2 * sgen-gc.c: Simple generational GC.
5 * Paolo Molaro (lupus@ximian.com)
6 * Rodrigo Kumpera (kumpera@gmail.com)
8 * Copyright 2005-2011 Novell, Inc (http://www.novell.com)
9 * Copyright 2011 Xamarin Inc (http://www.xamarin.com)
11 * Thread start/stop adapted from Boehm's GC:
12 * Copyright (c) 1994 by Xerox Corporation. All rights reserved.
13 * Copyright (c) 1996 by Silicon Graphics. All rights reserved.
14 * Copyright (c) 1998 by Fergus Henderson. All rights reserved.
15 * Copyright (c) 2000-2004 by Hewlett-Packard Company. All rights reserved.
16 * Copyright 2001-2003 Ximian, Inc
17 * Copyright 2003-2010 Novell, Inc.
18 * Copyright 2011 Xamarin, Inc.
19 * Copyright (C) 2012 Xamarin Inc
21 * Licensed under the MIT license. See LICENSE file in the project root for full license information.
23 * Important: allocation provides always zeroed memory, having to do
24 * a memset after allocation is deadly for performance.
25 * Memory usage at startup is currently as follows:
27 * 64 KB internal space
29 * We should provide a small memory config with half the sizes
31 * We currently try to make as few mono assumptions as possible:
32 * 1) 2-word header with no GC pointers in it (first vtable, second to store the
34 * 2) gc descriptor is the second word in the vtable (first word in the class)
35 * 3) 8 byte alignment is the minimum and enough (not true for special structures (SIMD), FIXME)
36 * 4) there is a function to get an object's size and the number of
37 * elements in an array.
38 * 5) we know the special way bounds are allocated for complex arrays
39 * 6) we know about proxies and how to treat them when domains are unloaded
41 * Always try to keep stack usage to a minimum: no recursive behaviour
42 * and no large stack allocs.
44 * General description.
45 * Objects are initially allocated in a nursery using a fast bump-pointer technique.
46 * When the nursery is full we start a nursery collection: this is performed with a
48 * When the old generation is full we start a copying GC of the old generation as well:
49 * this will be changed to mark&sweep with copying when fragmentation becomes to severe
50 * in the future. Maybe we'll even do both during the same collection like IMMIX.
52 * The things that complicate this description are:
53 * *) pinned objects: we can't move them so we need to keep track of them
54 * *) no precise info of the thread stacks and registers: we need to be able to
55 * quickly find the objects that may be referenced conservatively and pin them
56 * (this makes the first issues more important)
57 * *) large objects are too expensive to be dealt with using copying GC: we handle them
58 * with mark/sweep during major collections
59 * *) some objects need to not move even if they are small (interned strings, Type handles):
60 * we use mark/sweep for them, too: they are not allocated in the nursery, but inside
61 * PinnedChunks regions
67 *) we could have a function pointer in MonoClass to implement
68 customized write barriers for value types
70 *) investigate the stuff needed to advance a thread to a GC-safe
71 point (single-stepping, read from unmapped memory etc) and implement it.
72 This would enable us to inline allocations and write barriers, for example,
73 or at least parts of them, like the write barrier checks.
74 We may need this also for handling precise info on stacks, even simple things
75 as having uninitialized data on the stack and having to wait for the prolog
76 to zero it. Not an issue for the last frame that we scan conservatively.
77 We could always not trust the value in the slots anyway.
79 *) modify the jit to save info about references in stack locations:
80 this can be done just for locals as a start, so that at least
81 part of the stack is handled precisely.
83 *) test/fix endianess issues
85 *) Implement a card table as the write barrier instead of remembered
86 sets? Card tables are not easy to implement with our current
87 memory layout. We have several different kinds of major heap
88 objects: Small objects in regular blocks, small objects in pinned
89 chunks and LOS objects. If we just have a pointer we have no way
90 to tell which kind of object it points into, therefore we cannot
91 know where its card table is. The least we have to do to make
92 this happen is to get rid of write barriers for indirect stores.
95 *) Get rid of write barriers for indirect stores. We can do this by
96 telling the GC to wbarrier-register an object once we do an ldloca
97 or ldelema on it, and to unregister it once it's not used anymore
98 (it can only travel downwards on the stack). The problem with
99 unregistering is that it needs to happen eventually no matter
100 what, even if exceptions are thrown, the thread aborts, etc.
101 Rodrigo suggested that we could do only the registering part and
102 let the collector find out (pessimistically) when it's safe to
103 unregister, namely when the stack pointer of the thread that
104 registered the object is higher than it was when the registering
105 happened. This might make for a good first implementation to get
106 some data on performance.
108 *) Some sort of blacklist support? Blacklists is a concept from the
109 Boehm GC: if during a conservative scan we find pointers to an
110 area which we might use as heap, we mark that area as unusable, so
111 pointer retention by random pinning pointers is reduced.
113 *) experiment with max small object size (very small right now - 2kb,
114 because it's tied to the max freelist size)
116 *) add an option to mmap the whole heap in one chunk: it makes for many
117 simplifications in the checks (put the nursery at the top and just use a single
118 check for inclusion/exclusion): the issue this has is that on 32 bit systems it's
119 not flexible (too much of the address space may be used by default or we can't
120 increase the heap as needed) and we'd need a race-free mechanism to return memory
121 back to the system (mprotect(PROT_NONE) will still keep the memory allocated if it
122 was written to, munmap is needed, but the following mmap may not find the same segment
125 *) memzero the major fragments after restarting the world and optionally a smaller
128 *) investigate having fragment zeroing threads
130 *) separate locks for finalization and other minor stuff to reduce
133 *) try a different copying order to improve memory locality
135 *) a thread abort after a store but before the write barrier will
136 prevent the write barrier from executing
138 *) specialized dynamically generated markers/copiers
140 *) Dynamically adjust TLAB size to the number of threads. If we have
141 too many threads that do allocation, we might need smaller TLABs,
142 and we might get better performance with larger TLABs if we only
143 have a handful of threads. We could sum up the space left in all
144 assigned TLABs and if that's more than some percentage of the
145 nursery size, reduce the TLAB size.
147 *) Explore placing unreachable objects on unused nursery memory.
148 Instead of memset'ng a region to zero, place an int[] covering it.
149 A good place to start is add_nursery_frag. The tricky thing here is
150 placing those objects atomically outside of a collection.
152 *) Allocation should use asymmetric Dekker synchronization:
153 http://blogs.oracle.com/dave/resource/Asymmetric-Dekker-Synchronization.txt
154 This should help weak consistency archs.
161 #define _XOPEN_SOURCE
162 #define _DARWIN_C_SOURCE
168 #ifdef HAVE_PTHREAD_H
171 #ifdef HAVE_PTHREAD_NP_H
172 #include <pthread_np.h>
180 #include "mono/sgen/sgen-gc.h"
181 #include "mono/sgen/sgen-cardtable.h"
182 #include "mono/sgen/sgen-protocol.h"
183 #include "mono/sgen/sgen-memory-governor.h"
184 #include "mono/sgen/sgen-hash-table.h"
185 #include "mono/sgen/sgen-cardtable.h"
186 #include "mono/sgen/sgen-pinning.h"
187 #include "mono/sgen/sgen-workers.h"
188 #include "mono/sgen/sgen-client.h"
189 #include "mono/sgen/sgen-pointer-queue.h"
190 #include "mono/sgen/gc-internal-agnostic.h"
191 #include "mono/utils/mono-proclib.h"
192 #include "mono/utils/mono-memory-model.h"
193 #include "mono/utils/hazard-pointer.h"
195 #include <mono/utils/memcheck.h>
197 #undef pthread_create
199 #undef pthread_detach
202 * ######################################################################
203 * ######## Types and constants used by the GC.
204 * ######################################################################
207 /* 0 means not initialized, 1 is initialized, -1 means in progress */
208 static int gc_initialized = 0;
209 /* If set, check if we need to do something every X allocations */
210 gboolean has_per_allocation_action;
211 /* If set, do a heap check every X allocation */
212 guint32 verify_before_allocs = 0;
213 /* If set, do a minor collection before every X allocation */
214 guint32 collect_before_allocs = 0;
215 /* If set, do a whole heap check before each collection */
216 static gboolean whole_heap_check_before_collection = FALSE;
217 /* If set, do a remset consistency check at various opportunities */
218 static gboolean remset_consistency_checks = FALSE;
219 /* If set, do a mod union consistency check before each finishing collection pause */
220 static gboolean mod_union_consistency_check = FALSE;
221 /* If set, check whether mark bits are consistent after major collections */
222 static gboolean check_mark_bits_after_major_collection = FALSE;
223 /* If set, check that all nursery objects are pinned/not pinned, depending on context */
224 static gboolean check_nursery_objects_pinned = FALSE;
225 /* If set, do a few checks when the concurrent collector is used */
226 static gboolean do_concurrent_checks = FALSE;
227 /* If set, do a plausibility check on the scan_starts before and after
229 static gboolean do_scan_starts_check = FALSE;
231 static gboolean disable_minor_collections = FALSE;
232 static gboolean disable_major_collections = FALSE;
233 static gboolean do_verify_nursery = FALSE;
234 static gboolean do_dump_nursery_content = FALSE;
235 static gboolean enable_nursery_canaries = FALSE;
237 static gboolean precleaning_enabled = TRUE;
239 #ifdef HEAVY_STATISTICS
240 guint64 stat_objects_alloced_degraded = 0;
241 guint64 stat_bytes_alloced_degraded = 0;
243 guint64 stat_copy_object_called_nursery = 0;
244 guint64 stat_objects_copied_nursery = 0;
245 guint64 stat_copy_object_called_major = 0;
246 guint64 stat_objects_copied_major = 0;
248 guint64 stat_scan_object_called_nursery = 0;
249 guint64 stat_scan_object_called_major = 0;
251 guint64 stat_slots_allocated_in_vain;
253 guint64 stat_nursery_copy_object_failed_from_space = 0;
254 guint64 stat_nursery_copy_object_failed_forwarded = 0;
255 guint64 stat_nursery_copy_object_failed_pinned = 0;
256 guint64 stat_nursery_copy_object_failed_to_space = 0;
258 static guint64 stat_wbarrier_add_to_global_remset = 0;
259 static guint64 stat_wbarrier_arrayref_copy = 0;
260 static guint64 stat_wbarrier_generic_store = 0;
261 static guint64 stat_wbarrier_generic_store_atomic = 0;
262 static guint64 stat_wbarrier_set_root = 0;
265 static guint64 stat_pinned_objects = 0;
267 static guint64 time_minor_pre_collection_fragment_clear = 0;
268 static guint64 time_minor_pinning = 0;
269 static guint64 time_minor_scan_remsets = 0;
270 static guint64 time_minor_scan_pinned = 0;
271 static guint64 time_minor_scan_roots = 0;
272 static guint64 time_minor_finish_gray_stack = 0;
273 static guint64 time_minor_fragment_creation = 0;
275 static guint64 time_major_pre_collection_fragment_clear = 0;
276 static guint64 time_major_pinning = 0;
277 static guint64 time_major_scan_pinned = 0;
278 static guint64 time_major_scan_roots = 0;
279 static guint64 time_major_scan_mod_union = 0;
280 static guint64 time_major_finish_gray_stack = 0;
281 static guint64 time_major_free_bigobjs = 0;
282 static guint64 time_major_los_sweep = 0;
283 static guint64 time_major_sweep = 0;
284 static guint64 time_major_fragment_creation = 0;
286 static guint64 time_max = 0;
288 static SGEN_TV_DECLARE (time_major_conc_collection_start);
289 static SGEN_TV_DECLARE (time_major_conc_collection_end);
291 static SGEN_TV_DECLARE (last_minor_collection_start_tv);
292 static SGEN_TV_DECLARE (last_minor_collection_end_tv);
294 int gc_debug_level = 0;
299 mono_gc_flush_info (void)
301 fflush (gc_debug_file);
305 #define TV_DECLARE SGEN_TV_DECLARE
306 #define TV_GETTIME SGEN_TV_GETTIME
307 #define TV_ELAPSED SGEN_TV_ELAPSED
309 static SGEN_TV_DECLARE (sgen_init_timestamp);
311 NurseryClearPolicy nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
313 #define object_is_forwarded SGEN_OBJECT_IS_FORWARDED
314 #define object_is_pinned SGEN_OBJECT_IS_PINNED
315 #define pin_object SGEN_PIN_OBJECT
317 #define ptr_in_nursery sgen_ptr_in_nursery
319 #define LOAD_VTABLE SGEN_LOAD_VTABLE
322 nursery_canaries_enabled (void)
324 return enable_nursery_canaries;
327 #define safe_object_get_size sgen_safe_object_get_size
329 #if defined(HAVE_CONC_GC_AS_DEFAULT)
330 /* Use concurrent major on deskstop platforms */
331 #define DEFAULT_MAJOR_INIT sgen_marksweep_conc_init
332 #define DEFAULT_MAJOR_NAME "marksweep-conc"
334 #define DEFAULT_MAJOR_INIT sgen_marksweep_init
335 #define DEFAULT_MAJOR_NAME "marksweep"
339 * ######################################################################
340 * ######## Global data.
