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 typedef SgenGrayQueue GrayQueue;
404 /* forward declarations */
405 static void scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx);
407 static void pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx);
408 static void finish_gray_stack (int generation, ScanCopyContext ctx);
411 SgenMajorCollector major_collector;
412 SgenMinorCollector sgen_minor_collector;
413 /* FIXME: get rid of this */
414 static GrayQueue gray_queue;
416 static SgenRememberedSet remset;
418 /* The gray queue to use from the main collection thread. */
419 #define WORKERS_DISTRIBUTE_GRAY_QUEUE (&gray_queue)
422 * The gray queue a worker job must use. If we're not parallel or
423 * concurrent, we use the main gray queue.
425 static SgenGrayQueue*
426 sgen_workers_get_job_gray_queue (WorkerData *worker_data)
428 return worker_data ? &worker_data->private_gray_queue : WORKERS_DISTRIBUTE_GRAY_QUEUE;
432 gray_queue_enable_redirect (SgenGrayQueue *queue)
434 SGEN_ASSERT (0, concurrent_collection_in_progress, "Where are we redirecting the gray queue to, without a concurrent collection?");
436 sgen_gray_queue_set_alloc_prepare (queue, sgen_workers_take_from_queue_and_awake);
437 sgen_workers_take_from_queue_and_awake (queue);
441 sgen_scan_area_with_callback (char *start, char *end, IterateObjectCallbackFunc callback, void *data, gboolean allow_flags, gboolean fail_on_canaries)
443 while (start < end) {
447 if (!*(void**)start) {
448 start += sizeof (void*); /* should be ALLOC_ALIGN, really */
453 if (!(obj = (char *)SGEN_OBJECT_IS_FORWARDED (start)))
459 if (!sgen_client_object_is_array_fill ((GCObject*)obj)) {
460 CHECK_CANARY_FOR_OBJECT ((GCObject*)obj, fail_on_canaries);
461 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
462 callback ((GCObject*)obj, size, data);
463 CANARIFY_SIZE (size);
465 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
473 * sgen_add_to_global_remset:
475 * The global remset contains locations which point into newspace after
476 * a minor collection. This can happen if the objects they point to are pinned.
478 * LOCKING: If called from a parallel collector, the global remset
479 * lock must be held. For serial collectors that is not necessary.
482 sgen_add_to_global_remset (gpointer ptr, GCObject *obj)
484 SGEN_ASSERT (5, sgen_ptr_in_nursery (obj), "Target pointer of global remset must be in the nursery");
486 HEAVY_STAT (++stat_wbarrier_add_to_global_remset);
488 if (!major_collector.is_concurrent) {
489 SGEN_ASSERT (5, current_collection_generation != -1, "Global remsets can only be added during collections");
491 if (current_collection_generation == -1)
492 SGEN_ASSERT (5, sgen_concurrent_collection_in_progress (), "Global remsets outside of collection pauses can only be added by the concurrent collector");
495 if (!object_is_pinned (obj))
496 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");
497 else if (sgen_cement_lookup_or_register (obj))
500 remset.record_pointer (ptr);
502 sgen_pin_stats_register_global_remset (obj);
504 SGEN_LOG (8, "Adding global remset for %p", ptr);
505 binary_protocol_global_remset (ptr, obj, (gpointer)SGEN_LOAD_VTABLE (obj));
509 * sgen_drain_gray_stack:
511 * Scan objects in the gray stack until the stack is empty. This should be called
512 * frequently after each object is copied, to achieve better locality and cache
517 sgen_drain_gray_stack (ScanCopyContext ctx)
519 ScanObjectFunc scan_func = ctx.ops->scan_object;
520 GrayQueue *queue = ctx.queue;
522 if (ctx.ops->drain_gray_stack)
523 return ctx.ops->drain_gray_stack (queue);
528 GRAY_OBJECT_DEQUEUE (queue, &obj, &desc);
531 SGEN_LOG (9, "Precise gray object scan %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
532 scan_func (obj, desc, queue);
538 * Addresses in the pin queue are already sorted. This function finds
539 * the object header for each address and pins the object. The
540 * addresses must be inside the nursery section. The (start of the)
541 * address array is overwritten with the addresses of the actually
542 * pinned objects. Return the number of pinned objects.
545 pin_objects_from_nursery_pin_queue (gboolean do_scan_objects, ScanCopyContext ctx)
547 GCMemSection *section = nursery_section;
548 void **start = sgen_pinning_get_entry (section->pin_queue_first_entry);
549 void **end = sgen_pinning_get_entry (section->pin_queue_last_entry);
550 void *start_nursery = section->data;
551 void *end_nursery = section->next_data;
556 void *pinning_front = start_nursery;
558 void **definitely_pinned = start;
559 ScanObjectFunc scan_func = ctx.ops->scan_object;
560 SgenGrayQueue *queue = ctx.queue;
562 sgen_nursery_allocator_prepare_for_pinning ();
564 while (start < end) {
565 GCObject *obj_to_pin = NULL;
566 size_t obj_to_pin_size = 0;
571 SGEN_ASSERT (0, addr >= start_nursery && addr < end_nursery, "Potential pinning address out of range");
572 SGEN_ASSERT (0, addr >= last, "Pin queue not sorted");
579 SGEN_LOG (5, "Considering pinning addr %p", addr);
580 /* We've already processed everything up to pinning_front. */
581 if (addr < pinning_front) {
587 * Find the closest scan start <= addr. We might search backward in the
588 * scan_starts array because entries might be NULL. In the worst case we
589 * start at start_nursery.
591 idx = ((char*)addr - (char*)section->data) / SCAN_START_SIZE;
592 SGEN_ASSERT (0, idx < section->num_scan_start, "Scan start index out of range");
593 search_start = (void*)section->scan_starts [idx];
594 if (!search_start || search_start > addr) {
597 search_start = section->scan_starts [idx];
598 if (search_start && search_start <= addr)
601 if (!search_start || search_start > addr)
602 search_start = start_nursery;
606 * If the pinning front is closer than the scan start we found, start
607 * searching at the front.
609 if (search_start < pinning_front)
610 search_start = pinning_front;
613 * Now addr should be in an object a short distance from search_start.
615 * search_start must point to zeroed mem or point to an object.
618 size_t obj_size, canarified_obj_size;
621 if (!*(void**)search_start) {
622 search_start = (void*)ALIGN_UP ((mword)search_start + sizeof (gpointer));
623 /* The loop condition makes sure we don't overrun addr. */
627 canarified_obj_size = obj_size = ALIGN_UP (safe_object_get_size ((GCObject*)search_start));
630 * Filler arrays are marked by an invalid sync word. We don't
631 * consider them for pinning. They are not delimited by canaries,
634 if (!sgen_client_object_is_array_fill ((GCObject*)search_start)) {
635 CHECK_CANARY_FOR_OBJECT (search_start, TRUE);
636 CANARIFY_SIZE (canarified_obj_size);
638 if (addr >= search_start && (char*)addr < (char*)search_start + obj_size) {
639 /* This is the object we're looking for. */
640 obj_to_pin = (GCObject*)search_start;
641 obj_to_pin_size = canarified_obj_size;
646 /* Skip to the next object */
647 search_start = (void*)((char*)search_start + canarified_obj_size);
648 } while (search_start <= addr);
650 /* We've searched past the address we were looking for. */
652 pinning_front = search_start;
653 goto next_pin_queue_entry;
657 * We've found an object to pin. It might still be a dummy array, but we
658 * can advance the pinning front in any case.
660 pinning_front = (char*)obj_to_pin + obj_to_pin_size;
663 * If this is a dummy array marking the beginning of a nursery
664 * fragment, we don't pin it.
666 if (sgen_client_object_is_array_fill (obj_to_pin))
667 goto next_pin_queue_entry;
670 * Finally - pin the object!
672 desc = sgen_obj_get_descriptor_safe (obj_to_pin);
673 if (do_scan_objects) {
674 scan_func (obj_to_pin, desc, queue);
676 SGEN_LOG (4, "Pinned object %p, vtable %p (%s), count %d\n",
677 obj_to_pin, *(void**)obj_to_pin, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj_to_pin)), count);
678 binary_protocol_pin (obj_to_pin,
679 (gpointer)LOAD_VTABLE (obj_to_pin),
680 safe_object_get_size (obj_to_pin));
682 pin_object (obj_to_pin);
683 GRAY_OBJECT_ENQUEUE (queue, obj_to_pin, desc);
684 sgen_pin_stats_register_object (obj_to_pin, GENERATION_NURSERY);
685 definitely_pinned [count] = obj_to_pin;
688 if (concurrent_collection_in_progress)
689 sgen_pinning_register_pinned_in_nursery (obj_to_pin);
691 next_pin_queue_entry:
695 sgen_client_nursery_objects_pinned (definitely_pinned, count);
696 stat_pinned_objects += count;
701 pin_objects_in_nursery (gboolean do_scan_objects, ScanCopyContext ctx)
705 if (nursery_section->pin_queue_first_entry == nursery_section->pin_queue_last_entry)
708 reduced_to = pin_objects_from_nursery_pin_queue (do_scan_objects, ctx);
709 nursery_section->pin_queue_last_entry = nursery_section->pin_queue_first_entry + reduced_to;
713 * This function is only ever called (via `collector_pin_object()` in `sgen-copy-object.h`)
714 * when we can't promote an object because we're out of memory.
717 sgen_pin_object (GCObject *object, GrayQueue *queue)
719 SGEN_ASSERT (0, sgen_ptr_in_nursery (object), "We're only supposed to use this for pinning nursery objects when out of memory.");
722 * All pinned objects are assumed to have been staged, so we need to stage as well.
723 * Also, the count of staged objects shows that "late pinning" happened.
725 sgen_pin_stage_ptr (object);
727 SGEN_PIN_OBJECT (object);
728 binary_protocol_pin (object, (gpointer)LOAD_VTABLE (object), safe_object_get_size (object));
731 sgen_pin_stats_register_object (object, GENERATION_NURSERY);
733 GRAY_OBJECT_ENQUEUE (queue, object, sgen_obj_get_descriptor_safe (object));
736 /* Sort the addresses in array in increasing order.
737 * Done using a by-the book heap sort. Which has decent and stable performance, is pretty cache efficient.
740 sgen_sort_addresses (void **array, size_t size)
745 for (i = 1; i < size; ++i) {
748 size_t parent = (child - 1) / 2;
750 if (array [parent] >= array [child])
753 tmp = array [parent];
754 array [parent] = array [child];
761 for (i = size - 1; i > 0; --i) {
764 array [i] = array [0];
770 while (root * 2 + 1 <= end) {
771 size_t child = root * 2 + 1;
773 if (child < end && array [child] < array [child + 1])
775 if (array [root] >= array [child])
779 array [root] = array [child];
788 * Scan the memory between start and end and queue values which could be pointers
789 * to the area between start_nursery and end_nursery for later consideration.
790 * Typically used for thread stacks.
