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 /* FIXME: get rid of this */
403 typedef SgenGrayQueue GrayQueue;
405 /* forward declarations */
406 static void scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx);
408 static void pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx);
409 static void finish_gray_stack (int generation, ScanCopyContext ctx);
412 SgenMajorCollector major_collector;
413 SgenMinorCollector sgen_minor_collector;
414 /* FIXME: get rid of this */
415 SgenGrayQueue gc_thread_gray_queue;
417 static SgenRememberedSet remset;
420 * The gray queue a worker job must use. If we're not parallel or
421 * concurrent, we use the main gray queue.
423 static SgenGrayQueue*
424 sgen_workers_get_job_gray_queue (WorkerData *worker_data, SgenGrayQueue *default_gray_queue)
426 return worker_data ? &worker_data->private_gray_queue : default_gray_queue;
430 gray_queue_enable_redirect (SgenGrayQueue *queue)
432 SGEN_ASSERT (0, concurrent_collection_in_progress, "Where are we redirecting the gray queue to, without a concurrent collection?");
434 sgen_gray_queue_set_alloc_prepare (queue, sgen_workers_take_from_queue_and_awake);
435 sgen_workers_take_from_queue_and_awake (queue);
439 sgen_scan_area_with_callback (char *start, char *end, IterateObjectCallbackFunc callback, void *data, gboolean allow_flags, gboolean fail_on_canaries)
441 while (start < end) {
445 if (!*(void**)start) {
446 start += sizeof (void*); /* should be ALLOC_ALIGN, really */
451 if (!(obj = (char *)SGEN_OBJECT_IS_FORWARDED (start)))
457 if (!sgen_client_object_is_array_fill ((GCObject*)obj)) {
458 CHECK_CANARY_FOR_OBJECT ((GCObject*)obj, fail_on_canaries);
459 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
460 callback ((GCObject*)obj, size, data);
461 CANARIFY_SIZE (size);
463 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
471 * sgen_add_to_global_remset:
473 * The global remset contains locations which point into newspace after
474 * a minor collection. This can happen if the objects they point to are pinned.
476 * LOCKING: If called from a parallel collector, the global remset
477 * lock must be held. For serial collectors that is not necessary.
480 sgen_add_to_global_remset (gpointer ptr, GCObject *obj)
482 SGEN_ASSERT (5, sgen_ptr_in_nursery (obj), "Target pointer of global remset must be in the nursery");
484 HEAVY_STAT (++stat_wbarrier_add_to_global_remset);
486 if (!major_collector.is_concurrent) {
487 SGEN_ASSERT (5, current_collection_generation != -1, "Global remsets can only be added during collections");
489 if (current_collection_generation == -1)
490 SGEN_ASSERT (5, sgen_concurrent_collection_in_progress (), "Global remsets outside of collection pauses can only be added by the concurrent collector");
493 if (!object_is_pinned (obj))
494 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");
495 else if (sgen_cement_lookup_or_register (obj))
498 remset.record_pointer (ptr);
500 sgen_pin_stats_register_global_remset (obj);
502 SGEN_LOG (8, "Adding global remset for %p", ptr);
503 binary_protocol_global_remset (ptr, obj, (gpointer)SGEN_LOAD_VTABLE (obj));
507 * sgen_drain_gray_stack:
509 * Scan objects in the gray stack until the stack is empty. This should be called
510 * frequently after each object is copied, to achieve better locality and cache
515 sgen_drain_gray_stack (ScanCopyContext ctx)
517 ScanObjectFunc scan_func = ctx.ops->scan_object;
518 GrayQueue *queue = ctx.queue;
520 if (ctx.ops->drain_gray_stack)
521 return ctx.ops->drain_gray_stack (queue);
526 GRAY_OBJECT_DEQUEUE (queue, &obj, &desc);
529 SGEN_LOG (9, "Precise gray object scan %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
530 scan_func (obj, desc, queue);
536 * Addresses in the pin queue are already sorted. This function finds
537 * the object header for each address and pins the object. The
538 * addresses must be inside the nursery section. The (start of the)
539 * address array is overwritten with the addresses of the actually
540 * pinned objects. Return the number of pinned objects.
543 pin_objects_from_nursery_pin_queue (gboolean do_scan_objects, ScanCopyContext ctx)
545 GCMemSection *section = nursery_section;
546 void **start = sgen_pinning_get_entry (section->pin_queue_first_entry);
547 void **end = sgen_pinning_get_entry (section->pin_queue_last_entry);
548 void *start_nursery = section->data;
549 void *end_nursery = section->next_data;
554 void *pinning_front = start_nursery;
556 void **definitely_pinned = start;
557 ScanObjectFunc scan_func = ctx.ops->scan_object;
558 SgenGrayQueue *queue = ctx.queue;
560 sgen_nursery_allocator_prepare_for_pinning ();
562 while (start < end) {
563 GCObject *obj_to_pin = NULL;
564 size_t obj_to_pin_size = 0;
569 SGEN_ASSERT (0, addr >= start_nursery && addr < end_nursery, "Potential pinning address out of range");
570 SGEN_ASSERT (0, addr >= last, "Pin queue not sorted");
577 SGEN_LOG (5, "Considering pinning addr %p", addr);
578 /* We've already processed everything up to pinning_front. */
579 if (addr < pinning_front) {
585 * Find the closest scan start <= addr. We might search backward in the
586 * scan_starts array because entries might be NULL. In the worst case we
587 * start at start_nursery.
589 idx = ((char*)addr - (char*)section->data) / SCAN_START_SIZE;
590 SGEN_ASSERT (0, idx < section->num_scan_start, "Scan start index out of range");
591 search_start = (void*)section->scan_starts [idx];
592 if (!search_start || search_start > addr) {
595 search_start = section->scan_starts [idx];
596 if (search_start && search_start <= addr)
599 if (!search_start || search_start > addr)
600 search_start = start_nursery;
604 * If the pinning front is closer than the scan start we found, start
605 * searching at the front.
607 if (search_start < pinning_front)
608 search_start = pinning_front;
611 * Now addr should be in an object a short distance from search_start.
613 * search_start must point to zeroed mem or point to an object.
616 size_t obj_size, canarified_obj_size;
619 if (!*(void**)search_start) {
620 search_start = (void*)ALIGN_UP ((mword)search_start + sizeof (gpointer));
621 /* The loop condition makes sure we don't overrun addr. */
625 canarified_obj_size = obj_size = ALIGN_UP (safe_object_get_size ((GCObject*)search_start));
628 * Filler arrays are marked by an invalid sync word. We don't
629 * consider them for pinning. They are not delimited by canaries,
632 if (!sgen_client_object_is_array_fill ((GCObject*)search_start)) {
633 CHECK_CANARY_FOR_OBJECT (search_start, TRUE);
634 CANARIFY_SIZE (canarified_obj_size);
636 if (addr >= search_start && (char*)addr < (char*)search_start + obj_size) {
637 /* This is the object we're looking for. */
638 obj_to_pin = (GCObject*)search_start;
639 obj_to_pin_size = canarified_obj_size;
644 /* Skip to the next object */
645 search_start = (void*)((char*)search_start + canarified_obj_size);
646 } while (search_start <= addr);
648 /* We've searched past the address we were looking for. */
650 pinning_front = search_start;
651 goto next_pin_queue_entry;
655 * We've found an object to pin. It might still be a dummy array, but we
656 * can advance the pinning front in any case.
658 pinning_front = (char*)obj_to_pin + obj_to_pin_size;
661 * If this is a dummy array marking the beginning of a nursery
662 * fragment, we don't pin it.
664 if (sgen_client_object_is_array_fill (obj_to_pin))
665 goto next_pin_queue_entry;
668 * Finally - pin the object!
670 desc = sgen_obj_get_descriptor_safe (obj_to_pin);
671 if (do_scan_objects) {
672 scan_func (obj_to_pin, desc, queue);
674 SGEN_LOG (4, "Pinned object %p, vtable %p (%s), count %d\n",
675 obj_to_pin, *(void**)obj_to_pin, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj_to_pin)), count);
676 binary_protocol_pin (obj_to_pin,
677 (gpointer)LOAD_VTABLE (obj_to_pin),
678 safe_object_get_size (obj_to_pin));
680 pin_object (obj_to_pin);
681 GRAY_OBJECT_ENQUEUE (queue, obj_to_pin, desc);
682 sgen_pin_stats_register_object (obj_to_pin, GENERATION_NURSERY);
683 definitely_pinned [count] = obj_to_pin;
686 if (concurrent_collection_in_progress)
687 sgen_pinning_register_pinned_in_nursery (obj_to_pin);
689 next_pin_queue_entry:
693 sgen_client_nursery_objects_pinned (definitely_pinned, count);
694 stat_pinned_objects += count;
699 pin_objects_in_nursery (gboolean do_scan_objects, ScanCopyContext ctx)
703 if (nursery_section->pin_queue_first_entry == nursery_section->pin_queue_last_entry)
706 reduced_to = pin_objects_from_nursery_pin_queue (do_scan_objects, ctx);
707 nursery_section->pin_queue_last_entry = nursery_section->pin_queue_first_entry + reduced_to;
711 * This function is only ever called (via `collector_pin_object()` in `sgen-copy-object.h`)
712 * when we can't promote an object because we're out of memory.
715 sgen_pin_object (GCObject *object, GrayQueue *queue)
717 SGEN_ASSERT (0, sgen_ptr_in_nursery (object), "We're only supposed to use this for pinning nursery objects when out of memory.");
720 * All pinned objects are assumed to have been staged, so we need to stage as well.
721 * Also, the count of staged objects shows that "late pinning" happened.
723 sgen_pin_stage_ptr (object);
725 SGEN_PIN_OBJECT (object);
726 binary_protocol_pin (object, (gpointer)LOAD_VTABLE (object), safe_object_get_size (object));
729 sgen_pin_stats_register_object (object, GENERATION_NURSERY);
731 GRAY_OBJECT_ENQUEUE (queue, object, sgen_obj_get_descriptor_safe (object));
734 /* Sort the addresses in array in increasing order.
735 * Done using a by-the book heap sort. Which has decent and stable performance, is pretty cache efficient.
738 sgen_sort_addresses (void **array, size_t size)
743 for (i = 1; i < size; ++i) {
746 size_t parent = (child - 1) / 2;
748 if (array [parent] >= array [child])
751 tmp = array [parent];
752 array [parent] = array [child];
759 for (i = size - 1; i > 0; --i) {
762 array [i] = array [0];
768 while (root * 2 + 1 <= end) {
769 size_t child = root * 2 + 1;
771 if (child < end && array [child] < array [child + 1])
773 if (array [root] >= array [child])
777 array [root] = array [child];
786 * Scan the memory between start and end and queue values which could be pointers
787 * to the area between start_nursery and end_nursery for later consideration.
788 * Typically used for thread stacks.