341 * ######################################################################
343 MonoCoopMutex gc_mutex;
345 #define SCAN_START_SIZE SGEN_SCAN_START_SIZE
347 size_t degraded_mode = 0;
349 static mword bytes_pinned_from_failed_allocation = 0;
351 GCMemSection *nursery_section = NULL;
352 static volatile mword lowest_heap_address = ~(mword)0;
353 static volatile mword highest_heap_address = 0;
355 MonoCoopMutex sgen_interruption_mutex;
357 int current_collection_generation = -1;
358 static volatile gboolean concurrent_collection_in_progress = FALSE;
360 /* objects that are ready to be finalized */
361 static SgenPointerQueue fin_ready_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
362 static SgenPointerQueue critical_fin_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
364 /* registered roots: the key to the hash is the root start address */
366 * Different kinds of roots are kept separate to speed up pin_from_roots () for example.
368 SgenHashTable roots_hash [ROOT_TYPE_NUM] = {
369 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
370 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
371 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL)
373 static mword roots_size = 0; /* amount of memory in the root set */
375 /* The size of a TLAB */
376 /* The bigger the value, the less often we have to go to the slow path to allocate a new
377 * one, but the more space is wasted by threads not allocating much memory.
379 * FIXME: Make this self-tuning for each thread.
381 guint32 tlab_size = (1024 * 4);
383 #define MAX_SMALL_OBJ_SIZE SGEN_MAX_SMALL_OBJ_SIZE
385 #define ALLOC_ALIGN SGEN_ALLOC_ALIGN
387 #define ALIGN_UP SGEN_ALIGN_UP
389 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
390 MonoNativeThreadId main_gc_thread = NULL;
393 /*Object was pinned during the current collection*/
394 static mword objects_pinned;
397 * ######################################################################
398 * ######## Macros and function declarations.
399 * ######################################################################
402 /* forward declarations */
403 static void scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx);
405 static void pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx);
406 static void finish_gray_stack (int generation, ScanCopyContext ctx);
409 SgenMajorCollector major_collector;
410 SgenMinorCollector sgen_minor_collector;
412 static SgenRememberedSet remset;
415 * The gray queue a worker job must use. If we're not parallel or
416 * concurrent, we use the main gray queue.
418 static SgenGrayQueue*
419 sgen_workers_get_job_gray_queue (WorkerData *worker_data, SgenGrayQueue *default_gray_queue)
422 return &worker_data->private_gray_queue;
423 SGEN_ASSERT (0, default_gray_queue, "Why don't we have a default gray queue when we're not running in a worker thread?");
424 return default_gray_queue;
428 gray_queue_enable_redirect (SgenGrayQueue *queue)
430 SGEN_ASSERT (0, concurrent_collection_in_progress, "Where are we redirecting the gray queue to, without a concurrent collection?");
432 sgen_gray_queue_set_alloc_prepare (queue, sgen_workers_take_from_queue_and_awake);
433 sgen_workers_take_from_queue_and_awake (queue);
437 sgen_scan_area_with_callback (char *start, char *end, IterateObjectCallbackFunc callback, void *data, gboolean allow_flags, gboolean fail_on_canaries)
439 while (start < end) {
443 if (!*(void**)start) {
444 start += sizeof (void*); /* should be ALLOC_ALIGN, really */
449 if (!(obj = (char *)SGEN_OBJECT_IS_FORWARDED (start)))
455 if (!sgen_client_object_is_array_fill ((GCObject*)obj)) {
456 CHECK_CANARY_FOR_OBJECT ((GCObject*)obj, fail_on_canaries);
457 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
458 callback ((GCObject*)obj, size, data);
459 CANARIFY_SIZE (size);
461 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
469 * sgen_add_to_global_remset:
471 * The global remset contains locations which point into newspace after
472 * a minor collection. This can happen if the objects they point to are pinned.
474 * LOCKING: If called from a parallel collector, the global remset
475 * lock must be held. For serial collectors that is not necessary.
478 sgen_add_to_global_remset (gpointer ptr, GCObject *obj)
480 SGEN_ASSERT (5, sgen_ptr_in_nursery (obj), "Target pointer of global remset must be in the nursery");
482 HEAVY_STAT (++stat_wbarrier_add_to_global_remset);
484 if (!major_collector.is_concurrent) {
485 SGEN_ASSERT (5, current_collection_generation != -1, "Global remsets can only be added during collections");
487 if (current_collection_generation == -1)
488 SGEN_ASSERT (5, sgen_concurrent_collection_in_progress (), "Global remsets outside of collection pauses can only be added by the concurrent collector");
491 if (!object_is_pinned (obj))
492 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");
493 else if (sgen_cement_lookup_or_register (obj))
496 remset.record_pointer (ptr);
498 sgen_pin_stats_register_global_remset (obj);
500 SGEN_LOG (8, "Adding global remset for %p", ptr);
501 binary_protocol_global_remset (ptr, obj, (gpointer)SGEN_LOAD_VTABLE (obj));
505 * sgen_drain_gray_stack:
507 * Scan objects in the gray stack until the stack is empty. This should be called
508 * frequently after each object is copied, to achieve better locality and cache
513 sgen_drain_gray_stack (ScanCopyContext ctx)
515 ScanObjectFunc scan_func = ctx.ops->scan_object;
516 SgenGrayQueue *queue = ctx.queue;
518 if (ctx.ops->drain_gray_stack)
519 return ctx.ops->drain_gray_stack (queue);
524 GRAY_OBJECT_DEQUEUE (queue, &obj, &desc);
527 SGEN_LOG (9, "Precise gray object scan %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
528 scan_func (obj, desc, queue);
534 * Addresses in the pin queue are already sorted. This function finds
535 * the object header for each address and pins the object. The
536 * addresses must be inside the nursery section. The (start of the)
537 * address array is overwritten with the addresses of the actually
538 * pinned objects. Return the number of pinned objects.
541 pin_objects_from_nursery_pin_queue (gboolean do_scan_objects, ScanCopyContext ctx)
543 GCMemSection *section = nursery_section;
544 void **start = sgen_pinning_get_entry (section->pin_queue_first_entry);
545 void **end = sgen_pinning_get_entry (section->pin_queue_last_entry);
546 void *start_nursery = section->data;
547 void *end_nursery = section->next_data;
552 void *pinning_front = start_nursery;
554 void **definitely_pinned = start;
555 ScanObjectFunc scan_func = ctx.ops->scan_object;
556 SgenGrayQueue *queue = ctx.queue;
558 sgen_nursery_allocator_prepare_for_pinning ();
560 while (start < end) {
561 GCObject *obj_to_pin = NULL;
562 size_t obj_to_pin_size = 0;
567 SGEN_ASSERT (0, addr >= start_nursery && addr < end_nursery, "Potential pinning address out of range");
568 SGEN_ASSERT (0, addr >= last, "Pin queue not sorted");
575 SGEN_LOG (5, "Considering pinning addr %p", addr);
576 /* We've already processed everything up to pinning_front. */
577 if (addr < pinning_front) {
583 * Find the closest scan start <= addr. We might search backward in the
584 * scan_starts array because entries might be NULL. In the worst case we
585 * start at start_nursery.
587 idx = ((char*)addr - (char*)section->data) / SCAN_START_SIZE;
588 SGEN_ASSERT (0, idx < section->num_scan_start, "Scan start index out of range");
589 search_start = (void*)section->scan_starts [idx];
590 if (!search_start || search_start > addr) {
593 search_start = section->scan_starts [idx];
594 if (search_start && search_start <= addr)
597 if (!search_start || search_start > addr)
598 search_start = start_nursery;
602 * If the pinning front is closer than the scan start we found, start
603 * searching at the front.
605 if (search_start < pinning_front)
606 search_start = pinning_front;
609 * Now addr should be in an object a short distance from search_start.
611 * search_start must point to zeroed mem or point to an object.
614 size_t obj_size, canarified_obj_size;
617 if (!*(void**)search_start) {
618 search_start = (void*)ALIGN_UP ((mword)search_start + sizeof (gpointer));
619 /* The loop condition makes sure we don't overrun addr. */
623 canarified_obj_size = obj_size = ALIGN_UP (safe_object_get_size ((GCObject*)search_start));
626 * Filler arrays are marked by an invalid sync word. We don't
627 * consider them for pinning. They are not delimited by canaries,
630 if (!sgen_client_object_is_array_fill ((GCObject*)search_start)) {
631 CHECK_CANARY_FOR_OBJECT (search_start, TRUE);
632 CANARIFY_SIZE (canarified_obj_size);
634 if (addr >= search_start && (char*)addr < (char*)search_start + obj_size) {
635 /* This is the object we're looking for. */
636 obj_to_pin = (GCObject*)search_start;
637 obj_to_pin_size = canarified_obj_size;
642 /* Skip to the next object */
643 search_start = (void*)((char*)search_start + canarified_obj_size);
644 } while (search_start <= addr);
646 /* We've searched past the address we were looking for. */
648 pinning_front = search_start;
649 goto next_pin_queue_entry;
653 * We've found an object to pin. It might still be a dummy array, but we
654 * can advance the pinning front in any case.
656 pinning_front = (char*)obj_to_pin + obj_to_pin_size;
659 * If this is a dummy array marking the beginning of a nursery
660 * fragment, we don't pin it.
662 if (sgen_client_object_is_array_fill (obj_to_pin))
663 goto next_pin_queue_entry;
666 * Finally - pin the object!
668 desc = sgen_obj_get_descriptor_safe (obj_to_pin);
669 if (do_scan_objects) {
670 scan_func (obj_to_pin, desc, queue);
672 SGEN_LOG (4, "Pinned object %p, vtable %p (%s), count %d\n",
673 obj_to_pin, *(void**)obj_to_pin, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj_to_pin)), count);
674 binary_protocol_pin (obj_to_pin,
675 (gpointer)LOAD_VTABLE (obj_to_pin),
676 safe_object_get_size (obj_to_pin));
678 pin_object (obj_to_pin);
679 GRAY_OBJECT_ENQUEUE (queue, obj_to_pin, desc);
680 sgen_pin_stats_register_object (obj_to_pin, GENERATION_NURSERY);
681 definitely_pinned [count] = obj_to_pin;
684 if (concurrent_collection_in_progress)
685 sgen_pinning_register_pinned_in_nursery (obj_to_pin);
687 next_pin_queue_entry:
691 sgen_client_nursery_objects_pinned (definitely_pinned, count);
692 stat_pinned_objects += count;
697 pin_objects_in_nursery (gboolean do_scan_objects, ScanCopyContext ctx)
701 if (nursery_section->pin_queue_first_entry == nursery_section->pin_queue_last_entry)
704 reduced_to = pin_objects_from_nursery_pin_queue (do_scan_objects, ctx);
705 nursery_section->pin_queue_last_entry = nursery_section->pin_queue_first_entry + reduced_to;
709 * This function is only ever called (via `collector_pin_object()` in `sgen-copy-object.h`)
710 * when we can't promote an object because we're out of memory.
713 sgen_pin_object (GCObject *object, SgenGrayQueue *queue)
715 SGEN_ASSERT (0, sgen_ptr_in_nursery (object), "We're only supposed to use this for pinning nursery objects when out of memory.");
718 * All pinned objects are assumed to have been staged, so we need to stage as well.
719 * Also, the count of staged objects shows that "late pinning" happened.
721 sgen_pin_stage_ptr (object);
723 SGEN_PIN_OBJECT (object);
724 binary_protocol_pin (object, (gpointer)LOAD_VTABLE (object), safe_object_get_size (object));
727 sgen_pin_stats_register_object (object, GENERATION_NURSERY);
729 GRAY_OBJECT_ENQUEUE (queue, object, sgen_obj_get_descriptor_safe (object));
732 /* Sort the addresses in array in increasing order.
733 * Done using a by-the book heap sort. Which has decent and stable performance, is pretty cache efficient.
736 sgen_sort_addresses (void **array, size_t size)
741 for (i = 1; i < size; ++i) {
744 size_t parent = (child - 1) / 2;
746 if (array [parent] >= array [child])
749 tmp = array [parent];
750 array [parent] = array [child];
757 for (i = size - 1; i > 0; --i) {
760 array [i] = array [0];
766 while (root * 2 + 1 <= end) {
767 size_t child = root * 2 + 1;
769 if (child < end && array [child] < array [child + 1])
771 if (array [root] >= array [child])
775 array [root] = array [child];
784 * Scan the memory between start and end and queue values which could be pointers
785 * to the area between start_nursery and end_nursery for later consideration.
786 * Typically used for thread stacks.
