2 * sgen-gc.c: Simple generational GC.
5 * Paolo Molaro (lupus@ximian.com)
6 * Rodrigo Kumpera (kumpera@gmail.com)
8 * Copyright 2005-2011 Novell, Inc (http://www.novell.com)
9 * Copyright 2011 Xamarin Inc (http://www.xamarin.com)
11 * Thread start/stop adapted from Boehm's GC:
12 * Copyright (c) 1994 by Xerox Corporation. All rights reserved.
13 * Copyright (c) 1996 by Silicon Graphics. All rights reserved.
14 * Copyright (c) 1998 by Fergus Henderson. All rights reserved.
15 * Copyright (c) 2000-2004 by Hewlett-Packard Company. All rights reserved.
16 * Copyright 2001-2003 Ximian, Inc
17 * Copyright 2003-2010 Novell, Inc.
18 * Copyright 2011 Xamarin, Inc.
19 * Copyright (C) 2012 Xamarin Inc
21 * Licensed under the MIT license. See LICENSE file in the project root for full license information.
23 * Important: allocation provides always zeroed memory, having to do
24 * a memset after allocation is deadly for performance.
25 * Memory usage at startup is currently as follows:
27 * 64 KB internal space
29 * We should provide a small memory config with half the sizes
31 * We currently try to make as few mono assumptions as possible:
32 * 1) 2-word header with no GC pointers in it (first vtable, second to store the
34 * 2) gc descriptor is the second word in the vtable (first word in the class)
35 * 3) 8 byte alignment is the minimum and enough (not true for special structures (SIMD), FIXME)
36 * 4) there is a function to get an object's size and the number of
37 * elements in an array.
38 * 5) we know the special way bounds are allocated for complex arrays
39 * 6) we know about proxies and how to treat them when domains are unloaded
41 * Always try to keep stack usage to a minimum: no recursive behaviour
42 * and no large stack allocs.
44 * General description.
45 * Objects are initially allocated in a nursery using a fast bump-pointer technique.
46 * When the nursery is full we start a nursery collection: this is performed with a
48 * When the old generation is full we start a copying GC of the old generation as well:
49 * this will be changed to mark&sweep with copying when fragmentation becomes to severe
50 * in the future. Maybe we'll even do both during the same collection like IMMIX.
52 * The things that complicate this description are:
53 * *) pinned objects: we can't move them so we need to keep track of them
54 * *) no precise info of the thread stacks and registers: we need to be able to
55 * quickly find the objects that may be referenced conservatively and pin them
56 * (this makes the first issues more important)
57 * *) large objects are too expensive to be dealt with using copying GC: we handle them
58 * with mark/sweep during major collections
59 * *) some objects need to not move even if they are small (interned strings, Type handles):
60 * we use mark/sweep for them, too: they are not allocated in the nursery, but inside
61 * PinnedChunks regions
67 *) we could have a function pointer in MonoClass to implement
68 customized write barriers for value types
70 *) investigate the stuff needed to advance a thread to a GC-safe
71 point (single-stepping, read from unmapped memory etc) and implement it.
72 This would enable us to inline allocations and write barriers, for example,
73 or at least parts of them, like the write barrier checks.
74 We may need this also for handling precise info on stacks, even simple things
75 as having uninitialized data on the stack and having to wait for the prolog
76 to zero it. Not an issue for the last frame that we scan conservatively.
77 We could always not trust the value in the slots anyway.
79 *) modify the jit to save info about references in stack locations:
80 this can be done just for locals as a start, so that at least
81 part of the stack is handled precisely.
83 *) test/fix endianess issues
85 *) Implement a card table as the write barrier instead of remembered
86 sets? Card tables are not easy to implement with our current
87 memory layout. We have several different kinds of major heap
88 objects: Small objects in regular blocks, small objects in pinned
89 chunks and LOS objects. If we just have a pointer we have no way
90 to tell which kind of object it points into, therefore we cannot
91 know where its card table is. The least we have to do to make
92 this happen is to get rid of write barriers for indirect stores.
95 *) Get rid of write barriers for indirect stores. We can do this by
96 telling the GC to wbarrier-register an object once we do an ldloca
97 or ldelema on it, and to unregister it once it's not used anymore
98 (it can only travel downwards on the stack). The problem with
99 unregistering is that it needs to happen eventually no matter
100 what, even if exceptions are thrown, the thread aborts, etc.
101 Rodrigo suggested that we could do only the registering part and
102 let the collector find out (pessimistically) when it's safe to
103 unregister, namely when the stack pointer of the thread that
104 registered the object is higher than it was when the registering
105 happened. This might make for a good first implementation to get
106 some data on performance.
108 *) Some sort of blacklist support? Blacklists is a concept from the
109 Boehm GC: if during a conservative scan we find pointers to an
110 area which we might use as heap, we mark that area as unusable, so
111 pointer retention by random pinning pointers is reduced.
113 *) experiment with max small object size (very small right now - 2kb,
114 because it's tied to the max freelist size)
116 *) add an option to mmap the whole heap in one chunk: it makes for many
117 simplifications in the checks (put the nursery at the top and just use a single
118 check for inclusion/exclusion): the issue this has is that on 32 bit systems it's
119 not flexible (too much of the address space may be used by default or we can't
120 increase the heap as needed) and we'd need a race-free mechanism to return memory
121 back to the system (mprotect(PROT_NONE) will still keep the memory allocated if it
122 was written to, munmap is needed, but the following mmap may not find the same segment
125 *) memzero the major fragments after restarting the world and optionally a smaller
128 *) investigate having fragment zeroing threads
130 *) separate locks for finalization and other minor stuff to reduce
133 *) try a different copying order to improve memory locality
135 *) a thread abort after a store but before the write barrier will
136 prevent the write barrier from executing
138 *) specialized dynamically generated markers/copiers
140 *) Dynamically adjust TLAB size to the number of threads. If we have
141 too many threads that do allocation, we might need smaller TLABs,
142 and we might get better performance with larger TLABs if we only
143 have a handful of threads. We could sum up the space left in all
144 assigned TLABs and if that's more than some percentage of the
145 nursery size, reduce the TLAB size.
147 *) Explore placing unreachable objects on unused nursery memory.
148 Instead of memset'ng a region to zero, place an int[] covering it.
149 A good place to start is add_nursery_frag. The tricky thing here is
150 placing those objects atomically outside of a collection.
152 *) Allocation should use asymmetric Dekker synchronization:
153 http://blogs.oracle.com/dave/resource/Asymmetric-Dekker-Synchronization.txt
154 This should help weak consistency archs.
161 #define _XOPEN_SOURCE
162 #define _DARWIN_C_SOURCE
168 #ifdef HAVE_PTHREAD_H
171 #ifdef HAVE_PTHREAD_NP_H
172 #include <pthread_np.h>
180 #include "mono/sgen/sgen-gc.h"
181 #include "mono/sgen/sgen-cardtable.h"
182 #include "mono/sgen/sgen-protocol.h"
183 #include "mono/sgen/sgen-memory-governor.h"
184 #include "mono/sgen/sgen-hash-table.h"
185 #include "mono/sgen/sgen-cardtable.h"
186 #include "mono/sgen/sgen-pinning.h"
187 #include "mono/sgen/sgen-workers.h"
188 #include "mono/sgen/sgen-client.h"
189 #include "mono/sgen/sgen-pointer-queue.h"
190 #include "mono/sgen/gc-internal-agnostic.h"
191 #include "mono/utils/mono-proclib.h"
192 #include "mono/utils/mono-memory-model.h"
193 #include "mono/utils/hazard-pointer.h"
195 #include <mono/utils/memcheck.h>
197 #undef pthread_create
199 #undef pthread_detach
202 * ######################################################################
203 * ######## Types and constants used by the GC.
204 * ######################################################################
207 /* 0 means not initialized, 1 is initialized, -1 means in progress */
208 static int gc_initialized = 0;
209 /* If set, check if we need to do something every X allocations */
210 gboolean has_per_allocation_action;
211 /* If set, do a heap check every X allocation */
212 guint32 verify_before_allocs = 0;
213 /* If set, do a minor collection before every X allocation */
214 guint32 collect_before_allocs = 0;
215 /* If set, do a whole heap check before each collection */
216 static gboolean whole_heap_check_before_collection = FALSE;
217 /* If set, do a remset consistency check at various opportunities */
218 static gboolean remset_consistency_checks = FALSE;
219 /* If set, do a mod union consistency check before each finishing collection pause */
220 static gboolean mod_union_consistency_check = FALSE;
221 /* If set, check whether mark bits are consistent after major collections */
222 static gboolean check_mark_bits_after_major_collection = FALSE;
223 /* If set, check that all nursery objects are pinned/not pinned, depending on context */
224 static gboolean check_nursery_objects_pinned = FALSE;
225 /* If set, do a few checks when the concurrent collector is used */
226 static gboolean do_concurrent_checks = FALSE;
227 /* If set, do a plausibility check on the scan_starts before and after
229 static gboolean do_scan_starts_check = FALSE;
231 static gboolean disable_minor_collections = FALSE;
232 static gboolean disable_major_collections = FALSE;
233 static gboolean do_verify_nursery = FALSE;
234 static gboolean do_dump_nursery_content = FALSE;
235 static gboolean enable_nursery_canaries = FALSE;
237 static gboolean precleaning_enabled = TRUE;
239 #ifdef HEAVY_STATISTICS
240 guint64 stat_objects_alloced_degraded = 0;
241 guint64 stat_bytes_alloced_degraded = 0;
243 guint64 stat_copy_object_called_nursery = 0;
244 guint64 stat_objects_copied_nursery = 0;
245 guint64 stat_copy_object_called_major = 0;
246 guint64 stat_objects_copied_major = 0;
248 guint64 stat_scan_object_called_nursery = 0;
249 guint64 stat_scan_object_called_major = 0;
251 guint64 stat_slots_allocated_in_vain;
253 guint64 stat_nursery_copy_object_failed_from_space = 0;
254 guint64 stat_nursery_copy_object_failed_forwarded = 0;
255 guint64 stat_nursery_copy_object_failed_pinned = 0;
256 guint64 stat_nursery_copy_object_failed_to_space = 0;
258 static guint64 stat_wbarrier_add_to_global_remset = 0;
259 static guint64 stat_wbarrier_arrayref_copy = 0;
260 static guint64 stat_wbarrier_generic_store = 0;
261 static guint64 stat_wbarrier_generic_store_atomic = 0;
262 static guint64 stat_wbarrier_set_root = 0;
265 static guint64 stat_pinned_objects = 0;
267 static guint64 time_minor_pre_collection_fragment_clear = 0;
268 static guint64 time_minor_pinning = 0;
269 static guint64 time_minor_scan_remsets = 0;
270 static guint64 time_minor_scan_pinned = 0;
271 static guint64 time_minor_scan_roots = 0;
272 static guint64 time_minor_finish_gray_stack = 0;
273 static guint64 time_minor_fragment_creation = 0;
275 static guint64 time_major_pre_collection_fragment_clear = 0;
276 static guint64 time_major_pinning = 0;
277 static guint64 time_major_scan_pinned = 0;
278 static guint64 time_major_scan_roots = 0;
279 static guint64 time_major_scan_mod_union = 0;
280 static guint64 time_major_finish_gray_stack = 0;
281 static guint64 time_major_free_bigobjs = 0;
282 static guint64 time_major_los_sweep = 0;
283 static guint64 time_major_sweep = 0;
284 static guint64 time_major_fragment_creation = 0;
286 static guint64 time_max = 0;
288 static SGEN_TV_DECLARE (time_major_conc_collection_start);
289 static SGEN_TV_DECLARE (time_major_conc_collection_end);
291 static SGEN_TV_DECLARE (last_minor_collection_start_tv);
292 static SGEN_TV_DECLARE (last_minor_collection_end_tv);
294 int gc_debug_level = 0;
299 mono_gc_flush_info (void)
301 fflush (gc_debug_file);
305 #define TV_DECLARE SGEN_TV_DECLARE
306 #define TV_GETTIME SGEN_TV_GETTIME
307 #define TV_ELAPSED SGEN_TV_ELAPSED
309 static SGEN_TV_DECLARE (sgen_init_timestamp);
311 NurseryClearPolicy nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
313 #define object_is_forwarded SGEN_OBJECT_IS_FORWARDED
314 #define object_is_pinned SGEN_OBJECT_IS_PINNED
315 #define pin_object SGEN_PIN_OBJECT
317 #define ptr_in_nursery sgen_ptr_in_nursery
319 #define LOAD_VTABLE SGEN_LOAD_VTABLE
322 nursery_canaries_enabled (void)
324 return enable_nursery_canaries;
327 #define safe_object_get_size sgen_safe_object_get_size
329 #if defined(HAVE_CONC_GC_AS_DEFAULT)
330 /* Use concurrent major on deskstop platforms */
331 #define DEFAULT_MAJOR_INIT sgen_marksweep_conc_init
332 #define DEFAULT_MAJOR_NAME "marksweep-conc"
334 #define DEFAULT_MAJOR_INIT sgen_marksweep_init
335 #define DEFAULT_MAJOR_NAME "marksweep"
339 * ######################################################################
340 * ######## Global data.
