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;
335 gboolean sgen_try_free_some_memory;
337 #define SCAN_START_SIZE SGEN_SCAN_START_SIZE
339 size_t degraded_mode = 0;
341 static mword bytes_pinned_from_failed_allocation = 0;
343 GCMemSection *nursery_section = NULL;
344 static volatile mword lowest_heap_address = ~(mword)0;
345 static volatile mword highest_heap_address = 0;
347 MonoCoopMutex sgen_interruption_mutex;
349 int current_collection_generation = -1;
350 static volatile gboolean concurrent_collection_in_progress = FALSE;
352 /* objects that are ready to be finalized */
353 static SgenPointerQueue fin_ready_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
354 static SgenPointerQueue critical_fin_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
356 /* registered roots: the key to the hash is the root start address */
358 * Different kinds of roots are kept separate to speed up pin_from_roots () for example.
360 SgenHashTable roots_hash [ROOT_TYPE_NUM] = {
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),
363 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL)
365 static mword roots_size = 0; /* amount of memory in the root set */
367 /* The size of a TLAB */
368 /* The bigger the value, the less often we have to go to the slow path to allocate a new
369 * one, but the more space is wasted by threads not allocating much memory.
371 * FIXME: Make this self-tuning for each thread.
373 guint32 tlab_size = (1024 * 4);
375 #define MAX_SMALL_OBJ_SIZE SGEN_MAX_SMALL_OBJ_SIZE
377 #define ALLOC_ALIGN SGEN_ALLOC_ALIGN
379 #define ALIGN_UP SGEN_ALIGN_UP
381 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
382 MonoNativeThreadId main_gc_thread = NULL;
385 /*Object was pinned during the current collection*/
386 static mword objects_pinned;
389 * ######################################################################
390 * ######## Macros and function declarations.
391 * ######################################################################
394 typedef SgenGrayQueue GrayQueue;
396 /* forward declarations */
397 static void scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx);
399 static void pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx);
400 static void finish_gray_stack (int generation, ScanCopyContext ctx);
403 SgenMajorCollector major_collector;
404 SgenMinorCollector sgen_minor_collector;
405 /* FIXME: get rid of this */
406 static GrayQueue gray_queue;
408 static SgenRememberedSet remset;
410 /* The gray queue to use from the main collection thread. */
411 #define WORKERS_DISTRIBUTE_GRAY_QUEUE (&gray_queue)
414 * The gray queue a worker job must use. If we're not parallel or
415 * concurrent, we use the main gray queue.
417 static SgenGrayQueue*
418 sgen_workers_get_job_gray_queue (WorkerData *worker_data)
420 return worker_data ? &worker_data->private_gray_queue : WORKERS_DISTRIBUTE_GRAY_QUEUE;
424 gray_queue_redirect (SgenGrayQueue *queue)
426 gboolean wake = FALSE;
429 GrayQueueSection *section = sgen_gray_object_dequeue_section (queue);
432 sgen_section_gray_queue_enqueue ((SgenSectionGrayQueue *)queue->alloc_prepare_data, section);
437 g_assert (concurrent_collection_in_progress);
438 sgen_workers_ensure_awake ();
443 gray_queue_enable_redirect (SgenGrayQueue *queue)
445 if (!concurrent_collection_in_progress)
448 sgen_gray_queue_set_alloc_prepare (queue, gray_queue_redirect, sgen_workers_get_distribute_section_gray_queue ());
449 gray_queue_redirect (queue);
453 sgen_scan_area_with_callback (char *start, char *end, IterateObjectCallbackFunc callback, void *data, gboolean allow_flags, gboolean fail_on_canaries)
455 while (start < end) {
459 if (!*(void**)start) {
460 start += sizeof (void*); /* should be ALLOC_ALIGN, really */
465 if (!(obj = (char *)SGEN_OBJECT_IS_FORWARDED (start)))
471 if (!sgen_client_object_is_array_fill ((GCObject*)obj)) {
472 CHECK_CANARY_FOR_OBJECT ((GCObject*)obj, fail_on_canaries);
473 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
474 callback ((GCObject*)obj, size, data);
475 CANARIFY_SIZE (size);
477 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
485 * sgen_add_to_global_remset:
487 * The global remset contains locations which point into newspace after
488 * a minor collection. This can happen if the objects they point to are pinned.
490 * LOCKING: If called from a parallel collector, the global remset
491 * lock must be held. For serial collectors that is not necessary.
494 sgen_add_to_global_remset (gpointer ptr, GCObject *obj)
496 SGEN_ASSERT (5, sgen_ptr_in_nursery (obj), "Target pointer of global remset must be in the nursery");
498 HEAVY_STAT (++stat_wbarrier_add_to_global_remset);
500 if (!major_collector.is_concurrent) {
501 SGEN_ASSERT (5, current_collection_generation != -1, "Global remsets can only be added during collections");
503 if (current_collection_generation == -1)
504 SGEN_ASSERT (5, sgen_concurrent_collection_in_progress (), "Global remsets outside of collection pauses can only be added by the concurrent collector");
507 if (!object_is_pinned (obj))
508 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");
509 else if (sgen_cement_lookup_or_register (obj))
512 remset.record_pointer (ptr);
514 sgen_pin_stats_register_global_remset (obj);
516 SGEN_LOG (8, "Adding global remset for %p", ptr);
517 binary_protocol_global_remset (ptr, obj, (gpointer)SGEN_LOAD_VTABLE (obj));
521 * sgen_drain_gray_stack:
523 * Scan objects in the gray stack until the stack is empty. This should be called
524 * frequently after each object is copied, to achieve better locality and cache
529 sgen_drain_gray_stack (ScanCopyContext ctx)
531 ScanObjectFunc scan_func = ctx.ops->scan_object;
532 GrayQueue *queue = ctx.queue;
534 if (ctx.ops->drain_gray_stack)
535 return ctx.ops->drain_gray_stack (queue);
540 GRAY_OBJECT_DEQUEUE (queue, &obj, &desc);
543 SGEN_LOG (9, "Precise gray object scan %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
544 scan_func (obj, desc, queue);
550 * Addresses in the pin queue are already sorted. This function finds
551 * the object header for each address and pins the object. The
552 * addresses must be inside the nursery section. The (start of the)
553 * address array is overwritten with the addresses of the actually
554 * pinned objects. Return the number of pinned objects.
557 pin_objects_from_nursery_pin_queue (gboolean do_scan_objects, ScanCopyContext ctx)
559 GCMemSection *section = nursery_section;
560 void **start = sgen_pinning_get_entry (section->pin_queue_first_entry);
561 void **end = sgen_pinning_get_entry (section->pin_queue_last_entry);
562 void *start_nursery = section->data;
563 void *end_nursery = section->next_data;
568 void *pinning_front = start_nursery;
570 void **definitely_pinned = start;
571 ScanObjectFunc scan_func = ctx.ops->scan_object;
572 SgenGrayQueue *queue = ctx.queue;
574 sgen_nursery_allocator_prepare_for_pinning ();
576 while (start < end) {
577 GCObject *obj_to_pin = NULL;
578 size_t obj_to_pin_size = 0;
583 SGEN_ASSERT (0, addr >= start_nursery && addr < end_nursery, "Potential pinning address out of range");
584 SGEN_ASSERT (0, addr >= last, "Pin queue not sorted");
591 SGEN_LOG (5, "Considering pinning addr %p", addr);
592 /* We've already processed everything up to pinning_front. */
593 if (addr < pinning_front) {
599 * Find the closest scan start <= addr. We might search backward in the
600 * scan_starts array because entries might be NULL. In the worst case we
601 * start at start_nursery.
603 idx = ((char*)addr - (char*)section->data) / SCAN_START_SIZE;
604 SGEN_ASSERT (0, idx < section->num_scan_start, "Scan start index out of range");
605 search_start = (void*)section->scan_starts [idx];
606 if (!search_start || search_start > addr) {
609 search_start = section->scan_starts [idx];
610 if (search_start && search_start <= addr)
613 if (!search_start || search_start > addr)
614 search_start = start_nursery;
618 * If the pinning front is closer than the scan start we found, start
619 * searching at the front.
621 if (search_start < pinning_front)
622 search_start = pinning_front;
625 * Now addr should be in an object a short distance from search_start.
627 * search_start must point to zeroed mem or point to an object.
630 size_t obj_size, canarified_obj_size;
633 if (!*(void**)search_start) {
634 search_start = (void*)ALIGN_UP ((mword)search_start + sizeof (gpointer));
635 /* The loop condition makes sure we don't overrun addr. */
639 canarified_obj_size = obj_size = ALIGN_UP (safe_object_get_size ((GCObject*)search_start));
642 * Filler arrays are marked by an invalid sync word. We don't
643 * consider them for pinning. They are not delimited by canaries,
646 if (!sgen_client_object_is_array_fill ((GCObject*)search_start)) {
647 CHECK_CANARY_FOR_OBJECT (search_start, TRUE);
648 CANARIFY_SIZE (canarified_obj_size);
650 if (addr >= search_start && (char*)addr < (char*)search_start + obj_size) {
651 /* This is the object we're looking for. */
652 obj_to_pin = (GCObject*)search_start;
653 obj_to_pin_size = canarified_obj_size;
658 /* Skip to the next object */
659 search_start = (void*)((char*)search_start + canarified_obj_size);
660 } while (search_start <= addr);
662 /* We've searched past the address we were looking for. */
664 pinning_front = search_start;
665 goto next_pin_queue_entry;
669 * We've found an object to pin. It might still be a dummy array, but we
670 * can advance the pinning front in any case.
672 pinning_front = (char*)obj_to_pin + obj_to_pin_size;
675 * If this is a dummy array marking the beginning of a nursery
676 * fragment, we don't pin it.
678 if (sgen_client_object_is_array_fill (obj_to_pin))
679 goto next_pin_queue_entry;
682 * Finally - pin the object!
684 desc = sgen_obj_get_descriptor_safe (obj_to_pin);
685 if (do_scan_objects) {
686 scan_func (obj_to_pin, desc, queue);
688 SGEN_LOG (4, "Pinned object %p, vtable %p (%s), count %d\n",
689 obj_to_pin, *(void**)obj_to_pin, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj_to_pin)), count);
690 binary_protocol_pin (obj_to_pin,
691 (gpointer)LOAD_VTABLE (obj_to_pin),
692 safe_object_get_size (obj_to_pin));
694 pin_object (obj_to_pin);
695 GRAY_OBJECT_ENQUEUE (queue, obj_to_pin, desc);
696 sgen_pin_stats_register_object (obj_to_pin, obj_to_pin_size);
697 definitely_pinned [count] = obj_to_pin;
701 next_pin_queue_entry:
705 sgen_client_nursery_objects_pinned (definitely_pinned, count);
706 stat_pinned_objects += count;
711 pin_objects_in_nursery (gboolean do_scan_objects, ScanCopyContext ctx)
715 if (nursery_section->pin_queue_first_entry == nursery_section->pin_queue_last_entry)
718 reduced_to = pin_objects_from_nursery_pin_queue (do_scan_objects, ctx);
719 nursery_section->pin_queue_last_entry = nursery_section->pin_queue_first_entry + reduced_to;
723 * This function is only ever called (via `collector_pin_object()` in `sgen-copy-object.h`)
724 * when we can't promote an object because we're out of memory.
727 sgen_pin_object (GCObject *object, GrayQueue *queue)
730 * All pinned objects are assumed to have been staged, so we need to stage as well.