789 sgen_conservatively_pin_objects_from (void **start, void **end, void *start_nursery, void *end_nursery, int pin_type)
793 SGEN_ASSERT (0, ((mword)start & (SIZEOF_VOID_P - 1)) == 0, "Why are we scanning for references in unaligned memory ?");
795 #if defined(VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE) && !defined(_WIN64)
796 VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE (start, (char*)end - (char*)start);
799 while (start < end) {
801 * *start can point to the middle of an object
802 * note: should we handle pointing at the end of an object?
803 * pinning in C# code disallows pointing at the end of an object
804 * but there is some small chance that an optimizing C compiler
805 * may keep the only reference to an object by pointing
806 * at the end of it. We ignore this small chance for now.
807 * Pointers to the end of an object are indistinguishable
808 * from pointers to the start of the next object in memory
809 * so if we allow that we'd need to pin two objects...
810 * We queue the pointer in an array, the
811 * array will then be sorted and uniqued. This way
812 * we can coalesce several pinning pointers and it should
813 * be faster since we'd do a memory scan with increasing
814 * addresses. Note: we can align the address to the allocation
815 * alignment, so the unique process is more effective.
817 mword addr = (mword)*start;
818 addr &= ~(ALLOC_ALIGN - 1);
819 if (addr >= (mword)start_nursery && addr < (mword)end_nursery) {
820 SGEN_LOG (6, "Pinning address %p from %p", (void*)addr, start);
821 sgen_pin_stage_ptr ((void*)addr);
822 binary_protocol_pin_stage (start, (void*)addr);
823 sgen_pin_stats_register_address ((char*)addr, pin_type);
829 SGEN_LOG (7, "found %d potential pinned heap pointers", count);
833 * The first thing we do in a collection is to identify pinned objects.
834 * This function considers all the areas of memory that need to be
835 * conservatively scanned.
838 pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx)
842 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);
843 /* objects pinned from the API are inside these roots */
844 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_PINNED], void **, start_root, RootRecord *, root) {
845 SGEN_LOG (6, "Pinned roots %p-%p", start_root, root->end_root);
846 sgen_conservatively_pin_objects_from (start_root, (void**)root->end_root, start_nursery, end_nursery, PIN_TYPE_OTHER);
847 } SGEN_HASH_TABLE_FOREACH_END;
848 /* now deal with the thread stacks
849 * in the future we should be able to conservatively scan only:
850 * *) the cpu registers
851 * *) the unmanaged stack frames
852 * *) the _last_ managed stack frame
853 * *) pointers slots in managed frames
855 sgen_client_scan_thread_data (start_nursery, end_nursery, FALSE, ctx);
859 single_arg_user_copy_or_mark (GCObject **obj, void *gc_data)
861 ScanCopyContext *ctx = (ScanCopyContext *)gc_data;
862 ctx->ops->copy_or_mark_object (obj, ctx->queue);
866 * The memory area from start_root to end_root contains pointers to objects.
867 * Their position is precisely described by @desc (this means that the pointer
868 * can be either NULL or the pointer to the start of an object).
869 * This functions copies them to to_space updates them.
871 * This function is not thread-safe!
874 precisely_scan_objects_from (void** start_root, void** end_root, char* n_start, char *n_end, SgenDescriptor desc, ScanCopyContext ctx)
876 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
877 SgenGrayQueue *queue = ctx.queue;
879 switch (desc & ROOT_DESC_TYPE_MASK) {
880 case ROOT_DESC_BITMAP:
881 desc >>= ROOT_DESC_TYPE_SHIFT;
883 if ((desc & 1) && *start_root) {
884 copy_func ((GCObject**)start_root, queue);
885 SGEN_LOG (9, "Overwrote root at %p with %p", start_root, *start_root);
891 case ROOT_DESC_COMPLEX: {
892 gsize *bitmap_data = (gsize *)sgen_get_complex_descriptor_bitmap (desc);
893 gsize bwords = (*bitmap_data) - 1;
894 void **start_run = start_root;
896 while (bwords-- > 0) {
897 gsize bmap = *bitmap_data++;
898 void **objptr = start_run;
900 if ((bmap & 1) && *objptr) {
901 copy_func ((GCObject**)objptr, queue);
902 SGEN_LOG (9, "Overwrote root at %p with %p", objptr, *objptr);
907 start_run += GC_BITS_PER_WORD;
911 case ROOT_DESC_USER: {
912 SgenUserRootMarkFunc marker = sgen_get_user_descriptor_func (desc);
913 marker (start_root, single_arg_user_copy_or_mark, &ctx);
916 case ROOT_DESC_RUN_LEN:
917 g_assert_not_reached ();
919 g_assert_not_reached ();
924 reset_heap_boundaries (void)
926 lowest_heap_address = ~(mword)0;
927 highest_heap_address = 0;
931 sgen_update_heap_boundaries (mword low, mword high)
936 old = lowest_heap_address;
939 } while (SGEN_CAS_PTR ((gpointer*)&lowest_heap_address, (gpointer)low, (gpointer)old) != (gpointer)old);
942 old = highest_heap_address;
945 } while (SGEN_CAS_PTR ((gpointer*)&highest_heap_address, (gpointer)high, (gpointer)old) != (gpointer)old);
949 * Allocate and setup the data structures needed to be able to allocate objects
950 * in the nursery. The nursery is stored in nursery_section.
955 GCMemSection *section;
962 SGEN_LOG (2, "Allocating nursery size: %zu", (size_t)sgen_nursery_size);
963 /* later we will alloc a larger area for the nursery but only activate
964 * what we need. The rest will be used as expansion if we have too many pinned
965 * objects in the existing nursery.
967 /* FIXME: handle OOM */
968 section = (GCMemSection *)sgen_alloc_internal (INTERNAL_MEM_SECTION);
970 alloc_size = sgen_nursery_size;
972 /* If there isn't enough space even for the nursery we should simply abort. */
973 g_assert (sgen_memgov_try_alloc_space (alloc_size, SPACE_NURSERY));
975 data = (char *)major_collector.alloc_heap (alloc_size, alloc_size, DEFAULT_NURSERY_BITS);
976 sgen_update_heap_boundaries ((mword)data, (mword)(data + sgen_nursery_size));
977 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 ());
978 section->data = section->next_data = data;
979 section->size = alloc_size;
980 section->end_data = data + sgen_nursery_size;
981 scan_starts = (alloc_size + SCAN_START_SIZE - 1) / SCAN_START_SIZE;
982 section->scan_starts = (char **)sgen_alloc_internal_dynamic (sizeof (char*) * scan_starts, INTERNAL_MEM_SCAN_STARTS, TRUE);
983 section->num_scan_start = scan_starts;
985 nursery_section = section;
987 sgen_nursery_allocator_set_nursery_bounds (data, data + sgen_nursery_size);
991 mono_gc_get_logfile (void)
993 return gc_debug_file;
997 scan_finalizer_entries (SgenPointerQueue *fin_queue, ScanCopyContext ctx)
999 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
1000 SgenGrayQueue *queue = ctx.queue;
1003 for (i = 0; i < fin_queue->next_slot; ++i) {
1004 GCObject *obj = (GCObject *)fin_queue->data [i];
1007 SGEN_LOG (5, "Scan of fin ready object: %p (%s)\n", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
1008 copy_func ((GCObject**)&fin_queue->data [i], queue);
1013 generation_name (int generation)
1015 switch (generation) {
1016 case GENERATION_NURSERY: return "nursery";
1017 case GENERATION_OLD: return "old";
1018 default: g_assert_not_reached ();
1023 sgen_generation_name (int generation)
1025 return generation_name (generation);
1029 finish_gray_stack (int generation, ScanCopyContext ctx)
1033 int done_with_ephemerons, ephemeron_rounds = 0;
1034 char *start_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_start () : NULL;
1035 char *end_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_end () : (char*)-1;
1036 SgenGrayQueue *queue = ctx.queue;
1038 binary_protocol_finish_gray_stack_start (sgen_timestamp (), generation);
1040 * We copied all the reachable objects. Now it's the time to copy
1041 * the objects that were not referenced by the roots, but by the copied objects.
1042 * we built a stack of objects pointed to by gray_start: they are
1043 * additional roots and we may add more items as we go.
1044 * We loop until gray_start == gray_objects which means no more objects have
1045 * been added. Note this is iterative: no recursion is involved.
1046 * We need to walk the LO list as well in search of marked big objects
1047 * (use a flag since this is needed only on major collections). We need to loop
1048 * here as well, so keep a counter of marked LO (increasing it in copy_object).
1049 * To achieve better cache locality and cache usage, we drain the gray stack
1050 * frequently, after each object is copied, and just finish the work here.
1052 sgen_drain_gray_stack (ctx);
1054 SGEN_LOG (2, "%s generation done", generation_name (generation));
1057 Reset bridge data, we might have lingering data from a previous collection if this is a major
1058 collection trigged by minor overflow.
1060 We must reset the gathered bridges since their original block might be evacuated due to major
1061 fragmentation in the meanwhile and the bridge code should not have to deal with that.
1063 if (sgen_client_bridge_need_processing ())
1064 sgen_client_bridge_reset_data ();
1067 * Mark all strong toggleref objects. This must be done before we walk ephemerons or finalizers
1068 * to ensure they see the full set of live objects.
1070 sgen_client_mark_togglerefs (start_addr, end_addr, ctx);
1073 * Walk the ephemeron tables marking all values with reachable keys. This must be completely done
1074 * before processing finalizable objects and non-tracking weak links to avoid finalizing/clearing
1075 * objects that are in fact reachable.
1077 done_with_ephemerons = 0;
1079 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1080 sgen_drain_gray_stack (ctx);
1082 } while (!done_with_ephemerons);
1084 if (sgen_client_bridge_need_processing ()) {
1085 /*Make sure the gray stack is empty before we process bridge objects so we get liveness right*/
1086 sgen_drain_gray_stack (ctx);
1087 sgen_collect_bridge_objects (generation, ctx);
1088 if (generation == GENERATION_OLD)
1089 sgen_collect_bridge_objects (GENERATION_NURSERY, ctx);
1092 Do the first bridge step here, as the collector liveness state will become useless after that.
1094 An important optimization is to only proccess the possibly dead part of the object graph and skip
1095 over all live objects as we transitively know everything they point must be alive too.
1097 The above invariant is completely wrong if we let the gray queue be drained and mark/copy everything.
1099 This has the unfortunate side effect of making overflow collections perform the first step twice, but
1100 given we now have heuristics that perform major GC in anticipation of minor overflows this should not
1103 sgen_client_bridge_processing_stw_step ();
1107 Make sure we drain the gray stack before processing disappearing links and finalizers.
1108 If we don't make sure it is empty we might wrongly see a live object as dead.
1110 sgen_drain_gray_stack (ctx);
1113 We must clear weak links that don't track resurrection before processing object ready for
1114 finalization so they can be cleared before that.
1116 sgen_null_link_in_range (generation, ctx, FALSE);
1117 if (generation == GENERATION_OLD)
1118 sgen_null_link_in_range (GENERATION_NURSERY, ctx, FALSE);
1121 /* walk the finalization queue and move also the objects that need to be
1122 * finalized: use the finalized objects as new roots so the objects they depend
1123 * on are also not reclaimed. As with the roots above, only objects in the nursery
1124 * are marked/copied.
1126 sgen_finalize_in_range (generation, ctx);
1127 if (generation == GENERATION_OLD)
1128 sgen_finalize_in_range (GENERATION_NURSERY, ctx);
1129 /* drain the new stack that might have been created */
1130 SGEN_LOG (6, "Precise scan of gray area post fin");
1131 sgen_drain_gray_stack (ctx);
1134 * This must be done again after processing finalizable objects since CWL slots are cleared only after the key is finalized.
1136 done_with_ephemerons = 0;
1138 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1139 sgen_drain_gray_stack (ctx);
1141 } while (!done_with_ephemerons);
1143 sgen_client_clear_unreachable_ephemerons (ctx);
1146 * We clear togglerefs only after all possible chances of revival are done.
1147 * This is semantically more inline with what users expect and it allows for
1148 * user finalizers to correctly interact with TR objects.
1150 sgen_client_clear_togglerefs (start_addr, end_addr, ctx);
1153 SGEN_LOG (2, "Finalize queue handling scan for %s generation: %lld usecs %d ephemeron rounds", generation_name (generation), (long long)TV_ELAPSED (atv, btv), ephemeron_rounds);
1156 * handle disappearing links
1157 * Note we do this after checking the finalization queue because if an object
1158 * survives (at least long enough to be finalized) we don't clear the link.
