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 * This library is free software; you can redistribute it and/or
22 * modify it under the terms of the GNU Library General Public
23 * License 2.0 as published by the Free Software Foundation;
25 * This library is distributed in the hope that it will be useful,
26 * but WITHOUT ANY WARRANTY; without even the implied warranty of
27 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
28 * Library General Public License for more details.
30 * You should have received a copy of the GNU Library General Public
31 * License 2.0 along with this library; if not, write to the Free
32 * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
34 * Important: allocation provides always zeroed memory, having to do
35 * a memset after allocation is deadly for performance.
36 * Memory usage at startup is currently as follows:
38 * 64 KB internal space
40 * We should provide a small memory config with half the sizes
42 * We currently try to make as few mono assumptions as possible:
43 * 1) 2-word header with no GC pointers in it (first vtable, second to store the
45 * 2) gc descriptor is the second word in the vtable (first word in the class)
46 * 3) 8 byte alignment is the minimum and enough (not true for special structures (SIMD), FIXME)
47 * 4) there is a function to get an object's size and the number of
48 * elements in an array.
49 * 5) we know the special way bounds are allocated for complex arrays
50 * 6) we know about proxies and how to treat them when domains are unloaded
52 * Always try to keep stack usage to a minimum: no recursive behaviour
53 * and no large stack allocs.
55 * General description.
56 * Objects are initially allocated in a nursery using a fast bump-pointer technique.
57 * When the nursery is full we start a nursery collection: this is performed with a
59 * When the old generation is full we start a copying GC of the old generation as well:
60 * this will be changed to mark&sweep with copying when fragmentation becomes to severe
61 * in the future. Maybe we'll even do both during the same collection like IMMIX.
63 * The things that complicate this description are:
64 * *) pinned objects: we can't move them so we need to keep track of them
65 * *) no precise info of the thread stacks and registers: we need to be able to
66 * quickly find the objects that may be referenced conservatively and pin them
67 * (this makes the first issues more important)
68 * *) large objects are too expensive to be dealt with using copying GC: we handle them
69 * with mark/sweep during major collections
70 * *) some objects need to not move even if they are small (interned strings, Type handles):
71 * we use mark/sweep for them, too: they are not allocated in the nursery, but inside
72 * PinnedChunks regions
78 *) we could have a function pointer in MonoClass to implement
79 customized write barriers for value types
81 *) investigate the stuff needed to advance a thread to a GC-safe
82 point (single-stepping, read from unmapped memory etc) and implement it.
83 This would enable us to inline allocations and write barriers, for example,
84 or at least parts of them, like the write barrier checks.
85 We may need this also for handling precise info on stacks, even simple things
86 as having uninitialized data on the stack and having to wait for the prolog
87 to zero it. Not an issue for the last frame that we scan conservatively.
88 We could always not trust the value in the slots anyway.
90 *) modify the jit to save info about references in stack locations:
91 this can be done just for locals as a start, so that at least
92 part of the stack is handled precisely.
94 *) test/fix endianess issues
96 *) Implement a card table as the write barrier instead of remembered
97 sets? Card tables are not easy to implement with our current
98 memory layout. We have several different kinds of major heap
99 objects: Small objects in regular blocks, small objects in pinned
100 chunks and LOS objects. If we just have a pointer we have no way
101 to tell which kind of object it points into, therefore we cannot
102 know where its card table is. The least we have to do to make
103 this happen is to get rid of write barriers for indirect stores.
106 *) Get rid of write barriers for indirect stores. We can do this by
107 telling the GC to wbarrier-register an object once we do an ldloca
108 or ldelema on it, and to unregister it once it's not used anymore
109 (it can only travel downwards on the stack). The problem with
110 unregistering is that it needs to happen eventually no matter
111 what, even if exceptions are thrown, the thread aborts, etc.
112 Rodrigo suggested that we could do only the registering part and
113 let the collector find out (pessimistically) when it's safe to
114 unregister, namely when the stack pointer of the thread that
115 registered the object is higher than it was when the registering
116 happened. This might make for a good first implementation to get
117 some data on performance.
119 *) Some sort of blacklist support? Blacklists is a concept from the
120 Boehm GC: if during a conservative scan we find pointers to an
121 area which we might use as heap, we mark that area as unusable, so
122 pointer retention by random pinning pointers is reduced.
124 *) experiment with max small object size (very small right now - 2kb,
125 because it's tied to the max freelist size)
127 *) add an option to mmap the whole heap in one chunk: it makes for many
128 simplifications in the checks (put the nursery at the top and just use a single
129 check for inclusion/exclusion): the issue this has is that on 32 bit systems it's
130 not flexible (too much of the address space may be used by default or we can't
131 increase the heap as needed) and we'd need a race-free mechanism to return memory
132 back to the system (mprotect(PROT_NONE) will still keep the memory allocated if it
133 was written to, munmap is needed, but the following mmap may not find the same segment
136 *) memzero the major fragments after restarting the world and optionally a smaller
139 *) investigate having fragment zeroing threads
141 *) separate locks for finalization and other minor stuff to reduce
144 *) try a different copying order to improve memory locality
146 *) a thread abort after a store but before the write barrier will
147 prevent the write barrier from executing
149 *) specialized dynamically generated markers/copiers
151 *) Dynamically adjust TLAB size to the number of threads. If we have
152 too many threads that do allocation, we might need smaller TLABs,
153 and we might get better performance with larger TLABs if we only
154 have a handful of threads. We could sum up the space left in all
155 assigned TLABs and if that's more than some percentage of the
156 nursery size, reduce the TLAB size.
158 *) Explore placing unreachable objects on unused nursery memory.
159 Instead of memset'ng a region to zero, place an int[] covering it.
160 A good place to start is add_nursery_frag. The tricky thing here is
161 placing those objects atomically outside of a collection.
163 *) Allocation should use asymmetric Dekker synchronization:
164 http://blogs.oracle.com/dave/resource/Asymmetric-Dekker-Synchronization.txt
165 This should help weak consistency archs.
172 #define _XOPEN_SOURCE
173 #define _DARWIN_C_SOURCE
179 #ifdef HAVE_PTHREAD_H
182 #ifdef HAVE_PTHREAD_NP_H
183 #include <pthread_np.h>
191 #include "mono/sgen/sgen-gc.h"
192 #include "mono/sgen/sgen-cardtable.h"
193 #include "mono/sgen/sgen-protocol.h"
194 #include "mono/sgen/sgen-memory-governor.h"
195 #include "mono/sgen/sgen-hash-table.h"
196 #include "mono/sgen/sgen-cardtable.h"
197 #include "mono/sgen/sgen-pinning.h"
198 #include "mono/sgen/sgen-workers.h"
199 #include "mono/sgen/sgen-client.h"
200 #include "mono/sgen/sgen-pointer-queue.h"
201 #include "mono/sgen/gc-internal-agnostic.h"
202 #include "mono/utils/mono-proclib.h"
203 #include "mono/utils/mono-memory-model.h"
204 #include "mono/utils/hazard-pointer.h"
206 #include <mono/utils/memcheck.h>
208 #undef pthread_create
210 #undef pthread_detach
213 * ######################################################################
214 * ######## Types and constants used by the GC.
215 * ######################################################################
218 /* 0 means not initialized, 1 is initialized, -1 means in progress */
219 static int gc_initialized = 0;
220 /* If set, check if we need to do something every X allocations */
221 gboolean has_per_allocation_action;
222 /* If set, do a heap check every X allocation */
223 guint32 verify_before_allocs = 0;
224 /* If set, do a minor collection before every X allocation */
225 guint32 collect_before_allocs = 0;
226 /* If set, do a whole heap check before each collection */
227 static gboolean whole_heap_check_before_collection = FALSE;
228 /* If set, do a heap consistency check before each minor collection */
229 static gboolean consistency_check_at_minor_collection = FALSE;
230 /* If set, do a mod union consistency check before each finishing collection pause */
231 static gboolean mod_union_consistency_check = FALSE;
232 /* If set, check whether mark bits are consistent after major collections */
233 static gboolean check_mark_bits_after_major_collection = FALSE;
234 /* If set, check that all nursery objects are pinned/not pinned, depending on context */
235 static gboolean check_nursery_objects_pinned = FALSE;
236 /* If set, do a few checks when the concurrent collector is used */
237 static gboolean do_concurrent_checks = FALSE;
238 /* If set, do a plausibility check on the scan_starts before and after
240 static gboolean do_scan_starts_check = FALSE;
242 static gboolean disable_minor_collections = FALSE;
243 static gboolean disable_major_collections = FALSE;
244 static gboolean do_verify_nursery = FALSE;
245 static gboolean do_dump_nursery_content = FALSE;
246 static gboolean enable_nursery_canaries = FALSE;
248 static gboolean precleaning_enabled = TRUE;
250 #ifdef HEAVY_STATISTICS
251 guint64 stat_objects_alloced_degraded = 0;
252 guint64 stat_bytes_alloced_degraded = 0;
254 guint64 stat_copy_object_called_nursery = 0;
255 guint64 stat_objects_copied_nursery = 0;
256 guint64 stat_copy_object_called_major = 0;
257 guint64 stat_objects_copied_major = 0;
259 guint64 stat_scan_object_called_nursery = 0;
260 guint64 stat_scan_object_called_major = 0;
262 guint64 stat_slots_allocated_in_vain;
264 guint64 stat_nursery_copy_object_failed_from_space = 0;
265 guint64 stat_nursery_copy_object_failed_forwarded = 0;
266 guint64 stat_nursery_copy_object_failed_pinned = 0;
267 guint64 stat_nursery_copy_object_failed_to_space = 0;
269 static guint64 stat_wbarrier_add_to_global_remset = 0;
270 static guint64 stat_wbarrier_arrayref_copy = 0;
271 static guint64 stat_wbarrier_generic_store = 0;
272 static guint64 stat_wbarrier_generic_store_atomic = 0;
273 static guint64 stat_wbarrier_set_root = 0;
276 static guint64 stat_pinned_objects = 0;
278 static guint64 time_minor_pre_collection_fragment_clear = 0;
279 static guint64 time_minor_pinning = 0;
280 static guint64 time_minor_scan_remsets = 0;
281 static guint64 time_minor_scan_pinned = 0;
282 static guint64 time_minor_scan_roots = 0;
283 static guint64 time_minor_finish_gray_stack = 0;
284 static guint64 time_minor_fragment_creation = 0;
286 static guint64 time_major_pre_collection_fragment_clear = 0;
287 static guint64 time_major_pinning = 0;
288 static guint64 time_major_scan_pinned = 0;
289 static guint64 time_major_scan_roots = 0;
290 static guint64 time_major_scan_mod_union = 0;
291 static guint64 time_major_finish_gray_stack = 0;
292 static guint64 time_major_free_bigobjs = 0;
293 static guint64 time_major_los_sweep = 0;
294 static guint64 time_major_sweep = 0;
295 static guint64 time_major_fragment_creation = 0;
297 static guint64 time_max = 0;
299 static SGEN_TV_DECLARE (time_major_conc_collection_start);
300 static SGEN_TV_DECLARE (time_major_conc_collection_end);
302 static SGEN_TV_DECLARE (last_minor_collection_start_tv);
303 static SGEN_TV_DECLARE (last_minor_collection_end_tv);
305 int gc_debug_level = 0;
310 mono_gc_flush_info (void)
312 fflush (gc_debug_file);
316 #define TV_DECLARE SGEN_TV_DECLARE
317 #define TV_GETTIME SGEN_TV_GETTIME
318 #define TV_ELAPSED SGEN_TV_ELAPSED
320 static SGEN_TV_DECLARE (sgen_init_timestamp);
322 NurseryClearPolicy nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
324 #define object_is_forwarded SGEN_OBJECT_IS_FORWARDED
325 #define object_is_pinned SGEN_OBJECT_IS_PINNED
326 #define pin_object SGEN_PIN_OBJECT
328 #define ptr_in_nursery sgen_ptr_in_nursery
330 #define LOAD_VTABLE SGEN_LOAD_VTABLE
333 nursery_canaries_enabled (void)
335 return enable_nursery_canaries;
338 #define safe_object_get_size sgen_safe_object_get_size
341 * ######################################################################
342 * ######## Global data.
343 * ######################################################################
345 MonoCoopMutex gc_mutex;
346 gboolean sgen_try_free_some_memory;
348 #define SCAN_START_SIZE SGEN_SCAN_START_SIZE
350 size_t degraded_mode = 0;
352 static mword bytes_pinned_from_failed_allocation = 0;
354 GCMemSection *nursery_section = NULL;
355 static volatile mword lowest_heap_address = ~(mword)0;
356 static volatile mword highest_heap_address = 0;
358 MonoCoopMutex sgen_interruption_mutex;
360 int current_collection_generation = -1;
361 static volatile gboolean concurrent_collection_in_progress = FALSE;
363 /* objects that are ready to be finalized */
364 static SgenPointerQueue fin_ready_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
365 static SgenPointerQueue critical_fin_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
367 /* registered roots: the key to the hash is the root start address */
369 * Different kinds of roots are kept separate to speed up pin_from_roots () for example.
