, sgen_get_pinned_count ());
2310 sgen_dump_pin_queue ();
2314 g_assert (sgen_gray_object_queue_is_empty (&gray_queue));
2316 TV_GETTIME (gc_total_end);
2317 time_max = MAX (time_max, TV_ELAPSED (gc_total_start, gc_total_end));
2319 sgen_restart_world (oldest_generation_collected, infos);
2323 * ######################################################################
2324 * ######## Memory allocation from the OS
2325 * ######################################################################
2326 * This section of code deals with getting memory from the OS and
2327 * allocating memory for GC-internal data structures.
2328 * Internal memory can be handled with a freelist for small objects.
2334 G_GNUC_UNUSED static void
2335 report_internal_mem_usage (void)
2337 printf ("Internal memory usage:\n");
2338 sgen_report_internal_mem_usage ();
2339 printf ("Pinned memory usage:\n");
2340 major_collector.report_pinned_memory_usage ();
2344 * ######################################################################
2345 * ######## Finalization support
2346 * ######################################################################
2350 * If the object has been forwarded it means it's still referenced from a root.
2351 * If it is pinned it's still alive as well.
2352 * A LOS object is only alive if we have pinned it.
2353 * Return TRUE if @obj is ready to be finalized.
2355 static inline gboolean
2356 sgen_is_object_alive (GCObject *object)
2358 if (ptr_in_nursery (object))
2359 return sgen_nursery_is_object_alive (object);
2361 return sgen_major_is_object_alive (object);
2365 * This function returns true if @object is either alive and belongs to the
2366 * current collection - major collections are full heap, so old gen objects
2367 * are never alive during a minor collection.
2370 sgen_is_object_alive_and_on_current_collection (GCObject *object)
2372 if (ptr_in_nursery (object))
2373 return sgen_nursery_is_object_alive (object);
2375 if (current_collection_generation == GENERATION_NURSERY)
2378 return sgen_major_is_object_alive (object);
2383 sgen_gc_is_object_ready_for_finalization (GCObject *object)
2385 return !sgen_is_object_alive (object);
2389 sgen_queue_finalization_entry (GCObject *obj)
2391 gboolean critical = sgen_client_object_has_critical_finalizer (obj);
2393 sgen_pointer_queue_add (critical ? &critical_fin_queue : &fin_ready_queue, obj);
2395 sgen_client_object_queued_for_finalization (obj);
2399 sgen_object_is_live (GCObject *obj)
2401 return sgen_is_object_alive_and_on_current_collection (obj);
2405 * `System.GC.WaitForPendingFinalizers` first checks `sgen_have_pending_finalizers()` to
2406 * determine whether it can exit quickly. The latter must therefore only return FALSE if
2407 * all finalizers have really finished running.
2409 * `sgen_gc_invoke_finalizers()` first dequeues a finalizable object, and then finalizes it.
2410 * This means that just checking whether the queues are empty leaves the possibility that an
2411 * object might have been dequeued but not yet finalized. That's why we need the additional
2412 * flag `pending_unqueued_finalizer`.
2415 static volatile gboolean pending_unqueued_finalizer = FALSE;
2418 sgen_gc_invoke_finalizers (void)
2422 g_assert (!pending_unqueued_finalizer);
2424 /* FIXME: batch to reduce lock contention */
2425 while (sgen_have_pending_finalizers ()) {
2431 * We need to set `pending_unqueued_finalizer` before dequeing the
2432 * finalizable object.
2434 if (!sgen_pointer_queue_is_empty (&fin_ready_queue)) {
2435 pending_unqueued_finalizer = TRUE;
2436 mono_memory_write_barrier ();
2437 obj = (GCObject *)sgen_pointer_queue_pop (&fin_ready_queue);
2438 } else if (!sgen_pointer_queue_is_empty (&critical_fin_queue)) {
2439 pending_unqueued_finalizer = TRUE;
2440 mono_memory_write_barrier ();
2441 obj = (GCObject *)sgen_pointer_queue_pop (&critical_fin_queue);
2447 SGEN_LOG (7, "Finalizing object %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
2455 /* the object is on the stack so it is pinned */
2456 /*g_print ("Calling finalizer for object: %p (%s)\n", obj, sgen_client_object_safe_name (obj));*/
2457 sgen_client_run_finalize (obj);
2460 if (pending_unqueued_finalizer) {
2461 mono_memory_write_barrier ();
2462 pending_unqueued_finalizer = FALSE;
2469 sgen_have_pending_finalizers (void)
2471 return pending_unqueued_finalizer || !sgen_pointer_queue_is_empty (&fin_ready_queue) || !sgen_pointer_queue_is_empty (&critical_fin_queue);
2475 * ######################################################################
2476 * ######## registered roots support
2477 * ######################################################################
2481 * We do not coalesce roots.