793 sgen_conservatively_pin_objects_from (void **start, void **end, void *start_nursery, void *end_nursery, int pin_type)
797 SGEN_ASSERT (0, ((mword)start & (SIZEOF_VOID_P - 1)) == 0, "Why are we scanning for references in unaligned memory ?");
799 #if defined(VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE) && !defined(_WIN64)
800 VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE (start, (char*)end - (char*)start);
803 while (start < end) {
805 * *start can point to the middle of an object
806 * note: should we handle pointing at the end of an object?
807 * pinning in C# code disallows pointing at the end of an object
808 * but there is some small chance that an optimizing C compiler
809 * may keep the only reference to an object by pointing
810 * at the end of it. We ignore this small chance for now.
811 * Pointers to the end of an object are indistinguishable
812 * from pointers to the start of the next object in memory
813 * so if we allow that we'd need to pin two objects...
814 * We queue the pointer in an array, the
815 * array will then be sorted and uniqued. This way
816 * we can coalesce several pinning pointers and it should
817 * be faster since we'd do a memory scan with increasing
818 * addresses. Note: we can align the address to the allocation
819 * alignment, so the unique process is more effective.
821 mword addr = (mword)*start;
822 addr &= ~(ALLOC_ALIGN - 1);
823 if (addr >= (mword)start_nursery && addr < (mword)end_nursery) {
824 SGEN_LOG (6, "Pinning address %p from %p", (void*)addr, start);
825 sgen_pin_stage_ptr ((void*)addr);
826 binary_protocol_pin_stage (start, (void*)addr);
827 sgen_pin_stats_register_address ((char*)addr, pin_type);
833 SGEN_LOG (7, "found %d potential pinned heap pointers", count);
837 * The first thing we do in a collection is to identify pinned objects.
838 * This function considers all the areas of memory that need to be
839 * conservatively scanned.
842 pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx)
846 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);
847 /* objects pinned from the API are inside these roots */
848 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_PINNED], void **, start_root, RootRecord *, root) {
849 SGEN_LOG (6, "Pinned roots %p-%p", start_root, root->end_root);
850 sgen_conservatively_pin_objects_from (start_root, (void**)root->end_root, start_nursery, end_nursery, PIN_TYPE_OTHER);
851 } SGEN_HASH_TABLE_FOREACH_END;
852 /* now deal with the thread stacks
853 * in the future we should be able to conservatively scan only:
854 * *) the cpu registers
855 * *) the unmanaged stack frames
856 * *) the _last_ managed stack frame
857 * *) pointers slots in managed frames
859 sgen_client_scan_thread_data (start_nursery, end_nursery, FALSE, ctx);
863 single_arg_user_copy_or_mark (GCObject **obj, void *gc_data)
865 ScanCopyContext *ctx = (ScanCopyContext *)gc_data;
866 ctx->ops->copy_or_mark_object (obj, ctx->queue);
870 * The memory area from start_root to end_root contains pointers to objects.
871 * Their position is precisely described by @desc (this means that the pointer
872 * can be either NULL or the pointer to the start of an object).
873 * This functions copies them to to_space updates them.
875 * This function is not thread-safe!
878 precisely_scan_objects_from (void** start_root, void** end_root, char* n_start, char *n_end, SgenDescriptor desc, ScanCopyContext ctx)
880 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
881 SgenGrayQueue *queue = ctx.queue;
883 switch (desc & ROOT_DESC_TYPE_MASK) {
884 case ROOT_DESC_BITMAP:
885 desc >>= ROOT_DESC_TYPE_SHIFT;
887 if ((desc & 1) && *start_root) {
888 copy_func ((GCObject**)start_root, queue);
889 SGEN_LOG (9, "Overwrote root at %p with %p", start_root, *start_root);
895 case ROOT_DESC_COMPLEX: {
896 gsize *bitmap_data = (gsize *)sgen_get_complex_descriptor_bitmap (desc);
897 gsize bwords = (*bitmap_data) - 1;
898 void **start_run = start_root;
900 while (bwords-- > 0) {
901 gsize bmap = *bitmap_data++;
902 void **objptr = start_run;
904 if ((bmap & 1) && *objptr) {
905 copy_func ((GCObject**)objptr, queue);
906 SGEN_LOG (9, "Overwrote root at %p with %p", objptr, *objptr);
911 start_run += GC_BITS_PER_WORD;
915 case ROOT_DESC_USER: {
916 SgenUserRootMarkFunc marker = sgen_get_user_descriptor_func (desc);
917 marker (start_root, single_arg_user_copy_or_mark, &ctx);
920 case ROOT_DESC_RUN_LEN:
921 g_assert_not_reached ();
923 g_assert_not_reached ();
928 reset_heap_boundaries (void)
930 lowest_heap_address = ~(mword)0;
931 highest_heap_address = 0;
935 sgen_update_heap_boundaries (mword low, mword high)
940 old = lowest_heap_address;
943 } while (SGEN_CAS_PTR ((gpointer*)&lowest_heap_address, (gpointer)low, (gpointer)old) != (gpointer)old);
946 old = highest_heap_address;
949 } while (SGEN_CAS_PTR ((gpointer*)&highest_heap_address, (gpointer)high, (gpointer)old) != (gpointer)old);
953 * Allocate and setup the data structures needed to be able to allocate objects
954 * in the nursery. The nursery is stored in nursery_section.
959 GCMemSection *section;
966 SGEN_LOG (2, "Allocating nursery size: %zu", (size_t)sgen_nursery_size);
967 /* later we will alloc a larger area for the nursery but only activate
968 * what we need. The rest will be used as expansion if we have too many pinned
969 * objects in the existing nursery.
971 /* FIXME: handle OOM */
972 section = (GCMemSection *)sgen_alloc_internal (INTERNAL_MEM_SECTION);
974 alloc_size = sgen_nursery_size;
976 /* If there isn't enough space even for the nursery we should simply abort. */
977 g_assert (sgen_memgov_try_alloc_space (alloc_size, SPACE_NURSERY));
979 data = (char *)major_collector.alloc_heap (alloc_size, alloc_size, DEFAULT_NURSERY_BITS);
980 sgen_update_heap_boundaries ((mword)data, (mword)(data + sgen_nursery_size));
981 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 ());
982 section->data = section->next_data = data;
983 section->size = alloc_size;
984 section->end_data = data + sgen_nursery_size;
985 scan_starts = (alloc_size + SCAN_START_SIZE - 1) / SCAN_START_SIZE;
986 section->scan_starts = (char **)sgen_alloc_internal_dynamic (sizeof (char*) * scan_starts, INTERNAL_MEM_SCAN_STARTS, TRUE);
987 section->num_scan_start = scan_starts;
989 nursery_section = section;
991 sgen_nursery_allocator_set_nursery_bounds (data, data + sgen_nursery_size);
995 mono_gc_get_logfile (void)
997 return gc_debug_file;
1001 scan_finalizer_entries (SgenPointerQueue *fin_queue, ScanCopyContext ctx)
1003 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
1004 SgenGrayQueue *queue = ctx.queue;
1007 for (i = 0; i < fin_queue->next_slot; ++i) {
1008 GCObject *obj = (GCObject *)fin_queue->data [i];
1011 SGEN_LOG (5, "Scan of fin ready object: %p (%s)\n", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
1012 copy_func ((GCObject**)&fin_queue->data [i], queue);
1017 generation_name (int generation)
1019 switch (generation) {
1020 case GENERATION_NURSERY: return "nursery";
1021 case GENERATION_OLD: return "old";
1022 default: g_assert_not_reached ();
1027 sgen_generation_name (int generation)
1029 return generation_name (generation);
1033 finish_gray_stack (int generation, ScanCopyContext ctx)
1037 int done_with_ephemerons, ephemeron_rounds = 0;
1038 char *start_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_start () : NULL;
1039 char *end_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_end () : (char*)-1;
1040 SgenGrayQueue *queue = ctx.queue;
1042 binary_protocol_finish_gray_stack_start (sgen_timestamp (), generation);
1044 * We copied all the reachable objects. Now it's the time to copy
1045 * the objects that were not referenced by the roots, but by the copied objects.
1046 * we built a stack of objects pointed to by gray_start: they are
1047 * additional roots and we may add more items as we go.
1048 * We loop until gray_start == gray_objects which means no more objects have
1049 * been added. Note this is iterative: no recursion is involved.
1050 * We need to walk the LO list as well in search of marked big objects
1051 * (use a flag since this is needed only on major collections). We need to loop
1052 * here as well, so keep a counter of marked LO (increasing it in copy_object).
1053 * To achieve better cache locality and cache usage, we drain the gray stack
1054 * frequently, after each object is copied, and just finish the work here.
1056 sgen_drain_gray_stack (ctx);
1058 SGEN_LOG (2, "%s generation done", generation_name (generation));
1061 Reset bridge data, we might have lingering data from a previous collection if this is a major
1062 collection trigged by minor overflow.
1064 We must reset the gathered bridges since their original block might be evacuated due to major
1065 fragmentation in the meanwhile and the bridge code should not have to deal with that.
1067 if (sgen_client_bridge_need_processing ())
1068 sgen_client_bridge_reset_data ();
1071 * Mark all strong toggleref objects. This must be done before we walk ephemerons or finalizers
1072 * to ensure they see the full set of live objects.
1074 sgen_client_mark_togglerefs (start_addr, end_addr, ctx);
1077 * Walk the ephemeron tables marking all values with reachable keys. This must be completely done
1078 * before processing finalizable objects and non-tracking weak links to avoid finalizing/clearing
1079 * objects that are in fact reachable.
1081 done_with_ephemerons = 0;
1083 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1084 sgen_drain_gray_stack (ctx);
1086 } while (!done_with_ephemerons);
1088 if (sgen_client_bridge_need_processing ()) {
1089 /*Make sure the gray stack is empty before we process bridge objects so we get liveness right*/
1090 sgen_drain_gray_stack (ctx);
1091 sgen_collect_bridge_objects (generation, ctx);
1092 if (generation == GENERATION_OLD)
1093 sgen_collect_bridge_objects (GENERATION_NURSERY, ctx);
1096 Do the first bridge step here, as the collector liveness state will become useless after that.
1098 An important optimization is to only proccess the possibly dead part of the object graph and skip
1099 over all live objects as we transitively know everything they point must be alive too.
1101 The above invariant is completely wrong if we let the gray queue be drained and mark/copy everything.
1103 This has the unfortunate side effect of making overflow collections perform the first step twice, but
1104 given we now have heuristics that perform major GC in anticipation of minor overflows this should not
1107 sgen_client_bridge_processing_stw_step ();
1111 Make sure we drain the gray stack before processing disappearing links and finalizers.
1112 If we don't make sure it is empty we might wrongly see a live object as dead.
1114 sgen_drain_gray_stack (ctx);
1117 We must clear weak links that don't track resurrection before processing object ready for
1118 finalization so they can be cleared before that.