791 sgen_conservatively_pin_objects_from (void **start, void **end, void *start_nursery, void *end_nursery, int pin_type)
795 SGEN_ASSERT (0, ((mword)start & (SIZEOF_VOID_P - 1)) == 0, "Why are we scanning for references in unaligned memory ?");
797 #if defined(VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE) && !defined(_WIN64)
798 VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE (start, (char*)end - (char*)start);
801 while (start < end) {
803 * *start can point to the middle of an object
804 * note: should we handle pointing at the end of an object?
805 * pinning in C# code disallows pointing at the end of an object
806 * but there is some small chance that an optimizing C compiler
807 * may keep the only reference to an object by pointing
808 * at the end of it. We ignore this small chance for now.
809 * Pointers to the end of an object are indistinguishable
810 * from pointers to the start of the next object in memory
811 * so if we allow that we'd need to pin two objects...
812 * We queue the pointer in an array, the
813 * array will then be sorted and uniqued. This way
814 * we can coalesce several pinning pointers and it should
815 * be faster since we'd do a memory scan with increasing
816 * addresses. Note: we can align the address to the allocation
817 * alignment, so the unique process is more effective.
819 mword addr = (mword)*start;
820 addr &= ~(ALLOC_ALIGN - 1);
821 if (addr >= (mword)start_nursery && addr < (mword)end_nursery) {
822 SGEN_LOG (6, "Pinning address %p from %p", (void*)addr, start);
823 sgen_pin_stage_ptr ((void*)addr);
824 binary_protocol_pin_stage (start, (void*)addr);
825 sgen_pin_stats_register_address ((char*)addr, pin_type);
831 SGEN_LOG (7, "found %d potential pinned heap pointers", count);
835 * The first thing we do in a collection is to identify pinned objects.
836 * This function considers all the areas of memory that need to be
837 * conservatively scanned.
840 pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx)
844 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);
845 /* objects pinned from the API are inside these roots */
846 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_PINNED], void **, start_root, RootRecord *, root) {
847 SGEN_LOG (6, "Pinned roots %p-%p", start_root, root->end_root);
848 sgen_conservatively_pin_objects_from (start_root, (void**)root->end_root, start_nursery, end_nursery, PIN_TYPE_OTHER);
849 } SGEN_HASH_TABLE_FOREACH_END;
850 /* now deal with the thread stacks
851 * in the future we should be able to conservatively scan only:
852 * *) the cpu registers
853 * *) the unmanaged stack frames
854 * *) the _last_ managed stack frame
855 * *) pointers slots in managed frames
857 sgen_client_scan_thread_data (start_nursery, end_nursery, FALSE, ctx);
861 single_arg_user_copy_or_mark (GCObject **obj, void *gc_data)
863 ScanCopyContext *ctx = (ScanCopyContext *)gc_data;
864 ctx->ops->copy_or_mark_object (obj, ctx->queue);
868 * The memory area from start_root to end_root contains pointers to objects.
869 * Their position is precisely described by @desc (this means that the pointer
870 * can be either NULL or the pointer to the start of an object).
871 * This functions copies them to to_space updates them.
873 * This function is not thread-safe!
876 precisely_scan_objects_from (void** start_root, void** end_root, char* n_start, char *n_end, SgenDescriptor desc, ScanCopyContext ctx)
878 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
879 SgenGrayQueue *queue = ctx.queue;
881 switch (desc & ROOT_DESC_TYPE_MASK) {
882 case ROOT_DESC_BITMAP:
883 desc >>= ROOT_DESC_TYPE_SHIFT;
885 if ((desc & 1) && *start_root) {
886 copy_func ((GCObject**)start_root, queue);
887 SGEN_LOG (9, "Overwrote root at %p with %p", start_root, *start_root);
893 case ROOT_DESC_COMPLEX: {
894 gsize *bitmap_data = (gsize *)sgen_get_complex_descriptor_bitmap (desc);
895 gsize bwords = (*bitmap_data) - 1;
896 void **start_run = start_root;
898 while (bwords-- > 0) {
899 gsize bmap = *bitmap_data++;
900 void **objptr = start_run;
902 if ((bmap & 1) && *objptr) {
903 copy_func ((GCObject**)objptr, queue);
904 SGEN_LOG (9, "Overwrote root at %p with %p", objptr, *objptr);
909 start_run += GC_BITS_PER_WORD;
913 case ROOT_DESC_USER: {
914 SgenUserRootMarkFunc marker = sgen_get_user_descriptor_func (desc);
915 marker (start_root, single_arg_user_copy_or_mark, &ctx);
918 case ROOT_DESC_RUN_LEN:
919 g_assert_not_reached ();
921 g_assert_not_reached ();
926 reset_heap_boundaries (void)
928 lowest_heap_address = ~(mword)0;
929 highest_heap_address = 0;
933 sgen_update_heap_boundaries (mword low, mword high)
938 old = lowest_heap_address;
941 } while (SGEN_CAS_PTR ((gpointer*)&lowest_heap_address, (gpointer)low, (gpointer)old) != (gpointer)old);
944 old = highest_heap_address;
947 } while (SGEN_CAS_PTR ((gpointer*)&highest_heap_address, (gpointer)high, (gpointer)old) != (gpointer)old);
951 * Allocate and setup the data structures needed to be able to allocate objects
952 * in the nursery. The nursery is stored in nursery_section.
957 GCMemSection *section;
964 SGEN_LOG (2, "Allocating nursery size: %zu", (size_t)sgen_nursery_size);
965 /* later we will alloc a larger area for the nursery but only activate
966 * what we need. The rest will be used as expansion if we have too many pinned
967 * objects in the existing nursery.
969 /* FIXME: handle OOM */
970 section = (GCMemSection *)sgen_alloc_internal (INTERNAL_MEM_SECTION);
972 alloc_size = sgen_nursery_size;
974 /* If there isn't enough space even for the nursery we should simply abort. */
975 g_assert (sgen_memgov_try_alloc_space (alloc_size, SPACE_NURSERY));
977 data = (char *)major_collector.alloc_heap (alloc_size, alloc_size, DEFAULT_NURSERY_BITS);
978 sgen_update_heap_boundaries ((mword)data, (mword)(data + sgen_nursery_size));
979 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 ());
980 section->data = section->next_data = data;
981 section->size = alloc_size;
982 section->end_data = data + sgen_nursery_size;
983 scan_starts = (alloc_size + SCAN_START_SIZE - 1) / SCAN_START_SIZE;
984 section->scan_starts = (char **)sgen_alloc_internal_dynamic (sizeof (char*) * scan_starts, INTERNAL_MEM_SCAN_STARTS, TRUE);
985 section->num_scan_start = scan_starts;
987 nursery_section = section;
989 sgen_nursery_allocator_set_nursery_bounds (data, data + sgen_nursery_size);
993 mono_gc_get_logfile (void)
995 return gc_debug_file;
999 scan_finalizer_entries (SgenPointerQueue *fin_queue, ScanCopyContext ctx)
1001 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
1002 SgenGrayQueue *queue = ctx.queue;
1005 for (i = 0; i < fin_queue->next_slot; ++i) {
1006 GCObject *obj = (GCObject *)fin_queue->data [i];
1009 SGEN_LOG (5, "Scan of fin ready object: %p (%s)\n", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
1010 copy_func ((GCObject**)&fin_queue->data [i], queue);
1015 generation_name (int generation)
1017 switch (generation) {
1018 case GENERATION_NURSERY: return "nursery";
1019 case GENERATION_OLD: return "old";
1020 default: g_assert_not_reached ();
1025 sgen_generation_name (int generation)
1027 return generation_name (generation);
1031 finish_gray_stack (int generation, ScanCopyContext ctx)
1035 int done_with_ephemerons, ephemeron_rounds = 0;
1036 char *start_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_start () : NULL;
1037 char *end_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_end () : (char*)-1;
1038 SgenGrayQueue *queue = ctx.queue;
1040 binary_protocol_finish_gray_stack_start (sgen_timestamp (), generation);
1042 * We copied all the reachable objects. Now it's the time to copy
1043 * the objects that were not referenced by the roots, but by the copied objects.
1044 * we built a stack of objects pointed to by gray_start: they are
1045 * additional roots and we may add more items as we go.
1046 * We loop until gray_start == gray_objects which means no more objects have
1047 * been added. Note this is iterative: no recursion is involved.
1048 * We need to walk the LO list as well in search of marked big objects
1049 * (use a flag since this is needed only on major collections). We need to loop
1050 * here as well, so keep a counter of marked LO (increasing it in copy_object).
1051 * To achieve better cache locality and cache usage, we drain the gray stack
1052 * frequently, after each object is copied, and just finish the work here.
1054 sgen_drain_gray_stack (ctx);
1056 SGEN_LOG (2, "%s generation done", generation_name (generation));
1059 Reset bridge data, we might have lingering data from a previous collection if this is a major
1060 collection trigged by minor overflow.
1062 We must reset the gathered bridges since their original block might be evacuated due to major
1063 fragmentation in the meanwhile and the bridge code should not have to deal with that.
1065 if (sgen_client_bridge_need_processing ())
1066 sgen_client_bridge_reset_data ();
1069 * Mark all strong toggleref objects. This must be done before we walk ephemerons or finalizers
1070 * to ensure they see the full set of live objects.
1072 sgen_client_mark_togglerefs (start_addr, end_addr, ctx);
1075 * Walk the ephemeron tables marking all values with reachable keys. This must be completely done
1076 * before processing finalizable objects and non-tracking weak links to avoid finalizing/clearing
1077 * objects that are in fact reachable.
1079 done_with_ephemerons = 0;
1081 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1082 sgen_drain_gray_stack (ctx);
1084 } while (!done_with_ephemerons);
1086 if (sgen_client_bridge_need_processing ()) {
1087 /*Make sure the gray stack is empty before we process bridge objects so we get liveness right*/
1088 sgen_drain_gray_stack (ctx);
1089 sgen_collect_bridge_objects (generation, ctx);
1090 if (generation == GENERATION_OLD)
1091 sgen_collect_bridge_objects (GENERATION_NURSERY, ctx);
1094 Do the first bridge step here, as the collector liveness state will become useless after that.
1096 An important optimization is to only proccess the possibly dead part of the object graph and skip
1097 over all live objects as we transitively know everything they point must be alive too.
1099 The above invariant is completely wrong if we let the gray queue be drained and mark/copy everything.
1101 This has the unfortunate side effect of making overflow collections perform the first step twice, but
1102 given we now have heuristics that perform major GC in anticipation of minor overflows this should not
1105 sgen_client_bridge_processing_stw_step ();
1109 Make sure we drain the gray stack before processing disappearing links and finalizers.
1110 If we don't make sure it is empty we might wrongly see a live object as dead.
1112 sgen_drain_gray_stack (ctx);
1115 We must clear weak links that don't track resurrection before processing object ready for
1116 finalization so they can be cleared before that.