371 SgenHashTable roots_hash [ROOT_TYPE_NUM] = {
372 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
373 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
374 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL)
376 static mword roots_size = 0; /* amount of memory in the root set */
378 /* The size of a TLAB */
379 /* The bigger the value, the less often we have to go to the slow path to allocate a new
380 * one, but the more space is wasted by threads not allocating much memory.
382 * FIXME: Make this self-tuning for each thread.
384 guint32 tlab_size = (1024 * 4);
386 #define MAX_SMALL_OBJ_SIZE SGEN_MAX_SMALL_OBJ_SIZE
388 #define ALLOC_ALIGN SGEN_ALLOC_ALIGN
390 #define ALIGN_UP SGEN_ALIGN_UP
392 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
393 MonoNativeThreadId main_gc_thread = NULL;
396 /*Object was pinned during the current collection*/
397 static mword objects_pinned;
400 * ######################################################################
401 * ######## Macros and function declarations.
402 * ######################################################################
405 typedef SgenGrayQueue GrayQueue;
407 /* forward declarations */
408 static void scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx);
410 static void pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx);
411 static void finish_gray_stack (int generation, ScanCopyContext ctx);
414 SgenMajorCollector major_collector;
415 SgenMinorCollector sgen_minor_collector;
416 /* FIXME: get rid of this */
417 static GrayQueue gray_queue;
419 static SgenRememberedSet remset;
421 /* The gray queue to use from the main collection thread. */
422 #define WORKERS_DISTRIBUTE_GRAY_QUEUE (&gray_queue)
425 * The gray queue a worker job must use. If we're not parallel or
426 * concurrent, we use the main gray queue.
428 static SgenGrayQueue*
429 sgen_workers_get_job_gray_queue (WorkerData *worker_data)
431 return worker_data ? &worker_data->private_gray_queue : WORKERS_DISTRIBUTE_GRAY_QUEUE;
435 gray_queue_redirect (SgenGrayQueue *queue)
437 gboolean wake = FALSE;
440 GrayQueueSection *section = sgen_gray_object_dequeue_section (queue);
443 sgen_section_gray_queue_enqueue ((SgenSectionGrayQueue *)queue->alloc_prepare_data, section);
448 g_assert (concurrent_collection_in_progress);
449 sgen_workers_ensure_awake ();
454 gray_queue_enable_redirect (SgenGrayQueue *queue)
456 if (!concurrent_collection_in_progress)
459 sgen_gray_queue_set_alloc_prepare (queue, gray_queue_redirect, sgen_workers_get_distribute_section_gray_queue ());
460 gray_queue_redirect (queue);
464 sgen_scan_area_with_callback (char *start, char *end, IterateObjectCallbackFunc callback, void *data, gboolean allow_flags, gboolean fail_on_canaries)
466 while (start < end) {
470 if (!*(void**)start) {
471 start += sizeof (void*); /* should be ALLOC_ALIGN, really */
476 if (!(obj = (char *)SGEN_OBJECT_IS_FORWARDED (start)))
482 if (!sgen_client_object_is_array_fill ((GCObject*)obj)) {
483 CHECK_CANARY_FOR_OBJECT ((GCObject*)obj, fail_on_canaries);
484 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
485 callback ((GCObject*)obj, size, data);
486 CANARIFY_SIZE (size);
488 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
496 * sgen_add_to_global_remset:
498 * The global remset contains locations which point into newspace after
499 * a minor collection. This can happen if the objects they point to are pinned.
501 * LOCKING: If called from a parallel collector, the global remset
502 * lock must be held. For serial collectors that is not necessary.
505 sgen_add_to_global_remset (gpointer ptr, GCObject *obj)
507 SGEN_ASSERT (5, sgen_ptr_in_nursery (obj), "Target pointer of global remset must be in the nursery");
509 HEAVY_STAT (++stat_wbarrier_add_to_global_remset);
511 if (!major_collector.is_concurrent) {
512 SGEN_ASSERT (5, current_collection_generation != -1, "Global remsets can only be added during collections");
514 if (current_collection_generation == -1)
515 SGEN_ASSERT (5, sgen_concurrent_collection_in_progress (), "Global remsets outside of collection pauses can only be added by the concurrent collector");
518 if (!object_is_pinned (obj))
519 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");
520 else if (sgen_cement_lookup_or_register (obj))
523 remset.record_pointer (ptr);
525 sgen_pin_stats_register_global_remset (obj);
527 SGEN_LOG (8, "Adding global remset for %p", ptr);
528 binary_protocol_global_remset (ptr, obj, (gpointer)SGEN_LOAD_VTABLE (obj));
532 * sgen_drain_gray_stack:
534 * Scan objects in the gray stack until the stack is empty. This should be called
535 * frequently after each object is copied, to achieve better locality and cache
540 sgen_drain_gray_stack (ScanCopyContext ctx)
542 ScanObjectFunc scan_func = ctx.ops->scan_object;
543 GrayQueue *queue = ctx.queue;
545 if (ctx.ops->drain_gray_stack)
546 return ctx.ops->drain_gray_stack (queue);
551 GRAY_OBJECT_DEQUEUE (queue, &obj, &desc);
554 SGEN_LOG (9, "Precise gray object scan %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
555 scan_func (obj, desc, queue);
561 * Addresses in the pin queue are already sorted. This function finds
562 * the object header for each address and pins the object. The
563 * addresses must be inside the nursery section. The (start of the)
564 * address array is overwritten with the addresses of the actually
565 * pinned objects. Return the number of pinned objects.
568 pin_objects_from_nursery_pin_queue (gboolean do_scan_objects, ScanCopyContext ctx)
570 GCMemSection *section = nursery_section;
571 void **start = sgen_pinning_get_entry (section->pin_queue_first_entry);
572 void **end = sgen_pinning_get_entry (section->pin_queue_last_entry);
573 void *start_nursery = section->data;
574 void *end_nursery = section->next_data;
579 void *pinning_front = start_nursery;
581 void **definitely_pinned = start;
582 ScanObjectFunc scan_func = ctx.ops->scan_object;
583 SgenGrayQueue *queue = ctx.queue;
585 sgen_nursery_allocator_prepare_for_pinning ();
587 while (start < end) {
588 GCObject *obj_to_pin = NULL;
589 size_t obj_to_pin_size = 0;
594 SGEN_ASSERT (0, addr >= start_nursery && addr < end_nursery, "Potential pinning address out of range");
595 SGEN_ASSERT (0, addr >= last, "Pin queue not sorted");
602 SGEN_LOG (5, "Considering pinning addr %p", addr);
603 /* We've already processed everything up to pinning_front. */
604 if (addr < pinning_front) {
610 * Find the closest scan start <= addr. We might search backward in the
611 * scan_starts array because entries might be NULL. In the worst case we
612 * start at start_nursery.
614 idx = ((char*)addr - (char*)section->data) / SCAN_START_SIZE;
615 SGEN_ASSERT (0, idx < section->num_scan_start, "Scan start index out of range");
616 search_start = (void*)section->scan_starts [idx];
617 if (!search_start || search_start > addr) {
620 search_start = section->scan_starts [idx];
621 if (search_start && search_start <= addr)
624 if (!search_start || search_start > addr)
625 search_start = start_nursery;
629 * If the pinning front is closer than the scan start we found, start
630 * searching at the front.
632 if (search_start < pinning_front)
633 search_start = pinning_front;
636 * Now addr should be in an object a short distance from search_start.
638 * search_start must point to zeroed mem or point to an object.
641 size_t obj_size, canarified_obj_size;
644 if (!*(void**)search_start) {
645 search_start = (void*)ALIGN_UP ((mword)search_start + sizeof (gpointer));
646 /* The loop condition makes sure we don't overrun addr. */
650 canarified_obj_size = obj_size = ALIGN_UP (safe_object_get_size ((GCObject*)search_start));
653 * Filler arrays are marked by an invalid sync word. We don't
654 * consider them for pinning. They are not delimited by canaries,
657 if (!sgen_client_object_is_array_fill ((GCObject*)search_start)) {
658 CHECK_CANARY_FOR_OBJECT (search_start, TRUE);
659 CANARIFY_SIZE (canarified_obj_size);
661 if (addr >= search_start && (char*)addr < (char*)search_start + obj_size) {
662 /* This is the object we're looking for. */
663 obj_to_pin = (GCObject*)search_start;
664 obj_to_pin_size = canarified_obj_size;
669 /* Skip to the next object */
670 search_start = (void*)((char*)search_start + canarified_obj_size);
671 } while (search_start <= addr);
673 /* We've searched past the address we were looking for. */
675 pinning_front = search_start;
676 goto next_pin_queue_entry;
680 * We've found an object to pin. It might still be a dummy array, but we
681 * can advance the pinning front in any case.
683 pinning_front = (char*)obj_to_pin + obj_to_pin_size;
686 * If this is a dummy array marking the beginning of a nursery
687 * fragment, we don't pin it.
689 if (sgen_client_object_is_array_fill (obj_to_pin))
690 goto next_pin_queue_entry;
693 * Finally - pin the object!
695 desc = sgen_obj_get_descriptor_safe (obj_to_pin);
696 if (do_scan_objects) {
697 scan_func (obj_to_pin, desc, queue);
699 SGEN_LOG (4, "Pinned object %p, vtable %p (%s), count %d\n",
700 obj_to_pin, *(void**)obj_to_pin, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj_to_pin)), count);
701 binary_protocol_pin (obj_to_pin,
702 (gpointer)LOAD_VTABLE (obj_to_pin),
703 safe_object_get_size (obj_to_pin));
705 pin_object (obj_to_pin);
706 GRAY_OBJECT_ENQUEUE (queue, obj_to_pin, desc);
707 sgen_pin_stats_register_object (obj_to_pin, obj_to_pin_size);
708 definitely_pinned [count] = obj_to_pin;
712 next_pin_queue_entry:
716 sgen_client_nursery_objects_pinned (definitely_pinned, count);
717 stat_pinned_objects += count;
722 pin_objects_in_nursery (gboolean do_scan_objects, ScanCopyContext ctx)
726 if (nursery_section->pin_queue_first_entry == nursery_section->pin_queue_last_entry)
729 reduced_to = pin_objects_from_nursery_pin_queue (do_scan_objects, ctx);
730 nursery_section->pin_queue_last_entry = nursery_section->pin_queue_first_entry + reduced_to;
734 * This function is only ever called (via `collector_pin_object()` in `sgen-copy-object.h`)
735 * when we can't promote an object because we're out of memory.
738 sgen_pin_object (GCObject *object, GrayQueue *queue)
741 * All pinned objects are assumed to have been staged, so we need to stage as well.
742 * Also, the count of staged objects shows that "late pinning" happened.
744 sgen_pin_stage_ptr (object);
746 SGEN_PIN_OBJECT (object);
747 binary_protocol_pin (object, (gpointer)LOAD_VTABLE (object), safe_object_get_size (object));
750 sgen_pin_stats_register_object (object, safe_object_get_size (object));
752 GRAY_OBJECT_ENQUEUE (queue, object, sgen_obj_get_descriptor_safe (object));
755 /* Sort the addresses in array in increasing order.
756 * Done using a by-the book heap sort. Which has decent and stable performance, is pretty cache efficient.
759 sgen_sort_addresses (void **array, size_t size)
764 for (i = 1; i < size; ++i) {
767 size_t parent = (child - 1) / 2;
769 if (array [parent] >= array [child])
772 tmp = array [parent];
773 array [parent] = array [child];
780 for (i = size - 1; i > 0; --i) {
783 array [i] = array [0];
789 while (root * 2 + 1 <= end) {
790 size_t child = root * 2 + 1;
792 if (child < end && array [child] < array [child + 1])
794 if (array [root] >= array [child])
798 array [root] = array [child];
807 * Scan the memory between start and end and queue values which could be pointers
808 * to the area between start_nursery and end_nursery for later consideration.
809 * Typically used for thread stacks.
812 sgen_conservatively_pin_objects_from (void **start, void **end, void *start_nursery, void *end_nursery, int pin_type)
816 SGEN_ASSERT (0, ((mword)start & (SIZEOF_VOID_P - 1)) == 0, "Why are we scanning for references in unaligned memory ?");
818 #if defined(VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE) && !defined(_WIN64)
819 VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE (start, (char*)end - (char*)start);
822 while (start < end) {
824 * *start can point to the middle of an object
825 * note: should we handle pointing at the end of an object?
826 * pinning in C# code disallows pointing at the end of an object
827 * but there is some small chance that an optimizing C compiler
828 * may keep the only reference to an object by pointing
829 * at the end of it. We ignore this small chance for now.
830 * Pointers to the end of an object are indistinguishable
831 * from pointers to the start of the next object in memory
832 * so if we allow that we'd need to pin two objects...