1120 sgen_null_link_in_range (generation, ctx, FALSE);
1121 if (generation == GENERATION_OLD)
1122 sgen_null_link_in_range (GENERATION_NURSERY, ctx, FALSE);
1125 /* walk the finalization queue and move also the objects that need to be
1126 * finalized: use the finalized objects as new roots so the objects they depend
1127 * on are also not reclaimed. As with the roots above, only objects in the nursery
1128 * are marked/copied.
1130 sgen_finalize_in_range (generation, ctx);
1131 if (generation == GENERATION_OLD)
1132 sgen_finalize_in_range (GENERATION_NURSERY, ctx);
1133 /* drain the new stack that might have been created */
1134 SGEN_LOG (6, "Precise scan of gray area post fin");
1135 sgen_drain_gray_stack (ctx);
1138 * This must be done again after processing finalizable objects since CWL slots are cleared only after the key is finalized.
1140 done_with_ephemerons = 0;
1142 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1143 sgen_drain_gray_stack (ctx);
1145 } while (!done_with_ephemerons);
1147 sgen_client_clear_unreachable_ephemerons (ctx);
1150 * We clear togglerefs only after all possible chances of revival are done.
1151 * This is semantically more inline with what users expect and it allows for
1152 * user finalizers to correctly interact with TR objects.
1154 sgen_client_clear_togglerefs (start_addr, end_addr, ctx);
1157 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);
1160 * handle disappearing links
1161 * Note we do this after checking the finalization queue because if an object
1162 * survives (at least long enough to be finalized) we don't clear the link.
1163 * This also deals with a possible issue with the monitor reclamation: with the Boehm
1164 * GC a finalized object my lose the monitor because it is cleared before the finalizer is
1167 g_assert (sgen_gray_object_queue_is_empty (queue));
1169 sgen_null_link_in_range (generation, ctx, TRUE);
1170 if (generation == GENERATION_OLD)
1171 sgen_null_link_in_range (GENERATION_NURSERY, ctx, TRUE);
1172 if (sgen_gray_object_queue_is_empty (queue))
1174 sgen_drain_gray_stack (ctx);
1177 g_assert (sgen_gray_object_queue_is_empty (queue));
1179 sgen_gray_object_queue_trim_free_list (queue);
1180 binary_protocol_finish_gray_stack_end (sgen_timestamp (), generation);
1184 sgen_check_section_scan_starts (GCMemSection *section)
1187 for (i = 0; i < section->num_scan_start; ++i) {
1188 if (section->scan_starts [i]) {
1189 mword size = safe_object_get_size ((GCObject*) section->scan_starts [i]);
1190 SGEN_ASSERT (0, size >= SGEN_CLIENT_MINIMUM_OBJECT_SIZE && size <= MAX_SMALL_OBJ_SIZE, "Weird object size at scan starts.");
1196 check_scan_starts (void)
1198 if (!do_scan_starts_check)
1200 sgen_check_section_scan_starts (nursery_section);
1201 major_collector.check_scan_starts ();
1205 scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx)
1209 SGEN_HASH_TABLE_FOREACH (&roots_hash [root_type], void **, start_root, RootRecord *, root) {
1210 SGEN_LOG (6, "Precise root scan %p-%p (desc: %p)", start_root, root->end_root, (void*)root->root_desc);
1211 precisely_scan_objects_from (start_root, (void**)root->end_root, addr_start, addr_end, root->root_desc, ctx);
1212 } SGEN_HASH_TABLE_FOREACH_END;
1218 static gboolean inited = FALSE;
1223 mono_counters_register ("Collection max time", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME | MONO_COUNTER_MONOTONIC, &time_max);
1225 mono_counters_register ("Minor fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pre_collection_fragment_clear);
1226 mono_counters_register ("Minor pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pinning);
1227 mono_counters_register ("Minor scan remembered set", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_remsets);
1228 mono_counters_register ("Minor scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_pinned);
1229 mono_counters_register ("Minor scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_roots);
1230 mono_counters_register ("Minor fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_fragment_creation);
1232 mono_counters_register ("Major fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pre_collection_fragment_clear);
1233 mono_counters_register ("Major pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pinning);
1234 mono_counters_register ("Major scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_pinned);
1235 mono_counters_register ("Major scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_roots);
1236 mono_counters_register ("Major scan mod union", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_mod_union);
1237 mono_counters_register ("Major finish gray stack", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_finish_gray_stack);
1238 mono_counters_register ("Major free big objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_free_bigobjs);
1239 mono_counters_register ("Major LOS sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_los_sweep);
1240 mono_counters_register ("Major sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_sweep);
1241 mono_counters_register ("Major fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_fragment_creation);
1243 mono_counters_register ("Number of pinned objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_pinned_objects);
1245 #ifdef HEAVY_STATISTICS
1246 mono_counters_register ("WBarrier remember pointer", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_add_to_global_remset);
1247 mono_counters_register ("WBarrier arrayref copy", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_arrayref_copy);
1248 mono_counters_register ("WBarrier generic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store);
1249 mono_counters_register ("WBarrier generic atomic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store_atomic);
1250 mono_counters_register ("WBarrier set root", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_set_root);
1252 mono_counters_register ("# objects allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_alloced_degraded);
1253 mono_counters_register ("bytes allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_bytes_alloced_degraded);
1255 mono_counters_register ("# copy_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_nursery);
1256 mono_counters_register ("# objects copied (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_nursery);
1257 mono_counters_register ("# copy_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_major);
1258 mono_counters_register ("# objects copied (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_major);
1260 mono_counters_register ("# scan_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_nursery);
1261 mono_counters_register ("# scan_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_major);
1263 mono_counters_register ("Slots allocated in vain", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_slots_allocated_in_vain);
1265 mono_counters_register ("# nursery copy_object() failed from space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_from_space);
1266 mono_counters_register ("# nursery copy_object() failed forwarded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_forwarded);
1267 mono_counters_register ("# nursery copy_object() failed pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_pinned);
1268 mono_counters_register ("# nursery copy_object() failed to space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_to_space);
1270 sgen_nursery_allocator_init_heavy_stats ();
1278 reset_pinned_from_failed_allocation (void)
1280 bytes_pinned_from_failed_allocation = 0;
1284 sgen_set_pinned_from_failed_allocation (mword objsize)
1286 bytes_pinned_from_failed_allocation += objsize;
1290 sgen_collection_is_concurrent (void)
1292 switch (current_collection_generation) {
1293 case GENERATION_NURSERY:
1295 case GENERATION_OLD:
1296 return concurrent_collection_in_progress;
1298 g_error ("Invalid current generation %d", current_collection_generation);
1304 sgen_concurrent_collection_in_progress (void)
1306 return concurrent_collection_in_progress;
1310 SgenThreadPoolJob job;
1311 SgenObjectOperations *ops;
1315 job_remembered_set_scan (void *worker_data_untyped, SgenThreadPoolJob *job)
1317 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1318 ScanJob *job_data = (ScanJob*)job;
1319 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1320 remset.scan_remsets (ctx);
1324 SgenThreadPoolJob job;
1325 SgenObjectOperations *ops;
1329 } ScanFromRegisteredRootsJob;
1332 job_scan_from_registered_roots (void *worker_data_untyped, SgenThreadPoolJob *job)
1334 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1335 ScanFromRegisteredRootsJob *job_data = (ScanFromRegisteredRootsJob*)job;
1336 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1338 scan_from_registered_roots (job_data->heap_start, job_data->heap_end, job_data->root_type, ctx);
1342 SgenThreadPoolJob job;
1343 SgenObjectOperations *ops;
1346 } ScanThreadDataJob;
1349 job_scan_thread_data (void *worker_data_untyped, SgenThreadPoolJob *job)
1351 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1352 ScanThreadDataJob *job_data = (ScanThreadDataJob*)job;
1353 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1355 sgen_client_scan_thread_data (job_data->heap_start, job_data->heap_end, TRUE, ctx);
1359 SgenThreadPoolJob job;
1360 SgenObjectOperations *ops;
1361 SgenPointerQueue *queue;
1362 } ScanFinalizerEntriesJob;
1365 job_scan_finalizer_entries (void *worker_data_untyped, SgenThreadPoolJob *job)
1367 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1368 ScanFinalizerEntriesJob *job_data = (ScanFinalizerEntriesJob*)job;
1369 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1371 scan_finalizer_entries (job_data->queue, ctx);
1375 job_scan_major_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1377 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1378 ScanJob *job_data = (ScanJob*)job;
1379 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1381 g_assert (concurrent_collection_in_progress);
1382 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx);
1386 job_scan_los_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1388 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1389 ScanJob *job_data = (ScanJob*)job;
1390 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1392 g_assert (concurrent_collection_in_progress);
1393 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx);
1397 job_mod_union_preclean (void *worker_data_untyped, SgenThreadPoolJob *job)
1399 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1400 ScanJob *job_data = (ScanJob*)job;
1401 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1403 g_assert (concurrent_collection_in_progress);
1405 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx);
1406 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx);
1408 sgen_scan_pin_queue_objects (ctx);
1412 init_gray_queue (gboolean use_workers)
1415 sgen_workers_init_distribute_gray_queue ();
1416 sgen_gray_object_queue_init (&gray_queue, NULL);
1420 enqueue_scan_from_roots_jobs (char *heap_start, char *heap_end, SgenObjectOperations *ops, gboolean enqueue)
1422 ScanFromRegisteredRootsJob *scrrj;
1423 ScanThreadDataJob *stdj;
1424 ScanFinalizerEntriesJob *sfej;
1426 /* registered roots, this includes static fields */
1428 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots normal", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1430 scrrj->heap_start = heap_start;
1431 scrrj->heap_end = heap_end;
1432 scrrj->root_type = ROOT_TYPE_NORMAL;
1433 sgen_workers_enqueue_job (&scrrj->job, enqueue);
1435 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots wbarrier", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1437 scrrj->heap_start = heap_start;
1438 scrrj->heap_end = heap_end;
1439 scrrj->root_type = ROOT_TYPE_WBARRIER;
1440 sgen_workers_enqueue_job (&scrrj->job, enqueue);
1444 stdj = (ScanThreadDataJob*)sgen_thread_pool_job_alloc ("scan thread data", job_scan_thread_data, sizeof (ScanThreadDataJob));
1446 stdj->heap_start = heap_start;
1447 stdj->heap_end = heap_end;
1448 sgen_workers_enqueue_job (&stdj->job, enqueue);
1450 /* Scan the list of objects ready for finalization. */
1452 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1453 sfej->queue = &fin_ready_queue;
1455 sgen_workers_enqueue_job (&sfej->job, enqueue);
1457 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan critical finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1458 sfej->queue = &critical_fin_queue;
1460 sgen_workers_enqueue_job (&sfej->job, enqueue);
1464 * Perform a nursery collection.
1466 * Return whether any objects were late-pinned due to being out of memory.