731 * Also, the count of staged objects shows that "late pinning" happened.
733 sgen_pin_stage_ptr (object);
735 SGEN_PIN_OBJECT (object);
736 binary_protocol_pin (object, (gpointer)LOAD_VTABLE (object), safe_object_get_size (object));
739 sgen_pin_stats_register_object (object, safe_object_get_size (object));
741 GRAY_OBJECT_ENQUEUE (queue, object, sgen_obj_get_descriptor_safe (object));
744 /* Sort the addresses in array in increasing order.
745 * Done using a by-the book heap sort. Which has decent and stable performance, is pretty cache efficient.
748 sgen_sort_addresses (void **array, size_t size)
753 for (i = 1; i < size; ++i) {
756 size_t parent = (child - 1) / 2;
758 if (array [parent] >= array [child])
761 tmp = array [parent];
762 array [parent] = array [child];
769 for (i = size - 1; i > 0; --i) {
772 array [i] = array [0];
778 while (root * 2 + 1 <= end) {
779 size_t child = root * 2 + 1;
781 if (child < end && array [child] < array [child + 1])
783 if (array [root] >= array [child])
787 array [root] = array [child];
796 * Scan the memory between start and end and queue values which could be pointers
797 * to the area between start_nursery and end_nursery for later consideration.
798 * Typically used for thread stacks.
801 sgen_conservatively_pin_objects_from (void **start, void **end, void *start_nursery, void *end_nursery, int pin_type)
805 SGEN_ASSERT (0, ((mword)start & (SIZEOF_VOID_P - 1)) == 0, "Why are we scanning for references in unaligned memory ?");
807 #if defined(VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE) && !defined(_WIN64)
808 VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE (start, (char*)end - (char*)start);
811 while (start < end) {
813 * *start can point to the middle of an object
814 * note: should we handle pointing at the end of an object?
815 * pinning in C# code disallows pointing at the end of an object
816 * but there is some small chance that an optimizing C compiler
817 * may keep the only reference to an object by pointing
818 * at the end of it. We ignore this small chance for now.
819 * Pointers to the end of an object are indistinguishable
820 * from pointers to the start of the next object in memory
821 * so if we allow that we'd need to pin two objects...
822 * We queue the pointer in an array, the
823 * array will then be sorted and uniqued. This way
824 * we can coalesce several pinning pointers and it should
825 * be faster since we'd do a memory scan with increasing
826 * addresses. Note: we can align the address to the allocation
827 * alignment, so the unique process is more effective.
829 mword addr = (mword)*start;
830 addr &= ~(ALLOC_ALIGN - 1);
831 if (addr >= (mword)start_nursery && addr < (mword)end_nursery) {
832 SGEN_LOG (6, "Pinning address %p from %p", (void*)addr, start);
833 sgen_pin_stage_ptr ((void*)addr);
834 binary_protocol_pin_stage (start, (void*)addr);
835 sgen_pin_stats_register_address ((char*)addr, pin_type);
841 SGEN_LOG (7, "found %d potential pinned heap pointers", count);
845 * The first thing we do in a collection is to identify pinned objects.
846 * This function considers all the areas of memory that need to be
847 * conservatively scanned.
850 pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx)
854 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);
855 /* objects pinned from the API are inside these roots */
856 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_PINNED], void **, start_root, RootRecord *, root) {
857 SGEN_LOG (6, "Pinned roots %p-%p", start_root, root->end_root);
858 sgen_conservatively_pin_objects_from (start_root, (void**)root->end_root, start_nursery, end_nursery, PIN_TYPE_OTHER);
859 } SGEN_HASH_TABLE_FOREACH_END;
860 /* now deal with the thread stacks
861 * in the future we should be able to conservatively scan only:
862 * *) the cpu registers
863 * *) the unmanaged stack frames
864 * *) the _last_ managed stack frame
865 * *) pointers slots in managed frames
867 sgen_client_scan_thread_data (start_nursery, end_nursery, FALSE, ctx);
871 single_arg_user_copy_or_mark (GCObject **obj, void *gc_data)
873 ScanCopyContext *ctx = (ScanCopyContext *)gc_data;
874 ctx->ops->copy_or_mark_object (obj, ctx->queue);
878 * The memory area from start_root to end_root contains pointers to objects.
879 * Their position is precisely described by @desc (this means that the pointer
880 * can be either NULL or the pointer to the start of an object).
881 * This functions copies them to to_space updates them.
883 * This function is not thread-safe!
886 precisely_scan_objects_from (void** start_root, void** end_root, char* n_start, char *n_end, SgenDescriptor desc, ScanCopyContext ctx)
888 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
889 SgenGrayQueue *queue = ctx.queue;
891 switch (desc & ROOT_DESC_TYPE_MASK) {
892 case ROOT_DESC_BITMAP:
893 desc >>= ROOT_DESC_TYPE_SHIFT;
895 if ((desc & 1) && *start_root) {
896 copy_func ((GCObject**)start_root, queue);
897 SGEN_LOG (9, "Overwrote root at %p with %p", start_root, *start_root);
903 case ROOT_DESC_COMPLEX: {
904 gsize *bitmap_data = (gsize *)sgen_get_complex_descriptor_bitmap (desc);
905 gsize bwords = (*bitmap_data) - 1;
906 void **start_run = start_root;
908 while (bwords-- > 0) {
909 gsize bmap = *bitmap_data++;
910 void **objptr = start_run;
912 if ((bmap & 1) && *objptr) {
913 copy_func ((GCObject**)objptr, queue);
914 SGEN_LOG (9, "Overwrote root at %p with %p", objptr, *objptr);
919 start_run += GC_BITS_PER_WORD;
923 case ROOT_DESC_USER: {
924 SgenUserRootMarkFunc marker = sgen_get_user_descriptor_func (desc);
925 marker (start_root, single_arg_user_copy_or_mark, &ctx);
928 case ROOT_DESC_RUN_LEN:
929 g_assert_not_reached ();
931 g_assert_not_reached ();
936 reset_heap_boundaries (void)
938 lowest_heap_address = ~(mword)0;
939 highest_heap_address = 0;
943 sgen_update_heap_boundaries (mword low, mword high)
948 old = lowest_heap_address;
951 } while (SGEN_CAS_PTR ((gpointer*)&lowest_heap_address, (gpointer)low, (gpointer)old) != (gpointer)old);
954 old = highest_heap_address;
957 } while (SGEN_CAS_PTR ((gpointer*)&highest_heap_address, (gpointer)high, (gpointer)old) != (gpointer)old);
961 * Allocate and setup the data structures needed to be able to allocate objects
962 * in the nursery. The nursery is stored in nursery_section.
967 GCMemSection *section;
974 SGEN_LOG (2, "Allocating nursery size: %zu", (size_t)sgen_nursery_size);
975 /* later we will alloc a larger area for the nursery but only activate
976 * what we need. The rest will be used as expansion if we have too many pinned
977 * objects in the existing nursery.
979 /* FIXME: handle OOM */
980 section = (GCMemSection *)sgen_alloc_internal (INTERNAL_MEM_SECTION);
982 alloc_size = sgen_nursery_size;
984 /* If there isn't enough space even for the nursery we should simply abort. */
985 g_assert (sgen_memgov_try_alloc_space (alloc_size, SPACE_NURSERY));
987 data = (char *)major_collector.alloc_heap (alloc_size, alloc_size, DEFAULT_NURSERY_BITS);
988 sgen_update_heap_boundaries ((mword)data, (mword)(data + sgen_nursery_size));
989 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 ());
990 section->data = section->next_data = data;
991 section->size = alloc_size;
992 section->end_data = data + sgen_nursery_size;
993 scan_starts = (alloc_size + SCAN_START_SIZE - 1) / SCAN_START_SIZE;
994 section->scan_starts = (char **)sgen_alloc_internal_dynamic (sizeof (char*) * scan_starts, INTERNAL_MEM_SCAN_STARTS, TRUE);
995 section->num_scan_start = scan_starts;
997 nursery_section = section;
999 sgen_nursery_allocator_set_nursery_bounds (data, data + sgen_nursery_size);
1003 mono_gc_get_logfile (void)
1005 return gc_debug_file;
1009 scan_finalizer_entries (SgenPointerQueue *fin_queue, ScanCopyContext ctx)
1011 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
1012 SgenGrayQueue *queue = ctx.queue;
1015 for (i = 0; i < fin_queue->next_slot; ++i) {
1016 GCObject *obj = (GCObject *)fin_queue->data [i];
1019 SGEN_LOG (5, "Scan of fin ready object: %p (%s)\n", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
1020 copy_func ((GCObject**)&fin_queue->data [i], queue);
1025 generation_name (int generation)
1027 switch (generation) {
1028 case GENERATION_NURSERY: return "nursery";
1029 case GENERATION_OLD: return "old";
1030 default: g_assert_not_reached ();
1035 sgen_generation_name (int generation)
1037 return generation_name (generation);
1041 finish_gray_stack (int generation, ScanCopyContext ctx)
1045 int done_with_ephemerons, ephemeron_rounds = 0;
1046 char *start_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_start () : NULL;
1047 char *end_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_end () : (char*)-1;
1048 SgenGrayQueue *queue = ctx.queue;
1050 binary_protocol_finish_gray_stack_start (sgen_timestamp (), generation);
1052 * We copied all the reachable objects. Now it's the time to copy
1053 * the objects that were not referenced by the roots, but by the copied objects.
1054 * we built a stack of objects pointed to by gray_start: they are
1055 * additional roots and we may add more items as we go.
1056 * We loop until gray_start == gray_objects which means no more objects have
1057 * been added. Note this is iterative: no recursion is involved.
1058 * We need to walk the LO list as well in search of marked big objects
1059 * (use a flag since this is needed only on major collections). We need to loop
1060 * here as well, so keep a counter of marked LO (increasing it in copy_object).
1061 * To achieve better cache locality and cache usage, we drain the gray stack
1062 * frequently, after each object is copied, and just finish the work here.
1064 sgen_drain_gray_stack (ctx);
1066 SGEN_LOG (2, "%s generation done", generation_name (generation));
1069 Reset bridge data, we might have lingering data from a previous collection if this is a major
1070 collection trigged by minor overflow.
1072 We must reset the gathered bridges since their original block might be evacuated due to major
1073 fragmentation in the meanwhile and the bridge code should not have to deal with that.
1075 if (sgen_client_bridge_need_processing ())
1076 sgen_client_bridge_reset_data ();
1079 * Walk the ephemeron tables marking all values with reachable keys. This must be completely done
1080 * before processing finalizable objects and non-tracking weak links to avoid finalizing/clearing
1081 * objects that are in fact reachable.
1083 done_with_ephemerons = 0;
1085 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1086 sgen_drain_gray_stack (ctx);
1088 } while (!done_with_ephemerons);
1090 sgen_client_mark_togglerefs (start_addr, end_addr, ctx);
1092 if (sgen_client_bridge_need_processing ()) {
1093 /*Make sure the gray stack is empty before we process bridge objects so we get liveness right*/
1094 sgen_drain_gray_stack (ctx);
1095 sgen_collect_bridge_objects (generation, ctx);
1096 if (generation == GENERATION_OLD)
1097 sgen_collect_bridge_objects (GENERATION_NURSERY, ctx);
1100 Do the first bridge step here, as the collector liveness state will become useless after that.
1102 An important optimization is to only proccess the possibly dead part of the object graph and skip
1103 over all live objects as we transitively know everything they point must be alive too.