1159 * This also deals with a possible issue with the monitor reclamation: with the Boehm
1160 * GC a finalized object my lose the monitor because it is cleared before the finalizer is
1163 g_assert (sgen_gray_object_queue_is_empty (queue));
1165 sgen_null_link_in_range (generation, ctx, TRUE);
1166 if (generation == GENERATION_OLD)
1167 sgen_null_link_in_range (GENERATION_NURSERY, ctx, TRUE);
1168 if (sgen_gray_object_queue_is_empty (queue))
1170 sgen_drain_gray_stack (ctx);
1173 g_assert (sgen_gray_object_queue_is_empty (queue));
1175 sgen_gray_object_queue_trim_free_list (queue);
1176 binary_protocol_finish_gray_stack_end (sgen_timestamp (), generation);
1180 sgen_check_section_scan_starts (GCMemSection *section)
1183 for (i = 0; i < section->num_scan_start; ++i) {
1184 if (section->scan_starts [i]) {
1185 mword size = safe_object_get_size ((GCObject*) section->scan_starts [i]);
1186 SGEN_ASSERT (0, size >= SGEN_CLIENT_MINIMUM_OBJECT_SIZE && size <= MAX_SMALL_OBJ_SIZE, "Weird object size at scan starts.");
1192 check_scan_starts (void)
1194 if (!do_scan_starts_check)
1196 sgen_check_section_scan_starts (nursery_section);
1197 major_collector.check_scan_starts ();
1201 scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx)
1205 SGEN_HASH_TABLE_FOREACH (&roots_hash [root_type], void **, start_root, RootRecord *, root) {
1206 SGEN_LOG (6, "Precise root scan %p-%p (desc: %p)", start_root, root->end_root, (void*)root->root_desc);
1207 precisely_scan_objects_from (start_root, (void**)root->end_root, addr_start, addr_end, root->root_desc, ctx);
1208 } SGEN_HASH_TABLE_FOREACH_END;
1214 static gboolean inited = FALSE;
1219 mono_counters_register ("Collection max time", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME | MONO_COUNTER_MONOTONIC, &time_max);
1221 mono_counters_register ("Minor fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pre_collection_fragment_clear);
1222 mono_counters_register ("Minor pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pinning);
1223 mono_counters_register ("Minor scan remembered set", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_remsets);
1224 mono_counters_register ("Minor scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_pinned);
1225 mono_counters_register ("Minor scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_roots);
1226 mono_counters_register ("Minor fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_fragment_creation);
1228 mono_counters_register ("Major fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pre_collection_fragment_clear);
1229 mono_counters_register ("Major pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pinning);
1230 mono_counters_register ("Major scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_pinned);
1231 mono_counters_register ("Major scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_roots);
1232 mono_counters_register ("Major scan mod union", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_mod_union);
1233 mono_counters_register ("Major finish gray stack", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_finish_gray_stack);
1234 mono_counters_register ("Major free big objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_free_bigobjs);
1235 mono_counters_register ("Major LOS sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_los_sweep);
1236 mono_counters_register ("Major sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_sweep);
1237 mono_counters_register ("Major fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_fragment_creation);
1239 mono_counters_register ("Number of pinned objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_pinned_objects);
1241 #ifdef HEAVY_STATISTICS
1242 mono_counters_register ("WBarrier remember pointer", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_add_to_global_remset);
1243 mono_counters_register ("WBarrier arrayref copy", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_arrayref_copy);
1244 mono_counters_register ("WBarrier generic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store);
1245 mono_counters_register ("WBarrier generic atomic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store_atomic);
1246 mono_counters_register ("WBarrier set root", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_set_root);
1248 mono_counters_register ("# objects allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_alloced_degraded);
1249 mono_counters_register ("bytes allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_bytes_alloced_degraded);
1251 mono_counters_register ("# copy_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_nursery);
1252 mono_counters_register ("# objects copied (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_nursery);
1253 mono_counters_register ("# copy_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_major);
1254 mono_counters_register ("# objects copied (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_major);
1256 mono_counters_register ("# scan_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_nursery);
1257 mono_counters_register ("# scan_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_major);
1259 mono_counters_register ("Slots allocated in vain", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_slots_allocated_in_vain);
1261 mono_counters_register ("# nursery copy_object() failed from space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_from_space);
1262 mono_counters_register ("# nursery copy_object() failed forwarded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_forwarded);
1263 mono_counters_register ("# nursery copy_object() failed pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_pinned);
1264 mono_counters_register ("# nursery copy_object() failed to space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_to_space);
1266 sgen_nursery_allocator_init_heavy_stats ();
1274 reset_pinned_from_failed_allocation (void)
1276 bytes_pinned_from_failed_allocation = 0;
1280 sgen_set_pinned_from_failed_allocation (mword objsize)
1282 bytes_pinned_from_failed_allocation += objsize;
1286 sgen_collection_is_concurrent (void)
1288 switch (current_collection_generation) {
1289 case GENERATION_NURSERY:
1291 case GENERATION_OLD:
1292 return concurrent_collection_in_progress;
1294 g_error ("Invalid current generation %d", current_collection_generation);
1300 sgen_concurrent_collection_in_progress (void)
1302 return concurrent_collection_in_progress;
1306 SgenThreadPoolJob job;
1307 SgenObjectOperations *ops;
1308 SgenGrayQueue *gc_thread_gray_queue;
1311 static ScanCopyContext
1312 scan_copy_context_for_scan_job (void *worker_data_untyped, ScanJob *job)
1314 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1316 return CONTEXT_FROM_OBJECT_OPERATIONS (job->ops, sgen_workers_get_job_gray_queue (worker_data, job->gc_thread_gray_queue));
1320 job_remembered_set_scan (void *worker_data_untyped, SgenThreadPoolJob *job)
1322 remset.scan_remsets (scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job));
1330 } ScanFromRegisteredRootsJob;
1333 job_scan_from_registered_roots (void *worker_data_untyped, SgenThreadPoolJob *job)
1335 ScanFromRegisteredRootsJob *job_data = (ScanFromRegisteredRootsJob*)job;
1336 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1338 scan_from_registered_roots (job_data->heap_start, job_data->heap_end, job_data->root_type, ctx);
1345 } ScanThreadDataJob;
1348 job_scan_thread_data (void *worker_data_untyped, SgenThreadPoolJob *job)
1350 ScanThreadDataJob *job_data = (ScanThreadDataJob*)job;
1351 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1353 sgen_client_scan_thread_data (job_data->heap_start, job_data->heap_end, TRUE, ctx);
1358 SgenPointerQueue *queue;
1359 } ScanFinalizerEntriesJob;
1362 job_scan_finalizer_entries (void *worker_data_untyped, SgenThreadPoolJob *job)
1364 ScanFinalizerEntriesJob *job_data = (ScanFinalizerEntriesJob*)job;
1365 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1367 scan_finalizer_entries (job_data->queue, ctx);
1371 job_scan_major_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1373 ScanJob *job_data = (ScanJob*)job;
1374 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, job_data);
1376 g_assert (concurrent_collection_in_progress);
1377 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx);
1381 job_scan_los_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1383 ScanJob *job_data = (ScanJob*)job;
1384 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, job_data);
1386 g_assert (concurrent_collection_in_progress);
1387 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx);
1391 job_mod_union_preclean (void *worker_data_untyped, SgenThreadPoolJob *job)
1393 ScanJob *job_data = (ScanJob*)job;
1394 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, job_data);
1396 g_assert (concurrent_collection_in_progress);
1398 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx);
1399 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx);
1401 sgen_scan_pin_queue_objects (ctx);
1405 init_gray_queue (SgenGrayQueue *gc_thread_gray_queue, gboolean use_workers)
1408 sgen_workers_init_distribute_gray_queue ();
1409 sgen_gray_object_queue_init (gc_thread_gray_queue, NULL, TRUE);
1413 enqueue_scan_from_roots_jobs (SgenGrayQueue *gc_thread_gray_queue, char *heap_start, char *heap_end, SgenObjectOperations *ops, gboolean enqueue)
1415 ScanFromRegisteredRootsJob *scrrj;
1416 ScanThreadDataJob *stdj;
1417 ScanFinalizerEntriesJob *sfej;
1419 /* registered roots, this includes static fields */
1421 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots normal", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1422 scrrj->scan_job.ops = ops;
1423 scrrj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1424 scrrj->heap_start = heap_start;
1425 scrrj->heap_end = heap_end;
1426 scrrj->root_type = ROOT_TYPE_NORMAL;
1427 sgen_workers_enqueue_job (&scrrj->scan_job.job, enqueue);
1429 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots wbarrier", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1430 scrrj->scan_job.ops = ops;
1431 scrrj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1432 scrrj->heap_start = heap_start;
1433 scrrj->heap_end = heap_end;
1434 scrrj->root_type = ROOT_TYPE_WBARRIER;
1435 sgen_workers_enqueue_job (&scrrj->scan_job.job, enqueue);
1439 stdj = (ScanThreadDataJob*)sgen_thread_pool_job_alloc ("scan thread data", job_scan_thread_data, sizeof (ScanThreadDataJob));
1440 stdj->scan_job.ops = ops;
1441 stdj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1442 stdj->heap_start = heap_start;
1443 stdj->heap_end = heap_end;
1444 sgen_workers_enqueue_job (&stdj->scan_job.job, enqueue);
1446 /* Scan the list of objects ready for finalization. */
1448 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1449 sfej->scan_job.ops = ops;
1450 sfej->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1451 sfej->queue = &fin_ready_queue;
1452 sgen_workers_enqueue_job (&sfej->scan_job.job, enqueue);
1454 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan critical finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1455 sfej->scan_job.ops = ops;
1456 sfej->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1457 sfej->queue = &critical_fin_queue;
1458 sgen_workers_enqueue_job (&sfej->scan_job.job, enqueue);
1462 * Perform a nursery collection.
1464 * Return whether any objects were late-pinned due to being out of memory.
1467 collect_nursery (const char *reason, gboolean is_overflow, SgenGrayQueue *unpin_queue, gboolean finish_up_concurrent_mark)
1469 gboolean needs_major;
1470 size_t max_garbage_amount;
1472 mword fragment_total;
1474 SgenGrayQueue gc_thread_gray_queue;
1475 SgenObjectOperations *object_ops = &sgen_minor_collector.serial_ops;
1476 ScanCopyContext ctx;
1480 if (disable_minor_collections)
1483 TV_GETTIME (last_minor_collection_start_tv);
1484 atv = last_minor_collection_start_tv;
1486 binary_protocol_collection_begin (gc_stats.minor_gc_count, GENERATION_NURSERY);
1488 if (do_verify_nursery || do_dump_nursery_content)
1489 sgen_debug_verify_nursery (do_dump_nursery_content);
1491 current_collection_generation = GENERATION_NURSERY;
1493 SGEN_ASSERT (0, !sgen_collection_is_concurrent (), "Why is the nursery collection concurrent?");
1495 reset_pinned_from_failed_allocation ();
1497 check_scan_starts ();
1499 sgen_nursery_alloc_prepare_for_minor ();
1503 nursery_next = sgen_nursery_alloc_get_upper_alloc_bound ();
1504 /* FIXME: optimize later to use the higher address where an object can be present */
1505 nursery_next = MAX (nursery_next, sgen_get_nursery_end ());
1507 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 ()));
1508 max_garbage_amount = nursery_next - sgen_get_nursery_start ();
1509 g_assert (nursery_section->size >= max_garbage_amount);
1511 /* world must be stopped already */
1513 time_minor_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1515 sgen_client_pre_collection_checks ();
1517 nursery_section->next_data = nursery_next;
1519 major_collector.start_nursery_collection ();
1521 sgen_memgov_minor_collection_start ();
1523 init_gray_queue (&gc_thread_gray_queue, FALSE);
1524 ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, &gc_thread_gray_queue);
1526 gc_stats.minor_gc_count ++;
1528 if (whole_heap_check_before_collection) {
1529 sgen_clear_nursery_fragments ();
1530 sgen_check_whole_heap (finish_up_concurrent_mark);
1533 sgen_process_fin_stage_entries ();
1535 /* pin from pinned handles */
1536 sgen_init_pinning ();
1537 sgen_client_binary_protocol_mark_start (GENERATION_NURSERY);
1538 pin_from_roots (sgen_get_nursery_start (), nursery_next, ctx);
1539 /* pin cemented objects */
1540 sgen_pin_cemented_objects ();
1541 /* identify pinned objects */
1542 sgen_optimize_pin_queue ();
1543 sgen_pinning_setup_section (nursery_section);
1545 pin_objects_in_nursery (FALSE, ctx);
1546 sgen_pinning_trim_queue_to_section (nursery_section);
1548 if (remset_consistency_checks)
1549 sgen_check_remset_consistency ();
1552 time_minor_pinning += TV_ELAPSED (btv, atv);
1553 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), (long long)TV_ELAPSED (btv, atv));
1554 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1556 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan remset", job_remembered_set_scan, sizeof (ScanJob));
1557 sj->ops = object_ops;
1558 sj->gc_thread_gray_queue = &gc_thread_gray_queue;
1559 sgen_workers_enqueue_job (&sj->job, FALSE);
1561 /* we don't have complete write barrier yet, so we scan all the old generation sections */
1563 time_minor_scan_remsets += TV_ELAPSED (atv, btv);
1564 SGEN_LOG (2, "Old generation scan: %lld usecs", (long long)TV_ELAPSED (atv, btv));
1566 sgen_pin_stats_report ();
1568 /* FIXME: Why do we do this at this specific, seemingly random, point? */
1569 sgen_client_collecting_minor (&fin_ready_queue, &critical_fin_queue);
1572 time_minor_scan_pinned += TV_ELAPSED (btv, atv);
1574 enqueue_scan_from_roots_jobs (&gc_thread_gray_queue, sgen_get_nursery_start (), nursery_next, object_ops, FALSE);
1577 time_minor_scan_roots += TV_ELAPSED (atv, btv);
1579 finish_gray_stack (GENERATION_NURSERY, ctx);
1582 time_minor_finish_gray_stack += TV_ELAPSED (btv, atv);
1583 sgen_client_binary_protocol_mark_end (GENERATION_NURSERY);
1585 if (objects_pinned) {
1586 sgen_optimize_pin_queue ();
1587 sgen_pinning_setup_section (nursery_section);
1591 * This is the latest point at which we can do this check, because
1592 * sgen_build_nursery_fragments() unpins nursery objects again.
