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>
182 #include "mono/sgen/sgen-gc.h"
183 #include "mono/sgen/sgen-cardtable.h"
184 #include "mono/sgen/sgen-protocol.h"
185 #include "mono/sgen/sgen-memory-governor.h"
186 #include "mono/sgen/sgen-hash-table.h"
187 #include "mono/sgen/sgen-pinning.h"
188 #include "mono/sgen/sgen-workers.h"
189 #include "mono/sgen/sgen-client.h"
190 #include "mono/sgen/sgen-pointer-queue.h"
191 #include "mono/sgen/gc-internal-agnostic.h"
192 #include "mono/utils/mono-proclib.h"
193 #include "mono/utils/mono-memory-model.h"
194 #include "mono/utils/hazard-pointer.h"
196 #include <mono/utils/memcheck.h>
197 #include <mono/utils/mono-mmap-internals.h>
198 #include <mono/utils/unlocked.h>
200 #undef pthread_create
202 #undef pthread_detach
205 * ######################################################################
206 * ######## Types and constants used by the GC.
207 * ######################################################################
210 /* 0 means not initialized, 1 is initialized, -1 means in progress */
211 static int gc_initialized = 0;
212 /* If set, check if we need to do something every X allocations */
213 gboolean has_per_allocation_action;
214 /* If set, do a heap check every X allocation */
215 guint32 verify_before_allocs = 0;
216 /* If set, do a minor collection before every X allocation */
217 guint32 collect_before_allocs = 0;
218 /* If set, do a whole heap check before each collection */
219 static gboolean whole_heap_check_before_collection = FALSE;
220 /* If set, do a remset consistency check at various opportunities */
221 static gboolean remset_consistency_checks = FALSE;
222 /* If set, do a mod union consistency check before each finishing collection pause */
223 static gboolean mod_union_consistency_check = FALSE;
224 /* If set, check whether mark bits are consistent after major collections */
225 static gboolean check_mark_bits_after_major_collection = FALSE;
226 /* If set, check that all nursery objects are pinned/not pinned, depending on context */
227 static gboolean check_nursery_objects_pinned = FALSE;
228 /* If set, do a few checks when the concurrent collector is used */
229 static gboolean do_concurrent_checks = FALSE;
230 /* If set, do a plausibility check on the scan_starts before and after
232 static gboolean do_scan_starts_check = FALSE;
234 static gboolean disable_minor_collections = FALSE;
235 static gboolean disable_major_collections = FALSE;
236 static gboolean do_verify_nursery = FALSE;
237 static gboolean do_dump_nursery_content = FALSE;
238 static gboolean enable_nursery_canaries = FALSE;
240 static gboolean precleaning_enabled = TRUE;
241 static gboolean dynamic_nursery = FALSE;
242 static size_t min_nursery_size = 0;
243 static size_t max_nursery_size = 0;
245 #ifdef HEAVY_STATISTICS
246 guint64 stat_objects_alloced_degraded = 0;
247 guint64 stat_bytes_alloced_degraded = 0;
249 guint64 stat_copy_object_called_nursery = 0;
250 guint64 stat_objects_copied_nursery = 0;
251 guint64 stat_copy_object_called_major = 0;
252 guint64 stat_objects_copied_major = 0;
254 guint64 stat_scan_object_called_nursery = 0;
255 guint64 stat_scan_object_called_major = 0;
257 guint64 stat_slots_allocated_in_vain;
259 guint64 stat_nursery_copy_object_failed_from_space = 0;
260 guint64 stat_nursery_copy_object_failed_forwarded = 0;
261 guint64 stat_nursery_copy_object_failed_pinned = 0;
262 guint64 stat_nursery_copy_object_failed_to_space = 0;
264 static guint64 stat_wbarrier_add_to_global_remset = 0;
265 static guint64 stat_wbarrier_arrayref_copy = 0;
266 static guint64 stat_wbarrier_generic_store = 0;
267 static guint64 stat_wbarrier_generic_store_atomic = 0;
268 static guint64 stat_wbarrier_set_root = 0;
271 static guint64 stat_pinned_objects = 0;
273 static guint64 time_minor_pre_collection_fragment_clear = 0;
274 static guint64 time_minor_pinning = 0;
275 static guint64 time_minor_scan_remsets = 0;
276 static guint64 time_minor_scan_major_blocks = 0;
277 static guint64 time_minor_scan_los = 0;
278 static guint64 time_minor_scan_pinned = 0;
279 static guint64 time_minor_scan_roots = 0;
280 static guint64 time_minor_finish_gray_stack = 0;
281 static guint64 time_minor_fragment_creation = 0;
283 static guint64 time_major_pre_collection_fragment_clear = 0;
284 static guint64 time_major_pinning = 0;
285 static guint64 time_major_scan_pinned = 0;
286 static guint64 time_major_scan_roots = 0;
287 static guint64 time_major_scan_mod_union_blocks = 0;
288 static guint64 time_major_scan_mod_union_los = 0;
289 static guint64 time_major_finish_gray_stack = 0;
290 static guint64 time_major_free_bigobjs = 0;
291 static guint64 time_major_los_sweep = 0;
292 static guint64 time_major_sweep = 0;
293 static guint64 time_major_fragment_creation = 0;
295 static guint64 time_max = 0;
297 static int sgen_max_pause_time = SGEN_DEFAULT_MAX_PAUSE_TIME;
298 static float sgen_max_pause_margin = SGEN_DEFAULT_MAX_PAUSE_MARGIN;
300 static SGEN_TV_DECLARE (time_major_conc_collection_start);
301 static SGEN_TV_DECLARE (time_major_conc_collection_end);
303 int gc_debug_level = 0;
305 static char* gc_params_options;
306 static char* gc_debug_options;
310 mono_gc_flush_info (void)
312 fflush (gc_debug_file);
316 #define TV_DECLARE SGEN_TV_DECLARE
317 #define TV_GETTIME SGEN_TV_GETTIME
318 #define TV_ELAPSED SGEN_TV_ELAPSED
320 static SGEN_TV_DECLARE (sgen_init_timestamp);
322 NurseryClearPolicy nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
324 #define object_is_forwarded SGEN_OBJECT_IS_FORWARDED
325 #define object_is_pinned SGEN_OBJECT_IS_PINNED
326 #define pin_object SGEN_PIN_OBJECT
328 #define ptr_in_nursery sgen_ptr_in_nursery
330 #define LOAD_VTABLE SGEN_LOAD_VTABLE
333 nursery_canaries_enabled (void)
335 return enable_nursery_canaries;
338 #define safe_object_get_size sgen_safe_object_get_size
340 #if defined(HAVE_CONC_GC_AS_DEFAULT)
341 /* Use concurrent major on deskstop platforms */
342 #define DEFAULT_MAJOR SGEN_MAJOR_CONCURRENT
344 #define DEFAULT_MAJOR SGEN_MAJOR_SERIAL
350 SGEN_MAJOR_CONCURRENT,
351 SGEN_MAJOR_CONCURRENT_PARALLEL
357 SGEN_MINOR_SIMPLE_PARALLEL,
364 SGEN_MODE_THROUGHPUT,
369 * ######################################################################
370 * ######## Global data.
371 * ######################################################################
373 MonoCoopMutex gc_mutex;
375 #define SCAN_START_SIZE SGEN_SCAN_START_SIZE
377 size_t degraded_mode = 0;
379 static mword bytes_pinned_from_failed_allocation = 0;
381 GCMemSection *nursery_section = NULL;
382 static volatile mword lowest_heap_address = ~(mword)0;
383 static volatile mword highest_heap_address = 0;
385 MonoCoopMutex sgen_interruption_mutex;
387 int current_collection_generation = -1;
388 volatile gboolean concurrent_collection_in_progress = FALSE;
390 /* objects that are ready to be finalized */
391 static SgenPointerQueue fin_ready_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
392 static SgenPointerQueue critical_fin_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
394 /* registered roots: the key to the hash is the root start address */
396 * Different kinds of roots are kept separate to speed up pin_from_roots () for example.
398 SgenHashTable roots_hash [ROOT_TYPE_NUM] = {
399 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
400 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
401 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL)
403 static mword roots_size = 0; /* amount of memory in the root set */
405 /* The size of a TLAB */
406 /* The bigger the value, the less often we have to go to the slow path to allocate a new
407 * one, but the more space is wasted by threads not allocating much memory.
409 * FIXME: Make this self-tuning for each thread.
411 guint32 tlab_size = (1024 * 4);
413 #define MAX_SMALL_OBJ_SIZE SGEN_MAX_SMALL_OBJ_SIZE
415 #define ALLOC_ALIGN SGEN_ALLOC_ALIGN
417 #define ALIGN_UP SGEN_ALIGN_UP
419 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
420 MonoNativeThreadId main_gc_thread = NULL;
423 /*Object was pinned during the current collection*/
424 static mword objects_pinned;
427 * ######################################################################
428 * ######## Macros and function declarations.
429 * ######################################################################
432 /* forward declarations */
433 static void scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx);
435 static void pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx);
436 static void finish_gray_stack (int generation, ScanCopyContext ctx);
439 SgenMajorCollector major_collector;
440 SgenMinorCollector sgen_minor_collector;
442 static SgenRememberedSet remset;
445 * The gray queue a worker job must use. If we're not parallel or
446 * concurrent, we use the main gray queue.
448 static SgenGrayQueue*
449 sgen_workers_get_job_gray_queue (WorkerData *worker_data, SgenGrayQueue *default_gray_queue)
452 return &worker_data->private_gray_queue;
453 SGEN_ASSERT (0, default_gray_queue, "Why don't we have a default gray queue when we're not running in a worker thread?");
454 return default_gray_queue;
458 gray_queue_redirect (SgenGrayQueue *queue)
460 sgen_workers_take_from_queue (current_collection_generation, queue);
464 sgen_scan_area_with_callback (char *start, char *end, IterateObjectCallbackFunc callback, void *data, gboolean allow_flags, gboolean fail_on_canaries)
466 while (start < end) {
470 if (!*(void**)start) {
471 start += sizeof (void*); /* should be ALLOC_ALIGN, really */
476 if (!(obj = (char *)SGEN_OBJECT_IS_FORWARDED (start)))
482 if (!sgen_client_object_is_array_fill ((GCObject*)obj)) {
483 CHECK_CANARY_FOR_OBJECT ((GCObject*)obj, fail_on_canaries);
484 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
485 callback ((GCObject*)obj, size, data);
486 CANARIFY_SIZE (size);
488 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
496 * sgen_add_to_global_remset:
498 * The global remset contains locations which point into newspace after
499 * a minor collection. This can happen if the objects they point to are pinned.
501 * LOCKING: If called from a parallel collector, the global remset
502 * lock must be held. For serial collectors that is not necessary.