1118 sgen_null_link_in_range (generation, ctx, FALSE);
1119 if (generation == GENERATION_OLD)
1120 sgen_null_link_in_range (GENERATION_NURSERY, ctx, FALSE);
1123 /* walk the finalization queue and move also the objects that need to be
1124 * finalized: use the finalized objects as new roots so the objects they depend
1125 * on are also not reclaimed. As with the roots above, only objects in the nursery
1126 * are marked/copied.
1128 sgen_finalize_in_range (generation, ctx);
1129 if (generation == GENERATION_OLD)
1130 sgen_finalize_in_range (GENERATION_NURSERY, ctx);
1131 /* drain the new stack that might have been created */
1132 SGEN_LOG (6, "Precise scan of gray area post fin");
1133 sgen_drain_gray_stack (ctx);
1136 * This must be done again after processing finalizable objects since CWL slots are cleared only after the key is finalized.
1138 done_with_ephemerons = 0;
1140 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1141 sgen_drain_gray_stack (ctx);
1143 } while (!done_with_ephemerons);
1145 sgen_client_clear_unreachable_ephemerons (ctx);
1148 * We clear togglerefs only after all possible chances of revival are done.
1149 * This is semantically more inline with what users expect and it allows for
1150 * user finalizers to correctly interact with TR objects.
1152 sgen_client_clear_togglerefs (start_addr, end_addr, ctx);
1155 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);
1158 * handle disappearing links
1159 * Note we do this after checking the finalization queue because if an object
1160 * survives (at least long enough to be finalized) we don't clear the link.
1161 * This also deals with a possible issue with the monitor reclamation: with the Boehm
1162 * GC a finalized object my lose the monitor because it is cleared before the finalizer is
1165 g_assert (sgen_gray_object_queue_is_empty (queue));
1167 sgen_null_link_in_range (generation, ctx, TRUE);
1168 if (generation == GENERATION_OLD)
1169 sgen_null_link_in_range (GENERATION_NURSERY, ctx, TRUE);
1170 if (sgen_gray_object_queue_is_empty (queue))
1172 sgen_drain_gray_stack (ctx);
1175 g_assert (sgen_gray_object_queue_is_empty (queue));
1177 sgen_gray_object_queue_trim_free_list (queue);
1178 binary_protocol_finish_gray_stack_end (sgen_timestamp (), generation);
1182 sgen_check_section_scan_starts (GCMemSection *section)
1185 for (i = 0; i < section->num_scan_start; ++i) {
1186 if (section->scan_starts [i]) {
1187 mword size = safe_object_get_size ((GCObject*) section->scan_starts [i]);
1188 SGEN_ASSERT (0, size >= SGEN_CLIENT_MINIMUM_OBJECT_SIZE && size <= MAX_SMALL_OBJ_SIZE, "Weird object size at scan starts.");
1194 check_scan_starts (void)
1196 if (!do_scan_starts_check)
1198 sgen_check_section_scan_starts (nursery_section);
1199 major_collector.check_scan_starts ();
1203 scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx)
1207 SGEN_HASH_TABLE_FOREACH (&roots_hash [root_type], void **, start_root, RootRecord *, root) {
1208 SGEN_LOG (6, "Precise root scan %p-%p (desc: %p)", start_root, root->end_root, (void*)root->root_desc);
1209 precisely_scan_objects_from (start_root, (void**)root->end_root, addr_start, addr_end, root->root_desc, ctx);
1210 } SGEN_HASH_TABLE_FOREACH_END;
1216 static gboolean inited = FALSE;
1221 mono_counters_register ("Collection max time", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME | MONO_COUNTER_MONOTONIC, &time_max);
1223 mono_counters_register ("Minor fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pre_collection_fragment_clear);
1224 mono_counters_register ("Minor pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pinning);
1225 mono_counters_register ("Minor scan remembered set", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_remsets);
1226 mono_counters_register ("Minor scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_pinned);
1227 mono_counters_register ("Minor scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_roots);
1228 mono_counters_register ("Minor fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_fragment_creation);
1230 mono_counters_register ("Major fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pre_collection_fragment_clear);
1231 mono_counters_register ("Major pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pinning);
1232 mono_counters_register ("Major scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_pinned);
1233 mono_counters_register ("Major scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_roots);
1234 mono_counters_register ("Major scan mod union", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_mod_union);
1235 mono_counters_register ("Major finish gray stack", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_finish_gray_stack);
1236 mono_counters_register ("Major free big objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_free_bigobjs);
1237 mono_counters_register ("Major LOS sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_los_sweep);
1238 mono_counters_register ("Major sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_sweep);
1239 mono_counters_register ("Major fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_fragment_creation);
1241 mono_counters_register ("Number of pinned objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_pinned_objects);
1243 #ifdef HEAVY_STATISTICS
1244 mono_counters_register ("WBarrier remember pointer", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_add_to_global_remset);
1245 mono_counters_register ("WBarrier arrayref copy", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_arrayref_copy);
1246 mono_counters_register ("WBarrier generic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store);
1247 mono_counters_register ("WBarrier generic atomic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store_atomic);
1248 mono_counters_register ("WBarrier set root", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_set_root);
1250 mono_counters_register ("# objects allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_alloced_degraded);
1251 mono_counters_register ("bytes allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_bytes_alloced_degraded);
1253 mono_counters_register ("# copy_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_nursery);
1254 mono_counters_register ("# objects copied (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_nursery);
1255 mono_counters_register ("# copy_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_major);
1256 mono_counters_register ("# objects copied (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_major);
1258 mono_counters_register ("# scan_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_nursery);
1259 mono_counters_register ("# scan_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_major);
1261 mono_counters_register ("Slots allocated in vain", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_slots_allocated_in_vain);
1263 mono_counters_register ("# nursery copy_object() failed from space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_from_space);
1264 mono_counters_register ("# nursery copy_object() failed forwarded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_forwarded);
1265 mono_counters_register ("# nursery copy_object() failed pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_pinned);
1266 mono_counters_register ("# nursery copy_object() failed to space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_to_space);
1268 sgen_nursery_allocator_init_heavy_stats ();
1276 reset_pinned_from_failed_allocation (void)
1278 bytes_pinned_from_failed_allocation = 0;
1282 sgen_set_pinned_from_failed_allocation (mword objsize)
1284 bytes_pinned_from_failed_allocation += objsize;
1288 sgen_collection_is_concurrent (void)
1290 switch (current_collection_generation) {
1291 case GENERATION_NURSERY:
1293 case GENERATION_OLD:
1294 return concurrent_collection_in_progress;
1296 g_error ("Invalid current generation %d", current_collection_generation);
1302 sgen_concurrent_collection_in_progress (void)
1304 return concurrent_collection_in_progress;
1308 SgenThreadPoolJob job;
1309 SgenObjectOperations *ops;
1310 SgenGrayQueue *gc_thread_gray_queue;
1313 static ScanCopyContext
1314 scan_copy_context_for_scan_job (void *worker_data_untyped, ScanJob *job)
1316 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1318 return CONTEXT_FROM_OBJECT_OPERATIONS (job->ops, sgen_workers_get_job_gray_queue (worker_data, job->gc_thread_gray_queue));
1322 job_remembered_set_scan (void *worker_data_untyped, SgenThreadPoolJob *job)
1324 remset.scan_remsets (scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job));
1332 } ScanFromRegisteredRootsJob;
1335 job_scan_from_registered_roots (void *worker_data_untyped, SgenThreadPoolJob *job)
1337 ScanFromRegisteredRootsJob *job_data = (ScanFromRegisteredRootsJob*)job;
1338 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1340 scan_from_registered_roots (job_data->heap_start, job_data->heap_end, job_data->root_type, ctx);
1347 } ScanThreadDataJob;
1350 job_scan_thread_data (void *worker_data_untyped, SgenThreadPoolJob *job)
1352 ScanThreadDataJob *job_data = (ScanThreadDataJob*)job;
1353 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1355 sgen_client_scan_thread_data (job_data->heap_start, job_data->heap_end, TRUE, ctx);
1360 SgenPointerQueue *queue;
1361 } ScanFinalizerEntriesJob;
1364 job_scan_finalizer_entries (void *worker_data_untyped, SgenThreadPoolJob *job)
1366 ScanFinalizerEntriesJob *job_data = (ScanFinalizerEntriesJob*)job;
1367 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1369 scan_finalizer_entries (job_data->queue, ctx);
1373 job_scan_major_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1375 ScanJob *job_data = (ScanJob*)job;
1376 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, job_data);
1378 g_assert (concurrent_collection_in_progress);
1379 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx);
1383 job_scan_los_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1385 ScanJob *job_data = (ScanJob*)job;
1386 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, job_data);
1388 g_assert (concurrent_collection_in_progress);
1389 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx);
1393 job_mod_union_preclean (void *worker_data_untyped, SgenThreadPoolJob *job)
1395 ScanJob *job_data = (ScanJob*)job;
1396 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, job_data);
1398 g_assert (concurrent_collection_in_progress);
1400 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx);
1401 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx);
1403 sgen_scan_pin_queue_objects (ctx);
1407 init_gray_queue (gboolean use_workers)
1410 sgen_workers_init_distribute_gray_queue ();
1411 sgen_gray_object_queue_init (&gc_thread_gray_queue, NULL);
1415 enqueue_scan_from_roots_jobs (SgenGrayQueue *gc_thread_gray_queue, char *heap_start, char *heap_end, SgenObjectOperations *ops, gboolean enqueue)
1417 ScanFromRegisteredRootsJob *scrrj;
1418 ScanThreadDataJob *stdj;
1419 ScanFinalizerEntriesJob *sfej;
1421 /* registered roots, this includes static fields */
1423 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots normal", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1424 scrrj->scan_job.ops = ops;
1425 scrrj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1426 scrrj->heap_start = heap_start;
1427 scrrj->heap_end = heap_end;
1428 scrrj->root_type = ROOT_TYPE_NORMAL;
1429 sgen_workers_enqueue_job (&scrrj->scan_job.job, enqueue);
1431 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots wbarrier", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1432 scrrj->scan_job.ops = ops;
1433 scrrj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1434 scrrj->heap_start = heap_start;
1435 scrrj->heap_end = heap_end;
1436 scrrj->root_type = ROOT_TYPE_WBARRIER;
1437 sgen_workers_enqueue_job (&scrrj->scan_job.job, enqueue);
1441 stdj = (ScanThreadDataJob*)sgen_thread_pool_job_alloc ("scan thread data", job_scan_thread_data, sizeof (ScanThreadDataJob));
1442 stdj->scan_job.ops = ops;
1443 stdj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1444 stdj->heap_start = heap_start;
1445 stdj->heap_end = heap_end;
1446 sgen_workers_enqueue_job (&stdj->scan_job.job, enqueue);
1448 /* Scan the list of objects ready for finalization. */
1450 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1451 sfej->scan_job.ops = ops;
1452 sfej->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1453 sfej->queue = &fin_ready_queue;
1454 sgen_workers_enqueue_job (&sfej->scan_job.job, enqueue);
1456 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan critical finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1457 sfej->scan_job.ops = ops;
1458 sfej->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1459 sfej->queue = &critical_fin_queue;
1460 sgen_workers_enqueue_job (&sfej->scan_job.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, &gc_thread_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 sj->gc_thread_gray_queue = &gc_thread_gray_queue;
1559 sgen_workers_enqueue_job (&sj->job, FALSE);
1561 /* we don't have complete write barrier yet, so we scan all the old generation sections */
1563 time_minor_scan_remsets += TV_ELAPSED (atv, btv);
1564 SGEN_LOG (2, "Old generation scan: %lld usecs", (long long)TV_ELAPSED (atv, btv));
1566 sgen_pin_stats_report ();
1568 /* FIXME: Why do we do this at this specific, seemingly random, point? */
1569 sgen_client_collecting_minor (&fin_ready_queue, &critical_fin_queue);
1572 time_minor_scan_pinned += TV_ELAPSED (btv, atv);
1574 enqueue_scan_from_roots_jobs (&gc_thread_gray_queue, sgen_get_nursery_start (), nursery_next, object_ops, FALSE);
1577 time_minor_scan_roots += TV_ELAPSED (atv, btv);
1579 finish_gray_stack (GENERATION_NURSERY, ctx);
1582 time_minor_finish_gray_stack += TV_ELAPSED (btv, atv);
1583 sgen_client_binary_protocol_mark_end (GENERATION_NURSERY);
1585 if (objects_pinned) {
1586 sgen_optimize_pin_queue ();
1587 sgen_pinning_setup_section (nursery_section);
1591 * This is the latest point at which we can do this check, because
1592 * sgen_build_nursery_fragments() unpins nursery objects again.