341 * ######################################################################
343 MonoCoopMutex gc_mutex;
345 #define SCAN_START_SIZE SGEN_SCAN_START_SIZE
347 size_t degraded_mode = 0;
349 static mword bytes_pinned_from_failed_allocation = 0;
351 GCMemSection *nursery_section = NULL;
352 static volatile mword lowest_heap_address = ~(mword)0;
353 static volatile mword highest_heap_address = 0;
355 MonoCoopMutex sgen_interruption_mutex;
357 int current_collection_generation = -1;
358 static volatile gboolean concurrent_collection_in_progress = FALSE;
360 /* objects that are ready to be finalized */
361 static SgenPointerQueue fin_ready_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
362 static SgenPointerQueue critical_fin_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
364 /* registered roots: the key to the hash is the root start address */
366 * Different kinds of roots are kept separate to speed up pin_from_roots () for example.
368 SgenHashTable roots_hash [ROOT_TYPE_NUM] = {
369 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
370 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
371 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL)
373 static mword roots_size = 0; /* amount of memory in the root set */
375 /* The size of a TLAB */
376 /* The bigger the value, the less often we have to go to the slow path to allocate a new
377 * one, but the more space is wasted by threads not allocating much memory.
379 * FIXME: Make this self-tuning for each thread.
381 guint32 tlab_size = (1024 * 4);
383 #define MAX_SMALL_OBJ_SIZE SGEN_MAX_SMALL_OBJ_SIZE
385 #define ALLOC_ALIGN SGEN_ALLOC_ALIGN
387 #define ALIGN_UP SGEN_ALIGN_UP
389 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
390 MonoNativeThreadId main_gc_thread = NULL;
393 /*Object was pinned during the current collection*/
394 static mword objects_pinned;
397 * ######################################################################
398 * ######## Macros and function declarations.
399 * ######################################################################
402 typedef SgenGrayQueue GrayQueue;
404 /* forward declarations */
405 static void scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx);
407 static void pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx);
408 static void finish_gray_stack (int generation, ScanCopyContext ctx);
411 SgenMajorCollector major_collector;
412 SgenMinorCollector sgen_minor_collector;
413 /* FIXME: get rid of this */
414 static GrayQueue gray_queue;
416 static SgenRememberedSet remset;
418 /* The gray queue to use from the main collection thread. */
419 #define WORKERS_DISTRIBUTE_GRAY_QUEUE (&gray_queue)
422 * The gray queue a worker job must use. If we're not parallel or
423 * concurrent, we use the main gray queue.
425 static SgenGrayQueue*
426 sgen_workers_get_job_gray_queue (WorkerData *worker_data)
428 return worker_data ? &worker_data->private_gray_queue : WORKERS_DISTRIBUTE_GRAY_QUEUE;
432 gray_queue_redirect (SgenGrayQueue *queue)
434 gboolean wake = FALSE;
437 GrayQueueSection *section = sgen_gray_object_dequeue_section (queue);
440 sgen_section_gray_queue_enqueue ((SgenSectionGrayQueue *)queue->alloc_prepare_data, section);
445 g_assert (concurrent_collection_in_progress);
446 sgen_workers_ensure_awake ();
451 gray_queue_enable_redirect (SgenGrayQueue *queue)
453 if (!concurrent_collection_in_progress)
456 sgen_gray_queue_set_alloc_prepare (queue, gray_queue_redirect, sgen_workers_get_distribute_section_gray_queue ());
457 gray_queue_redirect (queue);
461 sgen_scan_area_with_callback (char *start, char *end, IterateObjectCallbackFunc callback, void *data, gboolean allow_flags, gboolean fail_on_canaries)
463 while (start < end) {
467 if (!*(void**)start) {
468 start += sizeof (void*); /* should be ALLOC_ALIGN, really */
473 if (!(obj = (char *)SGEN_OBJECT_IS_FORWARDED (start)))
479 if (!sgen_client_object_is_array_fill ((GCObject*)obj)) {
480 CHECK_CANARY_FOR_OBJECT ((GCObject*)obj, fail_on_canaries);
481 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
482 callback ((GCObject*)obj, size, data);
483 CANARIFY_SIZE (size);
485 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
493 * sgen_add_to_global_remset:
495 * The global remset contains locations which point into newspace after
496 * a minor collection. This can happen if the objects they point to are pinned.
498 * LOCKING: If called from a parallel collector, the global remset
499 * lock must be held. For serial collectors that is not necessary.
502 sgen_add_to_global_remset (gpointer ptr, GCObject *obj)
504 SGEN_ASSERT (5, sgen_ptr_in_nursery (obj), "Target pointer of global remset must be in the nursery");
506 HEAVY_STAT (++stat_wbarrier_add_to_global_remset);
508 if (!major_collector.is_concurrent) {
509 SGEN_ASSERT (5, current_collection_generation != -1, "Global remsets can only be added during collections");
511 if (current_collection_generation == -1)
512 SGEN_ASSERT (5, sgen_concurrent_collection_in_progress (), "Global remsets outside of collection pauses can only be added by the concurrent collector");
515 if (!object_is_pinned (obj))
516 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");
517 else if (sgen_cement_lookup_or_register (obj))
520 remset.record_pointer (ptr);
522 sgen_pin_stats_register_global_remset (obj);
524 SGEN_LOG (8, "Adding global remset for %p", ptr);
525 binary_protocol_global_remset (ptr, obj, (gpointer)SGEN_LOAD_VTABLE (obj));
529 * sgen_drain_gray_stack:
531 * Scan objects in the gray stack until the stack is empty. This should be called
532 * frequently after each object is copied, to achieve better locality and cache
537 sgen_drain_gray_stack (ScanCopyContext ctx)
539 ScanObjectFunc scan_func = ctx.ops->scan_object;
540 GrayQueue *queue = ctx.queue;
542 if (ctx.ops->drain_gray_stack)
543 return ctx.ops->drain_gray_stack (queue);
548 GRAY_OBJECT_DEQUEUE (queue, &obj, &desc);
551 SGEN_LOG (9, "Precise gray object scan %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
552 scan_func (obj, desc, queue);
558 * Addresses in the pin queue are already sorted. This function finds
559 * the object header for each address and pins the object. The
560 * addresses must be inside the nursery section. The (start of the)
561 * address array is overwritten with the addresses of the actually
562 * pinned objects. Return the number of pinned objects.
565 pin_objects_from_nursery_pin_queue (gboolean do_scan_objects, ScanCopyContext ctx)
567 GCMemSection *section = nursery_section;
568 void **start = sgen_pinning_get_entry (section->pin_queue_first_entry);
569 void **end = sgen_pinning_get_entry (section->pin_queue_last_entry);
570 void *start_nursery = section->data;
571 void *end_nursery = section->next_data;
576 void *pinning_front = start_nursery;
578 void **definitely_pinned = start;
579 ScanObjectFunc scan_func = ctx.ops->scan_object;
580 SgenGrayQueue *queue = ctx.queue;
582 sgen_nursery_allocator_prepare_for_pinning ();
584 while (start < end) {
585 GCObject *obj_to_pin = NULL;
586 size_t obj_to_pin_size = 0;
591 SGEN_ASSERT (0, addr >= start_nursery && addr < end_nursery, "Potential pinning address out of range");
592 SGEN_ASSERT (0, addr >= last, "Pin queue not sorted");
599 SGEN_LOG (5, "Considering pinning addr %p", addr);
600 /* We've already processed everything up to pinning_front. */
601 if (addr < pinning_front) {
607 * Find the closest scan start <= addr. We might search backward in the
608 * scan_starts array because entries might be NULL. In the worst case we
609 * start at start_nursery.
611 idx = ((char*)addr - (char*)section->data) / SCAN_START_SIZE;
612 SGEN_ASSERT (0, idx < section->num_scan_start, "Scan start index out of range");
613 search_start = (void*)section->scan_starts [idx];
614 if (!search_start || search_start > addr) {
617 search_start = section->scan_starts [idx];
618 if (search_start && search_start <= addr)
621 if (!search_start || search_start > addr)
622 search_start = start_nursery;
626 * If the pinning front is closer than the scan start we found, start
627 * searching at the front.
629 if (search_start < pinning_front)
630 search_start = pinning_front;
633 * Now addr should be in an object a short distance from search_start.
635 * search_start must point to zeroed mem or point to an object.
638 size_t obj_size, canarified_obj_size;
641 if (!*(void**)search_start) {
642 search_start = (void*)ALIGN_UP ((mword)search_start + sizeof (gpointer));
643 /* The loop condition makes sure we don't overrun addr. */
647 canarified_obj_size = obj_size = ALIGN_UP (safe_object_get_size ((GCObject*)search_start));
650 * Filler arrays are marked by an invalid sync word. We don't
651 * consider them for pinning. They are not delimited by canaries,
654 if (!sgen_client_object_is_array_fill ((GCObject*)search_start)) {
655 CHECK_CANARY_FOR_OBJECT (search_start, TRUE);
656 CANARIFY_SIZE (canarified_obj_size);
658 if (addr >= search_start && (char*)addr < (char*)search_start + obj_size) {
659 /* This is the object we're looking for. */
660 obj_to_pin = (GCObject*)search_start;
661 obj_to_pin_size = canarified_obj_size;
666 /* Skip to the next object */
667 search_start = (void*)((char*)search_start + canarified_obj_size);
668 } while (search_start <= addr);
670 /* We've searched past the address we were looking for. */
672 pinning_front = search_start;
673 goto next_pin_queue_entry;
677 * We've found an object to pin. It might still be a dummy array, but we
678 * can advance the pinning front in any case.
680 pinning_front = (char*)obj_to_pin + obj_to_pin_size;
683 * If this is a dummy array marking the beginning of a nursery
684 * fragment, we don't pin it.
686 if (sgen_client_object_is_array_fill (obj_to_pin))
687 goto next_pin_queue_entry;
690 * Finally - pin the object!
692 desc = sgen_obj_get_descriptor_safe (obj_to_pin);
693 if (do_scan_objects) {
694 scan_func (obj_to_pin, desc, queue);
696 SGEN_LOG (4, "Pinned object %p, vtable %p (%s), count %d\n",
697 obj_to_pin, *(void**)obj_to_pin, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj_to_pin)), count);
698 binary_protocol_pin (obj_to_pin,
699 (gpointer)LOAD_VTABLE (obj_to_pin),
700 safe_object_get_size (obj_to_pin));
702 pin_object (obj_to_pin);
703 GRAY_OBJECT_ENQUEUE (queue, obj_to_pin, desc);
704 sgen_pin_stats_register_object (obj_to_pin, GENERATION_NURSERY);
705 definitely_pinned [count] = obj_to_pin;
708 if (concurrent_collection_in_progress)
709 sgen_pinning_register_pinned_in_nursery (obj_to_pin);
711 next_pin_queue_entry:
715 sgen_client_nursery_objects_pinned (definitely_pinned, count);
716 stat_pinned_objects += count;
721 pin_objects_in_nursery (gboolean do_scan_objects, ScanCopyContext ctx)
725 if (nursery_section->pin_queue_first_entry == nursery_section->pin_queue_last_entry)
728 reduced_to = pin_objects_from_nursery_pin_queue (do_scan_objects, ctx);
729 nursery_section->pin_queue_last_entry = nursery_section->pin_queue_first_entry + reduced_to;
733 * This function is only ever called (via `collector_pin_object()` in `sgen-copy-object.h`)
734 * when we can't promote an object because we're out of memory.
737 sgen_pin_object (GCObject *object, GrayQueue *queue)
739 SGEN_ASSERT (0, sgen_ptr_in_nursery (object), "We're only supposed to use this for pinning nursery objects when out of memory.");
742 * All pinned objects are assumed to have been staged, so we need to stage as well.
743 * Also, the count of staged objects shows that "late pinning" happened.
745 sgen_pin_stage_ptr (object);
747 SGEN_PIN_OBJECT (object);
748 binary_protocol_pin (object, (gpointer)LOAD_VTABLE (object), safe_object_get_size (object));
751 sgen_pin_stats_register_object (object, GENERATION_NURSERY);
753 GRAY_OBJECT_ENQUEUE (queue, object, sgen_obj_get_descriptor_safe (object));
756 /* Sort the addresses in array in increasing order.
757 * Done using a by-the book heap sort. Which has decent and stable performance, is pretty cache efficient.
760 sgen_sort_addresses (void **array, size_t size)
765 for (i = 1; i < size; ++i) {
768 size_t parent = (child - 1) / 2;
770 if (array [parent] >= array [child])
773 tmp = array [parent];
774 array [parent] = array [child];
781 for (i = size - 1; i > 0; --i) {
784 array [i] = array [0];
790 while (root * 2 + 1 <= end) {
791 size_t child = root * 2 + 1;
793 if (child < end && array [child] < array [child + 1])
795 if (array [root] >= array [child])
799 array [root] = array [child];
808 * Scan the memory between start and end and queue values which could be pointers
809 * to the area between start_nursery and end_nursery for later consideration.
810 * Typically used for thread stacks.