505 sgen_add_to_global_remset (gpointer ptr, GCObject *obj)
507 SGEN_ASSERT (5, sgen_ptr_in_nursery (obj), "Target pointer of global remset must be in the nursery");
509 HEAVY_STAT (++stat_wbarrier_add_to_global_remset);
511 if (!major_collector.is_concurrent) {
512 SGEN_ASSERT (5, current_collection_generation != -1, "Global remsets can only be added during collections");
514 if (current_collection_generation == -1)
515 SGEN_ASSERT (5, sgen_concurrent_collection_in_progress (), "Global remsets outside of collection pauses can only be added by the concurrent collector");
518 if (!object_is_pinned (obj))
519 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");
520 else if (sgen_cement_lookup_or_register (obj))
523 remset.record_pointer (ptr);
525 sgen_pin_stats_register_global_remset (obj);
527 SGEN_LOG (8, "Adding global remset for %p", ptr);
528 binary_protocol_global_remset (ptr, obj, (gpointer)SGEN_LOAD_VTABLE (obj));
532 * sgen_drain_gray_stack:
534 * Scan objects in the gray stack until the stack is empty. This should be called
535 * frequently after each object is copied, to achieve better locality and cache
540 sgen_drain_gray_stack (ScanCopyContext ctx)
542 SGEN_ASSERT (0, ctx.ops->drain_gray_stack, "Why do we have a scan/copy context with a missing drain gray stack function?");
544 return ctx.ops->drain_gray_stack (ctx.queue);
548 * Addresses in the pin queue are already sorted. This function finds
549 * the object header for each address and pins the object. The
550 * addresses must be inside the nursery section. The (start of the)
551 * address array is overwritten with the addresses of the actually
552 * pinned objects. Return the number of pinned objects.
555 pin_objects_from_nursery_pin_queue (gboolean do_scan_objects, ScanCopyContext ctx)
557 GCMemSection *section = nursery_section;
558 void **start = sgen_pinning_get_entry (section->pin_queue_first_entry);
559 void **end = sgen_pinning_get_entry (section->pin_queue_last_entry);
560 void *start_nursery = section->data;
561 void *end_nursery = section->end_data;
566 void *pinning_front = start_nursery;
568 void **definitely_pinned = start;
569 ScanObjectFunc scan_func = ctx.ops->scan_object;
570 SgenGrayQueue *queue = ctx.queue;
572 sgen_nursery_allocator_prepare_for_pinning ();
574 while (start < end) {
575 GCObject *obj_to_pin = NULL;
576 size_t obj_to_pin_size = 0;
581 SGEN_ASSERT (0, addr >= start_nursery && addr < end_nursery, "Potential pinning address out of range");
582 SGEN_ASSERT (0, addr >= last, "Pin queue not sorted");
589 SGEN_LOG (5, "Considering pinning addr %p", addr);
590 /* We've already processed everything up to pinning_front. */
591 if (addr < pinning_front) {
597 * Find the closest scan start <= addr. We might search backward in the
598 * scan_starts array because entries might be NULL. In the worst case we
599 * start at start_nursery.
601 idx = ((char*)addr - (char*)section->data) / SCAN_START_SIZE;
602 SGEN_ASSERT (0, idx < section->num_scan_start, "Scan start index out of range");
603 search_start = (void*)section->scan_starts [idx];
604 if (!search_start || search_start > addr) {
607 search_start = section->scan_starts [idx];
608 if (search_start && search_start <= addr)
611 if (!search_start || search_start > addr)
612 search_start = start_nursery;
616 * If the pinning front is closer than the scan start we found, start
617 * searching at the front.
619 if (search_start < pinning_front)
620 search_start = pinning_front;
623 * Now addr should be in an object a short distance from search_start.
625 * search_start must point to zeroed mem or point to an object.
628 size_t obj_size, canarified_obj_size;
631 if (!*(void**)search_start) {
632 search_start = (void*)ALIGN_UP ((mword)search_start + sizeof (gpointer));
633 /* The loop condition makes sure we don't overrun addr. */
637 canarified_obj_size = obj_size = ALIGN_UP (safe_object_get_size ((GCObject*)search_start));
640 * Filler arrays are marked by an invalid sync word. We don't
641 * consider them for pinning. They are not delimited by canaries,
644 if (!sgen_client_object_is_array_fill ((GCObject*)search_start)) {
645 CHECK_CANARY_FOR_OBJECT (search_start, TRUE);
646 CANARIFY_SIZE (canarified_obj_size);
648 if (addr >= search_start && (char*)addr < (char*)search_start + obj_size) {
649 /* This is the object we're looking for. */
650 obj_to_pin = (GCObject*)search_start;
651 obj_to_pin_size = canarified_obj_size;
656 /* Skip to the next object */
657 search_start = (void*)((char*)search_start + canarified_obj_size);
658 } while (search_start <= addr);
660 /* We've searched past the address we were looking for. */
662 pinning_front = search_start;
663 goto next_pin_queue_entry;
667 * We've found an object to pin. It might still be a dummy array, but we
668 * can advance the pinning front in any case.
670 pinning_front = (char*)obj_to_pin + obj_to_pin_size;
673 * If this is a dummy array marking the beginning of a nursery
674 * fragment, we don't pin it.
676 if (sgen_client_object_is_array_fill (obj_to_pin))
677 goto next_pin_queue_entry;
680 * Finally - pin the object!
682 desc = sgen_obj_get_descriptor_safe (obj_to_pin);
683 if (do_scan_objects) {
684 scan_func (obj_to_pin, desc, queue);
686 SGEN_LOG (4, "Pinned object %p, vtable %p (%s), count %d\n",
687 obj_to_pin, *(void**)obj_to_pin, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj_to_pin)), count);
688 binary_protocol_pin (obj_to_pin,
689 (gpointer)LOAD_VTABLE (obj_to_pin),
690 safe_object_get_size (obj_to_pin));
692 pin_object (obj_to_pin);
693 GRAY_OBJECT_ENQUEUE_SERIAL (queue, obj_to_pin, desc);
694 sgen_pin_stats_register_object (obj_to_pin, GENERATION_NURSERY);
695 definitely_pinned [count] = obj_to_pin;
698 if (concurrent_collection_in_progress)
699 sgen_pinning_register_pinned_in_nursery (obj_to_pin);
701 next_pin_queue_entry:
705 sgen_client_nursery_objects_pinned (definitely_pinned, count);
706 stat_pinned_objects += count;
711 pin_objects_in_nursery (gboolean do_scan_objects, ScanCopyContext ctx)
715 if (nursery_section->pin_queue_first_entry == nursery_section->pin_queue_last_entry)
718 reduced_to = pin_objects_from_nursery_pin_queue (do_scan_objects, ctx);
719 nursery_section->pin_queue_last_entry = nursery_section->pin_queue_first_entry + reduced_to;
723 * This function is only ever called (via `collector_pin_object()` in `sgen-copy-object.h`)
724 * when we can't promote an object because we're out of memory.
727 sgen_pin_object (GCObject *object, SgenGrayQueue *queue)
729 SGEN_ASSERT (0, sgen_ptr_in_nursery (object), "We're only supposed to use this for pinning nursery objects when out of memory.");
732 * All pinned objects are assumed to have been staged, so we need to stage as well.
733 * Also, the count of staged objects shows that "late pinning" happened.
735 sgen_pin_stage_ptr (object);
737 SGEN_PIN_OBJECT (object);
738 binary_protocol_pin (object, (gpointer)LOAD_VTABLE (object), safe_object_get_size (object));
741 sgen_pin_stats_register_object (object, GENERATION_NURSERY);
743 GRAY_OBJECT_ENQUEUE_SERIAL (queue, object, sgen_obj_get_descriptor_safe (object));
746 /* Sort the addresses in array in increasing order.
747 * Done using a by-the book heap sort. Which has decent and stable performance, is pretty cache efficient.
750 sgen_sort_addresses (void **array, size_t size)
755 for (i = 1; i < size; ++i) {
758 size_t parent = (child - 1) / 2;
760 if (array [parent] >= array [child])
763 tmp = array [parent];
764 array [parent] = array [child];
771 for (i = size - 1; i > 0; --i) {
774 array [i] = array [0];
780 while (root * 2 + 1 <= end) {
781 size_t child = root * 2 + 1;
783 if (child < end && array [child] < array [child + 1])
785 if (array [root] >= array [child])
789 array [root] = array [child];
798 * Scan the memory between start and end and queue values which could be pointers
799 * to the area between start_nursery and end_nursery for later consideration.
800 * Typically used for thread stacks.
803 sgen_conservatively_pin_objects_from (void **start, void **end, void *start_nursery, void *end_nursery, int pin_type)
807 SGEN_ASSERT (0, ((mword)start & (SIZEOF_VOID_P - 1)) == 0, "Why are we scanning for references in unaligned memory ?");
809 #if defined(VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE) && !defined(_WIN64)
810 VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE (start, (char*)end - (char*)start);
813 while (start < end) {
815 * *start can point to the middle of an object
816 * note: should we handle pointing at the end of an object?
817 * pinning in C# code disallows pointing at the end of an object
818 * but there is some small chance that an optimizing C compiler
819 * may keep the only reference to an object by pointing
820 * at the end of it. We ignore this small chance for now.
821 * Pointers to the end of an object are indistinguishable
822 * from pointers to the start of the next object in memory
823 * so if we allow that we'd need to pin two objects...
824 * We queue the pointer in an array, the
825 * array will then be sorted and uniqued. This way
826 * we can coalesce several pinning pointers and it should
827 * be faster since we'd do a memory scan with increasing
828 * addresses. Note: we can align the address to the allocation
829 * alignment, so the unique process is more effective.
831 mword addr = (mword)*start;
832 addr &= ~(ALLOC_ALIGN - 1);
833 if (addr >= (mword)start_nursery && addr < (mword)end_nursery) {
834 SGEN_LOG (6, "Pinning address %p from %p", (void*)addr, start);
835 sgen_pin_stage_ptr ((void*)addr);
836 binary_protocol_pin_stage (start, (void*)addr);
837 sgen_pin_stats_register_address ((char*)addr, pin_type);
843 SGEN_LOG (7, "found %d potential pinned heap pointers", count);
847 * The first thing we do in a collection is to identify pinned objects.
848 * This function considers all the areas of memory that need to be
849 * conservatively scanned.