1594 if (remset_consistency_checks)
1595 sgen_check_remset_consistency ();
1597 /* walk the pin_queue, build up the fragment list of free memory, unmark
1598 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1601 sgen_client_binary_protocol_reclaim_start (GENERATION_NURSERY);
1602 fragment_total = sgen_build_nursery_fragments (nursery_section, unpin_queue);
1603 if (!fragment_total)
1606 /* Clear TLABs for all threads */
1607 sgen_clear_tlabs ();
1609 sgen_client_binary_protocol_reclaim_end (GENERATION_NURSERY);
1611 time_minor_fragment_creation += TV_ELAPSED (atv, btv);
1612 SGEN_LOG (2, "Fragment creation: %lld usecs, %lu bytes available", (long long)TV_ELAPSED (atv, btv), (unsigned long)fragment_total);
1614 if (remset_consistency_checks)
1615 sgen_check_major_refs ();
1617 major_collector.finish_nursery_collection ();
1619 TV_GETTIME (last_minor_collection_end_tv);
1620 gc_stats.minor_gc_time += TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv);
1622 sgen_debug_dump_heap ("minor", gc_stats.minor_gc_count - 1, NULL);
1624 /* prepare the pin queue for the next collection */
1625 sgen_finish_pinning ();
1626 if (sgen_have_pending_finalizers ()) {
1627 SGEN_LOG (4, "Finalizer-thread wakeup");
1628 sgen_client_finalize_notify ();
1630 sgen_pin_stats_reset ();
1631 /* clear cemented hash */
1632 sgen_cement_clear_below_threshold ();
1634 g_assert (sgen_gray_object_queue_is_empty (&gc_thread_gray_queue));
1636 remset.finish_minor_collection ();
1638 check_scan_starts ();
1640 binary_protocol_flush_buffers (FALSE);
1642 sgen_memgov_minor_collection_end (reason, is_overflow);
1644 /*objects are late pinned because of lack of memory, so a major is a good call*/
1645 needs_major = objects_pinned > 0;
1646 current_collection_generation = -1;
1649 binary_protocol_collection_end (gc_stats.minor_gc_count - 1, GENERATION_NURSERY, 0, 0);
1651 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1652 sgen_check_nursery_objects_pinned (unpin_queue != NULL);
1658 COPY_OR_MARK_FROM_ROOTS_SERIAL,
1659 COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT,
1660 COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT
1661 } CopyOrMarkFromRootsMode;
1664 major_copy_or_mark_from_roots (size_t *old_next_pin_slot, CopyOrMarkFromRootsMode mode, SgenObjectOperations *object_ops)
1669 /* FIXME: only use these values for the precise scan
1670 * note that to_space pointers should be excluded anyway...
1672 char *heap_start = NULL;
1673 char *heap_end = (char*)-1;
1674 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, &gc_thread_gray_queue);
1675 gboolean concurrent = mode != COPY_OR_MARK_FROM_ROOTS_SERIAL;
1677 SGEN_ASSERT (0, !!concurrent == !!concurrent_collection_in_progress, "We've been called with the wrong mode.");
1679 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1680 /*This cleans up unused fragments */
1681 sgen_nursery_allocator_prepare_for_pinning ();
1683 if (do_concurrent_checks)
1684 sgen_debug_check_nursery_is_clean ();
1686 /* The concurrent collector doesn't touch the nursery. */
1687 sgen_nursery_alloc_prepare_for_major ();
1690 init_gray_queue (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1694 /* Pinning depends on this */
1695 sgen_clear_nursery_fragments ();
1697 if (whole_heap_check_before_collection)
1698 sgen_check_whole_heap (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT);
1701 time_major_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1703 if (!sgen_collection_is_concurrent ())
1704 nursery_section->next_data = sgen_get_nursery_end ();
1705 /* we should also coalesce scanning from sections close to each other
1706 * and deal with pointers outside of the sections later.
1711 sgen_client_pre_collection_checks ();
1713 if (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1714 /* Remsets are not useful for a major collection */
1715 remset.clear_cards ();
1718 sgen_process_fin_stage_entries ();
1721 sgen_init_pinning ();
1722 SGEN_LOG (6, "Collecting pinned addresses");
1723 pin_from_roots ((void*)lowest_heap_address, (void*)highest_heap_address, ctx);
1724 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1725 /* Pin cemented objects that were forced */
1726 sgen_pin_cemented_objects ();
1728 sgen_optimize_pin_queue ();
1729 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1731 * Cemented objects that are in the pinned list will be marked. When
1732 * marking concurrently we won't mark mod-union cards for these objects.
1733 * Instead they will remain cemented until the next major collection,
1734 * when we will recheck if they are still pinned in the roots.
1736 sgen_cement_force_pinned ();
1739 sgen_client_collecting_major_1 ();
1742 * pin_queue now contains all candidate pointers, sorted and
1743 * uniqued. We must do two passes now to figure out which
1744 * objects are pinned.
1746 * The first is to find within the pin_queue the area for each
1747 * section. This requires that the pin_queue be sorted. We
1748 * also process the LOS objects and pinned chunks here.
1750 * The second, destructive, pass is to reduce the section
1751 * areas to pointers to the actually pinned objects.
1753 SGEN_LOG (6, "Pinning from sections");
1754 /* first pass for the sections */
1755 sgen_find_section_pin_queue_start_end (nursery_section);
1756 /* identify possible pointers to the insize of large objects */
1757 SGEN_LOG (6, "Pinning from large objects");
1758 for (bigobj = los_object_list; bigobj; bigobj = bigobj->next) {
1760 if (sgen_find_optimized_pin_queue_area ((char*)bigobj->data, (char*)bigobj->data + sgen_los_object_size (bigobj), &dummy, &dummy)) {
1761 binary_protocol_pin (bigobj->data, (gpointer)LOAD_VTABLE (bigobj->data), safe_object_get_size (bigobj->data));
1763 if (sgen_los_object_is_pinned (bigobj->data)) {
1764 SGEN_ASSERT (0, mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, "LOS objects can only be pinned here after concurrent marking.");
1767 sgen_los_pin_object (bigobj->data);
1768 if (SGEN_OBJECT_HAS_REFERENCES (bigobj->data))
1769 GRAY_OBJECT_ENQUEUE (&gc_thread_gray_queue, bigobj->data, sgen_obj_get_descriptor ((GCObject*)bigobj->data));
1770 sgen_pin_stats_register_object (bigobj->data, GENERATION_OLD);
1771 SGEN_LOG (6, "Marked large object %p (%s) size: %lu from roots", bigobj->data,
1772 sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (bigobj->data)),
1773 (unsigned long)sgen_los_object_size (bigobj));
1775 sgen_client_pinned_los_object (bigobj->data);
1779 pin_objects_in_nursery (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT, ctx);
1780 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1781 sgen_check_nursery_objects_pinned (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1783 major_collector.pin_objects (&gc_thread_gray_queue);
1784 if (old_next_pin_slot)
1785 *old_next_pin_slot = sgen_get_pinned_count ();
1788 * We don't actually pin when starting a concurrent collection, so the remset
1789 * consistency check won't work.
1791 if (remset_consistency_checks && mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT)
1792 sgen_check_remset_consistency ();
1795 time_major_pinning += TV_ELAPSED (atv, btv);
1796 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), (long long)TV_ELAPSED (atv, btv));
1797 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1799 major_collector.init_to_space ();
1801 SGEN_ASSERT (0, sgen_workers_all_done (), "Why are the workers not done when we start or finish a major collection?");
1802 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1803 if (sgen_workers_have_idle_work ()) {
1805 * We force the finish of the worker with the new object ops context
1806 * which can also do copying. We need to have finished pinning.
