3 * Simple generational GC.
6 * Paolo Molaro (lupus@ximian.com)
7 * Rodrigo Kumpera (kumpera@gmail.com)
9 * Copyright 2005-2011 Novell, Inc (http://www.novell.com)
10 * Copyright 2011 Xamarin Inc (http://www.xamarin.com)
12 * Thread start/stop adapted from Boehm's GC:
13 * Copyright (c) 1994 by Xerox Corporation. All rights reserved.
14 * Copyright (c) 1996 by Silicon Graphics. All rights reserved.
15 * Copyright (c) 1998 by Fergus Henderson. All rights reserved.
16 * Copyright (c) 2000-2004 by Hewlett-Packard Company. All rights reserved.
17 * Copyright 2001-2003 Ximian, Inc
18 * Copyright 2003-2010 Novell, Inc.
19 * Copyright 2011 Xamarin, Inc.
20 * Copyright (C) 2012 Xamarin Inc
22 * Licensed under the MIT license. See LICENSE file in the project root for full license information.
24 * Important: allocation provides always zeroed memory, having to do
25 * a memset after allocation is deadly for performance.
26 * Memory usage at startup is currently as follows:
28 * 64 KB internal space
30 * We should provide a small memory config with half the sizes
32 * We currently try to make as few mono assumptions as possible:
33 * 1) 2-word header with no GC pointers in it (first vtable, second to store the
35 * 2) gc descriptor is the second word in the vtable (first word in the class)
36 * 3) 8 byte alignment is the minimum and enough (not true for special structures (SIMD), FIXME)
37 * 4) there is a function to get an object's size and the number of
38 * elements in an array.
39 * 5) we know the special way bounds are allocated for complex arrays
40 * 6) we know about proxies and how to treat them when domains are unloaded
42 * Always try to keep stack usage to a minimum: no recursive behaviour
43 * and no large stack allocs.
45 * General description.
46 * Objects are initially allocated in a nursery using a fast bump-pointer technique.
47 * When the nursery is full we start a nursery collection: this is performed with a
49 * When the old generation is full we start a copying GC of the old generation as well:
50 * this will be changed to mark&sweep with copying when fragmentation becomes to severe
51 * in the future. Maybe we'll even do both during the same collection like IMMIX.
53 * The things that complicate this description are:
54 * *) pinned objects: we can't move them so we need to keep track of them
55 * *) no precise info of the thread stacks and registers: we need to be able to
56 * quickly find the objects that may be referenced conservatively and pin them
57 * (this makes the first issues more important)
58 * *) large objects are too expensive to be dealt with using copying GC: we handle them
59 * with mark/sweep during major collections
60 * *) some objects need to not move even if they are small (interned strings, Type handles):
61 * we use mark/sweep for them, too: they are not allocated in the nursery, but inside
62 * PinnedChunks regions
68 *) we could have a function pointer in MonoClass to implement
69 customized write barriers for value types
71 *) investigate the stuff needed to advance a thread to a GC-safe
72 point (single-stepping, read from unmapped memory etc) and implement it.
73 This would enable us to inline allocations and write barriers, for example,
74 or at least parts of them, like the write barrier checks.
75 We may need this also for handling precise info on stacks, even simple things
76 as having uninitialized data on the stack and having to wait for the prolog
77 to zero it. Not an issue for the last frame that we scan conservatively.
78 We could always not trust the value in the slots anyway.
80 *) modify the jit to save info about references in stack locations:
81 this can be done just for locals as a start, so that at least
82 part of the stack is handled precisely.
84 *) test/fix endianess issues
86 *) Implement a card table as the write barrier instead of remembered
87 sets? Card tables are not easy to implement with our current
88 memory layout. We have several different kinds of major heap
89 objects: Small objects in regular blocks, small objects in pinned
90 chunks and LOS objects. If we just have a pointer we have no way
91 to tell which kind of object it points into, therefore we cannot
92 know where its card table is. The least we have to do to make
93 this happen is to get rid of write barriers for indirect stores.
96 *) Get rid of write barriers for indirect stores. We can do this by
97 telling the GC to wbarrier-register an object once we do an ldloca
98 or ldelema on it, and to unregister it once it's not used anymore
99 (it can only travel downwards on the stack). The problem with
100 unregistering is that it needs to happen eventually no matter
101 what, even if exceptions are thrown, the thread aborts, etc.
102 Rodrigo suggested that we could do only the registering part and
103 let the collector find out (pessimistically) when it's safe to
104 unregister, namely when the stack pointer of the thread that
105 registered the object is higher than it was when the registering
106 happened. This might make for a good first implementation to get
107 some data on performance.
109 *) Some sort of blacklist support? Blacklists is a concept from the
110 Boehm GC: if during a conservative scan we find pointers to an
111 area which we might use as heap, we mark that area as unusable, so
112 pointer retention by random pinning pointers is reduced.
114 *) experiment with max small object size (very small right now - 2kb,
115 because it's tied to the max freelist size)
117 *) add an option to mmap the whole heap in one chunk: it makes for many
118 simplifications in the checks (put the nursery at the top and just use a single
119 check for inclusion/exclusion): the issue this has is that on 32 bit systems it's
120 not flexible (too much of the address space may be used by default or we can't
121 increase the heap as needed) and we'd need a race-free mechanism to return memory
122 back to the system (mprotect(PROT_NONE) will still keep the memory allocated if it
123 was written to, munmap is needed, but the following mmap may not find the same segment
126 *) memzero the major fragments after restarting the world and optionally a smaller
129 *) investigate having fragment zeroing threads
131 *) separate locks for finalization and other minor stuff to reduce
134 *) try a different copying order to improve memory locality
136 *) a thread abort after a store but before the write barrier will
137 prevent the write barrier from executing
139 *) specialized dynamically generated markers/copiers
141 *) Dynamically adjust TLAB size to the number of threads. If we have
142 too many threads that do allocation, we might need smaller TLABs,
143 and we might get better performance with larger TLABs if we only
144 have a handful of threads. We could sum up the space left in all
145 assigned TLABs and if that's more than some percentage of the
146 nursery size, reduce the TLAB size.
148 *) Explore placing unreachable objects on unused nursery memory.
149 Instead of memset'ng a region to zero, place an int[] covering it.
150 A good place to start is add_nursery_frag. The tricky thing here is
151 placing those objects atomically outside of a collection.
153 *) Allocation should use asymmetric Dekker synchronization:
154 http://blogs.oracle.com/dave/resource/Asymmetric-Dekker-Synchronization.txt
155 This should help weak consistency archs.
162 #define _XOPEN_SOURCE
163 #define _DARWIN_C_SOURCE
169 #ifdef HAVE_PTHREAD_H
172 #ifdef HAVE_PTHREAD_NP_H
173 #include <pthread_np.h>
181 #include "mono/sgen/sgen-gc.h"
182 #include "mono/sgen/sgen-cardtable.h"
183 #include "mono/sgen/sgen-protocol.h"
184 #include "mono/sgen/sgen-memory-governor.h"
185 #include "mono/sgen/sgen-hash-table.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_major_blocks = 0;
271 static guint64 time_minor_scan_los = 0;
272 static guint64 time_minor_scan_pinned = 0;
273 static guint64 time_minor_scan_roots = 0;
274 static guint64 time_minor_finish_gray_stack = 0;
275 static guint64 time_minor_fragment_creation = 0;
277 static guint64 time_major_pre_collection_fragment_clear = 0;
278 static guint64 time_major_pinning = 0;
279 static guint64 time_major_scan_pinned = 0;
280 static guint64 time_major_scan_roots = 0;
281 static guint64 time_major_scan_mod_union = 0;
282 static guint64 time_major_finish_gray_stack = 0;
283 static guint64 time_major_free_bigobjs = 0;
284 static guint64 time_major_los_sweep = 0;
285 static guint64 time_major_sweep = 0;
286 static guint64 time_major_fragment_creation = 0;
288 static guint64 time_max = 0;
290 static SGEN_TV_DECLARE (time_major_conc_collection_start);
291 static SGEN_TV_DECLARE (time_major_conc_collection_end);
293 int gc_debug_level = 0;
295 static char* gc_params_options;
296 static char* gc_debug_options;
300 mono_gc_flush_info (void)
302 fflush (gc_debug_file);
306 #define TV_DECLARE SGEN_TV_DECLARE
307 #define TV_GETTIME SGEN_TV_GETTIME
308 #define TV_ELAPSED SGEN_TV_ELAPSED
310 static SGEN_TV_DECLARE (sgen_init_timestamp);
312 NurseryClearPolicy nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
314 #define object_is_forwarded SGEN_OBJECT_IS_FORWARDED
315 #define object_is_pinned SGEN_OBJECT_IS_PINNED
316 #define pin_object SGEN_PIN_OBJECT
318 #define ptr_in_nursery sgen_ptr_in_nursery
320 #define LOAD_VTABLE SGEN_LOAD_VTABLE
323 nursery_canaries_enabled (void)
325 return enable_nursery_canaries;
328 #define safe_object_get_size sgen_safe_object_get_size
330 #if defined(HAVE_CONC_GC_AS_DEFAULT)
331 /* Use concurrent major on deskstop platforms */
332 #define DEFAULT_MAJOR_INIT sgen_marksweep_conc_init
333 #define DEFAULT_MAJOR_NAME "marksweep-conc"
335 #define DEFAULT_MAJOR_INIT sgen_marksweep_init
336 #define DEFAULT_MAJOR_NAME "marksweep"
340 * ######################################################################
341 * ######## Global data.
342 * ######################################################################
344 MonoCoopMutex gc_mutex;
346 #define SCAN_START_SIZE SGEN_SCAN_START_SIZE
348 size_t degraded_mode = 0;
350 static mword bytes_pinned_from_failed_allocation = 0;
352 GCMemSection *nursery_section = NULL;
353 static volatile mword lowest_heap_address = ~(mword)0;
354 static volatile mword highest_heap_address = 0;
356 MonoCoopMutex sgen_interruption_mutex;
358 int current_collection_generation = -1;
359 static volatile gboolean concurrent_collection_in_progress = FALSE;
361 /* objects that are ready to be finalized */
362 static SgenPointerQueue fin_ready_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
363 static SgenPointerQueue critical_fin_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
365 /* registered roots: the key to the hash is the root start address */
367 * Different kinds of roots are kept separate to speed up pin_from_roots () for example.
369 SgenHashTable roots_hash [ROOT_TYPE_NUM] = {
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),
372 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL)
374 static mword roots_size = 0; /* amount of memory in the root set */
376 /* The size of a TLAB */
377 /* The bigger the value, the less often we have to go to the slow path to allocate a new
378 * one, but the more space is wasted by threads not allocating much memory.
380 * FIXME: Make this self-tuning for each thread.
382 guint32 tlab_size = (1024 * 4);
384 #define MAX_SMALL_OBJ_SIZE SGEN_MAX_SMALL_OBJ_SIZE
386 #define ALLOC_ALIGN SGEN_ALLOC_ALIGN
388 #define ALIGN_UP SGEN_ALIGN_UP
390 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
391 MonoNativeThreadId main_gc_thread = NULL;
394 /*Object was pinned during the current collection*/
395 static mword objects_pinned;
398 * ######################################################################
399 * ######## Macros and function declarations.
400 * ######################################################################
403 /* forward declarations */
404 static void scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx);
406 static void pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx);
407 static void finish_gray_stack (int generation, ScanCopyContext ctx);
410 SgenMajorCollector major_collector;
411 SgenMinorCollector sgen_minor_collector;
413 static SgenRememberedSet remset;
416 * The gray queue a worker job must use. If we're not parallel or
417 * concurrent, we use the main gray queue.
419 static SgenGrayQueue*
420 sgen_workers_get_job_gray_queue (WorkerData *worker_data, SgenGrayQueue *default_gray_queue)
423 return &worker_data->private_gray_queue;
424 SGEN_ASSERT (0, default_gray_queue, "Why don't we have a default gray queue when we're not running in a worker thread?");
425 return default_gray_queue;
429 gray_queue_redirect (SgenGrayQueue *queue)
431 sgen_workers_take_from_queue (queue);
435 sgen_scan_area_with_callback (char *start, char *end, IterateObjectCallbackFunc callback, void *data, gboolean allow_flags, gboolean fail_on_canaries)
437 while (start < end) {
441 if (!*(void**)start) {
442 start += sizeof (void*); /* should be ALLOC_ALIGN, really */
447 if (!(obj = (char *)SGEN_OBJECT_IS_FORWARDED (start)))
453 if (!sgen_client_object_is_array_fill ((GCObject*)obj)) {
454 CHECK_CANARY_FOR_OBJECT ((GCObject*)obj, fail_on_canaries);
455 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
456 callback ((GCObject*)obj, size, data);
457 CANARIFY_SIZE (size);
459 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
467 * sgen_add_to_global_remset:
469 * The global remset contains locations which point into newspace after
470 * a minor collection. This can happen if the objects they point to are pinned.
472 * LOCKING: If called from a parallel collector, the global remset
473 * lock must be held. For serial collectors that is not necessary.
476 sgen_add_to_global_remset (gpointer ptr, GCObject *obj)
478 SGEN_ASSERT (5, sgen_ptr_in_nursery (obj), "Target pointer of global remset must be in the nursery");
480 HEAVY_STAT (++stat_wbarrier_add_to_global_remset);
482 if (!major_collector.is_concurrent) {
483 SGEN_ASSERT (5, current_collection_generation != -1, "Global remsets can only be added during collections");
485 if (current_collection_generation == -1)
486 SGEN_ASSERT (5, sgen_concurrent_collection_in_progress (), "Global remsets outside of collection pauses can only be added by the concurrent collector");
489 if (!object_is_pinned (obj))
490 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");
491 else if (sgen_cement_lookup_or_register (obj))
494 remset.record_pointer (ptr);
496 sgen_pin_stats_register_global_remset (obj);
498 SGEN_LOG (8, "Adding global remset for %p", ptr);
499 binary_protocol_global_remset (ptr, obj, (gpointer)SGEN_LOAD_VTABLE (obj));
503 * sgen_drain_gray_stack:
505 * Scan objects in the gray stack until the stack is empty. This should be called
506 * frequently after each object is copied, to achieve better locality and cache
511 sgen_drain_gray_stack (ScanCopyContext ctx)
513 SGEN_ASSERT (0, ctx.ops->drain_gray_stack, "Why do we have a scan/copy context with a missing drain gray stack function?");
515 return ctx.ops->drain_gray_stack (ctx.queue);
519 * Addresses in the pin queue are already sorted. This function finds
520 * the object header for each address and pins the object. The
521 * addresses must be inside the nursery section. The (start of the)
522 * address array is overwritten with the addresses of the actually
523 * pinned objects. Return the number of pinned objects.