1105 The above invariant is completely wrong if we let the gray queue be drained and mark/copy everything.
1107 This has the unfortunate side effect of making overflow collections perform the first step twice, but
1108 given we now have heuristics that perform major GC in anticipation of minor overflows this should not
1111 sgen_client_bridge_processing_stw_step ();
1115 Make sure we drain the gray stack before processing disappearing links and finalizers.
1116 If we don't make sure it is empty we might wrongly see a live object as dead.
1118 sgen_drain_gray_stack (ctx);
1121 We must clear weak links that don't track resurrection before processing object ready for
1122 finalization so they can be cleared before that.
1124 sgen_null_link_in_range (generation, ctx, FALSE);
1125 if (generation == GENERATION_OLD)
1126 sgen_null_link_in_range (GENERATION_NURSERY, ctx, FALSE);
1129 /* walk the finalization queue and move also the objects that need to be
1130 * finalized: use the finalized objects as new roots so the objects they depend
1131 * on are also not reclaimed. As with the roots above, only objects in the nursery
1132 * are marked/copied.
1134 sgen_finalize_in_range (generation, ctx);
1135 if (generation == GENERATION_OLD)
1136 sgen_finalize_in_range (GENERATION_NURSERY, ctx);
1137 /* drain the new stack that might have been created */
1138 SGEN_LOG (6, "Precise scan of gray area post fin");
1139 sgen_drain_gray_stack (ctx);
1142 * This must be done again after processing finalizable objects since CWL slots are cleared only after the key is finalized.
1144 done_with_ephemerons = 0;
1146 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1147 sgen_drain_gray_stack (ctx);
1149 } while (!done_with_ephemerons);
1151 sgen_client_clear_unreachable_ephemerons (ctx);
1154 * We clear togglerefs only after all possible chances of revival are done.
1155 * This is semantically more inline with what users expect and it allows for
1156 * user finalizers to correctly interact with TR objects.
1158 sgen_client_clear_togglerefs (start_addr, end_addr, ctx);
1161 SGEN_LOG (2, "Finalize queue handling scan for %s generation: %ld usecs %d ephemeron rounds", generation_name (generation), TV_ELAPSED (atv, btv), ephemeron_rounds);
1164 * handle disappearing links
1165 * Note we do this after checking the finalization queue because if an object
1166 * survives (at least long enough to be finalized) we don't clear the link.
1167 * This also deals with a possible issue with the monitor reclamation: with the Boehm
1168 * GC a finalized object my lose the monitor because it is cleared before the finalizer is
1171 g_assert (sgen_gray_object_queue_is_empty (queue));
1173 sgen_null_link_in_range (generation, ctx, TRUE);
1174 if (generation == GENERATION_OLD)
1175 sgen_null_link_in_range (GENERATION_NURSERY, ctx, TRUE);
1176 if (sgen_gray_object_queue_is_empty (queue))
1178 sgen_drain_gray_stack (ctx);
1181 g_assert (sgen_gray_object_queue_is_empty (queue));
1183 sgen_gray_object_queue_trim_free_list (queue);
1184 binary_protocol_finish_gray_stack_end (sgen_timestamp (), generation);
1188 sgen_check_section_scan_starts (GCMemSection *section)
1191 for (i = 0; i < section->num_scan_start; ++i) {
1192 if (section->scan_starts [i]) {
1193 mword size = safe_object_get_size ((GCObject*) section->scan_starts [i]);
1194 SGEN_ASSERT (0, size >= SGEN_CLIENT_MINIMUM_OBJECT_SIZE && size <= MAX_SMALL_OBJ_SIZE, "Weird object size at scan starts.");
1200 check_scan_starts (void)
1202 if (!do_scan_starts_check)
1204 sgen_check_section_scan_starts (nursery_section);
1205 major_collector.check_scan_starts ();
1209 scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx)
1213 SGEN_HASH_TABLE_FOREACH (&roots_hash [root_type], void **, start_root, RootRecord *, root) {
1214 SGEN_LOG (6, "Precise root scan %p-%p (desc: %p)", start_root, root->end_root, (void*)root->root_desc);
1215 precisely_scan_objects_from (start_root, (void**)root->end_root, addr_start, addr_end, root->root_desc, ctx);
1216 } SGEN_HASH_TABLE_FOREACH_END;
1222 static gboolean inited = FALSE;
1227 mono_counters_register ("Collection max time", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME | MONO_COUNTER_MONOTONIC, &time_max);
1229 mono_counters_register ("Minor fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pre_collection_fragment_clear);
1230 mono_counters_register ("Minor pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pinning);
1231 mono_counters_register ("Minor scan remembered set", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_remsets);
1232 mono_counters_register ("Minor scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_pinned);
1233 mono_counters_register ("Minor scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_roots);
1234 mono_counters_register ("Minor fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_fragment_creation);
1236 mono_counters_register ("Major fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pre_collection_fragment_clear);
1237 mono_counters_register ("Major pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pinning);
1238 mono_counters_register ("Major scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_pinned);
1239 mono_counters_register ("Major scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_roots);
1240 mono_counters_register ("Major scan mod union", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_mod_union);
1241 mono_counters_register ("Major finish gray stack", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_finish_gray_stack);
1242 mono_counters_register ("Major free big objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_free_bigobjs);
1243 mono_counters_register ("Major LOS sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_los_sweep);
1244 mono_counters_register ("Major sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_sweep);
1245 mono_counters_register ("Major fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_fragment_creation);
1247 mono_counters_register ("Number of pinned objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_pinned_objects);
1249 #ifdef HEAVY_STATISTICS
1250 mono_counters_register ("WBarrier remember pointer", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_add_to_global_remset);
1251 mono_counters_register ("WBarrier arrayref copy", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_arrayref_copy);
1252 mono_counters_register ("WBarrier generic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store);
1253 mono_counters_register ("WBarrier generic atomic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store_atomic);
1254 mono_counters_register ("WBarrier set root", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_set_root);
1256 mono_counters_register ("# objects allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_alloced_degraded);
1257 mono_counters_register ("bytes allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_bytes_alloced_degraded);
1259 mono_counters_register ("# copy_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_nursery);
1260 mono_counters_register ("# objects copied (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_nursery);
1261 mono_counters_register ("# copy_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_major);
1262 mono_counters_register ("# objects copied (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_major);
1264 mono_counters_register ("# scan_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_nursery);
1265 mono_counters_register ("# scan_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_major);
1267 mono_counters_register ("Slots allocated in vain", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_slots_allocated_in_vain);
1269 mono_counters_register ("# nursery copy_object() failed from space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_from_space);
1270 mono_counters_register ("# nursery copy_object() failed forwarded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_forwarded);
1271 mono_counters_register ("# nursery copy_object() failed pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_pinned);
1272 mono_counters_register ("# nursery copy_object() failed to space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_to_space);
1274 sgen_nursery_allocator_init_heavy_stats ();
1282 reset_pinned_from_failed_allocation (void)
1284 bytes_pinned_from_failed_allocation = 0;
1288 sgen_set_pinned_from_failed_allocation (mword objsize)
1290 bytes_pinned_from_failed_allocation += objsize;
1294 sgen_collection_is_concurrent (void)
1296 switch (current_collection_generation) {
1297 case GENERATION_NURSERY:
1299 case GENERATION_OLD:
1300 return concurrent_collection_in_progress;
1302 g_error ("Invalid current generation %d", current_collection_generation);
1308 sgen_concurrent_collection_in_progress (void)
1310 return concurrent_collection_in_progress;
1314 SgenThreadPoolJob job;
1315 SgenObjectOperations *ops;
1319 job_remembered_set_scan (void *worker_data_untyped, SgenThreadPoolJob *job)
1321 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1322 ScanJob *job_data = (ScanJob*)job;
1323 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1324 remset.scan_remsets (ctx);
1328 SgenThreadPoolJob job;
1329 SgenObjectOperations *ops;
1333 } ScanFromRegisteredRootsJob;
1336 job_scan_from_registered_roots (void *worker_data_untyped, SgenThreadPoolJob *job)
1338 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1339 ScanFromRegisteredRootsJob *job_data = (ScanFromRegisteredRootsJob*)job;
1340 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1342 scan_from_registered_roots (job_data->heap_start, job_data->heap_end, job_data->root_type, ctx);
1346 SgenThreadPoolJob job;
1347 SgenObjectOperations *ops;
1350 } ScanThreadDataJob;
1353 job_scan_thread_data (void *worker_data_untyped, SgenThreadPoolJob *job)
1355 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1356 ScanThreadDataJob *job_data = (ScanThreadDataJob*)job;
1357 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1359 sgen_client_scan_thread_data (job_data->heap_start, job_data->heap_end, TRUE, ctx);
1363 SgenThreadPoolJob job;
1364 SgenObjectOperations *ops;
1365 SgenPointerQueue *queue;
1366 } ScanFinalizerEntriesJob;
1369 job_scan_finalizer_entries (void *worker_data_untyped, SgenThreadPoolJob *job)
1371 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1372 ScanFinalizerEntriesJob *job_data = (ScanFinalizerEntriesJob*)job;
1373 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1375 scan_finalizer_entries (job_data->queue, ctx);
1379 job_scan_major_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1381 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1382 ScanJob *job_data = (ScanJob*)job;
1383 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1385 g_assert (concurrent_collection_in_progress);
1386 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx);
1390 job_scan_los_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1392 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1393 ScanJob *job_data = (ScanJob*)job;
1394 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1396 g_assert (concurrent_collection_in_progress);
1397 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx);
1401 job_mod_union_preclean (void *worker_data_untyped, SgenThreadPoolJob *job)
1403 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1404 ScanJob *job_data = (ScanJob*)job;
1405 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1407 g_assert (concurrent_collection_in_progress);
1409 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx);
1410 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx);
1414 init_gray_queue (gboolean use_workers)
1417 sgen_workers_init_distribute_gray_queue ();
1418 sgen_gray_object_queue_init (&gray_queue, NULL);
1422 enqueue_scan_from_roots_jobs (char *heap_start, char *heap_end, SgenObjectOperations *ops, gboolean enqueue)
1424 ScanFromRegisteredRootsJob *scrrj;
1425 ScanThreadDataJob *stdj;
1426 ScanFinalizerEntriesJob *sfej;
1428 /* registered roots, this includes static fields */
1430 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots normal", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1432 scrrj->heap_start = heap_start;
1433 scrrj->heap_end = heap_end;
1434 scrrj->root_type = ROOT_TYPE_NORMAL;
1435 sgen_workers_enqueue_job (&scrrj->job, enqueue);
1437 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots wbarrier", 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_WBARRIER;
1442 sgen_workers_enqueue_job (&scrrj->job, enqueue);
1446 stdj = (ScanThreadDataJob*)sgen_thread_pool_job_alloc ("scan thread data", job_scan_thread_data, sizeof (ScanThreadDataJob));
1447 stdj->heap_start = heap_start;
1448 stdj->heap_end = heap_end;
1449 sgen_workers_enqueue_job (&stdj->job, enqueue);
1451 /* Scan the list of objects ready for finalization. */
1453 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1454 sfej->queue = &fin_ready_queue;
1456 sgen_workers_enqueue_job (&sfej->job, enqueue);
1458 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan critical finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1459 sfej->queue = &critical_fin_queue;
1461 sgen_workers_enqueue_job (&sfej->job, enqueue);
1465 * Perform a nursery collection.