1808 sgen_workers_start_all_workers (object_ops, NULL);
1809 sgen_workers_join ();
1813 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1814 main_gc_thread = mono_native_thread_self ();
1817 sgen_client_collecting_major_2 ();
1820 time_major_scan_pinned += TV_ELAPSED (btv, atv);
1822 sgen_client_collecting_major_3 (&fin_ready_queue, &critical_fin_queue);
1824 enqueue_scan_from_roots_jobs (&gc_thread_gray_queue, heap_start, heap_end, object_ops, FALSE);
1827 time_major_scan_roots += TV_ELAPSED (atv, btv);
1830 * We start the concurrent worker after pinning and after we scanned the roots
1831 * in order to make sure that the worker does not finish before handling all
1834 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1835 if (precleaning_enabled) {
1837 /* Mod union preclean job */
1838 sj = (ScanJob*)sgen_thread_pool_job_alloc ("preclean mod union cardtable", job_mod_union_preclean, sizeof (ScanJob));
1839 sj->ops = object_ops;
1840 sj->gc_thread_gray_queue = &gc_thread_gray_queue;
1841 sgen_workers_start_all_workers (object_ops, &sj->job);
1843 sgen_workers_start_all_workers (object_ops, NULL);
1845 gray_queue_enable_redirect (&gc_thread_gray_queue);
1848 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1851 /* Mod union card table */
1852 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan mod union cardtable", job_scan_major_mod_union_card_table, sizeof (ScanJob));
1853 sj->ops = object_ops;
1854 sj->gc_thread_gray_queue = &gc_thread_gray_queue;
1855 sgen_workers_enqueue_job (&sj->job, FALSE);
1857 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan LOS mod union cardtable", job_scan_los_mod_union_card_table, sizeof (ScanJob));
1858 sj->ops = object_ops;
1859 sj->gc_thread_gray_queue = &gc_thread_gray_queue;
1860 sgen_workers_enqueue_job (&sj->job, FALSE);
1863 time_major_scan_mod_union += TV_ELAPSED (btv, atv);
1866 sgen_pin_stats_report ();
1870 major_finish_copy_or_mark (CopyOrMarkFromRootsMode mode)
1872 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1873 sgen_finish_pinning ();
1875 sgen_pin_stats_reset ();
1877 if (do_concurrent_checks)
1878 sgen_debug_check_nursery_is_clean ();
1883 major_start_collection (const char *reason, gboolean concurrent, size_t *old_next_pin_slot)
1885 SgenObjectOperations *object_ops;
1887 binary_protocol_collection_begin (gc_stats.major_gc_count, GENERATION_OLD);
1889 current_collection_generation = GENERATION_OLD;
1891 sgen_workers_assert_gray_queue_is_empty ();
1894 sgen_cement_reset ();
1897 g_assert (major_collector.is_concurrent);
1898 concurrent_collection_in_progress = TRUE;
1900 object_ops = &major_collector.major_ops_concurrent_start;
1902 object_ops = &major_collector.major_ops_serial;
1905 reset_pinned_from_failed_allocation ();
1907 sgen_memgov_major_collection_start (concurrent, reason);
1909 //count_ref_nonref_objs ();
1910 //consistency_check ();
1912 check_scan_starts ();
1915 SGEN_LOG (1, "Start major collection %d", gc_stats.major_gc_count);
1916 gc_stats.major_gc_count ++;
1918 if (major_collector.start_major_collection)
1919 major_collector.start_major_collection ();
1921 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);
1922 major_finish_copy_or_mark (concurrent ? COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT : COPY_OR_MARK_FROM_ROOTS_SERIAL);
1926 major_finish_collection (const char *reason, gboolean is_overflow, size_t old_next_pin_slot, gboolean forced)
1928 ScannedObjectCounts counts;
1929 SgenObjectOperations *object_ops;
1930 mword fragment_total;
1936 if (concurrent_collection_in_progress) {
1937 object_ops = &major_collector.major_ops_concurrent_finish;
1939 major_copy_or_mark_from_roots (NULL, COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, object_ops);
1941 major_finish_copy_or_mark (COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT);
1943 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1944 main_gc_thread = NULL;
1947 object_ops = &major_collector.major_ops_serial;
1950 sgen_workers_assert_gray_queue_is_empty ();
1952 /* all the objects in the heap */
1953 finish_gray_stack (GENERATION_OLD, CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, &gc_thread_gray_queue));
1955 time_major_finish_gray_stack += TV_ELAPSED (btv, atv);
1957 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after joining");
1959 if (objects_pinned) {
1960 g_assert (!concurrent_collection_in_progress);
1963 * This is slow, but we just OOM'd.
1965 * See comment at `sgen_pin_queue_clear_discarded_entries` for how the pin
1966 * queue is laid out at this point.
1968 sgen_pin_queue_clear_discarded_entries (nursery_section, old_next_pin_slot);
1970 * We need to reestablish all pinned nursery objects in the pin queue
1971 * because they're needed for fragment creation. Unpinning happens by
1972 * walking the whole queue, so it's not necessary to reestablish where major
1973 * heap block pins are - all we care is that they're still in there
1976 sgen_optimize_pin_queue ();
1977 sgen_find_section_pin_queue_start_end (nursery_section);
1981 reset_heap_boundaries ();
1982 sgen_update_heap_boundaries ((mword)sgen_get_nursery_start (), (mword)sgen_get_nursery_end ());
1984 /* walk the pin_queue, build up the fragment list of free memory, unmark
1985 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1988 fragment_total = sgen_build_nursery_fragments (nursery_section, NULL);
1989 if (!fragment_total)
1991 SGEN_LOG (4, "Free space in nursery after major %ld", (long)fragment_total);
1993 if (do_concurrent_checks && concurrent_collection_in_progress)
1994 sgen_debug_check_nursery_is_clean ();
1996 /* prepare the pin queue for the next collection */
1997 sgen_finish_pinning ();
1999 /* Clear TLABs for all threads */
2000 sgen_clear_tlabs ();
2002 sgen_pin_stats_reset ();
2004 sgen_cement_clear_below_threshold ();
2006 if (check_mark_bits_after_major_collection)
2007 sgen_check_heap_marked (concurrent_collection_in_progress);
2010 time_major_fragment_creation += TV_ELAPSED (atv, btv);
2012 binary_protocol_sweep_begin (GENERATION_OLD, !major_collector.sweeps_lazily);
2013 sgen_memgov_major_pre_sweep ();
2016 time_major_free_bigobjs += TV_ELAPSED (btv, atv);
2021 time_major_los_sweep += TV_ELAPSED (atv, btv);
2023 major_collector.sweep ();
2025 binary_protocol_sweep_end (GENERATION_OLD, !major_collector.sweeps_lazily);
2028 time_major_sweep += TV_ELAPSED (btv, atv);
2030 sgen_debug_dump_heap ("major", gc_stats.major_gc_count - 1, reason);
2032 if (sgen_have_pending_finalizers ()) {
2033 SGEN_LOG (4, "Finalizer-thread wakeup");
2034 sgen_client_finalize_notify ();
2037 g_assert (sgen_gray_object_queue_is_empty (&gc_thread_gray_queue));
2039 sgen_memgov_major_collection_end (forced, concurrent_collection_in_progress, reason, is_overflow);
2040 current_collection_generation = -1;
2042 memset (&counts, 0, sizeof (ScannedObjectCounts));
2043 major_collector.finish_major_collection (&counts);
2045 sgen_workers_assert_gray_queue_is_empty ();
2047 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after major collection has finished");
2048 if (concurrent_collection_in_progress)
2049 concurrent_collection_in_progress = FALSE;
2051 check_scan_starts ();
2053 binary_protocol_flush_buffers (FALSE);
2055 //consistency_check ();
2057 binary_protocol_collection_end (gc_stats.major_gc_count - 1, GENERATION_OLD, counts.num_scanned_objects, counts.num_unique_scanned_objects);
2061 major_do_collection (const char *reason, gboolean is_overflow, gboolean forced)
2063 TV_DECLARE (time_start);
2064 TV_DECLARE (time_end);
2065 size_t old_next_pin_slot;
2067 if (disable_major_collections)
2070 if (major_collector.get_and_reset_num_major_objects_marked) {
2071 long long num_marked = major_collector.get_and_reset_num_major_objects_marked ();
2072 g_assert (!num_marked);
2075 /* world must be stopped already */
2076 TV_GETTIME (time_start);
2078 major_start_collection (reason, FALSE, &old_next_pin_slot);
2079 major_finish_collection (reason, is_overflow, old_next_pin_slot, forced);
2081 TV_GETTIME (time_end);
2082 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2084 /* FIXME: also report this to the user, preferably in gc-end. */
2085 if (major_collector.get_and_reset_num_major_objects_marked)
2086 major_collector.get_and_reset_num_major_objects_marked ();
2088 return bytes_pinned_from_failed_allocation > 0;
2092 major_start_concurrent_collection (const char *reason)
2094 TV_DECLARE (time_start);
2095 TV_DECLARE (time_end);
2096 long long num_objects_marked;
2098 if (disable_major_collections)
2101 TV_GETTIME (time_start);
2102 SGEN_TV_GETTIME (time_major_conc_collection_start);
2104 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2105 g_assert (num_objects_marked == 0);
2107 binary_protocol_concurrent_start ();
2109 // FIXME: store reason and pass it when finishing
2110 major_start_collection (reason, TRUE, NULL);
2111 SGEN_ASSERT (0, sgen_gray_object_queue_is_empty (&gc_thread_gray_queue), "What's left in the gray queue?");
2113 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2115 TV_GETTIME (time_end);
2116 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2118 current_collection_generation = -1;
2122 * Returns whether the major collection has finished.
2125 major_should_finish_concurrent_collection (void)
2127 SGEN_ASSERT (0, sgen_gray_object_queue_is_empty (&gc_thread_gray_queue), "Why is the gray queue not empty before we have started doing anything?");
2128 return sgen_workers_all_done ();
2132 major_update_concurrent_collection (void)
2134 TV_DECLARE (total_start);
2135 TV_DECLARE (total_end);
2137 TV_GETTIME (total_start);
2139 binary_protocol_concurrent_update ();
2141 major_collector.update_cardtable_mod_union ();
2142 sgen_los_update_cardtable_mod_union ();
2144 TV_GETTIME (total_end);
2145 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end);
2149 major_finish_concurrent_collection (gboolean forced)
2151 TV_DECLARE (total_start);
2152 TV_DECLARE (total_end);
2154 TV_GETTIME (total_start);
2156 binary_protocol_concurrent_finish ();
2159 * We need to stop all workers since we're updating the cardtable below.
2160 * The workers will be resumed with a finishing pause context to avoid
2161 * additional cardtable and object scanning.
2163 sgen_workers_stop_all_workers ();
2165 SGEN_TV_GETTIME (time_major_conc_collection_end);
2166 gc_stats.major_gc_time_concurrent += SGEN_TV_ELAPSED (time_major_conc_collection_start, time_major_conc_collection_end);
2168 major_collector.update_cardtable_mod_union ();
2169 sgen_los_update_cardtable_mod_union ();
2171 if (mod_union_consistency_check)
2172 sgen_check_mod_union_consistency ();
2174 current_collection_generation = GENERATION_OLD;
2175 sgen_cement_reset ();
2176 major_finish_collection ("finishing", FALSE, -1, forced);
2178 if (whole_heap_check_before_collection)
2179 sgen_check_whole_heap (FALSE);
2181 TV_GETTIME (total_end);
2182 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end) - TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv);
2184 current_collection_generation = -1;
2188 * Ensure an allocation request for @size will succeed by freeing enough memory.
2190 * LOCKING: The GC lock MUST be held.
2193 sgen_ensure_free_space (size_t size, int generation)
2195 int generation_to_collect = -1;
2196 const char *reason = NULL;
2198 if (generation == GENERATION_OLD) {
2199 if (sgen_need_major_collection (size)) {
2200 reason = "LOS overflow";
2201 generation_to_collect = GENERATION_OLD;
2204 if (degraded_mode) {
2205 if (sgen_need_major_collection (size)) {
2206 reason = "Degraded mode overflow";
2207 generation_to_collect = GENERATION_OLD;
2209 } else if (sgen_need_major_collection (size)) {
2210 reason = concurrent_collection_in_progress ? "Forced finish concurrent collection" : "Minor allowance";
2211 generation_to_collect = GENERATION_OLD;
2213 generation_to_collect = GENERATION_NURSERY;
2214 reason = "Nursery full";
2218 if (generation_to_collect == -1) {
2219 if (concurrent_collection_in_progress && sgen_workers_all_done ()) {
2220 generation_to_collect = GENERATION_OLD;
2221 reason = "Finish concurrent collection";
2225 if (generation_to_collect == -1)
2227 sgen_perform_collection (size, generation_to_collect, reason, FALSE, TRUE);
2231 * LOCKING: Assumes the GC lock is held.