2484 sgen_register_root (char *start, size_t size, SgenDescriptor descr, int root_type, int source, const char *msg)
2486 RootRecord new_root;
2489 for (i = 0; i < ROOT_TYPE_NUM; ++i) {
2490 RootRecord *root = (RootRecord *)sgen_hash_table_lookup (&roots_hash [i], start);
2491 /* we allow changing the size and the descriptor (for thread statics etc) */
2493 size_t old_size = root->end_root - start;
2494 root->end_root = start + size;
2495 SGEN_ASSERT (0, !!root->root_desc == !!descr, "Can't change whether a root is precise or conservative.");
2496 SGEN_ASSERT (0, root->source == source, "Can't change a root's source identifier.");
2497 SGEN_ASSERT (0, !!root->msg == !!msg, "Can't change a root's message.");
2498 root->root_desc = descr;
2500 roots_size -= old_size;
2506 new_root.end_root = start + size;
2507 new_root.root_desc = descr;
2508 new_root.source = source;
2511 sgen_hash_table_replace (&roots_hash [root_type], start, &new_root, NULL);
2514 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);
2521 sgen_deregister_root (char* addr)
2527 for (root_type = 0; root_type < ROOT_TYPE_NUM; ++root_type) {
2528 if (sgen_hash_table_remove (&roots_hash [root_type], addr, &root))
2529 roots_size -= (root.end_root - addr);
2535 * ######################################################################
2536 * ######## Thread handling (stop/start code)
2537 * ######################################################################
2541 sgen_get_current_collection_generation (void)
2543 return current_collection_generation;
2547 sgen_thread_register (SgenThreadInfo* info, void *stack_bottom_fallback)
2549 #ifndef HAVE_KW_THREAD
2550 info->tlab_start = info->tlab_next = info->tlab_temp_end = info->tlab_real_end = NULL;
2553 sgen_init_tlab_info (info);
2555 sgen_client_thread_register (info, stack_bottom_fallback);
2561 sgen_thread_unregister (SgenThreadInfo *p)
2563 sgen_client_thread_unregister (p);
2567 * ######################################################################
2568 * ######## Write barriers
2569 * ######################################################################
2573 * Note: the write barriers first do the needed GC work and then do the actual store:
2574 * this way the value is visible to the conservative GC scan after the write barrier
2575 * itself. If a GC interrupts the barrier in the middle, value will be kept alive by
2576 * the conservative scan, otherwise by the remembered set scan.
2580 mono_gc_wbarrier_arrayref_copy (gpointer dest_ptr, gpointer src_ptr, int count)
2582 HEAVY_STAT (++stat_wbarrier_arrayref_copy);
2583 /*This check can be done without taking a lock since dest_ptr array is pinned*/
2584 if (ptr_in_nursery (dest_ptr) || count <= 0) {
2585 mono_gc_memmove_aligned (dest_ptr, src_ptr, count * sizeof (gpointer));
2589 #ifdef SGEN_HEAVY_BINARY_PROTOCOL
2590 if (binary_protocol_is_heavy_enabled ()) {
2592 for (i = 0; i < count; ++i) {
2593 gpointer dest = (gpointer*)dest_ptr + i;
2594 gpointer obj = *((gpointer*)src_ptr + i);
2596 binary_protocol_wbarrier (dest, obj, (gpointer)LOAD_VTABLE (obj));
2601 remset.wbarrier_arrayref_copy (dest_ptr, src_ptr, count);
2605 mono_gc_wbarrier_generic_nostore (gpointer ptr)
2609 HEAVY_STAT (++stat_wbarrier_generic_store);
2611 sgen_client_wbarrier_generic_nostore_check (ptr);
2613 obj = *(gpointer*)ptr;
2615 binary_protocol_wbarrier (ptr, obj, (gpointer)LOAD_VTABLE (obj));
2618 * We need to record old->old pointer locations for the
2619 * concurrent collector.