1594 if (remset_consistency_checks)
1595 sgen_check_remset_consistency ();
1597 /* walk the pin_queue, build up the fragment list of free memory, unmark
1598 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1601 sgen_client_binary_protocol_reclaim_start (GENERATION_NURSERY);
1602 fragment_total = sgen_build_nursery_fragments (nursery_section, unpin_queue);
1603 if (!fragment_total)
1606 /* Clear TLABs for all threads */
1607 sgen_clear_tlabs ();
1609 sgen_client_binary_protocol_reclaim_end (GENERATION_NURSERY);
1611 time_minor_fragment_creation += TV_ELAPSED (atv, btv);
1612 SGEN_LOG (2, "Fragment creation: %lld usecs, %lu bytes available", (long long)TV_ELAPSED (atv, btv), (unsigned long)fragment_total);
1614 if (remset_consistency_checks)
1615 sgen_check_major_refs ();
1617 major_collector.finish_nursery_collection ();
1619 TV_GETTIME (last_minor_collection_end_tv);
1620 gc_stats.minor_gc_time += TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv);
1622 sgen_debug_dump_heap ("minor", gc_stats.minor_gc_count - 1, NULL);
1624 /* prepare the pin queue for the next collection */
1625 sgen_finish_pinning ();
1626 if (sgen_have_pending_finalizers ()) {
1627 SGEN_LOG (4, "Finalizer-thread wakeup");
1628 sgen_client_finalize_notify ();
1630 sgen_pin_stats_reset ();
1631 /* clear cemented hash */
1632 sgen_cement_clear_below_threshold ();
1634 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
1636 remset.finish_minor_collection ();
1638 check_scan_starts ();
1640 binary_protocol_flush_buffers (FALSE);
1642 sgen_memgov_minor_collection_end (reason, is_overflow);
1644 /*objects are late pinned because of lack of memory, so a major is a good call*/
1645 needs_major = objects_pinned > 0;
1646 current_collection_generation = -1;
1649 binary_protocol_collection_end (gc_stats.minor_gc_count - 1, GENERATION_NURSERY, 0, 0);
1651 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1652 sgen_check_nursery_objects_pinned (unpin_queue != NULL);
1658 COPY_OR_MARK_FROM_ROOTS_SERIAL,
1659 COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT,
1660 COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT
1661 } CopyOrMarkFromRootsMode;
1664 major_copy_or_mark_from_roots (SgenGrayQueue *gc_thread_gray_queue, size_t *old_next_pin_slot, CopyOrMarkFromRootsMode mode, SgenObjectOperations *object_ops)
1669 /* FIXME: only use these values for the precise scan
1670 * note that to_space pointers should be excluded anyway...
1672 char *heap_start = NULL;
1673 char *heap_end = (char*)-1;
1674 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, gc_thread_gray_queue);
1675 gboolean concurrent = mode != COPY_OR_MARK_FROM_ROOTS_SERIAL;
1677 SGEN_ASSERT (0, !!concurrent == !!concurrent_collection_in_progress, "We've been called with the wrong mode.");
1679 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1680 /*This cleans up unused fragments */
1681 sgen_nursery_allocator_prepare_for_pinning ();
1683 if (do_concurrent_checks)
1684 sgen_debug_check_nursery_is_clean ();
1686 /* The concurrent collector doesn't touch the nursery. */
1687 sgen_nursery_alloc_prepare_for_major ();
1692 /* Pinning depends on this */
1693 sgen_clear_nursery_fragments ();
1695 if (whole_heap_check_before_collection)
1696 sgen_check_whole_heap (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT);
1699 time_major_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1701 if (!sgen_collection_is_concurrent ())
1702 nursery_section->next_data = sgen_get_nursery_end ();
1703 /* we should also coalesce scanning from sections close to each other
1704 * and deal with pointers outside of the sections later.
1709 sgen_client_pre_collection_checks ();
1711 if (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1712 /* Remsets are not useful for a major collection */
1713 remset.clear_cards ();
1716 sgen_process_fin_stage_entries ();
1719 sgen_init_pinning ();
1720 SGEN_LOG (6, "Collecting pinned addresses");
1721 pin_from_roots ((void*)lowest_heap_address, (void*)highest_heap_address, ctx);
1722 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1723 /* Pin cemented objects that were forced */
1724 sgen_pin_cemented_objects ();
1726 sgen_optimize_pin_queue ();
1727 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1729 * Cemented objects that are in the pinned list will be marked. When
1730 * marking concurrently we won't mark mod-union cards for these objects.
1731 * Instead they will remain cemented until the next major collection,
1732 * when we will recheck if they are still pinned in the roots.
1734 sgen_cement_force_pinned ();
1737 sgen_client_collecting_major_1 ();
1740 * pin_queue now contains all candidate pointers, sorted and
1741 * uniqued. We must do two passes now to figure out which
1742 * objects are pinned.
1744 * The first is to find within the pin_queue the area for each
1745 * section. This requires that the pin_queue be sorted. We
1746 * also process the LOS objects and pinned chunks here.
1748 * The second, destructive, pass is to reduce the section
1749 * areas to pointers to the actually pinned objects.
1751 SGEN_LOG (6, "Pinning from sections");
1752 /* first pass for the sections */
1753 sgen_find_section_pin_queue_start_end (nursery_section);
1754 /* identify possible pointers to the insize of large objects */
1755 SGEN_LOG (6, "Pinning from large objects");
1756 for (bigobj = los_object_list; bigobj; bigobj = bigobj->next) {
1758 if (sgen_find_optimized_pin_queue_area ((char*)bigobj->data, (char*)bigobj->data + sgen_los_object_size (bigobj), &dummy, &dummy)) {
1759 binary_protocol_pin (bigobj->data, (gpointer)LOAD_VTABLE (bigobj->data), safe_object_get_size (bigobj->data));
1761 if (sgen_los_object_is_pinned (bigobj->data)) {
1762 SGEN_ASSERT (0, mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, "LOS objects can only be pinned here after concurrent marking.");
1765 sgen_los_pin_object (bigobj->data);
1766 if (SGEN_OBJECT_HAS_REFERENCES (bigobj->data))
1767 GRAY_OBJECT_ENQUEUE (gc_thread_gray_queue, bigobj->data, sgen_obj_get_descriptor ((GCObject*)bigobj->data));
1768 sgen_pin_stats_register_object (bigobj->data, GENERATION_OLD);
1769 SGEN_LOG (6, "Marked large object %p (%s) size: %lu from roots", bigobj->data,
1770 sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (bigobj->data)),
1771 (unsigned long)sgen_los_object_size (bigobj));
1773 sgen_client_pinned_los_object (bigobj->data);
1777 pin_objects_in_nursery (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT, ctx);
1778 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1779 sgen_check_nursery_objects_pinned (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1781 major_collector.pin_objects (gc_thread_gray_queue);
1782 if (old_next_pin_slot)
1783 *old_next_pin_slot = sgen_get_pinned_count ();
1786 * We don't actually pin when starting a concurrent collection, so the remset
1787 * consistency check won't work.
1789 if (remset_consistency_checks && mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT)
1790 sgen_check_remset_consistency ();
1793 time_major_pinning += TV_ELAPSED (atv, btv);
1794 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), (long long)TV_ELAPSED (atv, btv));
1795 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1797 major_collector.init_to_space ();
1799 SGEN_ASSERT (0, sgen_workers_all_done (), "Why are the workers not done when we start or finish a major collection?");
1800 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1801 if (sgen_workers_have_idle_work ()) {
1803 * We force the finish of the worker with the new object ops context
1804 * which can also do copying. We need to have finished pinning.
1806 sgen_workers_start_all_workers (object_ops, NULL);
1807 sgen_workers_join ();
1811 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1812 main_gc_thread = mono_native_thread_self ();
1815 sgen_client_collecting_major_2 ();
1818 time_major_scan_pinned += TV_ELAPSED (btv, atv);
1820 sgen_client_collecting_major_3 (&fin_ready_queue, &critical_fin_queue);
1822 enqueue_scan_from_roots_jobs (gc_thread_gray_queue, heap_start, heap_end, object_ops, FALSE);
1825 time_major_scan_roots += TV_ELAPSED (atv, btv);
1828 * We start the concurrent worker after pinning and after we scanned the roots
1829 * in order to make sure that the worker does not finish before handling all
1832 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1833 if (precleaning_enabled) {
1835 /* Mod union preclean job */
1836 sj = (ScanJob*)sgen_thread_pool_job_alloc ("preclean mod union cardtable", job_mod_union_preclean, sizeof (ScanJob));
1837 sj->ops = object_ops;
1838 sj->gc_thread_gray_queue = NULL;
1839 sgen_workers_start_all_workers (object_ops, &sj->job);
1841 sgen_workers_start_all_workers (object_ops, NULL);
1843 gray_queue_enable_redirect (gc_thread_gray_queue);
1846 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1849 /* Mod union card table */
1850 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan mod union cardtable", job_scan_major_mod_union_card_table, sizeof (ScanJob));
1851 sj->ops = object_ops;
1852 sj->gc_thread_gray_queue = gc_thread_gray_queue;
1853 sgen_workers_enqueue_job (&sj->job, FALSE);
1855 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan LOS mod union cardtable", job_scan_los_mod_union_card_table, sizeof (ScanJob));
1856 sj->ops = object_ops;
1857 sj->gc_thread_gray_queue = gc_thread_gray_queue;
1858 sgen_workers_enqueue_job (&sj->job, FALSE);
1861 time_major_scan_mod_union += TV_ELAPSED (btv, atv);
1864 sgen_pin_stats_report ();
1866 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1867 sgen_finish_pinning ();
1869 sgen_pin_stats_reset ();
1871 if (do_concurrent_checks)
1872 sgen_debug_check_nursery_is_clean ();
1877 major_start_collection (SgenGrayQueue *gc_thread_gray_queue, const char *reason, gboolean concurrent, size_t *old_next_pin_slot)
1879 SgenObjectOperations *object_ops;
1881 binary_protocol_collection_begin (gc_stats.major_gc_count, GENERATION_OLD);
1883 current_collection_generation = GENERATION_OLD;
1885 sgen_workers_assert_gray_queue_is_empty ();
1888 sgen_cement_reset ();
1891 g_assert (major_collector.is_concurrent);
1892 concurrent_collection_in_progress = TRUE;
1894 object_ops = &major_collector.major_ops_concurrent_start;
1896 object_ops = &major_collector.major_ops_serial;
1899 reset_pinned_from_failed_allocation ();
1901 sgen_memgov_major_collection_start (concurrent, reason);
1903 //count_ref_nonref_objs ();
1904 //consistency_check ();
1906 check_scan_starts ();
1909 SGEN_LOG (1, "Start major collection %d", gc_stats.major_gc_count);
1910 gc_stats.major_gc_count ++;
1912 if (major_collector.start_major_collection)
1913 major_collector.start_major_collection ();
1915 major_copy_or_mark_from_roots (gc_thread_gray_queue, old_next_pin_slot, concurrent ? COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT : COPY_OR_MARK_FROM_ROOTS_SERIAL, object_ops);
1919 major_finish_collection (SgenGrayQueue *gc_thread_gray_queue, const char *reason, gboolean is_overflow, 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 (gc_thread_gray_queue, NULL, COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, object_ops);
1934 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1935 main_gc_thread = NULL;
1938 object_ops = &major_collector.major_ops_serial;
1941 sgen_workers_assert_gray_queue_is_empty ();
1943 finish_gray_stack (GENERATION_OLD, CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, gc_thread_gray_queue));
1945 time_major_finish_gray_stack += TV_ELAPSED (btv, atv);
1947 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after joining");
1949 if (objects_pinned) {
1950 g_assert (!concurrent_collection_in_progress);
1953 * This is slow, but we just OOM'd.
1955 * See comment at `sgen_pin_queue_clear_discarded_entries` for how the pin
1956 * queue is laid out at this point.