2234 sgen_perform_collection (size_t requested_size, int generation_to_collect, const char *reason, gboolean wait_to_finish, gboolean stw)
2236 TV_DECLARE (gc_total_start);
2237 TV_DECLARE (gc_total_end);
2238 int overflow_generation_to_collect = -1;
2239 int oldest_generation_collected = generation_to_collect;
2240 const char *overflow_reason = NULL;
2241 gboolean finish_concurrent = concurrent_collection_in_progress && (major_should_finish_concurrent_collection () || generation_to_collect == GENERATION_OLD);
2243 binary_protocol_collection_requested (generation_to_collect, requested_size, wait_to_finish ? 1 : 0);
2245 SGEN_ASSERT (0, generation_to_collect == GENERATION_NURSERY || generation_to_collect == GENERATION_OLD, "What generation is this?");
2248 sgen_stop_world (generation_to_collect);
2250 SGEN_ASSERT (0, sgen_is_world_stopped (), "We can only collect if the world is stopped");
2253 TV_GETTIME (gc_total_start);
2255 // FIXME: extract overflow reason
2256 // FIXME: minor overflow for concurrent case
2257 if (generation_to_collect == GENERATION_NURSERY && !finish_concurrent) {
2258 if (concurrent_collection_in_progress)
2259 major_update_concurrent_collection ();
2261 if (collect_nursery (reason, FALSE, NULL, FALSE) && !concurrent_collection_in_progress) {
2262 overflow_generation_to_collect = GENERATION_OLD;
2263 overflow_reason = "Minor overflow";
2265 } else if (finish_concurrent) {
2266 major_finish_concurrent_collection (wait_to_finish);
2267 oldest_generation_collected = GENERATION_OLD;
2269 SGEN_ASSERT (0, generation_to_collect == GENERATION_OLD, "We should have handled nursery collections above");
2270 if (major_collector.is_concurrent && !wait_to_finish) {
2271 collect_nursery ("Concurrent start", FALSE, NULL, FALSE);
2272 major_start_concurrent_collection (reason);
2273 oldest_generation_collected = GENERATION_NURSERY;
2274 } else if (major_do_collection (reason, FALSE, wait_to_finish)) {
2275 overflow_generation_to_collect = GENERATION_NURSERY;
2276 overflow_reason = "Excessive pinning";
2280 if (overflow_generation_to_collect != -1) {
2281 SGEN_ASSERT (0, !concurrent_collection_in_progress, "We don't yet support overflow collections with the concurrent collector");
2284 * We need to do an overflow collection, either because we ran out of memory
2285 * or the nursery is fully pinned.
2288 if (overflow_generation_to_collect == GENERATION_NURSERY)
2289 collect_nursery (overflow_reason, TRUE, NULL, FALSE);
2291 major_do_collection (overflow_reason, TRUE, wait_to_finish);
2293 oldest_generation_collected = MAX (oldest_generation_collected, overflow_generation_to_collect);
2296 SGEN_LOG (2, "Heap size: %lu, LOS size: %lu", (unsigned long)sgen_gc_get_total_heap_allocation (), (unsigned long)los_memory_usage);
2298 /* this also sets the proper pointers for the next allocation */
2299 if (generation_to_collect == GENERATION_NURSERY && !sgen_can_alloc_size (requested_size)) {
2300 /* TypeBuilder and MonoMethod are killing mcs with fragmentation */
2301 SGEN_LOG (1, "nursery collection didn't find enough room for %zd alloc (%zd pinned)", requested_size, sgen_get_pinned_count ());
2302 sgen_dump_pin_queue ();
2306 g_assert (sgen_gray_object_queue_is_empty (&gc_thread_gray_queue));
2308 TV_GETTIME (gc_total_end);
2309 time_max = MAX (time_max, TV_ELAPSED (gc_total_start, gc_total_end));
2312 sgen_restart_world (oldest_generation_collected);
2316 * ######################################################################
2317 * ######## Memory allocation from the OS
2318 * ######################################################################
2319 * This section of code deals with getting memory from the OS and
2320 * allocating memory for GC-internal data structures.
2321 * Internal memory can be handled with a freelist for small objects.
2327 G_GNUC_UNUSED static void
2328 report_internal_mem_usage (void)
2330 printf ("Internal memory usage:\n");
2331 sgen_report_internal_mem_usage ();
2332 printf ("Pinned memory usage:\n");
2333 major_collector.report_pinned_memory_usage ();
2337 * ######################################################################
2338 * ######## Finalization support
2339 * ######################################################################
2343 * If the object has been forwarded it means it's still referenced from a root.
2344 * If it is pinned it's still alive as well.
2345 * A LOS object is only alive if we have pinned it.
2346 * Return TRUE if @obj is ready to be finalized.
2348 static inline gboolean
2349 sgen_is_object_alive (GCObject *object)
2351 if (ptr_in_nursery (object))
2352 return sgen_nursery_is_object_alive (object);
2354 return sgen_major_is_object_alive (object);
2358 * This function returns true if @object is either alive and belongs to the
2359 * current collection - major collections are full heap, so old gen objects
2360 * are never alive during a minor collection.
2363 sgen_is_object_alive_and_on_current_collection (GCObject *object)
2365 if (ptr_in_nursery (object))
2366 return sgen_nursery_is_object_alive (object);
2368 if (current_collection_generation == GENERATION_NURSERY)
2371 return sgen_major_is_object_alive (object);
2376 sgen_gc_is_object_ready_for_finalization (GCObject *object)
2378 return !sgen_is_object_alive (object);
2382 sgen_queue_finalization_entry (GCObject *obj)
2384 gboolean critical = sgen_client_object_has_critical_finalizer (obj);
2386 sgen_pointer_queue_add (critical ? &critical_fin_queue : &fin_ready_queue, obj);
2388 sgen_client_object_queued_for_finalization (obj);
2392 sgen_object_is_live (GCObject *obj)
2394 return sgen_is_object_alive_and_on_current_collection (obj);
2398 * `System.GC.WaitForPendingFinalizers` first checks `sgen_have_pending_finalizers()` to
2399 * determine whether it can exit quickly. The latter must therefore only return FALSE if
2400 * all finalizers have really finished running.
2402 * `sgen_gc_invoke_finalizers()` first dequeues a finalizable object, and then finalizes it.
2403 * This means that just checking whether the queues are empty leaves the possibility that an
2404 * object might have been dequeued but not yet finalized. That's why we need the additional
2405 * flag `pending_unqueued_finalizer`.
2408 static volatile gboolean pending_unqueued_finalizer = FALSE;
2411 sgen_gc_invoke_finalizers (void)
2415 g_assert (!pending_unqueued_finalizer);
2417 /* FIXME: batch to reduce lock contention */
2418 while (sgen_have_pending_finalizers ()) {
2424 * We need to set `pending_unqueued_finalizer` before dequeing the
2425 * finalizable object.
2427 if (!sgen_pointer_queue_is_empty (&fin_ready_queue)) {
2428 pending_unqueued_finalizer = TRUE;
2429 mono_memory_write_barrier ();
2430 obj = (GCObject *)sgen_pointer_queue_pop (&fin_ready_queue);
2431 } else if (!sgen_pointer_queue_is_empty (&critical_fin_queue)) {
2432 pending_unqueued_finalizer = TRUE;
2433 mono_memory_write_barrier ();
2434 obj = (GCObject *)sgen_pointer_queue_pop (&critical_fin_queue);
2440 SGEN_LOG (7, "Finalizing object %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
2448 /* the object is on the stack so it is pinned */
2449 /*g_print ("Calling finalizer for object: %p (%s)\n", obj, sgen_client_object_safe_name (obj));*/
2450 sgen_client_run_finalize (obj);
2453 if (pending_unqueued_finalizer) {
2454 mono_memory_write_barrier ();
2455 pending_unqueued_finalizer = FALSE;
2462 sgen_have_pending_finalizers (void)
2464 return pending_unqueued_finalizer || !sgen_pointer_queue_is_empty (&fin_ready_queue) || !sgen_pointer_queue_is_empty (&critical_fin_queue);
2468 * ######################################################################
2469 * ######## registered roots support
2470 * ######################################################################
2474 * We do not coalesce roots.
2477 sgen_register_root (char *start, size_t size, SgenDescriptor descr, int root_type, int source, const char *msg)
2479 RootRecord new_root;
2482 for (i = 0; i < ROOT_TYPE_NUM; ++i) {
2483 RootRecord *root = (RootRecord *)sgen_hash_table_lookup (&roots_hash [i], start);
2484 /* we allow changing the size and the descriptor (for thread statics etc) */
2486 size_t old_size = root->end_root - start;
2487 root->end_root = start + size;
2488 SGEN_ASSERT (0, !!root->root_desc == !!descr, "Can't change whether a root is precise or conservative.");
2489 SGEN_ASSERT (0, root->source == source, "Can't change a root's source identifier.");
2490 SGEN_ASSERT (0, !!root->msg == !!msg, "Can't change a root's message.");
2491 root->root_desc = descr;
2493 roots_size -= old_size;
2499 new_root.end_root = start + size;
2500 new_root.root_desc = descr;
2501 new_root.source = source;
2504 sgen_hash_table_replace (&roots_hash [root_type], start, &new_root, NULL);
2507 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);
2514 sgen_deregister_root (char* addr)
2520 for (root_type = 0; root_type < ROOT_TYPE_NUM; ++root_type) {
2521 if (sgen_hash_table_remove (&roots_hash [root_type], addr, &root))
2522 roots_size -= (root.end_root - addr);
2528 * ######################################################################
2529 * ######## Thread handling (stop/start code)
2530 * ######################################################################
2534 sgen_get_current_collection_generation (void)
2536 return current_collection_generation;
2540 sgen_thread_register (SgenThreadInfo* info, void *stack_bottom_fallback)
2542 #ifndef HAVE_KW_THREAD
2543 info->tlab_start = info->tlab_next = info->tlab_temp_end = info->tlab_real_end = NULL;
2546 sgen_init_tlab_info (info);
2548 sgen_client_thread_register (info, stack_bottom_fallback);
2554 sgen_thread_unregister (SgenThreadInfo *p)
2556 sgen_client_thread_unregister (p);
2560 * ######################################################################
2561 * ######## Write barriers
2562 * ######################################################################
2566 * Note: the write barriers first do the needed GC work and then do the actual store:
2567 * this way the value is visible to the conservative GC scan after the write barrier
2568 * itself. If a GC interrupts the barrier in the middle, value will be kept alive by
2569 * the conservative scan, otherwise by the remembered set scan.