1467 * Return whether any objects were late-pinned due to being out of memory.
1470 collect_nursery (SgenGrayQueue *unpin_queue, gboolean finish_up_concurrent_mark)
1472 gboolean needs_major;
1473 size_t max_garbage_amount;
1475 mword fragment_total;
1477 SgenObjectOperations *object_ops = &sgen_minor_collector.serial_ops;
1478 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, &gray_queue);
1482 if (disable_minor_collections)
1485 TV_GETTIME (last_minor_collection_start_tv);
1486 atv = last_minor_collection_start_tv;
1488 binary_protocol_collection_begin (gc_stats.minor_gc_count, GENERATION_NURSERY);
1490 if (do_verify_nursery || do_dump_nursery_content)
1491 sgen_debug_verify_nursery (do_dump_nursery_content);
1493 current_collection_generation = GENERATION_NURSERY;
1495 SGEN_ASSERT (0, !sgen_collection_is_concurrent (), "Why is the nursery collection concurrent?");
1497 reset_pinned_from_failed_allocation ();
1499 check_scan_starts ();
1501 sgen_nursery_alloc_prepare_for_minor ();
1505 nursery_next = sgen_nursery_alloc_get_upper_alloc_bound ();
1506 /* FIXME: optimize later to use the higher address where an object can be present */
1507 nursery_next = MAX (nursery_next, sgen_get_nursery_end ());
1509 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 ()));
1510 max_garbage_amount = nursery_next - sgen_get_nursery_start ();
1511 g_assert (nursery_section->size >= max_garbage_amount);
1513 /* world must be stopped already */
1515 time_minor_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1517 sgen_client_pre_collection_checks ();
1519 nursery_section->next_data = nursery_next;
1521 major_collector.start_nursery_collection ();
1523 sgen_memgov_minor_collection_start ();
1525 init_gray_queue (FALSE);
1527 gc_stats.minor_gc_count ++;
1529 if (whole_heap_check_before_collection) {
1530 sgen_clear_nursery_fragments ();
1531 sgen_check_whole_heap (finish_up_concurrent_mark);
1533 if (consistency_check_at_minor_collection)
1534 sgen_check_consistency ();
1536 sgen_process_fin_stage_entries ();
1538 /* pin from pinned handles */
1539 sgen_init_pinning ();
1540 sgen_client_binary_protocol_mark_start (GENERATION_NURSERY);
1541 pin_from_roots (sgen_get_nursery_start (), nursery_next, ctx);
1542 /* pin cemented objects */
1543 sgen_pin_cemented_objects ();
1544 /* identify pinned objects */
1545 sgen_optimize_pin_queue ();
1546 sgen_pinning_setup_section (nursery_section);
1548 pin_objects_in_nursery (FALSE, ctx);
1549 sgen_pinning_trim_queue_to_section (nursery_section);
1552 time_minor_pinning += TV_ELAPSED (btv, atv);
1553 SGEN_LOG (2, "Finding pinned pointers: %zd in %ld usecs", sgen_get_pinned_count (), TV_ELAPSED (btv, atv));
1554 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1556 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan remset", job_remembered_set_scan, sizeof (ScanJob));
1557 sj->ops = object_ops;
1558 sgen_workers_enqueue_job (&sj->job, FALSE);
1560 /* we don't have complete write barrier yet, so we scan all the old generation sections */
1562 time_minor_scan_remsets += TV_ELAPSED (atv, btv);
1563 SGEN_LOG (2, "Old generation scan: %ld usecs", TV_ELAPSED (atv, btv));
1565 sgen_pin_stats_print_class_stats ();
1567 /* FIXME: Why do we do this at this specific, seemingly random, point? */
1568 sgen_client_collecting_minor (&fin_ready_queue, &critical_fin_queue);
1571 time_minor_scan_pinned += TV_ELAPSED (btv, atv);
1573 enqueue_scan_from_roots_jobs (sgen_get_nursery_start (), nursery_next, object_ops, FALSE);
1576 time_minor_scan_roots += TV_ELAPSED (atv, btv);
1578 finish_gray_stack (GENERATION_NURSERY, ctx);
1581 time_minor_finish_gray_stack += TV_ELAPSED (btv, atv);
1582 sgen_client_binary_protocol_mark_end (GENERATION_NURSERY);
1584 if (objects_pinned) {
1585 sgen_optimize_pin_queue ();
1586 sgen_pinning_setup_section (nursery_section);
1589 /* walk the pin_queue, build up the fragment list of free memory, unmark
1590 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1593 sgen_client_binary_protocol_reclaim_start (GENERATION_NURSERY);
1594 fragment_total = sgen_build_nursery_fragments (nursery_section, unpin_queue);
1595 if (!fragment_total)
1598 /* Clear TLABs for all threads */
1599 sgen_clear_tlabs ();
1601 sgen_client_binary_protocol_reclaim_end (GENERATION_NURSERY);
1603 time_minor_fragment_creation += TV_ELAPSED (atv, btv);
1604 SGEN_LOG (2, "Fragment creation: %ld usecs, %lu bytes available", TV_ELAPSED (atv, btv), (unsigned long)fragment_total);
1606 if (consistency_check_at_minor_collection)
1607 sgen_check_major_refs ();
1609 major_collector.finish_nursery_collection ();
1611 TV_GETTIME (last_minor_collection_end_tv);
1612 gc_stats.minor_gc_time += TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv);
1614 sgen_debug_dump_heap ("minor", gc_stats.minor_gc_count - 1, NULL);
1616 /* prepare the pin queue for the next collection */
1617 sgen_finish_pinning ();
1618 if (sgen_have_pending_finalizers ()) {
1619 SGEN_LOG (4, "Finalizer-thread wakeup");
1620 sgen_client_finalize_notify ();
1622 sgen_pin_stats_reset ();
1623 /* clear cemented hash */
1624 sgen_cement_clear_below_threshold ();
1626 g_assert (sgen_gray_object_queue_is_empty (&gray_queue));
1628 remset.finish_minor_collection ();
1630 check_scan_starts ();
1632 binary_protocol_flush_buffers (FALSE);
1634 sgen_memgov_minor_collection_end ();
1636 /*objects are late pinned because of lack of memory, so a major is a good call*/
1637 needs_major = objects_pinned > 0;
1638 current_collection_generation = -1;
1641 binary_protocol_collection_end (gc_stats.minor_gc_count - 1, GENERATION_NURSERY, 0, 0);
1643 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1644 sgen_check_nursery_objects_pinned (unpin_queue != NULL);
1650 COPY_OR_MARK_FROM_ROOTS_SERIAL,
1651 COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT,
1652 COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT
1653 } CopyOrMarkFromRootsMode;
1656 major_copy_or_mark_from_roots (size_t *old_next_pin_slot, CopyOrMarkFromRootsMode mode, SgenObjectOperations *object_ops)
1661 /* FIXME: only use these values for the precise scan
1662 * note that to_space pointers should be excluded anyway...
1664 char *heap_start = NULL;
1665 char *heap_end = (char*)-1;
1666 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, WORKERS_DISTRIBUTE_GRAY_QUEUE);
1667 gboolean concurrent = mode != COPY_OR_MARK_FROM_ROOTS_SERIAL;
1669 SGEN_ASSERT (0, !!concurrent == !!concurrent_collection_in_progress, "We've been called with the wrong mode.");
1671 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1672 /*This cleans up unused fragments */
1673 sgen_nursery_allocator_prepare_for_pinning ();
1675 if (do_concurrent_checks)
1676 sgen_debug_check_nursery_is_clean ();
1678 /* The concurrent collector doesn't touch the nursery. */
1679 sgen_nursery_alloc_prepare_for_major ();
1682 init_gray_queue (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1686 /* Pinning depends on this */
1687 sgen_clear_nursery_fragments ();
1689 if (whole_heap_check_before_collection)
1690 sgen_check_whole_heap (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT);
1693 time_major_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1695 if (!sgen_collection_is_concurrent ())
1696 nursery_section->next_data = sgen_get_nursery_end ();
1697 /* we should also coalesce scanning from sections close to each other
1698 * and deal with pointers outside of the sections later.
1703 sgen_client_pre_collection_checks ();
1706 /* Remsets are not useful for a major collection */
1707 remset.clear_cards ();
1710 sgen_process_fin_stage_entries ();
1713 sgen_init_pinning ();
1714 SGEN_LOG (6, "Collecting pinned addresses");
1715 pin_from_roots ((void*)lowest_heap_address, (void*)highest_heap_address, ctx);
1717 sgen_optimize_pin_queue ();
1719 sgen_client_collecting_major_1 ();
1722 * pin_queue now contains all candidate pointers, sorted and
1723 * uniqued. We must do two passes now to figure out which
1724 * objects are pinned.
1726 * The first is to find within the pin_queue the area for each
1727 * section. This requires that the pin_queue be sorted. We
1728 * also process the LOS objects and pinned chunks here.
1730 * The second, destructive, pass is to reduce the section
1731 * areas to pointers to the actually pinned objects.
1733 SGEN_LOG (6, "Pinning from sections");
1734 /* first pass for the sections */
1735 sgen_find_section_pin_queue_start_end (nursery_section);
1736 /* identify possible pointers to the insize of large objects */
1737 SGEN_LOG (6, "Pinning from large objects");
1738 for (bigobj = los_object_list; bigobj; bigobj = bigobj->next) {
1740 if (sgen_find_optimized_pin_queue_area ((char*)bigobj->data, (char*)bigobj->data + sgen_los_object_size (bigobj), &dummy, &dummy)) {
1741 binary_protocol_pin (bigobj->data, (gpointer)LOAD_VTABLE (bigobj->data), safe_object_get_size (bigobj->data));
1743 if (sgen_los_object_is_pinned (bigobj->data)) {
1744 SGEN_ASSERT (0, mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, "LOS objects can only be pinned here after concurrent marking.");
1747 sgen_los_pin_object (bigobj->data);
1748 if (SGEN_OBJECT_HAS_REFERENCES (bigobj->data))
1749 GRAY_OBJECT_ENQUEUE (WORKERS_DISTRIBUTE_GRAY_QUEUE, bigobj->data, sgen_obj_get_descriptor ((GCObject*)bigobj->data));
1750 sgen_pin_stats_register_object (bigobj->data, safe_object_get_size (bigobj->data));
1751 SGEN_LOG (6, "Marked large object %p (%s) size: %lu from roots", bigobj->data,
1752 sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (bigobj->data)),
1753 (unsigned long)sgen_los_object_size (bigobj));
1755 sgen_client_pinned_los_object (bigobj->data);
1759 pin_objects_in_nursery (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT, ctx);
1760 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1761 sgen_check_nursery_objects_pinned (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1763 major_collector.pin_objects (WORKERS_DISTRIBUTE_GRAY_QUEUE);
1764 if (old_next_pin_slot)
1765 *old_next_pin_slot = sgen_get_pinned_count ();
1768 time_major_pinning += TV_ELAPSED (atv, btv);
1769 SGEN_LOG (2, "Finding pinned pointers: %zd in %ld usecs", sgen_get_pinned_count (), TV_ELAPSED (atv, btv));
1770 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1772 major_collector.init_to_space ();
1774 SGEN_ASSERT (0, sgen_workers_all_done (), "Why are the workers not done when we start or finish a major collection?");
1776 * The concurrent collector doesn't move objects, neither on
1777 * the major heap nor in the nursery, so we can mark even
1778 * before pinning has finished. For the non-concurrent
1779 * collector we start the workers after pinning.