1958 sgen_pin_queue_clear_discarded_entries (nursery_section, old_next_pin_slot);
1960 * We need to reestablish all pinned nursery objects in the pin queue
1961 * because they're needed for fragment creation. Unpinning happens by
1962 * walking the whole queue, so it's not necessary to reestablish where major
1963 * heap block pins are - all we care is that they're still in there
1966 sgen_optimize_pin_queue ();
1967 sgen_find_section_pin_queue_start_end (nursery_section);
1971 reset_heap_boundaries ();
1972 sgen_update_heap_boundaries ((mword)sgen_get_nursery_start (), (mword)sgen_get_nursery_end ());
1974 if (whole_heap_check_before_collection)
1975 sgen_check_whole_heap (FALSE);
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 sgen_memgov_major_collection_end (forced, concurrent_collection_in_progress, reason, is_overflow);
2031 current_collection_generation = -1;
2033 memset (&counts, 0, sizeof (ScannedObjectCounts));
2034 major_collector.finish_major_collection (&counts);
2036 sgen_workers_assert_gray_queue_is_empty ();
2038 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after major collection has finished");
2039 if (concurrent_collection_in_progress)
2040 concurrent_collection_in_progress = FALSE;
2042 check_scan_starts ();
2044 binary_protocol_flush_buffers (FALSE);
2046 //consistency_check ();
2048 binary_protocol_collection_end (gc_stats.major_gc_count - 1, GENERATION_OLD, counts.num_scanned_objects, counts.num_unique_scanned_objects);
2052 major_do_collection (const char *reason, gboolean is_overflow, gboolean forced)
2054 TV_DECLARE (time_start);
2055 TV_DECLARE (time_end);
2056 size_t old_next_pin_slot;
2057 SgenGrayQueue gc_thread_gray_queue;
2059 if (disable_major_collections)
2062 if (major_collector.get_and_reset_num_major_objects_marked) {
2063 long long num_marked = major_collector.get_and_reset_num_major_objects_marked ();
2064 g_assert (!num_marked);
2067 /* world must be stopped already */
2068 TV_GETTIME (time_start);
2070 init_gray_queue (&gc_thread_gray_queue, FALSE);
2071 major_start_collection (&gc_thread_gray_queue, reason, FALSE, &old_next_pin_slot);
2072 major_finish_collection (&gc_thread_gray_queue, reason, is_overflow, old_next_pin_slot, forced);
2073 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2075 TV_GETTIME (time_end);
2076 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2078 /* FIXME: also report this to the user, preferably in gc-end. */
2079 if (major_collector.get_and_reset_num_major_objects_marked)
2080 major_collector.get_and_reset_num_major_objects_marked ();
2082 return bytes_pinned_from_failed_allocation > 0;
2086 major_start_concurrent_collection (const char *reason)
2088 TV_DECLARE (time_start);
2089 TV_DECLARE (time_end);
2090 long long num_objects_marked;
2091 SgenGrayQueue gc_thread_gray_queue;
2093 if (disable_major_collections)
2096 TV_GETTIME (time_start);
2097 SGEN_TV_GETTIME (time_major_conc_collection_start);
2099 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2100 g_assert (num_objects_marked == 0);
2102 binary_protocol_concurrent_start ();
2104 init_gray_queue (&gc_thread_gray_queue, TRUE);
2105 // FIXME: store reason and pass it when finishing
2106 major_start_collection (&gc_thread_gray_queue, reason, TRUE, NULL);
2107 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2109 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2111 TV_GETTIME (time_end);
2112 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2114 current_collection_generation = -1;
2118 * Returns whether the major collection has finished.
2121 major_should_finish_concurrent_collection (void)
2123 return sgen_workers_all_done ();
2127 major_update_concurrent_collection (void)
2129 TV_DECLARE (total_start);
2130 TV_DECLARE (total_end);
2132 TV_GETTIME (total_start);
2134 binary_protocol_concurrent_update ();
2136 major_collector.update_cardtable_mod_union ();
2137 sgen_los_update_cardtable_mod_union ();
2139 TV_GETTIME (total_end);
2140 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end);
2144 major_finish_concurrent_collection (gboolean forced)
2146 SgenGrayQueue gc_thread_gray_queue;
2147 TV_DECLARE (total_start);
2148 TV_DECLARE (total_end);
2150 TV_GETTIME (total_start);
2152 binary_protocol_concurrent_finish ();
2155 * We need to stop all workers since we're updating the cardtable below.
2156 * The workers will be resumed with a finishing pause context to avoid
2157 * additional cardtable and object scanning.
2159 sgen_workers_stop_all_workers ();
2161 SGEN_TV_GETTIME (time_major_conc_collection_end);
2162 gc_stats.major_gc_time_concurrent += SGEN_TV_ELAPSED (time_major_conc_collection_start, time_major_conc_collection_end);
2164 major_collector.update_cardtable_mod_union ();
2165 sgen_los_update_cardtable_mod_union ();
2167 if (mod_union_consistency_check)
2168 sgen_check_mod_union_consistency ();
2170 current_collection_generation = GENERATION_OLD;
2171 sgen_cement_reset ();
2172 init_gray_queue (&gc_thread_gray_queue, FALSE);
2173 major_finish_collection (&gc_thread_gray_queue, "finishing", FALSE, -1, forced);
2174 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2176 TV_GETTIME (total_end);
2177 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end) - TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv);
2179 current_collection_generation = -1;
2183 * Ensure an allocation request for @size will succeed by freeing enough memory.
2185 * LOCKING: The GC lock MUST be held.
2188 sgen_ensure_free_space (size_t size, int generation)
2190 int generation_to_collect = -1;
2191 const char *reason = NULL;
2193 if (generation == GENERATION_OLD) {
2194 if (sgen_need_major_collection (size)) {
2195 reason = "LOS overflow";
2196 generation_to_collect = GENERATION_OLD;
2199 if (degraded_mode) {
2200 if (sgen_need_major_collection (size)) {
2201 reason = "Degraded mode overflow";
2202 generation_to_collect = GENERATION_OLD;
2204 } else if (sgen_need_major_collection (size)) {
2205 reason = concurrent_collection_in_progress ? "Forced finish concurrent collection" : "Minor allowance";
2206 generation_to_collect = GENERATION_OLD;
2208 generation_to_collect = GENERATION_NURSERY;
2209 reason = "Nursery full";
2213 if (generation_to_collect == -1) {
2214 if (concurrent_collection_in_progress && sgen_workers_all_done ()) {
2215 generation_to_collect = GENERATION_OLD;
2216 reason = "Finish concurrent collection";
2220 if (generation_to_collect == -1)
2222 sgen_perform_collection (size, generation_to_collect, reason, FALSE, TRUE);
2226 * LOCKING: Assumes the GC lock is held.
2229 sgen_perform_collection (size_t requested_size, int generation_to_collect, const char *reason, gboolean wait_to_finish, gboolean stw)
2231 TV_DECLARE (gc_total_start);
2232 TV_DECLARE (gc_total_end);
2233 int overflow_generation_to_collect = -1;
2234 int oldest_generation_collected = generation_to_collect;
2235 const char *overflow_reason = NULL;
2236 gboolean finish_concurrent = concurrent_collection_in_progress && (major_should_finish_concurrent_collection () || generation_to_collect == GENERATION_OLD);
2238 binary_protocol_collection_requested (generation_to_collect, requested_size, wait_to_finish ? 1 : 0);
2240 SGEN_ASSERT (0, generation_to_collect == GENERATION_NURSERY || generation_to_collect == GENERATION_OLD, "What generation is this?");
2243 sgen_stop_world (generation_to_collect);
2245 SGEN_ASSERT (0, sgen_is_world_stopped (), "We can only collect if the world is stopped");
2248 TV_GETTIME (gc_total_start);
2250 // FIXME: extract overflow reason
2251 // FIXME: minor overflow for concurrent case
2252 if (generation_to_collect == GENERATION_NURSERY && !finish_concurrent) {
2253 if (concurrent_collection_in_progress)
2254 major_update_concurrent_collection ();
2256 if (collect_nursery (reason, FALSE, NULL, FALSE) && !concurrent_collection_in_progress) {
2257 overflow_generation_to_collect = GENERATION_OLD;
2258 overflow_reason = "Minor overflow";
2260 } else if (finish_concurrent) {
2261 major_finish_concurrent_collection (wait_to_finish);
2262 oldest_generation_collected = GENERATION_OLD;
2264 SGEN_ASSERT (0, generation_to_collect == GENERATION_OLD, "We should have handled nursery collections above");
2265 if (major_collector.is_concurrent && !wait_to_finish) {
2266 collect_nursery ("Concurrent start", FALSE, NULL, FALSE);
2267 major_start_concurrent_collection (reason);
2268 oldest_generation_collected = GENERATION_NURSERY;
2269 } else if (major_do_collection (reason, FALSE, wait_to_finish)) {
2270 overflow_generation_to_collect = GENERATION_NURSERY;
2271 overflow_reason = "Excessive pinning";
2275 if (overflow_generation_to_collect != -1) {
2276 SGEN_ASSERT (0, !concurrent_collection_in_progress, "We don't yet support overflow collections with the concurrent collector");
2279 * We need to do an overflow collection, either because we ran out of memory
2280 * or the nursery is fully pinned.
2283 if (overflow_generation_to_collect == GENERATION_NURSERY)
2284 collect_nursery (overflow_reason, TRUE, NULL, FALSE);
2286 major_do_collection (overflow_reason, TRUE, wait_to_finish);
2288 oldest_generation_collected = MAX (oldest_generation_collected, overflow_generation_to_collect);
2291 SGEN_LOG (2, "Heap size: %lu, LOS size: %lu", (unsigned long)sgen_gc_get_total_heap_allocation (), (unsigned long)los_memory_usage);
2293 /* this also sets the proper pointers for the next allocation */
2294 if (generation_to_collect == GENERATION_NURSERY && !sgen_can_alloc_size (requested_size)) {
2295 /* TypeBuilder and MonoMethod are killing mcs with fragmentation */
2296 SGEN_LOG (1, "nursery collection didn't find enough room for %zd alloc (%zd pinned)", requested_size, sgen_get_pinned_count ());
2297 sgen_dump_pin_queue ();
2301 TV_GETTIME (gc_total_end);
2302 time_max = MAX (time_max, TV_ELAPSED (gc_total_start, gc_total_end));
2305 sgen_restart_world (oldest_generation_collected);
2309 * ######################################################################
2310 * ######## Memory allocation from the OS
2311 * ######################################################################
2312 * This section of code deals with getting memory from the OS and
2313 * allocating memory for GC-internal data structures.
2314 * Internal memory can be handled with a freelist for small objects.
2320 G_GNUC_UNUSED static void
2321 report_internal_mem_usage (void)
2323 printf ("Internal memory usage:\n");
2324 sgen_report_internal_mem_usage ();
2325 printf ("Pinned memory usage:\n");
2326 major_collector.report_pinned_memory_usage ();
2330 * ######################################################################
2331 * ######## Finalization support
2332 * ######################################################################
2336 * If the object has been forwarded it means it's still referenced from a root.
2337 * If it is pinned it's still alive as well.
2338 * A LOS object is only alive if we have pinned it.
2339 * Return TRUE if @obj is ready to be finalized.
2341 static inline gboolean
2342 sgen_is_object_alive (GCObject *object)
2344 if (ptr_in_nursery (object))
2345 return sgen_nursery_is_object_alive (object);
2347 return sgen_major_is_object_alive (object);
2351 * This function returns true if @object is either alive and belongs to the
2352 * current collection - major collections are full heap, so old gen objects
2353 * are never alive during a minor collection.
2356 sgen_is_object_alive_and_on_current_collection (GCObject *object)
2358 if (ptr_in_nursery (object))
2359 return sgen_nursery_is_object_alive (object);
2361 if (current_collection_generation == GENERATION_NURSERY)
2364 return sgen_major_is_object_alive (object);
2369 sgen_gc_is_object_ready_for_finalization (GCObject *object)
2371 return !sgen_is_object_alive (object);
2375 sgen_queue_finalization_entry (GCObject *obj)
2377 gboolean critical = sgen_client_object_has_critical_finalizer (obj);
2379 sgen_pointer_queue_add (critical ? &critical_fin_queue : &fin_ready_queue, obj);
2381 sgen_client_object_queued_for_finalization (obj);
2385 sgen_object_is_live (GCObject *obj)
2387 return sgen_is_object_alive_and_on_current_collection (obj);
2391 * `System.GC.WaitForPendingFinalizers` first checks `sgen_have_pending_finalizers()` to
2392 * determine whether it can exit quickly. The latter must therefore only return FALSE if
2393 * all finalizers have really finished running.
2395 * `sgen_gc_invoke_finalizers()` first dequeues a finalizable object, and then finalizes it.
2396 * This means that just checking whether the queues are empty leaves the possibility that an
2397 * object might have been dequeued but not yet finalized. That's why we need the additional
2398 * flag `pending_unqueued_finalizer`.