2621 if (!ptr_in_nursery (obj) && !concurrent_collection_in_progress) {
2622 SGEN_LOG (8, "Skipping remset at %p", ptr);
2626 SGEN_LOG (8, "Adding remset at %p", ptr);
2628 remset.wbarrier_generic_nostore (ptr);
2632 mono_gc_wbarrier_generic_store (gpointer ptr, GCObject* value)
2634 SGEN_LOG (8, "Wbarrier store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2635 SGEN_UPDATE_REFERENCE_ALLOW_NULL (ptr, value);
2636 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2637 mono_gc_wbarrier_generic_nostore (ptr);
2638 sgen_dummy_use (value);
2641 /* Same as mono_gc_wbarrier_generic_store () but performs the store
2642 * as an atomic operation with release semantics.
2645 mono_gc_wbarrier_generic_store_atomic (gpointer ptr, GCObject *value)
2647 HEAVY_STAT (++stat_wbarrier_generic_store_atomic);
2649 SGEN_LOG (8, "Wbarrier atomic store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2651 InterlockedWritePointer ((volatile gpointer *)ptr, value);
2653 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2654 mono_gc_wbarrier_generic_nostore (ptr);
2656 sgen_dummy_use (value);
2660 sgen_wbarrier_value_copy_bitmap (gpointer _dest, gpointer _src, int size, unsigned bitmap)
2662 GCObject **dest = (GCObject **)_dest;
2663 GCObject **src = (GCObject **)_src;
2667 mono_gc_wbarrier_generic_store (dest, *src);
2672 size -= SIZEOF_VOID_P;
2678 * ######################################################################
2679 * ######## Other mono public interface functions.
2680 * ######################################################################
2684 sgen_gc_collect (int generation)
2689 sgen_perform_collection (0, generation, "user request", TRUE);
2694 sgen_gc_collection_count (int generation)
2696 if (generation == 0)
2697 return gc_stats.minor_gc_count;
2698 return gc_stats.major_gc_count;
2702 sgen_gc_get_used_size (void)
2706 tot = los_memory_usage;
2707 tot += nursery_section->next_data - nursery_section->data;
2708 tot += major_collector.get_used_size ();
2709 /* FIXME: account for pinned objects */
2715 sgen_env_var_error (const char *env_var, const char *fallback, const char *description_format, ...)
2719 va_start (ap, description_format);
2721 fprintf (stderr, "Warning: In environment variable `%s': ", env_var);
2722 vfprintf (stderr, description_format, ap);
2724 fprintf (stderr, " - %s", fallback);
2725 fprintf (stderr, "\n");
2731 parse_double_in_interval (const char *env_var, const char *opt_name, const char *opt, double min, double max, double *result)
2734 double val = strtod (opt, &endptr);
2735 if (endptr == opt) {
2736 sgen_env_var_error (env_var, "Using default value.", "`%s` must be a number.", opt_name);
2739 else if (val < min || val > max) {
2740 sgen_env_var_error (env_var, "Using default value.", "`%s` must be between %.2f - %.2f.", opt_name, min, max);
2752 char *major_collector_opt = NULL;
2753 char *minor_collector_opt = NULL;
2754 size_t max_heap = 0;
2755 size_t soft_limit = 0;
2757 gboolean debug_print_allowance = FALSE;
2758 double allowance_ratio = 0, save_target = 0;
2759 gboolean cement_enabled = TRUE;
2762 result = InterlockedCompareExchange (&gc_initialized, -1, 0);
2765 /* already inited */
2768 /* being inited by another thread */
2769 mono_thread_info_usleep (1000);
2772 /* we will init it */
2775 g_assert_not_reached ();
2777 } while (result != 0);
2779 SGEN_TV_GETTIME (sgen_init_timestamp);
2781 #ifdef SGEN_WITHOUT_MONO
2782 mono_thread_smr_init ();
2785 mono_coop_mutex_init (&gc_mutex);
2787 gc_debug_file = stderr;
2789 mono_coop_mutex_init (&sgen_interruption_mutex);
2791 if ((env = g_getenv (MONO_GC_PARAMS_NAME))) {