852 pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx)
856 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);
857 /* objects pinned from the API are inside these roots */
858 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_PINNED], void **, start_root, RootRecord *, root) {
859 SGEN_LOG (6, "Pinned roots %p-%p", start_root, root->end_root);
860 sgen_conservatively_pin_objects_from (start_root, (void**)root->end_root, start_nursery, end_nursery, PIN_TYPE_OTHER);
861 } SGEN_HASH_TABLE_FOREACH_END;
862 /* now deal with the thread stacks
863 * in the future we should be able to conservatively scan only:
864 * *) the cpu registers
865 * *) the unmanaged stack frames
866 * *) the _last_ managed stack frame
867 * *) pointers slots in managed frames
869 sgen_client_scan_thread_data (start_nursery, end_nursery, FALSE, ctx);
873 single_arg_user_copy_or_mark (GCObject **obj, void *gc_data)
875 ScanCopyContext *ctx = (ScanCopyContext *)gc_data;
876 ctx->ops->copy_or_mark_object (obj, ctx->queue);
880 * The memory area from start_root to end_root contains pointers to objects.
881 * Their position is precisely described by @desc (this means that the pointer
882 * can be either NULL or the pointer to the start of an object).
883 * This functions copies them to to_space updates them.
885 * This function is not thread-safe!
888 precisely_scan_objects_from (void** start_root, void** end_root, char* n_start, char *n_end, SgenDescriptor desc, ScanCopyContext ctx)
890 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
891 ScanPtrFieldFunc scan_field_func = ctx.ops->scan_ptr_field;
892 SgenGrayQueue *queue = ctx.queue;
894 switch (desc & ROOT_DESC_TYPE_MASK) {
895 case ROOT_DESC_BITMAP:
896 desc >>= ROOT_DESC_TYPE_SHIFT;
898 if ((desc & 1) && *start_root) {
899 copy_func ((GCObject**)start_root, queue);
900 SGEN_LOG (9, "Overwrote root at %p with %p", start_root, *start_root);
906 case ROOT_DESC_COMPLEX: {
907 gsize *bitmap_data = (gsize *)sgen_get_complex_descriptor_bitmap (desc);
908 gsize bwords = (*bitmap_data) - 1;
909 void **start_run = start_root;
911 while (bwords-- > 0) {
912 gsize bmap = *bitmap_data++;
913 void **objptr = start_run;
915 if ((bmap & 1) && *objptr) {
916 copy_func ((GCObject**)objptr, queue);
917 SGEN_LOG (9, "Overwrote root at %p with %p", objptr, *objptr);
922 start_run += GC_BITS_PER_WORD;
926 case ROOT_DESC_VECTOR: {
929 for (p = start_root; p < end_root; p++) {
931 scan_field_func (NULL, (GCObject**)p, queue);
935 case ROOT_DESC_USER: {
936 SgenUserRootMarkFunc marker = sgen_get_user_descriptor_func (desc);
937 marker (start_root, single_arg_user_copy_or_mark, &ctx);
940 case ROOT_DESC_RUN_LEN:
941 g_assert_not_reached ();
943 g_assert_not_reached ();
948 reset_heap_boundaries (void)
950 lowest_heap_address = ~(mword)0;
951 highest_heap_address = 0;
955 sgen_update_heap_boundaries (mword low, mword high)
960 old = lowest_heap_address;
963 } while (SGEN_CAS_PTR ((gpointer*)&lowest_heap_address, (gpointer)low, (gpointer)old) != (gpointer)old);
966 old = highest_heap_address;
969 } while (SGEN_CAS_PTR ((gpointer*)&highest_heap_address, (gpointer)high, (gpointer)old) != (gpointer)old);
973 * Allocate and setup the data structures needed to be able to allocate objects
974 * in the nursery. The nursery is stored in nursery_section.
977 alloc_nursery (gboolean dynamic, size_t min_size, size_t max_size)
984 min_size = SGEN_DEFAULT_NURSERY_MIN_SIZE;
986 max_size = SGEN_DEFAULT_NURSERY_MAX_SIZE;
988 SGEN_ASSERT (0, min_size == max_size, "We can't have nursery ranges for static configuration.");
990 min_size = max_size = SGEN_DEFAULT_NURSERY_SIZE;
993 SGEN_ASSERT (0, !nursery_section, "Why are we allocating the nursery twice?");
994 SGEN_LOG (2, "Allocating nursery size: %zu, initial %zu", max_size, min_size);
996 /* FIXME: handle OOM */
997 nursery_section = (GCMemSection *)sgen_alloc_internal (INTERNAL_MEM_SECTION);
999 /* If there isn't enough space even for the nursery we should simply abort. */
1000 g_assert (sgen_memgov_try_alloc_space (max_size, SPACE_NURSERY));
1003 * The nursery section range represents the memory section where objects
1004 * can be found. This is used when iterating for objects in the nursery,
1005 * pinning etc. sgen_nursery_max_size represents the total allocated space
1006 * for the nursery. sgen_nursery_size represents the current size of the
1007 * nursery and it is used for allocation limits, heuristics etc. The
1008 * nursery section is not always identical to the current nursery size
1009 * because it can contain pinned objects from when the nursery was larger.
1011 * sgen_nursery_size <= nursery_section size <= sgen_nursery_max_size
1013 data = (char *)major_collector.alloc_heap (max_size, max_size);
1014 sgen_update_heap_boundaries ((mword)data, (mword)(data + max_size));
1015 nursery_section->data = data;
1016 nursery_section->end_data = data + min_size;
1017 scan_starts = (max_size + SCAN_START_SIZE - 1) / SCAN_START_SIZE;
1018 nursery_section->scan_starts = (char **)sgen_alloc_internal_dynamic (sizeof (char*) * scan_starts, INTERNAL_MEM_SCAN_STARTS, TRUE);
1019 nursery_section->num_scan_start = scan_starts;
1021 sgen_nursery_allocator_set_nursery_bounds (data, min_size, max_size);
1025 mono_gc_get_logfile (void)
1027 return gc_debug_file;
1031 mono_gc_params_set (const char* options)
1033 if (gc_params_options)
1034 g_free (gc_params_options);
1036 gc_params_options = g_strdup (options);
1040 mono_gc_debug_set (const char* options)
1042 if (gc_debug_options)
1043 g_free (gc_debug_options);
1045 gc_debug_options = g_strdup (options);
1049 scan_finalizer_entries (SgenPointerQueue *fin_queue, ScanCopyContext ctx)
1051 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
1052 SgenGrayQueue *queue = ctx.queue;
1055 for (i = 0; i < fin_queue->next_slot; ++i) {
1056 GCObject *obj = (GCObject *)fin_queue->data [i];
1059 SGEN_LOG (5, "Scan of fin ready object: %p (%s)\n", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
1060 copy_func ((GCObject**)&fin_queue->data [i], queue);
1065 generation_name (int generation)
1067 switch (generation) {
1068 case GENERATION_NURSERY: return "nursery";
1069 case GENERATION_OLD: return "old";
1070 default: g_assert_not_reached ();
1075 sgen_generation_name (int generation)
1077 return generation_name (generation);
1081 finish_gray_stack (int generation, ScanCopyContext ctx)
1085 int done_with_ephemerons, ephemeron_rounds = 0;
1086 char *start_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_start () : NULL;
1087 char *end_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_end () : (char*)-1;
1088 SgenGrayQueue *queue = ctx.queue;
1090 binary_protocol_finish_gray_stack_start (sgen_timestamp (), generation);
1092 * We copied all the reachable objects. Now it's the time to copy
1093 * the objects that were not referenced by the roots, but by the copied objects.
1094 * we built a stack of objects pointed to by gray_start: they are
1095 * additional roots and we may add more items as we go.
1096 * We loop until gray_start == gray_objects which means no more objects have
1097 * been added. Note this is iterative: no recursion is involved.
1098 * We need to walk the LO list as well in search of marked big objects
1099 * (use a flag since this is needed only on major collections). We need to loop
1100 * here as well, so keep a counter of marked LO (increasing it in copy_object).
1101 * To achieve better cache locality and cache usage, we drain the gray stack
1102 * frequently, after each object is copied, and just finish the work here.
1104 sgen_drain_gray_stack (ctx);
1106 SGEN_LOG (2, "%s generation done", generation_name (generation));
1109 Reset bridge data, we might have lingering data from a previous collection if this is a major
1110 collection trigged by minor overflow.
1112 We must reset the gathered bridges since their original block might be evacuated due to major
1113 fragmentation in the meanwhile and the bridge code should not have to deal with that.
1115 if (sgen_client_bridge_need_processing ())
1116 sgen_client_bridge_reset_data ();
1119 * Mark all strong toggleref objects. This must be done before we walk ephemerons or finalizers
1120 * to ensure they see the full set of live objects.
1122 sgen_client_mark_togglerefs (start_addr, end_addr, ctx);
1125 * Walk the ephemeron tables marking all values with reachable keys. This must be completely done
1126 * before processing finalizable objects and non-tracking weak links to avoid finalizing/clearing
1127 * objects that are in fact reachable.
1129 done_with_ephemerons = 0;
1131 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1132 sgen_drain_gray_stack (ctx);
1134 } while (!done_with_ephemerons);
1136 if (sgen_client_bridge_need_processing ()) {
1137 /*Make sure the gray stack is empty before we process bridge objects so we get liveness right*/
1138 sgen_drain_gray_stack (ctx);
1139 sgen_collect_bridge_objects (generation, ctx);
1140 if (generation == GENERATION_OLD)
1141 sgen_collect_bridge_objects (GENERATION_NURSERY, ctx);
1144 Do the first bridge step here, as the collector liveness state will become useless after that.
1146 An important optimization is to only proccess the possibly dead part of the object graph and skip
1147 over all live objects as we transitively know everything they point must be alive too.
1149 The above invariant is completely wrong if we let the gray queue be drained and mark/copy everything.
1151 This has the unfortunate side effect of making overflow collections perform the first step twice, but
1152 given we now have heuristics that perform major GC in anticipation of minor overflows this should not
1155 sgen_client_bridge_processing_stw_step ();
1159 Make sure we drain the gray stack before processing disappearing links and finalizers.
1160 If we don't make sure it is empty we might wrongly see a live object as dead.
1162 sgen_drain_gray_stack (ctx);
1165 We must clear weak links that don't track resurrection before processing object ready for
1166 finalization so they can be cleared before that.
1168 sgen_null_link_in_range (generation, ctx, FALSE);
1169 if (generation == GENERATION_OLD)
1170 sgen_null_link_in_range (GENERATION_NURSERY, ctx, FALSE);
1173 /* walk the finalization queue and move also the objects that need to be
1174 * finalized: use the finalized objects as new roots so the objects they depend
1175 * on are also not reclaimed. As with the roots above, only objects in the nursery
1176 * are marked/copied.