2401 static volatile gboolean pending_unqueued_finalizer = FALSE;
2404 sgen_gc_invoke_finalizers (void)
2408 g_assert (!pending_unqueued_finalizer);
2410 /* FIXME: batch to reduce lock contention */
2411 while (sgen_have_pending_finalizers ()) {
2417 * We need to set `pending_unqueued_finalizer` before dequeing the
2418 * finalizable object.
2420 if (!sgen_pointer_queue_is_empty (&fin_ready_queue)) {
2421 pending_unqueued_finalizer = TRUE;
2422 mono_memory_write_barrier ();
2423 obj = (GCObject *)sgen_pointer_queue_pop (&fin_ready_queue);
2424 } else if (!sgen_pointer_queue_is_empty (&critical_fin_queue)) {
2425 pending_unqueued_finalizer = TRUE;
2426 mono_memory_write_barrier ();
2427 obj = (GCObject *)sgen_pointer_queue_pop (&critical_fin_queue);
2433 SGEN_LOG (7, "Finalizing object %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
2441 /* the object is on the stack so it is pinned */
2442 /*g_print ("Calling finalizer for object: %p (%s)\n", obj, sgen_client_object_safe_name (obj));*/
2443 sgen_client_run_finalize (obj);
2446 if (pending_unqueued_finalizer) {
2447 mono_memory_write_barrier ();
2448 pending_unqueued_finalizer = FALSE;
2455 sgen_have_pending_finalizers (void)
2457 return pending_unqueued_finalizer || !sgen_pointer_queue_is_empty (&fin_ready_queue) || !sgen_pointer_queue_is_empty (&critical_fin_queue);
2461 * ######################################################################
2462 * ######## registered roots support
2463 * ######################################################################
2467 * We do not coalesce roots.
2470 sgen_register_root (char *start, size_t size, SgenDescriptor descr, int root_type, int source, const char *msg)
2472 RootRecord new_root;
2475 for (i = 0; i < ROOT_TYPE_NUM; ++i) {
2476 RootRecord *root = (RootRecord *)sgen_hash_table_lookup (&roots_hash [i], start);
2477 /* we allow changing the size and the descriptor (for thread statics etc) */
2479 size_t old_size = root->end_root - start;
2480 root->end_root = start + size;
2481 SGEN_ASSERT (0, !!root->root_desc == !!descr, "Can't change whether a root is precise or conservative.");
2482 SGEN_ASSERT (0, root->source == source, "Can't change a root's source identifier.");
2483 SGEN_ASSERT (0, !!root->msg == !!msg, "Can't change a root's message.");
2484 root->root_desc = descr;
2486 roots_size -= old_size;
2492 new_root.end_root = start + size;
2493 new_root.root_desc = descr;
2494 new_root.source = source;
2497 sgen_hash_table_replace (&roots_hash [root_type], start, &new_root, NULL);
2500 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);
2507 sgen_deregister_root (char* addr)
2513 for (root_type = 0; root_type < ROOT_TYPE_NUM; ++root_type) {
2514 if (sgen_hash_table_remove (&roots_hash [root_type], addr, &root))
2515 roots_size -= (root.end_root - addr);
2521 * ######################################################################
2522 * ######## Thread handling (stop/start code)
2523 * ######################################################################
2527 sgen_get_current_collection_generation (void)
2529 return current_collection_generation;
2533 sgen_thread_register (SgenThreadInfo* info, void *stack_bottom_fallback)
2535 #ifndef HAVE_KW_THREAD
2536 info->tlab_start = info->tlab_next = info->tlab_temp_end = info->tlab_real_end = NULL;
2539 sgen_init_tlab_info (info);
2541 sgen_client_thread_register (info, stack_bottom_fallback);
2547 sgen_thread_unregister (SgenThreadInfo *p)
2549 sgen_client_thread_unregister (p);
2553 * ######################################################################
2554 * ######## Write barriers
2555 * ######################################################################
2559 * Note: the write barriers first do the needed GC work and then do the actual store:
2560 * this way the value is visible to the conservative GC scan after the write barrier
2561 * itself. If a GC interrupts the barrier in the middle, value will be kept alive by
2562 * the conservative scan, otherwise by the remembered set scan.
2566 mono_gc_wbarrier_arrayref_copy (gpointer dest_ptr, gpointer src_ptr, int count)
2568 HEAVY_STAT (++stat_wbarrier_arrayref_copy);
2569 /*This check can be done without taking a lock since dest_ptr array is pinned*/
2570 if (ptr_in_nursery (dest_ptr) || count <= 0) {
2571 mono_gc_memmove_aligned (dest_ptr, src_ptr, count * sizeof (gpointer));
2575 #ifdef SGEN_HEAVY_BINARY_PROTOCOL
2576 if (binary_protocol_is_heavy_enabled ()) {
2578 for (i = 0; i < count; ++i) {
2579 gpointer dest = (gpointer*)dest_ptr + i;
2580 gpointer obj = *((gpointer*)src_ptr + i);
2582 binary_protocol_wbarrier (dest, obj, (gpointer)LOAD_VTABLE (obj));
2587 remset.wbarrier_arrayref_copy (dest_ptr, src_ptr, count);
2591 mono_gc_wbarrier_generic_nostore (gpointer ptr)
2595 HEAVY_STAT (++stat_wbarrier_generic_store);
2597 sgen_client_wbarrier_generic_nostore_check (ptr);
2599 obj = *(gpointer*)ptr;
2601 binary_protocol_wbarrier (ptr, obj, (gpointer)LOAD_VTABLE (obj));
2604 * We need to record old->old pointer locations for the
2605 * concurrent collector.
2607 if (!ptr_in_nursery (obj) && !concurrent_collection_in_progress) {
2608 SGEN_LOG (8, "Skipping remset at %p", ptr);
2612 SGEN_LOG (8, "Adding remset at %p", ptr);
2614 remset.wbarrier_generic_nostore (ptr);
2618 mono_gc_wbarrier_generic_store (gpointer ptr, GCObject* value)
2620 SGEN_LOG (8, "Wbarrier store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2621 SGEN_UPDATE_REFERENCE_ALLOW_NULL (ptr, value);
2622 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2623 mono_gc_wbarrier_generic_nostore (ptr);
2624 sgen_dummy_use (value);
2627 /* Same as mono_gc_wbarrier_generic_store () but performs the store
2628 * as an atomic operation with release semantics.
2631 mono_gc_wbarrier_generic_store_atomic (gpointer ptr, GCObject *value)
2633 HEAVY_STAT (++stat_wbarrier_generic_store_atomic);
2635 SGEN_LOG (8, "Wbarrier atomic store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2637 InterlockedWritePointer ((volatile gpointer *)ptr, value);
2639 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2640 mono_gc_wbarrier_generic_nostore (ptr);
2642 sgen_dummy_use (value);
2646 sgen_wbarrier_value_copy_bitmap (gpointer _dest, gpointer _src, int size, unsigned bitmap)
2648 GCObject **dest = (GCObject **)_dest;
2649 GCObject **src = (GCObject **)_src;
2653 mono_gc_wbarrier_generic_store (dest, *src);
2658 size -= SIZEOF_VOID_P;
2664 * ######################################################################
2665 * ######## Other mono public interface functions.
2666 * ######################################################################
2670 sgen_gc_collect (int generation)
2675 sgen_perform_collection (0, generation, "user request", TRUE, TRUE);
2680 sgen_gc_collection_count (int generation)
2682 if (generation == 0)
2683 return gc_stats.minor_gc_count;
2684 return gc_stats.major_gc_count;
2688 sgen_gc_get_used_size (void)
2692 tot = los_memory_usage;
2693 tot += nursery_section->next_data - nursery_section->data;
2694 tot += major_collector.get_used_size ();
2695 /* FIXME: account for pinned objects */
2701 sgen_env_var_error (const char *env_var, const char *fallback, const char *description_format, ...)
2705 va_start (ap, description_format);
2707 fprintf (stderr, "Warning: In environment variable `%s': ", env_var);
2708 vfprintf (stderr, description_format, ap);
2710 fprintf (stderr, " - %s", fallback);
2711 fprintf (stderr, "\n");
2717 parse_double_in_interval (const char *env_var, const char *opt_name, const char *opt, double min, double max, double *result)
2720 double val = strtod (opt, &endptr);
2721 if (endptr == opt) {
2722 sgen_env_var_error (env_var, "Using default value.", "`%s` must be a number.", opt_name);
2725 else if (val < min || val > max) {
2726 sgen_env_var_error (env_var, "Using default value.", "`%s` must be between %.2f - %.2f.", opt_name, min, max);
2738 char *major_collector_opt = NULL;
2739 char *minor_collector_opt = NULL;
2740 size_t max_heap = 0;
2741 size_t soft_limit = 0;
2743 gboolean debug_print_allowance = FALSE;
2744 double allowance_ratio = 0, save_target = 0;
2745 gboolean cement_enabled = TRUE;
2748 result = InterlockedCompareExchange (&gc_initialized, -1, 0);
2751 /* already inited */
2754 /* being inited by another thread */
2755 mono_thread_info_usleep (1000);
2758 /* we will init it */
2761 g_assert_not_reached ();
2763 } while (result != 0);
2765 SGEN_TV_GETTIME (sgen_init_timestamp);
2767 #ifdef SGEN_WITHOUT_MONO
2768 mono_thread_smr_init ();
2771 mono_coop_mutex_init (&gc_mutex);
2773 gc_debug_file = stderr;
2775 mono_coop_mutex_init (&sgen_interruption_mutex);
2777 if ((env = g_getenv (MONO_GC_PARAMS_NAME))) {
2778 opts = g_strsplit (env, ",", -1);
2779 for (ptr = opts; *ptr; ++ptr) {
2781 if (g_str_has_prefix (opt, "major=")) {
2782 opt = strchr (opt, '=') + 1;
2783 major_collector_opt = g_strdup (opt);
2784 } else if (g_str_has_prefix (opt, "minor=")) {
2785 opt = strchr (opt, '=') + 1;
2786 minor_collector_opt = g_strdup (opt);
2794 sgen_init_internal_allocator ();
2795 sgen_init_nursery_allocator ();
2796 sgen_init_fin_weak_hash ();
2797 sgen_init_hash_table ();
2798 sgen_init_descriptors ();
2799 sgen_init_gray_queues ();
2800 sgen_init_allocator ();
2801 sgen_init_gchandles ();
2803 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_SECTION, SGEN_SIZEOF_GC_MEM_SECTION);
2804 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_GRAY_QUEUE, sizeof (GrayQueueSection));
2806 sgen_client_init ();
2808 if (!minor_collector_opt) {
2809 sgen_simple_nursery_init (&sgen_minor_collector);
2811 if (!strcmp (minor_collector_opt, "simple")) {
2813 sgen_simple_nursery_init (&sgen_minor_collector);
2814 } else if (!strcmp (minor_collector_opt, "split")) {
2815 sgen_split_nursery_init (&sgen_minor_collector);
2817 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `simple` instead.", "Unknown minor collector `%s'.", minor_collector_opt);
2818 goto use_simple_nursery;
2822 if (!major_collector_opt) {
2824 DEFAULT_MAJOR_INIT (&major_collector);
2825 } else if (!strcmp (major_collector_opt, "marksweep")) {
2826 sgen_marksweep_init (&major_collector);
2827 } else if (!strcmp (major_collector_opt, "marksweep-conc")) {
2828 sgen_marksweep_conc_init (&major_collector);
2830 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `" DEFAULT_MAJOR_NAME "` instead.", "Unknown major collector `%s'.", major_collector_opt);
2831 goto use_default_major;
2834 sgen_nursery_size = DEFAULT_NURSERY_SIZE;
2837 gboolean usage_printed = FALSE;
2839 for (ptr = opts; *ptr; ++ptr) {
2841 if (!strcmp (opt, ""))
2843 if (g_str_has_prefix (opt, "major="))
2845 if (g_str_has_prefix (opt, "minor="))
2847 if (g_str_has_prefix (opt, "max-heap-size=")) {
2848 size_t page_size = mono_pagesize ();
2849 size_t max_heap_candidate = 0;
2850 opt = strchr (opt, '=') + 1;
2851 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &max_heap_candidate)) {
2852 max_heap = (max_heap_candidate + page_size - 1) & ~(size_t)(page_size - 1);
2853 if (max_heap != max_heap_candidate)
2854 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Rounding up.", "`max-heap-size` size must be a multiple of %d.", page_size);
2856 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`max-heap-size` must be an integer.");
2860 if (g_str_has_prefix (opt, "soft-heap-limit=")) {
2861 opt = strchr (opt, '=') + 1;
2862 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &soft_limit)) {
2863 if (soft_limit <= 0) {
2864 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be positive.");
2868 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be an integer.");
2874 if (g_str_has_prefix (opt, "nursery-size=")) {