1781 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1782 if (precleaning_enabled) {
1784 /* Mod union preclean job */
1785 sj = (ScanJob*)sgen_thread_pool_job_alloc ("preclean mod union cardtable", job_mod_union_preclean, sizeof (ScanJob));
1786 sj->ops = object_ops;
1787 sgen_workers_start_all_workers (object_ops, &sj->job);
1789 sgen_workers_start_all_workers (object_ops, NULL);
1791 gray_queue_enable_redirect (WORKERS_DISTRIBUTE_GRAY_QUEUE);
1792 } else if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1793 if (sgen_workers_have_idle_work ()) {
1794 sgen_workers_start_all_workers (object_ops, NULL);
1795 sgen_workers_join ();
1799 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1800 main_gc_thread = mono_native_thread_self ();
1803 sgen_client_collecting_major_2 ();
1806 time_major_scan_pinned += TV_ELAPSED (btv, atv);
1808 sgen_client_collecting_major_3 (&fin_ready_queue, &critical_fin_queue);
1811 * FIXME: is this the right context? It doesn't seem to contain a copy function
1812 * unless we're concurrent.
1814 enqueue_scan_from_roots_jobs (heap_start, heap_end, object_ops, mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1817 time_major_scan_roots += TV_ELAPSED (atv, btv);
1819 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1822 /* Mod union card table */
1823 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan mod union cardtable", job_scan_major_mod_union_card_table, sizeof (ScanJob));
1824 sj->ops = object_ops;
1825 sgen_workers_enqueue_job (&sj->job, FALSE);
1827 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan LOS mod union cardtable", job_scan_los_mod_union_card_table, sizeof (ScanJob));
1828 sj->ops = object_ops;
1829 sgen_workers_enqueue_job (&sj->job, FALSE);
1832 time_major_scan_mod_union += TV_ELAPSED (btv, atv);
1835 sgen_pin_stats_print_class_stats ();
1839 major_finish_copy_or_mark (CopyOrMarkFromRootsMode mode)
1841 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1843 * Prepare the pin queue for the next collection. Since pinning runs on the worker
1844 * threads we must wait for the jobs to finish before we can reset it.
1846 sgen_workers_wait_for_jobs_finished ();
1847 sgen_finish_pinning ();
1849 sgen_pin_stats_reset ();
1851 if (do_concurrent_checks)
1852 sgen_debug_check_nursery_is_clean ();
1857 major_start_collection (gboolean concurrent, size_t *old_next_pin_slot)
1859 SgenObjectOperations *object_ops;
1861 binary_protocol_collection_begin (gc_stats.major_gc_count, GENERATION_OLD);
1863 current_collection_generation = GENERATION_OLD;
1865 g_assert (sgen_section_gray_queue_is_empty (sgen_workers_get_distribute_section_gray_queue ()));
1868 sgen_cement_reset ();
1871 g_assert (major_collector.is_concurrent);
1872 concurrent_collection_in_progress = TRUE;
1874 object_ops = &major_collector.major_ops_concurrent_start;
1876 object_ops = &major_collector.major_ops_serial;
1879 reset_pinned_from_failed_allocation ();
1881 sgen_memgov_major_collection_start ();
1883 //count_ref_nonref_objs ();
1884 //consistency_check ();
1886 check_scan_starts ();
1889 SGEN_LOG (1, "Start major collection %d", gc_stats.major_gc_count);
1890 gc_stats.major_gc_count ++;
1892 if (major_collector.start_major_collection)
1893 major_collector.start_major_collection ();
1895 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);
1896 major_finish_copy_or_mark (concurrent ? COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT : COPY_OR_MARK_FROM_ROOTS_SERIAL);
1900 major_finish_collection (const char *reason, size_t old_next_pin_slot, gboolean forced)
1902 ScannedObjectCounts counts;
1903 SgenObjectOperations *object_ops;
1904 mword fragment_total;
1910 if (concurrent_collection_in_progress) {
1911 object_ops = &major_collector.major_ops_concurrent_finish;
1913 major_copy_or_mark_from_roots (NULL, COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, object_ops);
1915 major_finish_copy_or_mark (COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT);
1917 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1918 main_gc_thread = NULL;
1921 object_ops = &major_collector.major_ops_serial;
1924 g_assert (sgen_section_gray_queue_is_empty (sgen_workers_get_distribute_section_gray_queue ()));
1926 /* all the objects in the heap */
1927 finish_gray_stack (GENERATION_OLD, CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, &gray_queue));
1929 time_major_finish_gray_stack += TV_ELAPSED (btv, atv);
1931 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after joining");
1933 if (objects_pinned) {
1934 g_assert (!concurrent_collection_in_progress);
1937 * This is slow, but we just OOM'd.
1939 * See comment at `sgen_pin_queue_clear_discarded_entries` for how the pin
1940 * queue is laid out at this point.
1942 sgen_pin_queue_clear_discarded_entries (nursery_section, old_next_pin_slot);
1944 * We need to reestablish all pinned nursery objects in the pin queue
1945 * because they're needed for fragment creation. Unpinning happens by
1946 * walking the whole queue, so it's not necessary to reestablish where major
1947 * heap block pins are - all we care is that they're still in there
1950 sgen_optimize_pin_queue ();
1951 sgen_find_section_pin_queue_start_end (nursery_section);
1955 reset_heap_boundaries ();
1956 sgen_update_heap_boundaries ((mword)sgen_get_nursery_start (), (mword)sgen_get_nursery_end ());
1958 /* walk the pin_queue, build up the fragment list of free memory, unmark
1959 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1962 fragment_total = sgen_build_nursery_fragments (nursery_section, NULL);
1963 if (!fragment_total)
1965 SGEN_LOG (4, "Free space in nursery after major %ld", (long)fragment_total);
1967 if (do_concurrent_checks && concurrent_collection_in_progress)
1968 sgen_debug_check_nursery_is_clean ();
1970 /* prepare the pin queue for the next collection */
1971 sgen_finish_pinning ();
1973 /* Clear TLABs for all threads */
1974 sgen_clear_tlabs ();
1976 sgen_pin_stats_reset ();
1978 sgen_cement_clear_below_threshold ();
1980 if (check_mark_bits_after_major_collection)
1981 sgen_check_heap_marked (concurrent_collection_in_progress);
1984 time_major_fragment_creation += TV_ELAPSED (atv, btv);
1986 binary_protocol_sweep_begin (GENERATION_OLD, !major_collector.sweeps_lazily);
1987 sgen_memgov_major_pre_sweep ();
1990 time_major_free_bigobjs += TV_ELAPSED (btv, atv);
1995 time_major_los_sweep += TV_ELAPSED (atv, btv);
1997 major_collector.sweep ();
1999 binary_protocol_sweep_end (GENERATION_OLD, !major_collector.sweeps_lazily);
2002 time_major_sweep += TV_ELAPSED (btv, atv);
2004 sgen_debug_dump_heap ("major", gc_stats.major_gc_count - 1, reason);
2006 if (sgen_have_pending_finalizers ()) {
2007 SGEN_LOG (4, "Finalizer-thread wakeup");
2008 sgen_client_finalize_notify ();
2011 g_assert (sgen_gray_object_queue_is_empty (&gray_queue));
2013 sgen_memgov_major_collection_end (forced);
2014 current_collection_generation = -1;
2016 memset (&counts, 0, sizeof (ScannedObjectCounts));
2017 major_collector.finish_major_collection (&counts);
2019 g_assert (sgen_section_gray_queue_is_empty (sgen_workers_get_distribute_section_gray_queue ()));
2021 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after major collection has finished");
2022 if (concurrent_collection_in_progress)
2023 concurrent_collection_in_progress = FALSE;
2025 check_scan_starts ();
2027 binary_protocol_flush_buffers (FALSE);
2029 //consistency_check ();
2031 binary_protocol_collection_end (gc_stats.major_gc_count - 1, GENERATION_OLD, counts.num_scanned_objects, counts.num_unique_scanned_objects);
2035 major_do_collection (const char *reason, gboolean forced)
2037 TV_DECLARE (time_start);
2038 TV_DECLARE (time_end);
2039 size_t old_next_pin_slot;
2041 if (disable_major_collections)
2044 if (major_collector.get_and_reset_num_major_objects_marked) {
2045 long long num_marked = major_collector.get_and_reset_num_major_objects_marked ();
2046 g_assert (!num_marked);
2049 /* world must be stopped already */
2050 TV_GETTIME (time_start);
2052 major_start_collection (FALSE, &old_next_pin_slot);
2053 major_finish_collection (reason, old_next_pin_slot, forced);
2055 TV_GETTIME (time_end);
2056 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2058 /* FIXME: also report this to the user, preferably in gc-end. */
2059 if (major_collector.get_and_reset_num_major_objects_marked)
2060 major_collector.get_and_reset_num_major_objects_marked ();
2062 return bytes_pinned_from_failed_allocation > 0;
2066 major_start_concurrent_collection (const char *reason)
2068 TV_DECLARE (time_start);
2069 TV_DECLARE (time_end);
2070 long long num_objects_marked;
2072 if (disable_major_collections)
2075 TV_GETTIME (time_start);
2076 SGEN_TV_GETTIME (time_major_conc_collection_start);
2078 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2079 g_assert (num_objects_marked == 0);
2081 binary_protocol_concurrent_start ();
2083 // FIXME: store reason and pass it when finishing
2084 major_start_collection (TRUE, NULL);
2086 gray_queue_redirect (&gray_queue);
2088 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2090 TV_GETTIME (time_end);
2091 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2093 current_collection_generation = -1;
2097 * Returns whether the major collection has finished.
2100 major_should_finish_concurrent_collection (void)
2102 SGEN_ASSERT (0, sgen_gray_object_queue_is_empty (&gray_queue), "Why is the gray queue not empty before we have started doing anything?");
2103 return sgen_workers_all_done ();
2107 major_update_concurrent_collection (void)
2109 TV_DECLARE (total_start);
2110 TV_DECLARE (total_end);
2112 TV_GETTIME (total_start);
2114 binary_protocol_concurrent_update ();
2116 major_collector.update_cardtable_mod_union ();
2117 sgen_los_update_cardtable_mod_union ();
2119 TV_GETTIME (total_end);
2120 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end);
2124 major_finish_concurrent_collection (gboolean forced)
2126 TV_DECLARE (total_start);
2127 TV_DECLARE (total_end);
2129 TV_GETTIME (total_start);
2131 binary_protocol_concurrent_finish ();
2134 * We need to stop all workers since we're updating the cardtable below.
2135 * The workers will be resumed with a finishing pause context to avoid
2136 * additional cardtable and object scanning.
2138 sgen_workers_stop_all_workers ();
2140 SGEN_TV_GETTIME (time_major_conc_collection_end);
2141 gc_stats.major_gc_time_concurrent += SGEN_TV_ELAPSED (time_major_conc_collection_start, time_major_conc_collection_end);
2143 major_collector.update_cardtable_mod_union ();
2144 sgen_los_update_cardtable_mod_union ();
2146 if (mod_union_consistency_check)
2147 sgen_check_mod_union_consistency ();
2149 current_collection_generation = GENERATION_OLD;
2150 sgen_cement_reset ();
2151 major_finish_collection ("finishing", -1, forced);
2153 if (whole_heap_check_before_collection)
2154 sgen_check_whole_heap (FALSE);
2156 TV_GETTIME (total_end);
2157 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end) - TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv);
2159 current_collection_generation = -1;
2163 * Ensure an allocation request for @size will succeed by freeing enough memory.