2792 opts = g_strsplit (env, ",", -1);
2793 for (ptr = opts; *ptr; ++ptr) {
2795 if (g_str_has_prefix (opt, "major=")) {
2796 opt = strchr (opt, '=') + 1;
2797 major_collector_opt = g_strdup (opt);
2798 } else if (g_str_has_prefix (opt, "minor=")) {
2799 opt = strchr (opt, '=') + 1;
2800 minor_collector_opt = g_strdup (opt);
2808 sgen_init_internal_allocator ();
2809 sgen_init_nursery_allocator ();
2810 sgen_init_fin_weak_hash ();
2811 sgen_init_hash_table ();
2812 sgen_init_descriptors ();
2813 sgen_init_gray_queues ();
2814 sgen_init_allocator ();
2815 sgen_init_gchandles ();
2817 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_SECTION, SGEN_SIZEOF_GC_MEM_SECTION);
2818 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_GRAY_QUEUE, sizeof (GrayQueueSection));
2820 sgen_client_init ();
2822 if (!minor_collector_opt) {
2823 sgen_simple_nursery_init (&sgen_minor_collector);
2825 if (!strcmp (minor_collector_opt, "simple")) {
2827 sgen_simple_nursery_init (&sgen_minor_collector);
2828 } else if (!strcmp (minor_collector_opt, "split")) {
2829 sgen_split_nursery_init (&sgen_minor_collector);
2831 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `simple` instead.", "Unknown minor collector `%s'.", minor_collector_opt);
2832 goto use_simple_nursery;
2836 if (!major_collector_opt || !strcmp (major_collector_opt, "marksweep")) {
2837 use_marksweep_major:
2838 sgen_marksweep_init (&major_collector);
2839 } else if (!major_collector_opt || !strcmp (major_collector_opt, "marksweep-conc")) {
2840 sgen_marksweep_conc_init (&major_collector);
2842 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `marksweep` instead.", "Unknown major collector `%s'.", major_collector_opt);
2843 goto use_marksweep_major;
2846 sgen_nursery_size = DEFAULT_NURSERY_SIZE;
2849 gboolean usage_printed = FALSE;
2851 for (ptr = opts; *ptr; ++ptr) {
2853 if (!strcmp (opt, ""))
2855 if (g_str_has_prefix (opt, "major="))
2857 if (g_str_has_prefix (opt, "minor="))
2859 if (g_str_has_prefix (opt, "max-heap-size=")) {
2860 size_t page_size = mono_pagesize ();
2861 size_t max_heap_candidate = 0;
2862 opt = strchr (opt, '=') + 1;
2863 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &max_heap_candidate)) {
2864 max_heap = (max_heap_candidate + page_size - 1) & ~(size_t)(page_size - 1);
2865 if (max_heap != max_heap_candidate)
2866 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Rounding up.", "`max-heap-size` size must be a multiple of %d.", page_size);
2868 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`max-heap-size` must be an integer.");
2872 if (g_str_has_prefix (opt, "soft-heap-limit=")) {
2873 opt = strchr (opt, '=') + 1;
2874 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &soft_limit)) {
2875 if (soft_limit <= 0) {
2876 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be positive.");
2880 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be an integer.");
2886 if (g_str_has_prefix (opt, "nursery-size=")) {
2888 opt = strchr (opt, '=') + 1;
2889 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &val)) {
2890 if ((val & (val - 1))) {
2891 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be a power of two.");
2895 if (val < SGEN_MAX_NURSERY_WASTE) {
2896 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.",
2897 "`nursery-size` must be at least %d bytes.", SGEN_MAX_NURSERY_WASTE);
2901 sgen_nursery_size = val;
2902 sgen_nursery_bits = 0;
2903 while (ONE_P << (++ sgen_nursery_bits) != sgen_nursery_size)
2906 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be an integer.");
2912 if (g_str_has_prefix (opt, "save-target-ratio=")) {