833 * We queue the pointer in an array, the
834 * array will then be sorted and uniqued. This way
835 * we can coalesce several pinning pointers and it should
836 * be faster since we'd do a memory scan with increasing
837 * addresses. Note: we can align the address to the allocation
838 * alignment, so the unique process is more effective.
840 mword addr = (mword)*start;
841 addr &= ~(ALLOC_ALIGN - 1);
842 if (addr >= (mword)start_nursery && addr < (mword)end_nursery) {
843 SGEN_LOG (6, "Pinning address %p from %p", (void*)addr, start);
844 sgen_pin_stage_ptr ((void*)addr);
845 binary_protocol_pin_stage (start, (void*)addr);
846 sgen_pin_stats_register_address ((char*)addr, pin_type);
852 SGEN_LOG (7, "found %d potential pinned heap pointers", count);
856 * The first thing we do in a collection is to identify pinned objects.
857 * This function considers all the areas of memory that need to be
858 * conservatively scanned.
861 pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx)
865 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);
866 /* objects pinned from the API are inside these roots */
867 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_PINNED], void **, start_root, RootRecord *, root) {
868 SGEN_LOG (6, "Pinned roots %p-%p", start_root, root->end_root);
869 sgen_conservatively_pin_objects_from (start_root, (void**)root->end_root, start_nursery, end_nursery, PIN_TYPE_OTHER);
870 } SGEN_HASH_TABLE_FOREACH_END;
871 /* now deal with the thread stacks
872 * in the future we should be able to conservatively scan only:
873 * *) the cpu registers
874 * *) the unmanaged stack frames
875 * *) the _last_ managed stack frame
876 * *) pointers slots in managed frames
878 sgen_client_scan_thread_data (start_nursery, end_nursery, FALSE, ctx);
882 single_arg_user_copy_or_mark (GCObject **obj, void *gc_data)
884 ScanCopyContext *ctx = (ScanCopyContext *)gc_data;
885 ctx->ops->copy_or_mark_object (obj, ctx->queue);
889 * The memory area from start_root to end_root contains pointers to objects.
890 * Their position is precisely described by @desc (this means that the pointer
891 * can be either NULL or the pointer to the start of an object).
892 * This functions copies them to to_space updates them.
894 * This function is not thread-safe!
897 precisely_scan_objects_from (void** start_root, void** end_root, char* n_start, char *n_end, SgenDescriptor desc, ScanCopyContext ctx)
899 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
900 SgenGrayQueue *queue = ctx.queue;
902 switch (desc & ROOT_DESC_TYPE_MASK) {
903 case ROOT_DESC_BITMAP:
904 desc >>= ROOT_DESC_TYPE_SHIFT;
906 if ((desc & 1) && *start_root) {
907 copy_func ((GCObject**)start_root, queue);
908 SGEN_LOG (9, "Overwrote root at %p with %p", start_root, *start_root);
914 case ROOT_DESC_COMPLEX: {
915 gsize *bitmap_data = (gsize *)sgen_get_complex_descriptor_bitmap (desc);
916 gsize bwords = (*bitmap_data) - 1;
917 void **start_run = start_root;
919 while (bwords-- > 0) {
920 gsize bmap = *bitmap_data++;
921 void **objptr = start_run;
923 if ((bmap & 1) && *objptr) {
924 copy_func ((GCObject**)objptr, queue);
925 SGEN_LOG (9, "Overwrote root at %p with %p", objptr, *objptr);
930 start_run += GC_BITS_PER_WORD;
934 case ROOT_DESC_USER: {
935 SgenUserRootMarkFunc marker = sgen_get_user_descriptor_func (desc);
936 marker (start_root, single_arg_user_copy_or_mark, &ctx);
939 case ROOT_DESC_RUN_LEN:
940 g_assert_not_reached ();
942 g_assert_not_reached ();
947 reset_heap_boundaries (void)
949 lowest_heap_address = ~(mword)0;
950 highest_heap_address = 0;
954 sgen_update_heap_boundaries (mword low, mword high)
959 old = lowest_heap_address;
962 } while (SGEN_CAS_PTR ((gpointer*)&lowest_heap_address, (gpointer)low, (gpointer)old) != (gpointer)old);
965 old = highest_heap_address;
968 } while (SGEN_CAS_PTR ((gpointer*)&highest_heap_address, (gpointer)high, (gpointer)old) != (gpointer)old);
972 * Allocate and setup the data structures needed to be able to allocate objects
973 * in the nursery. The nursery is stored in nursery_section.
978 GCMemSection *section;
985 SGEN_LOG (2, "Allocating nursery size: %zu", (size_t)sgen_nursery_size);
986 /* later we will alloc a larger area for the nursery but only activate
987 * what we need. The rest will be used as expansion if we have too many pinned
988 * objects in the existing nursery.
990 /* FIXME: handle OOM */
991 section = (GCMemSection *)sgen_alloc_internal (INTERNAL_MEM_SECTION);
993 alloc_size = sgen_nursery_size;
995 /* If there isn't enough space even for the nursery we should simply abort. */
996 g_assert (sgen_memgov_try_alloc_space (alloc_size, SPACE_NURSERY));
998 data = (char *)major_collector.alloc_heap (alloc_size, alloc_size, DEFAULT_NURSERY_BITS);
999 sgen_update_heap_boundaries ((mword)data, (mword)(data + sgen_nursery_size));
1000 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 ());
1001 section->data = section->next_data = data;
1002 section->size = alloc_size;
1003 section->end_data = data + sgen_nursery_size;
1004 scan_starts = (alloc_size + SCAN_START_SIZE - 1) / SCAN_START_SIZE;
1005 section->scan_starts = (char **)sgen_alloc_internal_dynamic (sizeof (char*) * scan_starts, INTERNAL_MEM_SCAN_STARTS, TRUE);
1006 section->num_scan_start = scan_starts;
1008 nursery_section = section;
1010 sgen_nursery_allocator_set_nursery_bounds (data, data + sgen_nursery_size);
1014 mono_gc_get_logfile (void)
1016 return gc_debug_file;
1020 scan_finalizer_entries (SgenPointerQueue *fin_queue, ScanCopyContext ctx)
1022 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
1023 SgenGrayQueue *queue = ctx.queue;
1026 for (i = 0; i < fin_queue->next_slot; ++i) {
1027 GCObject *obj = (GCObject *)fin_queue->data [i];
1030 SGEN_LOG (5, "Scan of fin ready object: %p (%s)\n", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
1031 copy_func ((GCObject**)&fin_queue->data [i], queue);
1036 generation_name (int generation)
1038 switch (generation) {
1039 case GENERATION_NURSERY: return "nursery";
1040 case GENERATION_OLD: return "old";
1041 default: g_assert_not_reached ();
1046 sgen_generation_name (int generation)
1048 return generation_name (generation);
1052 finish_gray_stack (int generation, ScanCopyContext ctx)
1056 int done_with_ephemerons, ephemeron_rounds = 0;
1057 char *start_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_start () : NULL;
1058 char *end_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_end () : (char*)-1;
1059 SgenGrayQueue *queue = ctx.queue;
1061 binary_protocol_finish_gray_stack_start (sgen_timestamp (), generation);
1063 * We copied all the reachable objects. Now it's the time to copy
1064 * the objects that were not referenced by the roots, but by the copied objects.
1065 * we built a stack of objects pointed to by gray_start: they are
1066 * additional roots and we may add more items as we go.
1067 * We loop until gray_start == gray_objects which means no more objects have
1068 * been added. Note this is iterative: no recursion is involved.
1069 * We need to walk the LO list as well in search of marked big objects
1070 * (use a flag since this is needed only on major collections). We need to loop
1071 * here as well, so keep a counter of marked LO (increasing it in copy_object).
1072 * To achieve better cache locality and cache usage, we drain the gray stack
1073 * frequently, after each object is copied, and just finish the work here.
1075 sgen_drain_gray_stack (ctx);
1077 SGEN_LOG (2, "%s generation done", generation_name (generation));
1080 Reset bridge data, we might have lingering data from a previous collection if this is a major
1081 collection trigged by minor overflow.
1083 We must reset the gathered bridges since their original block might be evacuated due to major
1084 fragmentation in the meanwhile and the bridge code should not have to deal with that.
1086 if (sgen_client_bridge_need_processing ())
1087 sgen_client_bridge_reset_data ();
1090 * Walk the ephemeron tables marking all values with reachable keys. This must be completely done
1091 * before processing finalizable objects and non-tracking weak links to avoid finalizing/clearing
1092 * objects that are in fact reachable.
1094 done_with_ephemerons = 0;
1096 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1097 sgen_drain_gray_stack (ctx);
1099 } while (!done_with_ephemerons);
1101 sgen_client_mark_togglerefs (start_addr, end_addr, ctx);
1103 if (sgen_client_bridge_need_processing ()) {
1104 /*Make sure the gray stack is empty before we process bridge objects so we get liveness right*/
1105 sgen_drain_gray_stack (ctx);
1106 sgen_collect_bridge_objects (generation, ctx);
1107 if (generation == GENERATION_OLD)
1108 sgen_collect_bridge_objects (GENERATION_NURSERY, ctx);
1111 Do the first bridge step here, as the collector liveness state will become useless after that.
1113 An important optimization is to only proccess the possibly dead part of the object graph and skip
1114 over all live objects as we transitively know everything they point must be alive too.
1116 The above invariant is completely wrong if we let the gray queue be drained and mark/copy everything.
1118 This has the unfortunate side effect of making overflow collections perform the first step twice, but
1119 given we now have heuristics that perform major GC in anticipation of minor overflows this should not
1122 sgen_client_bridge_processing_stw_step ();
1126 Make sure we drain the gray stack before processing disappearing links and finalizers.
1127 If we don't make sure it is empty we might wrongly see a live object as dead.
1129 sgen_drain_gray_stack (ctx);
1132 We must clear weak links that don't track resurrection before processing object ready for
1133 finalization so they can be cleared before that.
1135 sgen_null_link_in_range (generation, ctx, FALSE);
1136 if (generation == GENERATION_OLD)
1137 sgen_null_link_in_range (GENERATION_NURSERY, ctx, FALSE);
1140 /* walk the finalization queue and move also the objects that need to be
1141 * finalized: use the finalized objects as new roots so the objects they depend
1142 * on are also not reclaimed. As with the roots above, only objects in the nursery
1143 * are marked/copied.
1145 sgen_finalize_in_range (generation, ctx);
1146 if (generation == GENERATION_OLD)
1147 sgen_finalize_in_range (GENERATION_NURSERY, ctx);
1148 /* drain the new stack that might have been created */
1149 SGEN_LOG (6, "Precise scan of gray area post fin");
1150 sgen_drain_gray_stack (ctx);
1153 * This must be done again after processing finalizable objects since CWL slots are cleared only after the key is finalized.
1155 done_with_ephemerons = 0;
1157 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1158 sgen_drain_gray_stack (ctx);
1160 } while (!done_with_ephemerons);
1162 sgen_client_clear_unreachable_ephemerons (ctx);
1165 * We clear togglerefs only after all possible chances of revival are done.
1166 * This is semantically more inline with what users expect and it allows for
1167 * user finalizers to correctly interact with TR objects.
1169 sgen_client_clear_togglerefs (start_addr, end_addr, ctx);
1172 SGEN_LOG (2, "Finalize queue handling scan for %s generation: %lld usecs %d ephemeron rounds", generation_name (generation), TV_ELAPSED (atv, btv), ephemeron_rounds);
1175 * handle disappearing links
1176 * Note we do this after checking the finalization queue because if an object
1177 * survives (at least long enough to be finalized) we don't clear the link.