526 pin_objects_from_nursery_pin_queue (gboolean do_scan_objects, ScanCopyContext ctx)
528 GCMemSection *section = nursery_section;
529 void **start = sgen_pinning_get_entry (section->pin_queue_first_entry);
530 void **end = sgen_pinning_get_entry (section->pin_queue_last_entry);
531 void *start_nursery = section->data;
532 void *end_nursery = section->next_data;
537 void *pinning_front = start_nursery;
539 void **definitely_pinned = start;
540 ScanObjectFunc scan_func = ctx.ops->scan_object;
541 SgenGrayQueue *queue = ctx.queue;
543 sgen_nursery_allocator_prepare_for_pinning ();
545 while (start < end) {
546 GCObject *obj_to_pin = NULL;
547 size_t obj_to_pin_size = 0;
552 SGEN_ASSERT (0, addr >= start_nursery && addr < end_nursery, "Potential pinning address out of range");
553 SGEN_ASSERT (0, addr >= last, "Pin queue not sorted");
560 SGEN_LOG (5, "Considering pinning addr %p", addr);
561 /* We've already processed everything up to pinning_front. */
562 if (addr < pinning_front) {
568 * Find the closest scan start <= addr. We might search backward in the
569 * scan_starts array because entries might be NULL. In the worst case we
570 * start at start_nursery.
572 idx = ((char*)addr - (char*)section->data) / SCAN_START_SIZE;
573 SGEN_ASSERT (0, idx < section->num_scan_start, "Scan start index out of range");
574 search_start = (void*)section->scan_starts [idx];
575 if (!search_start || search_start > addr) {
578 search_start = section->scan_starts [idx];
579 if (search_start && search_start <= addr)
582 if (!search_start || search_start > addr)
583 search_start = start_nursery;
587 * If the pinning front is closer than the scan start we found, start
588 * searching at the front.
590 if (search_start < pinning_front)
591 search_start = pinning_front;
594 * Now addr should be in an object a short distance from search_start.
596 * search_start must point to zeroed mem or point to an object.
599 size_t obj_size, canarified_obj_size;
602 if (!*(void**)search_start) {
603 search_start = (void*)ALIGN_UP ((mword)search_start + sizeof (gpointer));
604 /* The loop condition makes sure we don't overrun addr. */
608 canarified_obj_size = obj_size = ALIGN_UP (safe_object_get_size ((GCObject*)search_start));
611 * Filler arrays are marked by an invalid sync word. We don't
612 * consider them for pinning. They are not delimited by canaries,
615 if (!sgen_client_object_is_array_fill ((GCObject*)search_start)) {
616 CHECK_CANARY_FOR_OBJECT (search_start, TRUE);
617 CANARIFY_SIZE (canarified_obj_size);
619 if (addr >= search_start && (char*)addr < (char*)search_start + obj_size) {
620 /* This is the object we're looking for. */
621 obj_to_pin = (GCObject*)search_start;
622 obj_to_pin_size = canarified_obj_size;
627 /* Skip to the next object */
628 search_start = (void*)((char*)search_start + canarified_obj_size);
629 } while (search_start <= addr);
631 /* We've searched past the address we were looking for. */
633 pinning_front = search_start;
634 goto next_pin_queue_entry;
638 * We've found an object to pin. It might still be a dummy array, but we
639 * can advance the pinning front in any case.
641 pinning_front = (char*)obj_to_pin + obj_to_pin_size;
644 * If this is a dummy array marking the beginning of a nursery
645 * fragment, we don't pin it.
647 if (sgen_client_object_is_array_fill (obj_to_pin))
648 goto next_pin_queue_entry;
651 * Finally - pin the object!
653 desc = sgen_obj_get_descriptor_safe (obj_to_pin);
654 if (do_scan_objects) {
655 scan_func (obj_to_pin, desc, queue);
657 SGEN_LOG (4, "Pinned object %p, vtable %p (%s), count %d\n",
658 obj_to_pin, *(void**)obj_to_pin, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj_to_pin)), count);
659 binary_protocol_pin (obj_to_pin,
660 (gpointer)LOAD_VTABLE (obj_to_pin),
661 safe_object_get_size (obj_to_pin));
663 pin_object (obj_to_pin);
664 GRAY_OBJECT_ENQUEUE_SERIAL (queue, obj_to_pin, desc);
665 sgen_pin_stats_register_object (obj_to_pin, GENERATION_NURSERY);
666 definitely_pinned [count] = obj_to_pin;
669 if (concurrent_collection_in_progress)
670 sgen_pinning_register_pinned_in_nursery (obj_to_pin);
672 next_pin_queue_entry:
676 sgen_client_nursery_objects_pinned (definitely_pinned, count);
677 stat_pinned_objects += count;
682 pin_objects_in_nursery (gboolean do_scan_objects, ScanCopyContext ctx)
686 if (nursery_section->pin_queue_first_entry == nursery_section->pin_queue_last_entry)
689 reduced_to = pin_objects_from_nursery_pin_queue (do_scan_objects, ctx);
690 nursery_section->pin_queue_last_entry = nursery_section->pin_queue_first_entry + reduced_to;
694 * This function is only ever called (via `collector_pin_object()` in `sgen-copy-object.h`)
695 * when we can't promote an object because we're out of memory.
698 sgen_pin_object (GCObject *object, SgenGrayQueue *queue)
700 SGEN_ASSERT (0, sgen_ptr_in_nursery (object), "We're only supposed to use this for pinning nursery objects when out of memory.");
703 * All pinned objects are assumed to have been staged, so we need to stage as well.
704 * Also, the count of staged objects shows that "late pinning" happened.
706 sgen_pin_stage_ptr (object);
708 SGEN_PIN_OBJECT (object);
709 binary_protocol_pin (object, (gpointer)LOAD_VTABLE (object), safe_object_get_size (object));
712 sgen_pin_stats_register_object (object, GENERATION_NURSERY);
714 GRAY_OBJECT_ENQUEUE_SERIAL (queue, object, sgen_obj_get_descriptor_safe (object));
717 /* Sort the addresses in array in increasing order.
718 * Done using a by-the book heap sort. Which has decent and stable performance, is pretty cache efficient.
721 sgen_sort_addresses (void **array, size_t size)
726 for (i = 1; i < size; ++i) {
729 size_t parent = (child - 1) / 2;
731 if (array [parent] >= array [child])
734 tmp = array [parent];
735 array [parent] = array [child];
742 for (i = size - 1; i > 0; --i) {
745 array [i] = array [0];
751 while (root * 2 + 1 <= end) {
752 size_t child = root * 2 + 1;
754 if (child < end && array [child] < array [child + 1])
756 if (array [root] >= array [child])
760 array [root] = array [child];
769 * Scan the memory between start and end and queue values which could be pointers
770 * to the area between start_nursery and end_nursery for later consideration.
771 * Typically used for thread stacks.
774 sgen_conservatively_pin_objects_from (void **start, void **end, void *start_nursery, void *end_nursery, int pin_type)
778 SGEN_ASSERT (0, ((mword)start & (SIZEOF_VOID_P - 1)) == 0, "Why are we scanning for references in unaligned memory ?");
780 #if defined(VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE) && !defined(_WIN64)
781 VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE (start, (char*)end - (char*)start);
784 while (start < end) {
786 * *start can point to the middle of an object
787 * note: should we handle pointing at the end of an object?
788 * pinning in C# code disallows pointing at the end of an object
789 * but there is some small chance that an optimizing C compiler
790 * may keep the only reference to an object by pointing
791 * at the end of it. We ignore this small chance for now.
792 * Pointers to the end of an object are indistinguishable
793 * from pointers to the start of the next object in memory
794 * so if we allow that we'd need to pin two objects...
795 * We queue the pointer in an array, the
796 * array will then be sorted and uniqued. This way
797 * we can coalesce several pinning pointers and it should
798 * be faster since we'd do a memory scan with increasing
799 * addresses. Note: we can align the address to the allocation
800 * alignment, so the unique process is more effective.
802 mword addr = (mword)*start;
803 addr &= ~(ALLOC_ALIGN - 1);
804 if (addr >= (mword)start_nursery && addr < (mword)end_nursery) {
805 SGEN_LOG (6, "Pinning address %p from %p", (void*)addr, start);
806 sgen_pin_stage_ptr ((void*)addr);
807 binary_protocol_pin_stage (start, (void*)addr);
808 sgen_pin_stats_register_address ((char*)addr, pin_type);
814 SGEN_LOG (7, "found %d potential pinned heap pointers", count);
818 * The first thing we do in a collection is to identify pinned objects.
819 * This function considers all the areas of memory that need to be
820 * conservatively scanned.
823 pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx)
827 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);
828 /* objects pinned from the API are inside these roots */
829 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_PINNED], void **, start_root, RootRecord *, root) {
830 SGEN_LOG (6, "Pinned roots %p-%p", start_root, root->end_root);
831 sgen_conservatively_pin_objects_from (start_root, (void**)root->end_root, start_nursery, end_nursery, PIN_TYPE_OTHER);
832 } SGEN_HASH_TABLE_FOREACH_END;
833 /* now deal with the thread stacks
834 * in the future we should be able to conservatively scan only:
835 * *) the cpu registers
836 * *) the unmanaged stack frames
837 * *) the _last_ managed stack frame
838 * *) pointers slots in managed frames
840 sgen_client_scan_thread_data (start_nursery, end_nursery, FALSE, ctx);
844 single_arg_user_copy_or_mark (GCObject **obj, void *gc_data)
846 ScanCopyContext *ctx = (ScanCopyContext *)gc_data;
847 ctx->ops->copy_or_mark_object (obj, ctx->queue);
851 * The memory area from start_root to end_root contains pointers to objects.
852 * Their position is precisely described by @desc (this means that the pointer
853 * can be either NULL or the pointer to the start of an object).
854 * This functions copies them to to_space updates them.
856 * This function is not thread-safe!
859 precisely_scan_objects_from (void** start_root, void** end_root, char* n_start, char *n_end, SgenDescriptor desc, ScanCopyContext ctx)
861 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
862 ScanPtrFieldFunc scan_field_func = ctx.ops->scan_ptr_field;
863 SgenGrayQueue *queue = ctx.queue;
865 switch (desc & ROOT_DESC_TYPE_MASK) {
866 case ROOT_DESC_BITMAP:
867 desc >>= ROOT_DESC_TYPE_SHIFT;
869 if ((desc & 1) && *start_root) {
870 copy_func ((GCObject**)start_root, queue);
871 SGEN_LOG (9, "Overwrote root at %p with %p", start_root, *start_root);
877 case ROOT_DESC_COMPLEX: {
878 gsize *bitmap_data = (gsize *)sgen_get_complex_descriptor_bitmap (desc);
879 gsize bwords = (*bitmap_data) - 1;
880 void **start_run = start_root;
882 while (bwords-- > 0) {
883 gsize bmap = *bitmap_data++;
884 void **objptr = start_run;
886 if ((bmap & 1) && *objptr) {
887 copy_func ((GCObject**)objptr, queue);
888 SGEN_LOG (9, "Overwrote root at %p with %p", objptr, *objptr);
893 start_run += GC_BITS_PER_WORD;
897 case ROOT_DESC_VECTOR: {
900 for (p = start_root; p < end_root; p++) {
902 scan_field_func (NULL, (GCObject**)p, queue);
906 case ROOT_DESC_USER: {
907 SgenUserRootMarkFunc marker = sgen_get_user_descriptor_func (desc);
908 marker (start_root, single_arg_user_copy_or_mark, &ctx);
911 case ROOT_DESC_RUN_LEN:
912 g_assert_not_reached ();
914 g_assert_not_reached ();
919 reset_heap_boundaries (void)
921 lowest_heap_address = ~(mword)0;
922 highest_heap_address = 0;
926 sgen_update_heap_boundaries (mword low, mword high)
931 old = lowest_heap_address;
934 } while (SGEN_CAS_PTR ((gpointer*)&lowest_heap_address, (gpointer)low, (gpointer)old) != (gpointer)old);
937 old = highest_heap_address;
940 } while (SGEN_CAS_PTR ((gpointer*)&highest_heap_address, (gpointer)high, (gpointer)old) != (gpointer)old);
944 * Allocate and setup the data structures needed to be able to allocate objects
945 * in the nursery. The nursery is stored in nursery_section.
950 GCMemSection *section;
957 SGEN_LOG (2, "Allocating nursery size: %zu", (size_t)sgen_nursery_size);
958 /* later we will alloc a larger area for the nursery but only activate
959 * what we need. The rest will be used as expansion if we have too many pinned
960 * objects in the existing nursery.