2914 opt = strchr (opt, '=') + 1;
2915 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "save-target-ratio", opt,
2916 SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO, &val)) {
2921 if (g_str_has_prefix (opt, "default-allowance-ratio=")) {
2923 opt = strchr (opt, '=') + 1;
2924 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "default-allowance-ratio", opt,
2925 SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, &val)) {
2926 allowance_ratio = val;
2931 if (!strcmp (opt, "cementing")) {
2932 cement_enabled = TRUE;
2935 if (!strcmp (opt, "no-cementing")) {
2936 cement_enabled = FALSE;
2940 if (!strcmp (opt, "precleaning")) {
2941 precleaning_enabled = TRUE;
2944 if (!strcmp (opt, "no-precleaning")) {
2945 precleaning_enabled = FALSE;
2949 if (major_collector.handle_gc_param && major_collector.handle_gc_param (opt))
2952 if (sgen_minor_collector.handle_gc_param && sgen_minor_collector.handle_gc_param (opt))
2955 if (sgen_client_handle_gc_param (opt))
2958 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Ignoring.", "Unknown option `%s`.", opt);
2963 fprintf (stderr, "\n%s must be a comma-delimited list of one or more of the following:\n", MONO_GC_PARAMS_NAME);
2964 fprintf (stderr, " max-heap-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
2965 fprintf (stderr, " soft-heap-limit=n (where N is an integer, possibly with a k, m or a g suffix)\n");
2966 fprintf (stderr, " nursery-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
2967 fprintf (stderr, " major=COLLECTOR (where COLLECTOR is `marksweep', `marksweep-conc', `marksweep-par')\n");
2968 fprintf (stderr, " minor=COLLECTOR (where COLLECTOR is `simple' or `split')\n");
2969 fprintf (stderr, " wbarrier=WBARRIER (where WBARRIER is `remset' or `cardtable')\n");
2970 fprintf (stderr, " [no-]cementing\n");
2971 if (major_collector.print_gc_param_usage)
2972 major_collector.print_gc_param_usage ();
2973 if (sgen_minor_collector.print_gc_param_usage)
2974 sgen_minor_collector.print_gc_param_usage ();
2975 sgen_client_print_gc_params_usage ();
2976 fprintf (stderr, " Experimental options:\n");
2977 fprintf (stderr, " save-target-ratio=R (where R must be between %.2f - %.2f).\n", SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO);
2978 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);
2979 fprintf (stderr, "\n");
2981 usage_printed = TRUE;
2986 if (major_collector_opt)
2987 g_free (major_collector_opt);
2989 if (minor_collector_opt)
2990 g_free (minor_collector_opt);
2994 sgen_pinning_init ();
2995 sgen_cement_init (cement_enabled);
2997 if ((env = g_getenv (MONO_GC_DEBUG_NAME))) {
2998 gboolean usage_printed = FALSE;
3000 opts = g_strsplit (env, ",", -1);
3001 for (ptr = opts; ptr && *ptr; ptr ++) {
3003 if (!strcmp (opt, ""))
3005 if (opt [0] >= '0' && opt [0] <= '9') {
3006 gc_debug_level = atoi (opt);
3011 char *rf = g_strdup_printf ("%s.%d", opt, mono_process_current_pid ());
3012 gc_debug_file = fopen (rf, "wb");
3014 gc_debug_file = stderr;
3017 } else if (!strcmp (opt, "print-allowance")) {
3018 debug_print_allowance = TRUE;
3019 } else if (!strcmp (opt, "print-pinning")) {
3020 sgen_pin_stats_enable ();
3021 } else if (!strcmp (opt, "verify-before-allocs")) {
3022 verify_before_allocs = 1;
3023 has_per_allocation_action = TRUE;
3024 } else if (g_str_has_prefix (opt, "verify-before-allocs=")) {
3025 char *arg = strchr (opt, '=') + 1;
3026 verify_before_allocs = atoi (arg);
3027 has_per_allocation_action = TRUE;
3028 } else if (!strcmp (opt, "collect-before-allocs")) {
3029 collect_before_allocs = 1;
3030 has_per_allocation_action = TRUE;