2573 mono_gc_wbarrier_arrayref_copy (gpointer dest_ptr, gpointer src_ptr, int count)
2575 HEAVY_STAT (++stat_wbarrier_arrayref_copy);
2576 /*This check can be done without taking a lock since dest_ptr array is pinned*/
2577 if (ptr_in_nursery (dest_ptr) || count <= 0) {
2578 mono_gc_memmove_aligned (dest_ptr, src_ptr, count * sizeof (gpointer));
2582 #ifdef SGEN_HEAVY_BINARY_PROTOCOL
2583 if (binary_protocol_is_heavy_enabled ()) {
2585 for (i = 0; i < count; ++i) {
2586 gpointer dest = (gpointer*)dest_ptr + i;
2587 gpointer obj = *((gpointer*)src_ptr + i);
2589 binary_protocol_wbarrier (dest, obj, (gpointer)LOAD_VTABLE (obj));
2594 remset.wbarrier_arrayref_copy (dest_ptr, src_ptr, count);
2598 mono_gc_wbarrier_generic_nostore (gpointer ptr)
2602 HEAVY_STAT (++stat_wbarrier_generic_store);
2604 sgen_client_wbarrier_generic_nostore_check (ptr);
2606 obj = *(gpointer*)ptr;
2608 binary_protocol_wbarrier (ptr, obj, (gpointer)LOAD_VTABLE (obj));
2611 * We need to record old->old pointer locations for the
2612 * concurrent collector.
2614 if (!ptr_in_nursery (obj) && !concurrent_collection_in_progress) {
2615 SGEN_LOG (8, "Skipping remset at %p", ptr);
2619 SGEN_LOG (8, "Adding remset at %p", ptr);
2621 remset.wbarrier_generic_nostore (ptr);
2625 mono_gc_wbarrier_generic_store (gpointer ptr, GCObject* value)
2627 SGEN_LOG (8, "Wbarrier store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2628 SGEN_UPDATE_REFERENCE_ALLOW_NULL (ptr, value);
2629 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2630 mono_gc_wbarrier_generic_nostore (ptr);
2631 sgen_dummy_use (value);
2634 /* Same as mono_gc_wbarrier_generic_store () but performs the store
2635 * as an atomic operation with release semantics.
2638 mono_gc_wbarrier_generic_store_atomic (gpointer ptr, GCObject *value)
2640 HEAVY_STAT (++stat_wbarrier_generic_store_atomic);
2642 SGEN_LOG (8, "Wbarrier atomic store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2644 InterlockedWritePointer ((volatile gpointer *)ptr, value);
2646 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2647 mono_gc_wbarrier_generic_nostore (ptr);
2649 sgen_dummy_use (value);
2653 sgen_wbarrier_value_copy_bitmap (gpointer _dest, gpointer _src, int size, unsigned bitmap)
2655 GCObject **dest = (GCObject **)_dest;
2656 GCObject **src = (GCObject **)_src;
2660 mono_gc_wbarrier_generic_store (dest, *src);
2665 size -= SIZEOF_VOID_P;
2671 * ######################################################################
2672 * ######## Other mono public interface functions.
2673 * ######################################################################
2677 sgen_gc_collect (int generation)
2682 sgen_perform_collection (0, generation, "user request", TRUE, TRUE);
2687 sgen_gc_collection_count (int generation)
2689 if (generation == 0)
2690 return gc_stats.minor_gc_count;
2691 return gc_stats.major_gc_count;
2695 sgen_gc_get_used_size (void)
2699 tot = los_memory_usage;
2700 tot += nursery_section->next_data - nursery_section->data;
2701 tot += major_collector.get_used_size ();
2702 /* FIXME: account for pinned objects */
2708 sgen_env_var_error (const char *env_var, const char *fallback, const char *description_format, ...)
2712 va_start (ap, description_format);
2714 fprintf (stderr, "Warning: In environment variable `%s': ", env_var);
2715 vfprintf (stderr, description_format, ap);
2717 fprintf (stderr, " - %s", fallback);
2718 fprintf (stderr, "\n");
2724 parse_double_in_interval (const char *env_var, const char *opt_name, const char *opt, double min, double max, double *result)
2727 double val = strtod (opt, &endptr);
2728 if (endptr == opt) {
2729 sgen_env_var_error (env_var, "Using default value.", "`%s` must be a number.", opt_name);
2732 else if (val < min || val > max) {
2733 sgen_env_var_error (env_var, "Using default value.", "`%s` must be between %.2f - %.2f.", opt_name, min, max);
2745 char *major_collector_opt = NULL;
2746 char *minor_collector_opt = NULL;
2747 size_t max_heap = 0;
2748 size_t soft_limit = 0;
2750 gboolean debug_print_allowance = FALSE;
2751 double allowance_ratio = 0, save_target = 0;
2752 gboolean cement_enabled = TRUE;
2755 result = InterlockedCompareExchange (&gc_initialized, -1, 0);
2758 /* already inited */
2761 /* being inited by another thread */
2762 mono_thread_info_usleep (1000);
2765 /* we will init it */
2768 g_assert_not_reached ();
2770 } while (result != 0);
2772 SGEN_TV_GETTIME (sgen_init_timestamp);
2774 #ifdef SGEN_WITHOUT_MONO
2775 mono_thread_smr_init ();
2778 mono_coop_mutex_init (&gc_mutex);
2780 gc_debug_file = stderr;
2782 mono_coop_mutex_init (&sgen_interruption_mutex);
2784 if ((env = g_getenv (MONO_GC_PARAMS_NAME))) {
2785 opts = g_strsplit (env, ",", -1);
2786 for (ptr = opts; *ptr; ++ptr) {
2788 if (g_str_has_prefix (opt, "major=")) {
2789 opt = strchr (opt, '=') + 1;
2790 major_collector_opt = g_strdup (opt);
2791 } else if (g_str_has_prefix (opt, "minor=")) {
2792 opt = strchr (opt, '=') + 1;
2793 minor_collector_opt = g_strdup (opt);
2801 sgen_init_internal_allocator ();
2802 sgen_init_nursery_allocator ();
2803 sgen_init_fin_weak_hash ();
2804 sgen_init_hash_table ();
2805 sgen_init_descriptors ();
2806 sgen_init_gray_queues ();
2807 sgen_init_allocator ();
2808 sgen_init_gchandles ();
2810 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_SECTION, SGEN_SIZEOF_GC_MEM_SECTION);
2811 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_GRAY_QUEUE, sizeof (GrayQueueSection));
2813 sgen_client_init ();
2815 if (!minor_collector_opt) {
2816 sgen_simple_nursery_init (&sgen_minor_collector);
2818 if (!strcmp (minor_collector_opt, "simple")) {
2820 sgen_simple_nursery_init (&sgen_minor_collector);
2821 } else if (!strcmp (minor_collector_opt, "split")) {
2822 sgen_split_nursery_init (&sgen_minor_collector);
2824 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `simple` instead.", "Unknown minor collector `%s'.", minor_collector_opt);
2825 goto use_simple_nursery;
2829 if (!major_collector_opt) {
2831 DEFAULT_MAJOR_INIT (&major_collector);
2832 } else if (!strcmp (major_collector_opt, "marksweep")) {
2833 sgen_marksweep_init (&major_collector);
2834 } else if (!strcmp (major_collector_opt, "marksweep-conc")) {
2835 sgen_marksweep_conc_init (&major_collector);
2837 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `" DEFAULT_MAJOR_NAME "` instead.", "Unknown major collector `%s'.", major_collector_opt);
2838 goto use_default_major;
2841 sgen_nursery_size = DEFAULT_NURSERY_SIZE;
2844 gboolean usage_printed = FALSE;
2846 for (ptr = opts; *ptr; ++ptr) {
2848 if (!strcmp (opt, ""))
2850 if (g_str_has_prefix (opt, "major="))
2852 if (g_str_has_prefix (opt, "minor="))
2854 if (g_str_has_prefix (opt, "max-heap-size=")) {
2855 size_t page_size = mono_pagesize ();
2856 size_t max_heap_candidate = 0;
2857 opt = strchr (opt, '=') + 1;
2858 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &max_heap_candidate)) {
2859 max_heap = (max_heap_candidate + page_size - 1) & ~(size_t)(page_size - 1);
2860 if (max_heap != max_heap_candidate)
2861 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Rounding up.", "`max-heap-size` size must be a multiple of %d.", page_size);
2863 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`max-heap-size` must be an integer.");
2867 if (g_str_has_prefix (opt, "soft-heap-limit=")) {
2868 opt = strchr (opt, '=') + 1;
2869 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &soft_limit)) {
2870 if (soft_limit <= 0) {
2871 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be positive.");
2875 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be an integer.");
2881 if (g_str_has_prefix (opt, "nursery-size=")) {
2883 opt = strchr (opt, '=') + 1;
2884 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &val)) {
2885 if ((val & (val - 1))) {
2886 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be a power of two.");
2890 if (val < SGEN_MAX_NURSERY_WASTE) {
2891 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.",
2892 "`nursery-size` must be at least %d bytes.", SGEN_MAX_NURSERY_WASTE);
2896 sgen_nursery_size = val;
2897 sgen_nursery_bits = 0;
2898 while (ONE_P << (++ sgen_nursery_bits) != sgen_nursery_size)
2901 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be an integer.");
2907 if (g_str_has_prefix (opt, "save-target-ratio=")) {
2909 opt = strchr (opt, '=') + 1;
2910 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "save-target-ratio", opt,
2911 SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO, &val)) {
2916 if (g_str_has_prefix (opt, "default-allowance-ratio=")) {
2918 opt = strchr (opt, '=') + 1;
2919 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "default-allowance-ratio", opt,
2920 SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, SGEN_MAX_ALLOWANCE_NURSERY_SIZE_RATIO, &val)) {
2921 allowance_ratio = val;
2926 if (!strcmp (opt, "cementing")) {
2927 cement_enabled = TRUE;
2930 if (!strcmp (opt, "no-cementing")) {
2931 cement_enabled = FALSE;
2935 if (!strcmp (opt, "precleaning")) {
2936 precleaning_enabled = TRUE;
2939 if (!strcmp (opt, "no-precleaning")) {
2940 precleaning_enabled = FALSE;
2944 if (major_collector.handle_gc_param && major_collector.handle_gc_param (opt))
2947 if (sgen_minor_collector.handle_gc_param && sgen_minor_collector.handle_gc_param (opt))
2950 if (sgen_client_handle_gc_param (opt))
2953 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Ignoring.", "Unknown option `%s`.", opt);
2958 fprintf (stderr, "\n%s must be a comma-delimited list of one or more of the following:\n", MONO_GC_PARAMS_NAME);
2959 fprintf (stderr, " max-heap-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