962 /* FIXME: handle OOM */
963 section = (GCMemSection *)sgen_alloc_internal (INTERNAL_MEM_SECTION);
965 alloc_size = sgen_nursery_size;
967 /* If there isn't enough space even for the nursery we should simply abort. */
968 g_assert (sgen_memgov_try_alloc_space (alloc_size, SPACE_NURSERY));
970 data = (char *)major_collector.alloc_heap (alloc_size, alloc_size, DEFAULT_NURSERY_BITS);
971 sgen_update_heap_boundaries ((mword)data, (mword)(data + sgen_nursery_size));
972 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 ());
973 section->data = section->next_data = data;
974 section->size = alloc_size;
975 section->end_data = data + sgen_nursery_size;
976 scan_starts = (alloc_size + SCAN_START_SIZE - 1) / SCAN_START_SIZE;
977 section->scan_starts = (char **)sgen_alloc_internal_dynamic (sizeof (char*) * scan_starts, INTERNAL_MEM_SCAN_STARTS, TRUE);
978 section->num_scan_start = scan_starts;
980 nursery_section = section;
982 sgen_nursery_allocator_set_nursery_bounds (data, data + sgen_nursery_size);
986 mono_gc_get_logfile (void)
988 return gc_debug_file;
992 mono_gc_params_set (const char* options)
994 if (gc_params_options)
995 g_free (gc_params_options);
997 gc_params_options = g_strdup (options);
1001 mono_gc_debug_set (const char* options)
1003 if (gc_debug_options)
1004 g_free (gc_debug_options);
1006 gc_debug_options = g_strdup (options);
1010 scan_finalizer_entries (SgenPointerQueue *fin_queue, ScanCopyContext ctx)
1012 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
1013 SgenGrayQueue *queue = ctx.queue;
1016 for (i = 0; i < fin_queue->next_slot; ++i) {
1017 GCObject *obj = (GCObject *)fin_queue->data [i];
1020 SGEN_LOG (5, "Scan of fin ready object: %p (%s)\n", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
1021 copy_func ((GCObject**)&fin_queue->data [i], queue);
1026 generation_name (int generation)
1028 switch (generation) {
1029 case GENERATION_NURSERY: return "nursery";
1030 case GENERATION_OLD: return "old";
1031 default: g_assert_not_reached ();
1036 sgen_generation_name (int generation)
1038 return generation_name (generation);
1042 finish_gray_stack (int generation, ScanCopyContext ctx)
1046 int done_with_ephemerons, ephemeron_rounds = 0;
1047 char *start_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_start () : NULL;
1048 char *end_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_end () : (char*)-1;
1049 SgenGrayQueue *queue = ctx.queue;
1051 binary_protocol_finish_gray_stack_start (sgen_timestamp (), generation);
1053 * We copied all the reachable objects. Now it's the time to copy
1054 * the objects that were not referenced by the roots, but by the copied objects.
1055 * we built a stack of objects pointed to by gray_start: they are
1056 * additional roots and we may add more items as we go.
1057 * We loop until gray_start == gray_objects which means no more objects have
1058 * been added. Note this is iterative: no recursion is involved.
1059 * We need to walk the LO list as well in search of marked big objects
1060 * (use a flag since this is needed only on major collections). We need to loop
1061 * here as well, so keep a counter of marked LO (increasing it in copy_object).
1062 * To achieve better cache locality and cache usage, we drain the gray stack
1063 * frequently, after each object is copied, and just finish the work here.
1065 sgen_drain_gray_stack (ctx);
1067 SGEN_LOG (2, "%s generation done", generation_name (generation));
1070 Reset bridge data, we might have lingering data from a previous collection if this is a major
1071 collection trigged by minor overflow.
1073 We must reset the gathered bridges since their original block might be evacuated due to major
1074 fragmentation in the meanwhile and the bridge code should not have to deal with that.
1076 if (sgen_client_bridge_need_processing ())
1077 sgen_client_bridge_reset_data ();
1080 * Mark all strong toggleref objects. This must be done before we walk ephemerons or finalizers
1081 * to ensure they see the full set of live objects.
1083 sgen_client_mark_togglerefs (start_addr, end_addr, ctx);
1086 * Walk the ephemeron tables marking all values with reachable keys. This must be completely done
1087 * before processing finalizable objects and non-tracking weak links to avoid finalizing/clearing
1088 * objects that are in fact reachable.
1090 done_with_ephemerons = 0;
1092 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1093 sgen_drain_gray_stack (ctx);
1095 } while (!done_with_ephemerons);
1097 if (sgen_client_bridge_need_processing ()) {
1098 /*Make sure the gray stack is empty before we process bridge objects so we get liveness right*/
1099 sgen_drain_gray_stack (ctx);
1100 sgen_collect_bridge_objects (generation, ctx);
1101 if (generation == GENERATION_OLD)
1102 sgen_collect_bridge_objects (GENERATION_NURSERY, ctx);
1105 Do the first bridge step here, as the collector liveness state will become useless after that.
1107 An important optimization is to only proccess the possibly dead part of the object graph and skip
1108 over all live objects as we transitively know everything they point must be alive too.
1110 The above invariant is completely wrong if we let the gray queue be drained and mark/copy everything.
1112 This has the unfortunate side effect of making overflow collections perform the first step twice, but
1113 given we now have heuristics that perform major GC in anticipation of minor overflows this should not
1116 sgen_client_bridge_processing_stw_step ();
1120 Make sure we drain the gray stack before processing disappearing links and finalizers.
1121 If we don't make sure it is empty we might wrongly see a live object as dead.
1123 sgen_drain_gray_stack (ctx);
1126 We must clear weak links that don't track resurrection before processing object ready for
1127 finalization so they can be cleared before that.
1129 sgen_null_link_in_range (generation, ctx, FALSE);
1130 if (generation == GENERATION_OLD)
1131 sgen_null_link_in_range (GENERATION_NURSERY, ctx, FALSE);
1134 /* walk the finalization queue and move also the objects that need to be
1135 * finalized: use the finalized objects as new roots so the objects they depend
1136 * on are also not reclaimed. As with the roots above, only objects in the nursery
1137 * are marked/copied.
1139 sgen_finalize_in_range (generation, ctx);
1140 if (generation == GENERATION_OLD)
1141 sgen_finalize_in_range (GENERATION_NURSERY, ctx);
1142 /* drain the new stack that might have been created */
1143 SGEN_LOG (6, "Precise scan of gray area post fin");
1144 sgen_drain_gray_stack (ctx);
1147 * This must be done again after processing finalizable objects since CWL slots are cleared only after the key is finalized.
1149 done_with_ephemerons = 0;
1151 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1152 sgen_drain_gray_stack (ctx);
1154 } while (!done_with_ephemerons);
1156 sgen_client_clear_unreachable_ephemerons (ctx);
1159 * We clear togglerefs only after all possible chances of revival are done.
1160 * This is semantically more inline with what users expect and it allows for
1161 * user finalizers to correctly interact with TR objects.
1163 sgen_client_clear_togglerefs (start_addr, end_addr, ctx);
1166 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);
1169 * handle disappearing links
1170 * Note we do this after checking the finalization queue because if an object
1171 * survives (at least long enough to be finalized) we don't clear the link.
1172 * This also deals with a possible issue with the monitor reclamation: with the Boehm
1173 * GC a finalized object my lose the monitor because it is cleared before the finalizer is
1176 g_assert (sgen_gray_object_queue_is_empty (queue));
1178 sgen_null_link_in_range (generation, ctx, TRUE);
1179 if (generation == GENERATION_OLD)
1180 sgen_null_link_in_range (GENERATION_NURSERY, ctx, TRUE);
1181 if (sgen_gray_object_queue_is_empty (queue))
1183 sgen_drain_gray_stack (ctx);
1186 g_assert (sgen_gray_object_queue_is_empty (queue));
1188 binary_protocol_finish_gray_stack_end (sgen_timestamp (), generation);
1192 sgen_check_section_scan_starts (GCMemSection *section)
1195 for (i = 0; i < section->num_scan_start; ++i) {
1196 if (section->scan_starts [i]) {
1197 mword size = safe_object_get_size ((GCObject*) section->scan_starts [i]);
1198 SGEN_ASSERT (0, size >= SGEN_CLIENT_MINIMUM_OBJECT_SIZE && size <= MAX_SMALL_OBJ_SIZE, "Weird object size at scan starts.");
1204 check_scan_starts (void)
1206 if (!do_scan_starts_check)
1208 sgen_check_section_scan_starts (nursery_section);
1209 major_collector.check_scan_starts ();
1213 scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx)
1217 SGEN_HASH_TABLE_FOREACH (&roots_hash [root_type], void **, start_root, RootRecord *, root) {
1218 SGEN_LOG (6, "Precise root scan %p-%p (desc: %p)", start_root, root->end_root, (void*)root->root_desc);
1219 precisely_scan_objects_from (start_root, (void**)root->end_root, addr_start, addr_end, root->root_desc, ctx);
1220 } SGEN_HASH_TABLE_FOREACH_END;
1226 static gboolean inited = FALSE;
1231 mono_counters_register ("Collection max time", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME | MONO_COUNTER_MONOTONIC, &time_max);
1233 mono_counters_register ("Minor fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pre_collection_fragment_clear);
1234 mono_counters_register ("Minor pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pinning);
1235 mono_counters_register ("Minor scan remembered set", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_remsets);
1236 mono_counters_register ("Minor scan major blocks", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_major_blocks);
1237 mono_counters_register ("Minor scan los", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_los);
1238 mono_counters_register ("Minor scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_pinned);
1239 mono_counters_register ("Minor scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_roots);
1240 mono_counters_register ("Minor fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_fragment_creation);
1242 mono_counters_register ("Major fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pre_collection_fragment_clear);
1243 mono_counters_register ("Major pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pinning);
1244 mono_counters_register ("Major scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_pinned);
1245 mono_counters_register ("Major scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_roots);
1246 mono_counters_register ("Major scan mod union", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_mod_union);
1247 mono_counters_register ("Major finish gray stack", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_finish_gray_stack);
1248 mono_counters_register ("Major free big objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_free_bigobjs);
1249 mono_counters_register ("Major LOS sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_los_sweep);
1250 mono_counters_register ("Major sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_sweep);
1251 mono_counters_register ("Major fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_fragment_creation);
1253 mono_counters_register ("Number of pinned objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_pinned_objects);
1255 #ifdef HEAVY_STATISTICS
1256 mono_counters_register ("WBarrier remember pointer", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_add_to_global_remset);
1257 mono_counters_register ("WBarrier arrayref copy", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_arrayref_copy);
1258 mono_counters_register ("WBarrier generic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store);
1259 mono_counters_register ("WBarrier generic atomic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store_atomic);
1260 mono_counters_register ("WBarrier set root", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_set_root);
1262 mono_counters_register ("# objects allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_alloced_degraded);
1263 mono_counters_register ("bytes allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_bytes_alloced_degraded);
1265 mono_counters_register ("# copy_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_nursery);
1266 mono_counters_register ("# objects copied (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_nursery);
1267 mono_counters_register ("# copy_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_major);
1268 mono_counters_register ("# objects copied (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_major);
1270 mono_counters_register ("# scan_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_nursery);
1271 mono_counters_register ("# scan_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_major);
1273 mono_counters_register ("Slots allocated in vain", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_slots_allocated_in_vain);
1275 mono_counters_register ("# nursery copy_object() failed from space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_from_space);
1276 mono_counters_register ("# nursery copy_object() failed forwarded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_forwarded);
1277 mono_counters_register ("# nursery copy_object() failed pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_pinned);
1278 mono_counters_register ("# nursery copy_object() failed to space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_to_space);
1280 sgen_nursery_allocator_init_heavy_stats ();
1288 reset_pinned_from_failed_allocation (void)
1290 bytes_pinned_from_failed_allocation = 0;
1294 sgen_set_pinned_from_failed_allocation (mword objsize)
1296 bytes_pinned_from_failed_allocation += objsize;
1300 sgen_collection_is_concurrent (void)
1302 switch (current_collection_generation) {
1303 case GENERATION_NURSERY:
1305 case GENERATION_OLD:
1306 return concurrent_collection_in_progress;
1308 g_error ("Invalid current generation %d", current_collection_generation);
1314 sgen_concurrent_collection_in_progress (void)
1316 return concurrent_collection_in_progress;
1320 SgenThreadPoolJob job;
1321 SgenObjectOperations *ops;
1322 SgenGrayQueue *gc_thread_gray_queue;
1330 static ScanCopyContext
1331 scan_copy_context_for_scan_job (void *worker_data_untyped, ScanJob *job)
1333 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1335 return CONTEXT_FROM_OBJECT_OPERATIONS (job->ops, sgen_workers_get_job_gray_queue (worker_data, job->gc_thread_gray_queue));