1469 collect_nursery (const char *reason, gboolean is_overflow, SgenGrayQueue *unpin_queue, gboolean finish_up_concurrent_mark)
1471 gboolean needs_major;
1472 size_t max_garbage_amount;
1474 mword fragment_total;
1476 SgenObjectOperations *object_ops = &sgen_minor_collector.serial_ops;
1477 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, &gray_queue);
1481 if (disable_minor_collections)
1484 TV_GETTIME (last_minor_collection_start_tv);
1485 atv = last_minor_collection_start_tv;
1487 binary_protocol_collection_begin (gc_stats.minor_gc_count, GENERATION_NURSERY);
1489 if (do_verify_nursery || do_dump_nursery_content)
1490 sgen_debug_verify_nursery (do_dump_nursery_content);
1492 current_collection_generation = GENERATION_NURSERY;
1494 SGEN_ASSERT (0, !sgen_collection_is_concurrent (), "Why is the nursery collection concurrent?");
1496 reset_pinned_from_failed_allocation ();
1498 check_scan_starts ();
1500 sgen_nursery_alloc_prepare_for_minor ();
1504 nursery_next = sgen_nursery_alloc_get_upper_alloc_bound ();
1505 /* FIXME: optimize later to use the higher address where an object can be present */
1506 nursery_next = MAX (nursery_next, sgen_get_nursery_end ());
1508 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 ()));
1509 max_garbage_amount = nursery_next - sgen_get_nursery_start ();
1510 g_assert (nursery_section->size >= max_garbage_amount);
1512 /* world must be stopped already */
1514 time_minor_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1516 sgen_client_pre_collection_checks ();
1518 nursery_section->next_data = nursery_next;
1520 major_collector.start_nursery_collection ();
1522 sgen_memgov_minor_collection_start ();
1524 init_gray_queue (FALSE);
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 sgen_workers_enqueue_job (&sj->job, FALSE);
1560 /* we don't have complete write barrier yet, so we scan all the old generation sections */
1562 time_minor_scan_remsets += TV_ELAPSED (atv, btv);
1563 SGEN_LOG (2, "Old generation scan: %lld usecs", (long long)TV_ELAPSED (atv, btv));
1565 sgen_pin_stats_report ();
1567 /* FIXME: Why do we do this at this specific, seemingly random, point? */
1568 sgen_client_collecting_minor (&fin_ready_queue, &critical_fin_queue);
1571 time_minor_scan_pinned += TV_ELAPSED (btv, atv);
1573 enqueue_scan_from_roots_jobs (sgen_get_nursery_start (), nursery_next, object_ops, FALSE);
1576 time_minor_scan_roots += TV_ELAPSED (atv, btv);
1578 finish_gray_stack (GENERATION_NURSERY, ctx);
1581 time_minor_finish_gray_stack += TV_ELAPSED (btv, atv);
1582 sgen_client_binary_protocol_mark_end (GENERATION_NURSERY);
1584 if (objects_pinned) {
1585 sgen_optimize_pin_queue ();
1586 sgen_pinning_setup_section (nursery_section);
1590 * This is the latest point at which we can do this check, because
1591 * sgen_build_nursery_fragments() unpins nursery objects again.
1593 if (remset_consistency_checks)
1594 sgen_check_remset_consistency ();
1596 /* walk the pin_queue, build up the fragment list of free memory, unmark
1597 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1600 sgen_client_binary_protocol_reclaim_start (GENERATION_NURSERY);
1601 fragment_total = sgen_build_nursery_fragments (nursery_section, unpin_queue);
1602 if (!fragment_total)
1605 /* Clear TLABs for all threads */
1606 sgen_clear_tlabs ();
1608 sgen_client_binary_protocol_reclaim_end (GENERATION_NURSERY);
1610 time_minor_fragment_creation += TV_ELAPSED (atv, btv);
1611 SGEN_LOG (2, "Fragment creation: %lld usecs, %lu bytes available", (long long)TV_ELAPSED (atv, btv), (unsigned long)fragment_total);
1613 if (remset_consistency_checks)
1614 sgen_check_major_refs ();
1616 major_collector.finish_nursery_collection ();
1618 TV_GETTIME (last_minor_collection_end_tv);
1619 gc_stats.minor_gc_time += TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv);
1621 sgen_debug_dump_heap ("minor", gc_stats.minor_gc_count - 1, NULL);
1623 /* prepare the pin queue for the next collection */
1624 sgen_finish_pinning ();
1625 if (sgen_have_pending_finalizers ()) {
1626 SGEN_LOG (4, "Finalizer-thread wakeup");
1627 sgen_client_finalize_notify ();
1629 sgen_pin_stats_reset ();
1630 /* clear cemented hash */
1631 sgen_cement_clear_below_threshold ();
1633 g_assert (sgen_gray_object_queue_is_empty (&gray_queue));
1635 remset.finish_minor_collection ();
1637 check_scan_starts ();
1639 binary_protocol_flush_buffers (FALSE);
1641 sgen_memgov_minor_collection_end (reason, is_overflow);
1643 /*objects are late pinned because of lack of memory, so a major is a good call*/
1644 needs_major = objects_pinned > 0;
1645 current_collection_generation = -1;
1648 binary_protocol_collection_end (gc_stats.minor_gc_count - 1, GENERATION_NURSERY, 0, 0);
1650 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1651 sgen_check_nursery_objects_pinned (unpin_queue != NULL);
1657 COPY_OR_MARK_FROM_ROOTS_SERIAL,
1658 COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT,
1659 COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT
1660 } CopyOrMarkFromRootsMode;
1663 major_copy_or_mark_from_roots (size_t *old_next_pin_slot, CopyOrMarkFromRootsMode mode, SgenObjectOperations *object_ops)
1668 /* FIXME: only use these values for the precise scan
1669 * note that to_space pointers should be excluded anyway...
1671 char *heap_start = NULL;
1672 char *heap_end = (char*)-1;
1673 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, WORKERS_DISTRIBUTE_GRAY_QUEUE);
1674 gboolean concurrent = mode != COPY_OR_MARK_FROM_ROOTS_SERIAL;
1676 SGEN_ASSERT (0, !!concurrent == !!concurrent_collection_in_progress, "We've been called with the wrong mode.");
1678 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1679 /*This cleans up unused fragments */
1680 sgen_nursery_allocator_prepare_for_pinning ();
1682 if (do_concurrent_checks)
1683 sgen_debug_check_nursery_is_clean ();
1685 /* The concurrent collector doesn't touch the nursery. */
1686 sgen_nursery_alloc_prepare_for_major ();
1689 init_gray_queue (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1693 /* Pinning depends on this */
1694 sgen_clear_nursery_fragments ();
1696 if (whole_heap_check_before_collection)
1697 sgen_check_whole_heap (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT);
1700 time_major_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1702 if (!sgen_collection_is_concurrent ())
1703 nursery_section->next_data = sgen_get_nursery_end ();
1704 /* we should also coalesce scanning from sections close to each other
1705 * and deal with pointers outside of the sections later.
1710 sgen_client_pre_collection_checks ();
1712 if (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1713 /* Remsets are not useful for a major collection */
1714 remset.clear_cards ();
1717 sgen_process_fin_stage_entries ();
1720 sgen_init_pinning ();
1721 SGEN_LOG (6, "Collecting pinned addresses");
1722 pin_from_roots ((void*)lowest_heap_address, (void*)highest_heap_address, ctx);
1723 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1724 /* Pin cemented objects that were forced */
1725 sgen_pin_cemented_objects ();
1727 sgen_optimize_pin_queue ();
1728 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1730 * Cemented objects that are in the pinned list will be marked. When
1731 * marking concurrently we won't mark mod-union cards for these objects.
1732 * Instead they will remain cemented until the next major collection,
1733 * when we will recheck if they are still pinned in the roots.
1735 sgen_cement_force_pinned ();
1738 sgen_client_collecting_major_1 ();
1741 * pin_queue now contains all candidate pointers, sorted and
1742 * uniqued. We must do two passes now to figure out which
1743 * objects are pinned.
1745 * The first is to find within the pin_queue the area for each
1746 * section. This requires that the pin_queue be sorted. We
1747 * also process the LOS objects and pinned chunks here.
1749 * The second, destructive, pass is to reduce the section
1750 * areas to pointers to the actually pinned objects.
1752 SGEN_LOG (6, "Pinning from sections");
1753 /* first pass for the sections */
1754 sgen_find_section_pin_queue_start_end (nursery_section);
1755 /* identify possible pointers to the insize of large objects */
1756 SGEN_LOG (6, "Pinning from large objects");
1757 for (bigobj = los_object_list; bigobj; bigobj = bigobj->next) {
1759 if (sgen_find_optimized_pin_queue_area ((char*)bigobj->data, (char*)bigobj->data + sgen_los_object_size (bigobj), &dummy, &dummy)) {
1760 binary_protocol_pin (bigobj->data, (gpointer)LOAD_VTABLE (bigobj->data), safe_object_get_size (bigobj->data));
1762 if (sgen_los_object_is_pinned (bigobj->data)) {
1763 SGEN_ASSERT (0, mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, "LOS objects can only be pinned here after concurrent marking.");
1766 sgen_los_pin_object (bigobj->data);
1767 if (SGEN_OBJECT_HAS_REFERENCES (bigobj->data))
1768 GRAY_OBJECT_ENQUEUE (WORKERS_DISTRIBUTE_GRAY_QUEUE, bigobj->data, sgen_obj_get_descriptor ((GCObject*)bigobj->data));
1769 sgen_pin_stats_register_object (bigobj->data, GENERATION_OLD);
1770 SGEN_LOG (6, "Marked large object %p (%s) size: %lu from roots", bigobj->data,
1771 sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (bigobj->data)),
1772 (unsigned long)sgen_los_object_size (bigobj));
1774 sgen_client_pinned_los_object (bigobj->data);
1778 pin_objects_in_nursery (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT, ctx);
1779 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1780 sgen_check_nursery_objects_pinned (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1782 major_collector.pin_objects (WORKERS_DISTRIBUTE_GRAY_QUEUE);
1783 if (old_next_pin_slot)
1784 *old_next_pin_slot = sgen_get_pinned_count ();
1787 * We don't actually pin when starting a concurrent collection, so the remset
1788 * consistency check won't work.
1790 if (remset_consistency_checks && mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT)
1791 sgen_check_remset_consistency ();
1794 time_major_pinning += TV_ELAPSED (atv, btv);
1795 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), (long long)TV_ELAPSED (atv, btv));
1796 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1798 major_collector.init_to_space ();
1800 SGEN_ASSERT (0, sgen_workers_all_done (), "Why are the workers not done when we start or finish a major collection?");
1801 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1802 if (sgen_workers_have_idle_work ()) {
1804 * We force the finish of the worker with the new object ops context
1805 * which can also do copying. We need to have finished pinning.