1178 sgen_finalize_in_range (generation, ctx);
1179 if (generation == GENERATION_OLD)
1180 sgen_finalize_in_range (GENERATION_NURSERY, ctx);
1181 /* drain the new stack that might have been created */
1182 SGEN_LOG (6, "Precise scan of gray area post fin");
1183 sgen_drain_gray_stack (ctx);
1186 * This must be done again after processing finalizable objects since CWL slots are cleared only after the key is finalized.
1188 done_with_ephemerons = 0;
1190 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1191 sgen_drain_gray_stack (ctx);
1193 } while (!done_with_ephemerons);
1195 sgen_client_clear_unreachable_ephemerons (ctx);
1198 * We clear togglerefs only after all possible chances of revival are done.
1199 * This is semantically more inline with what users expect and it allows for
1200 * user finalizers to correctly interact with TR objects.
1202 sgen_client_clear_togglerefs (start_addr, end_addr, ctx);
1205 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);
1208 * handle disappearing links
1209 * Note we do this after checking the finalization queue because if an object
1210 * survives (at least long enough to be finalized) we don't clear the link.
1211 * This also deals with a possible issue with the monitor reclamation: with the Boehm
1212 * GC a finalized object my lose the monitor because it is cleared before the finalizer is
1215 g_assert (sgen_gray_object_queue_is_empty (queue));
1217 sgen_null_link_in_range (generation, ctx, TRUE);
1218 if (generation == GENERATION_OLD)
1219 sgen_null_link_in_range (GENERATION_NURSERY, ctx, TRUE);
1220 if (sgen_gray_object_queue_is_empty (queue))
1222 sgen_drain_gray_stack (ctx);
1225 g_assert (sgen_gray_object_queue_is_empty (queue));
1227 binary_protocol_finish_gray_stack_end (sgen_timestamp (), generation);
1231 sgen_check_section_scan_starts (GCMemSection *section)
1234 for (i = 0; i < section->num_scan_start; ++i) {
1235 if (section->scan_starts [i]) {
1236 mword size = safe_object_get_size ((GCObject*) section->scan_starts [i]);
1237 SGEN_ASSERT (0, size >= SGEN_CLIENT_MINIMUM_OBJECT_SIZE && size <= MAX_SMALL_OBJ_SIZE, "Weird object size at scan starts.");
1243 check_scan_starts (void)
1245 if (!do_scan_starts_check)
1247 sgen_check_section_scan_starts (nursery_section);
1248 major_collector.check_scan_starts ();
1252 scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx)
1256 SGEN_HASH_TABLE_FOREACH (&roots_hash [root_type], void **, start_root, RootRecord *, root) {
1257 SGEN_LOG (6, "Precise root scan %p-%p (desc: %p)", start_root, root->end_root, (void*)root->root_desc);
1258 precisely_scan_objects_from (start_root, (void**)root->end_root, addr_start, addr_end, root->root_desc, ctx);
1259 } SGEN_HASH_TABLE_FOREACH_END;
1265 static gboolean inited = FALSE;
1270 mono_counters_register ("Collection max time", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME | MONO_COUNTER_MONOTONIC, &time_max);
1272 mono_counters_register ("Minor fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pre_collection_fragment_clear);
1273 mono_counters_register ("Minor pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pinning);
1274 mono_counters_register ("Minor scan remembered set", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_remsets);
1275 mono_counters_register ("Minor scan major blocks", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_major_blocks);
1276 mono_counters_register ("Minor scan los", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_los);
1277 mono_counters_register ("Minor scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_pinned);
1278 mono_counters_register ("Minor scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_roots);
1279 mono_counters_register ("Minor fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_fragment_creation);
1281 mono_counters_register ("Major fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pre_collection_fragment_clear);
1282 mono_counters_register ("Major pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pinning);
1283 mono_counters_register ("Major scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_pinned);
1284 mono_counters_register ("Major scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_roots);
1285 mono_counters_register ("Major scan mod union blocks", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_mod_union_blocks);
1286 mono_counters_register ("Major scan mod union los", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_mod_union_los);
1287 mono_counters_register ("Major finish gray stack", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_finish_gray_stack);
1288 mono_counters_register ("Major free big objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_free_bigobjs);
1289 mono_counters_register ("Major LOS sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_los_sweep);
1290 mono_counters_register ("Major sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_sweep);
1291 mono_counters_register ("Major fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_fragment_creation);
1293 mono_counters_register ("Number of pinned objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_pinned_objects);
1295 #ifdef HEAVY_STATISTICS
1296 mono_counters_register ("WBarrier remember pointer", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_add_to_global_remset);
1297 mono_counters_register ("WBarrier arrayref copy", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_arrayref_copy);
1298 mono_counters_register ("WBarrier generic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store);
1299 mono_counters_register ("WBarrier generic atomic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store_atomic);
1300 mono_counters_register ("WBarrier set root", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_set_root);
1302 mono_counters_register ("# objects allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_alloced_degraded);
1303 mono_counters_register ("bytes allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_bytes_alloced_degraded);
1305 mono_counters_register ("# copy_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_nursery);
1306 mono_counters_register ("# objects copied (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_nursery);
1307 mono_counters_register ("# copy_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_major);
1308 mono_counters_register ("# objects copied (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_major);
1310 mono_counters_register ("# scan_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_nursery);
1311 mono_counters_register ("# scan_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_major);
1313 mono_counters_register ("Slots allocated in vain", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_slots_allocated_in_vain);
1315 mono_counters_register ("# nursery copy_object() failed from space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_from_space);
1316 mono_counters_register ("# nursery copy_object() failed forwarded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_forwarded);
1317 mono_counters_register ("# nursery copy_object() failed pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_pinned);
1318 mono_counters_register ("# nursery copy_object() failed to space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_to_space);
1320 sgen_nursery_allocator_init_heavy_stats ();
1328 reset_pinned_from_failed_allocation (void)
1330 bytes_pinned_from_failed_allocation = 0;
1334 sgen_set_pinned_from_failed_allocation (mword objsize)
1336 bytes_pinned_from_failed_allocation += objsize;
1340 sgen_collection_is_concurrent (void)
1342 switch (current_collection_generation) {
1343 case GENERATION_NURSERY:
1345 case GENERATION_OLD:
1346 return concurrent_collection_in_progress;
1348 g_error ("Invalid current generation %d", current_collection_generation);
1354 sgen_concurrent_collection_in_progress (void)
1356 return concurrent_collection_in_progress;
1360 SgenThreadPoolJob job;
1361 SgenObjectOperations *ops;
1362 SgenGrayQueue *gc_thread_gray_queue;
1367 int job_index, job_split_count;
1371 static ScanCopyContext
1372 scan_copy_context_for_scan_job (void *worker_data_untyped, ScanJob *job)
1374 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1378 * For jobs enqueued on workers we set the ops at job runtime in order
1379 * to be able to profit from on the fly optimized object ops or other
1380 * object ops changes, like forced concurrent finish.
1382 SGEN_ASSERT (0, sgen_workers_is_worker_thread (mono_native_thread_id_get ()), "We need a context for the scan job");
1383 job->ops = sgen_workers_get_idle_func_object_ops (worker_data);
1386 return CONTEXT_FROM_OBJECT_OPERATIONS (job->ops, sgen_workers_get_job_gray_queue (worker_data, job->gc_thread_gray_queue));
1394 } ScanFromRegisteredRootsJob;
1397 job_scan_from_registered_roots (void *worker_data_untyped, SgenThreadPoolJob *job)
1399 ScanFromRegisteredRootsJob *job_data = (ScanFromRegisteredRootsJob*)job;
1400 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1402 scan_from_registered_roots (job_data->heap_start, job_data->heap_end, job_data->root_type, ctx);
1409 } ScanThreadDataJob;
1412 job_scan_thread_data (void *worker_data_untyped, SgenThreadPoolJob *job)
1414 ScanThreadDataJob *job_data = (ScanThreadDataJob*)job;
1415 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1417 sgen_client_scan_thread_data (job_data->heap_start, job_data->heap_end, TRUE, ctx);
1422 SgenPointerQueue *queue;
1423 } ScanFinalizerEntriesJob;
1426 job_scan_finalizer_entries (void *worker_data_untyped, SgenThreadPoolJob *job)
1428 ScanFinalizerEntriesJob *job_data = (ScanFinalizerEntriesJob*)job;
1429 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1431 scan_finalizer_entries (job_data->queue, ctx);
1435 job_scan_wbroots (void *worker_data_untyped, SgenThreadPoolJob *job)
1437 ScanJob *job_data = (ScanJob*)job;
1438 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, job_data);
1440 sgen_wbroots_scan_card_table (ctx);
1444 job_scan_major_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1446 SGEN_TV_DECLARE (atv);
1447 SGEN_TV_DECLARE (btv);
1448 ParallelScanJob *job_data = (ParallelScanJob*)job;
1449 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1451 SGEN_TV_GETTIME (atv);
1452 major_collector.scan_card_table (CARDTABLE_SCAN_GLOBAL, ctx, job_data->job_index, job_data->job_split_count, job_data->data);
1453 SGEN_TV_GETTIME (btv);
1454 time_minor_scan_major_blocks += SGEN_TV_ELAPSED (atv, btv);
1456 if (worker_data_untyped)
1457 ((WorkerData*)worker_data_untyped)->major_scan_time += SGEN_TV_ELAPSED (atv, btv);
1461 job_scan_los_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1463 SGEN_TV_DECLARE (atv);
1464 SGEN_TV_DECLARE (btv);
1465 ParallelScanJob *job_data = (ParallelScanJob*)job;
1466 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1468 SGEN_TV_GETTIME (atv);
1469 sgen_los_scan_card_table (CARDTABLE_SCAN_GLOBAL, ctx, job_data->job_index, job_data->job_split_count);
1470 SGEN_TV_GETTIME (btv);
1471 time_minor_scan_los += SGEN_TV_ELAPSED (atv, btv);
1473 if (worker_data_untyped)
1474 ((WorkerData*)worker_data_untyped)->los_scan_time += SGEN_TV_ELAPSED (atv, btv);
1478 job_scan_major_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1480 SGEN_TV_DECLARE (atv);
1481 SGEN_TV_DECLARE (btv);
1482 ParallelScanJob *job_data = (ParallelScanJob*)job;
1483 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1485 g_assert (concurrent_collection_in_progress);
1486 SGEN_TV_GETTIME (atv);
1487 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx, job_data->job_index, job_data->job_split_count, job_data->data);
1488 SGEN_TV_GETTIME (btv);
1489 time_major_scan_mod_union_blocks += SGEN_TV_ELAPSED (atv, btv);
1491 if (worker_data_untyped)
1492 ((WorkerData*)worker_data_untyped)->major_scan_time += SGEN_TV_ELAPSED (atv, btv);
1496 job_scan_los_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1498 SGEN_TV_DECLARE (atv);
1499 SGEN_TV_DECLARE (btv);
1500 ParallelScanJob *job_data = (ParallelScanJob*)job;