1178 * This also deals with a possible issue with the monitor reclamation: with the Boehm
1179 * GC a finalized object my lose the monitor because it is cleared before the finalizer is
1182 g_assert (sgen_gray_object_queue_is_empty (queue));
1184 sgen_null_link_in_range (generation, ctx, TRUE);
1185 if (generation == GENERATION_OLD)
1186 sgen_null_link_in_range (GENERATION_NURSERY, ctx, TRUE);
1187 if (sgen_gray_object_queue_is_empty (queue))
1189 sgen_drain_gray_stack (ctx);
1192 g_assert (sgen_gray_object_queue_is_empty (queue));
1194 sgen_gray_object_queue_trim_free_list (queue);
1195 binary_protocol_finish_gray_stack_end (sgen_timestamp (), generation);
1199 sgen_check_section_scan_starts (GCMemSection *section)
1202 for (i = 0; i < section->num_scan_start; ++i) {
1203 if (section->scan_starts [i]) {
1204 mword size = safe_object_get_size ((GCObject*) section->scan_starts [i]);
1205 SGEN_ASSERT (0, size >= SGEN_CLIENT_MINIMUM_OBJECT_SIZE && size <= MAX_SMALL_OBJ_SIZE, "Weird object size at scan starts.");
1211 check_scan_starts (void)
1213 if (!do_scan_starts_check)
1215 sgen_check_section_scan_starts (nursery_section);
1216 major_collector.check_scan_starts ();
1220 scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx)
1224 SGEN_HASH_TABLE_FOREACH (&roots_hash [root_type], void **, start_root, RootRecord *, root) {
1225 SGEN_LOG (6, "Precise root scan %p-%p (desc: %p)", start_root, root->end_root, (void*)root->root_desc);
1226 precisely_scan_objects_from (start_root, (void**)root->end_root, addr_start, addr_end, root->root_desc, ctx);
1227 } SGEN_HASH_TABLE_FOREACH_END;
1233 static gboolean inited = FALSE;
1238 mono_counters_register ("Collection max time", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME | MONO_COUNTER_MONOTONIC, &time_max);
1240 mono_counters_register ("Minor fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pre_collection_fragment_clear);
1241 mono_counters_register ("Minor pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pinning);
1242 mono_counters_register ("Minor scan remembered set", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_remsets);
1243 mono_counters_register ("Minor scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_pinned);
1244 mono_counters_register ("Minor scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_roots);
1245 mono_counters_register ("Minor fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_fragment_creation);
1247 mono_counters_register ("Major fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pre_collection_fragment_clear);
1248 mono_counters_register ("Major pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pinning);
1249 mono_counters_register ("Major scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_pinned);
1250 mono_counters_register ("Major scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_roots);
1251 mono_counters_register ("Major scan mod union", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_mod_union);
1252 mono_counters_register ("Major finish gray stack", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_finish_gray_stack);
1253 mono_counters_register ("Major free big objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_free_bigobjs);
1254 mono_counters_register ("Major LOS sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_los_sweep);
1255 mono_counters_register ("Major sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_sweep);
1256 mono_counters_register ("Major fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_fragment_creation);
1258 mono_counters_register ("Number of pinned objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_pinned_objects);
1260 #ifdef HEAVY_STATISTICS
1261 mono_counters_register ("WBarrier remember pointer", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_add_to_global_remset);
1262 mono_counters_register ("WBarrier arrayref copy", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_arrayref_copy);
1263 mono_counters_register ("WBarrier generic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store);
1264 mono_counters_register ("WBarrier generic atomic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store_atomic);
1265 mono_counters_register ("WBarrier set root", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_set_root);
1267 mono_counters_register ("# objects allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_alloced_degraded);
1268 mono_counters_register ("bytes allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_bytes_alloced_degraded);
1270 mono_counters_register ("# copy_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_nursery);
1271 mono_counters_register ("# objects copied (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_nursery);
1272 mono_counters_register ("# copy_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_major);
1273 mono_counters_register ("# objects copied (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_major);
1275 mono_counters_register ("# scan_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_nursery);
1276 mono_counters_register ("# scan_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_major);
1278 mono_counters_register ("Slots allocated in vain", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_slots_allocated_in_vain);
1280 mono_counters_register ("# nursery copy_object() failed from space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_from_space);
1281 mono_counters_register ("# nursery copy_object() failed forwarded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_forwarded);
1282 mono_counters_register ("# nursery copy_object() failed pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_pinned);
1283 mono_counters_register ("# nursery copy_object() failed to space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_to_space);
1285 sgen_nursery_allocator_init_heavy_stats ();
1293 reset_pinned_from_failed_allocation (void)
1295 bytes_pinned_from_failed_allocation = 0;
1299 sgen_set_pinned_from_failed_allocation (mword objsize)
1301 bytes_pinned_from_failed_allocation += objsize;
1305 sgen_collection_is_concurrent (void)
1307 switch (current_collection_generation) {
1308 case GENERATION_NURSERY:
1310 case GENERATION_OLD:
1311 return concurrent_collection_in_progress;
1313 g_error ("Invalid current generation %d", current_collection_generation);
1319 sgen_concurrent_collection_in_progress (void)
1321 return concurrent_collection_in_progress;
1325 SgenThreadPoolJob job;
1326 SgenObjectOperations *ops;
1330 job_remembered_set_scan (void *worker_data_untyped, SgenThreadPoolJob *job)
1332 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1333 ScanJob *job_data = (ScanJob*)job;
1334 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1335 remset.scan_remsets (ctx);
1339 SgenThreadPoolJob job;
1340 SgenObjectOperations *ops;
1344 } ScanFromRegisteredRootsJob;
1347 job_scan_from_registered_roots (void *worker_data_untyped, SgenThreadPoolJob *job)
1349 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1350 ScanFromRegisteredRootsJob *job_data = (ScanFromRegisteredRootsJob*)job;
1351 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1353 scan_from_registered_roots (job_data->heap_start, job_data->heap_end, job_data->root_type, ctx);
1357 SgenThreadPoolJob job;
1358 SgenObjectOperations *ops;
1361 } ScanThreadDataJob;
1364 job_scan_thread_data (void *worker_data_untyped, SgenThreadPoolJob *job)
1366 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1367 ScanThreadDataJob *job_data = (ScanThreadDataJob*)job;
1368 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1370 sgen_client_scan_thread_data (job_data->heap_start, job_data->heap_end, TRUE, ctx);
1374 SgenThreadPoolJob job;
1375 SgenObjectOperations *ops;
1376 SgenPointerQueue *queue;
1377 } ScanFinalizerEntriesJob;
1380 job_scan_finalizer_entries (void *worker_data_untyped, SgenThreadPoolJob *job)
1382 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1383 ScanFinalizerEntriesJob *job_data = (ScanFinalizerEntriesJob*)job;
1384 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1386 scan_finalizer_entries (job_data->queue, ctx);
1390 job_scan_major_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 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx);
1401 job_scan_los_mod_union_card_table (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);
1408 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx);
1412 job_mod_union_preclean (void *worker_data_untyped, SgenThreadPoolJob *job)
1414 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1415 ScanJob *job_data = (ScanJob*)job;
1416 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1418 g_assert (concurrent_collection_in_progress);
1420 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx);
1421 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx);
1425 init_gray_queue (gboolean use_workers)
1428 sgen_workers_init_distribute_gray_queue ();
1429 sgen_gray_object_queue_init (&gray_queue, NULL);
1433 enqueue_scan_from_roots_jobs (char *heap_start, char *heap_end, SgenObjectOperations *ops, gboolean enqueue)
1435 ScanFromRegisteredRootsJob *scrrj;
1436 ScanThreadDataJob *stdj;
1437 ScanFinalizerEntriesJob *sfej;
1439 /* registered roots, this includes static fields */
1441 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots normal", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1443 scrrj->heap_start = heap_start;
1444 scrrj->heap_end = heap_end;
1445 scrrj->root_type = ROOT_TYPE_NORMAL;
1446 sgen_workers_enqueue_job (&scrrj->job, enqueue);
1448 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots wbarrier", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1450 scrrj->heap_start = heap_start;
1451 scrrj->heap_end = heap_end;
1452 scrrj->root_type = ROOT_TYPE_WBARRIER;
1453 sgen_workers_enqueue_job (&scrrj->job, enqueue);
1457 stdj = (ScanThreadDataJob*)sgen_thread_pool_job_alloc ("scan thread data", job_scan_thread_data, sizeof (ScanThreadDataJob));
1458 stdj->heap_start = heap_start;
1459 stdj->heap_end = heap_end;
1460 sgen_workers_enqueue_job (&stdj->job, enqueue);
1462 /* Scan the list of objects ready for finalization. */
1464 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1465 sfej->queue = &fin_ready_queue;
1467 sgen_workers_enqueue_job (&sfej->job, enqueue);
1469 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan critical finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1470 sfej->queue = &critical_fin_queue;
1472 sgen_workers_enqueue_job (&sfej->job, enqueue);
1476 * Perform a nursery collection.
1478 * Return whether any objects were late-pinned due to being out of memory.
1481 collect_nursery (SgenGrayQueue *unpin_queue, gboolean finish_up_concurrent_mark)
1483 gboolean needs_major;
1484 size_t max_garbage_amount;
1486 mword fragment_total;
1488 SgenObjectOperations *object_ops = &sgen_minor_collector.serial_ops;
1489 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, &gray_queue);
1493 if (disable_minor_collections)
1496 TV_GETTIME (last_minor_collection_start_tv);
1497 atv = last_minor_collection_start_tv;
1499 binary_protocol_collection_begin (gc_stats.minor_gc_count, GENERATION_NURSERY);
1501 if (do_verify_nursery || do_dump_nursery_content)
1502 sgen_debug_verify_nursery (do_dump_nursery_content);
1504 current_collection_generation = GENERATION_NURSERY;
1506 SGEN_ASSERT (0, !sgen_collection_is_concurrent (), "Why is the nursery collection concurrent?");
1508 reset_pinned_from_failed_allocation ();
1510 check_scan_starts ();
1512 sgen_nursery_alloc_prepare_for_minor ();
1516 nursery_next = sgen_nursery_alloc_get_upper_alloc_bound ();
1517 /* FIXME: optimize later to use the higher address where an object can be present */
1518 nursery_next = MAX (nursery_next, sgen_get_nursery_end ());
1520 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 ()));
1521 max_garbage_amount = nursery_next - sgen_get_nursery_start ();
1522 g_assert (nursery_section->size >= max_garbage_amount);
1524 /* world must be stopped already */
1526 time_minor_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1528 sgen_client_pre_collection_checks ();
1530 nursery_section->next_data = nursery_next;
1532 major_collector.start_nursery_collection ();
1534 sgen_memgov_minor_collection_start ();
1536 init_gray_queue (FALSE);
1538 gc_stats.minor_gc_count ++;
1540 if (whole_heap_check_before_collection) {
1541 sgen_clear_nursery_fragments ();
1542 sgen_check_whole_heap (finish_up_concurrent_mark);
1544 if (consistency_check_at_minor_collection)
1545 sgen_check_consistency ();
1547 sgen_process_fin_stage_entries ();
1549 /* pin from pinned handles */
1550 sgen_init_pinning ();
1551 sgen_client_binary_protocol_mark_start (GENERATION_NURSERY);
1552 pin_from_roots (sgen_get_nursery_start (), nursery_next, ctx);
1553 /* pin cemented objects */
1554 sgen_pin_cemented_objects ();
1555 /* identify pinned objects */
1556 sgen_optimize_pin_queue ();
1557 sgen_pinning_setup_section (nursery_section);
1559 pin_objects_in_nursery (FALSE, ctx);
1560 sgen_pinning_trim_queue_to_section (nursery_section);
1563 time_minor_pinning += TV_ELAPSED (btv, atv);
1564 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), TV_ELAPSED (btv, atv));
1565 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1567 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan remset", job_remembered_set_scan, sizeof (ScanJob));
1568 sj->ops = object_ops;
1569 sgen_workers_enqueue_job (&sj->job, FALSE);
1571 /* we don't have complete write barrier yet, so we scan all the old generation sections */
1573 time_minor_scan_remsets += TV_ELAPSED (atv, btv);
1574 SGEN_LOG (2, "Old generation scan: %lld usecs", TV_ELAPSED (atv, btv));
1576 sgen_pin_stats_print_class_stats ();
1578 /* FIXME: Why do we do this at this specific, seemingly random, point? */
1579 sgen_client_collecting_minor (&fin_ready_queue, &critical_fin_queue);
1582 time_minor_scan_pinned += TV_ELAPSED (btv, atv);
1584 enqueue_scan_from_roots_jobs (sgen_get_nursery_start (), nursery_next, object_ops, FALSE);
1587 time_minor_scan_roots += TV_ELAPSED (atv, btv);
1589 finish_gray_stack (GENERATION_NURSERY, ctx);
1592 time_minor_finish_gray_stack += TV_ELAPSED (btv, atv);
1593 sgen_client_binary_protocol_mark_end (GENERATION_NURSERY);
1595 if (objects_pinned) {
1596 sgen_optimize_pin_queue ();
1597 sgen_pinning_setup_section (nursery_section);
1600 /* walk the pin_queue, build up the fragment list of free memory, unmark
1601 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1604 sgen_client_binary_protocol_reclaim_start (GENERATION_NURSERY);
1605 fragment_total = sgen_build_nursery_fragments (nursery_section, unpin_queue);
1606 if (!fragment_total)
1609 /* Clear TLABs for all threads */
1610 sgen_clear_tlabs ();
1612 sgen_client_binary_protocol_reclaim_end (GENERATION_NURSERY);
1614 time_minor_fragment_creation += TV_ELAPSED (atv, btv);
1615 SGEN_LOG (2, "Fragment creation: %lld usecs, %lu bytes available", TV_ELAPSED (atv, btv), (unsigned long)fragment_total);
1617 if (consistency_check_at_minor_collection)
1618 sgen_check_major_refs ();
1620 major_collector.finish_nursery_collection ();
1622 TV_GETTIME (last_minor_collection_end_tv);
1623 gc_stats.minor_gc_time += TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv);
1625 sgen_debug_dump_heap ("minor", gc_stats.minor_gc_count - 1, NULL);
1627 /* prepare the pin queue for the next collection */
1628 sgen_finish_pinning ();
1629 if (sgen_have_pending_finalizers ()) {
1630 SGEN_LOG (4, "Finalizer-thread wakeup");
1631 sgen_client_finalize_notify ();
1633 sgen_pin_stats_reset ();
1634 /* clear cemented hash */
1635 sgen_cement_clear_below_threshold ();
1637 g_assert (sgen_gray_object_queue_is_empty (&gray_queue));
1639 remset.finish_minor_collection ();
1641 check_scan_starts ();
1643 binary_protocol_flush_buffers (FALSE);
1645 sgen_memgov_minor_collection_end ();
1647 /*objects are late pinned because of lack of memory, so a major is a good call*/
1648 needs_major = objects_pinned > 0;
1649 current_collection_generation = -1;
1652 binary_protocol_collection_end (gc_stats.minor_gc_count - 1, GENERATION_NURSERY, 0, 0);
1654 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1655 sgen_check_nursery_objects_pinned (unpin_queue != NULL);
1661 COPY_OR_MARK_FROM_ROOTS_SERIAL,
1662 COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT,
1663 COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT
1664 } CopyOrMarkFromRootsMode;
1667 major_copy_or_mark_from_roots (size_t *old_next_pin_slot, CopyOrMarkFromRootsMode mode, SgenObjectOperations *object_ops)
1672 /* FIXME: only use these values for the precise scan
1673 * note that to_space pointers should be excluded anyway...