1807 sgen_workers_start_all_workers (object_ops, NULL);
1808 sgen_workers_join ();
1812 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1813 main_gc_thread = mono_native_thread_self ();
1816 sgen_client_collecting_major_2 ();
1819 time_major_scan_pinned += TV_ELAPSED (btv, atv);
1821 sgen_client_collecting_major_3 (&fin_ready_queue, &critical_fin_queue);
1823 enqueue_scan_from_roots_jobs (heap_start, heap_end, object_ops, FALSE);
1826 time_major_scan_roots += TV_ELAPSED (atv, btv);
1829 * We start the concurrent worker after pinning and after we scanned the roots
1830 * in order to make sure that the worker does not finish before handling all
1833 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1834 if (precleaning_enabled) {
1836 /* Mod union preclean job */
1837 sj = (ScanJob*)sgen_thread_pool_job_alloc ("preclean mod union cardtable", job_mod_union_preclean, sizeof (ScanJob));
1838 sj->ops = object_ops;
1839 sgen_workers_start_all_workers (object_ops, &sj->job);
1841 sgen_workers_start_all_workers (object_ops, NULL);
1843 gray_queue_enable_redirect (WORKERS_DISTRIBUTE_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 sgen_workers_enqueue_job (&sj->job, FALSE);
1854 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan LOS mod union cardtable", job_scan_los_mod_union_card_table, sizeof (ScanJob));
1855 sj->ops = object_ops;
1856 sgen_workers_enqueue_job (&sj->job, FALSE);
1859 time_major_scan_mod_union += TV_ELAPSED (btv, atv);
1862 sgen_pin_stats_report ();
1866 major_finish_copy_or_mark (CopyOrMarkFromRootsMode mode)
1868 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1869 sgen_finish_pinning ();
1871 sgen_pin_stats_reset ();
1873 if (do_concurrent_checks)
1874 sgen_debug_check_nursery_is_clean ();
1879 major_start_collection (const char *reason, gboolean concurrent, size_t *old_next_pin_slot)
1881 SgenObjectOperations *object_ops;
1883 binary_protocol_collection_begin (gc_stats.major_gc_count, GENERATION_OLD);
1885 current_collection_generation = GENERATION_OLD;
1887 sgen_workers_assert_gray_queue_is_empty ();
1890 sgen_cement_reset ();
1893 g_assert (major_collector.is_concurrent);
1894 concurrent_collection_in_progress = TRUE;
1896 object_ops = &major_collector.major_ops_concurrent_start;
1898 object_ops = &major_collector.major_ops_serial;
1901 reset_pinned_from_failed_allocation ();
1903 sgen_memgov_major_collection_start (concurrent, reason);
1905 //count_ref_nonref_objs ();
1906 //consistency_check ();
1908 check_scan_starts ();
1911 SGEN_LOG (1, "Start major collection %d", gc_stats.major_gc_count);
1912 gc_stats.major_gc_count ++;
1914 if (major_collector.start_major_collection)
1915 major_collector.start_major_collection ();
1917 major_copy_or_mark_from_roots (old_next_pin_slot, concurrent ? COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT : COPY_OR_MARK_FROM_ROOTS_SERIAL, object_ops);
1918 major_finish_copy_or_mark (concurrent ? COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT : COPY_OR_MARK_FROM_ROOTS_SERIAL);
1922 major_finish_collection (const char *reason, gboolean is_overflow, size_t old_next_pin_slot, gboolean forced)
1924 ScannedObjectCounts counts;
1925 SgenObjectOperations *object_ops;
1926 mword fragment_total;
1932 if (concurrent_collection_in_progress) {
1933 object_ops = &major_collector.major_ops_concurrent_finish;
1935 major_copy_or_mark_from_roots (NULL, COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, object_ops);
1937 major_finish_copy_or_mark (COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT);
1939 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1940 main_gc_thread = NULL;
1943 object_ops = &major_collector.major_ops_serial;
1946 sgen_workers_assert_gray_queue_is_empty ();
1948 /* all the objects in the heap */
1949 finish_gray_stack (GENERATION_OLD, CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, &gray_queue));
1951 time_major_finish_gray_stack += TV_ELAPSED (btv, atv);
1953 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after joining");
1955 if (objects_pinned) {
1956 g_assert (!concurrent_collection_in_progress);
1959 * This is slow, but we just OOM'd.
1961 * See comment at `sgen_pin_queue_clear_discarded_entries` for how the pin
1962 * queue is laid out at this point.
1964 sgen_pin_queue_clear_discarded_entries (nursery_section, old_next_pin_slot);
1966 * We need to reestablish all pinned nursery objects in the pin queue
1967 * because they're needed for fragment creation. Unpinning happens by
1968 * walking the whole queue, so it's not necessary to reestablish where major
1969 * heap block pins are - all we care is that they're still in there
1972 sgen_optimize_pin_queue ();
1973 sgen_find_section_pin_queue_start_end (nursery_section);
1977 reset_heap_boundaries ();
1978 sgen_update_heap_boundaries ((mword)sgen_get_nursery_start (), (mword)sgen_get_nursery_end ());
1980 /* walk the pin_queue, build up the fragment list of free memory, unmark
1981 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1984 fragment_total = sgen_build_nursery_fragments (nursery_section, NULL);
1985 if (!fragment_total)
1987 SGEN_LOG (4, "Free space in nursery after major %ld", (long)fragment_total);
1989 if (do_concurrent_checks && concurrent_collection_in_progress)
1990 sgen_debug_check_nursery_is_clean ();
1992 /* prepare the pin queue for the next collection */
1993 sgen_finish_pinning ();
1995 /* Clear TLABs for all threads */
1996 sgen_clear_tlabs ();
1998 sgen_pin_stats_reset ();
2000 sgen_cement_clear_below_threshold ();
2002 if (check_mark_bits_after_major_collection)
2003 sgen_check_heap_marked (concurrent_collection_in_progress);
2006 time_major_fragment_creation += TV_ELAPSED (atv, btv);
2008 binary_protocol_sweep_begin (GENERATION_OLD, !major_collector.sweeps_lazily);
2009 sgen_memgov_major_pre_sweep ();
2012 time_major_free_bigobjs += TV_ELAPSED (btv, atv);
2017 time_major_los_sweep += TV_ELAPSED (atv, btv);
2019 major_collector.sweep ();
2021 binary_protocol_sweep_end (GENERATION_OLD, !major_collector.sweeps_lazily);
2024 time_major_sweep += TV_ELAPSED (btv, atv);
2026 sgen_debug_dump_heap ("major", gc_stats.major_gc_count - 1, reason);
2028 if (sgen_have_pending_finalizers ()) {
2029 SGEN_LOG (4, "Finalizer-thread wakeup");
2030 sgen_client_finalize_notify ();
2033 g_assert (sgen_gray_object_queue_is_empty (&gray_queue));
2035 sgen_memgov_major_collection_end (forced, concurrent_collection_in_progress, reason, is_overflow);
2036 current_collection_generation = -1;
2038 memset (&counts, 0, sizeof (ScannedObjectCounts));
2039 major_collector.finish_major_collection (&counts);
2041 sgen_workers_assert_gray_queue_is_empty ();
2043 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after major collection has finished");
2044 if (concurrent_collection_in_progress)
2045 concurrent_collection_in_progress = FALSE;
2047 check_scan_starts ();
2049 binary_protocol_flush_buffers (FALSE);
2051 //consistency_check ();
2053 binary_protocol_collection_end (gc_stats.major_gc_count - 1, GENERATION_OLD, counts.num_scanned_objects, counts.num_unique_scanned_objects);
2057 major_do_collection (const char *reason, gboolean is_overflow, gboolean forced)
2059 TV_DECLARE (time_start);
2060 TV_DECLARE (time_end);
2061 size_t old_next_pin_slot;
2063 if (disable_major_collections)
2066 if (major_collector.get_and_reset_num_major_objects_marked) {
2067 long long num_marked = major_collector.get_and_reset_num_major_objects_marked ();
2068 g_assert (!num_marked);
2071 /* world must be stopped already */
2072 TV_GETTIME (time_start);
2074 major_start_collection (reason, FALSE, &old_next_pin_slot);
2075 major_finish_collection (reason, is_overflow, old_next_pin_slot, forced);
2077 TV_GETTIME (time_end);
2078 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2080 /* FIXME: also report this to the user, preferably in gc-end. */
2081 if (major_collector.get_and_reset_num_major_objects_marked)
2082 major_collector.get_and_reset_num_major_objects_marked ();
2084 return bytes_pinned_from_failed_allocation > 0;
2088 major_start_concurrent_collection (const char *reason)
2090 TV_DECLARE (time_start);
2091 TV_DECLARE (time_end);
2092 long long num_objects_marked;
2094 if (disable_major_collections)
2097 TV_GETTIME (time_start);
2098 SGEN_TV_GETTIME (time_major_conc_collection_start);
2100 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2101 g_assert (num_objects_marked == 0);
2103 binary_protocol_concurrent_start ();
2105 // FIXME: store reason and pass it when finishing
2106 major_start_collection (reason, TRUE, NULL);
2107 SGEN_ASSERT (0, sgen_gray_object_queue_is_empty (&gray_queue), "What's left in the 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 SGEN_ASSERT (0, sgen_gray_object_queue_is_empty (&gray_queue), "Why is the gray queue not empty before we have started doing anything?");
2124 return sgen_workers_all_done ();
2128 major_update_concurrent_collection (void)
2130 TV_DECLARE (total_start);
2131 TV_DECLARE (total_end);
2133 TV_GETTIME (total_start);
2135 binary_protocol_concurrent_update ();
2137 major_collector.update_cardtable_mod_union ();
2138 sgen_los_update_cardtable_mod_union ();
2140 TV_GETTIME (total_end);
2141 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end);
2145 major_finish_concurrent_collection (gboolean forced)
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 major_finish_collection ("finishing", FALSE, -1, forced);
2174 if (whole_heap_check_before_collection)
2175 sgen_check_whole_heap (FALSE);
2177 TV_GETTIME (total_end);
2178 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end) - TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv);
2180 current_collection_generation = -1;
2184 * Ensure an allocation request for @size will succeed by freeing enough memory.
2186 * LOCKING: The GC lock MUST be held.
2189 sgen_ensure_free_space (size_t size, int generation)
2191 int generation_to_collect = -1;
2192 const char *reason = NULL;
2194 if (generation == GENERATION_OLD) {
2195 if (sgen_need_major_collection (size)) {
2196 reason = "LOS overflow";
2197 generation_to_collect = GENERATION_OLD;
2200 if (degraded_mode) {
2201 if (sgen_need_major_collection (size)) {
2202 reason = "Degraded mode overflow";
2203 generation_to_collect = GENERATION_OLD;
2205 } else if (sgen_need_major_collection (size)) {
2206 reason = concurrent_collection_in_progress ? "Forced finish concurrent collection" : "Minor allowance";
2207 generation_to_collect = GENERATION_OLD;
2209 generation_to_collect = GENERATION_NURSERY;
2210 reason = "Nursery full";
2214 if (generation_to_collect == -1) {
2215 if (concurrent_collection_in_progress && sgen_workers_all_done ()) {
2216 generation_to_collect = GENERATION_OLD;
2217 reason = "Finish concurrent collection";
2221 if (generation_to_collect == -1)
2223 sgen_perform_collection (size, generation_to_collect, reason, FALSE, TRUE);
2227 * LOCKING: Assumes the GC lock is held.