1501 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1503 g_assert (concurrent_collection_in_progress);
1504 SGEN_TV_GETTIME (atv);
1505 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx, job_data->job_index, job_data->job_split_count);
1506 SGEN_TV_GETTIME (btv);
1507 time_major_scan_mod_union_los += SGEN_TV_ELAPSED (atv, btv);
1509 if (worker_data_untyped)
1510 ((WorkerData*)worker_data_untyped)->los_scan_time += SGEN_TV_ELAPSED (atv, btv);
1514 job_major_mod_union_preclean (void *worker_data_untyped, SgenThreadPoolJob *job)
1516 SGEN_TV_DECLARE (atv);
1517 SGEN_TV_DECLARE (btv);
1518 ParallelScanJob *job_data = (ParallelScanJob*)job;
1519 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1521 g_assert (concurrent_collection_in_progress);
1522 SGEN_TV_GETTIME (atv);
1523 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx, job_data->job_index, job_data->job_split_count, job_data->data);
1524 SGEN_TV_GETTIME (btv);
1526 g_assert (worker_data_untyped);
1527 ((WorkerData*)worker_data_untyped)->major_scan_time += SGEN_TV_ELAPSED (atv, btv);
1531 job_los_mod_union_preclean (void *worker_data_untyped, SgenThreadPoolJob *job)
1533 SGEN_TV_DECLARE (atv);
1534 SGEN_TV_DECLARE (btv);
1535 ParallelScanJob *job_data = (ParallelScanJob*)job;
1536 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1538 g_assert (concurrent_collection_in_progress);
1539 SGEN_TV_GETTIME (atv);
1540 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx, job_data->job_index, job_data->job_split_count);
1541 SGEN_TV_GETTIME (btv);
1543 g_assert (worker_data_untyped);
1544 ((WorkerData*)worker_data_untyped)->los_scan_time += SGEN_TV_ELAPSED (atv, btv);
1548 job_scan_last_pinned (void *worker_data_untyped, SgenThreadPoolJob *job)
1550 ScanJob *job_data = (ScanJob*)job;
1551 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, job_data);
1553 g_assert (concurrent_collection_in_progress);
1555 sgen_scan_pin_queue_objects (ctx);
1559 workers_finish_callback (void)
1561 ParallelScanJob *psj;
1563 size_t num_major_sections = major_collector.get_num_major_sections ();
1564 int split_count = sgen_workers_get_job_split_count (GENERATION_OLD);
1566 /* Mod union preclean jobs */
1567 for (i = 0; i < split_count; i++) {
1568 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("preclean major mod union cardtable", job_major_mod_union_preclean, sizeof (ParallelScanJob));
1569 psj->scan_job.gc_thread_gray_queue = NULL;
1571 psj->job_split_count = split_count;
1572 psj->data = num_major_sections / split_count;
1573 sgen_workers_enqueue_job (GENERATION_OLD, &psj->scan_job.job, TRUE);
1576 for (i = 0; i < split_count; i++) {
1577 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("preclean los mod union cardtable", job_los_mod_union_preclean, sizeof (ParallelScanJob));
1578 psj->scan_job.gc_thread_gray_queue = NULL;
1580 psj->job_split_count = split_count;
1581 sgen_workers_enqueue_job (GENERATION_OLD, &psj->scan_job.job, TRUE);
1584 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan last pinned", job_scan_last_pinned, sizeof (ScanJob));
1585 sj->gc_thread_gray_queue = NULL;
1586 sgen_workers_enqueue_job (GENERATION_OLD, &sj->job, TRUE);
1590 init_gray_queue (SgenGrayQueue *gc_thread_gray_queue)
1592 sgen_gray_object_queue_init (gc_thread_gray_queue, NULL, TRUE);
1596 enqueue_scan_remembered_set_jobs (SgenGrayQueue *gc_thread_gray_queue, SgenObjectOperations *ops, gboolean enqueue)
1598 int i, split_count = sgen_workers_get_job_split_count (GENERATION_NURSERY);
1599 size_t num_major_sections = major_collector.get_num_major_sections ();
1602 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan wbroots", job_scan_wbroots, sizeof (ScanJob));
1604 sj->gc_thread_gray_queue = gc_thread_gray_queue;
1605 sgen_workers_enqueue_job (GENERATION_NURSERY, &sj->job, enqueue);
1607 for (i = 0; i < split_count; i++) {
1608 ParallelScanJob *psj;
1610 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("scan major remsets", job_scan_major_card_table, sizeof (ParallelScanJob));
1611 psj->scan_job.ops = ops;
1612 psj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1614 psj->job_split_count = split_count;
1615 psj->data = num_major_sections / split_count;
1616 sgen_workers_enqueue_job (GENERATION_NURSERY, &psj->scan_job.job, enqueue);
1618 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("scan LOS remsets", job_scan_los_card_table, sizeof (ParallelScanJob));
1619 psj->scan_job.ops = ops;
1620 psj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1622 psj->job_split_count = split_count;
1623 sgen_workers_enqueue_job (GENERATION_NURSERY, &psj->scan_job.job, enqueue);
1628 enqueue_scan_from_roots_jobs (SgenGrayQueue *gc_thread_gray_queue, char *heap_start, char *heap_end, SgenObjectOperations *ops, gboolean enqueue)
1630 ScanFromRegisteredRootsJob *scrrj;
1631 ScanThreadDataJob *stdj;
1632 ScanFinalizerEntriesJob *sfej;
1634 /* registered roots, this includes static fields */
1636 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots normal", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1637 scrrj->scan_job.ops = ops;
1638 scrrj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1639 scrrj->heap_start = heap_start;
1640 scrrj->heap_end = heap_end;
1641 scrrj->root_type = ROOT_TYPE_NORMAL;
1642 sgen_workers_enqueue_job (current_collection_generation, &scrrj->scan_job.job, enqueue);
1644 if (current_collection_generation == GENERATION_OLD) {
1645 /* During minors we scan the cardtable for these roots instead */
1646 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots wbarrier", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1647 scrrj->scan_job.ops = ops;
1648 scrrj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1649 scrrj->heap_start = heap_start;
1650 scrrj->heap_end = heap_end;
1651 scrrj->root_type = ROOT_TYPE_WBARRIER;
1652 sgen_workers_enqueue_job (current_collection_generation, &scrrj->scan_job.job, enqueue);
1657 stdj = (ScanThreadDataJob*)sgen_thread_pool_job_alloc ("scan thread data", job_scan_thread_data, sizeof (ScanThreadDataJob));
1658 stdj->scan_job.ops = ops;
1659 stdj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1660 stdj->heap_start = heap_start;
1661 stdj->heap_end = heap_end;
1662 sgen_workers_enqueue_job (current_collection_generation, &stdj->scan_job.job, enqueue);
1664 /* Scan the list of objects ready for finalization. */
1666 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1667 sfej->scan_job.ops = ops;
1668 sfej->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1669 sfej->queue = &fin_ready_queue;
1670 sgen_workers_enqueue_job (current_collection_generation, &sfej->scan_job.job, enqueue);
1672 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan critical finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1673 sfej->scan_job.ops = ops;
1674 sfej->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1675 sfej->queue = &critical_fin_queue;
1676 sgen_workers_enqueue_job (current_collection_generation, &sfej->scan_job.job, enqueue);
1680 * Perform a nursery collection.
1682 * Return whether any objects were late-pinned due to being out of memory.
1685 collect_nursery (const char *reason, gboolean is_overflow, SgenGrayQueue *unpin_queue)
1687 gboolean needs_major, is_parallel = FALSE;
1688 mword fragment_total;
1689 SgenGrayQueue gc_thread_gray_queue;
1690 SgenObjectOperations *object_ops_nopar, *object_ops_par = NULL;
1691 ScanCopyContext ctx;
1694 SGEN_TV_DECLARE (last_minor_collection_start_tv);
1695 SGEN_TV_DECLARE (last_minor_collection_end_tv);
1696 guint64 major_scan_start = time_minor_scan_major_blocks;
1697 guint64 los_scan_start = time_minor_scan_los;
1698 guint64 finish_gray_start = time_minor_finish_gray_stack;
1700 if (disable_minor_collections)
1703 TV_GETTIME (last_minor_collection_start_tv);
1704 atv = last_minor_collection_start_tv;
1706 binary_protocol_collection_begin (InterlockedRead (&gc_stats.minor_gc_count), GENERATION_NURSERY);
1708 object_ops_nopar = sgen_concurrent_collection_in_progress ()
1709 ? &sgen_minor_collector.serial_ops_with_concurrent_major
1710 : &sgen_minor_collector.serial_ops;
1711 if (sgen_minor_collector.is_parallel && sgen_nursery_size >= SGEN_PARALLEL_MINOR_MIN_NURSERY_SIZE) {
1712 object_ops_par = sgen_concurrent_collection_in_progress ()
1713 ? &sgen_minor_collector.parallel_ops_with_concurrent_major
1714 : &sgen_minor_collector.parallel_ops;
1718 if (do_verify_nursery || do_dump_nursery_content)
1719 sgen_debug_verify_nursery (do_dump_nursery_content);
1721 current_collection_generation = GENERATION_NURSERY;
1723 SGEN_ASSERT (0, !sgen_collection_is_concurrent (), "Why is the nursery collection concurrent?");
1725 reset_pinned_from_failed_allocation ();
1727 check_scan_starts ();
1729 sgen_nursery_alloc_prepare_for_minor ();
1734 SGEN_LOG (1, "Start nursery collection %" G_GINT32_FORMAT " %p-%p, size: %d", InterlockedRead (&gc_stats.minor_gc_count), nursery_section->data, nursery_section->end_data, (int)(nursery_section->end_data - nursery_section->data));
1736 /* world must be stopped already */
1738 time_minor_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1740 sgen_client_pre_collection_checks ();
1742 major_collector.start_nursery_collection ();
1744 sgen_memgov_minor_collection_start ();
1746 init_gray_queue (&gc_thread_gray_queue);
1747 ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops_nopar, &gc_thread_gray_queue);
1749 InterlockedIncrement (&gc_stats.minor_gc_count);
1751 sgen_process_fin_stage_entries ();
1753 /* pin from pinned handles */
1754 sgen_init_pinning ();
1755 if (concurrent_collection_in_progress)
1756 sgen_init_pinning_for_conc ();
1757 sgen_client_binary_protocol_mark_start (GENERATION_NURSERY);
1758 pin_from_roots (nursery_section->data, nursery_section->end_data, ctx);
1759 /* pin cemented objects */
1760 sgen_pin_cemented_objects ();
1761 /* identify pinned objects */
1762 sgen_optimize_pin_queue ();
1763 sgen_pinning_setup_section (nursery_section);
1765 pin_objects_in_nursery (FALSE, ctx);
1766 sgen_pinning_trim_queue_to_section (nursery_section);
1767 if (concurrent_collection_in_progress)
1768 sgen_finish_pinning_for_conc ();
1770 if (remset_consistency_checks)
1771 sgen_check_remset_consistency ();
1773 if (whole_heap_check_before_collection) {
1774 sgen_clear_nursery_fragments ();
1775 sgen_check_whole_heap (FALSE);
1779 time_minor_pinning += TV_ELAPSED (btv, atv);
1780 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), (long long)TV_ELAPSED (btv, atv));
1781 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1783 remset.start_scan_remsets ();
1785 enqueue_scan_remembered_set_jobs (&gc_thread_gray_queue, is_parallel ? NULL : object_ops_nopar, is_parallel);
1787 /* we don't have complete write barrier yet, so we scan all the old generation sections */
1789 time_minor_scan_remsets += TV_ELAPSED (atv, btv);
1790 SGEN_LOG (2, "Old generation scan: %lld usecs", (long long)TV_ELAPSED (atv, btv));
1792 sgen_pin_stats_report ();
1794 /* FIXME: Why do we do this at this specific, seemingly random, point? */
1795 sgen_client_collecting_minor (&fin_ready_queue, &critical_fin_queue);
1798 time_minor_scan_pinned += TV_ELAPSED (btv, atv);
1800 enqueue_scan_from_roots_jobs (&gc_thread_gray_queue, nursery_section->data, nursery_section->end_data, is_parallel ? NULL : object_ops_nopar, is_parallel);
1803 gray_queue_redirect (&gc_thread_gray_queue);
1804 sgen_workers_start_all_workers (GENERATION_NURSERY, object_ops_nopar, object_ops_par, NULL);
1805 sgen_workers_join (GENERATION_NURSERY);
1809 time_minor_scan_roots += TV_ELAPSED (atv, btv);
1811 finish_gray_stack (GENERATION_NURSERY, ctx);
1814 time_minor_finish_gray_stack += TV_ELAPSED (btv, atv);
1815 sgen_client_binary_protocol_mark_end (GENERATION_NURSERY);
1817 if (objects_pinned) {
1818 sgen_optimize_pin_queue ();
1819 sgen_pinning_setup_section (nursery_section);
1823 * This is the latest point at which we can do this check, because
1824 * sgen_build_nursery_fragments() unpins nursery objects again.