2165 * LOCKING: The GC lock MUST be held.
2168 sgen_ensure_free_space (size_t size, int generation)
2170 int generation_to_collect = -1;
2171 const char *reason = NULL;
2173 if (generation == GENERATION_OLD) {
2174 if (sgen_need_major_collection (size)) {
2175 reason = "LOS overflow";
2176 generation_to_collect = GENERATION_OLD;
2179 if (degraded_mode) {
2180 if (sgen_need_major_collection (size)) {
2181 reason = "Degraded mode overflow";
2182 generation_to_collect = GENERATION_OLD;
2184 } else if (sgen_need_major_collection (size)) {
2185 reason = concurrent_collection_in_progress ? "Forced finish concurrent collection" : "Minor allowance";
2186 generation_to_collect = GENERATION_OLD;
2188 generation_to_collect = GENERATION_NURSERY;
2189 reason = "Nursery full";
2193 if (generation_to_collect == -1) {
2194 if (concurrent_collection_in_progress && sgen_workers_all_done ()) {
2195 generation_to_collect = GENERATION_OLD;
2196 reason = "Finish concurrent collection";
2200 if (generation_to_collect == -1)
2202 sgen_perform_collection (size, generation_to_collect, reason, FALSE);
2206 * LOCKING: Assumes the GC lock is held.
2209 sgen_perform_collection (size_t requested_size, int generation_to_collect, const char *reason, gboolean wait_to_finish)
2211 TV_DECLARE (gc_start);
2212 TV_DECLARE (gc_end);
2213 TV_DECLARE (gc_total_start);
2214 TV_DECLARE (gc_total_end);
2215 GGTimingInfo infos [2];
2216 int overflow_generation_to_collect = -1;
2217 int oldest_generation_collected = generation_to_collect;
2218 const char *overflow_reason = NULL;
2220 binary_protocol_collection_requested (generation_to_collect, requested_size, wait_to_finish ? 1 : 0);
2222 SGEN_ASSERT (0, generation_to_collect == GENERATION_NURSERY || generation_to_collect == GENERATION_OLD, "What generation is this?");
2224 TV_GETTIME (gc_start);
2226 sgen_stop_world (generation_to_collect);
2228 TV_GETTIME (gc_total_start);
2230 if (concurrent_collection_in_progress) {
2232 * If the concurrent worker is finished or we are asked to do a major collection
2233 * then we finish the concurrent collection.
2235 gboolean finish = major_should_finish_concurrent_collection () || generation_to_collect == GENERATION_OLD;
2238 major_finish_concurrent_collection (wait_to_finish);
2239 oldest_generation_collected = GENERATION_OLD;
2241 SGEN_ASSERT (0, generation_to_collect == GENERATION_NURSERY, "Why aren't we finishing the concurrent collection?");
2242 major_update_concurrent_collection ();
2243 collect_nursery (NULL, FALSE);
2249 SGEN_ASSERT (0, !concurrent_collection_in_progress, "Why did this not get handled above?");
2252 * There's no concurrent collection in progress. Collect the generation we're asked
2253 * to collect. If the major collector is concurrent and we're not forced to wait,
2254 * start a concurrent collection.
2256 // FIXME: extract overflow reason
2257 if (generation_to_collect == GENERATION_NURSERY) {
2258 if (collect_nursery (NULL, FALSE)) {
2259 overflow_generation_to_collect = GENERATION_OLD;
2260 overflow_reason = "Minor overflow";
2263 if (major_collector.is_concurrent && !wait_to_finish) {
2264 collect_nursery (NULL, FALSE);
2265 major_start_concurrent_collection (reason);
2266 // FIXME: set infos[0] properly
2270 if (major_do_collection (reason, wait_to_finish)) {
2271 overflow_generation_to_collect = GENERATION_NURSERY;
2272 overflow_reason = "Excessive pinning";
2276 TV_GETTIME (gc_end);
2278 memset (infos, 0, sizeof (infos));
2279 infos [0].generation = generation_to_collect;
2280 infos [0].reason = reason;
2281 infos [0].is_overflow = FALSE;
2282 infos [1].generation = -1;
2283 infos [0].total_time = SGEN_TV_ELAPSED (gc_start, gc_end);
2285 SGEN_ASSERT (0, !concurrent_collection_in_progress, "Why did this not get handled above?");
2287 if (overflow_generation_to_collect != -1) {
2289 * We need to do an overflow collection, either because we ran out of memory
2290 * or the nursery is fully pinned.
2293 infos [1].generation = overflow_generation_to_collect;
2294 infos [1].reason = overflow_reason;
2295 infos [1].is_overflow = TRUE;
2298 if (overflow_generation_to_collect == GENERATION_NURSERY)
2299 collect_nursery (NULL, FALSE);
2301 major_do_collection (overflow_reason, wait_to_finish);
2303 TV_GETTIME (gc_end);
2304 infos [1].total_time = SGEN_TV_ELAPSED (gc_start, gc_end);
2306 oldest_generation_collected = MAX (oldest_generation_collected, overflow_generation_to_collect);
2309 SGEN_LOG (2, "Heap size: %lu, LOS size: %lu", (unsigned long)sgen_gc_get_total_heap_allocation (), (unsigned long)los_memory_usage);
2311 /* this also sets the proper pointers for the next allocation */
2312 if (generation_to_collect == GENERATION_NURSERY && !sgen_can_alloc_size (requested_size)) {
2313 /* TypeBuilder and MonoMethod are killing mcs with fragmentation */
2314 SGEN_LOG (1, "nursery collection didn't find enough room for %zd alloc (%zd pinned)", requested_size, sgen_get_pinned_count ());
2315 sgen_dump_pin_queue ();
2320 g_assert (sgen_gray_object_queue_is_empty (&gray_queue));
2322 TV_GETTIME (gc_total_end);
2323 time_max = MAX (time_max, TV_ELAPSED (gc_total_start, gc_total_end));
2325 sgen_restart_world (oldest_generation_collected, infos);
2329 * ######################################################################
2330 * ######## Memory allocation from the OS
2331 * ######################################################################
2332 * This section of code deals with getting memory from the OS and
2333 * allocating memory for GC-internal data structures.
2334 * Internal memory can be handled with a freelist for small objects.
2340 G_GNUC_UNUSED static void
2341 report_internal_mem_usage (void)
2343 printf ("Internal memory usage:\n");
2344 sgen_report_internal_mem_usage ();
2345 printf ("Pinned memory usage:\n");
2346 major_collector.report_pinned_memory_usage ();
2350 * ######################################################################
2351 * ######## Finalization support
2352 * ######################################################################
2356 * If the object has been forwarded it means it's still referenced from a root.
2357 * If it is pinned it's still alive as well.
2358 * A LOS object is only alive if we have pinned it.
2359 * Return TRUE if @obj is ready to be finalized.
2361 static inline gboolean
2362 sgen_is_object_alive (GCObject *object)
2364 if (ptr_in_nursery (object))
2365 return sgen_nursery_is_object_alive (object);
2367 return sgen_major_is_object_alive (object);
2371 * This function returns true if @object is either alive and belongs to the
2372 * current collection - major collections are full heap, so old gen objects
2373 * are never alive during a minor collection.
2376 sgen_is_object_alive_and_on_current_collection (GCObject *object)
2378 if (ptr_in_nursery (object))
2379 return sgen_nursery_is_object_alive (object);
2381 if (current_collection_generation == GENERATION_NURSERY)
2384 return sgen_major_is_object_alive (object);
2389 sgen_gc_is_object_ready_for_finalization (GCObject *object)
2391 return !sgen_is_object_alive (object);
2395 sgen_queue_finalization_entry (GCObject *obj)
2397 gboolean critical = sgen_client_object_has_critical_finalizer (obj);
2399 sgen_pointer_queue_add (critical ? &critical_fin_queue : &fin_ready_queue, obj);
2401 sgen_client_object_queued_for_finalization (obj);
2405 sgen_object_is_live (GCObject *obj)
2407 return sgen_is_object_alive_and_on_current_collection (obj);
2411 * `System.GC.WaitForPendingFinalizers` first checks `sgen_have_pending_finalizers()` to
2412 * determine whether it can exit quickly. The latter must therefore only return FALSE if
2413 * all finalizers have really finished running.
2415 * `sgen_gc_invoke_finalizers()` first dequeues a finalizable object, and then finalizes it.
2416 * This means that just checking whether the queues are empty leaves the possibility that an
2417 * object might have been dequeued but not yet finalized. That's why we need the additional
2418 * flag `pending_unqueued_finalizer`.
2421 static volatile gboolean pending_unqueued_finalizer = FALSE;
2424 sgen_gc_invoke_finalizers (void)
2428 g_assert (!pending_unqueued_finalizer);
2430 /* FIXME: batch to reduce lock contention */
2431 while (sgen_have_pending_finalizers ()) {
2437 * We need to set `pending_unqueued_finalizer` before dequeing the
2438 * finalizable object.
2440 if (!sgen_pointer_queue_is_empty (&fin_ready_queue)) {
2441 pending_unqueued_finalizer = TRUE;
2442 mono_memory_write_barrier ();
2443 obj = (GCObject *)sgen_pointer_queue_pop (&fin_ready_queue);
2444 } else if (!sgen_pointer_queue_is_empty (&critical_fin_queue)) {
2445 pending_unqueued_finalizer = TRUE;
2446 mono_memory_write_barrier ();
2447 obj = (GCObject *)sgen_pointer_queue_pop (&critical_fin_queue);
2453 SGEN_LOG (7, "Finalizing object %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
2461 /* the object is on the stack so it is pinned */
2462 /*g_print ("Calling finalizer for object: %p (%s)\n", obj, sgen_client_object_safe_name (obj));*/
2463 sgen_client_run_finalize (obj);
2466 if (pending_unqueued_finalizer) {
2467 mono_memory_write_barrier ();
2468 pending_unqueued_finalizer = FALSE;
2475 sgen_have_pending_finalizers (void)
2477 return pending_unqueued_finalizer || !sgen_pointer_queue_is_empty (&fin_ready_queue) || !sgen_pointer_queue_is_empty (&critical_fin_queue);
2481 * ######################################################################
2482 * ######## registered roots support
2483 * ######################################################################
2487 * We do not coalesce roots.