1339 job_remembered_set_scan (void *worker_data_untyped, SgenThreadPoolJob *job)
1341 SGEN_TV_DECLARE (atv);
1342 SGEN_TV_DECLARE (btv);
1343 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job);
1345 SGEN_TV_GETTIME (atv);
1346 sgen_get_major_collector ()->scan_card_table (CARDTABLE_SCAN_GLOBAL, ctx, 0, 1);
1347 SGEN_TV_GETTIME (btv);
1348 time_minor_scan_major_blocks += SGEN_TV_ELAPSED (atv, btv);
1350 sgen_los_scan_card_table (CARDTABLE_SCAN_GLOBAL, ctx, 0, 1);
1351 SGEN_TV_GETTIME (atv);
1352 time_minor_scan_los += SGEN_TV_ELAPSED (btv, atv);
1354 sgen_wbroots_scan_card_table (ctx);
1362 } ScanFromRegisteredRootsJob;
1365 job_scan_from_registered_roots (void *worker_data_untyped, SgenThreadPoolJob *job)
1367 ScanFromRegisteredRootsJob *job_data = (ScanFromRegisteredRootsJob*)job;
1368 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1370 scan_from_registered_roots (job_data->heap_start, job_data->heap_end, job_data->root_type, ctx);
1377 } ScanThreadDataJob;
1380 job_scan_thread_data (void *worker_data_untyped, SgenThreadPoolJob *job)
1382 ScanThreadDataJob *job_data = (ScanThreadDataJob*)job;
1383 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1385 sgen_client_scan_thread_data (job_data->heap_start, job_data->heap_end, TRUE, ctx);
1390 SgenPointerQueue *queue;
1391 } ScanFinalizerEntriesJob;
1394 job_scan_finalizer_entries (void *worker_data_untyped, SgenThreadPoolJob *job)
1396 ScanFinalizerEntriesJob *job_data = (ScanFinalizerEntriesJob*)job;
1397 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1399 scan_finalizer_entries (job_data->queue, ctx);
1403 job_scan_major_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1405 ParallelScanJob *job_data = (ParallelScanJob*)job;
1406 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1408 g_assert (concurrent_collection_in_progress);
1409 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx, job_data->job_index, sgen_workers_get_job_split_count ());
1413 job_scan_los_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1415 ParallelScanJob *job_data = (ParallelScanJob*)job;
1416 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1418 g_assert (concurrent_collection_in_progress);
1419 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx, job_data->job_index, sgen_workers_get_job_split_count ());
1423 job_major_mod_union_preclean (void *worker_data_untyped, SgenThreadPoolJob *job)
1425 ParallelScanJob *job_data = (ParallelScanJob*)job;
1426 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1428 g_assert (concurrent_collection_in_progress);
1430 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx, job_data->job_index, sgen_workers_get_job_split_count ());
1434 job_los_mod_union_preclean (void *worker_data_untyped, SgenThreadPoolJob *job)
1436 ParallelScanJob *job_data = (ParallelScanJob*)job;
1437 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1439 g_assert (concurrent_collection_in_progress);
1441 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx, job_data->job_index, sgen_workers_get_job_split_count ());
1445 job_scan_last_pinned (void *worker_data_untyped, SgenThreadPoolJob *job)
1447 ScanJob *job_data = (ScanJob*)job;
1448 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, job_data);
1450 g_assert (concurrent_collection_in_progress);
1452 sgen_scan_pin_queue_objects (ctx);
1456 workers_finish_callback (void)
1458 ParallelScanJob *psj;
1460 int split_count = sgen_workers_get_job_split_count ();
1462 /* Mod union preclean jobs */
1463 for (i = 0; i < split_count; i++) {
1464 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("preclean major mod union cardtable", job_major_mod_union_preclean, sizeof (ParallelScanJob));
1465 psj->scan_job.ops = sgen_workers_get_idle_func_object_ops ();
1466 psj->scan_job.gc_thread_gray_queue = NULL;
1468 sgen_workers_enqueue_job (&psj->scan_job.job, TRUE);
1471 for (i = 0; i < split_count; i++) {
1472 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("preclean los mod union cardtable", job_los_mod_union_preclean, sizeof (ParallelScanJob));
1473 psj->scan_job.ops = sgen_workers_get_idle_func_object_ops ();
1474 psj->scan_job.gc_thread_gray_queue = NULL;
1476 sgen_workers_enqueue_job (&psj->scan_job.job, TRUE);
1479 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan last pinned", job_scan_last_pinned, sizeof (ScanJob));
1480 sj->ops = sgen_workers_get_idle_func_object_ops ();
1481 sj->gc_thread_gray_queue = NULL;
1482 sgen_workers_enqueue_job (&sj->job, TRUE);
1486 init_gray_queue (SgenGrayQueue *gc_thread_gray_queue, gboolean use_workers)
1489 sgen_workers_init_distribute_gray_queue ();
1490 sgen_gray_object_queue_init (gc_thread_gray_queue, NULL, TRUE);
1494 enqueue_scan_from_roots_jobs (SgenGrayQueue *gc_thread_gray_queue, char *heap_start, char *heap_end, SgenObjectOperations *ops, gboolean enqueue)
1496 ScanFromRegisteredRootsJob *scrrj;
1497 ScanThreadDataJob *stdj;
1498 ScanFinalizerEntriesJob *sfej;
1500 /* registered roots, this includes static fields */
1502 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots normal", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1503 scrrj->scan_job.ops = ops;
1504 scrrj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1505 scrrj->heap_start = heap_start;
1506 scrrj->heap_end = heap_end;
1507 scrrj->root_type = ROOT_TYPE_NORMAL;
1508 sgen_workers_enqueue_job (&scrrj->scan_job.job, enqueue);
1510 if (current_collection_generation == GENERATION_OLD) {
1511 /* During minors we scan the cardtable for these roots instead */
1512 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots wbarrier", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1513 scrrj->scan_job.ops = ops;
1514 scrrj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1515 scrrj->heap_start = heap_start;
1516 scrrj->heap_end = heap_end;
1517 scrrj->root_type = ROOT_TYPE_WBARRIER;
1518 sgen_workers_enqueue_job (&scrrj->scan_job.job, enqueue);
1523 stdj = (ScanThreadDataJob*)sgen_thread_pool_job_alloc ("scan thread data", job_scan_thread_data, sizeof (ScanThreadDataJob));
1524 stdj->scan_job.ops = ops;
1525 stdj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1526 stdj->heap_start = heap_start;
1527 stdj->heap_end = heap_end;
1528 sgen_workers_enqueue_job (&stdj->scan_job.job, enqueue);
1530 /* Scan the list of objects ready for finalization. */
1532 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1533 sfej->scan_job.ops = ops;
1534 sfej->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1535 sfej->queue = &fin_ready_queue;
1536 sgen_workers_enqueue_job (&sfej->scan_job.job, enqueue);
1538 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan critical finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1539 sfej->scan_job.ops = ops;
1540 sfej->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1541 sfej->queue = &critical_fin_queue;
1542 sgen_workers_enqueue_job (&sfej->scan_job.job, enqueue);
1546 * Perform a nursery collection.
1548 * Return whether any objects were late-pinned due to being out of memory.
1551 collect_nursery (const char *reason, gboolean is_overflow, SgenGrayQueue *unpin_queue)
1553 gboolean needs_major, is_parallel = FALSE;
1554 size_t max_garbage_amount;
1556 mword fragment_total;
1558 SgenGrayQueue gc_thread_gray_queue;
1559 SgenObjectOperations *object_ops_nopar, *object_ops_par = NULL;
1560 ScanCopyContext ctx;
1563 SGEN_TV_DECLARE (last_minor_collection_start_tv);
1564 SGEN_TV_DECLARE (last_minor_collection_end_tv);
1566 if (disable_minor_collections)
1569 TV_GETTIME (last_minor_collection_start_tv);
1570 atv = last_minor_collection_start_tv;
1572 binary_protocol_collection_begin (gc_stats.minor_gc_count, GENERATION_NURSERY);
1574 if (sgen_concurrent_collection_in_progress ()) {
1575 /* FIXME Support parallel nursery collections with concurrent major */
1576 object_ops_nopar = &sgen_minor_collector.serial_ops_with_concurrent_major;
1578 object_ops_nopar = &sgen_minor_collector.serial_ops;
1579 if (sgen_minor_collector.is_parallel) {
1580 object_ops_par = &sgen_minor_collector.parallel_ops;
1585 if (do_verify_nursery || do_dump_nursery_content)
1586 sgen_debug_verify_nursery (do_dump_nursery_content);
1588 current_collection_generation = GENERATION_NURSERY;
1590 SGEN_ASSERT (0, !sgen_collection_is_concurrent (), "Why is the nursery collection concurrent?");
1592 reset_pinned_from_failed_allocation ();
1594 check_scan_starts ();
1596 sgen_nursery_alloc_prepare_for_minor ();
1600 nursery_next = sgen_nursery_alloc_get_upper_alloc_bound ();
1601 /* FIXME: optimize later to use the higher address where an object can be present */
1602 nursery_next = MAX (nursery_next, sgen_get_nursery_end ());
1604 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 ()));
1605 max_garbage_amount = nursery_next - sgen_get_nursery_start ();
1606 g_assert (nursery_section->size >= max_garbage_amount);
1608 /* world must be stopped already */
1610 time_minor_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1612 sgen_client_pre_collection_checks ();
1614 nursery_section->next_data = nursery_next;
1616 major_collector.start_nursery_collection ();
1618 sgen_memgov_minor_collection_start ();
1620 init_gray_queue (&gc_thread_gray_queue, is_parallel);
1621 ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops_nopar, &gc_thread_gray_queue);
1623 gc_stats.minor_gc_count ++;
1625 sgen_process_fin_stage_entries ();
1627 /* pin from pinned handles */
1628 sgen_init_pinning ();
1629 sgen_client_binary_protocol_mark_start (GENERATION_NURSERY);
1630 pin_from_roots (sgen_get_nursery_start (), nursery_next, ctx);
1631 /* pin cemented objects */
1632 sgen_pin_cemented_objects ();
1633 /* identify pinned objects */
1634 sgen_optimize_pin_queue ();
1635 sgen_pinning_setup_section (nursery_section);
1637 pin_objects_in_nursery (FALSE, ctx);
1638 sgen_pinning_trim_queue_to_section (nursery_section);
1640 if (remset_consistency_checks)
1641 sgen_check_remset_consistency ();
1643 if (whole_heap_check_before_collection) {
1644 sgen_clear_nursery_fragments ();
1645 sgen_check_whole_heap (FALSE);
1649 time_minor_pinning += TV_ELAPSED (btv, atv);
1650 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), (long long)TV_ELAPSED (btv, atv));
1651 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1653 remset.start_scan_remsets ();
1655 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan remset", job_remembered_set_scan, sizeof (ScanJob));
1656 sj->ops = is_parallel ? object_ops_par : object_ops_nopar;
1657 sj->gc_thread_gray_queue = &gc_thread_gray_queue;
1658 sgen_workers_enqueue_job (&sj->job, is_parallel);
1660 /* we don't have complete write barrier yet, so we scan all the old generation sections */
1662 time_minor_scan_remsets += TV_ELAPSED (atv, btv);
1663 SGEN_LOG (2, "Old generation scan: %lld usecs", (long long)TV_ELAPSED (atv, btv));
1665 sgen_pin_stats_report ();
1667 /* FIXME: Why do we do this at this specific, seemingly random, point? */
1668 sgen_client_collecting_minor (&fin_ready_queue, &critical_fin_queue);
1671 time_minor_scan_pinned += TV_ELAPSED (btv, atv);
1673 enqueue_scan_from_roots_jobs (&gc_thread_gray_queue, sgen_get_nursery_start (), nursery_next, is_parallel ? object_ops_par : object_ops_nopar, is_parallel);
1676 gray_queue_redirect (&gc_thread_gray_queue);
1677 sgen_workers_start_all_workers (object_ops_nopar, object_ops_par, NULL);
1678 sgen_workers_join ();
1682 time_minor_scan_roots += TV_ELAPSED (atv, btv);
1684 finish_gray_stack (GENERATION_NURSERY, ctx);
1687 time_minor_finish_gray_stack += TV_ELAPSED (btv, atv);
1688 sgen_client_binary_protocol_mark_end (GENERATION_NURSERY);
1690 if (objects_pinned) {
1691 sgen_optimize_pin_queue ();
1692 sgen_pinning_setup_section (nursery_section);
1696 * This is the latest point at which we can do this check, because
1697 * sgen_build_nursery_fragments() unpins nursery objects again.
1699 if (remset_consistency_checks)
1700 sgen_check_remset_consistency ();
1702 /* walk the pin_queue, build up the fragment list of free memory, unmark
1703 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1706 sgen_client_binary_protocol_reclaim_start (GENERATION_NURSERY);
1707 fragment_total = sgen_build_nursery_fragments (nursery_section, unpin_queue);
1708 if (!fragment_total)
1711 /* Clear TLABs for all threads */
1712 sgen_clear_tlabs ();
1714 sgen_client_binary_protocol_reclaim_end (GENERATION_NURSERY);
1716 time_minor_fragment_creation += TV_ELAPSED (atv, btv);
1717 SGEN_LOG (2, "Fragment creation: %lld usecs, %lu bytes available", (long long)TV_ELAPSED (atv, btv), (unsigned long)fragment_total);
1719 if (remset_consistency_checks)
1720 sgen_check_major_refs ();
1722 major_collector.finish_nursery_collection ();
1724 TV_GETTIME (last_minor_collection_end_tv);
1725 gc_stats.minor_gc_time += TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv);
1727 sgen_debug_dump_heap ("minor", gc_stats.minor_gc_count - 1, NULL);
1729 /* prepare the pin queue for the next collection */
1730 sgen_finish_pinning ();
1731 if (sgen_have_pending_finalizers ()) {
1732 SGEN_LOG (4, "Finalizer-thread wakeup");
1733 sgen_client_finalize_notify ();
1735 sgen_pin_stats_reset ();
1736 /* clear cemented hash */
1737 sgen_cement_clear_below_threshold ();
1739 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
1741 check_scan_starts ();
1743 binary_protocol_flush_buffers (FALSE);
1745 sgen_memgov_minor_collection_end (reason, is_overflow);
1747 /*objects are late pinned because of lack of memory, so a major is a good call*/
1748 needs_major = objects_pinned > 0;
1749 current_collection_generation = -1;
1752 binary_protocol_collection_end (gc_stats.minor_gc_count - 1, GENERATION_NURSERY, 0, 0);
1754 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1755 sgen_check_nursery_objects_pinned (unpin_queue != NULL);
1761 COPY_OR_MARK_FROM_ROOTS_SERIAL,
1762 COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT,
1763 COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT
1764 } CopyOrMarkFromRootsMode;
1767 major_copy_or_mark_from_roots (SgenGrayQueue *gc_thread_gray_queue, size_t *old_next_pin_slot, CopyOrMarkFromRootsMode mode, SgenObjectOperations *object_ops_nopar, SgenObjectOperations *object_ops_par)
1772 /* FIXME: only use these values for the precise scan
1773 * note that to_space pointers should be excluded anyway...