1826 if (remset_consistency_checks)
1827 sgen_check_remset_consistency ();
1830 if (sgen_max_pause_time) {
1834 duration = (int)(TV_ELAPSED (last_minor_collection_start_tv, btv) / 10000);
1835 if (duration > (sgen_max_pause_time * sgen_max_pause_margin))
1836 sgen_resize_nursery (TRUE);
1838 sgen_resize_nursery (FALSE);
1840 sgen_resize_nursery (FALSE);
1843 /* walk the pin_queue, build up the fragment list of free memory, unmark
1844 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1847 sgen_client_binary_protocol_reclaim_start (GENERATION_NURSERY);
1848 fragment_total = sgen_build_nursery_fragments (nursery_section, unpin_queue);
1849 if (!fragment_total)
1852 /* Clear TLABs for all threads */
1853 sgen_clear_tlabs ();
1855 sgen_client_binary_protocol_reclaim_end (GENERATION_NURSERY);
1857 time_minor_fragment_creation += TV_ELAPSED (atv, btv);
1858 SGEN_LOG (2, "Fragment creation: %lld usecs, %lu bytes available", (long long)TV_ELAPSED (atv, btv), (unsigned long)fragment_total);
1860 if (remset_consistency_checks)
1861 sgen_check_major_refs ();
1863 major_collector.finish_nursery_collection ();
1865 TV_GETTIME (last_minor_collection_end_tv);
1866 UnlockedAdd64 (&gc_stats.minor_gc_time, TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv));
1868 sgen_debug_dump_heap ("minor", InterlockedRead (&gc_stats.minor_gc_count) - 1, NULL);
1870 /* prepare the pin queue for the next collection */
1871 sgen_finish_pinning ();
1872 if (sgen_have_pending_finalizers ()) {
1873 SGEN_LOG (4, "Finalizer-thread wakeup");
1874 sgen_client_finalize_notify ();
1876 sgen_pin_stats_reset ();
1877 /* clear cemented hash */
1878 sgen_cement_clear_below_threshold ();
1880 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
1882 check_scan_starts ();
1884 binary_protocol_flush_buffers (FALSE);
1886 sgen_memgov_minor_collection_end (reason, is_overflow);
1888 /*objects are late pinned because of lack of memory, so a major is a good call*/
1889 needs_major = objects_pinned > 0;
1890 current_collection_generation = -1;
1894 binary_protocol_collection_end_stats (0, 0, time_minor_finish_gray_stack - finish_gray_start);
1896 binary_protocol_collection_end_stats (
1897 time_minor_scan_major_blocks - major_scan_start,
1898 time_minor_scan_los - los_scan_start,
1899 time_minor_finish_gray_stack - finish_gray_start);
1901 binary_protocol_collection_end (InterlockedRead (&gc_stats.minor_gc_count) - 1, GENERATION_NURSERY, 0, 0);
1903 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1904 sgen_check_nursery_objects_pinned (unpin_queue != NULL);
1910 COPY_OR_MARK_FROM_ROOTS_SERIAL,
1911 COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT,
1912 COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT
1913 } CopyOrMarkFromRootsMode;
1916 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)
1921 /* FIXME: only use these values for the precise scan
1922 * note that to_space pointers should be excluded anyway...
1924 char *heap_start = NULL;
1925 char *heap_end = (char*)-1;
1926 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops_nopar, gc_thread_gray_queue);
1927 gboolean concurrent = mode != COPY_OR_MARK_FROM_ROOTS_SERIAL;
1929 SGEN_ASSERT (0, !!concurrent == !!concurrent_collection_in_progress, "We've been called with the wrong mode.");
1931 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1932 /*This cleans up unused fragments */
1933 sgen_nursery_allocator_prepare_for_pinning ();
1935 if (do_concurrent_checks)
1936 sgen_debug_check_nursery_is_clean ();
1938 /* The concurrent collector doesn't touch the nursery. */
1939 sgen_nursery_alloc_prepare_for_major ();
1944 /* Pinning depends on this */
1945 sgen_clear_nursery_fragments ();
1947 if (whole_heap_check_before_collection)
1948 sgen_check_whole_heap (TRUE);
1951 time_major_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1955 sgen_client_pre_collection_checks ();
1957 if (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1958 /* Remsets are not useful for a major collection */
1959 remset.clear_cards ();
1962 sgen_process_fin_stage_entries ();
1965 sgen_init_pinning ();
1966 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT)
1967 sgen_init_pinning_for_conc ();
1968 SGEN_LOG (6, "Collecting pinned addresses");
1969 pin_from_roots ((void*)lowest_heap_address, (void*)highest_heap_address, ctx);
1970 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1971 /* Pin cemented objects that were forced */
1972 sgen_pin_cemented_objects ();
1974 sgen_optimize_pin_queue ();
1975 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1977 * Cemented objects that are in the pinned list will be marked. When
1978 * marking concurrently we won't mark mod-union cards for these objects.
1979 * Instead they will remain cemented until the next major collection,
1980 * when we will recheck if they are still pinned in the roots.
1982 sgen_cement_force_pinned ();
1985 sgen_client_collecting_major_1 ();
1988 * pin_queue now contains all candidate pointers, sorted and
1989 * uniqued. We must do two passes now to figure out which
1990 * objects are pinned.
1992 * The first is to find within the pin_queue the area for each
1993 * section. This requires that the pin_queue be sorted. We
1994 * also process the LOS objects and pinned chunks here.
1996 * The second, destructive, pass is to reduce the section
1997 * areas to pointers to the actually pinned objects.
1999 SGEN_LOG (6, "Pinning from sections");
2000 /* first pass for the sections */
2001 sgen_find_section_pin_queue_start_end (nursery_section);
2002 /* identify possible pointers to the insize of large objects */
2003 SGEN_LOG (6, "Pinning from large objects");
2004 for (bigobj = los_object_list; bigobj; bigobj = bigobj->next) {
2006 if (sgen_find_optimized_pin_queue_area ((char*)bigobj->data, (char*)bigobj->data + sgen_los_object_size (bigobj), &dummy, &dummy)) {
2007 binary_protocol_pin (bigobj->data, (gpointer)LOAD_VTABLE (bigobj->data), safe_object_get_size (bigobj->data));
2009 if (sgen_los_object_is_pinned (bigobj->data)) {
2010 SGEN_ASSERT (0, mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, "LOS objects can only be pinned here after concurrent marking.");
2013 sgen_los_pin_object (bigobj->data);
2014 if (SGEN_OBJECT_HAS_REFERENCES (bigobj->data))
2015 GRAY_OBJECT_ENQUEUE_SERIAL (gc_thread_gray_queue, bigobj->data, sgen_obj_get_descriptor ((GCObject*)bigobj->data));
2016 sgen_pin_stats_register_object (bigobj->data, GENERATION_OLD);
2017 SGEN_LOG (6, "Marked large object %p (%s) size: %lu from roots", bigobj->data,
2018 sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (bigobj->data)),
2019 (unsigned long)sgen_los_object_size (bigobj));
2021 sgen_client_pinned_los_object (bigobj->data);
2025 pin_objects_in_nursery (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT, ctx);
2026 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
2027 sgen_check_nursery_objects_pinned (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
2029 major_collector.pin_objects (gc_thread_gray_queue);
2030 if (old_next_pin_slot)
2031 *old_next_pin_slot = sgen_get_pinned_count ();
2034 time_major_pinning += TV_ELAPSED (atv, btv);
2035 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), (long long)TV_ELAPSED (atv, btv));
2036 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
2038 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT)
2039 sgen_finish_pinning_for_conc ();
2041 major_collector.init_to_space ();
2043 SGEN_ASSERT (0, sgen_workers_all_done (), "Why are the workers not done when we start or finish a major collection?");
2044 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
2045 if (object_ops_par != NULL)
2046 sgen_workers_set_num_active_workers (GENERATION_OLD, 0);
2047 if (object_ops_par == NULL && sgen_workers_have_idle_work (GENERATION_OLD)) {
2049 * We force the finish of the worker with the new object ops context
2050 * which can also do copying. We need to have finished pinning. On the
2051 * parallel collector, there is no need to drain the private queues
2052 * here, since we can do it as part of the finishing work, achieving
2053 * better work distribution.