3031 } else if (g_str_has_prefix (opt, "collect-before-allocs=")) {
3032 char *arg = strchr (opt, '=') + 1;
3033 has_per_allocation_action = TRUE;
3034 collect_before_allocs = atoi (arg);
3035 } else if (!strcmp (opt, "verify-before-collections")) {
3036 whole_heap_check_before_collection = TRUE;
3037 } else if (!strcmp (opt, "check-at-minor-collections")) {
3038 consistency_check_at_minor_collection = TRUE;
3039 nursery_clear_policy = CLEAR_AT_GC;
3040 } else if (!strcmp (opt, "mod-union-consistency-check")) {
3041 if (!major_collector.is_concurrent) {
3042 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`mod-union-consistency-check` only works with concurrent major collector.");
3045 mod_union_consistency_check = TRUE;
3046 } else if (!strcmp (opt, "check-mark-bits")) {
3047 check_mark_bits_after_major_collection = TRUE;
3048 } else if (!strcmp (opt, "check-nursery-pinned")) {
3049 check_nursery_objects_pinned = TRUE;
3050 } else if (!strcmp (opt, "clear-at-gc")) {
3051 nursery_clear_policy = CLEAR_AT_GC;
3052 } else if (!strcmp (opt, "clear-nursery-at-gc")) {
3053 nursery_clear_policy = CLEAR_AT_GC;
3054 } else if (!strcmp (opt, "clear-at-tlab-creation")) {
3055 nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
3056 } else if (!strcmp (opt, "debug-clear-at-tlab-creation")) {
3057 nursery_clear_policy = CLEAR_AT_TLAB_CREATION_DEBUG;
3058 } else if (!strcmp (opt, "check-scan-starts")) {
3059 do_scan_starts_check = TRUE;
3060 } else if (!strcmp (opt, "verify-nursery-at-minor-gc")) {
3061 do_verify_nursery = TRUE;
3062 } else if (!strcmp (opt, "check-concurrent")) {
3063 if (!major_collector.is_concurrent) {
3064 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`check-concurrent` only works with concurrent major collectors.");
3067 nursery_clear_policy = CLEAR_AT_GC;
3068 do_concurrent_checks = TRUE;
3069 } else if (!strcmp (opt, "dump-nursery-at-minor-gc")) {
3070 do_dump_nursery_content = TRUE;
3071 } else if (!strcmp (opt, "disable-minor")) {
3072 disable_minor_collections = TRUE;
3073 } else if (!strcmp (opt, "disable-major")) {
3074 disable_major_collections = TRUE;
3075 } else if (g_str_has_prefix (opt, "heap-dump=")) {
3076 char *filename = strchr (opt, '=') + 1;
3077 nursery_clear_policy = CLEAR_AT_GC;
3078 sgen_debug_enable_heap_dump (filename);
3079 } else if (g_str_has_prefix (opt, "binary-protocol=")) {
3080 char *filename = strchr (opt, '=') + 1;
3081 char *colon = strrchr (filename, ':');
3084 if (!mono_gc_parse_environment_string_extract_number (colon + 1, &limit)) {
3085 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring limit.", "Binary protocol file size limit must be an integer.");
3090 binary_protocol_init (filename, (long long)limit);
3091 } else if (!strcmp (opt, "nursery-canaries")) {
3092 do_verify_nursery = TRUE;
3093 enable_nursery_canaries = TRUE;
3094 } else if (!sgen_client_handle_gc_debug (opt)) {
3095 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "Unknown option `%s`.", opt);
3100 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);
3101 fprintf (stderr, "Valid <option>s are:\n");
3102 fprintf (stderr, " collect-before-allocs[=<n>]\n");
3103 fprintf (stderr, " verify-before-allocs[=<n>]\n");
3104 fprintf (stderr, " check-at-minor-collections\n");
3105 fprintf (stderr, " check-mark-bits\n");
3106 fprintf (stderr, " check-nursery-pinned\n");
3107 fprintf (stderr, " verify-before-collections\n");
3108 fprintf (stderr, " verify-nursery-at-minor-gc\n");
3109 fprintf (stderr, " dump-nursery-at-minor-gc\n");
3110 fprintf (stderr, " disable-minor\n");
3111 fprintf (stderr, " disable-major\n");