2490 sgen_register_root (char *start, size_t size, SgenDescriptor descr, int root_type, int source, const char *msg)
2492 RootRecord new_root;
2495 for (i = 0; i < ROOT_TYPE_NUM; ++i) {
2496 RootRecord *root = (RootRecord *)sgen_hash_table_lookup (&roots_hash [i], start);
2497 /* we allow changing the size and the descriptor (for thread statics etc) */
2499 size_t old_size = root->end_root - start;
2500 root->end_root = start + size;
2501 SGEN_ASSERT (0, !!root->root_desc == !!descr, "Can't change whether a root is precise or conservative.");
2502 SGEN_ASSERT (0, root->source == source, "Can't change a root's source identifier.");
2503 SGEN_ASSERT (0, !!root->msg == !!msg, "Can't change a root's message.");
2504 root->root_desc = descr;
2506 roots_size -= old_size;
2512 new_root.end_root = start + size;
2513 new_root.root_desc = descr;
2514 new_root.source = source;
2517 sgen_hash_table_replace (&roots_hash [root_type], start, &new_root, NULL);
2520 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);
2527 sgen_deregister_root (char* addr)
2533 for (root_type = 0; root_type < ROOT_TYPE_NUM; ++root_type) {
2534 if (sgen_hash_table_remove (&roots_hash [root_type], addr, &root))
2535 roots_size -= (root.end_root - addr);
2541 * ######################################################################
2542 * ######## Thread handling (stop/start code)
2543 * ######################################################################
2547 sgen_get_current_collection_generation (void)
2549 return current_collection_generation;
2553 sgen_thread_register (SgenThreadInfo* info, void *stack_bottom_fallback)
2555 #ifndef HAVE_KW_THREAD
2556 info->tlab_start = info->tlab_next = info->tlab_temp_end = info->tlab_real_end = NULL;
2559 sgen_init_tlab_info (info);
2561 sgen_client_thread_register (info, stack_bottom_fallback);
2567 sgen_thread_unregister (SgenThreadInfo *p)
2569 sgen_client_thread_unregister (p);
2573 * ######################################################################
2574 * ######## Write barriers
2575 * ######################################################################
2579 * Note: the write barriers first do the needed GC work and then do the actual store:
2580 * this way the value is visible to the conservative GC scan after the write barrier
2581 * itself. If a GC interrupts the barrier in the middle, value will be kept alive by
2582 * the conservative scan, otherwise by the remembered set scan.
2586 mono_gc_wbarrier_arrayref_copy (gpointer dest_ptr, gpointer src_ptr, int count)
2588 HEAVY_STAT (++stat_wbarrier_arrayref_copy);
2589 /*This check can be done without taking a lock since dest_ptr array is pinned*/
2590 if (ptr_in_nursery (dest_ptr) || count <= 0) {
2591 mono_gc_memmove_aligned (dest_ptr, src_ptr, count * sizeof (gpointer));
2595 #ifdef SGEN_HEAVY_BINARY_PROTOCOL
2596 if (binary_protocol_is_heavy_enabled ()) {
2598 for (i = 0; i < count; ++i) {
2599 gpointer dest = (gpointer*)dest_ptr + i;
2600 gpointer obj = *((gpointer*)src_ptr + i);
2602 binary_protocol_wbarrier (dest, obj, (gpointer)LOAD_VTABLE (obj));
2607 remset.wbarrier_arrayref_copy (dest_ptr, src_ptr, count);
2611 mono_gc_wbarrier_generic_nostore (gpointer ptr)
2615 HEAVY_STAT (++stat_wbarrier_generic_store);
2617 sgen_client_wbarrier_generic_nostore_check (ptr);
2619 obj = *(gpointer*)ptr;
2621 binary_protocol_wbarrier (ptr, obj, (gpointer)LOAD_VTABLE (obj));
2624 * We need to record old->old pointer locations for the
2625 * concurrent collector.
2627 if (!ptr_in_nursery (obj) && !concurrent_collection_in_progress) {
2628 SGEN_LOG (8, "Skipping remset at %p", ptr);
2632 SGEN_LOG (8, "Adding remset at %p", ptr);
2634 remset.wbarrier_generic_nostore (ptr);
2638 mono_gc_wbarrier_generic_store (gpointer ptr, GCObject* value)
2640 SGEN_LOG (8, "Wbarrier store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2641 SGEN_UPDATE_REFERENCE_ALLOW_NULL (ptr, value);
2642 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2643 mono_gc_wbarrier_generic_nostore (ptr);
2644 sgen_dummy_use (value);
2647 /* Same as mono_gc_wbarrier_generic_store () but performs the store
2648 * as an atomic operation with release semantics.
2651 mono_gc_wbarrier_generic_store_atomic (gpointer ptr, GCObject *value)
2653 HEAVY_STAT (++stat_wbarrier_generic_store_atomic);
2655 SGEN_LOG (8, "Wbarrier atomic store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2657 InterlockedWritePointer ((volatile gpointer *)ptr, value);
2659 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2660 mono_gc_wbarrier_generic_nostore (ptr);
2662 sgen_dummy_use (value);
2666 sgen_wbarrier_value_copy_bitmap (gpointer _dest, gpointer _src, int size, unsigned bitmap)
2668 GCObject **dest = (GCObject **)_dest;
2669 GCObject **src = (GCObject **)_src;
2673 mono_gc_wbarrier_generic_store (dest, *src);
2678 size -= SIZEOF_VOID_P;
2684 * ######################################################################
2685 * ######## Other mono public interface functions.
2686 * ######################################################################
2690 sgen_gc_collect (int generation)
2695 sgen_perform_collection (0, generation, "user request", TRUE);
2700 sgen_gc_collection_count (int generation)
2702 if (generation == 0)
2703 return gc_stats.minor_gc_count;
2704 return gc_stats.major_gc_count;
2708 sgen_gc_get_used_size (void)
2712 tot = los_memory_usage;
2713 tot += nursery_section->next_data - nursery_section->data;
2714 tot += major_collector.get_used_size ();
2715 /* FIXME: account for pinned objects */
2721 sgen_env_var_error (const char *env_var, const char *fallback, const char *description_format, ...)
2725 va_start (ap, description_format);
2727 fprintf (stderr, "Warning: In environment variable `%s': ", env_var);
2728 vfprintf (stderr, description_format, ap);
2730 fprintf (stderr, " - %s", fallback);
2731 fprintf (stderr, "\n");
2737 parse_double_in_interval (const char *env_var, const char *opt_name, const char *opt, double min, double max, double *result)
2740 double val = strtod (opt, &endptr);
2741 if (endptr == opt) {
2742 sgen_env_var_error (env_var, "Using default value.", "`%s` must be a number.", opt_name);
2745 else if (val < min || val > max) {
2746 sgen_env_var_error (env_var, "Using default value.", "`%s` must be between %.2f - %.2f.", opt_name, min, max);
2758 char *major_collector_opt = NULL;
2759 char *minor_collector_opt = NULL;
2760 size_t max_heap = 0;
2761 size_t soft_limit = 0;
2763 gboolean debug_print_allowance = FALSE;
2764 double allowance_ratio = 0, save_target = 0;
2765 gboolean cement_enabled = TRUE;
2768 result = InterlockedCompareExchange (&gc_initialized, -1, 0);
2771 /* already inited */
2774 /* being inited by another thread */
2775 mono_thread_info_usleep (1000);
2778 /* we will init it */
2781 g_assert_not_reached ();
2783 } while (result != 0);
2785 SGEN_TV_GETTIME (sgen_init_timestamp);
2787 #ifdef SGEN_WITHOUT_MONO
2788 mono_thread_smr_init ();
2791 mono_coop_mutex_init (&gc_mutex);
2793 gc_debug_file = stderr;
2795 mono_coop_mutex_init (&sgen_interruption_mutex);
2797 if ((env = g_getenv (MONO_GC_PARAMS_NAME))) {
2798 opts = g_strsplit (env, ",", -1);
2799 for (ptr = opts; *ptr; ++ptr) {
2801 if (g_str_has_prefix (opt, "major=")) {
2802 opt = strchr (opt, '=') + 1;
2803 major_collector_opt = g_strdup (opt);
2804 } else if (g_str_has_prefix (opt, "minor=")) {
2805 opt = strchr (opt, '=') + 1;
2806 minor_collector_opt = g_strdup (opt);
2814 sgen_init_internal_allocator ();
2815 sgen_init_nursery_allocator ();
2816 sgen_init_fin_weak_hash ();
2817 sgen_init_hash_table ();
2818 sgen_init_descriptors ();
2819 sgen_init_gray_queues ();
2820 sgen_init_allocator ();
2821 sgen_init_gchandles ();
2823 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_SECTION, SGEN_SIZEOF_GC_MEM_SECTION);
2824 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_GRAY_QUEUE, sizeof (GrayQueueSection));
2826 sgen_client_init ();
2828 if (!minor_collector_opt) {
2829 sgen_simple_nursery_init (&sgen_minor_collector);
2831 if (!strcmp (minor_collector_opt, "simple")) {
2833 sgen_simple_nursery_init (&sgen_minor_collector);
2834 } else if (!strcmp (minor_collector_opt, "split")) {
2835 sgen_split_nursery_init (&sgen_minor_collector);
2837 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `simple` instead.", "Unknown minor collector `%s'.", minor_collector_opt);
2838 goto use_simple_nursery;
2842 if (!major_collector_opt || !strcmp (major_collector_opt, "marksweep")) {
2843 use_marksweep_major:
2844 sgen_marksweep_init (&major_collector);
2845 } else if (!major_collector_opt || !strcmp (major_collector_opt, "marksweep-conc")) {
2846 sgen_marksweep_conc_init (&major_collector);
2848 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `marksweep` instead.", "Unknown major collector `%s'.", major_collector_opt);
2849 goto use_marksweep_major;
2852 sgen_nursery_size = DEFAULT_NURSERY_SIZE;
2855 gboolean usage_printed = FALSE;
2857 for (ptr = opts; *ptr; ++ptr) {
2859 if (!strcmp (opt, ""))
2861 if (g_str_has_prefix (opt, "major="))
2863 if (g_str_has_prefix (opt, "minor="))
2865 if (g_str_has_prefix (opt, "max-heap-size=")) {
2866 size_t page_size = mono_pagesize ();
2867 size_t max_heap_candidate = 0;
2868 opt = strchr (opt, '=') + 1;
2869 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &max_heap_candidate)) {
2870 max_heap = (max_heap_candidate + page_size - 1) & ~(size_t)(page_size - 1);
2871 if (max_heap != max_heap_candidate)
2872 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Rounding up.", "`max-heap-size` size must be a multiple of %d.", page_size);
2874 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`max-heap-size` must be an integer.");
2878 if (g_str_has_prefix (opt, "soft-heap-limit=")) {
2879 opt = strchr (opt, '=') + 1;
2880 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &soft_limit)) {
2881 if (soft_limit <= 0) {
2882 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be positive.");
2886 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be an integer.");
2892 if (g_str_has_prefix (opt, "nursery-size=")) {
2894 opt = strchr (opt, '=') + 1;
2895 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &val)) {
2896 if ((val & (val - 1))) {
2897 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be a power of two.");
2901 if (val < SGEN_MAX_NURSERY_WASTE) {
2902 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.",
2903 "`nursery-size` must be at least %d bytes.", SGEN_MAX_NURSERY_WASTE);
2907 sgen_nursery_size = val;
2908 sgen_nursery_bits = 0;
2909 while (ONE_P << (++ sgen_nursery_bits) != sgen_nursery_size)
2912 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be an integer.");
2918 if (g_str_has_prefix (opt, "save-target-ratio=")) {
2920 opt = strchr (opt, '=') + 1;
2921 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "save-target-ratio", opt,