1675 char *heap_start = NULL;
1676 char *heap_end = (char*)-1;
1677 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, WORKERS_DISTRIBUTE_GRAY_QUEUE);
1678 gboolean concurrent = mode != COPY_OR_MARK_FROM_ROOTS_SERIAL;
1680 SGEN_ASSERT (0, !!concurrent == !!concurrent_collection_in_progress, "We've been called with the wrong mode.");
1682 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1683 /*This cleans up unused fragments */
1684 sgen_nursery_allocator_prepare_for_pinning ();
1686 if (do_concurrent_checks)
1687 sgen_debug_check_nursery_is_clean ();
1689 /* The concurrent collector doesn't touch the nursery. */
1690 sgen_nursery_alloc_prepare_for_major ();
1693 init_gray_queue (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1697 /* Pinning depends on this */
1698 sgen_clear_nursery_fragments ();
1700 if (whole_heap_check_before_collection)
1701 sgen_check_whole_heap (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT);
1704 time_major_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1706 if (!sgen_collection_is_concurrent ())
1707 nursery_section->next_data = sgen_get_nursery_end ();
1708 /* we should also coalesce scanning from sections close to each other
1709 * and deal with pointers outside of the sections later.
1714 sgen_client_pre_collection_checks ();
1716 if (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1717 /* Remsets are not useful for a major collection */
1718 remset.clear_cards ();
1721 sgen_process_fin_stage_entries ();
1724 sgen_init_pinning ();
1725 SGEN_LOG (6, "Collecting pinned addresses");
1726 pin_from_roots ((void*)lowest_heap_address, (void*)highest_heap_address, ctx);
1727 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1728 /* Pin cemented objects that were forced */
1729 sgen_pin_cemented_objects ();
1731 sgen_optimize_pin_queue ();
1732 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1734 * Cemented objects that are in the pinned list will be marked. When
1735 * marking concurrently we won't mark mod-union cards for these objects.
1736 * Instead they will remain cemented until the next major collection,
1737 * when we will recheck if they are still pinned in the roots.
1739 sgen_cement_force_pinned ();
1742 sgen_client_collecting_major_1 ();
1745 * pin_queue now contains all candidate pointers, sorted and
1746 * uniqued. We must do two passes now to figure out which
1747 * objects are pinned.
1749 * The first is to find within the pin_queue the area for each
1750 * section. This requires that the pin_queue be sorted. We
1751 * also process the LOS objects and pinned chunks here.
1753 * The second, destructive, pass is to reduce the section
1754 * areas to pointers to the actually pinned objects.
1756 SGEN_LOG (6, "Pinning from sections");
1757 /* first pass for the sections */
1758 sgen_find_section_pin_queue_start_end (nursery_section);
1759 /* identify possible pointers to the insize of large objects */
1760 SGEN_LOG (6, "Pinning from large objects");
1761 for (bigobj = los_object_list; bigobj; bigobj = bigobj->next) {
1763 if (sgen_find_optimized_pin_queue_area ((char*)bigobj->data, (char*)bigobj->data + sgen_los_object_size (bigobj), &dummy, &dummy)) {
1764 binary_protocol_pin (bigobj->data, (gpointer)LOAD_VTABLE (bigobj->data), safe_object_get_size (bigobj->data));
1766 if (sgen_los_object_is_pinned (bigobj->data)) {
1767 SGEN_ASSERT (0, mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, "LOS objects can only be pinned here after concurrent marking.");
1770 sgen_los_pin_object (bigobj->data);
1771 if (SGEN_OBJECT_HAS_REFERENCES (bigobj->data))
1772 GRAY_OBJECT_ENQUEUE (WORKERS_DISTRIBUTE_GRAY_QUEUE, bigobj->data, sgen_obj_get_descriptor ((GCObject*)bigobj->data));
1773 sgen_pin_stats_register_object (bigobj->data, safe_object_get_size (bigobj->data));
1774 SGEN_LOG (6, "Marked large object %p (%s) size: %lu from roots", bigobj->data,
1775 sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (bigobj->data)),
1776 (unsigned long)sgen_los_object_size (bigobj));
1778 sgen_client_pinned_los_object (bigobj->data);
1782 pin_objects_in_nursery (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT, ctx);
1783 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1784 sgen_check_nursery_objects_pinned (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1786 major_collector.pin_objects (WORKERS_DISTRIBUTE_GRAY_QUEUE);
1787 if (old_next_pin_slot)
1788 *old_next_pin_slot = sgen_get_pinned_count ();
1791 time_major_pinning += TV_ELAPSED (atv, btv);
1792 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), TV_ELAPSED (atv, btv));
1793 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1795 major_collector.init_to_space ();
1797 SGEN_ASSERT (0, sgen_workers_all_done (), "Why are the workers not done when we start or finish a major collection?");
1798 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1799 if (sgen_workers_have_idle_work ()) {
1801 * We force the finish of the worker with the new object ops context
1802 * which can also do copying. We need to have finished pinning.
1804 sgen_workers_start_all_workers (object_ops, NULL);
1805 sgen_workers_join ();
1809 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1810 main_gc_thread = mono_native_thread_self ();
1813 sgen_client_collecting_major_2 ();
1816 time_major_scan_pinned += TV_ELAPSED (btv, atv);
1818 sgen_client_collecting_major_3 (&fin_ready_queue, &critical_fin_queue);
1820 enqueue_scan_from_roots_jobs (heap_start, heap_end, object_ops, FALSE);
1823 time_major_scan_roots += TV_ELAPSED (atv, btv);
1826 * We start the concurrent worker after pinning and after we scanned the roots
1827 * in order to make sure that the worker does not finish before handling all
1830 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1831 if (precleaning_enabled) {
1833 /* Mod union preclean job */
1834 sj = (ScanJob*)sgen_thread_pool_job_alloc ("preclean mod union cardtable", job_mod_union_preclean, sizeof (ScanJob));
1835 sj->ops = object_ops;
1836 sgen_workers_start_all_workers (object_ops, &sj->job);
1838 sgen_workers_start_all_workers (object_ops, NULL);
1840 gray_queue_enable_redirect (WORKERS_DISTRIBUTE_GRAY_QUEUE);
1843 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1846 /* Mod union card table */
1847 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan mod union cardtable", job_scan_major_mod_union_card_table, sizeof (ScanJob));
1848 sj->ops = object_ops;
1849 sgen_workers_enqueue_job (&sj->job, FALSE);
1851 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan LOS mod union cardtable", job_scan_los_mod_union_card_table, sizeof (ScanJob));
1852 sj->ops = object_ops;
1853 sgen_workers_enqueue_job (&sj->job, FALSE);
1856 time_major_scan_mod_union += TV_ELAPSED (btv, atv);
1859 sgen_pin_stats_print_class_stats ();
1863 major_finish_copy_or_mark (CopyOrMarkFromRootsMode mode)
1865 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1866 sgen_finish_pinning ();
1868 sgen_pin_stats_reset ();
1870 if (do_concurrent_checks)
1871 sgen_debug_check_nursery_is_clean ();
1876 major_start_collection (gboolean concurrent, size_t *old_next_pin_slot)
1878 SgenObjectOperations *object_ops;
1880 binary_protocol_collection_begin (gc_stats.major_gc_count, GENERATION_OLD);
1882 current_collection_generation = GENERATION_OLD;
1884 g_assert (sgen_section_gray_queue_is_empty (sgen_workers_get_distribute_section_gray_queue ()));
1887 sgen_cement_reset ();
1890 g_assert (major_collector.is_concurrent);
1891 concurrent_collection_in_progress = TRUE;
1893 object_ops = &major_collector.major_ops_concurrent_start;
1895 object_ops = &major_collector.major_ops_serial;
1898 reset_pinned_from_failed_allocation ();
1900 sgen_memgov_major_collection_start ();
1902 //count_ref_nonref_objs ();
1903 //consistency_check ();
1905 check_scan_starts ();
1908 SGEN_LOG (1, "Start major collection %d", gc_stats.major_gc_count);
1909 gc_stats.major_gc_count ++;
1911 if (major_collector.start_major_collection)
1912 major_collector.start_major_collection ();
1914 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);
1915 major_finish_copy_or_mark (concurrent ? COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT : COPY_OR_MARK_FROM_ROOTS_SERIAL);
1919 major_finish_collection (const char *reason, size_t old_next_pin_slot, gboolean forced)
1921 ScannedObjectCounts counts;
1922 SgenObjectOperations *object_ops;
1923 mword fragment_total;
1929 if (concurrent_collection_in_progress) {
1930 object_ops = &major_collector.major_ops_concurrent_finish;
1932 major_copy_or_mark_from_roots (NULL, COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, object_ops);
1934 major_finish_copy_or_mark (COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT);
1936 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1937 main_gc_thread = NULL;
1940 object_ops = &major_collector.major_ops_serial;
1943 g_assert (sgen_section_gray_queue_is_empty (sgen_workers_get_distribute_section_gray_queue ()));
1945 /* all the objects in the heap */
1946 finish_gray_stack (GENERATION_OLD, CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, &gray_queue));
1948 time_major_finish_gray_stack += TV_ELAPSED (btv, atv);
1950 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after joining");
1952 if (objects_pinned) {
1953 g_assert (!concurrent_collection_in_progress);
1956 * This is slow, but we just OOM'd.
1958 * See comment at `sgen_pin_queue_clear_discarded_entries` for how the pin
1959 * queue is laid out at this point.