2876 opt = strchr (opt, '=') + 1;
2877 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &val)) {
2878 if ((val & (val - 1))) {
2879 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be a power of two.");
2883 if (val < SGEN_MAX_NURSERY_WASTE) {
2884 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.",
2885 "`nursery-size` must be at least %d bytes.", SGEN_MAX_NURSERY_WASTE);
2889 sgen_nursery_size = val;
2890 sgen_nursery_bits = 0;
2891 while (ONE_P << (++ sgen_nursery_bits) != sgen_nursery_size)
2894 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be an integer.");
2900 if (g_str_has_prefix (opt, "save-target-ratio=")) {
2902 opt = strchr (opt, '=') + 1;
2903 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "save-target-ratio", opt,
2904 SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO, &val)) {
2909 if (g_str_has_prefix (opt, "default-allowance-ratio=")) {
2911 opt = strchr (opt, '=') + 1;
2912 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "default-allowance-ratio", opt,
2913 SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, SGEN_MAX_ALLOWANCE_NURSERY_SIZE_RATIO, &val)) {
2914 allowance_ratio = val;
2919 if (!strcmp (opt, "cementing")) {
2920 cement_enabled = TRUE;
2923 if (!strcmp (opt, "no-cementing")) {
2924 cement_enabled = FALSE;
2928 if (!strcmp (opt, "precleaning")) {
2929 precleaning_enabled = TRUE;
2932 if (!strcmp (opt, "no-precleaning")) {
2933 precleaning_enabled = FALSE;
2937 if (major_collector.handle_gc_param && major_collector.handle_gc_param (opt))
2940 if (sgen_minor_collector.handle_gc_param && sgen_minor_collector.handle_gc_param (opt))
2943 if (sgen_client_handle_gc_param (opt))
2946 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Ignoring.", "Unknown option `%s`.", opt);
2951 fprintf (stderr, "\n%s must be a comma-delimited list of one or more of the following:\n", MONO_GC_PARAMS_NAME);
2952 fprintf (stderr, " max-heap-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
2953 fprintf (stderr, " soft-heap-limit=n (where N is an integer, possibly with a k, m or a g suffix)\n");
2954 fprintf (stderr, " nursery-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
2955 fprintf (stderr, " major=COLLECTOR (where COLLECTOR is `marksweep', `marksweep-conc', `marksweep-par')\n");
2956 fprintf (stderr, " minor=COLLECTOR (where COLLECTOR is `simple' or `split')\n");
2957 fprintf (stderr, " wbarrier=WBARRIER (where WBARRIER is `remset' or `cardtable')\n");
2958 fprintf (stderr, " [no-]cementing\n");
2959 if (major_collector.print_gc_param_usage)
2960 major_collector.print_gc_param_usage ();
2961 if (sgen_minor_collector.print_gc_param_usage)
2962 sgen_minor_collector.print_gc_param_usage ();
2963 sgen_client_print_gc_params_usage ();
2964 fprintf (stderr, " Experimental options:\n");
2965 fprintf (stderr, " save-target-ratio=R (where R must be between %.2f - %.2f).\n", SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO);
2966 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);
2967 fprintf (stderr, "\n");
2969 usage_printed = TRUE;
2974 if (major_collector_opt)
2975 g_free (major_collector_opt);
2977 if (minor_collector_opt)
2978 g_free (minor_collector_opt);
2982 sgen_pinning_init ();
2983 sgen_cement_init (cement_enabled);
2985 if ((env = g_getenv (MONO_GC_DEBUG_NAME))) {
2986 gboolean usage_printed = FALSE;
2988 opts = g_strsplit (env, ",", -1);
2989 for (ptr = opts; ptr && *ptr; ptr ++) {
2991 if (!strcmp (opt, ""))
2993 if (opt [0] >= '0' && opt [0] <= '9') {
2994 gc_debug_level = atoi (opt);
2999 char *rf = g_strdup_printf ("%s.%d", opt, mono_process_current_pid ());
3000 gc_debug_file = fopen (rf, "wb");
3002 gc_debug_file = stderr;
3005 } else if (!strcmp (opt, "print-allowance")) {
3006 debug_print_allowance = TRUE;
3007 } else if (!strcmp (opt, "print-pinning")) {
3008 sgen_pin_stats_enable ();
3009 } else if (!strcmp (opt, "verify-before-allocs")) {
3010 verify_before_allocs = 1;
3011 has_per_allocation_action = TRUE;
3012 } else if (g_str_has_prefix (opt, "verify-before-allocs=")) {
3013 char *arg = strchr (opt, '=') + 1;
3014 verify_before_allocs = atoi (arg);
3015 has_per_allocation_action = TRUE;
3016 } else if (!strcmp (opt, "collect-before-allocs")) {
3017 collect_before_allocs = 1;
3018 has_per_allocation_action = TRUE;
3019 } else if (g_str_has_prefix (opt, "collect-before-allocs=")) {
3020 char *arg = strchr (opt, '=') + 1;
3021 has_per_allocation_action = TRUE;
3022 collect_before_allocs = atoi (arg);
3023 } else if (!strcmp (opt, "verify-before-collections")) {
3024 whole_heap_check_before_collection = TRUE;
3025 } else if (!strcmp (opt, "check-remset-consistency")) {
3026 remset_consistency_checks = TRUE;
3027 nursery_clear_policy = CLEAR_AT_GC;
3028 } else if (!strcmp (opt, "mod-union-consistency-check")) {
3029 if (!major_collector.is_concurrent) {
3030 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`mod-union-consistency-check` only works with concurrent major collector.");
3033 mod_union_consistency_check = TRUE;
3034 } else if (!strcmp (opt, "check-mark-bits")) {
3035 check_mark_bits_after_major_collection = TRUE;
3036 } else if (!strcmp (opt, "check-nursery-pinned")) {
3037 check_nursery_objects_pinned = TRUE;
3038 } else if (!strcmp (opt, "clear-at-gc")) {
3039 nursery_clear_policy = CLEAR_AT_GC;
3040 } else if (!strcmp (opt, "clear-nursery-at-gc")) {
3041 nursery_clear_policy = CLEAR_AT_GC;
3042 } else if (!strcmp (opt, "clear-at-tlab-creation")) {
3043 nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
3044 } else if (!strcmp (opt, "debug-clear-at-tlab-creation")) {
3045 nursery_clear_policy = CLEAR_AT_TLAB_CREATION_DEBUG;
3046 } else if (!strcmp (opt, "check-scan-starts")) {
3047 do_scan_starts_check = TRUE;
3048 } else if (!strcmp (opt, "verify-nursery-at-minor-gc")) {
3049 do_verify_nursery = TRUE;
3050 } else if (!strcmp (opt, "check-concurrent")) {
3051 if (!major_collector.is_concurrent) {
3052 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`check-concurrent` only works with concurrent major collectors.");
3055 nursery_clear_policy = CLEAR_AT_GC;
3056 do_concurrent_checks = TRUE;
3057 } else if (!strcmp (opt, "dump-nursery-at-minor-gc")) {
3058 do_dump_nursery_content = TRUE;
3059 } else if (!strcmp (opt, "disable-minor")) {
3060 disable_minor_collections = TRUE;
3061 } else if (!strcmp (opt, "disable-major")) {
3062 disable_major_collections = TRUE;
3063 } else if (g_str_has_prefix (opt, "heap-dump=")) {
3064 char *filename = strchr (opt, '=') + 1;
3065 nursery_clear_policy = CLEAR_AT_GC;
3066 sgen_debug_enable_heap_dump (filename);
3067 } else if (g_str_has_prefix (opt, "binary-protocol=")) {
3068 char *filename = strchr (opt, '=') + 1;
3069 char *colon = strrchr (filename, ':');
3072 if (!mono_gc_parse_environment_string_extract_number (colon + 1, &limit)) {
3073 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring limit.", "Binary protocol file size limit must be an integer.");
3078 binary_protocol_init (filename, (long long)limit);
3079 } else if (!strcmp (opt, "nursery-canaries")) {
3080 do_verify_nursery = TRUE;
3081 enable_nursery_canaries = TRUE;
3082 } else if (!sgen_client_handle_gc_debug (opt)) {
3083 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "Unknown option `%s`.", opt);
3088 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);
3089 fprintf (stderr, "Valid <option>s are:\n");
3090 fprintf (stderr, " collect-before-allocs[=<n>]\n");
3091 fprintf (stderr, " verify-before-allocs[=<n>]\n");
3092 fprintf (stderr, " check-remset-consistency\n");
3093 fprintf (stderr, " check-mark-bits\n");
3094 fprintf (stderr, " check-nursery-pinned\n");
3095 fprintf (stderr, " verify-before-collections\n");
3096 fprintf (stderr, " verify-nursery-at-minor-gc\n");
3097 fprintf (stderr, " dump-nursery-at-minor-gc\n");
3098 fprintf (stderr, " disable-minor\n");
3099 fprintf (stderr, " disable-major\n");
3100 fprintf (stderr, " check-concurrent\n");
3101 fprintf (stderr, " clear-[nursery-]at-gc\n");
3102 fprintf (stderr, " clear-at-tlab-creation\n");
3103 fprintf (stderr, " debug-clear-at-tlab-creation\n");
3104 fprintf (stderr, " check-scan-starts\n");
3105 fprintf (stderr, " print-allowance\n");
3106 fprintf (stderr, " print-pinning\n");
3107 fprintf (stderr, " heap-dump=<filename>\n");
3108 fprintf (stderr, " binary-protocol=<filename>[:<file-size-limit>]\n");
3109 fprintf (stderr, " nursery-canaries\n");
3110 sgen_client_print_gc_debug_usage ();
3111 fprintf (stderr, "\n");
3113 usage_printed = TRUE;
3119 if (check_mark_bits_after_major_collection)
3120 nursery_clear_policy = CLEAR_AT_GC;
3122 if (major_collector.post_param_init)
3123 major_collector.post_param_init (&major_collector);
3125 if (major_collector.needs_thread_pool)
3126 sgen_workers_init (1);
3128 sgen_memgov_init (max_heap, soft_limit, debug_print_allowance, allowance_ratio, save_target);
3130 memset (&remset, 0, sizeof (remset));
3132 sgen_card_table_init (&remset);
3134 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");
3138 sgen_init_bridge ();
3142 sgen_gc_initialized ()
3144 return gc_initialized > 0;
3148 sgen_get_nursery_clear_policy (void)
3150 return nursery_clear_policy;
3156 mono_coop_mutex_lock (&gc_mutex);
3160 sgen_gc_unlock (void)
3162 mono_coop_mutex_unlock (&gc_mutex);
3166 sgen_major_collector_iterate_live_block_ranges (sgen_cardtable_block_callback callback)
3168 major_collector.iterate_live_block_ranges (callback);
3172 sgen_major_collector_iterate_block_ranges (sgen_cardtable_block_callback callback)
3174 major_collector.iterate_block_ranges (callback);
3178 sgen_get_major_collector (void)
3180 return &major_collector;
3184 sgen_get_remset (void)
3190 count_cards (long long *major_total, long long *major_marked, long long *los_total, long long *los_marked)
3192 sgen_get_major_collector ()->count_cards (major_total, major_marked);
3193 sgen_los_count_cards (los_total, los_marked);
3196 static gboolean world_is_stopped = FALSE;
3198 /* LOCKING: assumes the GC lock is held */
3200 sgen_stop_world (int generation)
3202 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3204 SGEN_ASSERT (0, !world_is_stopped, "Why are we stopping a stopped world?");
3206 binary_protocol_world_stopping (generation, sgen_timestamp (), (gpointer) (gsize) mono_native_thread_id_get ());
3208 sgen_client_stop_world (generation);
3210 world_is_stopped = TRUE;
3212 if (binary_protocol_is_heavy_enabled ())
3213 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3214 binary_protocol_world_stopped (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3217 /* LOCKING: assumes the GC lock is held */
3219 sgen_restart_world (int generation)
3221 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3224 SGEN_ASSERT (0, world_is_stopped, "Why are we restarting a running world?");
3226 if (binary_protocol_is_heavy_enabled ())
3227 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3228 binary_protocol_world_restarting (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3230 world_is_stopped = FALSE;
3232 sgen_client_restart_world (generation, &stw_time);
3234 binary_protocol_world_restarted (generation, sgen_timestamp ());
3236 if (sgen_client_bridge_need_processing ())
3237 sgen_client_bridge_processing_finish (generation);
3239 sgen_memgov_collection_end (generation, stw_time);
3243 sgen_is_world_stopped (void)
3245 return world_is_stopped;
3249 sgen_check_whole_heap_stw (void)
3251 sgen_stop_world (0);
3252 sgen_clear_nursery_fragments ();
3253 sgen_check_whole_heap (FALSE);
3254 sgen_restart_world (0);
3258 sgen_timestamp (void)
3260 SGEN_TV_DECLARE (timestamp);
3261 SGEN_TV_GETTIME (timestamp);
3262 return SGEN_TV_ELAPSED (sgen_init_timestamp, timestamp);
3265 #endif /* HAVE_SGEN_GC */