3112 fprintf (stderr, " check-concurrent\n");
3113 fprintf (stderr, " clear-[nursery-]at-gc\n");
3114 fprintf (stderr, " clear-at-tlab-creation\n");
3115 fprintf (stderr, " debug-clear-at-tlab-creation\n");
3116 fprintf (stderr, " check-scan-starts\n");
3117 fprintf (stderr, " print-allowance\n");
3118 fprintf (stderr, " print-pinning\n");
3119 fprintf (stderr, " heap-dump=<filename>\n");
3120 fprintf (stderr, " binary-protocol=<filename>[:<file-size-limit>]\n");
3121 fprintf (stderr, " nursery-canaries\n");
3122 sgen_client_print_gc_debug_usage ();
3123 fprintf (stderr, "\n");
3125 usage_printed = TRUE;
3131 if (check_mark_bits_after_major_collection)
3132 nursery_clear_policy = CLEAR_AT_GC;
3134 if (major_collector.post_param_init)
3135 major_collector.post_param_init (&major_collector);
3137 if (major_collector.needs_thread_pool)
3138 sgen_workers_init (1);
3140 sgen_memgov_init (max_heap, soft_limit, debug_print_allowance, allowance_ratio, save_target);
3142 memset (&remset, 0, sizeof (remset));
3144 sgen_card_table_init (&remset);
3146 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");
3152 sgen_get_nursery_clear_policy (void)
3154 return nursery_clear_policy;
3160 mono_coop_mutex_lock (&gc_mutex);
3164 sgen_gc_unlock (void)
3166 mono_coop_mutex_unlock (&gc_mutex);
3170 sgen_major_collector_iterate_live_block_ranges (sgen_cardtable_block_callback callback)
3172 major_collector.iterate_live_block_ranges (callback);
3176 sgen_get_major_collector (void)
3178 return &major_collector;
3182 sgen_get_remset (void)
3188 count_cards (long long *major_total, long long *major_marked, long long *los_total, long long *los_marked)
3190 sgen_get_major_collector ()->count_cards (major_total, major_marked);
3191 sgen_los_count_cards (los_total, los_marked);
3194 static gboolean world_is_stopped = FALSE;
3196 /* LOCKING: assumes the GC lock is held */
3198 sgen_stop_world (int generation)
3200 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3202 SGEN_ASSERT (0, !world_is_stopped, "Why are we stopping a stopped world?");
3204 binary_protocol_world_stopping (generation, sgen_timestamp (), (gpointer) (gsize) mono_native_thread_id_get ());
3206 sgen_client_stop_world (generation);
3208 world_is_stopped = TRUE;
3210 if (binary_protocol_is_heavy_enabled ())
3211 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3212 binary_protocol_world_stopped (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3215 /* LOCKING: assumes the GC lock is held */
3217 sgen_restart_world (int generation, GGTimingInfo *timing)
3219 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3221 SGEN_ASSERT (0, world_is_stopped, "Why are we restarting a running world?");
3223 if (binary_protocol_is_heavy_enabled ())
3224 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3225 binary_protocol_world_restarting (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3227 world_is_stopped = FALSE;
3229 sgen_client_restart_world (generation, timing);
3231 binary_protocol_world_restarted (generation, sgen_timestamp ());
3233 if (sgen_client_bridge_need_processing ())
3234 sgen_client_bridge_processing_finish (generation);
3236 sgen_memgov_collection_end (generation, timing, timing ? 2 : 0);
3240 sgen_is_world_stopped (void)
3242 return world_is_stopped;
3246 sgen_check_whole_heap_stw (void)
3248 sgen_stop_world (0);
3249 sgen_clear_nursery_fragments ();
3250 sgen_check_whole_heap (FALSE);
3251 sgen_restart_world (0, NULL);
3255 sgen_timestamp (void)
3257 SGEN_TV_DECLARE (timestamp);
3258 SGEN_TV_GETTIME (timestamp);
3259 return SGEN_TV_ELAPSED (sgen_init_timestamp, timestamp);
3262 #endif /* HAVE_SGEN_GC */