1961 sgen_pin_queue_clear_discarded_entries (nursery_section, old_next_pin_slot);
1963 * We need to reestablish all pinned nursery objects in the pin queue
1964 * because they're needed for fragment creation. Unpinning happens by
1965 * walking the whole queue, so it's not necessary to reestablish where major
1966 * heap block pins are - all we care is that they're still in there
1969 sgen_optimize_pin_queue ();
1970 sgen_find_section_pin_queue_start_end (nursery_section);
1974 reset_heap_boundaries ();
1975 sgen_update_heap_boundaries ((mword)sgen_get_nursery_start (), (mword)sgen_get_nursery_end ());
1977 /* walk the pin_queue, build up the fragment list of free memory, unmark
1978 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1981 fragment_total = sgen_build_nursery_fragments (nursery_section, NULL);
1982 if (!fragment_total)
1984 SGEN_LOG (4, "Free space in nursery after major %ld", (long)fragment_total);
1986 if (do_concurrent_checks && concurrent_collection_in_progress)
1987 sgen_debug_check_nursery_is_clean ();
1989 /* prepare the pin queue for the next collection */
1990 sgen_finish_pinning ();
1992 /* Clear TLABs for all threads */
1993 sgen_clear_tlabs ();
1995 sgen_pin_stats_reset ();
1997 sgen_cement_clear_below_threshold ();
1999 if (check_mark_bits_after_major_collection)
2000 sgen_check_heap_marked (concurrent_collection_in_progress);
2003 time_major_fragment_creation += TV_ELAPSED (atv, btv);
2005 binary_protocol_sweep_begin (GENERATION_OLD, !major_collector.sweeps_lazily);
2006 sgen_memgov_major_pre_sweep ();
2009 time_major_free_bigobjs += TV_ELAPSED (btv, atv);
2014 time_major_los_sweep += TV_ELAPSED (atv, btv);
2016 major_collector.sweep ();
2018 binary_protocol_sweep_end (GENERATION_OLD, !major_collector.sweeps_lazily);
2021 time_major_sweep += TV_ELAPSED (btv, atv);
2023 sgen_debug_dump_heap ("major", gc_stats.major_gc_count - 1, reason);
2025 if (sgen_have_pending_finalizers ()) {
2026 SGEN_LOG (4, "Finalizer-thread wakeup");
2027 sgen_client_finalize_notify ();
2030 g_assert (sgen_gray_object_queue_is_empty (&gray_queue));
2032 sgen_memgov_major_collection_end (forced);
2033 current_collection_generation = -1;
2035 memset (&counts, 0, sizeof (ScannedObjectCounts));
2036 major_collector.finish_major_collection (&counts);
2038 g_assert (sgen_section_gray_queue_is_empty (sgen_workers_get_distribute_section_gray_queue ()));
2040 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after major collection has finished");
2041 if (concurrent_collection_in_progress)
2042 concurrent_collection_in_progress = FALSE;
2044 check_scan_starts ();
2046 binary_protocol_flush_buffers (FALSE);
2048 //consistency_check ();
2050 binary_protocol_collection_end (gc_stats.major_gc_count - 1, GENERATION_OLD, counts.num_scanned_objects, counts.num_unique_scanned_objects);
2054 major_do_collection (const char *reason, gboolean forced)
2056 TV_DECLARE (time_start);
2057 TV_DECLARE (time_end);
2058 size_t old_next_pin_slot;
2060 if (disable_major_collections)
2063 if (major_collector.get_and_reset_num_major_objects_marked) {
2064 long long num_marked = major_collector.get_and_reset_num_major_objects_marked ();
2065 g_assert (!num_marked);
2068 /* world must be stopped already */
2069 TV_GETTIME (time_start);
2071 major_start_collection (FALSE, &old_next_pin_slot);
2072 major_finish_collection (reason, old_next_pin_slot, forced);
2074 TV_GETTIME (time_end);
2075 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2077 /* FIXME: also report this to the user, preferably in gc-end. */
2078 if (major_collector.get_and_reset_num_major_objects_marked)
2079 major_collector.get_and_reset_num_major_objects_marked ();
2081 return bytes_pinned_from_failed_allocation > 0;
2085 major_start_concurrent_collection (const char *reason)
2087 TV_DECLARE (time_start);
2088 TV_DECLARE (time_end);
2089 long long num_objects_marked;
2091 if (disable_major_collections)
2094 TV_GETTIME (time_start);
2095 SGEN_TV_GETTIME (time_major_conc_collection_start);
2097 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2098 g_assert (num_objects_marked == 0);
2100 binary_protocol_concurrent_start ();
2102 // FIXME: store reason and pass it when finishing
2103 major_start_collection (TRUE, NULL);
2105 gray_queue_redirect (&gray_queue);
2107 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2109 TV_GETTIME (time_end);
2110 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2112 current_collection_generation = -1;
2116 * Returns whether the major collection has finished.
2119 major_should_finish_concurrent_collection (void)
2121 SGEN_ASSERT (0, sgen_gray_object_queue_is_empty (&gray_queue), "Why is the gray queue not empty before we have started doing anything?");
2122 return sgen_workers_all_done ();
2126 major_update_concurrent_collection (void)
2128 TV_DECLARE (total_start);
2129 TV_DECLARE (total_end);
2131 TV_GETTIME (total_start);
2133 binary_protocol_concurrent_update ();
2135 major_collector.update_cardtable_mod_union ();
2136 sgen_los_update_cardtable_mod_union ();
2138 TV_GETTIME (total_end);
2139 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end);
2143 major_finish_concurrent_collection (gboolean forced)
2145 TV_DECLARE (total_start);
2146 TV_DECLARE (total_end);
2148 TV_GETTIME (total_start);
2150 binary_protocol_concurrent_finish ();
2153 * We need to stop all workers since we're updating the cardtable below.
2154 * The workers will be resumed with a finishing pause context to avoid
2155 * additional cardtable and object scanning.
2157 sgen_workers_stop_all_workers ();
2159 SGEN_TV_GETTIME (time_major_conc_collection_end);
2160 gc_stats.major_gc_time_concurrent += SGEN_TV_ELAPSED (time_major_conc_collection_start, time_major_conc_collection_end);
2162 major_collector.update_cardtable_mod_union ();
2163 sgen_los_update_cardtable_mod_union ();
2165 if (mod_union_consistency_check)
2166 sgen_check_mod_union_consistency ();
2168 current_collection_generation = GENERATION_OLD;
2169 sgen_cement_reset ();
2170 major_finish_collection ("finishing", -1, forced);
2172 if (whole_heap_check_before_collection)
2173 sgen_check_whole_heap (FALSE);
2175 TV_GETTIME (total_end);
2176 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end) - TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv);
2178 current_collection_generation = -1;
2182 * Ensure an allocation request for @size will succeed by freeing enough memory.
2184 * LOCKING: The GC lock MUST be held.
2187 sgen_ensure_free_space (size_t size, int generation)
2189 int generation_to_collect = -1;
2190 const char *reason = NULL;
2192 if (generation == GENERATION_OLD) {
2193 if (sgen_need_major_collection (size)) {
2194 reason = "LOS overflow";
2195 generation_to_collect = GENERATION_OLD;
2198 if (degraded_mode) {
2199 if (sgen_need_major_collection (size)) {
2200 reason = "Degraded mode overflow";
2201 generation_to_collect = GENERATION_OLD;
2203 } else if (sgen_need_major_collection (size)) {
2204 reason = concurrent_collection_in_progress ? "Forced finish concurrent collection" : "Minor allowance";
2205 generation_to_collect = GENERATION_OLD;
2207 generation_to_collect = GENERATION_NURSERY;
2208 reason = "Nursery full";
2212 if (generation_to_collect == -1) {
2213 if (concurrent_collection_in_progress && sgen_workers_all_done ()) {
2214 generation_to_collect = GENERATION_OLD;
2215 reason = "Finish concurrent collection";
2219 if (generation_to_collect == -1)
2221 sgen_perform_collection (size, generation_to_collect, reason, FALSE);
2225 * LOCKING: Assumes the GC lock is held.
2228 sgen_perform_collection (size_t requested_size, int generation_to_collect, const char *reason, gboolean wait_to_finish)
2230 TV_DECLARE (gc_start);
2231 TV_DECLARE (gc_end);
2232 TV_DECLARE (gc_total_start);
2233 TV_DECLARE (gc_total_end);
2234 GGTimingInfo infos [2];
2235 int overflow_generation_to_collect = -1;
2236 int oldest_generation_collected = generation_to_collect;
2237 const char *overflow_reason = NULL;
2239 binary_protocol_collection_requested (generation_to_collect, requested_size, wait_to_finish ? 1 : 0);
2241 SGEN_ASSERT (0, generation_to_collect == GENERATION_NURSERY || generation_to_collect == GENERATION_OLD, "What generation is this?");
2243 TV_GETTIME (gc_start);
2245 sgen_stop_world (generation_to_collect);
2247 TV_GETTIME (gc_total_start);
2249 if (concurrent_collection_in_progress) {
2251 * If the concurrent worker is finished or we are asked to do a major collection
2252 * then we finish the concurrent collection.
2254 gboolean finish = major_should_finish_concurrent_collection () || generation_to_collect == GENERATION_OLD;
2257 major_finish_concurrent_collection (wait_to_finish);
2258 oldest_generation_collected = GENERATION_OLD;
2260 SGEN_ASSERT (0, generation_to_collect == GENERATION_NURSERY, "Why aren't we finishing the concurrent collection?");
2261 major_update_concurrent_collection ();
2262 collect_nursery (NULL, FALSE);
2268 SGEN_ASSERT (0, !concurrent_collection_in_progress, "Why did this not get handled above?");
2271 * There's no concurrent collection in progress. Collect the generation we're asked
2272 * to collect. If the major collector is concurrent and we're not forced to wait,
2273 * start a concurrent collection.
2275 // FIXME: extract overflow reason
2276 if (generation_to_collect == GENERATION_NURSERY) {
2277 if (collect_nursery (NULL, FALSE)) {
2278 overflow_generation_to_collect = GENERATION_OLD;
2279 overflow_reason = "Minor overflow";
2282 if (major_collector.is_concurrent && !wait_to_finish) {
2283 collect_nursery (NULL, FALSE);
2284 major_start_concurrent_collection (reason);
2285 // FIXME: set infos[0] properly
2289 if (major_do_collection (reason, wait_to_finish)) {
2290 overflow_generation_to_collect = GENERATION_NURSERY;
2291 overflow_reason = "Excessive pinning";
2295 TV_GETTIME (gc_end);
2297 memset (infos, 0, sizeof (infos));
2298 infos [0].generation = generation_to_collect;
2299 infos [0].reason = reason;
2300 infos [0].is_overflow = FALSE;
2301 infos [1].generation = -1;
2302 infos [0].total_time = SGEN_TV_ELAPSED (gc_start, gc_end);
2304 SGEN_ASSERT (0, !concurrent_collection_in_progress, "Why did this not get handled above?");
2306 if (overflow_generation_to_collect != -1) {
2308 * We need to do an overflow collection, either because we ran out of memory
2309 * or the nursery is fully pinned.
2312 infos [1].generation = overflow_generation_to_collect;
2313 infos [1].reason = overflow_reason;
2314 infos [1].is_overflow = TRUE;
2317 if (overflow_generation_to_collect == GENERATION_NURSERY)
2318 collect_nursery (NULL, FALSE);
2320 major_do_collection (overflow_reason, wait_to_finish);
2322 TV_GETTIME (gc_end);
2323 infos [1].total_time = SGEN_TV_ELAPSED (gc_start, gc_end);
2325 oldest_generation_collected = MAX (oldest_generation_collected, overflow_generation_to_collect);
2328 SGEN_LOG (2, "Heap size: %lu, LOS size: %lu", (unsigned long)sgen_gc_get_total_heap_allocation (), (unsigned long)los_memory_usage);
2330 /* this also sets the proper pointers for the next allocation */
2331 if (generation_to_collect == GENERATION_NURSERY && !sgen_can_alloc_size (requested_size)) {
2332 /* TypeBuilder and MonoMethod are killing mcs with fragmentation */
2333 SGEN_LOG (1, "nursery collection didn't find enough room for %zd alloc (%zd pinned)", requested_size, sgen_get_pinned_count ());
2334 sgen_dump_pin_queue ();
2339 g_assert (sgen_gray_object_queue_is_empty (&gray_queue));
2341 TV_GETTIME (gc_total_end);
2342 time_max = MAX (time_max, TV_ELAPSED (gc_total_start, gc_total_end));
2344 sgen_restart_world (oldest_generation_collected, infos);
2348 * ######################################################################
2349 * ######## Memory allocation from the OS
2350 * ######################################################################
2351 * This section of code deals with getting memory from the OS and
2352 * allocating memory for GC-internal data structures.
2353 * Internal memory can be handled with a freelist for small objects.
2359 G_GNUC_UNUSED static void
2360 report_internal_mem_usage (void)
2362 printf ("Internal memory usage:\n");
2363 sgen_report_internal_mem_usage ();
2364 printf ("Pinned memory usage:\n");
2365 major_collector.report_pinned_memory_usage ();
2369 * ######################################################################
2370 * ######## Finalization support
2371 * ######################################################################
2375 * If the object has been forwarded it means it's still referenced from a root.
2376 * If it is pinned it's still alive as well.
2377 * A LOS object is only alive if we have pinned it.
2378 * Return TRUE if @obj is ready to be finalized.
2380 static inline gboolean
2381 sgen_is_object_alive (GCObject *object)
2383 if (ptr_in_nursery (object))
2384 return sgen_nursery_is_object_alive (object);
2386 return sgen_major_is_object_alive (object);
2390 * This function returns true if @object is either alive and belongs to the
2391 * current collection - major collections are full heap, so old gen objects
2392 * are never alive during a minor collection.
2395 sgen_is_object_alive_and_on_current_collection (GCObject *object)
2397 if (ptr_in_nursery (object))
2398 return sgen_nursery_is_object_alive (object);
2400 if (current_collection_generation == GENERATION_NURSERY)
2403 return sgen_major_is_object_alive (object);
2408 sgen_gc_is_object_ready_for_finalization (GCObject *object)
2410 return !sgen_is_object_alive (object);
2414 sgen_queue_finalization_entry (GCObject *obj)
2416 gboolean critical = sgen_client_object_has_critical_finalizer (obj);
2418 sgen_pointer_queue_add (critical ? &critical_fin_queue : &fin_ready_queue, obj);
2420 sgen_client_object_queued_for_finalization (obj);
2424 sgen_object_is_live (GCObject *obj)
2426 return sgen_is_object_alive_and_on_current_collection (obj);
2430 * `System.GC.WaitForPendingFinalizers` first checks `sgen_have_pending_finalizers()` to
2431 * determine whether it can exit quickly. The latter must therefore only return FALSE if
2432 * all finalizers have really finished running.
2434 * `sgen_gc_invoke_finalizers()` first dequeues a finalizable object, and then finalizes it.