2230 sgen_perform_collection (size_t requested_size, int generation_to_collect, const char *reason, gboolean wait_to_finish, gboolean stw)
2232 TV_DECLARE (gc_total_start);
2233 TV_DECLARE (gc_total_end);
2234 int overflow_generation_to_collect = -1;
2235 int oldest_generation_collected = generation_to_collect;
2236 const char *overflow_reason = NULL;
2237 gboolean finish_concurrent = concurrent_collection_in_progress && (major_should_finish_concurrent_collection () || generation_to_collect == GENERATION_OLD);
2239 binary_protocol_collection_requested (generation_to_collect, requested_size, wait_to_finish ? 1 : 0);
2241 SGEN_ASSERT (0, generation_to_collect == GENERATION_NURSERY || generation_to_collect == GENERATION_OLD, "What generation is this?");
2244 sgen_stop_world (generation_to_collect);
2246 SGEN_ASSERT (0, sgen_is_world_stopped (), "We can only collect if the world is stopped");
2249 TV_GETTIME (gc_total_start);
2251 // FIXME: extract overflow reason
2252 // FIXME: minor overflow for concurrent case
2253 if (generation_to_collect == GENERATION_NURSERY && !finish_concurrent) {
2254 if (concurrent_collection_in_progress)
2255 major_update_concurrent_collection ();
2257 if (collect_nursery (reason, FALSE, NULL, FALSE) && !concurrent_collection_in_progress) {
2258 overflow_generation_to_collect = GENERATION_OLD;
2259 overflow_reason = "Minor overflow";
2261 } else if (finish_concurrent) {
2262 major_finish_concurrent_collection (wait_to_finish);
2263 oldest_generation_collected = GENERATION_OLD;
2265 SGEN_ASSERT (0, generation_to_collect == GENERATION_OLD, "We should have handled nursery collections above");
2266 if (major_collector.is_concurrent && !wait_to_finish) {
2267 collect_nursery ("Concurrent start", FALSE, NULL, FALSE);
2268 major_start_concurrent_collection (reason);
2269 oldest_generation_collected = GENERATION_NURSERY;
2270 } else if (major_do_collection (reason, FALSE, wait_to_finish)) {
2271 overflow_generation_to_collect = GENERATION_NURSERY;
2272 overflow_reason = "Excessive pinning";
2276 if (overflow_generation_to_collect != -1) {
2277 SGEN_ASSERT (0, !concurrent_collection_in_progress, "We don't yet support overflow collections with the concurrent collector");
2280 * We need to do an overflow collection, either because we ran out of memory
2281 * or the nursery is fully pinned.
2284 if (overflow_generation_to_collect == GENERATION_NURSERY)
2285 collect_nursery (overflow_reason, TRUE, NULL, FALSE);
2287 major_do_collection (overflow_reason, TRUE, wait_to_finish);
2289 oldest_generation_collected = MAX (oldest_generation_collected, overflow_generation_to_collect);
2292 SGEN_LOG (2, "Heap size: %lu, LOS size: %lu", (unsigned long)sgen_gc_get_total_heap_allocation (), (unsigned long)los_memory_usage);
2294 /* this also sets the proper pointers for the next allocation */
2295 if (generation_to_collect == GENERATION_NURSERY && !sgen_can_alloc_size (requested_size)) {
2296 /* TypeBuilder and MonoMethod are killing mcs with fragmentation */
2297 SGEN_LOG (1, "nursery collection didn't find enough room for %zd alloc (%zd pinned)", requested_size, sgen_get_pinned_count ());
2298 sgen_dump_pin_queue ();
2302 g_assert (sgen_gray_object_queue_is_empty (&gray_queue));
2304 TV_GETTIME (gc_total_end);
2305 time_max = MAX (time_max, TV_ELAPSED (gc_total_start, gc_total_end));
2308 sgen_restart_world (oldest_generation_collected);
2312 * ######################################################################
2313 * ######## Memory allocation from the OS
2314 * ######################################################################
2315 * This section of code deals with getting memory from the OS and
2316 * allocating memory for GC-internal data structures.
2317 * Internal memory can be handled with a freelist for small objects.
2323 G_GNUC_UNUSED static void
2324 report_internal_mem_usage (void)
2326 printf ("Internal memory usage:\n");
2327 sgen_report_internal_mem_usage ();
2328 printf ("Pinned memory usage:\n");
2329 major_collector.report_pinned_memory_usage ();
2333 * ######################################################################
2334 * ######## Finalization support
2335 * ######################################################################
2339 * If the object has been forwarded it means it's still referenced from a root.
2340 * If it is pinned it's still alive as well.
2341 * A LOS object is only alive if we have pinned it.
2342 * Return TRUE if @obj is ready to be finalized.
2344 static inline gboolean
2345 sgen_is_object_alive (GCObject *object)
2347 if (ptr_in_nursery (object))
2348 return sgen_nursery_is_object_alive (object);
2350 return sgen_major_is_object_alive (object);
2354 * This function returns true if @object is either alive and belongs to the
2355 * current collection - major collections are full heap, so old gen objects
2356 * are never alive during a minor collection.
2359 sgen_is_object_alive_and_on_current_collection (GCObject *object)
2361 if (ptr_in_nursery (object))
2362 return sgen_nursery_is_object_alive (object);
2364 if (current_collection_generation == GENERATION_NURSERY)
2367 return sgen_major_is_object_alive (object);
2372 sgen_gc_is_object_ready_for_finalization (GCObject *object)
2374 return !sgen_is_object_alive (object);
2378 sgen_queue_finalization_entry (GCObject *obj)
2380 gboolean critical = sgen_client_object_has_critical_finalizer (obj);
2382 sgen_pointer_queue_add (critical ? &critical_fin_queue : &fin_ready_queue, obj);
2384 sgen_client_object_queued_for_finalization (obj);
2388 sgen_object_is_live (GCObject *obj)
2390 return sgen_is_object_alive_and_on_current_collection (obj);
2394 * `System.GC.WaitForPendingFinalizers` first checks `sgen_have_pending_finalizers()` to
2395 * determine whether it can exit quickly. The latter must therefore only return FALSE if
2396 * all finalizers have really finished running.
2398 * `sgen_gc_invoke_finalizers()` first dequeues a finalizable object, and then finalizes it.
2399 * This means that just checking whether the queues are empty leaves the possibility that an
2400 * object might have been dequeued but not yet finalized. That's why we need the additional
2401 * flag `pending_unqueued_finalizer`.
2404 static volatile gboolean pending_unqueued_finalizer = FALSE;
2407 sgen_gc_invoke_finalizers (void)
2411 g_assert (!pending_unqueued_finalizer);
2413 /* FIXME: batch to reduce lock contention */
2414 while (sgen_have_pending_finalizers ()) {
2420 * We need to set `pending_unqueued_finalizer` before dequeing the
2421 * finalizable object.
2423 if (!sgen_pointer_queue_is_empty (&fin_ready_queue)) {
2424 pending_unqueued_finalizer = TRUE;
2425 mono_memory_write_barrier ();
2426 obj = (GCObject *)sgen_pointer_queue_pop (&fin_ready_queue);
2427 } else if (!sgen_pointer_queue_is_empty (&critical_fin_queue)) {
2428 pending_unqueued_finalizer = TRUE;
2429 mono_memory_write_barrier ();
2430 obj = (GCObject *)sgen_pointer_queue_pop (&critical_fin_queue);
2436 SGEN_LOG (7, "Finalizing object %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
2444 /* the object is on the stack so it is pinned */
2445 /*g_print ("Calling finalizer for object: %p (%s)\n", obj, sgen_client_object_safe_name (obj));*/
2446 sgen_client_run_finalize (obj);
2449 if (pending_unqueued_finalizer) {
2450 mono_memory_write_barrier ();
2451 pending_unqueued_finalizer = FALSE;
2458 sgen_have_pending_finalizers (void)
2460 return pending_unqueued_finalizer || !sgen_pointer_queue_is_empty (&fin_ready_queue) || !sgen_pointer_queue_is_empty (&critical_fin_queue);
2464 * ######################################################################
2465 * ######## registered roots support
2466 * ######################################################################
2470 * We do not coalesce roots.
2473 sgen_register_root (char *start, size_t size, SgenDescriptor descr, int root_type, int source, const char *msg)
2475 RootRecord new_root;
2478 for (i = 0; i < ROOT_TYPE_NUM; ++i) {
2479 RootRecord *root = (RootRecord *)sgen_hash_table_lookup (&roots_hash [i], start);
2480 /* we allow changing the size and the descriptor (for thread statics etc) */
2482 size_t old_size = root->end_root - start;
2483 root->end_root = start + size;
2484 SGEN_ASSERT (0, !!root->root_desc == !!descr, "Can't change whether a root is precise or conservative.");
2485 SGEN_ASSERT (0, root->source == source, "Can't change a root's source identifier.");
2486 SGEN_ASSERT (0, !!root->msg == !!msg, "Can't change a root's message.");
2487 root->root_desc = descr;
2489 roots_size -= old_size;
2495 new_root.end_root = start + size;
2496 new_root.root_desc = descr;
2497 new_root.source = source;
2500 sgen_hash_table_replace (&roots_hash [root_type], start, &new_root, NULL);
2503 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);
2510 sgen_deregister_root (char* addr)
2516 for (root_type = 0; root_type < ROOT_TYPE_NUM; ++root_type) {
2517 if (sgen_hash_table_remove (&roots_hash [root_type], addr, &root))
2518 roots_size -= (root.end_root - addr);
2524 * ######################################################################
2525 * ######## Thread handling (stop/start code)
2526 * ######################################################################
2530 sgen_get_current_collection_generation (void)
2532 return current_collection_generation;
2536 sgen_thread_register (SgenThreadInfo* info, void *stack_bottom_fallback)
2538 #ifndef HAVE_KW_THREAD
2539 info->tlab_start = info->tlab_next = info->tlab_temp_end = info->tlab_real_end = NULL;
2542 sgen_init_tlab_info (info);
2544 sgen_client_thread_register (info, stack_bottom_fallback);
2550 sgen_thread_unregister (SgenThreadInfo *p)
2552 sgen_client_thread_unregister (p);
2556 * ######################################################################
2557 * ######## Write barriers
2558 * ######################################################################
2562 * Note: the write barriers first do the needed GC work and then do the actual store:
2563 * this way the value is visible to the conservative GC scan after the write barrier
2564 * itself. If a GC interrupts the barrier in the middle, value will be kept alive by
2565 * the conservative scan, otherwise by the remembered set scan.