1775 char *heap_start = NULL;
1776 char *heap_end = (char*)-1;
1777 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops_nopar, gc_thread_gray_queue);
1778 gboolean concurrent = mode != COPY_OR_MARK_FROM_ROOTS_SERIAL;
1780 SGEN_ASSERT (0, !!concurrent == !!concurrent_collection_in_progress, "We've been called with the wrong mode.");
1782 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1783 /*This cleans up unused fragments */
1784 sgen_nursery_allocator_prepare_for_pinning ();
1786 if (do_concurrent_checks)
1787 sgen_debug_check_nursery_is_clean ();
1789 /* The concurrent collector doesn't touch the nursery. */
1790 sgen_nursery_alloc_prepare_for_major ();
1795 /* Pinning depends on this */
1796 sgen_clear_nursery_fragments ();
1798 if (whole_heap_check_before_collection)
1799 sgen_check_whole_heap (TRUE);
1802 time_major_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1804 if (!sgen_collection_is_concurrent ())
1805 nursery_section->next_data = sgen_get_nursery_end ();
1806 /* we should also coalesce scanning from sections close to each other
1807 * and deal with pointers outside of the sections later.
1812 sgen_client_pre_collection_checks ();
1814 if (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1815 /* Remsets are not useful for a major collection */
1816 remset.clear_cards ();
1819 sgen_process_fin_stage_entries ();
1822 sgen_init_pinning ();
1823 SGEN_LOG (6, "Collecting pinned addresses");
1824 pin_from_roots ((void*)lowest_heap_address, (void*)highest_heap_address, ctx);
1825 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1826 /* Pin cemented objects that were forced */
1827 sgen_pin_cemented_objects ();
1829 sgen_optimize_pin_queue ();
1830 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1832 * Cemented objects that are in the pinned list will be marked. When
1833 * marking concurrently we won't mark mod-union cards for these objects.
1834 * Instead they will remain cemented until the next major collection,
1835 * when we will recheck if they are still pinned in the roots.
1837 sgen_cement_force_pinned ();
1840 sgen_client_collecting_major_1 ();
1843 * pin_queue now contains all candidate pointers, sorted and
1844 * uniqued. We must do two passes now to figure out which
1845 * objects are pinned.
1847 * The first is to find within the pin_queue the area for each
1848 * section. This requires that the pin_queue be sorted. We
1849 * also process the LOS objects and pinned chunks here.
1851 * The second, destructive, pass is to reduce the section
1852 * areas to pointers to the actually pinned objects.
1854 SGEN_LOG (6, "Pinning from sections");
1855 /* first pass for the sections */
1856 sgen_find_section_pin_queue_start_end (nursery_section);
1857 /* identify possible pointers to the insize of large objects */
1858 SGEN_LOG (6, "Pinning from large objects");
1859 for (bigobj = los_object_list; bigobj; bigobj = bigobj->next) {
1861 if (sgen_find_optimized_pin_queue_area ((char*)bigobj->data, (char*)bigobj->data + sgen_los_object_size (bigobj), &dummy, &dummy)) {
1862 binary_protocol_pin (bigobj->data, (gpointer)LOAD_VTABLE (bigobj->data), safe_object_get_size (bigobj->data));
1864 if (sgen_los_object_is_pinned (bigobj->data)) {
1865 SGEN_ASSERT (0, mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, "LOS objects can only be pinned here after concurrent marking.");
1868 sgen_los_pin_object (bigobj->data);
1869 if (SGEN_OBJECT_HAS_REFERENCES (bigobj->data))
1870 GRAY_OBJECT_ENQUEUE_SERIAL (gc_thread_gray_queue, bigobj->data, sgen_obj_get_descriptor ((GCObject*)bigobj->data));
1871 sgen_pin_stats_register_object (bigobj->data, GENERATION_OLD);
1872 SGEN_LOG (6, "Marked large object %p (%s) size: %lu from roots", bigobj->data,
1873 sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (bigobj->data)),
1874 (unsigned long)sgen_los_object_size (bigobj));
1876 sgen_client_pinned_los_object (bigobj->data);
1880 pin_objects_in_nursery (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT, ctx);
1881 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1882 sgen_check_nursery_objects_pinned (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1884 major_collector.pin_objects (gc_thread_gray_queue);
1885 if (old_next_pin_slot)
1886 *old_next_pin_slot = sgen_get_pinned_count ();
1889 time_major_pinning += TV_ELAPSED (atv, btv);
1890 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), (long long)TV_ELAPSED (atv, btv));
1891 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1893 major_collector.init_to_space ();
1895 SGEN_ASSERT (0, sgen_workers_all_done (), "Why are the workers not done when we start or finish a major collection?");
1896 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1897 if (object_ops_par != NULL)
1898 sgen_workers_set_num_active_workers (0);
1899 if (sgen_workers_have_idle_work ()) {
1901 * We force the finish of the worker with the new object ops context
1902 * which can also do copying. We need to have finished pinning.
1904 sgen_workers_start_all_workers (object_ops_nopar, object_ops_par, NULL);
1906 sgen_workers_join ();
1910 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
1911 main_gc_thread = mono_native_thread_self ();
1914 sgen_client_collecting_major_2 ();
1917 time_major_scan_pinned += TV_ELAPSED (btv, atv);
1919 sgen_client_collecting_major_3 (&fin_ready_queue, &critical_fin_queue);
1921 enqueue_scan_from_roots_jobs (gc_thread_gray_queue, heap_start, heap_end, object_ops_nopar, FALSE);
1924 time_major_scan_roots += TV_ELAPSED (atv, btv);
1927 * We start the concurrent worker after pinning and after we scanned the roots
1928 * in order to make sure that the worker does not finish before handling all
1931 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1932 sgen_workers_set_num_active_workers (1);
1933 gray_queue_redirect (gc_thread_gray_queue);
1934 if (precleaning_enabled) {
1935 sgen_workers_start_all_workers (object_ops_nopar, object_ops_par, workers_finish_callback);
1937 sgen_workers_start_all_workers (object_ops_nopar, object_ops_par, NULL);
1941 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1942 int i, split_count = sgen_workers_get_job_split_count ();
1943 gboolean parallel = object_ops_par != NULL;
1945 /* If we're not parallel we finish the collection on the gc thread */
1947 gray_queue_redirect (gc_thread_gray_queue);
1949 /* Mod union card table */
1950 for (i = 0; i < split_count; i++) {
1951 ParallelScanJob *psj;
1953 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("scan mod union cardtable", job_scan_major_mod_union_card_table, sizeof (ParallelScanJob));
1954 psj->scan_job.ops = object_ops_par ? object_ops_par : object_ops_nopar;
1955 psj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1957 sgen_workers_enqueue_job (&psj->scan_job.job, parallel);
1959 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("scan LOS mod union cardtable", job_scan_los_mod_union_card_table, sizeof (ParallelScanJob));
1960 psj->scan_job.ops = object_ops_par ? object_ops_par : object_ops_nopar;
1961 psj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1963 sgen_workers_enqueue_job (&psj->scan_job.job, parallel);
1968 * If we enqueue a job while workers are running we need to sgen_workers_ensure_awake
1969 * in order to make sure that we are running the idle func and draining all worker
1970 * gray queues. The operation of starting workers implies this, so we start them after
1971 * in order to avoid doing this operation twice. The workers will drain the main gray
1972 * stack that contained roots and pinned objects and also scan the mod union card
1975 sgen_workers_start_all_workers (object_ops_nopar, object_ops_par, NULL);
1976 sgen_workers_join ();
1980 sgen_pin_stats_report ();
1982 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1983 sgen_finish_pinning ();
1985 sgen_pin_stats_reset ();
1987 if (do_concurrent_checks)
1988 sgen_debug_check_nursery_is_clean ();
1993 major_start_collection (SgenGrayQueue *gc_thread_gray_queue, const char *reason, gboolean concurrent, size_t *old_next_pin_slot)
1995 SgenObjectOperations *object_ops_nopar, *object_ops_par = NULL;
1997 binary_protocol_collection_begin (gc_stats.major_gc_count, GENERATION_OLD);
1999 current_collection_generation = GENERATION_OLD;
2001 sgen_workers_assert_gray_queue_is_empty ();
2004 sgen_cement_reset ();
2007 g_assert (major_collector.is_concurrent);
2008 concurrent_collection_in_progress = TRUE;
2010 object_ops_nopar = &major_collector.major_ops_concurrent_start;
2011 if (major_collector.is_parallel)
2012 object_ops_par = &major_collector.major_ops_conc_par_start;
2015 object_ops_nopar = &major_collector.major_ops_serial;
2018 reset_pinned_from_failed_allocation ();
2020 sgen_memgov_major_collection_start (concurrent, reason);
2022 //count_ref_nonref_objs ();
2023 //consistency_check ();
2025 check_scan_starts ();
2028 SGEN_LOG (1, "Start major collection %d", gc_stats.major_gc_count);
2029 gc_stats.major_gc_count ++;
2031 if (major_collector.start_major_collection)
2032 major_collector.start_major_collection ();
2034 major_copy_or_mark_from_roots (gc_thread_gray_queue, old_next_pin_slot, concurrent ? COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT : COPY_OR_MARK_FROM_ROOTS_SERIAL, object_ops_nopar, object_ops_par);
2038 major_finish_collection (SgenGrayQueue *gc_thread_gray_queue, const char *reason, gboolean is_overflow, size_t old_next_pin_slot, gboolean forced)
2040 ScannedObjectCounts counts;
2041 SgenObjectOperations *object_ops_nopar;
2042 mword fragment_total;
2048 if (concurrent_collection_in_progress) {
2049 SgenObjectOperations *object_ops_par = NULL;
2051 object_ops_nopar = &major_collector.major_ops_concurrent_finish;
2052 if (major_collector.is_parallel)
2053 object_ops_par = &major_collector.major_ops_conc_par_finish;
2055 major_copy_or_mark_from_roots (gc_thread_gray_queue, NULL, COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, object_ops_nopar, object_ops_par);
2057 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
2058 main_gc_thread = NULL;
2061 object_ops_nopar = &major_collector.major_ops_serial;
2064 sgen_workers_assert_gray_queue_is_empty ();
2066 finish_gray_stack (GENERATION_OLD, CONTEXT_FROM_OBJECT_OPERATIONS (object_ops_nopar, gc_thread_gray_queue));
2068 time_major_finish_gray_stack += TV_ELAPSED (btv, atv);
2070 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after joining");
2072 if (objects_pinned) {
2073 g_assert (!concurrent_collection_in_progress);
2076 * This is slow, but we just OOM'd.
2078 * See comment at `sgen_pin_queue_clear_discarded_entries` for how the pin
2079 * queue is laid out at this point.
2081 sgen_pin_queue_clear_discarded_entries (nursery_section, old_next_pin_slot);
2083 * We need to reestablish all pinned nursery objects in the pin queue
2084 * because they're needed for fragment creation. Unpinning happens by
2085 * walking the whole queue, so it's not necessary to reestablish where major
2086 * heap block pins are - all we care is that they're still in there
2089 sgen_optimize_pin_queue ();
2090 sgen_find_section_pin_queue_start_end (nursery_section);
2094 reset_heap_boundaries ();
2095 sgen_update_heap_boundaries ((mword)sgen_get_nursery_start (), (mword)sgen_get_nursery_end ());
2097 /* walk the pin_queue, build up the fragment list of free memory, unmark
2098 * pinned objects as we go, memzero() the empty fragments so they are ready for the
2101 fragment_total = sgen_build_nursery_fragments (nursery_section, NULL);
2102 if (!fragment_total)
2104 SGEN_LOG (4, "Free space in nursery after major %ld", (long)fragment_total);
2106 if (do_concurrent_checks && concurrent_collection_in_progress)
2107 sgen_debug_check_nursery_is_clean ();
2109 /* prepare the pin queue for the next collection */
2110 sgen_finish_pinning ();
2112 /* Clear TLABs for all threads */
2113 sgen_clear_tlabs ();
2115 sgen_pin_stats_reset ();
2117 sgen_cement_clear_below_threshold ();
2119 if (check_mark_bits_after_major_collection)
2120 sgen_check_heap_marked (concurrent_collection_in_progress);
2123 time_major_fragment_creation += TV_ELAPSED (atv, btv);
2125 binary_protocol_sweep_begin (GENERATION_OLD, !major_collector.sweeps_lazily);
2126 sgen_memgov_major_pre_sweep ();
2129 time_major_free_bigobjs += TV_ELAPSED (btv, atv);
2134 time_major_los_sweep += TV_ELAPSED (atv, btv);
2136 major_collector.sweep ();
2138 binary_protocol_sweep_end (GENERATION_OLD, !major_collector.sweeps_lazily);
2141 time_major_sweep += TV_ELAPSED (btv, atv);
2143 sgen_debug_dump_heap ("major", gc_stats.major_gc_count - 1, reason);
2145 if (sgen_have_pending_finalizers ()) {
2146 SGEN_LOG (4, "Finalizer-thread wakeup");
2147 sgen_client_finalize_notify ();
2150 sgen_memgov_major_collection_end (forced, concurrent_collection_in_progress, reason, is_overflow);
2151 current_collection_generation = -1;
2153 memset (&counts, 0, sizeof (ScannedObjectCounts));
2154 major_collector.finish_major_collection (&counts);
2156 sgen_workers_assert_gray_queue_is_empty ();
2158 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after major collection has finished");
2159 if (concurrent_collection_in_progress)
2160 concurrent_collection_in_progress = FALSE;
2162 check_scan_starts ();
2164 binary_protocol_flush_buffers (FALSE);
2166 //consistency_check ();
2168 binary_protocol_collection_end (gc_stats.major_gc_count - 1, GENERATION_OLD, counts.num_scanned_objects, counts.num_unique_scanned_objects);
2172 major_do_collection (const char *reason, gboolean is_overflow, gboolean forced)
2174 TV_DECLARE (time_start);
2175 TV_DECLARE (time_end);
2176 size_t old_next_pin_slot;
2177 SgenGrayQueue gc_thread_gray_queue;
2179 if (disable_major_collections)
2182 if (major_collector.get_and_reset_num_major_objects_marked) {
2183 long long num_marked = major_collector.get_and_reset_num_major_objects_marked ();
2184 g_assert (!num_marked);
2187 /* world must be stopped already */
2188 TV_GETTIME (time_start);
2190 init_gray_queue (&gc_thread_gray_queue, FALSE);
2191 major_start_collection (&gc_thread_gray_queue, reason, FALSE, &old_next_pin_slot);
2192 major_finish_collection (&gc_thread_gray_queue, reason, is_overflow, old_next_pin_slot, forced);
2193 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2195 TV_GETTIME (time_end);
2196 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2198 /* FIXME: also report this to the user, preferably in gc-end. */
2199 if (major_collector.get_and_reset_num_major_objects_marked)
2200 major_collector.get_and_reset_num_major_objects_marked ();
2202 return bytes_pinned_from_failed_allocation > 0;
2206 major_start_concurrent_collection (const char *reason)
2208 TV_DECLARE (time_start);
2209 TV_DECLARE (time_end);
2210 long long num_objects_marked;
2211 SgenGrayQueue gc_thread_gray_queue;
2213 if (disable_major_collections)
2216 TV_GETTIME (time_start);
2217 SGEN_TV_GETTIME (time_major_conc_collection_start);
2219 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2220 g_assert (num_objects_marked == 0);
2222 binary_protocol_concurrent_start ();
2224 init_gray_queue (&gc_thread_gray_queue, TRUE);
2225 // FIXME: store reason and pass it when finishing
2226 major_start_collection (&gc_thread_gray_queue, reason, TRUE, NULL);
2227 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2229 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2231 TV_GETTIME (time_end);
2232 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2234 current_collection_generation = -1;
2238 * Returns whether the major collection has finished.