2435 * This means that just checking whether the queues are empty leaves the possibility that an
2436 * object might have been dequeued but not yet finalized. That's why we need the additional
2437 * flag `pending_unqueued_finalizer`.
2440 static volatile gboolean pending_unqueued_finalizer = FALSE;
2443 sgen_gc_invoke_finalizers (void)
2447 g_assert (!pending_unqueued_finalizer);
2449 /* FIXME: batch to reduce lock contention */
2450 while (sgen_have_pending_finalizers ()) {
2456 * We need to set `pending_unqueued_finalizer` before dequeing the
2457 * finalizable object.
2459 if (!sgen_pointer_queue_is_empty (&fin_ready_queue)) {
2460 pending_unqueued_finalizer = TRUE;
2461 mono_memory_write_barrier ();
2462 obj = (GCObject *)sgen_pointer_queue_pop (&fin_ready_queue);
2463 } else if (!sgen_pointer_queue_is_empty (&critical_fin_queue)) {
2464 pending_unqueued_finalizer = TRUE;
2465 mono_memory_write_barrier ();
2466 obj = (GCObject *)sgen_pointer_queue_pop (&critical_fin_queue);
2472 SGEN_LOG (7, "Finalizing object %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
2480 /* the object is on the stack so it is pinned */
2481 /*g_print ("Calling finalizer for object: %p (%s)\n", obj, sgen_client_object_safe_name (obj));*/
2482 sgen_client_run_finalize (obj);
2485 if (pending_unqueued_finalizer) {
2486 mono_memory_write_barrier ();
2487 pending_unqueued_finalizer = FALSE;
2494 sgen_have_pending_finalizers (void)
2496 return pending_unqueued_finalizer || !sgen_pointer_queue_is_empty (&fin_ready_queue) || !sgen_pointer_queue_is_empty (&critical_fin_queue);
2500 * ######################################################################
2501 * ######## registered roots support
2502 * ######################################################################
2506 * We do not coalesce roots.
2509 sgen_register_root (char *start, size_t size, SgenDescriptor descr, int root_type, int source, const char *msg)
2511 RootRecord new_root;
2514 for (i = 0; i < ROOT_TYPE_NUM; ++i) {
2515 RootRecord *root = (RootRecord *)sgen_hash_table_lookup (&roots_hash [i], start);
2516 /* we allow changing the size and the descriptor (for thread statics etc) */
2518 size_t old_size = root->end_root - start;
2519 root->end_root = start + size;
2520 SGEN_ASSERT (0, !!root->root_desc == !!descr, "Can't change whether a root is precise or conservative.");
2521 SGEN_ASSERT (0, root->source == source, "Can't change a root's source identifier.");
2522 SGEN_ASSERT (0, !!root->msg == !!msg, "Can't change a root's message.");
2523 root->root_desc = descr;
2525 roots_size -= old_size;
2531 new_root.end_root = start + size;
2532 new_root.root_desc = descr;
2533 new_root.source = source;
2536 sgen_hash_table_replace (&roots_hash [root_type], start, &new_root, NULL);
2539 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);
2546 sgen_deregister_root (char* addr)
2552 for (root_type = 0; root_type < ROOT_TYPE_NUM; ++root_type) {
2553 if (sgen_hash_table_remove (&roots_hash [root_type], addr, &root))
2554 roots_size -= (root.end_root - addr);
2560 * ######################################################################
2561 * ######## Thread handling (stop/start code)
2562 * ######################################################################
2566 sgen_get_current_collection_generation (void)
2568 return current_collection_generation;
2572 sgen_thread_register (SgenThreadInfo* info, void *stack_bottom_fallback)
2574 #ifndef HAVE_KW_THREAD
2575 info->tlab_start = info->tlab_next = info->tlab_temp_end = info->tlab_real_end = NULL;
2578 sgen_init_tlab_info (info);
2580 sgen_client_thread_register (info, stack_bottom_fallback);
2586 sgen_thread_unregister (SgenThreadInfo *p)
2588 sgen_client_thread_unregister (p);
2592 * ######################################################################
2593 * ######## Write barriers
2594 * ######################################################################
2598 * Note: the write barriers first do the needed GC work and then do the actual store:
2599 * this way the value is visible to the conservative GC scan after the write barrier
2600 * itself. If a GC interrupts the barrier in the middle, value will be kept alive by
2601 * the conservative scan, otherwise by the remembered set scan.
2605 mono_gc_wbarrier_arrayref_copy (gpointer dest_ptr, gpointer src_ptr, int count)
2607 HEAVY_STAT (++stat_wbarrier_arrayref_copy);
2608 /*This check can be done without taking a lock since dest_ptr array is pinned*/
2609 if (ptr_in_nursery (dest_ptr) || count <= 0) {
2610 mono_gc_memmove_aligned (dest_ptr, src_ptr, count * sizeof (gpointer));
2614 #ifdef SGEN_HEAVY_BINARY_PROTOCOL
2615 if (binary_protocol_is_heavy_enabled ()) {
2617 for (i = 0; i < count; ++i) {
2618 gpointer dest = (gpointer*)dest_ptr + i;
2619 gpointer obj = *((gpointer*)src_ptr + i);
2621 binary_protocol_wbarrier (dest, obj, (gpointer)LOAD_VTABLE (obj));
2626 remset.wbarrier_arrayref_copy (dest_ptr, src_ptr, count);
2630 mono_gc_wbarrier_generic_nostore (gpointer ptr)
2634 HEAVY_STAT (++stat_wbarrier_generic_store);
2636 sgen_client_wbarrier_generic_nostore_check (ptr);
2638 obj = *(gpointer*)ptr;
2640 binary_protocol_wbarrier (ptr, obj, (gpointer)LOAD_VTABLE (obj));
2643 * We need to record old->old pointer locations for the
2644 * concurrent collector.
2646 if (!ptr_in_nursery (obj) && !concurrent_collection_in_progress) {
2647 SGEN_LOG (8, "Skipping remset at %p", ptr);
2651 SGEN_LOG (8, "Adding remset at %p", ptr);
2653 remset.wbarrier_generic_nostore (ptr);
2657 mono_gc_wbarrier_generic_store (gpointer ptr, GCObject* value)
2659 SGEN_LOG (8, "Wbarrier store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2660 SGEN_UPDATE_REFERENCE_ALLOW_NULL (ptr, value);
2661 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2662 mono_gc_wbarrier_generic_nostore (ptr);
2663 sgen_dummy_use (value);
2666 /* Same as mono_gc_wbarrier_generic_store () but performs the store
2667 * as an atomic operation with release semantics.
2670 mono_gc_wbarrier_generic_store_atomic (gpointer ptr, GCObject *value)
2672 HEAVY_STAT (++stat_wbarrier_generic_store_atomic);
2674 SGEN_LOG (8, "Wbarrier atomic store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2676 InterlockedWritePointer ((volatile gpointer *)ptr, value);
2678 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2679 mono_gc_wbarrier_generic_nostore (ptr);
2681 sgen_dummy_use (value);
2685 sgen_wbarrier_value_copy_bitmap (gpointer _dest, gpointer _src, int size, unsigned bitmap)
2687 GCObject **dest = (GCObject **)_dest;
2688 GCObject **src = (GCObject **)_src;
2692 mono_gc_wbarrier_generic_store (dest, *src);
2697 size -= SIZEOF_VOID_P;
2703 * ######################################################################
2704 * ######## Other mono public interface functions.
2705 * ######################################################################
2709 sgen_gc_collect (int generation)
2714 sgen_perform_collection (0, generation, "user request", TRUE);
2719 sgen_gc_collection_count (int generation)
2721 if (generation == 0)
2722 return gc_stats.minor_gc_count;
2723 return gc_stats.major_gc_count;
2727 sgen_gc_get_used_size (void)
2731 tot = los_memory_usage;
2732 tot += nursery_section->next_data - nursery_section->data;
2733 tot += major_collector.get_used_size ();
2734 /* FIXME: account for pinned objects */
2740 sgen_env_var_error (const char *env_var, const char *fallback, const char *description_format, ...)
2744 va_start (ap, description_format);
2746 fprintf (stderr, "Warning: In environment variable `%s': ", env_var);
2747 vfprintf (stderr, description_format, ap);
2749 fprintf (stderr, " - %s", fallback);
2750 fprintf (stderr, "\n");
2756 parse_double_in_interval (const char *env_var, const char *opt_name, const char *opt, double min, double max, double *result)
2759 double val = strtod (opt, &endptr);
2760 if (endptr == opt) {
2761 sgen_env_var_error (env_var, "Using default value.", "`%s` must be a number.", opt_name);
2764 else if (val < min || val > max) {
2765 sgen_env_var_error (env_var, "Using default value.", "`%s` must be between %.2f - %.2f.", opt_name, min, max);
2777 char *major_collector_opt = NULL;
2778 char *minor_collector_opt = NULL;
2779 size_t max_heap = 0;
2780 size_t soft_limit = 0;
2782 gboolean debug_print_allowance = FALSE;
2783 double allowance_ratio = 0, save_target = 0;
2784 gboolean cement_enabled = TRUE;
2787 result = InterlockedCompareExchange (&gc_initialized, -1, 0);
2790 /* already inited */
2793 /* being inited by another thread */
2794 mono_thread_info_usleep (1000);
2797 /* we will init it */
2800 g_assert_not_reached ();
2802 } while (result != 0);
2804 SGEN_TV_GETTIME (sgen_init_timestamp);
2806 #ifdef SGEN_WITHOUT_MONO
2807 mono_thread_smr_init ();
2810 mono_coop_mutex_init (&gc_mutex);
2812 gc_debug_file = stderr;
2814 mono_coop_mutex_init (&sgen_interruption_mutex);
2816 if ((env = g_getenv (MONO_GC_PARAMS_NAME))) {
2817 opts = g_strsplit (env, ",", -1);
2818 for (ptr = opts; *ptr; ++ptr) {
2820 if (g_str_has_prefix (opt, "major=")) {
2821 opt = strchr (opt, '=') + 1;
2822 major_collector_opt = g_strdup (opt);
2823 } else if (g_str_has_prefix (opt, "minor=")) {
2824 opt = strchr (opt, '=') + 1;
2825 minor_collector_opt = g_strdup (opt);
2833 sgen_init_internal_allocator ();
2834 sgen_init_nursery_allocator ();
2835 sgen_init_fin_weak_hash ();
2836 sgen_init_hash_table ();
2837 sgen_init_descriptors ();
2838 sgen_init_gray_queues ();
2839 sgen_init_allocator ();
2840 sgen_init_gchandles ();
2842 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_SECTION, SGEN_SIZEOF_GC_MEM_SECTION);
2843 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_GRAY_QUEUE, sizeof (GrayQueueSection));
2845 sgen_client_init ();
2847 if (!minor_collector_opt) {
2848 sgen_simple_nursery_init (&sgen_minor_collector);
2850 if (!strcmp (minor_collector_opt, "simple")) {
2852 sgen_simple_nursery_init (&sgen_minor_collector);
2853 } else if (!strcmp (minor_collector_opt, "split")) {
2854 sgen_split_nursery_init (&sgen_minor_collector);
2856 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `simple` instead.", "Unknown minor collector `%s'.", minor_collector_opt);
2857 goto use_simple_nursery;
2861 if (!major_collector_opt || !strcmp (major_collector_opt, "marksweep")) {
2862 use_marksweep_major:
2863 sgen_marksweep_init (&major_collector);
2864 } else if (!major_collector_opt || !strcmp (major_collector_opt, "marksweep-conc")) {
2865 sgen_marksweep_conc_init (&major_collector);
2867 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `marksweep` instead.", "Unknown major collector `%s'.", major_collector_opt);
2868 goto use_marksweep_major;
2871 sgen_nursery_size = DEFAULT_NURSERY_SIZE;
2874 gboolean usage_printed = FALSE;
2876 for (ptr = opts; *ptr; ++ptr) {
2878 if (!strcmp (opt, ""))
2880 if (g_str_has_prefix (opt, "major="))
2882 if (g_str_has_prefix (opt, "minor="))
2884 if (g_str_has_prefix (opt, "max-heap-size=")) {
2885 size_t page_size = mono_pagesize ();
2886 size_t max_heap_candidate = 0;
2887 opt = strchr (opt, '=') + 1;
2888 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &max_heap_candidate)) {
2889 max_heap = (max_heap_candidate + page_size - 1) & ~(size_t)(page_size - 1);
2890 if (max_heap != max_heap_candidate)
2891 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Rounding up.", "`max-heap-size` size must be a multiple of %d.", page_size);
2893 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`max-heap-size` must be an integer.");
2897 if (g_str_has_prefix (opt, "soft-heap-limit=")) {
2898 opt = strchr (opt, '=') + 1;
2899 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &soft_limit)) {
2900 if (soft_limit <= 0) {
2901 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be positive.");