2055 sgen_workers_start_all_workers (GENERATION_OLD, object_ops_nopar, object_ops_par, NULL);
2057 sgen_workers_join (GENERATION_OLD);
2061 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
2062 main_gc_thread = mono_native_thread_self ();
2065 sgen_client_collecting_major_2 ();
2068 time_major_scan_pinned += TV_ELAPSED (btv, atv);
2070 sgen_client_collecting_major_3 (&fin_ready_queue, &critical_fin_queue);
2072 enqueue_scan_from_roots_jobs (gc_thread_gray_queue, heap_start, heap_end, object_ops_nopar, FALSE);
2075 time_major_scan_roots += TV_ELAPSED (atv, btv);
2078 * We start the concurrent worker after pinning and after we scanned the roots
2079 * in order to make sure that the worker does not finish before handling all
2082 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
2083 sgen_workers_set_num_active_workers (GENERATION_OLD, 1);
2084 gray_queue_redirect (gc_thread_gray_queue);
2085 if (precleaning_enabled) {
2086 sgen_workers_start_all_workers (GENERATION_OLD, object_ops_nopar, object_ops_par, workers_finish_callback);
2088 sgen_workers_start_all_workers (GENERATION_OLD, object_ops_nopar, object_ops_par, NULL);
2092 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
2093 int i, split_count = sgen_workers_get_job_split_count (GENERATION_OLD);
2094 size_t num_major_sections = major_collector.get_num_major_sections ();
2095 gboolean parallel = object_ops_par != NULL;
2097 /* If we're not parallel we finish the collection on the gc thread */
2099 gray_queue_redirect (gc_thread_gray_queue);
2101 /* Mod union card table */
2102 for (i = 0; i < split_count; i++) {
2103 ParallelScanJob *psj;
2105 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("scan mod union cardtable", job_scan_major_mod_union_card_table, sizeof (ParallelScanJob));
2106 psj->scan_job.ops = parallel ? NULL : object_ops_nopar;
2107 psj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
2109 psj->job_split_count = split_count;
2110 psj->data = num_major_sections / split_count;
2111 sgen_workers_enqueue_job (GENERATION_OLD, &psj->scan_job.job, parallel);
2113 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("scan LOS mod union cardtable", job_scan_los_mod_union_card_table, sizeof (ParallelScanJob));
2114 psj->scan_job.ops = parallel ? NULL : object_ops_nopar;
2115 psj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
2117 psj->job_split_count = split_count;
2118 sgen_workers_enqueue_job (GENERATION_OLD, &psj->scan_job.job, parallel);
2123 * If we enqueue a job while workers are running we need to sgen_workers_ensure_awake
2124 * in order to make sure that we are running the idle func and draining all worker
2125 * gray queues. The operation of starting workers implies this, so we start them after
2126 * in order to avoid doing this operation twice. The workers will drain the main gray
2127 * stack that contained roots and pinned objects and also scan the mod union card
2130 sgen_workers_start_all_workers (GENERATION_OLD, object_ops_nopar, object_ops_par, NULL);
2131 sgen_workers_join (GENERATION_OLD);
2135 sgen_pin_stats_report ();
2137 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
2138 sgen_finish_pinning ();
2140 sgen_pin_stats_reset ();
2142 if (do_concurrent_checks)
2143 sgen_debug_check_nursery_is_clean ();
2148 major_start_collection (SgenGrayQueue *gc_thread_gray_queue, const char *reason, gboolean concurrent, size_t *old_next_pin_slot)
2150 SgenObjectOperations *object_ops_nopar, *object_ops_par = NULL;
2152 binary_protocol_collection_begin (InterlockedRead (&gc_stats.major_gc_count), GENERATION_OLD);
2154 current_collection_generation = GENERATION_OLD;
2156 sgen_workers_assert_gray_queue_is_empty (GENERATION_OLD);
2159 sgen_cement_reset ();
2162 g_assert (major_collector.is_concurrent);
2163 concurrent_collection_in_progress = TRUE;
2165 object_ops_nopar = &major_collector.major_ops_concurrent_start;
2166 if (major_collector.is_parallel)
2167 object_ops_par = &major_collector.major_ops_conc_par_start;
2170 object_ops_nopar = &major_collector.major_ops_serial;
2173 reset_pinned_from_failed_allocation ();
2175 sgen_memgov_major_collection_start (concurrent, reason);
2177 //count_ref_nonref_objs ();
2178 //consistency_check ();
2180 check_scan_starts ();
2183 SGEN_LOG (1, "Start major collection %" G_GINT32_FORMAT, InterlockedRead (&gc_stats.major_gc_count));
2184 InterlockedIncrement (&gc_stats.major_gc_count);
2186 if (major_collector.start_major_collection)
2187 major_collector.start_major_collection ();
2189 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);
2193 major_finish_collection (SgenGrayQueue *gc_thread_gray_queue, const char *reason, gboolean is_overflow, size_t old_next_pin_slot, gboolean forced)
2195 ScannedObjectCounts counts;
2196 SgenObjectOperations *object_ops_nopar;
2197 mword fragment_total;
2200 guint64 major_scan_start = time_major_scan_mod_union_blocks;
2201 guint64 los_scan_start = time_major_scan_mod_union_los;
2202 guint64 finish_gray_start = time_major_finish_gray_stack;
2204 if (concurrent_collection_in_progress) {
2205 SgenObjectOperations *object_ops_par = NULL;
2207 object_ops_nopar = &major_collector.major_ops_concurrent_finish;
2208 if (major_collector.is_parallel)
2209 object_ops_par = &major_collector.major_ops_conc_par_finish;
2211 major_copy_or_mark_from_roots (gc_thread_gray_queue, NULL, COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, object_ops_nopar, object_ops_par);
2213 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
2214 main_gc_thread = NULL;
2217 object_ops_nopar = &major_collector.major_ops_serial;
2220 sgen_workers_assert_gray_queue_is_empty (GENERATION_OLD);
2223 finish_gray_stack (GENERATION_OLD, CONTEXT_FROM_OBJECT_OPERATIONS (object_ops_nopar, gc_thread_gray_queue));
2225 time_major_finish_gray_stack += TV_ELAPSED (btv, atv);
2227 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after joining");
2229 if (objects_pinned) {
2230 g_assert (!concurrent_collection_in_progress);
2233 * This is slow, but we just OOM'd.
2235 * See comment at `sgen_pin_queue_clear_discarded_entries` for how the pin
2236 * queue is laid out at this point.
2238 sgen_pin_queue_clear_discarded_entries (nursery_section, old_next_pin_slot);
2240 * We need to reestablish all pinned nursery objects in the pin queue
2241 * because they're needed for fragment creation. Unpinning happens by
2242 * walking the whole queue, so it's not necessary to reestablish where major
2243 * heap block pins are - all we care is that they're still in there
2246 sgen_optimize_pin_queue ();
2247 sgen_find_section_pin_queue_start_end (nursery_section);
2251 reset_heap_boundaries ();
2252 sgen_update_heap_boundaries ((mword)sgen_get_nursery_start (), (mword)sgen_get_nursery_end ());
2254 /* walk the pin_queue, build up the fragment list of free memory, unmark
2255 * pinned objects as we go, memzero() the empty fragments so they are ready for the
2258 fragment_total = sgen_build_nursery_fragments (nursery_section, NULL);
2259 if (!fragment_total)
2261 SGEN_LOG (4, "Free space in nursery after major %ld", (long)fragment_total);
2263 if (do_concurrent_checks && concurrent_collection_in_progress)
2264 sgen_debug_check_nursery_is_clean ();
2266 /* prepare the pin queue for the next collection */
2267 sgen_finish_pinning ();
2269 /* Clear TLABs for all threads */
2270 sgen_clear_tlabs ();
2272 sgen_pin_stats_reset ();
2274 sgen_cement_clear_below_threshold ();
2276 if (check_mark_bits_after_major_collection)
2277 sgen_check_heap_marked (concurrent_collection_in_progress);
2280 time_major_fragment_creation += TV_ELAPSED (atv, btv);
2282 binary_protocol_sweep_begin (GENERATION_OLD, !major_collector.sweeps_lazily);
2283 sgen_memgov_major_pre_sweep ();
2286 time_major_free_bigobjs += TV_ELAPSED (btv, atv);
2291 time_major_los_sweep += TV_ELAPSED (atv, btv);
2293 major_collector.sweep ();
2295 binary_protocol_sweep_end (GENERATION_OLD, !major_collector.sweeps_lazily);
2298 time_major_sweep += TV_ELAPSED (btv, atv);
2300 sgen_debug_dump_heap ("major", InterlockedRead (&gc_stats.major_gc_count) - 1, reason);
2302 if (sgen_have_pending_finalizers ()) {
2303 SGEN_LOG (4, "Finalizer-thread wakeup");
2304 sgen_client_finalize_notify ();
2307 sgen_memgov_major_collection_end (forced, concurrent_collection_in_progress, reason, is_overflow);
2308 current_collection_generation = -1;
2310 memset (&counts, 0, sizeof (ScannedObjectCounts));
2311 major_collector.finish_major_collection (&counts);
2313 sgen_workers_assert_gray_queue_is_empty (GENERATION_OLD);
2315 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after major collection has finished");
2316 if (concurrent_collection_in_progress)
2317 concurrent_collection_in_progress = FALSE;
2319 check_scan_starts ();
2321 binary_protocol_flush_buffers (FALSE);
2323 //consistency_check ();
2324 if (major_collector.is_parallel)
2325 binary_protocol_collection_end_stats (0, 0, time_major_finish_gray_stack - finish_gray_start);
2327 binary_protocol_collection_end_stats (
2328 time_major_scan_mod_union_blocks - major_scan_start,
2329 time_major_scan_mod_union_los - los_scan_start,
2330 time_major_finish_gray_stack - finish_gray_start);
2332 binary_protocol_collection_end (InterlockedRead (&gc_stats.major_gc_count) - 1, GENERATION_OLD, counts.num_scanned_objects, counts.num_unique_scanned_objects);
2336 major_do_collection (const char *reason, gboolean is_overflow, gboolean forced)
2338 TV_DECLARE (time_start);
2339 TV_DECLARE (time_end);
2340 size_t old_next_pin_slot;
2341 SgenGrayQueue gc_thread_gray_queue;
2343 if (disable_major_collections)
2346 if (major_collector.get_and_reset_num_major_objects_marked) {
2347 long long num_marked = major_collector.get_and_reset_num_major_objects_marked ();
2348 g_assert (!num_marked);
2351 /* world must be stopped already */
2352 TV_GETTIME (time_start);
2354 init_gray_queue (&gc_thread_gray_queue);
2355 major_start_collection (&gc_thread_gray_queue, reason, FALSE, &old_next_pin_slot);
2356 major_finish_collection (&gc_thread_gray_queue, reason, is_overflow, old_next_pin_slot, forced);
2357 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2359 TV_GETTIME (time_end);
2360 UnlockedAdd64 (&gc_stats.major_gc_time, TV_ELAPSED (time_start, time_end));
2362 /* FIXME: also report this to the user, preferably in gc-end. */
2363 if (major_collector.get_and_reset_num_major_objects_marked)
2364 major_collector.get_and_reset_num_major_objects_marked ();
2366 return bytes_pinned_from_failed_allocation > 0;
2370 major_start_concurrent_collection (const char *reason)
2372 TV_DECLARE (time_start);
2373 TV_DECLARE (time_end);
2374 long long num_objects_marked;
2375 SgenGrayQueue gc_thread_gray_queue;
2377 if (disable_major_collections)
2380 TV_GETTIME (time_start);
2381 SGEN_TV_GETTIME (time_major_conc_collection_start);
2383 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2384 g_assert (num_objects_marked == 0);
2386 binary_protocol_concurrent_start ();
2388 init_gray_queue (&gc_thread_gray_queue);
2389 // FIXME: store reason and pass it when finishing
2390 major_start_collection (&gc_thread_gray_queue, reason, TRUE, NULL);
2391 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2393 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2395 TV_GETTIME (time_end);
2396 UnlockedAdd64 (&gc_stats.major_gc_time, TV_ELAPSED (time_start, time_end));
2398 current_collection_generation = -1;
2402 * Returns whether the major collection has finished.