2241 major_should_finish_concurrent_collection (void)
2243 return sgen_workers_all_done ();
2247 major_update_concurrent_collection (void)
2249 TV_DECLARE (total_start);
2250 TV_DECLARE (total_end);
2252 TV_GETTIME (total_start);
2254 binary_protocol_concurrent_update ();
2256 major_collector.update_cardtable_mod_union ();
2257 sgen_los_update_cardtable_mod_union ();
2259 TV_GETTIME (total_end);
2260 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end);
2264 major_finish_concurrent_collection (gboolean forced)
2266 SgenGrayQueue gc_thread_gray_queue;
2267 TV_DECLARE (total_start);
2268 TV_DECLARE (total_end);
2270 TV_GETTIME (total_start);
2272 binary_protocol_concurrent_finish ();
2275 * We need to stop all workers since we're updating the cardtable below.
2276 * The workers will be resumed with a finishing pause context to avoid
2277 * additional cardtable and object scanning.
2279 sgen_workers_stop_all_workers ();
2281 SGEN_TV_GETTIME (time_major_conc_collection_end);
2282 gc_stats.major_gc_time_concurrent += SGEN_TV_ELAPSED (time_major_conc_collection_start, time_major_conc_collection_end);
2284 major_collector.update_cardtable_mod_union ();
2285 sgen_los_update_cardtable_mod_union ();
2287 if (mod_union_consistency_check)
2288 sgen_check_mod_union_consistency ();
2290 current_collection_generation = GENERATION_OLD;
2291 sgen_cement_reset ();
2292 init_gray_queue (&gc_thread_gray_queue, FALSE);
2293 major_finish_collection (&gc_thread_gray_queue, "finishing", FALSE, -1, forced);
2294 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2296 TV_GETTIME (total_end);
2297 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end);
2299 current_collection_generation = -1;
2303 * Ensure an allocation request for @size will succeed by freeing enough memory.
2305 * LOCKING: The GC lock MUST be held.
2308 sgen_ensure_free_space (size_t size, int generation)
2310 int generation_to_collect = -1;
2311 const char *reason = NULL;
2313 if (generation == GENERATION_OLD) {
2314 if (sgen_need_major_collection (size)) {
2315 reason = "LOS overflow";
2316 generation_to_collect = GENERATION_OLD;
2319 if (degraded_mode) {
2320 if (sgen_need_major_collection (size)) {
2321 reason = "Degraded mode overflow";
2322 generation_to_collect = GENERATION_OLD;
2324 } else if (sgen_need_major_collection (size)) {
2325 reason = concurrent_collection_in_progress ? "Forced finish concurrent collection" : "Minor allowance";
2326 generation_to_collect = GENERATION_OLD;
2328 generation_to_collect = GENERATION_NURSERY;
2329 reason = "Nursery full";
2333 if (generation_to_collect == -1) {
2334 if (concurrent_collection_in_progress && sgen_workers_all_done ()) {
2335 generation_to_collect = GENERATION_OLD;
2336 reason = "Finish concurrent collection";
2340 if (generation_to_collect == -1)
2342 sgen_perform_collection (size, generation_to_collect, reason, FALSE, TRUE);
2346 * LOCKING: Assumes the GC lock is held.
2349 sgen_perform_collection (size_t requested_size, int generation_to_collect, const char *reason, gboolean wait_to_finish, gboolean stw)
2351 TV_DECLARE (gc_total_start);
2352 TV_DECLARE (gc_total_end);
2353 int overflow_generation_to_collect = -1;
2354 int oldest_generation_collected = generation_to_collect;
2355 const char *overflow_reason = NULL;
2356 gboolean finish_concurrent = concurrent_collection_in_progress && (major_should_finish_concurrent_collection () || generation_to_collect == GENERATION_OLD);
2358 binary_protocol_collection_requested (generation_to_collect, requested_size, wait_to_finish ? 1 : 0);
2360 SGEN_ASSERT (0, generation_to_collect == GENERATION_NURSERY || generation_to_collect == GENERATION_OLD, "What generation is this?");
2363 sgen_stop_world (generation_to_collect);
2365 SGEN_ASSERT (0, sgen_is_world_stopped (), "We can only collect if the world is stopped");
2368 TV_GETTIME (gc_total_start);
2370 // FIXME: extract overflow reason
2371 // FIXME: minor overflow for concurrent case
2372 if (generation_to_collect == GENERATION_NURSERY && !finish_concurrent) {
2373 if (concurrent_collection_in_progress)
2374 major_update_concurrent_collection ();
2376 if (collect_nursery (reason, FALSE, NULL) && !concurrent_collection_in_progress) {
2377 overflow_generation_to_collect = GENERATION_OLD;
2378 overflow_reason = "Minor overflow";
2380 } else if (finish_concurrent) {
2381 major_finish_concurrent_collection (wait_to_finish);
2382 oldest_generation_collected = GENERATION_OLD;
2384 SGEN_ASSERT (0, generation_to_collect == GENERATION_OLD, "We should have handled nursery collections above");
2385 if (major_collector.is_concurrent && !wait_to_finish) {
2386 collect_nursery ("Concurrent start", FALSE, NULL);
2387 major_start_concurrent_collection (reason);
2388 oldest_generation_collected = GENERATION_NURSERY;
2389 } else if (major_do_collection (reason, FALSE, wait_to_finish)) {
2390 overflow_generation_to_collect = GENERATION_NURSERY;
2391 overflow_reason = "Excessive pinning";
2395 if (overflow_generation_to_collect != -1) {
2396 SGEN_ASSERT (0, !concurrent_collection_in_progress, "We don't yet support overflow collections with the concurrent collector");
2399 * We need to do an overflow collection, either because we ran out of memory
2400 * or the nursery is fully pinned.
2403 if (overflow_generation_to_collect == GENERATION_NURSERY)
2404 collect_nursery (overflow_reason, TRUE, NULL);
2406 major_do_collection (overflow_reason, TRUE, wait_to_finish);
2408 oldest_generation_collected = MAX (oldest_generation_collected, overflow_generation_to_collect);
2411 SGEN_LOG (2, "Heap size: %lu, LOS size: %lu", (unsigned long)sgen_gc_get_total_heap_allocation (), (unsigned long)los_memory_usage);
2413 /* this also sets the proper pointers for the next allocation */
2414 if (generation_to_collect == GENERATION_NURSERY && !sgen_can_alloc_size (requested_size)) {
2415 /* TypeBuilder and MonoMethod are killing mcs with fragmentation */
2416 SGEN_LOG (1, "nursery collection didn't find enough room for %zd alloc (%zd pinned)", requested_size, sgen_get_pinned_count ());
2417 sgen_dump_pin_queue ();
2421 TV_GETTIME (gc_total_end);
2422 time_max = MAX (time_max, TV_ELAPSED (gc_total_start, gc_total_end));
2425 sgen_restart_world (oldest_generation_collected);
2429 * ######################################################################
2430 * ######## Memory allocation from the OS
2431 * ######################################################################
2432 * This section of code deals with getting memory from the OS and
2433 * allocating memory for GC-internal data structures.
2434 * Internal memory can be handled with a freelist for small objects.
2440 G_GNUC_UNUSED static void
2441 report_internal_mem_usage (void)
2443 printf ("Internal memory usage:\n");
2444 sgen_report_internal_mem_usage ();
2445 printf ("Pinned memory usage:\n");
2446 major_collector.report_pinned_memory_usage ();
2450 * ######################################################################
2451 * ######## Finalization support
2452 * ######################################################################
2456 * If the object has been forwarded it means it's still referenced from a root.
2457 * If it is pinned it's still alive as well.
2458 * A LOS object is only alive if we have pinned it.
2459 * Return TRUE if @obj is ready to be finalized.
2461 static inline gboolean
2462 sgen_is_object_alive (GCObject *object)
2464 if (ptr_in_nursery (object))
2465 return sgen_nursery_is_object_alive (object);
2467 return sgen_major_is_object_alive (object);
2471 * This function returns true if @object is either alive and belongs to the
2472 * current collection - major collections are full heap, so old gen objects
2473 * are never alive during a minor collection.
2476 sgen_is_object_alive_and_on_current_collection (GCObject *object)
2478 if (ptr_in_nursery (object))
2479 return sgen_nursery_is_object_alive (object);
2481 if (current_collection_generation == GENERATION_NURSERY)
2484 return sgen_major_is_object_alive (object);
2489 sgen_gc_is_object_ready_for_finalization (GCObject *object)
2491 return !sgen_is_object_alive (object);
2495 sgen_queue_finalization_entry (GCObject *obj)
2497 gboolean critical = sgen_client_object_has_critical_finalizer (obj);
2499 sgen_pointer_queue_add (critical ? &critical_fin_queue : &fin_ready_queue, obj);
2501 sgen_client_object_queued_for_finalization (obj);
2505 sgen_object_is_live (GCObject *obj)
2507 return sgen_is_object_alive_and_on_current_collection (obj);
2511 * `System.GC.WaitForPendingFinalizers` first checks `sgen_have_pending_finalizers()` to
2512 * determine whether it can exit quickly. The latter must therefore only return FALSE if
2513 * all finalizers have really finished running.
2515 * `sgen_gc_invoke_finalizers()` first dequeues a finalizable object, and then finalizes it.
2516 * This means that just checking whether the queues are empty leaves the possibility that an
2517 * object might have been dequeued but not yet finalized. That's why we need the additional
2518 * flag `pending_unqueued_finalizer`.
2521 static volatile gboolean pending_unqueued_finalizer = FALSE;
2522 volatile gboolean sgen_suspend_finalizers = FALSE;
2525 sgen_set_suspend_finalizers (void)
2527 sgen_suspend_finalizers = TRUE;
2531 sgen_gc_invoke_finalizers (void)
2535 g_assert (!pending_unqueued_finalizer);
2537 /* FIXME: batch to reduce lock contention */
2538 while (sgen_have_pending_finalizers ()) {
2544 * We need to set `pending_unqueued_finalizer` before dequeing the
2545 * finalizable object.
2547 if (!sgen_pointer_queue_is_empty (&fin_ready_queue)) {
2548 pending_unqueued_finalizer = TRUE;
2549 mono_memory_write_barrier ();
2550 obj = (GCObject *)sgen_pointer_queue_pop (&fin_ready_queue);
2551 } else if (!sgen_pointer_queue_is_empty (&critical_fin_queue)) {
2552 pending_unqueued_finalizer = TRUE;
2553 mono_memory_write_barrier ();
2554 obj = (GCObject *)sgen_pointer_queue_pop (&critical_fin_queue);
2560 SGEN_LOG (7, "Finalizing object %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
2568 /* the object is on the stack so it is pinned */
2569 /*g_print ("Calling finalizer for object: %p (%s)\n", obj, sgen_client_object_safe_name (obj));*/
2570 sgen_client_run_finalize (obj);
2573 if (pending_unqueued_finalizer) {
2574 mono_memory_write_barrier ();
2575 pending_unqueued_finalizer = FALSE;
2582 sgen_have_pending_finalizers (void)
2584 if (sgen_suspend_finalizers)
2586 return pending_unqueued_finalizer || !sgen_pointer_queue_is_empty (&fin_ready_queue) || !sgen_pointer_queue_is_empty (&critical_fin_queue);
2590 * ######################################################################
2591 * ######## registered roots support
2592 * ######################################################################
2596 * We do not coalesce roots.