2960 fprintf (stderr, " soft-heap-limit=n (where N is an integer, possibly with a k, m or a g suffix)\n");
2961 fprintf (stderr, " nursery-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
2962 fprintf (stderr, " major=COLLECTOR (where COLLECTOR is `marksweep', `marksweep-conc', `marksweep-par')\n");
2963 fprintf (stderr, " minor=COLLECTOR (where COLLECTOR is `simple' or `split')\n");
2964 fprintf (stderr, " wbarrier=WBARRIER (where WBARRIER is `remset' or `cardtable')\n");
2965 fprintf (stderr, " [no-]cementing\n");
2966 if (major_collector.print_gc_param_usage)
2967 major_collector.print_gc_param_usage ();
2968 if (sgen_minor_collector.print_gc_param_usage)
2969 sgen_minor_collector.print_gc_param_usage ();
2970 sgen_client_print_gc_params_usage ();
2971 fprintf (stderr, " Experimental options:\n");
2972 fprintf (stderr, " save-target-ratio=R (where R must be between %.2f - %.2f).\n", SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO);
2973 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);
2974 fprintf (stderr, "\n");
2976 usage_printed = TRUE;
2981 if (major_collector_opt)
2982 g_free (major_collector_opt);
2984 if (minor_collector_opt)
2985 g_free (minor_collector_opt);
2989 sgen_pinning_init ();
2990 sgen_cement_init (cement_enabled);
2992 if ((env = g_getenv (MONO_GC_DEBUG_NAME))) {
2993 gboolean usage_printed = FALSE;
2995 opts = g_strsplit (env, ",", -1);
2996 for (ptr = opts; ptr && *ptr; ptr ++) {
2998 if (!strcmp (opt, ""))
3000 if (opt [0] >= '0' && opt [0] <= '9') {
3001 gc_debug_level = atoi (opt);
3006 char *rf = g_strdup_printf ("%s.%d", opt, mono_process_current_pid ());
3007 gc_debug_file = fopen (rf, "wb");
3009 gc_debug_file = stderr;
3012 } else if (!strcmp (opt, "print-allowance")) {
3013 debug_print_allowance = TRUE;
3014 } else if (!strcmp (opt, "print-pinning")) {
3015 sgen_pin_stats_enable ();
3016 } else if (!strcmp (opt, "verify-before-allocs")) {
3017 verify_before_allocs = 1;
3018 has_per_allocation_action = TRUE;
3019 } else if (g_str_has_prefix (opt, "verify-before-allocs=")) {
3020 char *arg = strchr (opt, '=') + 1;
3021 verify_before_allocs = atoi (arg);
3022 has_per_allocation_action = TRUE;
3023 } else if (!strcmp (opt, "collect-before-allocs")) {
3024 collect_before_allocs = 1;
3025 has_per_allocation_action = TRUE;
3026 } else if (g_str_has_prefix (opt, "collect-before-allocs=")) {
3027 char *arg = strchr (opt, '=') + 1;
3028 has_per_allocation_action = TRUE;
3029 collect_before_allocs = atoi (arg);
3030 } else if (!strcmp (opt, "verify-before-collections")) {
3031 whole_heap_check_before_collection = TRUE;
3032 } else if (!strcmp (opt, "check-remset-consistency")) {
3033 remset_consistency_checks = TRUE;
3034 nursery_clear_policy = CLEAR_AT_GC;
3035 } else if (!strcmp (opt, "mod-union-consistency-check")) {
3036 if (!major_collector.is_concurrent) {
3037 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`mod-union-consistency-check` only works with concurrent major collector.");
3040 mod_union_consistency_check = TRUE;
3041 } else if (!strcmp (opt, "check-mark-bits")) {
3042 check_mark_bits_after_major_collection = TRUE;
3043 } else if (!strcmp (opt, "check-nursery-pinned")) {
3044 check_nursery_objects_pinned = TRUE;
3045 } else if (!strcmp (opt, "clear-at-gc")) {
3046 nursery_clear_policy = CLEAR_AT_GC;
3047 } else if (!strcmp (opt, "clear-nursery-at-gc")) {
3048 nursery_clear_policy = CLEAR_AT_GC;
3049 } else if (!strcmp (opt, "clear-at-tlab-creation")) {
3050 nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
3051 } else if (!strcmp (opt, "debug-clear-at-tlab-creation")) {
3052 nursery_clear_policy = CLEAR_AT_TLAB_CREATION_DEBUG;
3053 } else if (!strcmp (opt, "check-scan-starts")) {
3054 do_scan_starts_check = TRUE;
3055 } else if (!strcmp (opt, "verify-nursery-at-minor-gc")) {
3056 do_verify_nursery = TRUE;
3057 } else if (!strcmp (opt, "check-concurrent")) {
3058 if (!major_collector.is_concurrent) {
3059 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`check-concurrent` only works with concurrent major collectors.");
3062 nursery_clear_policy = CLEAR_AT_GC;
3063 do_concurrent_checks = TRUE;
3064 } else if (!strcmp (opt, "dump-nursery-at-minor-gc")) {
3065 do_dump_nursery_content = TRUE;
3066 } else if (!strcmp (opt, "disable-minor")) {
3067 disable_minor_collections = TRUE;
3068 } else if (!strcmp (opt, "disable-major")) {
3069 disable_major_collections = TRUE;
3070 } else if (g_str_has_prefix (opt, "heap-dump=")) {
3071 char *filename = strchr (opt, '=') + 1;
3072 nursery_clear_policy = CLEAR_AT_GC;
3073 sgen_debug_enable_heap_dump (filename);
3074 } else if (g_str_has_prefix (opt, "binary-protocol=")) {
3075 char *filename = strchr (opt, '=') + 1;
3076 char *colon = strrchr (filename, ':');
3079 if (!mono_gc_parse_environment_string_extract_number (colon + 1, &limit)) {
3080 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring limit.", "Binary protocol file size limit must be an integer.");
3085 binary_protocol_init (filename, (long long)limit);
3086 } else if (!strcmp (opt, "nursery-canaries")) {
3087 do_verify_nursery = TRUE;
3088 enable_nursery_canaries = TRUE;
3089 } else if (!sgen_client_handle_gc_debug (opt)) {
3090 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "Unknown option `%s`.", opt);
3095 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);
3096 fprintf (stderr, "Valid <option>s are:\n");
3097 fprintf (stderr, " collect-before-allocs[=<n>]\n");
3098 fprintf (stderr, " verify-before-allocs[=<n>]\n");
3099 fprintf (stderr, " check-remset-consistency\n");
3100 fprintf (stderr, " check-mark-bits\n");
3101 fprintf (stderr, " check-nursery-pinned\n");
3102 fprintf (stderr, " verify-before-collections\n");
3103 fprintf (stderr, " verify-nursery-at-minor-gc\n");
3104 fprintf (stderr, " dump-nursery-at-minor-gc\n");
3105 fprintf (stderr, " disable-minor\n");
3106 fprintf (stderr, " disable-major\n");
3107 fprintf (stderr, " check-concurrent\n");
3108 fprintf (stderr, " clear-[nursery-]at-gc\n");
3109 fprintf (stderr, " clear-at-tlab-creation\n");
3110 fprintf (stderr, " debug-clear-at-tlab-creation\n");
3111 fprintf (stderr, " check-scan-starts\n");
3112 fprintf (stderr, " print-allowance\n");
3113 fprintf (stderr, " print-pinning\n");
3114 fprintf (stderr, " heap-dump=<filename>\n");
3115 fprintf (stderr, " binary-protocol=<filename>[:<file-size-limit>]\n");
3116 fprintf (stderr, " nursery-canaries\n");
3117 sgen_client_print_gc_debug_usage ();
3118 fprintf (stderr, "\n");
3120 usage_printed = TRUE;
3126 if (check_mark_bits_after_major_collection)
3127 nursery_clear_policy = CLEAR_AT_GC;
3129 if (major_collector.post_param_init)
3130 major_collector.post_param_init (&major_collector);
3132 if (major_collector.needs_thread_pool)
3133 sgen_workers_init (1);
3135 sgen_memgov_init (max_heap, soft_limit, debug_print_allowance, allowance_ratio, save_target);
3137 memset (&remset, 0, sizeof (remset));
3139 sgen_card_table_init (&remset);
3141 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");
3145 sgen_init_bridge ();
3149 sgen_gc_initialized ()
3151 return gc_initialized > 0;
3155 sgen_get_nursery_clear_policy (void)
3157 return nursery_clear_policy;
3163 mono_coop_mutex_lock (&gc_mutex);
3167 sgen_gc_unlock (void)
3169 mono_coop_mutex_unlock (&gc_mutex);
3173 sgen_major_collector_iterate_live_block_ranges (sgen_cardtable_block_callback callback)
3175 major_collector.iterate_live_block_ranges (callback);
3179 sgen_major_collector_iterate_block_ranges (sgen_cardtable_block_callback callback)
3181 major_collector.iterate_block_ranges (callback);
3185 sgen_get_major_collector (void)
3187 return &major_collector;
3191 sgen_get_remset (void)
3197 count_cards (long long *major_total, long long *major_marked, long long *los_total, long long *los_marked)
3199 sgen_get_major_collector ()->count_cards (major_total, major_marked);
3200 sgen_los_count_cards (los_total, los_marked);
3203 static gboolean world_is_stopped = FALSE;
3205 /* LOCKING: assumes the GC lock is held */
3207 sgen_stop_world (int generation)
3209 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3211 SGEN_ASSERT (0, !world_is_stopped, "Why are we stopping a stopped world?");
3213 binary_protocol_world_stopping (generation, sgen_timestamp (), (gpointer) (gsize) mono_native_thread_id_get ());
3215 sgen_client_stop_world (generation);
3217 world_is_stopped = TRUE;
3219 if (binary_protocol_is_heavy_enabled ())
3220 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3221 binary_protocol_world_stopped (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3224 /* LOCKING: assumes the GC lock is held */
3226 sgen_restart_world (int generation)
3228 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3231 SGEN_ASSERT (0, world_is_stopped, "Why are we restarting a running world?");
3233 if (binary_protocol_is_heavy_enabled ())
3234 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3235 binary_protocol_world_restarting (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3237 world_is_stopped = FALSE;
3239 sgen_client_restart_world (generation, &stw_time);
3241 binary_protocol_world_restarted (generation, sgen_timestamp ());
3243 if (sgen_client_bridge_need_processing ())
3244 sgen_client_bridge_processing_finish (generation);
3246 sgen_memgov_collection_end (generation, stw_time);
3250 sgen_is_world_stopped (void)
3252 return world_is_stopped;
3256 sgen_check_whole_heap_stw (void)
3258 sgen_stop_world (0);
3259 sgen_clear_nursery_fragments ();
3260 sgen_check_whole_heap (FALSE);
3261 sgen_restart_world (0);
3265 sgen_timestamp (void)
3267 SGEN_TV_DECLARE (timestamp);
3268 SGEN_TV_GETTIME (timestamp);
3269 return SGEN_TV_ELAPSED (sgen_init_timestamp, timestamp);
3272 #endif /* HAVE_SGEN_GC */