813 sgen_conservatively_pin_objects_from (void **start, void **end, void *start_nursery, void *end_nursery, int pin_type)
817 SGEN_ASSERT (0, ((mword)start & (SIZEOF_VOID_P - 1)) == 0, "Why are we scanning for references in unaligned memory ?");
819 #if defined(VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE) && !defined(_WIN64)
820 VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE (start, (char*)end - (char*)start);
823 while (start < end) {
825 * *start can point to the middle of an object
826 * note: should we handle pointing at the end of an object?
827 * pinning in C# code disallows pointing at the end of an object
828 * but there is some small chance that an optimizing C compiler
829 * may keep the only reference to an object by pointing
830 * at the end of it. We ignore this small chance for now.
831 * Pointers to the end of an object are indistinguishable
832 * from pointers to the start of the next object in memory
833 * so if we allow that we'd need to pin two objects...
834 * We queue the pointer in an array, the
835 * array will then be sorted and uniqued. This way
836 * we can coalesce several pinning pointers and it should
837 * be faster since we'd do a memory scan with increasing
838 * addresses. Note: we can align the address to the allocation
839 * alignment, so the unique process is more effective.
841 mword addr = (mword)*start;
842 addr &= ~(ALLOC_ALIGN - 1);
843 if (addr >= (mword)start_nursery && addr < (mword)end_nursery) {
844 SGEN_LOG (6, "Pinning address %p from %p", (void*)addr, start);
845 sgen_pin_stage_ptr ((void*)addr);
846 binary_protocol_pin_stage (start, (void*)addr);
847 sgen_pin_stats_register_address ((char*)addr, pin_type);
853 SGEN_LOG (7, "found %d potential pinned heap pointers", count);
857 * The first thing we do in a collection is to identify pinned objects.
858 * This function considers all the areas of memory that need to be
859 * conservatively scanned.
862 pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx)
866 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);
867 /* objects pinned from the API are inside these roots */
868 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_PINNED], void **, start_root, RootRecord *, root) {
869 SGEN_LOG (6, "Pinned roots %p-%p", start_root, root->end_root);
870 sgen_conservatively_pin_objects_from (start_root, (void**)root->end_root, start_nursery, end_nursery, PIN_TYPE_OTHER);
871 } SGEN_HASH_TABLE_FOREACH_END;
872 /* now deal with the thread stacks
873 * in the future we should be able to conservatively scan only:
874 * *) the cpu registers
875 * *) the unmanaged stack frames
876 * *) the _last_ managed stack frame
877 * *) pointers slots in managed frames
879 sgen_client_scan_thread_data (start_nursery, end_nursery, FALSE, ctx);
883 single_arg_user_copy_or_mark (GCObject **obj, void *gc_data)
885 ScanCopyContext *ctx = (ScanCopyContext *)gc_data;
886 ctx->ops->copy_or_mark_object (obj, ctx->queue);
890 * The memory area from start_root to end_root contains pointers to objects.
891 * Their position is precisely described by @desc (this means that the pointer
892 * can be either NULL or the pointer to the start of an object).
893 * This functions copies them to to_space updates them.
895 * This function is not thread-safe!
898 precisely_scan_objects_from (void** start_root, void** end_root, char* n_start, char *n_end, SgenDescriptor desc, ScanCopyContext ctx)
900 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
901 SgenGrayQueue *queue = ctx.queue;
903 switch (desc & ROOT_DESC_TYPE_MASK) {
904 case ROOT_DESC_BITMAP:
905 desc >>= ROOT_DESC_TYPE_SHIFT;
907 if ((desc & 1) && *start_root) {
908 copy_func ((GCObject**)start_root, queue);
909 SGEN_LOG (9, "Overwrote root at %p with %p", start_root, *start_root);
915 case ROOT_DESC_COMPLEX: {
916 gsize *bitmap_data = (gsize *)sgen_get_complex_descriptor_bitmap (desc);
917 gsize bwords = (*bitmap_data) - 1;
918 void **start_run = start_root;
920 while (bwords-- > 0) {
921 gsize bmap = *bitmap_data++;
922 void **objptr = start_run;
924 if ((bmap & 1) && *objptr) {
925 copy_func ((GCObject**)objptr, queue);
926 SGEN_LOG (9, "Overwrote root at %p with %p", objptr, *objptr);
931 start_run += GC_BITS_PER_WORD;
935 case ROOT_DESC_USER: {
936 SgenUserRootMarkFunc marker = sgen_get_user_descriptor_func (desc);
937 marker (start_root, single_arg_user_copy_or_mark, &ctx);
940 case ROOT_DESC_RUN_LEN:
941 g_assert_not_reached ();
943 g_assert_not_reached ();
948 reset_heap_boundaries (void)
950 lowest_heap_address = ~(mword)0;
951 highest_heap_address = 0;
955 sgen_update_heap_boundaries (mword low, mword high)
960 old = lowest_heap_address;
963 } while (SGEN_CAS_PTR ((gpointer*)&lowest_heap_address, (gpointer)low, (gpointer)old) != (gpointer)old);
966 old = highest_heap_address;
969 } while (SGEN_CAS_PTR ((gpointer*)&highest_heap_address, (gpointer)high, (gpointer)old) != (gpointer)old);
973 * Allocate and setup the data structures needed to be able to allocate objects
974 * in the nursery. The nursery is stored in nursery_section.
979 GCMemSection *section;
986 SGEN_LOG (2, "Allocating nursery size: %zu", (size_t)sgen_nursery_size);
987 /* later we will alloc a larger area for the nursery but only activate
988 * what we need. The rest will be used as expansion if we have too many pinned
989 * objects in the existing nursery.
991 /* FIXME: handle OOM */
992 section = (GCMemSection *)sgen_alloc_internal (INTERNAL_MEM_SECTION);
994 alloc_size = sgen_nursery_size;
996 /* If there isn't enough space even for the nursery we should simply abort. */
997 g_assert (sgen_memgov_try_alloc_space (alloc_size, SPACE_NURSERY));
999 data = (char *)major_collector.alloc_heap (alloc_size, alloc_size, DEFAULT_NURSERY_BITS);
1000 sgen_update_heap_boundaries ((mword)data, (mword)(data + sgen_nursery_size));
1001 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 ());
1002 section->data = section->next_data = data;
1003 section->size = alloc_size;
1004 section->end_data = data + sgen_nursery_size;
1005 scan_starts = (alloc_size + SCAN_START_SIZE - 1) / SCAN_START_SIZE;
1006 section->scan_starts = (char **)sgen_alloc_internal_dynamic (sizeof (char*) * scan_starts, INTERNAL_MEM_SCAN_STARTS, TRUE);
1007 section->num_scan_start = scan_starts;
1009 nursery_section = section;
1011 sgen_nursery_allocator_set_nursery_bounds (data, data + sgen_nursery_size);
1015 mono_gc_get_logfile (void)
1017 return gc_debug_file;
1021 scan_finalizer_entries (SgenPointerQueue *fin_queue, ScanCopyContext ctx)
1023 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
1024 SgenGrayQueue *queue = ctx.queue;
1027 for (i = 0; i < fin_queue->next_slot; ++i) {
1028 GCObject *obj = (GCObject *)fin_queue->data [i];
1031 SGEN_LOG (5, "Scan of fin ready object: %p (%s)\n", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
1032 copy_func ((GCObject**)&fin_queue->data [i], queue);
1037 generation_name (int generation)
1039 switch (generation) {
1040 case GENERATION_NURSERY: return "nursery";
1041 case GENERATION_OLD: return "old";
1042 default: g_assert_not_reached ();
1047 sgen_generation_name (int generation)
1049 return generation_name (generation);
1053 finish_gray_stack (int generation, ScanCopyContext ctx)
1057 int done_with_ephemerons, ephemeron_rounds = 0;
1058 char *start_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_start () : NULL;
1059 char *end_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_end () : (char*)-1;
1060 SgenGrayQueue *queue = ctx.queue;
1062 binary_protocol_finish_gray_stack_start (sgen_timestamp (), generation);
1064 * We copied all the reachable objects. Now it's the time to copy
1065 * the objects that were not referenced by the roots, but by the copied objects.
1066 * we built a stack of objects pointed to by gray_start: they are
1067 * additional roots and we may add more items as we go.
1068 * We loop until gray_start == gray_objects which means no more objects have
1069 * been added. Note this is iterative: no recursion is involved.
1070 * We need to walk the LO list as well in search of marked big objects
1071 * (use a flag since this is needed only on major collections). We need to loop
1072 * here as well, so keep a counter of marked LO (increasing it in copy_object).
1073 * To achieve better cache locality and cache usage, we drain the gray stack
1074 * frequently, after each object is copied, and just finish the work here.
1076 sgen_drain_gray_stack (ctx);
1078 SGEN_LOG (2, "%s generation done", generation_name (generation));
1081 Reset bridge data, we might have lingering data from a previous collection if this is a major
1082 collection trigged by minor overflow.
1084 We must reset the gathered bridges since their original block might be evacuated due to major
1085 fragmentation in the meanwhile and the bridge code should not have to deal with that.
1087 if (sgen_client_bridge_need_processing ())
1088 sgen_client_bridge_reset_data ();
1091 * Mark all strong toggleref objects. This must be done before we walk ephemerons or finalizers
1092 * to ensure they see the full set of live objects.
1094 sgen_client_mark_togglerefs (start_addr, end_addr, ctx);
1097 * Walk the ephemeron tables marking all values with reachable keys. This must be completely done
1098 * before processing finalizable objects and non-tracking weak links to avoid finalizing/clearing
1099 * objects that are in fact reachable.
1101 done_with_ephemerons = 0;
1103 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1104 sgen_drain_gray_stack (ctx);
1106 } while (!done_with_ephemerons);
1108 if (sgen_client_bridge_need_processing ()) {
1109 /*Make sure the gray stack is empty before we process bridge objects so we get liveness right*/
1110 sgen_drain_gray_stack (ctx);
1111 sgen_collect_bridge_objects (generation, ctx);
1112 if (generation == GENERATION_OLD)
1113 sgen_collect_bridge_objects (GENERATION_NURSERY, ctx);
1116 Do the first bridge step here, as the collector liveness state will become useless after that.
1118 An important optimization is to only proccess the possibly dead part of the object graph and skip
1119 over all live objects as we transitively know everything they point must be alive too.
1121 The above invariant is completely wrong if we let the gray queue be drained and mark/copy everything.
1123 This has the unfortunate side effect of making overflow collections perform the first step twice, but
1124 given we now have heuristics that perform major GC in anticipation of minor overflows this should not
1127 sgen_client_bridge_processing_stw_step ();
1131 Make sure we drain the gray stack before processing disappearing links and finalizers.
1132 If we don't make sure it is empty we might wrongly see a live object as dead.
1134 sgen_drain_gray_stack (ctx);
1137 We must clear weak links that don't track resurrection before processing object ready for
1138 finalization so they can be cleared before that.
1140 sgen_null_link_in_range (generation, ctx, FALSE);
1141 if (generation == GENERATION_OLD)
1142 sgen_null_link_in_range (GENERATION_NURSERY, ctx, FALSE);
1145 /* walk the finalization queue and move also the objects that need to be
1146 * finalized: use the finalized objects as new roots so the objects they depend
1147 * on are also not reclaimed. As with the roots above, only objects in the nursery
1148 * are marked/copied.
1150 sgen_finalize_in_range (generation, ctx);
1151 if (generation == GENERATION_OLD)
1152 sgen_finalize_in_range (GENERATION_NURSERY, ctx);
1153 /* drain the new stack that might have been created */
1154 SGEN_LOG (6, "Precise scan of gray area post fin");
1155 sgen_drain_gray_stack (ctx);
1158 * This must be done again after processing finalizable objects since CWL slots are cleared only after the key is finalized.
1160 done_with_ephemerons = 0;
1162 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1163 sgen_drain_gray_stack (ctx);
1165 } while (!done_with_ephemerons);
1167 sgen_client_clear_unreachable_ephemerons (ctx);
1170 * We clear togglerefs only after all possible chances of revival are done.
1171 * This is semantically more inline with what users expect and it allows for
1172 * user finalizers to correctly interact with TR objects.
1174 sgen_client_clear_togglerefs (start_addr, end_addr, ctx);
1177 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);
1180 * handle disappearing links
1181 * Note we do this after checking the finalization queue because if an object
1182 * survives (at least long enough to be finalized) we don't clear the link.