2569 mono_gc_wbarrier_arrayref_copy (gpointer dest_ptr, gpointer src_ptr, int count)
2571 HEAVY_STAT (++stat_wbarrier_arrayref_copy);
2572 /*This check can be done without taking a lock since dest_ptr array is pinned*/
2573 if (ptr_in_nursery (dest_ptr) || count <= 0) {
2574 mono_gc_memmove_aligned (dest_ptr, src_ptr, count * sizeof (gpointer));
2578 #ifdef SGEN_HEAVY_BINARY_PROTOCOL
2579 if (binary_protocol_is_heavy_enabled ()) {
2581 for (i = 0; i < count; ++i) {
2582 gpointer dest = (gpointer*)dest_ptr + i;
2583 gpointer obj = *((gpointer*)src_ptr + i);
2585 binary_protocol_wbarrier (dest, obj, (gpointer)LOAD_VTABLE (obj));
2590 remset.wbarrier_arrayref_copy (dest_ptr, src_ptr, count);
2594 mono_gc_wbarrier_generic_nostore (gpointer ptr)
2598 HEAVY_STAT (++stat_wbarrier_generic_store);
2600 sgen_client_wbarrier_generic_nostore_check (ptr);
2602 obj = *(gpointer*)ptr;
2604 binary_protocol_wbarrier (ptr, obj, (gpointer)LOAD_VTABLE (obj));
2607 * We need to record old->old pointer locations for the
2608 * concurrent collector.
2610 if (!ptr_in_nursery (obj) && !concurrent_collection_in_progress) {
2611 SGEN_LOG (8, "Skipping remset at %p", ptr);
2615 SGEN_LOG (8, "Adding remset at %p", ptr);
2617 remset.wbarrier_generic_nostore (ptr);
2621 mono_gc_wbarrier_generic_store (gpointer ptr, GCObject* value)
2623 SGEN_LOG (8, "Wbarrier store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2624 SGEN_UPDATE_REFERENCE_ALLOW_NULL (ptr, value);
2625 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2626 mono_gc_wbarrier_generic_nostore (ptr);
2627 sgen_dummy_use (value);
2630 /* Same as mono_gc_wbarrier_generic_store () but performs the store
2631 * as an atomic operation with release semantics.
2634 mono_gc_wbarrier_generic_store_atomic (gpointer ptr, GCObject *value)
2636 HEAVY_STAT (++stat_wbarrier_generic_store_atomic);
2638 SGEN_LOG (8, "Wbarrier atomic store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2640 InterlockedWritePointer ((volatile gpointer *)ptr, value);
2642 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2643 mono_gc_wbarrier_generic_nostore (ptr);
2645 sgen_dummy_use (value);
2649 sgen_wbarrier_value_copy_bitmap (gpointer _dest, gpointer _src, int size, unsigned bitmap)
2651 GCObject **dest = (GCObject **)_dest;
2652 GCObject **src = (GCObject **)_src;
2656 mono_gc_wbarrier_generic_store (dest, *src);
2661 size -= SIZEOF_VOID_P;
2667 * ######################################################################
2668 * ######## Other mono public interface functions.
2669 * ######################################################################
2673 sgen_gc_collect (int generation)
2678 sgen_perform_collection (0, generation, "user request", TRUE, TRUE);
2683 sgen_gc_collection_count (int generation)
2685 if (generation == 0)
2686 return gc_stats.minor_gc_count;
2687 return gc_stats.major_gc_count;
2691 sgen_gc_get_used_size (void)
2695 tot = los_memory_usage;
2696 tot += nursery_section->next_data - nursery_section->data;
2697 tot += major_collector.get_used_size ();
2698 /* FIXME: account for pinned objects */
2704 sgen_env_var_error (const char *env_var, const char *fallback, const char *description_format, ...)
2708 va_start (ap, description_format);
2710 fprintf (stderr, "Warning: In environment variable `%s': ", env_var);
2711 vfprintf (stderr, description_format, ap);
2713 fprintf (stderr, " - %s", fallback);
2714 fprintf (stderr, "\n");
2720 parse_double_in_interval (const char *env_var, const char *opt_name, const char *opt, double min, double max, double *result)
2723 double val = strtod (opt, &endptr);
2724 if (endptr == opt) {
2725 sgen_env_var_error (env_var, "Using default value.", "`%s` must be a number.", opt_name);
2728 else if (val < min || val > max) {
2729 sgen_env_var_error (env_var, "Using default value.", "`%s` must be between %.2f - %.2f.", opt_name, min, max);
2741 char *major_collector_opt = NULL;
2742 char *minor_collector_opt = NULL;
2743 size_t max_heap = 0;
2744 size_t soft_limit = 0;
2746 gboolean debug_print_allowance = FALSE;
2747 double allowance_ratio = 0, save_target = 0;
2748 gboolean cement_enabled = TRUE;
2751 result = InterlockedCompareExchange (&gc_initialized, -1, 0);
2754 /* already inited */
2757 /* being inited by another thread */
2758 mono_thread_info_usleep (1000);
2761 /* we will init it */
2764 g_assert_not_reached ();
2766 } while (result != 0);
2768 SGEN_TV_GETTIME (sgen_init_timestamp);
2770 #ifdef SGEN_WITHOUT_MONO
2771 mono_thread_smr_init ();
2774 mono_coop_mutex_init (&gc_mutex);
2776 gc_debug_file = stderr;
2778 mono_coop_mutex_init (&sgen_interruption_mutex);
2780 if ((env = g_getenv (MONO_GC_PARAMS_NAME))) {
2781 opts = g_strsplit (env, ",", -1);
2782 for (ptr = opts; *ptr; ++ptr) {
2784 if (g_str_has_prefix (opt, "major=")) {
2785 opt = strchr (opt, '=') + 1;
2786 major_collector_opt = g_strdup (opt);
2787 } else if (g_str_has_prefix (opt, "minor=")) {
2788 opt = strchr (opt, '=') + 1;
2789 minor_collector_opt = g_strdup (opt);
2797 sgen_init_internal_allocator ();
2798 sgen_init_nursery_allocator ();
2799 sgen_init_fin_weak_hash ();
2800 sgen_init_hash_table ();
2801 sgen_init_descriptors ();
2802 sgen_init_gray_queues ();
2803 sgen_init_allocator ();
2804 sgen_init_gchandles ();
2806 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_SECTION, SGEN_SIZEOF_GC_MEM_SECTION);
2807 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_GRAY_QUEUE, sizeof (GrayQueueSection));
2809 sgen_client_init ();
2811 if (!minor_collector_opt) {
2812 sgen_simple_nursery_init (&sgen_minor_collector);
2814 if (!strcmp (minor_collector_opt, "simple")) {
2816 sgen_simple_nursery_init (&sgen_minor_collector);
2817 } else if (!strcmp (minor_collector_opt, "split")) {
2818 sgen_split_nursery_init (&sgen_minor_collector);
2820 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `simple` instead.", "Unknown minor collector `%s'.", minor_collector_opt);
2821 goto use_simple_nursery;
2825 if (!major_collector_opt) {
2827 DEFAULT_MAJOR_INIT (&major_collector);
2828 } else if (!strcmp (major_collector_opt, "marksweep")) {
2829 sgen_marksweep_init (&major_collector);
2830 } else if (!strcmp (major_collector_opt, "marksweep-conc")) {
2831 sgen_marksweep_conc_init (&major_collector);
2833 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `" DEFAULT_MAJOR_NAME "` instead.", "Unknown major collector `%s'.", major_collector_opt);
2834 goto use_default_major;
2837 sgen_nursery_size = DEFAULT_NURSERY_SIZE;
2840 gboolean usage_printed = FALSE;
2842 for (ptr = opts; *ptr; ++ptr) {
2844 if (!strcmp (opt, ""))
2846 if (g_str_has_prefix (opt, "major="))
2848 if (g_str_has_prefix (opt, "minor="))
2850 if (g_str_has_prefix (opt, "max-heap-size=")) {
2851 size_t page_size = mono_pagesize ();
2852 size_t max_heap_candidate = 0;
2853 opt = strchr (opt, '=') + 1;
2854 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &max_heap_candidate)) {
2855 max_heap = (max_heap_candidate + page_size - 1) & ~(size_t)(page_size - 1);
2856 if (max_heap != max_heap_candidate)
2857 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Rounding up.", "`max-heap-size` size must be a multiple of %d.", page_size);
2859 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`max-heap-size` must be an integer.");
2863 if (g_str_has_prefix (opt, "soft-heap-limit=")) {
2864 opt = strchr (opt, '=') + 1;
2865 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &soft_limit)) {
2866 if (soft_limit <= 0) {
2867 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be positive.");
2871 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be an integer.");
2877 if (g_str_has_prefix (opt, "nursery-size=")) {
2879 opt = strchr (opt, '=') + 1;
2880 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &val)) {
2881 if ((val & (val - 1))) {
2882 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be a power of two.");
2886 if (val < SGEN_MAX_NURSERY_WASTE) {
2887 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.",
2888 "`nursery-size` must be at least %d bytes.", SGEN_MAX_NURSERY_WASTE);
2892 sgen_nursery_size = val;
2893 sgen_nursery_bits = 0;
2894 while (ONE_P << (++ sgen_nursery_bits) != sgen_nursery_size)
2897 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be an integer.");
2903 if (g_str_has_prefix (opt, "save-target-ratio=")) {
2905 opt = strchr (opt, '=') + 1;
2906 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "save-target-ratio", opt,
2907 SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO, &val)) {
2912 if (g_str_has_prefix (opt, "default-allowance-ratio=")) {
2914 opt = strchr (opt, '=') + 1;
2915 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "default-allowance-ratio", opt,
2916 SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, SGEN_MAX_ALLOWANCE_NURSERY_SIZE_RATIO, &val)) {
2917 allowance_ratio = val;
2922 if (!strcmp (opt, "cementing")) {
2923 cement_enabled = TRUE;
2926 if (!strcmp (opt, "no-cementing")) {
2927 cement_enabled = FALSE;
2931 if (!strcmp (opt, "precleaning")) {
2932 precleaning_enabled = TRUE;
2935 if (!strcmp (opt, "no-precleaning")) {
2936 precleaning_enabled = FALSE;
2940 if (major_collector.handle_gc_param && major_collector.handle_gc_param (opt))
2943 if (sgen_minor_collector.handle_gc_param && sgen_minor_collector.handle_gc_param (opt))
2946 if (sgen_client_handle_gc_param (opt))
2949 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Ignoring.", "Unknown option `%s`.", opt);
2954 fprintf (stderr, "\n%s must be a comma-delimited list of one or more of the following:\n", MONO_GC_PARAMS_NAME);
2955 fprintf (stderr, " max-heap-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