2905 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be an integer.");
2911 if (g_str_has_prefix (opt, "nursery-size=")) {
2913 opt = strchr (opt, '=') + 1;
2914 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &val)) {
2915 if ((val & (val - 1))) {
2916 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be a power of two.");
2920 if (val < SGEN_MAX_NURSERY_WASTE) {
2921 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.",
2922 "`nursery-size` must be at least %d bytes.", SGEN_MAX_NURSERY_WASTE);
2926 sgen_nursery_size = val;
2927 sgen_nursery_bits = 0;
2928 while (ONE_P << (++ sgen_nursery_bits) != sgen_nursery_size)
2931 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be an integer.");
2937 if (g_str_has_prefix (opt, "save-target-ratio=")) {
2939 opt = strchr (opt, '=') + 1;
2940 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "save-target-ratio", opt,
2941 SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO, &val)) {
2946 if (g_str_has_prefix (opt, "default-allowance-ratio=")) {
2948 opt = strchr (opt, '=') + 1;
2949 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "default-allowance-ratio", opt,
2950 SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, &val)) {
2951 allowance_ratio = val;
2956 if (!strcmp (opt, "cementing")) {
2957 cement_enabled = TRUE;
2960 if (!strcmp (opt, "no-cementing")) {
2961 cement_enabled = FALSE;
2965 if (!strcmp (opt, "precleaning")) {
2966 precleaning_enabled = TRUE;
2969 if (!strcmp (opt, "no-precleaning")) {
2970 precleaning_enabled = FALSE;
2974 if (major_collector.handle_gc_param && major_collector.handle_gc_param (opt))
2977 if (sgen_minor_collector.handle_gc_param && sgen_minor_collector.handle_gc_param (opt))
2980 if (sgen_client_handle_gc_param (opt))
2983 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Ignoring.", "Unknown option `%s`.", opt);
2988 fprintf (stderr, "\n%s must be a comma-delimited list of one or more of the following:\n", MONO_GC_PARAMS_NAME);
2989 fprintf (stderr, " max-heap-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
2990 fprintf (stderr, " soft-heap-limit=n (where N is an integer, possibly with a k, m or a g suffix)\n");
2991 fprintf (stderr, " nursery-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
2992 fprintf (stderr, " major=COLLECTOR (where COLLECTOR is `marksweep', `marksweep-conc', `marksweep-par')\n");
2993 fprintf (stderr, " minor=COLLECTOR (where COLLECTOR is `simple' or `split')\n");
2994 fprintf (stderr, " wbarrier=WBARRIER (where WBARRIER is `remset' or `cardtable')\n");
2995 fprintf (stderr, " [no-]cementing\n");
2996 if (major_collector.print_gc_param_usage)
2997 major_collector.print_gc_param_usage ();
2998 if (sgen_minor_collector.print_gc_param_usage)
2999 sgen_minor_collector.print_gc_param_usage ();
3000 sgen_client_print_gc_params_usage ();
3001 fprintf (stderr, " Experimental options:\n");
3002 fprintf (stderr, " save-target-ratio=R (where R must be between %.2f - %.2f).\n", SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO);
3003 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);
3004 fprintf (stderr, "\n");
3006 usage_printed = TRUE;
3011 if (major_collector_opt)
3012 g_free (major_collector_opt);
3014 if (minor_collector_opt)
3015 g_free (minor_collector_opt);
3019 sgen_cement_init (cement_enabled);
3021 if ((env = g_getenv (MONO_GC_DEBUG_NAME))) {
3022 gboolean usage_printed = FALSE;
3024 opts = g_strsplit (env, ",", -1);
3025 for (ptr = opts; ptr && *ptr; ptr ++) {
3027 if (!strcmp (opt, ""))
3029 if (opt [0] >= '0' && opt [0] <= '9') {
3030 gc_debug_level = atoi (opt);
3035 char *rf = g_strdup_printf ("%s.%d", opt, mono_process_current_pid ());
3036 gc_debug_file = fopen (rf, "wb");
3038 gc_debug_file = stderr;
3041 } else if (!strcmp (opt, "print-allowance")) {
3042 debug_print_allowance = TRUE;
3043 } else if (!strcmp (opt, "print-pinning")) {
3044 sgen_pin_stats_enable ();
3045 } else if (!strcmp (opt, "verify-before-allocs")) {
3046 verify_before_allocs = 1;
3047 has_per_allocation_action = TRUE;
3048 } else if (g_str_has_prefix (opt, "verify-before-allocs=")) {
3049 char *arg = strchr (opt, '=') + 1;
3050 verify_before_allocs = atoi (arg);
3051 has_per_allocation_action = TRUE;
3052 } else if (!strcmp (opt, "collect-before-allocs")) {
3053 collect_before_allocs = 1;
3054 has_per_allocation_action = TRUE;
3055 } else if (g_str_has_prefix (opt, "collect-before-allocs=")) {
3056 char *arg = strchr (opt, '=') + 1;
3057 has_per_allocation_action = TRUE;
3058 collect_before_allocs = atoi (arg);
3059 } else if (!strcmp (opt, "verify-before-collections")) {
3060 whole_heap_check_before_collection = TRUE;
3061 } else if (!strcmp (opt, "check-at-minor-collections")) {
3062 consistency_check_at_minor_collection = TRUE;
3063 nursery_clear_policy = CLEAR_AT_GC;
3064 } else if (!strcmp (opt, "mod-union-consistency-check")) {
3065 if (!major_collector.is_concurrent) {
3066 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`mod-union-consistency-check` only works with concurrent major collector.");
3069 mod_union_consistency_check = TRUE;
3070 } else if (!strcmp (opt, "check-mark-bits")) {
3071 check_mark_bits_after_major_collection = TRUE;
3072 } else if (!strcmp (opt, "check-nursery-pinned")) {
3073 check_nursery_objects_pinned = TRUE;
3074 } else if (!strcmp (opt, "clear-at-gc")) {
3075 nursery_clear_policy = CLEAR_AT_GC;
3076 } else if (!strcmp (opt, "clear-nursery-at-gc")) {
3077 nursery_clear_policy = CLEAR_AT_GC;
3078 } else if (!strcmp (opt, "clear-at-tlab-creation")) {
3079 nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
3080 } else if (!strcmp (opt, "debug-clear-at-tlab-creation")) {
3081 nursery_clear_policy = CLEAR_AT_TLAB_CREATION_DEBUG;
3082 } else if (!strcmp (opt, "check-scan-starts")) {
3083 do_scan_starts_check = TRUE;
3084 } else if (!strcmp (opt, "verify-nursery-at-minor-gc")) {
3085 do_verify_nursery = TRUE;
3086 } else if (!strcmp (opt, "check-concurrent")) {
3087 if (!major_collector.is_concurrent) {
3088 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`check-concurrent` only works with concurrent major collectors.");
3091 nursery_clear_policy = CLEAR_AT_GC;
3092 do_concurrent_checks = TRUE;
3093 } else if (!strcmp (opt, "dump-nursery-at-minor-gc")) {
3094 do_dump_nursery_content = TRUE;
3095 } else if (!strcmp (opt, "disable-minor")) {
3096 disable_minor_collections = TRUE;
3097 } else if (!strcmp (opt, "disable-major")) {
3098 disable_major_collections = TRUE;
3099 } else if (g_str_has_prefix (opt, "heap-dump=")) {
3100 char *filename = strchr (opt, '=') + 1;
3101 nursery_clear_policy = CLEAR_AT_GC;
3102 sgen_debug_enable_heap_dump (filename);
3103 } else if (g_str_has_prefix (opt, "binary-protocol=")) {
3104 char *filename = strchr (opt, '=') + 1;
3105 char *colon = strrchr (filename, ':');
3108 if (!mono_gc_parse_environment_string_extract_number (colon + 1, &limit)) {
3109 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring limit.", "Binary protocol file size limit must be an integer.");
3114 binary_protocol_init (filename, (long long)limit);
3115 } else if (!strcmp (opt, "nursery-canaries")) {
3116 do_verify_nursery = TRUE;
3117 enable_nursery_canaries = TRUE;
3118 } else if (!sgen_client_handle_gc_debug (opt)) {
3119 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "Unknown option `%s`.", opt);
3124 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);
3125 fprintf (stderr, "Valid <option>s are:\n");
3126 fprintf (stderr, " collect-before-allocs[=<n>]\n");
3127 fprintf (stderr, " verify-before-allocs[=<n>]\n");
3128 fprintf (stderr, " check-at-minor-collections\n");
3129 fprintf (stderr, " check-mark-bits\n");
3130 fprintf (stderr, " check-nursery-pinned\n");
3131 fprintf (stderr, " verify-before-collections\n");
3132 fprintf (stderr, " verify-nursery-at-minor-gc\n");
3133 fprintf (stderr, " dump-nursery-at-minor-gc\n");
3134 fprintf (stderr, " disable-minor\n");
3135 fprintf (stderr, " disable-major\n");
3136 fprintf (stderr, " check-concurrent\n");
3137 fprintf (stderr, " clear-[nursery-]at-gc\n");
3138 fprintf (stderr, " clear-at-tlab-creation\n");
3139 fprintf (stderr, " debug-clear-at-tlab-creation\n");
3140 fprintf (stderr, " check-scan-starts\n");
3141 fprintf (stderr, " print-allowance\n");
3142 fprintf (stderr, " print-pinning\n");
3143 fprintf (stderr, " heap-dump=<filename>\n");
3144 fprintf (stderr, " binary-protocol=<filename>[:<file-size-limit>]\n");
3145 fprintf (stderr, " nursery-canaries\n");
3146 sgen_client_print_gc_debug_usage ();
3147 fprintf (stderr, "\n");
3149 usage_printed = TRUE;
3155 if (check_mark_bits_after_major_collection)
3156 nursery_clear_policy = CLEAR_AT_GC;
3158 if (major_collector.post_param_init)
3159 major_collector.post_param_init (&major_collector);
3161 if (major_collector.needs_thread_pool)
3162 sgen_workers_init (1);
3164 sgen_memgov_init (max_heap, soft_limit, debug_print_allowance, allowance_ratio, save_target);
3166 memset (&remset, 0, sizeof (remset));
3168 sgen_card_table_init (&remset);
3170 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");
3176 sgen_get_nursery_clear_policy (void)
3178 return nursery_clear_policy;
3184 mono_coop_mutex_lock (&gc_mutex);
3188 sgen_gc_unlock (void)
3190 gboolean try_free = sgen_try_free_some_memory;
3191 sgen_try_free_some_memory = FALSE;
3192 mono_coop_mutex_unlock (&gc_mutex);
3194 mono_thread_hazardous_try_free_some ();
3198 sgen_major_collector_iterate_live_block_ranges (sgen_cardtable_block_callback callback)
3200 major_collector.iterate_live_block_ranges (callback);
3204 sgen_get_major_collector (void)
3206 return &major_collector;
3210 sgen_get_remset (void)
3216 count_cards (long long *major_total, long long *major_marked, long long *los_total, long long *los_marked)
3218 sgen_get_major_collector ()->count_cards (major_total, major_marked);
3219 sgen_los_count_cards (los_total, los_marked);
3222 static gboolean world_is_stopped = FALSE;
3224 /* LOCKING: assumes the GC lock is held */
3226 sgen_stop_world (int generation)
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 stopping a stopped world?");
3232 binary_protocol_world_stopping (generation, sgen_timestamp (), (gpointer) (gsize) mono_native_thread_id_get ());
3234 sgen_client_stop_world (generation);
3236 world_is_stopped = TRUE;
3238 if (binary_protocol_is_heavy_enabled ())
3239 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3240 binary_protocol_world_stopped (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3243 /* LOCKING: assumes the GC lock is held */
3245 sgen_restart_world (int generation, GGTimingInfo *timing)
3247 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3249 SGEN_ASSERT (0, world_is_stopped, "Why are we restarting a running world?");
3251 if (binary_protocol_is_heavy_enabled ())
3252 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3253 binary_protocol_world_restarting (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3255 sgen_client_restart_world (generation, timing);
3257 world_is_stopped = FALSE;
3259 binary_protocol_world_restarted (generation, sgen_timestamp ());
3261 sgen_try_free_some_memory = TRUE;
3263 if (sgen_client_bridge_need_processing ())
3264 sgen_client_bridge_processing_finish (generation);
3266 sgen_memgov_collection_end (generation, timing, timing ? 2 : 0);
3270 sgen_is_world_stopped (void)
3272 return world_is_stopped;
3276 sgen_check_whole_heap_stw (void)
3278 sgen_stop_world (0);
3279 sgen_clear_nursery_fragments ();
3280 sgen_check_whole_heap (FALSE);
3281 sgen_restart_world (0, NULL);
3285 sgen_timestamp (void)
3287 SGEN_TV_DECLARE (timestamp);
3288 SGEN_TV_GETTIME (timestamp);
3289 return SGEN_TV_ELAPSED (sgen_init_timestamp, timestamp);
3292 #endif /* HAVE_SGEN_GC */