1183 * This also deals with a possible issue with the monitor reclamation: with the Boehm
1184 * GC a finalized object my lose the monitor because it is cleared before the finalizer is
1187 g_assert (sgen_gray_object_queue_is_empty (queue));
1189 sgen_null_link_in_range (generation, ctx, TRUE);
1190 if (generation == GENERATION_OLD)
1191 sgen_null_link_in_range (GENERATION_NURSERY, ctx, TRUE);
1192 if (sgen_gray_object_queue_is_empty (queue))
1194 sgen_drain_gray_stack (ctx);
1197 g_assert (sgen_gray_object_queue_is_empty (queue));
1199 sgen_gray_object_queue_trim_free_list (queue);
1200 binary_protocol_finish_gray_stack_end (sgen_timestamp (), generation);
1204 sgen_check_section_scan_starts (GCMemSection *section)
1207 for (i = 0; i < section->num_scan_start; ++i) {
1208 if (section->scan_starts [i]) {
1209 mword size = safe_object_get_size ((GCObject*) section->scan_starts [i]);
1210 SGEN_ASSERT (0, size >= SGEN_CLIENT_MINIMUM_OBJECT_SIZE && size <= MAX_SMALL_OBJ_SIZE, "Weird object size at scan starts.");
1216 check_scan_starts (void)
1218 if (!do_scan_starts_check)
1220 sgen_check_section_scan_starts (nursery_section);
1221 major_collector.check_scan_starts ();
1225 scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx)
1229 SGEN_HASH_TABLE_FOREACH (&roots_hash [root_type], void **, start_root, RootRecord *, root) {
1230 SGEN_LOG (6, "Precise root scan %p-%p (desc: %p)", start_root, root->end_root, (void*)root->root_desc);
1231 precisely_scan_objects_from (start_root, (void**)root->end_root, addr_start, addr_end, root->root_desc, ctx);
1232 } SGEN_HASH_TABLE_FOREACH_END;
1238 static gboolean inited = FALSE;
1243 mono_counters_register ("Collection max time", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME | MONO_COUNTER_MONOTONIC, &time_max);
1245 mono_counters_register ("Minor fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pre_collection_fragment_clear);
1246 mono_counters_register ("Minor pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pinning);
1247 mono_counters_register ("Minor scan remembered set", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_remsets);
1248 mono_counters_register ("Minor scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_pinned);
1249 mono_counters_register ("Minor scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_roots);
1250 mono_counters_register ("Minor fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_fragment_creation);
1252 mono_counters_register ("Major fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pre_collection_fragment_clear);
1253 mono_counters_register ("Major pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pinning);
1254 mono_counters_register ("Major scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_pinned);
1255 mono_counters_register ("Major scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_roots);
1256 mono_counters_register ("Major scan mod union", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_mod_union);
1257 mono_counters_register ("Major finish gray stack", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_finish_gray_stack);
1258 mono_counters_register ("Major free big objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_free_bigobjs);
1259 mono_counters_register ("Major LOS sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_los_sweep);
1260 mono_counters_register ("Major sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_sweep);
1261 mono_counters_register ("Major fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_fragment_creation);
1263 mono_counters_register ("Number of pinned objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_pinned_objects);
1265 #ifdef HEAVY_STATISTICS
1266 mono_counters_register ("WBarrier remember pointer", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_add_to_global_remset);
1267 mono_counters_register ("WBarrier arrayref copy", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_arrayref_copy);
1268 mono_counters_register ("WBarrier generic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store);
1269 mono_counters_register ("WBarrier generic atomic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store_atomic);
1270 mono_counters_register ("WBarrier set root", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_set_root);
1272 mono_counters_register ("# objects allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_alloced_degraded);
1273 mono_counters_register ("bytes allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_bytes_alloced_degraded);
1275 mono_counters_register ("# copy_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_nursery);
1276 mono_counters_register ("# objects copied (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_nursery);
1277 mono_counters_register ("# copy_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_major);
1278 mono_counters_register ("# objects copied (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_major);
1280 mono_counters_register ("# scan_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_nursery);
1281 mono_counters_register ("# scan_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_major);
1283 mono_counters_register ("Slots allocated in vain", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_slots_allocated_in_vain);
1285 mono_counters_register ("# nursery copy_object() failed from space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_from_space);
1286 mono_counters_register ("# nursery copy_object() failed forwarded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_forwarded);
1287 mono_counters_register ("# nursery copy_object() failed pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_pinned);
1288 mono_counters_register ("# nursery copy_object() failed to space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_to_space);
1290 sgen_nursery_allocator_init_heavy_stats ();
1298 reset_pinned_from_failed_allocation (void)
1300 bytes_pinned_from_failed_allocation = 0;
1304 sgen_set_pinned_from_failed_allocation (mword objsize)
1306 bytes_pinned_from_failed_allocation += objsize;
1310 sgen_collection_is_concurrent (void)
1312 switch (current_collection_generation) {
1313 case GENERATION_NURSERY:
1315 case GENERATION_OLD:
1316 return concurrent_collection_in_progress;
1318 g_error ("Invalid current generation %d", current_collection_generation);
1324 sgen_concurrent_collection_in_progress (void)
1326 return concurrent_collection_in_progress;
1330 SgenThreadPoolJob job;
1331 SgenObjectOperations *ops;
1335 job_remembered_set_scan (void *worker_data_untyped, SgenThreadPoolJob *job)
1337 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1338 ScanJob *job_data = (ScanJob*)job;
1339 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1340 remset.scan_remsets (ctx);
1344 SgenThreadPoolJob job;
1345 SgenObjectOperations *ops;
1349 } ScanFromRegisteredRootsJob;
1352 job_scan_from_registered_roots (void *worker_data_untyped, SgenThreadPoolJob *job)
1354 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1355 ScanFromRegisteredRootsJob *job_data = (ScanFromRegisteredRootsJob*)job;
1356 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1358 scan_from_registered_roots (job_data->heap_start, job_data->heap_end, job_data->root_type, ctx);
1362 SgenThreadPoolJob job;
1363 SgenObjectOperations *ops;
1366 } ScanThreadDataJob;
1369 job_scan_thread_data (void *worker_data_untyped, SgenThreadPoolJob *job)
1371 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1372 ScanThreadDataJob *job_data = (ScanThreadDataJob*)job;
1373 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1375 sgen_client_scan_thread_data (job_data->heap_start, job_data->heap_end, TRUE, ctx);
1379 SgenThreadPoolJob job;
1380 SgenObjectOperations *ops;
1381 SgenPointerQueue *queue;
1382 } ScanFinalizerEntriesJob;
1385 job_scan_finalizer_entries (void *worker_data_untyped, SgenThreadPoolJob *job)
1387 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1388 ScanFinalizerEntriesJob *job_data = (ScanFinalizerEntriesJob*)job;
1389 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1391 scan_finalizer_entries (job_data->queue, ctx);
1395 job_scan_major_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1397 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1398 ScanJob *job_data = (ScanJob*)job;
1399 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1401 g_assert (concurrent_collection_in_progress);
1402 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx);
1406 job_scan_los_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1408 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1409 ScanJob *job_data = (ScanJob*)job;
1410 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1412 g_assert (concurrent_collection_in_progress);
1413 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx);
1417 job_mod_union_preclean (void *worker_data_untyped, SgenThreadPoolJob *job)
1419 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1420 ScanJob *job_data = (ScanJob*)job;
1421 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (job_data->ops, sgen_workers_get_job_gray_queue (worker_data));
1423 g_assert (concurrent_collection_in_progress);
1425 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx);
1426 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx);
1428 sgen_scan_pin_queue_objects (ctx);
1432 init_gray_queue (gboolean use_workers)
1435 sgen_workers_init_distribute_gray_queue ();
1436 sgen_gray_object_queue_init (&gray_queue, NULL);
1440 enqueue_scan_from_roots_jobs (char *heap_start, char *heap_end, SgenObjectOperations *ops, gboolean enqueue)
1442 ScanFromRegisteredRootsJob *scrrj;
1443 ScanThreadDataJob *stdj;
1444 ScanFinalizerEntriesJob *sfej;
1446 /* registered roots, this includes static fields */
1448 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots normal", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1450 scrrj->heap_start = heap_start;
1451 scrrj->heap_end = heap_end;
1452 scrrj->root_type = ROOT_TYPE_NORMAL;
1453 sgen_workers_enqueue_job (&scrrj->job, enqueue);
1455 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots wbarrier", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1457 scrrj->heap_start = heap_start;
1458 scrrj->heap_end = heap_end;
1459 scrrj->root_type = ROOT_TYPE_WBARRIER;
1460 sgen_workers_enqueue_job (&scrrj->job, enqueue);
1464 stdj = (ScanThreadDataJob*)sgen_thread_pool_job_alloc ("scan thread data", job_scan_thread_data, sizeof (ScanThreadDataJob));
1466 stdj->heap_start = heap_start;
1467 stdj->heap_end = heap_end;
1468 sgen_workers_enqueue_job (&stdj->job, enqueue);
1470 /* Scan the list of objects ready for finalization. */
1472 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1473 sfej->queue = &fin_ready_queue;
1475 sgen_workers_enqueue_job (&sfej->job, enqueue);
1477 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan critical finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1478 sfej->queue = &critical_fin_queue;
1480 sgen_workers_enqueue_job (&sfej->job, enqueue);
1484 * Perform a nursery collection.
1486 * Return whether any objects were late-pinned due to being out of memory.
1489 collect_nursery (const char *reason, gboolean is_overflow, SgenGrayQueue *unpin_queue, gboolean finish_up_concurrent_mark)
1491 gboolean needs_major;
1492 size_t max_garbage_amount;
1494 mword fragment_total;
1496 SgenObjectOperations *object_ops = &sgen_minor_collector.serial_ops;
1497 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, &gray_queue);
1501 if (disable_minor_collections)
1504 TV_GETTIME (last_minor_collection_start_tv);
1505 atv = last_minor_collection_start_tv;
1507 binary_protocol_collection_begin (gc_stats.minor_gc_count, GENERATION_NURSERY);
1509 if (do_verify_nursery || do_dump_nursery_content)
1510 sgen_debug_verify_nursery (do_dump_nursery_content);
1512 current_collection_generation = GENERATION_NURSERY;
1514 SGEN_ASSERT (0, !sgen_collection_is_concurrent (), "Why is the nursery collection concurrent?");
1516 reset_pinned_from_failed_allocation ();
1518 check_scan_starts ();
1520 sgen_nursery_alloc_prepare_for_minor ();
1524 nursery_next = sgen_nursery_alloc_get_upper_alloc_bound ();
1525 /* FIXME: optimize later to use the higher address where an object can be present */
1526 nursery_next = MAX (nursery_next, sgen_get_nursery_end ());
1528 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 ()));
1529 max_garbage_amount = nursery_next - sgen_get_nursery_start ();
1530 g_assert (nursery_section->size >= max_garbage_amount);
1532 /* world must be stopped already */
1534 time_minor_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1536 sgen_client_pre_collection_checks ();
1538 nursery_section->next_data = nursery_next;
1540 major_collector.start_nursery_collection ();
1542 sgen_memgov_minor_collection_start ();
1544 init_gray_queue (FALSE);
1546 gc_stats.minor_gc_count ++;
1548 if (whole_heap_check_before_collection) {
1549 sgen_clear_nursery_fragments ();
1550 sgen_check_whole_heap (finish_up_concurrent_mark);
1553 sgen_process_fin_stage_entries ();
1555 /* pin from pinned handles */
1556 sgen_init_pinning ();
1557 sgen_client_binary_protocol_mark_start (GENERATION_NURSERY);
1558 pin_from_roots (sgen_get_nursery_start (), nursery_next, ctx);
1559 /* pin cemented objects */
1560 sgen_pin_cemented_objects ();
1561 /* identify pinned objects */
1562 sgen_optimize_pin_queue ();
1563 sgen_pinning_setup_section (nursery_section);
1565 pin_objects_in_nursery (FALSE, ctx);
1566 sgen_pinning_trim_queue_to_section (nursery_section);
1568 if (remset_consistency_checks)
1569 sgen_check_remset_consistency ();
1572 time_minor_pinning += TV_ELAPSED (btv, atv);
1573 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), (long long)TV_ELAPSED (btv, atv));
1574 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1576 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan remset", job_remembered_set_scan, sizeof (ScanJob));
1577 sj->ops = object_ops;
1578 sgen_workers_enqueue_job (&sj->job, FALSE);
1580 /* we don't have complete write barrier yet, so we scan all the old generation sections */
1582 time_minor_scan_remsets += TV_ELAPSED (atv, btv);
1583 SGEN_LOG (2, "Old generation scan: %lld usecs", (long long)TV_ELAPSED (atv, btv));
1585 sgen_pin_stats_report ();
1587 /* FIXME: Why do we do this at this specific, seemingly random, point? */
1588 sgen_client_collecting_minor (&fin_ready_queue, &critical_fin_queue);
1591 time_minor_scan_pinned += TV_ELAPSED (btv, atv);
1593 enqueue_scan_from_roots_jobs (sgen_get_nursery_start (), nursery_next, object_ops, FALSE);
1596 time_minor_scan_roots += TV_ELAPSED (atv, btv);
1598 finish_gray_stack (GENERATION_NURSERY, ctx);
1601 time_minor_finish_gray_stack += TV_ELAPSED (btv, atv);
1602 sgen_client_binary_protocol_mark_end (GENERATION_NURSERY);
1604 if (objects_pinned) {
1605 sgen_optimize_pin_queue ();
1606 sgen_pinning_setup_section (nursery_section);
1610 * This is the latest point at which we can do this check, because
1611 * sgen_build_nursery_fragments() unpins nursery objects again.