2956 fprintf (stderr, " soft-heap-limit=n (where N is an integer, possibly with a k, m or a g suffix)\n");
2957 fprintf (stderr, " nursery-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
2958 fprintf (stderr, " major=COLLECTOR (where COLLECTOR is `marksweep', `marksweep-conc', `marksweep-par')\n");
2959 fprintf (stderr, " minor=COLLECTOR (where COLLECTOR is `simple' or `split')\n");
2960 fprintf (stderr, " wbarrier=WBARRIER (where WBARRIER is `remset' or `cardtable')\n");
2961 fprintf (stderr, " [no-]cementing\n");
2962 if (major_collector.print_gc_param_usage)
2963 major_collector.print_gc_param_usage ();
2964 if (sgen_minor_collector.print_gc_param_usage)
2965 sgen_minor_collector.print_gc_param_usage ();
2966 sgen_client_print_gc_params_usage ();
2967 fprintf (stderr, " Experimental options:\n");
2968 fprintf (stderr, " save-target-ratio=R (where R must be between %.2f - %.2f).\n", SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO);
2969 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);
2970 fprintf (stderr, "\n");
2972 usage_printed = TRUE;
2977 if (major_collector_opt)
2978 g_free (major_collector_opt);
2980 if (minor_collector_opt)
2981 g_free (minor_collector_opt);
2985 sgen_pinning_init ();
2986 sgen_cement_init (cement_enabled);
2988 if ((env = g_getenv (MONO_GC_DEBUG_NAME))) {
2989 gboolean usage_printed = FALSE;
2991 opts = g_strsplit (env, ",", -1);
2992 for (ptr = opts; ptr && *ptr; ptr ++) {
2994 if (!strcmp (opt, ""))
2996 if (opt [0] >= '0' && opt [0] <= '9') {
2997 gc_debug_level = atoi (opt);
3002 char *rf = g_strdup_printf ("%s.%d", opt, mono_process_current_pid ());
3003 gc_debug_file = fopen (rf, "wb");
3005 gc_debug_file = stderr;
3008 } else if (!strcmp (opt, "print-allowance")) {
3009 debug_print_allowance = TRUE;
3010 } else if (!strcmp (opt, "print-pinning")) {
3011 sgen_pin_stats_enable ();
3012 } else if (!strcmp (opt, "verify-before-allocs")) {
3013 verify_before_allocs = 1;
3014 has_per_allocation_action = TRUE;
3015 } else if (g_str_has_prefix (opt, "verify-before-allocs=")) {
3016 char *arg = strchr (opt, '=') + 1;
3017 verify_before_allocs = atoi (arg);
3018 has_per_allocation_action = TRUE;
3019 } else if (!strcmp (opt, "collect-before-allocs")) {
3020 collect_before_allocs = 1;
3021 has_per_allocation_action = TRUE;
3022 } else if (g_str_has_prefix (opt, "collect-before-allocs=")) {
3023 char *arg = strchr (opt, '=') + 1;
3024 has_per_allocation_action = TRUE;
3025 collect_before_allocs = atoi (arg);
3026 } else if (!strcmp (opt, "verify-before-collections")) {
3027 whole_heap_check_before_collection = TRUE;
3028 } else if (!strcmp (opt, "check-remset-consistency")) {
3029 remset_consistency_checks = TRUE;
3030 nursery_clear_policy = CLEAR_AT_GC;
3031 } else if (!strcmp (opt, "mod-union-consistency-check")) {
3032 if (!major_collector.is_concurrent) {
3033 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`mod-union-consistency-check` only works with concurrent major collector.");
3036 mod_union_consistency_check = TRUE;
3037 } else if (!strcmp (opt, "check-mark-bits")) {
3038 check_mark_bits_after_major_collection = TRUE;
3039 } else if (!strcmp (opt, "check-nursery-pinned")) {
3040 check_nursery_objects_pinned = TRUE;
3041 } else if (!strcmp (opt, "clear-at-gc")) {
3042 nursery_clear_policy = CLEAR_AT_GC;
3043 } else if (!strcmp (opt, "clear-nursery-at-gc")) {
3044 nursery_clear_policy = CLEAR_AT_GC;
3045 } else if (!strcmp (opt, "clear-at-tlab-creation")) {
3046 nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
3047 } else if (!strcmp (opt, "debug-clear-at-tlab-creation")) {
3048 nursery_clear_policy = CLEAR_AT_TLAB_CREATION_DEBUG;
3049 } else if (!strcmp (opt, "check-scan-starts")) {
3050 do_scan_starts_check = TRUE;
3051 } else if (!strcmp (opt, "verify-nursery-at-minor-gc")) {
3052 do_verify_nursery = TRUE;
3053 } else if (!strcmp (opt, "check-concurrent")) {
3054 if (!major_collector.is_concurrent) {
3055 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`check-concurrent` only works with concurrent major collectors.");
3058 nursery_clear_policy = CLEAR_AT_GC;
3059 do_concurrent_checks = TRUE;
3060 } else if (!strcmp (opt, "dump-nursery-at-minor-gc")) {
3061 do_dump_nursery_content = TRUE;
3062 } else if (!strcmp (opt, "disable-minor")) {
3063 disable_minor_collections = TRUE;
3064 } else if (!strcmp (opt, "disable-major")) {
3065 disable_major_collections = TRUE;
3066 } else if (g_str_has_prefix (opt, "heap-dump=")) {
3067 char *filename = strchr (opt, '=') + 1;
3068 nursery_clear_policy = CLEAR_AT_GC;
3069 sgen_debug_enable_heap_dump (filename);
3070 } else if (g_str_has_prefix (opt, "binary-protocol=")) {
3071 char *filename = strchr (opt, '=') + 1;
3072 char *colon = strrchr (filename, ':');
3075 if (!mono_gc_parse_environment_string_extract_number (colon + 1, &limit)) {
3076 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring limit.", "Binary protocol file size limit must be an integer.");
3081 binary_protocol_init (filename, (long long)limit);
3082 } else if (!strcmp (opt, "nursery-canaries")) {
3083 do_verify_nursery = TRUE;
3084 enable_nursery_canaries = TRUE;
3085 } else if (!sgen_client_handle_gc_debug (opt)) {
3086 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "Unknown option `%s`.", opt);
3091 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);
3092 fprintf (stderr, "Valid <option>s are:\n");
3093 fprintf (stderr, " collect-before-allocs[=<n>]\n");
3094 fprintf (stderr, " verify-before-allocs[=<n>]\n");
3095 fprintf (stderr, " check-remset-consistency\n");
3096 fprintf (stderr, " check-mark-bits\n");
3097 fprintf (stderr, " check-nursery-pinned\n");
3098 fprintf (stderr, " verify-before-collections\n");
3099 fprintf (stderr, " verify-nursery-at-minor-gc\n");
3100 fprintf (stderr, " dump-nursery-at-minor-gc\n");
3101 fprintf (stderr, " disable-minor\n");
3102 fprintf (stderr, " disable-major\n");
3103 fprintf (stderr, " check-concurrent\n");
3104 fprintf (stderr, " clear-[nursery-]at-gc\n");
3105 fprintf (stderr, " clear-at-tlab-creation\n");
3106 fprintf (stderr, " debug-clear-at-tlab-creation\n");
3107 fprintf (stderr, " check-scan-starts\n");
3108 fprintf (stderr, " print-allowance\n");
3109 fprintf (stderr, " print-pinning\n");
3110 fprintf (stderr, " heap-dump=<filename>\n");
3111 fprintf (stderr, " binary-protocol=<filename>[:<file-size-limit>]\n");
3112 fprintf (stderr, " nursery-canaries\n");
3113 sgen_client_print_gc_debug_usage ();
3114 fprintf (stderr, "\n");
3116 usage_printed = TRUE;
3122 if (check_mark_bits_after_major_collection)
3123 nursery_clear_policy = CLEAR_AT_GC;
3125 if (major_collector.post_param_init)
3126 major_collector.post_param_init (&major_collector);
3128 if (major_collector.needs_thread_pool)
3129 sgen_workers_init (1);
3131 sgen_memgov_init (max_heap, soft_limit, debug_print_allowance, allowance_ratio, save_target);
3133 memset (&remset, 0, sizeof (remset));
3135 sgen_card_table_init (&remset);
3137 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");
3141 sgen_init_bridge ();
3145 sgen_gc_initialized ()
3147 return gc_initialized > 0;
3151 sgen_get_nursery_clear_policy (void)
3153 return nursery_clear_policy;
3159 mono_coop_mutex_lock (&gc_mutex);
3163 sgen_gc_unlock (void)
3165 mono_coop_mutex_unlock (&gc_mutex);
3169 sgen_major_collector_iterate_live_block_ranges (sgen_cardtable_block_callback callback)
3171 major_collector.iterate_live_block_ranges (callback);
3175 sgen_major_collector_iterate_block_ranges (sgen_cardtable_block_callback callback)
3177 major_collector.iterate_block_ranges (callback);
3181 sgen_get_major_collector (void)
3183 return &major_collector;
3187 sgen_get_remset (void)
3193 count_cards (long long *major_total, long long *major_marked, long long *los_total, long long *los_marked)
3195 sgen_get_major_collector ()->count_cards (major_total, major_marked);
3196 sgen_los_count_cards (los_total, los_marked);
3199 static gboolean world_is_stopped = FALSE;
3201 /* LOCKING: assumes the GC lock is held */
3203 sgen_stop_world (int generation)
3205 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3207 SGEN_ASSERT (0, !world_is_stopped, "Why are we stopping a stopped world?");
3209 binary_protocol_world_stopping (generation, sgen_timestamp (), (gpointer) (gsize) mono_native_thread_id_get ());
3211 sgen_client_stop_world (generation);
3213 world_is_stopped = TRUE;
3215 if (binary_protocol_is_heavy_enabled ())
3216 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3217 binary_protocol_world_stopped (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3220 /* LOCKING: assumes the GC lock is held */
3222 sgen_restart_world (int generation)
3224 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3227 SGEN_ASSERT (0, world_is_stopped, "Why are we restarting a running world?");
3229 if (binary_protocol_is_heavy_enabled ())
3230 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3231 binary_protocol_world_restarting (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3233 world_is_stopped = FALSE;
3235 sgen_client_restart_world (generation, &stw_time);
3237 binary_protocol_world_restarted (generation, sgen_timestamp ());
3239 if (sgen_client_bridge_need_processing ())
3240 sgen_client_bridge_processing_finish (generation);
3242 sgen_memgov_collection_end (generation, stw_time);
3246 sgen_is_world_stopped (void)
3248 return world_is_stopped;
3252 sgen_check_whole_heap_stw (void)
3254 sgen_stop_world (0);
3255 sgen_clear_nursery_fragments ();
3256 sgen_check_whole_heap (FALSE);
3257 sgen_restart_world (0);
3261 sgen_timestamp (void)
3263 SGEN_TV_DECLARE (timestamp);
3264 SGEN_TV_GETTIME (timestamp);
3265 return SGEN_TV_ELAPSED (sgen_init_timestamp, timestamp);
3268 #endif /* HAVE_SGEN_GC */