2922 SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO, &val)) {
2927 if (g_str_has_prefix (opt, "default-allowance-ratio=")) {
2929 opt = strchr (opt, '=') + 1;
2930 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "default-allowance-ratio", opt,
2931 SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, &val)) {
2932 allowance_ratio = val;
2937 if (!strcmp (opt, "cementing")) {
2938 cement_enabled = TRUE;
2941 if (!strcmp (opt, "no-cementing")) {
2942 cement_enabled = FALSE;
2946 if (!strcmp (opt, "precleaning")) {
2947 precleaning_enabled = TRUE;
2950 if (!strcmp (opt, "no-precleaning")) {
2951 precleaning_enabled = FALSE;
2955 if (major_collector.handle_gc_param && major_collector.handle_gc_param (opt))
2958 if (sgen_minor_collector.handle_gc_param && sgen_minor_collector.handle_gc_param (opt))
2961 if (sgen_client_handle_gc_param (opt))
2964 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Ignoring.", "Unknown option `%s`.", opt);
2969 fprintf (stderr, "\n%s must be a comma-delimited list of one or more of the following:\n", MONO_GC_PARAMS_NAME);
2970 fprintf (stderr, " max-heap-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
2971 fprintf (stderr, " soft-heap-limit=n (where N is an integer, possibly with a k, m or a g suffix)\n");
2972 fprintf (stderr, " nursery-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
2973 fprintf (stderr, " major=COLLECTOR (where COLLECTOR is `marksweep', `marksweep-conc', `marksweep-par')\n");
2974 fprintf (stderr, " minor=COLLECTOR (where COLLECTOR is `simple' or `split')\n");
2975 fprintf (stderr, " wbarrier=WBARRIER (where WBARRIER is `remset' or `cardtable')\n");
2976 fprintf (stderr, " [no-]cementing\n");
2977 if (major_collector.print_gc_param_usage)
2978 major_collector.print_gc_param_usage ();
2979 if (sgen_minor_collector.print_gc_param_usage)
2980 sgen_minor_collector.print_gc_param_usage ();
2981 sgen_client_print_gc_params_usage ();
2982 fprintf (stderr, " Experimental options:\n");
2983 fprintf (stderr, " save-target-ratio=R (where R must be between %.2f - %.2f).\n", SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO);
2984 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);
2985 fprintf (stderr, "\n");
2987 usage_printed = TRUE;
2992 if (major_collector_opt)
2993 g_free (major_collector_opt);
2995 if (minor_collector_opt)
2996 g_free (minor_collector_opt);
3000 sgen_cement_init (cement_enabled);
3002 if ((env = g_getenv (MONO_GC_DEBUG_NAME))) {
3003 gboolean usage_printed = FALSE;
3005 opts = g_strsplit (env, ",", -1);
3006 for (ptr = opts; ptr && *ptr; ptr ++) {
3008 if (!strcmp (opt, ""))
3010 if (opt [0] >= '0' && opt [0] <= '9') {
3011 gc_debug_level = atoi (opt);
3016 char *rf = g_strdup_printf ("%s.%d", opt, mono_process_current_pid ());
3017 gc_debug_file = fopen (rf, "wb");
3019 gc_debug_file = stderr;
3022 } else if (!strcmp (opt, "print-allowance")) {
3023 debug_print_allowance = TRUE;
3024 } else if (!strcmp (opt, "print-pinning")) {
3025 sgen_pin_stats_enable ();
3026 } else if (!strcmp (opt, "verify-before-allocs")) {
3027 verify_before_allocs = 1;
3028 has_per_allocation_action = TRUE;
3029 } else if (g_str_has_prefix (opt, "verify-before-allocs=")) {
3030 char *arg = strchr (opt, '=') + 1;
3031 verify_before_allocs = atoi (arg);
3032 has_per_allocation_action = TRUE;
3033 } else if (!strcmp (opt, "collect-before-allocs")) {
3034 collect_before_allocs = 1;
3035 has_per_allocation_action = TRUE;
3036 } else if (g_str_has_prefix (opt, "collect-before-allocs=")) {
3037 char *arg = strchr (opt, '=') + 1;
3038 has_per_allocation_action = TRUE;
3039 collect_before_allocs = atoi (arg);
3040 } else if (!strcmp (opt, "verify-before-collections")) {
3041 whole_heap_check_before_collection = TRUE;
3042 } else if (!strcmp (opt, "check-at-minor-collections")) {
3043 consistency_check_at_minor_collection = TRUE;
3044 nursery_clear_policy = CLEAR_AT_GC;
3045 } else if (!strcmp (opt, "mod-union-consistency-check")) {
3046 if (!major_collector.is_concurrent) {
3047 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`mod-union-consistency-check` only works with concurrent major collector.");
3050 mod_union_consistency_check = TRUE;
3051 } else if (!strcmp (opt, "check-mark-bits")) {
3052 check_mark_bits_after_major_collection = TRUE;
3053 } else if (!strcmp (opt, "check-nursery-pinned")) {
3054 check_nursery_objects_pinned = TRUE;
3055 } else if (!strcmp (opt, "clear-at-gc")) {
3056 nursery_clear_policy = CLEAR_AT_GC;
3057 } else if (!strcmp (opt, "clear-nursery-at-gc")) {
3058 nursery_clear_policy = CLEAR_AT_GC;
3059 } else if (!strcmp (opt, "clear-at-tlab-creation")) {
3060 nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
3061 } else if (!strcmp (opt, "debug-clear-at-tlab-creation")) {
3062 nursery_clear_policy = CLEAR_AT_TLAB_CREATION_DEBUG;
3063 } else if (!strcmp (opt, "check-scan-starts")) {
3064 do_scan_starts_check = TRUE;
3065 } else if (!strcmp (opt, "verify-nursery-at-minor-gc")) {
3066 do_verify_nursery = TRUE;
3067 } else if (!strcmp (opt, "check-concurrent")) {
3068 if (!major_collector.is_concurrent) {
3069 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`check-concurrent` only works with concurrent major collectors.");
3072 nursery_clear_policy = CLEAR_AT_GC;
3073 do_concurrent_checks = TRUE;
3074 } else if (!strcmp (opt, "dump-nursery-at-minor-gc")) {
3075 do_dump_nursery_content = TRUE;
3076 } else if (!strcmp (opt, "disable-minor")) {
3077 disable_minor_collections = TRUE;
3078 } else if (!strcmp (opt, "disable-major")) {
3079 disable_major_collections = TRUE;
3080 } else if (g_str_has_prefix (opt, "heap-dump=")) {
3081 char *filename = strchr (opt, '=') + 1;
3082 nursery_clear_policy = CLEAR_AT_GC;
3083 sgen_debug_enable_heap_dump (filename);
3084 } else if (g_str_has_prefix (opt, "binary-protocol=")) {
3085 char *filename = strchr (opt, '=') + 1;
3086 char *colon = strrchr (filename, ':');
3089 if (!mono_gc_parse_environment_string_extract_number (colon + 1, &limit)) {
3090 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring limit.", "Binary protocol file size limit must be an integer.");
3095 binary_protocol_init (filename, (long long)limit);
3096 } else if (!strcmp (opt, "nursery-canaries")) {
3097 do_verify_nursery = TRUE;
3098 enable_nursery_canaries = TRUE;
3099 } else if (!sgen_client_handle_gc_debug (opt)) {
3100 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "Unknown option `%s`.", opt);
3105 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);
3106 fprintf (stderr, "Valid <option>s are:\n");
3107 fprintf (stderr, " collect-before-allocs[=<n>]\n");
3108 fprintf (stderr, " verify-before-allocs[=<n>]\n");
3109 fprintf (stderr, " check-at-minor-collections\n");
3110 fprintf (stderr, " check-mark-bits\n");
3111 fprintf (stderr, " check-nursery-pinned\n");
3112 fprintf (stderr, " verify-before-collections\n");
3113 fprintf (stderr, " verify-nursery-at-minor-gc\n");
3114 fprintf (stderr, " dump-nursery-at-minor-gc\n");
3115 fprintf (stderr, " disable-minor\n");
3116 fprintf (stderr, " disable-major\n");
3117 fprintf (stderr, " check-concurrent\n");
3118 fprintf (stderr, " clear-[nursery-]at-gc\n");
3119 fprintf (stderr, " clear-at-tlab-creation\n");
3120 fprintf (stderr, " debug-clear-at-tlab-creation\n");
3121 fprintf (stderr, " check-scan-starts\n");
3122 fprintf (stderr, " print-allowance\n");
3123 fprintf (stderr, " print-pinning\n");
3124 fprintf (stderr, " heap-dump=<filename>\n");
3125 fprintf (stderr, " binary-protocol=<filename>[:<file-size-limit>]\n");
3126 fprintf (stderr, " nursery-canaries\n");
3127 sgen_client_print_gc_debug_usage ();
3128 fprintf (stderr, "\n");
3130 usage_printed = TRUE;
3136 if (check_mark_bits_after_major_collection)
3137 nursery_clear_policy = CLEAR_AT_GC;
3139 if (major_collector.post_param_init)
3140 major_collector.post_param_init (&major_collector);
3142 if (major_collector.needs_thread_pool)
3143 sgen_workers_init (1);
3145 sgen_memgov_init (max_heap, soft_limit, debug_print_allowance, allowance_ratio, save_target);
3147 memset (&remset, 0, sizeof (remset));
3149 sgen_card_table_init (&remset);
3151 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");
3157 sgen_get_nursery_clear_policy (void)
3159 return nursery_clear_policy;
3165 mono_coop_mutex_lock (&gc_mutex);
3169 sgen_gc_unlock (void)
3171 gboolean try_free = sgen_try_free_some_memory;
3172 sgen_try_free_some_memory = FALSE;
3173 mono_coop_mutex_unlock (&gc_mutex);
3175 mono_thread_hazardous_try_free_some ();
3179 sgen_major_collector_iterate_live_block_ranges (sgen_cardtable_block_callback callback)
3181 major_collector.iterate_live_block_ranges (callback);
3185 sgen_get_major_collector (void)
3187 return &major_collector;
3191 sgen_get_remset (void)
3197 count_cards (long long *major_total, long long *major_marked, long long *los_total, long long *los_marked)
3199 sgen_get_major_collector ()->count_cards (major_total, major_marked);
3200 sgen_los_count_cards (los_total, los_marked);
3203 static gboolean world_is_stopped = FALSE;
3205 /* LOCKING: assumes the GC lock is held */
3207 sgen_stop_world (int generation)
3209 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3211 SGEN_ASSERT (0, !world_is_stopped, "Why are we stopping a stopped world?");
3213 binary_protocol_world_stopping (generation, sgen_timestamp (), (gpointer) (gsize) mono_native_thread_id_get ());
3215 sgen_client_stop_world (generation);
3217 world_is_stopped = TRUE;
3219 if (binary_protocol_is_heavy_enabled ())
3220 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3221 binary_protocol_world_stopped (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3224 /* LOCKING: assumes the GC lock is held */
3226 sgen_restart_world (int generation, GGTimingInfo *timing)
3228 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3230 SGEN_ASSERT (0, world_is_stopped, "Why are we restarting a running world?");
3232 if (binary_protocol_is_heavy_enabled ())
3233 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3234 binary_protocol_world_restarting (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3236 sgen_client_restart_world (generation, timing);
3238 world_is_stopped = FALSE;
3240 binary_protocol_world_restarted (generation, sgen_timestamp ());
3242 sgen_try_free_some_memory = TRUE;
3244 if (sgen_client_bridge_need_processing ())
3245 sgen_client_bridge_processing_finish (generation);
3247 sgen_memgov_collection_end (generation, timing, timing ? 2 : 0);
3251 sgen_is_world_stopped (void)
3253 return world_is_stopped;
3257 sgen_check_whole_heap_stw (void)
3259 sgen_stop_world (0);
3260 sgen_clear_nursery_fragments ();
3261 sgen_check_whole_heap (FALSE);
3262 sgen_restart_world (0, NULL);
3266 sgen_timestamp (void)
3268 SGEN_TV_DECLARE (timestamp);
3269 SGEN_TV_GETTIME (timestamp);
3270 return SGEN_TV_ELAPSED (sgen_init_timestamp, timestamp);
3273 #endif /* HAVE_SGEN_GC */