1613 if (remset_consistency_checks)
1614 sgen_check_remset_consistency ();
1616 /* walk the pin_queue, build up the fragment list of free memory, unmark
1617 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1620 sgen_client_binary_protocol_reclaim_start (GENERATION_NURSERY);
1621 fragment_total = sgen_build_nursery_fragments (nursery_section, unpin_queue);
1622 if (!fragment_total)
1625 /* Clear TLABs for all threads */
1626 sgen_clear_tlabs ();
1628 sgen_client_binary_protocol_reclaim_end (GENERATION_NURSERY);
1630 time_minor_fragment_creation += TV_ELAPSED (atv, btv);
1631 SGEN_LOG (2, "Fragment creation: %lld usecs, %lu bytes available", (long long)TV_ELAPSED (atv, btv), (unsigned long)fragment_total);
1633 if (remset_consistency_checks)
1634 sgen_check_major_refs ();
1636 major_collector.finish_nursery_collection ();
1638 TV_GETTIME (last_minor_collection_end_tv);
1639 gc_stats.minor_gc_time += TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv);
1641 sgen_debug_dump_heap ("minor", gc_stats.minor_gc_count - 1, NULL);
1643 /* prepare the pin queue for the next collection */
1644 sgen_finish_pinning ();
1645 if (sgen_have_pending_finalizers ()) {
1646 SGEN_LOG (4, "Finalizer-thread wakeup");
1647 sgen_client_finalize_notify ();
1649 sgen_pin_stats_reset ();
1650 /* clear cemented hash */
1651 sgen_cement_clear_below_threshold ();
1653 g_assert (sgen_gray_object_queue_is_empty (&gray_queue));
1655 remset.finish_minor_collection ();
1657 check_scan_starts ();
1659 binary_protocol_flush_buffers (FALSE);
1661 sgen_memgov_minor_collection_end (reason, is_overflow);
1663 /*objects are late pinned because of lack of memory, so a major is a good call*/
1664 needs_major = objects_pinned > 0;
1665 current_collection_generation = -1;
1668 binary_protocol_collection_end (gc_stats.minor_gc_count - 1, GENERATION_NURSERY, 0, 0);
1670 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1671 sgen_check_nursery_objects_pinned (unpin_queue != NULL);
1677 COPY_OR_MARK_FROM_ROOTS_SERIAL,
1678 COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT,
1679 COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT
1680 } CopyOrMarkFromRootsMode;
1683 major_copy_or_mark_from_roots (size_t *old_next_pin_slot, CopyOrMarkFromRootsMode mode, SgenObjectOperations *object_ops)
1688 /* FIXME: only use these values for the precise scan
1689 * note that to_space pointers should be excluded anyway...
1691 char *heap_start = NULL;
1692 char *heap_end = (char*)-1;
1693 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, WORKERS_DISTRIBUTE_GRAY_QUEUE);
1694 gboolean concurrent = mode != COPY_OR_MARK_FROM_ROOTS_SERIAL;
1696 SGEN_ASSERT (0, !!concurrent == !!concurrent_collection_in_progress, "We've been called with the wrong mode.");
1698 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1699 /*This cleans up unused fragments */
1700 sgen_nursery_allocator_prepare_for_pinning ();
1702 if (do_concurrent_checks)
1703 sgen_debug_check_nursery_is_clean ();
1705 /* The concurrent collector doesn't touch the nursery. */
1706 sgen_nursery_alloc_prepare_for_major ();
1709 init_gray_queue (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1713 /* Pinning depends on this */
1714 sgen_clear_nursery_fragments ();
1716 if (whole_heap_check_before_collection)
1717 sgen_check_whole_heap (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT);
1720 time_major_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1722 if (!sgen_collection_is_concurrent ())
1723 nursery_section->next_data = sgen_get_nursery_end ();
1724 /* we should also coalesce scanning from sections close to each other
1725 * and deal with pointers outside of the sections later.
1730 sgen_client_pre_collection_checks ();
1732 if (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1733 /* Remsets are not useful for a major collection */
1734 remset.clear_cards ();
1737 sgen_process_fin_stage_entries ();
1740 sgen_init_pinning ();
1741 SGEN_LOG (6, "Collecting pinned addresses");
1742 pin_from_roots ((void*)lowest_heap_address, (void*)highest_heap_address, ctx);
1743 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1744 /* Pin cemented objects that were forced */
1745 sgen_pin_cemented_objects ();
1747 sgen_optimize_pin_queue ();
1748 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1750 * Cemented objects that are in the pinned list will be marked. When
1751 * marking concurrently we won't mark mod-union cards for these objects.
1752 * Instead they will remain cemented until the next major collection,
1753 * when we will recheck if they are still pinned in the roots.
1755 sgen_cement_force_pinned ();
1758 sgen_client_collecting_major_1 ();
1761 * pin_queue now contains all candidate pointers, sorted and
1762 * uniqued. We must do two passes now to figure out which
1763 * objects are pinned.
1765 * The first is to find within the pin_queue the area for each
1766 * section. This requires that the pin_queue be sorted. We
1767 * also process the LOS objects and pinned chunks here.
1769 * The second, destructive, pass is to reduce the section
1770 * areas to pointers to the actually pinned objects.
1772 SGEN_LOG (6, "Pinning from sections");
1773 /* first pass for the sections */
1774 sgen_find_section_pin_queue_start_end (nursery_section);
1775 /* identify possible pointers to the insize of large objects */
1776 SGEN_LOG (6, "Pinning from large objects");
1777 for (bigobj = los_object_list; bigobj; bigobj = bigobj->next) {
1779 if (sgen_find_optimized_pin_queue_area ((char*)bigobj->data, (char*)bigobj->data + sgen_los_object_size (bigobj), &dummy, &dummy)) {
1780 binary_protocol_pin (bigobj->data, (gpointer)LOAD_VTABLE (bigobj->data), safe_object_get_size (bigobj->data));
1782 if (sgen_los_object_is_pinned (bigobj->data)) {
1783 SGEN_ASSERT (0, mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, "LOS objects can only be pinned here after concurrent marking.");
1786 sgen_los_pin_object (bigobj->data);
1787 if (SGEN_OBJECT_HAS_REFERENCES (bigobj->data))
1788 GRAY_OBJECT_ENQUEUE (WORKERS_DISTRIBUTE_GRAY_QUEUE, bigobj->data, sgen_obj_get_descriptor ((GCObject*)bigobj->data));
1789 sgen_pin_stats_register_object (bigobj->data, GENERATION_OLD);
1790 SGEN_LOG (6, "Marked large object %p (%s) size: %lu from roots", bigobj->data,
1791 sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (bigobj->data)),
1792 (unsigned long)sgen_los_object_size (bigobj));
1794 sgen_client_pinned_los_object (bigobj->data);
1798 pin_objects_in_nursery (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT, ctx);
1799 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1800 sgen_check_nursery_objects_pinned (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1802 major_collector.pin_objects (WORKERS_DISTRIBUTE_GRAY_QUEUE);
1803 if (old_next_pin_slot)
1804 *old_next_pin_slot = sgen_get_pinned_count ();
1807 * We don't actually pin when starting a concurrent collection, so the remset
1808 * consistency check won't work.
1810 if (remset_consistency_checks && mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT)
1811 sgen_check_remset_consistency ();
1814 time_major_pinning += TV_ELAPSED (atv, btv);
1815 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), (long long)TV_ELAPSED (atv, btv));
1816 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1818 major_collector.init_to_space ();
1820 SGEN_ASSERT (0, sgen_workers_all_done (), "Why are the workers not done when we start or finish a major collection?");
1821 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1822 if (sgen_workers_have_idle_work ()) {
1824 * We force the finish of the worker with the new object ops context
1825 * which can also do copying. We need to have finished pinning.
1827 sgen_workers_start_all_workers (object_ops, NULL);
1828 sgen_workers_join ();
1832 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1833 main_gc_thread = mono_native_thread_self ();
1836 sgen_client_collecting_major_2 ();
1839 time_major_scan_pinned += TV_ELAPSED (btv, atv);
1841 sgen_client_collecting_major_3 (&fin_ready_queue, &critical_fin_queue);
1843 enqueue_scan_from_roots_jobs (heap_start, heap_end, object_ops, FALSE);
1846 time_major_scan_roots += TV_ELAPSED (atv, btv);
1849 * We start the concurrent worker after pinning and after we scanned the roots
1850 * in order to make sure that the worker does not finish before handling all
1853 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1854 if (precleaning_enabled) {
1856 /* Mod union preclean job */
1857 sj = (ScanJob*)sgen_thread_pool_job_alloc ("preclean mod union cardtable", job_mod_union_preclean, sizeof (ScanJob));
1858 sj->ops = object_ops;
1859 sgen_workers_start_all_workers (object_ops, &sj->job);
1861 sgen_workers_start_all_workers (object_ops, NULL);
1863 gray_queue_enable_redirect (WORKERS_DISTRIBUTE_GRAY_QUEUE);
1866 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1869 /* Mod union card table */
1870 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan mod union cardtable", job_scan_major_mod_union_card_table, sizeof (ScanJob));
1871 sj->ops = object_ops;
1872 sgen_workers_enqueue_job (&sj->job, FALSE);
1874 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan LOS mod union cardtable", job_scan_los_mod_union_card_table, sizeof (ScanJob));
1875 sj->ops = object_ops;
1876 sgen_workers_enqueue_job (&sj->job, FALSE);
1879 time_major_scan_mod_union += TV_ELAPSED (btv, atv);
1882 sgen_pin_stats_report ();
1886 major_finish_copy_or_mark (CopyOrMarkFromRootsMode mode)
1888 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1889 sgen_finish_pinning ();
1891 sgen_pin_stats_reset ();
1893 if (do_concurrent_checks)
1894 sgen_debug_check_nursery_is_clean ();
1899 major_start_collection (const char *reason, gboolean concurrent, size_t *old_next_pin_slot)
1901 SgenObjectOperations *object_ops;
1903 binary_protocol_collection_begin (gc_stats.major_gc_count, GENERATION_OLD);
1905 current_collection_generation = GENERATION_OLD;
1907 g_assert (sgen_section_gray_queue_is_empty (sgen_workers_get_distribute_section_gray_queue ()));
1910 sgen_cement_reset ();
1913 g_assert (major_collector.is_concurrent);
1914 concurrent_collection_in_progress = TRUE;
1916 object_ops = &major_collector.major_ops_concurrent_start;
1918 object_ops = &major_collector.major_ops_serial;
1921 reset_pinned_from_failed_allocation ();
1923 sgen_memgov_major_collection_start (concurrent, reason);
1925 //count_ref_nonref_objs ();
1926 //consistency_check ();
1928 check_scan_starts ();
1931 SGEN_LOG (1, "Start major collection %d", gc_stats.major_gc_count);
1932 gc_stats.major_gc_count ++;
1934 if (major_collector.start_major_collection)
1935 major_collector.start_major_collection ();
1937 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);
1938 major_finish_copy_or_mark (concurrent ? COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT : COPY_OR_MARK_FROM_ROOTS_SERIAL);
1942 major_finish_collection (const char *reason, gboolean is_overflow, size_t old_next_pin_slot, gboolean forced)
1944 ScannedObjectCounts counts;
1945 SgenObjectOperations *object_ops;
1946 mword fragment_total;
1952 if (concurrent_collection_in_progress) {
1953 object_ops = &major_collector.major_ops_concurrent_finish;
1955 major_copy_or_mark_from_roots (NULL, COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, object_ops);
1957 major_finish_copy_or_mark (COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT);
1959 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1960 main_gc_thread = NULL;
1963 object_ops = &major_collector.major_ops_serial;
1966 g_assert (sgen_section_gray_queue_is_empty (sgen_workers_get_distribute_section_gray_queue ()));
1968 /* all the objects in the heap */
1969 finish_gray_stack (GENERATION_OLD, CONTEXT_FROM_OBJECT_OPERATIONS (object_ops, &gray_queue));
1971 time_major_finish_gray_stack += TV_ELAPSED (btv, atv);
1973 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after joining");
1975 if (objects_pinned) {
1976 g_assert (!concurrent_collection_in_progress);
1979 * This is slow, but we just OOM'd.
1981 * See comment at `sgen_pin_queue_clear_discarded_entries` for how the pin
1982 * queue is laid out at this point.