2599 sgen_register_root (char *start, size_t size, SgenDescriptor descr, int root_type, int source, const char *msg)
2601 RootRecord new_root;
2604 for (i = 0; i < ROOT_TYPE_NUM; ++i) {
2605 RootRecord *root = (RootRecord *)sgen_hash_table_lookup (&roots_hash [i], start);
2606 /* we allow changing the size and the descriptor (for thread statics etc) */
2608 size_t old_size = root->end_root - start;
2609 root->end_root = start + size;
2610 SGEN_ASSERT (0, !!root->root_desc == !!descr, "Can't change whether a root is precise or conservative.");
2611 SGEN_ASSERT (0, root->source == source, "Can't change a root's source identifier.");
2612 SGEN_ASSERT (0, !!root->msg == !!msg, "Can't change a root's message.");
2613 root->root_desc = descr;
2615 roots_size -= old_size;
2621 new_root.end_root = start + size;
2622 new_root.root_desc = descr;
2623 new_root.source = source;
2626 sgen_hash_table_replace (&roots_hash [root_type], start, &new_root, NULL);
2629 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);
2636 sgen_deregister_root (char* addr)
2642 for (root_type = 0; root_type < ROOT_TYPE_NUM; ++root_type) {
2643 if (sgen_hash_table_remove (&roots_hash [root_type], addr, &root))
2644 roots_size -= (root.end_root - addr);
2650 sgen_wbroots_iterate_live_block_ranges (sgen_cardtable_block_callback cb)
2654 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_WBARRIER], void **, start_root, RootRecord *, root) {
2655 cb ((mword)start_root, (mword)root->end_root - (mword)start_root);
2656 } SGEN_HASH_TABLE_FOREACH_END;
2659 /* Root equivalent of sgen_client_cardtable_scan_object */
2661 sgen_wbroot_scan_card_table (void** start_root, mword size, ScanCopyContext ctx)
2663 ScanPtrFieldFunc scan_field_func = ctx.ops->scan_ptr_field;
2664 guint8 *card_data = sgen_card_table_get_card_scan_address ((mword)start_root);
2665 guint8 *card_base = card_data;
2666 mword card_count = sgen_card_table_number_of_cards_in_range ((mword)start_root, size);
2667 guint8 *card_data_end = card_data + card_count;
2668 mword extra_idx = 0;
2669 char *obj_start = sgen_card_table_align_pointer (start_root);
2670 char *obj_end = (char*)start_root + size;
2671 #ifdef SGEN_HAVE_OVERLAPPING_CARDS
2672 guint8 *overflow_scan_end = NULL;
2675 #ifdef SGEN_HAVE_OVERLAPPING_CARDS
2676 /*Check for overflow and if so, setup to scan in two steps*/
2677 if (card_data_end >= SGEN_SHADOW_CARDTABLE_END) {
2678 overflow_scan_end = sgen_shadow_cardtable + (card_data_end - SGEN_SHADOW_CARDTABLE_END);
2679 card_data_end = SGEN_SHADOW_CARDTABLE_END;
2685 card_data = sgen_find_next_card (card_data, card_data_end);
2687 for (; card_data < card_data_end; card_data = sgen_find_next_card (card_data + 1, card_data_end)) {
2688 size_t idx = (card_data - card_base) + extra_idx;
2689 char *start = (char*)(obj_start + idx * CARD_SIZE_IN_BYTES);
2690 char *card_end = start + CARD_SIZE_IN_BYTES;
2691 char *elem = start, *first_elem = start;
2694 * Don't clean first and last card on 32bit systems since they
2695 * may also be part from other roots.
2697 if (card_data != card_base && card_data != (card_data_end - 1))
2698 sgen_card_table_prepare_card_for_scanning (card_data);
2700 card_end = MIN (card_end, obj_end);
2702 if (elem < (char*)start_root)
2703 first_elem = elem = (char*)start_root;
2705 for (; elem < card_end; elem += SIZEOF_VOID_P) {
2706 if (*(GCObject**)elem)
2707 scan_field_func (NULL, (GCObject**)elem, ctx.queue);
2710 binary_protocol_card_scan (first_elem, elem - first_elem);
2713 #ifdef SGEN_HAVE_OVERLAPPING_CARDS
2714 if (overflow_scan_end) {
2715 extra_idx = card_data - card_base;
2716 card_base = card_data = sgen_shadow_cardtable;
2717 card_data_end = overflow_scan_end;
2718 overflow_scan_end = NULL;
2725 sgen_wbroots_scan_card_table (ScanCopyContext ctx)
2730 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_WBARRIER], void **, start_root, RootRecord *, root) {
2731 SGEN_ASSERT (0, (root->root_desc & ROOT_DESC_TYPE_MASK) == ROOT_DESC_VECTOR, "Unsupported root type");
2733 sgen_wbroot_scan_card_table (start_root, (mword)root->end_root - (mword)start_root, ctx);
2734 } SGEN_HASH_TABLE_FOREACH_END;
2738 * ######################################################################
2739 * ######## Thread handling (stop/start code)
2740 * ######################################################################
2744 sgen_get_current_collection_generation (void)
2746 return current_collection_generation;
2750 sgen_thread_register (SgenThreadInfo* info, void *stack_bottom_fallback)
2752 info->tlab_start = info->tlab_next = info->tlab_temp_end = info->tlab_real_end = NULL;
2754 sgen_client_thread_register (info, stack_bottom_fallback);
2760 sgen_thread_unregister (SgenThreadInfo *p)
2762 sgen_client_thread_unregister (p);
2766 * ######################################################################
2767 * ######## Write barriers
2768 * ######################################################################
2772 * Note: the write barriers first do the needed GC work and then do the actual store:
2773 * this way the value is visible to the conservative GC scan after the write barrier
2774 * itself. If a GC interrupts the barrier in the middle, value will be kept alive by
2775 * the conservative scan, otherwise by the remembered set scan.
2779 * mono_gc_wbarrier_arrayref_copy:
2782 mono_gc_wbarrier_arrayref_copy (gpointer dest_ptr, gpointer src_ptr, int count)
2784 HEAVY_STAT (++stat_wbarrier_arrayref_copy);
2785 /*This check can be done without taking a lock since dest_ptr array is pinned*/
2786 if (ptr_in_nursery (dest_ptr) || count <= 0) {
2787 mono_gc_memmove_aligned (dest_ptr, src_ptr, count * sizeof (gpointer));
2791 #ifdef SGEN_HEAVY_BINARY_PROTOCOL
2792 if (binary_protocol_is_heavy_enabled ()) {
2794 for (i = 0; i < count; ++i) {
2795 gpointer dest = (gpointer*)dest_ptr + i;
2796 gpointer obj = *((gpointer*)src_ptr + i);
2798 binary_protocol_wbarrier (dest, obj, (gpointer)LOAD_VTABLE (obj));
2803 remset.wbarrier_arrayref_copy (dest_ptr, src_ptr, count);
2807 * mono_gc_wbarrier_generic_nostore:
2810 mono_gc_wbarrier_generic_nostore (gpointer ptr)
2814 HEAVY_STAT (++stat_wbarrier_generic_store);
2816 sgen_client_wbarrier_generic_nostore_check (ptr);
2818 obj = *(gpointer*)ptr;
2820 binary_protocol_wbarrier (ptr, obj, (gpointer)LOAD_VTABLE (obj));
2823 * We need to record old->old pointer locations for the
2824 * concurrent collector.
2826 if (!ptr_in_nursery (obj) && !concurrent_collection_in_progress) {
2827 SGEN_LOG (8, "Skipping remset at %p", ptr);
2831 SGEN_LOG (8, "Adding remset at %p", ptr);
2833 remset.wbarrier_generic_nostore (ptr);
2837 * mono_gc_wbarrier_generic_store:
2840 mono_gc_wbarrier_generic_store (gpointer ptr, GCObject* value)
2842 SGEN_LOG (8, "Wbarrier store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2843 SGEN_UPDATE_REFERENCE_ALLOW_NULL (ptr, value);
2844 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2845 mono_gc_wbarrier_generic_nostore (ptr);
2846 sgen_dummy_use (value);
2850 * mono_gc_wbarrier_generic_store_atomic:
2851 * Same as \c mono_gc_wbarrier_generic_store but performs the store
2852 * as an atomic operation with release semantics.
2855 mono_gc_wbarrier_generic_store_atomic (gpointer ptr, GCObject *value)
2857 HEAVY_STAT (++stat_wbarrier_generic_store_atomic);
2859 SGEN_LOG (8, "Wbarrier atomic store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2861 InterlockedWritePointer ((volatile gpointer *)ptr, value);
2863 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2864 mono_gc_wbarrier_generic_nostore (ptr);
2866 sgen_dummy_use (value);
2870 sgen_wbarrier_value_copy_bitmap (gpointer _dest, gpointer _src, int size, unsigned bitmap)
2872 GCObject **dest = (GCObject **)_dest;
2873 GCObject **src = (GCObject **)_src;
2877 mono_gc_wbarrier_generic_store (dest, *src);
2882 size -= SIZEOF_VOID_P;
2888 * ######################################################################
2889 * ######## Other mono public interface functions.
2890 * ######################################################################
2894 sgen_gc_collect (int generation)
2899 sgen_perform_collection (0, generation, "user request", TRUE, TRUE);
2904 sgen_gc_collection_count (int generation)
2906 if (generation == 0)
2907 return gc_stats.minor_gc_count;
2908 return gc_stats.major_gc_count;
2912 sgen_gc_get_used_size (void)
2916 tot = los_memory_usage;
2917 tot += nursery_section->next_data - nursery_section->data;
2918 tot += major_collector.get_used_size ();
2919 /* FIXME: account for pinned objects */
2925 sgen_env_var_error (const char *env_var, const char *fallback, const char *description_format, ...)
2929 va_start (ap, description_format);
2931 fprintf (stderr, "Warning: In environment variable `%s': ", env_var);
2932 vfprintf (stderr, description_format, ap);
2934 fprintf (stderr, " - %s", fallback);
2935 fprintf (stderr, "\n");
2941 parse_double_in_interval (const char *env_var, const char *opt_name, const char *opt, double min, double max, double *result)
2944 double val = strtod (opt, &endptr);
2945 if (endptr == opt) {
2946 sgen_env_var_error (env_var, "Using default value.", "`%s` must be a number.", opt_name);
2949 else if (val < min || val > max) {
2950 sgen_env_var_error (env_var, "Using default value.", "`%s` must be between %.2f - %.2f.", opt_name, min, max);
2962 char *major_collector_opt = NULL;
2963 char *minor_collector_opt = NULL;
2964 char *params_opts = NULL;
2965 char *debug_opts = NULL;
2966 size_t max_heap = 0;
2967 size_t soft_limit = 0;
2969 gboolean debug_print_allowance = FALSE;
2970 double allowance_ratio = 0, save_target = 0;
2971 gboolean cement_enabled = TRUE;
2974 result = InterlockedCompareExchange (&gc_initialized, -1, 0);
2977 /* already inited */
2980 /* being inited by another thread */
2981 mono_thread_info_usleep (1000);
2984 /* we will init it */
2987 g_assert_not_reached ();
2989 } while (result != 0);
2991 SGEN_TV_GETTIME (sgen_init_timestamp);
2993 #ifdef SGEN_WITHOUT_MONO
2994 mono_thread_smr_init ();
2997 mono_coop_mutex_init (&gc_mutex);
2999 gc_debug_file = stderr;
3001 mono_coop_mutex_init (&sgen_interruption_mutex);
3003 if ((env = g_getenv (MONO_GC_PARAMS_NAME)) || gc_params_options) {
3004 params_opts = g_strdup_printf ("%s,%s", gc_params_options ? gc_params_options : "", env ? env : "");
3008 opts = g_strsplit (params_opts, ",", -1);
3009 for (ptr = opts; *ptr; ++ptr) {
3011 if (g_str_has_prefix (opt, "major=")) {
3012 opt = strchr (opt, '=') + 1;
3013 major_collector_opt = g_strdup (opt);
3014 } else if (g_str_has_prefix (opt, "minor=")) {
3015 opt = strchr (opt, '=') + 1;
3016 minor_collector_opt = g_strdup (opt);
3024 sgen_init_internal_allocator ();
3025 sgen_init_nursery_allocator ();
3026 sgen_init_fin_weak_hash ();
3027 sgen_init_hash_table ();
3028 sgen_init_descriptors ();
3029 sgen_init_gray_queues ();
3030 sgen_init_allocator ();
3031 sgen_init_gchandles ();
3033 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_SECTION, SGEN_SIZEOF_GC_MEM_SECTION);
3034 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_GRAY_QUEUE, sizeof (GrayQueueSection));
3036 sgen_client_init ();
3038 if (!minor_collector_opt) {
3039 sgen_simple_nursery_init (&sgen_minor_collector, FALSE);
3041 if (!strcmp (minor_collector_opt, "simple")) {
3043 sgen_simple_nursery_init (&sgen_minor_collector, FALSE);
3044 } else if (!strcmp (minor_collector_opt, "simple-par")) {
3045 sgen_simple_nursery_init (&sgen_minor_collector, TRUE);
3046 } else if (!strcmp (minor_collector_opt, "split")) {
3047 sgen_split_nursery_init (&sgen_minor_collector);
3049 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `simple` instead.", "Unknown minor collector `%s'.", minor_collector_opt);
3050 goto use_simple_nursery;
3054 if (!major_collector_opt) {
3056 DEFAULT_MAJOR_INIT (&major_collector);
3057 } else if (!strcmp (major_collector_opt, "marksweep")) {
3058 sgen_marksweep_init (&major_collector);
3059 } else if (!strcmp (major_collector_opt, "marksweep-conc")) {
3060 sgen_marksweep_conc_init (&major_collector);
3061 } else if (!strcmp (major_collector_opt, "marksweep-conc-par")) {
3062 sgen_marksweep_conc_par_init (&major_collector);
3064 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using `" DEFAULT_MAJOR_NAME "` instead.", "Unknown major collector `%s'.", major_collector_opt);
3065 goto use_default_major;
3068 sgen_nursery_size = DEFAULT_NURSERY_SIZE;
3071 gboolean usage_printed = FALSE;
3073 for (ptr = opts; *ptr; ++ptr) {
3075 if (!strcmp (opt, ""))
3077 if (g_str_has_prefix (opt, "major="))
3079 if (g_str_has_prefix (opt, "minor="))
3081 if (g_str_has_prefix (opt, "max-heap-size=")) {
3082 size_t page_size = mono_pagesize ();
3083 size_t max_heap_candidate = 0;
3084 opt = strchr (opt, '=') + 1;
3085 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &max_heap_candidate)) {
3086 max_heap = (max_heap_candidate + page_size - 1) & ~(size_t)(page_size - 1);
3087 if (max_heap != max_heap_candidate)
3088 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Rounding up.", "`max-heap-size` size must be a multiple of %d.", page_size);
3090 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`max-heap-size` must be an integer.");
3094 if (g_str_has_prefix (opt, "soft-heap-limit=")) {
3095 opt = strchr (opt, '=') + 1;
3096 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &soft_limit)) {
3097 if (soft_limit <= 0) {
3098 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be positive.");
3102 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be an integer.");
3108 if (g_str_has_prefix (opt, "nursery-size=")) {
3110 opt = strchr (opt, '=') + 1;
3111 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &val)) {
3112 if ((val & (val - 1))) {
3113 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be a power of two.");
3117 if (val < SGEN_MAX_NURSERY_WASTE) {
3118 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.",
3119 "`nursery-size` must be at least %d bytes.", SGEN_MAX_NURSERY_WASTE);
3123 sgen_nursery_size = val;
3124 sgen_nursery_bits = 0;
3125 while (ONE_P << (++ sgen_nursery_bits) != sgen_nursery_size)
3128 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be an integer.");
3134 if (g_str_has_prefix (opt, "save-target-ratio=")) {
3136 opt = strchr (opt, '=') + 1;
3137 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "save-target-ratio", opt,
3138 SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO, &val)) {
3143 if (g_str_has_prefix (opt, "default-allowance-ratio=")) {
3145 opt = strchr (opt, '=') + 1;
3146 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "default-allowance-ratio", opt,
3147 SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, SGEN_MAX_ALLOWANCE_NURSERY_SIZE_RATIO, &val)) {
3148 allowance_ratio = val;
3153 if (!strcmp (opt, "cementing")) {
3154 cement_enabled = TRUE;
3157 if (!strcmp (opt, "no-cementing")) {
3158 cement_enabled = FALSE;
3162 if (!strcmp (opt, "precleaning")) {
3163 precleaning_enabled = TRUE;
3166 if (!strcmp (opt, "no-precleaning")) {
3167 precleaning_enabled = FALSE;
3171 if (major_collector.handle_gc_param && major_collector.handle_gc_param (opt))
3174 if (sgen_minor_collector.handle_gc_param && sgen_minor_collector.handle_gc_param (opt))
3177 if (sgen_client_handle_gc_param (opt))
3180 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Ignoring.", "Unknown option `%s`.", opt);
3185 fprintf (stderr, "\n%s must be a comma-delimited list of one or more of the following:\n", MONO_GC_PARAMS_NAME);
3186 fprintf (stderr, " max-heap-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
3187 fprintf (stderr, " soft-heap-limit=n (where N is an integer, possibly with a k, m or a g suffix)\n");
3188 fprintf (stderr, " nursery-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
3189 fprintf (stderr, " major=COLLECTOR (where COLLECTOR is `marksweep', `marksweep-conc', `marksweep-par')\n");
3190 fprintf (stderr, " minor=COLLECTOR (where COLLECTOR is `simple' or `split')\n");
3191 fprintf (stderr, " wbarrier=WBARRIER (where WBARRIER is `remset' or `cardtable')\n");
3192 fprintf (stderr, " [no-]cementing\n");
3193 if (major_collector.print_gc_param_usage)
3194 major_collector.print_gc_param_usage ();
3195 if (sgen_minor_collector.print_gc_param_usage)
3196 sgen_minor_collector.print_gc_param_usage ();
3197 sgen_client_print_gc_params_usage ();
3198 fprintf (stderr, " Experimental options:\n");
3199 fprintf (stderr, " save-target-ratio=R (where R must be between %.2f - %.2f).\n", SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO);
3200 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);
3201 fprintf (stderr, "\n");
3203 usage_printed = TRUE;
3208 if (major_collector_opt)
3209 g_free (major_collector_opt);
3211 if (minor_collector_opt)
3212 g_free (minor_collector_opt);
3215 g_free (params_opts);
3219 sgen_pinning_init ();
3220 sgen_cement_init (cement_enabled);
3222 if ((env = g_getenv (MONO_GC_DEBUG_NAME)) || gc_debug_options) {
3223 debug_opts = g_strdup_printf ("%s,%s", gc_debug_options ? gc_debug_options : "", env ? env : "");
3227 gboolean usage_printed = FALSE;
3229 opts = g_strsplit (debug_opts, ",", -1);
3230 for (ptr = opts; ptr && *ptr; ptr ++) {
3232 if (!strcmp (opt, ""))
3234 if (opt [0] >= '0' && opt [0] <= '9') {
3235 gc_debug_level = atoi (opt);
3240 char *rf = g_strdup_printf ("%s.%d", opt, mono_process_current_pid ());
3241 gc_debug_file = fopen (rf, "wb");
3243 gc_debug_file = stderr;
3246 } else if (!strcmp (opt, "print-allowance")) {
3247 debug_print_allowance = TRUE;
3248 } else if (!strcmp (opt, "print-pinning")) {
3249 sgen_pin_stats_enable ();
3250 } else if (!strcmp (opt, "verify-before-allocs")) {
3251 verify_before_allocs = 1;
3252 has_per_allocation_action = TRUE;
3253 } else if (g_str_has_prefix (opt, "verify-before-allocs=")) {
3254 char *arg = strchr (opt, '=') + 1;
3255 verify_before_allocs = atoi (arg);
3256 has_per_allocation_action = TRUE;
3257 } else if (!strcmp (opt, "collect-before-allocs")) {
3258 collect_before_allocs = 1;
3259 has_per_allocation_action = TRUE;
3260 } else if (g_str_has_prefix (opt, "collect-before-allocs=")) {
3261 char *arg = strchr (opt, '=') + 1;
3262 has_per_allocation_action = TRUE;
3263 collect_before_allocs = atoi (arg);
3264 } else if (!strcmp (opt, "verify-before-collections")) {
3265 whole_heap_check_before_collection = TRUE;
3266 } else if (!strcmp (opt, "check-remset-consistency")) {
3267 remset_consistency_checks = TRUE;
3268 nursery_clear_policy = CLEAR_AT_GC;
3269 } else if (!strcmp (opt, "mod-union-consistency-check")) {
3270 if (!major_collector.is_concurrent) {
3271 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`mod-union-consistency-check` only works with concurrent major collector.");
3274 mod_union_consistency_check = TRUE;
3275 } else if (!strcmp (opt, "check-mark-bits")) {
3276 check_mark_bits_after_major_collection = TRUE;
3277 } else if (!strcmp (opt, "check-nursery-pinned")) {
3278 check_nursery_objects_pinned = TRUE;
3279 } else if (!strcmp (opt, "clear-at-gc")) {
3280 nursery_clear_policy = CLEAR_AT_GC;
3281 } else if (!strcmp (opt, "clear-nursery-at-gc")) {
3282 nursery_clear_policy = CLEAR_AT_GC;
3283 } else if (!strcmp (opt, "clear-at-tlab-creation")) {
3284 nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
3285 } else if (!strcmp (opt, "debug-clear-at-tlab-creation")) {
3286 nursery_clear_policy = CLEAR_AT_TLAB_CREATION_DEBUG;
3287 } else if (!strcmp (opt, "check-scan-starts")) {
3288 do_scan_starts_check = TRUE;
3289 } else if (!strcmp (opt, "verify-nursery-at-minor-gc")) {
3290 do_verify_nursery = TRUE;
3291 } else if (!strcmp (opt, "check-concurrent")) {
3292 if (!major_collector.is_concurrent) {
3293 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`check-concurrent` only works with concurrent major collectors.");
3296 nursery_clear_policy = CLEAR_AT_GC;
3297 do_concurrent_checks = TRUE;
3298 } else if (!strcmp (opt, "dump-nursery-at-minor-gc")) {
3299 do_dump_nursery_content = TRUE;
3300 } else if (!strcmp (opt, "disable-minor")) {
3301 disable_minor_collections = TRUE;
3302 } else if (!strcmp (opt, "disable-major")) {
3303 disable_major_collections = TRUE;
3304 } else if (g_str_has_prefix (opt, "heap-dump=")) {
3305 char *filename = strchr (opt, '=') + 1;
3306 nursery_clear_policy = CLEAR_AT_GC;
3307 sgen_debug_enable_heap_dump (filename);
3308 } else if (g_str_has_prefix (opt, "binary-protocol=")) {
3309 char *filename = strchr (opt, '=') + 1;
3310 char *colon = strrchr (filename, ':');
3313 if (!mono_gc_parse_environment_string_extract_number (colon + 1, &limit)) {
3314 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring limit.", "Binary protocol file size limit must be an integer.");
3319 binary_protocol_init (filename, (long long)limit);
3320 } else if (!strcmp (opt, "nursery-canaries")) {
3321 do_verify_nursery = TRUE;
3322 enable_nursery_canaries = TRUE;
3323 } else if (!sgen_client_handle_gc_debug (opt)) {
3324 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "Unknown option `%s`.", opt);
3329 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);
3330 fprintf (stderr, "Valid <option>s are:\n");
3331 fprintf (stderr, " collect-before-allocs[=<n>]\n");
3332 fprintf (stderr, " verify-before-allocs[=<n>]\n");
3333 fprintf (stderr, " check-remset-consistency\n");
3334 fprintf (stderr, " check-mark-bits\n");
3335 fprintf (stderr, " check-nursery-pinned\n");
3336 fprintf (stderr, " verify-before-collections\n");
3337 fprintf (stderr, " verify-nursery-at-minor-gc\n");
3338 fprintf (stderr, " dump-nursery-at-minor-gc\n");
3339 fprintf (stderr, " disable-minor\n");
3340 fprintf (stderr, " disable-major\n");
3341 fprintf (stderr, " check-concurrent\n");
3342 fprintf (stderr, " clear-[nursery-]at-gc\n");
3343 fprintf (stderr, " clear-at-tlab-creation\n");
3344 fprintf (stderr, " debug-clear-at-tlab-creation\n");
3345 fprintf (stderr, " check-scan-starts\n");
3346 fprintf (stderr, " print-allowance\n");
3347 fprintf (stderr, " print-pinning\n");
3348 fprintf (stderr, " heap-dump=<filename>\n");
3349 fprintf (stderr, " binary-protocol=<filename>[:<file-size-limit>]\n");
3350 fprintf (stderr, " nursery-canaries\n");
3351 sgen_client_print_gc_debug_usage ();
3352 fprintf (stderr, "\n");
3354 usage_printed = TRUE;
3361 g_free (debug_opts);
3363 if (check_mark_bits_after_major_collection)
3364 nursery_clear_policy = CLEAR_AT_GC;
3366 if (major_collector.post_param_init)
3367 major_collector.post_param_init (&major_collector);
3369 if (major_collector.needs_thread_pool || sgen_minor_collector.is_parallel) {
3370 int num_workers = 1;
3371 if (major_collector.is_parallel || sgen_minor_collector.is_parallel) {
3372 /* FIXME Detect the number of physical cores, instead of logical */
3373 num_workers = mono_cpu_count () / 2;
3374 if (num_workers < 1)
3377 sgen_workers_init (num_workers, (SgenWorkerCallback) major_collector.worker_init_cb);
3380 sgen_memgov_init (max_heap, soft_limit, debug_print_allowance, allowance_ratio, save_target);
3382 memset (&remset, 0, sizeof (remset));
3384 sgen_card_table_init (&remset);
3386 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");
3390 sgen_init_bridge ();
3394 sgen_gc_initialized ()
3396 return gc_initialized > 0;
3400 sgen_get_nursery_clear_policy (void)
3402 return nursery_clear_policy;
3408 mono_coop_mutex_lock (&gc_mutex);
3412 sgen_gc_unlock (void)
3414 mono_coop_mutex_unlock (&gc_mutex);
3418 sgen_major_collector_iterate_live_block_ranges (sgen_cardtable_block_callback callback)
3420 major_collector.iterate_live_block_ranges (callback);
3424 sgen_major_collector_iterate_block_ranges (sgen_cardtable_block_callback callback)
3426 major_collector.iterate_block_ranges (callback);
3430 sgen_get_major_collector (void)
3432 return &major_collector;
3436 sgen_get_minor_collector (void)
3438 return &sgen_minor_collector;
3442 sgen_get_remset (void)
3448 count_cards (long long *major_total, long long *major_marked, long long *los_total, long long *los_marked)
3450 sgen_get_major_collector ()->count_cards (major_total, major_marked);
3451 sgen_los_count_cards (los_total, los_marked);
3454 static gboolean world_is_stopped = FALSE;
3456 /* LOCKING: assumes the GC lock is held */
3458 sgen_stop_world (int generation)
3460 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3462 SGEN_ASSERT (0, !world_is_stopped, "Why are we stopping a stopped world?");
3464 binary_protocol_world_stopping (generation, sgen_timestamp (), (gpointer) (gsize) mono_native_thread_id_get ());
3466 sgen_client_stop_world (generation);
3468 world_is_stopped = TRUE;
3470 if (binary_protocol_is_heavy_enabled ())
3471 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3472 binary_protocol_world_stopped (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3475 /* LOCKING: assumes the GC lock is held */
3477 sgen_restart_world (int generation)
3479 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3482 SGEN_ASSERT (0, world_is_stopped, "Why are we restarting a running world?");
3484 if (binary_protocol_is_heavy_enabled ())
3485 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3486 binary_protocol_world_restarting (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3488 world_is_stopped = FALSE;
3490 sgen_client_restart_world (generation, &stw_time);
3492 binary_protocol_world_restarted (generation, sgen_timestamp ());
3494 if (sgen_client_bridge_need_processing ())
3495 sgen_client_bridge_processing_finish (generation);
3497 sgen_memgov_collection_end (generation, stw_time);
3501 sgen_is_world_stopped (void)
3503 return world_is_stopped;
3507 sgen_check_whole_heap_stw (void)
3509 sgen_stop_world (0);
3510 sgen_clear_nursery_fragments ();
3511 sgen_check_whole_heap (TRUE);
3512 sgen_restart_world (0);
3516 sgen_timestamp (void)
3518 SGEN_TV_DECLARE (timestamp);
3519 SGEN_TV_GETTIME (timestamp);
3520 return SGEN_TV_ELAPSED (sgen_init_timestamp, timestamp);
3523 #endif /* HAVE_SGEN_GC */