1984 sgen_pin_queue_clear_discarded_entries (nursery_section, old_next_pin_slot);
1986 * We need to reestablish all pinned nursery objects in the pin queue
1987 * because they're needed for fragment creation. Unpinning happens by
1988 * walking the whole queue, so it's not necessary to reestablish where major
1989 * heap block pins are - all we care is that they're still in there
1992 sgen_optimize_pin_queue ();
1993 sgen_find_section_pin_queue_start_end (nursery_section);
1997 reset_heap_boundaries ();
1998 sgen_update_heap_boundaries ((mword)sgen_get_nursery_start (), (mword)sgen_get_nursery_end ());
2000 /* walk the pin_queue, build up the fragment list of free memory, unmark
2001 * pinned objects as we go, memzero() the empty fragments so they are ready for the
2004 fragment_total = sgen_build_nursery_fragments (nursery_section, NULL);
2005 if (!fragment_total)
2007 SGEN_LOG (4, "Free space in nursery after major %ld", (long)fragment_total);
2009 if (do_concurrent_checks && concurrent_collection_in_progress)
2010 sgen_debug_check_nursery_is_clean ();
2012 /* prepare the pin queue for the next collection */
2013 sgen_finish_pinning ();
2015 /* Clear TLABs for all threads */
2016 sgen_clear_tlabs ();
2018 sgen_pin_stats_reset ();
2020 sgen_cement_clear_below_threshold ();
2022 if (check_mark_bits_after_major_collection)
2023 sgen_check_heap_marked (concurrent_collection_in_progress);
2026 time_major_fragment_creation += TV_ELAPSED (atv, btv);
2028 binary_protocol_sweep_begin (GENERATION_OLD, !major_collector.sweeps_lazily);
2029 sgen_memgov_major_pre_sweep ();
2032 time_major_free_bigobjs += TV_ELAPSED (btv, atv);
2037 time_major_los_sweep += TV_ELAPSED (atv, btv);
2039 major_collector.sweep ();
2041 binary_protocol_sweep_end (GENERATION_OLD, !major_collector.sweeps_lazily);
2044 time_major_sweep += TV_ELAPSED (btv, atv);
2046 sgen_debug_dump_heap ("major", gc_stats.major_gc_count - 1, reason);
2048 if (sgen_have_pending_finalizers ()) {
2049 SGEN_LOG (4, "Finalizer-thread wakeup");
2050 sgen_client_finalize_notify ();
2053 g_assert (sgen_gray_object_queue_is_empty (&gray_queue));
2055 sgen_memgov_major_collection_end (forced, concurrent_collection_in_progress, reason, is_overflow);
2056 current_collection_generation = -1;
2058 memset (&counts, 0, sizeof (ScannedObjectCounts));
2059 major_collector.finish_major_collection (&counts);
2061 g_assert (sgen_section_gray_queue_is_empty (sgen_workers_get_distribute_section_gray_queue ()));
2063 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after major collection has finished");
2064 if (concurrent_collection_in_progress)
2065 concurrent_collection_in_progress = FALSE;
2067 check_scan_starts ();
2069 binary_protocol_flush_buffers (FALSE);
2071 //consistency_check ();
2073 binary_protocol_collection_end (gc_stats.major_gc_count - 1, GENERATION_OLD, counts.num_scanned_objects, counts.num_unique_scanned_objects);
2077 major_do_collection (const char *reason, gboolean is_overflow, gboolean forced)
2079 TV_DECLARE (time_start);
2080 TV_DECLARE (time_end);
2081 size_t old_next_pin_slot;
2083 if (disable_major_collections)
2086 if (major_collector.get_and_reset_num_major_objects_marked) {
2087 long long num_marked = major_collector.get_and_reset_num_major_objects_marked ();
2088 g_assert (!num_marked);
2091 /* world must be stopped already */
2092 TV_GETTIME (time_start);
2094 major_start_collection (reason, FALSE, &old_next_pin_slot);
2095 major_finish_collection (reason, is_overflow, old_next_pin_slot, forced);
2097 TV_GETTIME (time_end);
2098 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2100 /* FIXME: also report this to the user, preferably in gc-end. */
2101 if (major_collector.get_and_reset_num_major_objects_marked)
2102 major_collector.get_and_reset_num_major_objects_marked ();
2104 return bytes_pinned_from_failed_allocation > 0;
2108 major_start_concurrent_collection (const char *reason)
2110 TV_DECLARE (time_start);
2111 TV_DECLARE (time_end);
2112 long long num_objects_marked;
2114 if (disable_major_collections)
2117 TV_GETTIME (time_start);
2118 SGEN_TV_GETTIME (time_major_conc_collection_start);
2120 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2121 g_assert (num_objects_marked == 0);
2123 binary_protocol_concurrent_start ();
2125 // FIXME: store reason and pass it when finishing
2126 major_start_collection (reason, TRUE, NULL);
2128 gray_queue_redirect (&gray_queue);
2130 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2132 TV_GETTIME (time_end);
2133 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2135 current_collection_generation = -1;
2139 * Returns whether the major collection has finished.
2142 major_should_finish_concurrent_collection (void)
2144 SGEN_ASSERT (0, sgen_gray_object_queue_is_empty (&gray_queue), "Why is the gray queue not empty before we have started doing anything?");
2145 return sgen_workers_all_done ();
2149 major_update_concurrent_collection (void)
2151 TV_DECLARE (total_start);
2152 TV_DECLARE (total_end);
2154 TV_GETTIME (total_start);
2156 binary_protocol_concurrent_update ();
2158 major_collector.update_cardtable_mod_union ();
2159 sgen_los_update_cardtable_mod_union ();
2161 TV_GETTIME (total_end);
2162 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end);
2166 major_finish_concurrent_collection (gboolean forced)
2168 TV_DECLARE (total_start);
2169 TV_DECLARE (total_end);
2171 TV_GETTIME (total_start);
2173 binary_protocol_concurrent_finish ();
2176 * We need to stop all workers since we're updating the cardtable below.
2177 * The workers will be resumed with a finishing pause context to avoid
2178 * additional cardtable and object scanning.
2180 sgen_workers_stop_all_workers ();
2182 SGEN_TV_GETTIME (time_major_conc_collection_end);
2183 gc_stats.major_gc_time_concurrent += SGEN_TV_ELAPSED (time_major_conc_collection_start, time_major_conc_collection_end);
2185 major_collector.update_cardtable_mod_union ();
2186 sgen_los_update_cardtable_mod_union ();
2188 if (mod_union_consistency_check)
2189 sgen_check_mod_union_consistency ();
2191 current_collection_generation = GENERATION_OLD;
2192 sgen_cement_reset ();
2193 major_finish_collection ("finishing", FALSE, -1, forced);
2195 if (whole_heap_check_before_collection)
2196 sgen_check_whole_heap (FALSE);
2198 TV_GETTIME (total_end);
2199 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end) - TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv);
2201 current_collection_generation = -1;
2205 * Ensure an allocation request for @size will succeed by freeing enough memory.
2207 * LOCKING: The GC lock MUST be held.
2210 sgen_ensure_free_space (size_t size, int generation)
2212 int generation_to_collect = -1;
2213 const char *reason = NULL;
2215 if (generation == GENERATION_OLD) {
2216 if (sgen_need_major_collection (size)) {
2217 reason = "LOS overflow";
2218 generation_to_collect = GENERATION_OLD;
2221 if (degraded_mode) {
2222 if (sgen_need_major_collection (size)) {
2223 reason = "Degraded mode overflow";
2224 generation_to_collect = GENERATION_OLD;
2226 } else if (sgen_need_major_collection (size)) {
2227 reason = concurrent_collection_in_progress ? "Forced finish concurrent collection" : "Minor allowance";
2228 generation_to_collect = GENERATION_OLD;
2230 generation_to_collect = GENERATION_NURSERY;
2231 reason = "Nursery full";
2235 if (generation_to_collect == -1) {
2236 if (concurrent_collection_in_progress && sgen_workers_all_done ()) {
2237 generation_to_collect = GENERATION_OLD;
2238 reason = "Finish concurrent collection";
2242 if (generation_to_collect == -1)
2244 sgen_perform_collection (size, generation_to_collect, reason, FALSE, TRUE);
2248 * LOCKING: Assumes the GC lock is held.
2251 sgen_perform_collection (size_t requested_size, int generation_to_collect, const char *reason, gboolean wait_to_finish, gboolean stw)
2253 TV_DECLARE (gc_total_start);
2254 TV_DECLARE (gc_total_end);
2255 int overflow_generation_to_collect = -1;
2256 int oldest_generation_collected = generation_to_collect;
2257 const char *overflow_reason = NULL;
2258 gboolean finish_concurrent = concurrent_collection_in_progress && (major_should_finish_concurrent_collection () || generation_to_collect == GENERATION_OLD);
2260 binary_protocol_collection_requested (generation_to_collect, requested_size, wait_to_finish ? 1 : 0);
2262 SGEN_ASSERT (0, generation_to_collect == GENERATION_NURSERY || generation_to_collect == GENERATION_OLD, "What generation is this?");
2265 sgen_stop_world (generation_to_collect);
2267 SGEN_ASSERT (0, sgen_is_world_stopped (), "We can only collect if the world is stopped");
2270 TV_GETTIME (gc_total_start);
2272 // FIXME: extract overflow reason
2273 // FIXME: minor overflow for concurrent case
2274 if (generation_to_collect == GENERATION_NURSERY && !finish_concurrent) {
2275 if (concurrent_collection_in_progress)
2276 major_update_concurrent_collection ();
2278 if (collect_nursery (reason, FALSE, NULL, FALSE) && !concurrent_collection_in_progress) {
2279 overflow_generation_to_collect = GENERATION_OLD;
2280 overflow_reason = "Minor overflow";
2282 } else if (finish_concurrent) {
2283 major_finish_concurrent_collection (wait_to_finish);
2284 oldest_generation_collected = GENERATION_OLD;
2286 SGEN_ASSERT (0, generation_to_collect == GENERATION_OLD, "We should have handled nursery collections above");
2287 if (major_collector.is_concurrent && !wait_to_finish) {
2288 collect_nursery ("Concurrent start", FALSE, NULL, FALSE);
2289 major_start_concurrent_collection (reason);
2290 oldest_generation_collected = GENERATION_NURSERY;
2291 } else if (major_do_collection (reason, FALSE, wait_to_finish)) {
2292 overflow_generation_to_collect = GENERATION_NURSERY;
2293 overflow_reason = "Excessive pinning";
2297 if (overflow_generation_to_collect != -1) {
2298 SGEN_ASSERT (0, !concurrent_collection_in_progress, "We don't yet support overflow collections with the concurrent collector");
2301 * We need to do an overflow collection, either because we ran out of memory
2302 * or the nursery is fully pinned.
2305 if (overflow_generation_to_collect == GENERATION_NURSERY)
2306 collect_nursery (overflow_reason, TRUE, NULL, FALSE);
2308 major_do_collection (overflow_reason, TRUE, wait_to_finish);
2310 oldest_generation_collected = MAX (oldest_generation_collected, overflow_generation_to_collect);
2313 SGEN_LOG (2, "Heap size: %lu, LOS size: %lu", (unsigned long)sgen_gc_get_total_heap_allocation (), (unsigned long)los_memory_usage);
2315 /* this also sets the proper pointers for the next allocation */
2316 if (generation_to_collect == GENERATION_NURSERY && !sgen_can_alloc_size (requested_size)) {
2317 /* TypeBuilder and MonoMethod are killing mcs with fragmentation */
2318 SGEN_LOG (1, "nursery collection didn't find enough room for %zd alloc (%zd pinned)", requested_size, sgen_get_pinned_count ());
2319 sgen_dump_pin_queue ();
2323 g_assert (sgen_gray_object_queue_is_empty (&gray_queue));
2325 TV_GETTIME (gc_total_end);
2326 time_max = MAX (time_max, TV_ELAPSED (gc_total_start, gc_total_end));
2329 sgen_restart_world (oldest_generation_collected);
2333 * ######################################################################
2334 * ######## Memory allocation from the OS
2335 * ######################################################################
2336 * This section of code deals with getting memory from the OS and
2337 * allocating memory for GC-internal data structures.
2338 * Internal memory can be handled with a freelist for small objects.
2344 G_GNUC_UNUSED static void
2345 report_internal_mem_usage (void)
2347 printf ("Internal memory usage:\n");
2348 sgen_report_internal_mem_usage ();
2349 printf ("Pinned memory usage:\n");
2350 major_collector.report_pinned_memory_usage ();
2354 * ######################################################################
2355 * ######## Finalization support
2356 * ######################################################################
2360 * If the object has been forwarded it means it's still referenced from a root.
2361 * If it is pinned it's still alive as well.
2362 * A LOS object is only alive if we have pinned it.
2363 * Return TRUE if @obj is ready to be finalized.
2365 static inline gboolean
2366 sgen_is_object_alive (GCObject *object)
2368 if (ptr_in_nursery (object))
2369 return sgen_nursery_is_object_alive (object);
2371 return sgen_major_is_object_alive (object);
2375 * This function returns true if @object is either alive and belongs to the
2376 * current collection - major collections are full heap, so old gen objects
2377 * are never alive during a minor collection.
2380 sgen_is_object_alive_and_on_current_collection (GCObject *object)
2382 if (ptr_in_nursery (object))
2383 return sgen_nursery_is_object_alive (object);
2385 if (current_collection_generation == GENERATION_NURSERY)
2388 return sgen_major_is_object_alive (object);
2393 sgen_gc_is_object_ready_for_finalization (GCObject *object)
2395 return !sgen_is_object_alive (object);
2399 sgen_queue_finalization_entry (GCObject *obj)
2401 gboolean critical = sgen_client_object_has_critical_finalizer (obj);
2403 sgen_pointer_queue_add (critical ? &critical_fin_queue : &fin_ready_queue, obj);
2405 sgen_client_object_queued_for_finalization (obj);
2409 sgen_object_is_live (GCObject *obj)
2411 return sgen_is_object_alive_and_on_current_collection (obj);
2415 * `System.GC.WaitForPendingFinalizers` first checks `sgen_have_pending_finalizers()` to
2416 * determine whether it can exit quickly. The latter must therefore only return FALSE if
2417 * all finalizers have really finished running.
2419 * `sgen_gc_invoke_finalizers()` first dequeues a finalizable object, and then finalizes it.
2420 * This means that just checking whether the queues are empty leaves the possibility that an
2421 * object might have been dequeued but not yet finalized. That's why we need the additional
2422 * flag `pending_unqueued_finalizer`.
2425 static volatile gboolean pending_unqueued_finalizer = FALSE;
2428 sgen_gc_invoke_finalizers (void)
2432 g_assert (!pending_unqueued_finalizer);
2434 /* FIXME: batch to reduce lock contention */
2435 while (sgen_have_pending_finalizers ()) {
2441 * We need to set `pending_unqueued_finalizer` before dequeing the
2442 * finalizable object.
2444 if (!sgen_pointer_queue_is_empty (&fin_ready_queue)) {
2445 pending_unqueued_finalizer = TRUE;
2446 mono_memory_write_barrier ();
2447 obj = (GCObject *)sgen_pointer_queue_pop (&fin_ready_queue);
2448 } else if (!sgen_pointer_queue_is_empty (&critical_fin_queue)) {
2449 pending_unqueued_finalizer = TRUE;
2450 mono_memory_write_barrier ();
2451 obj = (GCObject *)sgen_pointer_queue_pop (&critical_fin_queue);
2457 SGEN_LOG (7, "Finalizing object %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
2465 /* the object is on the stack so it is pinned */
2466 /*g_print ("Calling finalizer for object: %p (%s)\n", obj, sgen_client_object_safe_name (obj));*/
2467 sgen_client_run_finalize (obj);
2470 if (pending_unqueued_finalizer) {
2471 mono_memory_write_barrier ();
2472 pending_unqueued_finalizer = FALSE;
2479 sgen_have_pending_finalizers (void)
2481 return pending_unqueued_finalizer || !sgen_pointer_queue_is_empty (&fin_ready_queue) || !sgen_pointer_queue_is_empty (&critical_fin_queue);
2485 * ######################################################################
2486 * ######## registered roots support
2487 * ######################################################################
2491 * We do not coalesce roots.
2494 sgen_register_root (char *start, size_t size, SgenDescriptor descr, int root_type, int source, const char *msg)
2496 RootRecord new_root;
2499 for (i = 0; i < ROOT_TYPE_NUM; ++i) {
2500 RootRecord *root = (RootRecord *)sgen_hash_table_lookup (&roots_hash [i], start);
2501 /* we allow changing the size and the descriptor (for thread statics etc) */
2503 size_t old_size = root->end_root - start;
2504 root->end_root = start + size;
2505 SGEN_ASSERT (0, !!root->root_desc == !!descr, "Can't change whether a root is precise or conservative.");
2506 SGEN_ASSERT (0, root->source == source, "Can't change a root's source identifier.");
2507 SGEN_ASSERT (0, !!root->msg == !!msg, "Can't change a root's message.");
2508 root->root_desc = descr;
2510 roots_size -= old_size;
2516 new_root.end_root = start + size;
2517 new_root.root_desc = descr;
2518 new_root.source = source;
2521 sgen_hash_table_replace (&roots_hash [root_type], start, &new_root, NULL);
2524 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);
2531 sgen_deregister_root (char* addr)
2537 for (root_type = 0; root_type < ROOT_TYPE_NUM; ++root_type) {
2538 if (sgen_hash_table_remove (&roots_hash [root_type], addr, &root))
2539 roots_size -= (root.end_root - addr);
2545 * ######################################################################
2546 * ######## Thread handling (stop/start code)
2547 * ######################################################################
2551 sgen_get_current_collection_generation (void)
2553 return current_collection_generation;
2557 sgen_thread_register (SgenThreadInfo* info, void *stack_bottom_fallback)
2559 #ifndef HAVE_KW_THREAD
2560 info->tlab_start = info->tlab_next = info->tlab_temp_end = info->tlab_real_end = NULL;
2563 sgen_init_tlab_info (info);
2565 sgen_client_thread_register (info, stack_bottom_fallback);
2571 sgen_thread_unregister (SgenThreadInfo *p)
2573 sgen_client_thread_unregister (p);
2577 * ######################################################################
2578 * ######## Write barriers
2579 * ######################################################################
2583 * Note: the write barriers first do the needed GC work and then do the actual store:
2584 * this way the value is visible to the conservative GC scan after the write barrier
2585 * itself. If a GC interrupts the barrier in the middle, value will be kept alive by
2586 * the conservative scan, otherwise by the remembered set scan.
2590 mono_gc_wbarrier_arrayref_copy (gpointer dest_ptr, gpointer src_ptr, int count)
2592 HEAVY_STAT (++stat_wbarrier_arrayref_copy);
2593 /*This check can be done without taking a lock since dest_ptr array is pinned*/
2594 if (ptr_in_nursery (dest_ptr) || count <= 0) {
2595 mono_gc_memmove_aligned (dest_ptr, src_ptr, count * sizeof (gpointer));
2599 #ifdef SGEN_HEAVY_BINARY_PROTOCOL
2600 if (binary_protocol_is_heavy_enabled ()) {
2602 for (i = 0; i < count; ++i) {
2603 gpointer dest = (gpointer*)dest_ptr + i;
2604 gpointer obj = *((gpointer*)src_ptr + i);
2606 binary_protocol_wbarrier (dest, obj, (gpointer)LOAD_VTABLE (obj));
2611 remset.wbarrier_arrayref_copy (dest_ptr, src_ptr, count);
2615 mono_gc_wbarrier_generic_nostore (gpointer ptr)
2619 HEAVY_STAT (++stat_wbarrier_generic_store);
2621 sgen_client_wbarrier_generic_nostore_check (ptr);
2623 obj = *(gpointer*)ptr;
2625 binary_protocol_wbarrier (ptr, obj, (gpointer)LOAD_VTABLE (obj));
2628 * We need to record old->old pointer locations for the
2629 * concurrent collector.
2631 if (!ptr_in_nursery (obj) && !concurrent_collection_in_progress) {
2632 SGEN_LOG (8, "Skipping remset at %p", ptr);
2636 SGEN_LOG (8, "Adding remset at %p", ptr);
2638 remset.wbarrier_generic_nostore (ptr);
2642 mono_gc_wbarrier_generic_store (gpointer ptr, GCObject* value)
2644 SGEN_LOG (8, "Wbarrier store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2645 SGEN_UPDATE_REFERENCE_ALLOW_NULL (ptr, value);
2646 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2647 mono_gc_wbarrier_generic_nostore (ptr);
2648 sgen_dummy_use (value);
2651 /* Same as mono_gc_wbarrier_generic_store () but performs the store
2652 * as an atomic operation with release semantics.
2655 mono_gc_wbarrier_generic_store_atomic (gpointer ptr, GCObject *value)
2657 HEAVY_STAT (++stat_wbarrier_generic_store_atomic);
2659 SGEN_LOG (8, "Wbarrier atomic store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2661 InterlockedWritePointer ((volatile gpointer *)ptr, value);
2663 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2664 mono_gc_wbarrier_generic_nostore (ptr);
2666 sgen_dummy_use (value);
2670 sgen_wbarrier_value_copy_bitmap (gpointer _dest, gpointer _src, int size, unsigned bitmap)
2672 GCObject **dest = (GCObject **)_dest;
2673 GCObject **src = (GCObject **)_src;
2677 mono_gc_wbarrier_generic_store (dest, *src);
2682 size -= SIZEOF_VOID_P;
2688 * ######################################################################
2689 * ######## Other mono public interface functions.
2690 * ######################################################################
2694 sgen_gc_collect (int generation)
2699 sgen_perform_collection (0, generation, "user request", TRUE, TRUE);
2704 sgen_gc_collection_count (int generation)
2706 if (generation == 0)
2707 return gc_stats.minor_gc_count;
2708 return gc_stats.major_gc_count;
2712 sgen_gc_get_used_size (void)
2716 tot = los_memory_usage;
2717 tot += nursery_section->next_data - nursery_section->data;
2718 tot += major_collector.get_used_size ();
2719 /* FIXME: account for pinned objects */
2725 sgen_env_var_error (const char *env_var, const char *fallback, const char *description_format, ...)
2729 va_start (ap, description_format);
2731 fprintf (stderr, "Warning: In environment variable `%s': ", env_var);
2732 vfprintf (stderr, description_format, ap);
2734 fprintf (stderr, " - %s", fallback);
2735 fprintf (stderr, "\n");
2741 parse_double_in_interval (const char *env_var, const char *opt_name, const char *opt, double min, double max, double *result)
2744 double val = strtod (opt, &endptr);
2745 if (endptr == opt) {
2746 sgen_env_var_error (env_var, "Using default value.", "`%s` must be a number.", opt_name);
2749 else if (val < min || val > max) {
2750 sgen_env_var_error (env_var, "Using default value.", "`%s` must be between %.2f - %.2f.", opt_name, min, max);
2762 char *major_collector_opt = NULL;
2763 char *minor_collector_opt = NULL;
2764 size_t max_heap = 0;
2765 size_t soft_limit = 0;
2767 gboolean debug_print_allowance = FALSE;
2768 double allowance_ratio = 0, save_target = 0;
2769 gboolean cement_enabled = TRUE;
2772 result = InterlockedCompareExchange (&gc_initialized, -1, 0);
2775 /* already inited */
2778 /* being inited by another thread */
2779 mono_thread_info_usleep (1000);
2782 /* we will init it */
2785 g_assert_not_reached ();
2787 } while (result != 0);
2789 SGEN_TV_GETTIME (sgen_init_timestamp);
2791 #ifdef SGEN_WITHOUT_MONO
2792 mono_thread_smr_init ();
2795 mono_coop_mutex_init (&gc_mutex);
2797 gc_debug_file = stderr;
2799 mono_coop_mutex_init (&sgen_interruption_mutex);
2801 if ((env = g_getenv (MONO_GC_PARAMS_NAME))) {
2802 opts = g_strsplit (env, ",", -1);
2803 for (ptr = opts; *ptr; ++ptr) {
2805 if (g_str_has_prefix (opt, "major=")) {
2806 opt = strchr (opt, '=') + 1;
2807 major_collector_opt = g_strdup (opt);
2808 } else if (g_str_has_prefix (opt, "minor=")) {
2809 opt = strchr (opt, '=') + 1;
2810 minor_collector_opt = g_strdup (opt);
2818 sgen_init_internal_allocator ();
2819 sgen_init_nursery_allocator ();
2820 sgen_init_fin_weak_hash ();
2821 sgen_init_hash_table ();
2822 sgen_init_descriptors ();
2823 sgen_init_gray_queues ();
2824 sgen_init_allocator ();
2825 sgen_init_gchandles ();
2827 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_SECTION, SGEN_SIZEOF_GC_MEM_SECTION);
2828 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_GRAY_QUEUE, sizeof (GrayQueueSection));
2830 sgen_client_init ();
2832 if (!minor_collector_opt) {
2833 sgen_simple_nursery_init (&sgen_minor_collector);
2835 if (!strcmp (minor_collector_opt, "simple")) {
2837 sgen_simple_nursery_init (&sgen_minor_collector);
2838 } else if (!strcmp (minor_collector_opt, "split")) {
2839 sgen_split_nursery_init (&sgen_minor_collector);
2841 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `simple` instead.", "Unknown minor collector `%s'.", minor_collector_opt);
2842 goto use_simple_nursery;
2846 if (!major_collector_opt) {
2848 DEFAULT_MAJOR_INIT (&major_collector);
2849 } else if (!strcmp (major_collector_opt, "marksweep")) {
2850 sgen_marksweep_init (&major_collector);
2851 } else if (!strcmp (major_collector_opt, "marksweep-conc")) {
2852 sgen_marksweep_conc_init (&major_collector);
2854 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `" DEFAULT_MAJOR_NAME "` instead.", "Unknown major collector `%s'.", major_collector_opt);
2855 goto use_default_major;
2858 sgen_nursery_size = DEFAULT_NURSERY_SIZE;
2861 gboolean usage_printed = FALSE;
2863 for (ptr = opts; *ptr; ++ptr) {
2865 if (!strcmp (opt, ""))
2867 if (g_str_has_prefix (opt, "major="))
2869 if (g_str_has_prefix (opt, "minor="))
2871 if (g_str_has_prefix (opt, "max-heap-size=")) {
2872 size_t page_size = mono_pagesize ();
2873 size_t max_heap_candidate = 0;
2874 opt = strchr (opt, '=') + 1;
2875 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &max_heap_candidate)) {
2876 max_heap = (max_heap_candidate + page_size - 1) & ~(size_t)(page_size - 1);
2877 if (max_heap != max_heap_candidate)
2878 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Rounding up.", "`max-heap-size` size must be a multiple of %d.", page_size);
2880 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`max-heap-size` must be an integer.");
2884 if (g_str_has_prefix (opt, "soft-heap-limit=")) {
2885 opt = strchr (opt, '=') + 1;
2886 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &soft_limit)) {
2887 if (soft_limit <= 0) {
2888 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be positive.");
2892 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be an integer.");
2898 if (g_str_has_prefix (opt, "nursery-size=")) {
2900 opt = strchr (opt, '=') + 1;
2901 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &val)) {
2902 if ((val & (val - 1))) {
2903 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be a power of two.");
2907 if (val < SGEN_MAX_NURSERY_WASTE) {
2908 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.",
2909 "`nursery-size` must be at least %d bytes.", SGEN_MAX_NURSERY_WASTE);
2913 sgen_nursery_size = val;
2914 sgen_nursery_bits = 0;
2915 while (ONE_P << (++ sgen_nursery_bits) != sgen_nursery_size)
2918 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be an integer.");
2924 if (g_str_has_prefix (opt, "save-target-ratio=")) {
2926 opt = strchr (opt, '=') + 1;
2927 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "save-target-ratio", opt,
2928 SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO, &val)) {
2933 if (g_str_has_prefix (opt, "default-allowance-ratio=")) {
2935 opt = strchr (opt, '=') + 1;
2936 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "default-allowance-ratio", opt,
2937 SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, SGEN_MAX_ALLOWANCE_NURSERY_SIZE_RATIO, &val)) {
2938 allowance_ratio = val;
2943 if (!strcmp (opt, "cementing")) {
2944 cement_enabled = TRUE;
2947 if (!strcmp (opt, "no-cementing")) {
2948 cement_enabled = FALSE;
2952 if (!strcmp (opt, "precleaning")) {
2953 precleaning_enabled = TRUE;
2956 if (!strcmp (opt, "no-precleaning")) {
2957 precleaning_enabled = FALSE;
2961 if (major_collector.handle_gc_param && major_collector.handle_gc_param (opt))
2964 if (sgen_minor_collector.handle_gc_param && sgen_minor_collector.handle_gc_param (opt))
2967 if (sgen_client_handle_gc_param (opt))
2970 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Ignoring.", "Unknown option `%s`.", opt);
2975 fprintf (stderr, "\n%s must be a comma-delimited list of one or more of the following:\n", MONO_GC_PARAMS_NAME);
2976 fprintf (stderr, " max-heap-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
2977 fprintf (stderr, " soft-heap-limit=n (where N is an integer, possibly with a k, m or a g suffix)\n");
2978 fprintf (stderr, " nursery-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
2979 fprintf (stderr, " major=COLLECTOR (where COLLECTOR is `marksweep', `marksweep-conc', `marksweep-par')\n");
2980 fprintf (stderr, " minor=COLLECTOR (where COLLECTOR is `simple' or `split')\n");
2981 fprintf (stderr, " wbarrier=WBARRIER (where WBARRIER is `remset' or `cardtable')\n");
2982 fprintf (stderr, " [no-]cementing\n");
2983 if (major_collector.print_gc_param_usage)
2984 major_collector.print_gc_param_usage ();
2985 if (sgen_minor_collector.print_gc_param_usage)
2986 sgen_minor_collector.print_gc_param_usage ();
2987 sgen_client_print_gc_params_usage ();
2988 fprintf (stderr, " Experimental options:\n");
2989 fprintf (stderr, " save-target-ratio=R (where R must be between %.2f - %.2f).\n", SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO);
2990 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);
2991 fprintf (stderr, "\n");
2993 usage_printed = TRUE;
2998 if (major_collector_opt)
2999 g_free (major_collector_opt);
3001 if (minor_collector_opt)
3002 g_free (minor_collector_opt);
3006 sgen_pinning_init ();
3007 sgen_cement_init (cement_enabled);
3009 if ((env = g_getenv (MONO_GC_DEBUG_NAME))) {
3010 gboolean usage_printed = FALSE;
3012 opts = g_strsplit (env, ",", -1);
3013 for (ptr = opts; ptr && *ptr; ptr ++) {
3015 if (!strcmp (opt, ""))
3017 if (opt [0] >= '0' && opt [0] <= '9') {
3018 gc_debug_level = atoi (opt);
3023 char *rf = g_strdup_printf ("%s.%d", opt, mono_process_current_pid ());
3024 gc_debug_file = fopen (rf, "wb");
3026 gc_debug_file = stderr;
3029 } else if (!strcmp (opt, "print-allowance")) {
3030 debug_print_allowance = TRUE;
3031 } else if (!strcmp (opt, "print-pinning")) {
3032 sgen_pin_stats_enable ();
3033 } else if (!strcmp (opt, "verify-before-allocs")) {
3034 verify_before_allocs = 1;
3035 has_per_allocation_action = TRUE;
3036 } else if (g_str_has_prefix (opt, "verify-before-allocs=")) {
3037 char *arg = strchr (opt, '=') + 1;
3038 verify_before_allocs = atoi (arg);
3039 has_per_allocation_action = TRUE;
3040 } else if (!strcmp (opt, "collect-before-allocs")) {
3041 collect_before_allocs = 1;
3042 has_per_allocation_action = TRUE;
3043 } else if (g_str_has_prefix (opt, "collect-before-allocs=")) {
3044 char *arg = strchr (opt, '=') + 1;
3045 has_per_allocation_action = TRUE;
3046 collect_before_allocs = atoi (arg);
3047 } else if (!strcmp (opt, "verify-before-collections")) {
3048 whole_heap_check_before_collection = TRUE;
3049 } else if (!strcmp (opt, "check-remset-consistency")) {
3050 remset_consistency_checks = TRUE;
3051 nursery_clear_policy = CLEAR_AT_GC;
3052 } else if (!strcmp (opt, "mod-union-consistency-check")) {
3053 if (!major_collector.is_concurrent) {
3054 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`mod-union-consistency-check` only works with concurrent major collector.");
3057 mod_union_consistency_check = TRUE;
3058 } else if (!strcmp (opt, "check-mark-bits")) {
3059 check_mark_bits_after_major_collection = TRUE;
3060 } else if (!strcmp (opt, "check-nursery-pinned")) {
3061 check_nursery_objects_pinned = TRUE;
3062 } else if (!strcmp (opt, "clear-at-gc")) {
3063 nursery_clear_policy = CLEAR_AT_GC;
3064 } else if (!strcmp (opt, "clear-nursery-at-gc")) {
3065 nursery_clear_policy = CLEAR_AT_GC;
3066 } else if (!strcmp (opt, "clear-at-tlab-creation")) {
3067 nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
3068 } else if (!strcmp (opt, "debug-clear-at-tlab-creation")) {
3069 nursery_clear_policy = CLEAR_AT_TLAB_CREATION_DEBUG;
3070 } else if (!strcmp (opt, "check-scan-starts")) {
3071 do_scan_starts_check = TRUE;
3072 } else if (!strcmp (opt, "verify-nursery-at-minor-gc")) {
3073 do_verify_nursery = TRUE;
3074 } else if (!strcmp (opt, "check-concurrent")) {
3075 if (!major_collector.is_concurrent) {
3076 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`check-concurrent` only works with concurrent major collectors.");
3079 nursery_clear_policy = CLEAR_AT_GC;
3080 do_concurrent_checks = TRUE;
3081 } else if (!strcmp (opt, "dump-nursery-at-minor-gc")) {
3082 do_dump_nursery_content = TRUE;
3083 } else if (!strcmp (opt, "disable-minor")) {
3084 disable_minor_collections = TRUE;
3085 } else if (!strcmp (opt, "disable-major")) {
3086 disable_major_collections = TRUE;
3087 } else if (g_str_has_prefix (opt, "heap-dump=")) {
3088 char *filename = strchr (opt, '=') + 1;
3089 nursery_clear_policy = CLEAR_AT_GC;
3090 sgen_debug_enable_heap_dump (filename);
3091 } else if (g_str_has_prefix (opt, "binary-protocol=")) {
3092 char *filename = strchr (opt, '=') + 1;
3093 char *colon = strrchr (filename, ':');
3096 if (!mono_gc_parse_environment_string_extract_number (colon + 1, &limit)) {
3097 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring limit.", "Binary protocol file size limit must be an integer.");
3102 binary_protocol_init (filename, (long long)limit);
3103 } else if (!strcmp (opt, "nursery-canaries")) {
3104 do_verify_nursery = TRUE;
3105 enable_nursery_canaries = TRUE;
3106 } else if (!sgen_client_handle_gc_debug (opt)) {
3107 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "Unknown option `%s`.", opt);
3112 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);
3113 fprintf (stderr, "Valid <option>s are:\n");
3114 fprintf (stderr, " collect-before-allocs[=<n>]\n");
3115 fprintf (stderr, " verify-before-allocs[=<n>]\n");
3116 fprintf (stderr, " check-remset-consistency\n");
3117 fprintf (stderr, " check-mark-bits\n");
3118 fprintf (stderr, " check-nursery-pinned\n");
3119 fprintf (stderr, " verify-before-collections\n");
3120 fprintf (stderr, " verify-nursery-at-minor-gc\n");
3121 fprintf (stderr, " dump-nursery-at-minor-gc\n");
3122 fprintf (stderr, " disable-minor\n");
3123 fprintf (stderr, " disable-major\n");
3124 fprintf (stderr, " check-concurrent\n");
3125 fprintf (stderr, " clear-[nursery-]at-gc\n");
3126 fprintf (stderr, " clear-at-tlab-creation\n");
3127 fprintf (stderr, " debug-clear-at-tlab-creation\n");
3128 fprintf (stderr, " check-scan-starts\n");
3129 fprintf (stderr, " print-allowance\n");
3130 fprintf (stderr, " print-pinning\n");
3131 fprintf (stderr, " heap-dump=<filename>\n");
3132 fprintf (stderr, " binary-protocol=<filename>[:<file-size-limit>]\n");
3133 fprintf (stderr, " nursery-canaries\n");
3134 sgen_client_print_gc_debug_usage ();
3135 fprintf (stderr, "\n");
3137 usage_printed = TRUE;
3143 if (check_mark_bits_after_major_collection)
3144 nursery_clear_policy = CLEAR_AT_GC;
3146 if (major_collector.post_param_init)
3147 major_collector.post_param_init (&major_collector);
3149 if (major_collector.needs_thread_pool)
3150 sgen_workers_init (1);
3152 sgen_memgov_init (max_heap, soft_limit, debug_print_allowance, allowance_ratio, save_target);
3154 memset (&remset, 0, sizeof (remset));
3156 sgen_card_table_init (&remset);
3158 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");
3162 sgen_init_bridge ();
3166 sgen_gc_initialized ()
3168 return gc_initialized > 0;
3172 sgen_get_nursery_clear_policy (void)
3174 return nursery_clear_policy;
3180 mono_coop_mutex_lock (&gc_mutex);
3184 sgen_gc_unlock (void)
3186 mono_coop_mutex_unlock (&gc_mutex);
3190 sgen_major_collector_iterate_live_block_ranges (sgen_cardtable_block_callback callback)
3192 major_collector.iterate_live_block_ranges (callback);
3196 sgen_major_collector_iterate_block_ranges (sgen_cardtable_block_callback callback)
3198 major_collector.iterate_block_ranges (callback);
3202 sgen_get_major_collector (void)
3204 return &major_collector;
3208 sgen_get_remset (void)
3214 count_cards (long long *major_total, long long *major_marked, long long *los_total, long long *los_marked)
3216 sgen_get_major_collector ()->count_cards (major_total, major_marked);
3217 sgen_los_count_cards (los_total, los_marked);
3220 static gboolean world_is_stopped = FALSE;
3222 /* LOCKING: assumes the GC lock is held */
3224 sgen_stop_world (int generation)
3226 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3228 SGEN_ASSERT (0, !world_is_stopped, "Why are we stopping a stopped world?");
3230 binary_protocol_world_stopping (generation, sgen_timestamp (), (gpointer) (gsize) mono_native_thread_id_get ());
3232 sgen_client_stop_world (generation);
3234 world_is_stopped = TRUE;
3236 if (binary_protocol_is_heavy_enabled ())
3237 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3238 binary_protocol_world_stopped (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3241 /* LOCKING: assumes the GC lock is held */
3243 sgen_restart_world (int generation)
3245 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3248 SGEN_ASSERT (0, world_is_stopped, "Why are we restarting a running world?");
3250 if (binary_protocol_is_heavy_enabled ())
3251 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3252 binary_protocol_world_restarting (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3254 world_is_stopped = FALSE;
3256 sgen_client_restart_world (generation, &stw_time);
3258 binary_protocol_world_restarted (generation, sgen_timestamp ());
3260 if (sgen_client_bridge_need_processing ())
3261 sgen_client_bridge_processing_finish (generation);
3263 sgen_memgov_collection_end (generation, stw_time);
3267 sgen_is_world_stopped (void)
3269 return world_is_stopped;
3273 sgen_check_whole_heap_stw (void)
3275 sgen_stop_world (0);
3276 sgen_clear_nursery_fragments ();
3277 sgen_check_whole_heap (FALSE);
3278 sgen_restart_world (0);
3282 sgen_timestamp (void)
3284 SGEN_TV_DECLARE (timestamp);
3285 SGEN_TV_GETTIME (timestamp);
3286 return SGEN_TV_ELAPSED (sgen_init_timestamp, timestamp);
3289 #endif /* HAVE_SGEN_GC */