2405 major_should_finish_concurrent_collection (void)
2407 return sgen_workers_all_done ();
2411 major_update_concurrent_collection (void)
2413 TV_DECLARE (total_start);
2414 TV_DECLARE (total_end);
2416 TV_GETTIME (total_start);
2418 binary_protocol_concurrent_update ();
2420 major_collector.update_cardtable_mod_union ();
2421 sgen_los_update_cardtable_mod_union ();
2423 TV_GETTIME (total_end);
2424 UnlockedAdd64 (&gc_stats.major_gc_time, TV_ELAPSED (total_start, total_end));
2428 major_finish_concurrent_collection (gboolean forced)
2430 SgenGrayQueue gc_thread_gray_queue;
2431 TV_DECLARE (total_start);
2432 TV_DECLARE (total_end);
2434 TV_GETTIME (total_start);
2436 binary_protocol_concurrent_finish ();
2439 * We need to stop all workers since we're updating the cardtable below.
2440 * The workers will be resumed with a finishing pause context to avoid
2441 * additional cardtable and object scanning.
2443 sgen_workers_stop_all_workers (GENERATION_OLD);
2445 SGEN_TV_GETTIME (time_major_conc_collection_end);
2446 UnlockedAdd64 (&gc_stats.major_gc_time_concurrent, SGEN_TV_ELAPSED (time_major_conc_collection_start, time_major_conc_collection_end));
2448 major_collector.update_cardtable_mod_union ();
2449 sgen_los_update_cardtable_mod_union ();
2451 if (mod_union_consistency_check)
2452 sgen_check_mod_union_consistency ();
2454 current_collection_generation = GENERATION_OLD;
2455 sgen_cement_reset ();
2456 init_gray_queue (&gc_thread_gray_queue);
2457 major_finish_collection (&gc_thread_gray_queue, "finishing", FALSE, -1, forced);
2458 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2460 TV_GETTIME (total_end);
2461 UnlockedAdd64 (&gc_stats.major_gc_time, TV_ELAPSED (total_start, total_end));
2463 current_collection_generation = -1;
2467 * Ensure an allocation request for @size will succeed by freeing enough memory.
2469 * LOCKING: The GC lock MUST be held.
2472 sgen_ensure_free_space (size_t size, int generation)
2474 int generation_to_collect = -1;
2475 const char *reason = NULL;
2477 if (generation == GENERATION_OLD) {
2478 if (sgen_need_major_collection (size)) {
2479 reason = "LOS overflow";
2480 generation_to_collect = GENERATION_OLD;
2483 if (degraded_mode) {
2484 if (sgen_need_major_collection (size)) {
2485 reason = "Degraded mode overflow";
2486 generation_to_collect = GENERATION_OLD;
2488 } else if (sgen_need_major_collection (size)) {
2489 reason = concurrent_collection_in_progress ? "Forced finish concurrent collection" : "Minor allowance";
2490 generation_to_collect = GENERATION_OLD;
2492 generation_to_collect = GENERATION_NURSERY;
2493 reason = "Nursery full";
2497 if (generation_to_collect == -1) {
2498 if (concurrent_collection_in_progress && sgen_workers_all_done ()) {
2499 generation_to_collect = GENERATION_OLD;
2500 reason = "Finish concurrent collection";
2504 if (generation_to_collect == -1)
2506 sgen_perform_collection (size, generation_to_collect, reason, FALSE, TRUE);
2510 * LOCKING: Assumes the GC lock is held.
2513 sgen_perform_collection (size_t requested_size, int generation_to_collect, const char *reason, gboolean wait_to_finish, gboolean stw)
2515 TV_DECLARE (gc_total_start);
2516 TV_DECLARE (gc_total_end);
2517 int overflow_generation_to_collect = -1;
2518 int oldest_generation_collected = generation_to_collect;
2519 const char *overflow_reason = NULL;
2520 gboolean finish_concurrent = concurrent_collection_in_progress && (major_should_finish_concurrent_collection () || generation_to_collect == GENERATION_OLD);
2522 binary_protocol_collection_requested (generation_to_collect, requested_size, wait_to_finish ? 1 : 0);
2524 SGEN_ASSERT (0, generation_to_collect == GENERATION_NURSERY || generation_to_collect == GENERATION_OLD, "What generation is this?");
2527 sgen_stop_world (generation_to_collect);
2529 SGEN_ASSERT (0, sgen_is_world_stopped (), "We can only collect if the world is stopped");
2532 TV_GETTIME (gc_total_start);
2534 // FIXME: extract overflow reason
2535 // FIXME: minor overflow for concurrent case
2536 if (generation_to_collect == GENERATION_NURSERY && !finish_concurrent) {
2537 if (concurrent_collection_in_progress)
2538 major_update_concurrent_collection ();
2540 if (collect_nursery (reason, FALSE, NULL) && !concurrent_collection_in_progress) {
2541 overflow_generation_to_collect = GENERATION_OLD;
2542 overflow_reason = "Minor overflow";
2544 } else if (finish_concurrent) {
2545 major_finish_concurrent_collection (wait_to_finish);
2546 oldest_generation_collected = GENERATION_OLD;
2548 SGEN_ASSERT (0, generation_to_collect == GENERATION_OLD, "We should have handled nursery collections above");
2549 if (major_collector.is_concurrent && !wait_to_finish) {
2550 collect_nursery ("Concurrent start", FALSE, NULL);
2551 major_start_concurrent_collection (reason);
2552 oldest_generation_collected = GENERATION_NURSERY;
2553 } else if (major_do_collection (reason, FALSE, wait_to_finish)) {
2554 overflow_generation_to_collect = GENERATION_NURSERY;
2555 overflow_reason = "Excessive pinning";
2559 if (overflow_generation_to_collect != -1) {
2560 SGEN_ASSERT (0, !concurrent_collection_in_progress, "We don't yet support overflow collections with the concurrent collector");
2563 * We need to do an overflow collection, either because we ran out of memory
2564 * or the nursery is fully pinned.
2567 if (overflow_generation_to_collect == GENERATION_NURSERY)
2568 collect_nursery (overflow_reason, TRUE, NULL);
2570 major_do_collection (overflow_reason, TRUE, wait_to_finish);
2572 oldest_generation_collected = MAX (oldest_generation_collected, overflow_generation_to_collect);
2575 SGEN_LOG (2, "Heap size: %lu, LOS size: %lu", (unsigned long)sgen_gc_get_total_heap_allocation (), (unsigned long)los_memory_usage);
2577 /* this also sets the proper pointers for the next allocation */
2578 if (generation_to_collect == GENERATION_NURSERY && !sgen_can_alloc_size (requested_size)) {
2579 /* TypeBuilder and MonoMethod are killing mcs with fragmentation */
2580 SGEN_LOG (1, "nursery collection didn't find enough room for %zd alloc (%zd pinned)", requested_size, sgen_get_pinned_count ());
2581 sgen_dump_pin_queue ();
2585 TV_GETTIME (gc_total_end);
2586 time_max = MAX (time_max, TV_ELAPSED (gc_total_start, gc_total_end));
2589 sgen_restart_world (oldest_generation_collected);
2593 * ######################################################################
2594 * ######## Memory allocation from the OS
2595 * ######################################################################
2596 * This section of code deals with getting memory from the OS and
2597 * allocating memory for GC-internal data structures.
2598 * Internal memory can be handled with a freelist for small objects.
2604 G_GNUC_UNUSED static void
2605 report_internal_mem_usage (void)
2607 printf ("Internal memory usage:\n");
2608 sgen_report_internal_mem_usage ();
2609 printf ("Pinned memory usage:\n");
2610 major_collector.report_pinned_memory_usage ();
2614 * ######################################################################
2615 * ######## Finalization support
2616 * ######################################################################
2620 * If the object has been forwarded it means it's still referenced from a root.
2621 * If it is pinned it's still alive as well.
2622 * A LOS object is only alive if we have pinned it.
2623 * Return TRUE if @obj is ready to be finalized.
2625 static inline gboolean
2626 sgen_is_object_alive (GCObject *object)
2628 if (ptr_in_nursery (object))
2629 return sgen_nursery_is_object_alive (object);
2631 return sgen_major_is_object_alive (object);
2635 * This function returns true if @object is either alive and belongs to the
2636 * current collection - major collections are full heap, so old gen objects
2637 * are never alive during a minor collection.
2640 sgen_is_object_alive_and_on_current_collection (GCObject *object)
2642 if (ptr_in_nursery (object))
2643 return sgen_nursery_is_object_alive (object);
2645 if (current_collection_generation == GENERATION_NURSERY)
2648 return sgen_major_is_object_alive (object);
2653 sgen_gc_is_object_ready_for_finalization (GCObject *object)
2655 return !sgen_is_object_alive (object);
2659 sgen_queue_finalization_entry (GCObject *obj)
2661 gboolean critical = sgen_client_object_has_critical_finalizer (obj);
2663 sgen_pointer_queue_add (critical ? &critical_fin_queue : &fin_ready_queue, obj);
2665 sgen_client_object_queued_for_finalization (obj);
2669 sgen_object_is_live (GCObject *obj)
2671 return sgen_is_object_alive_and_on_current_collection (obj);
2675 * `System.GC.WaitForPendingFinalizers` first checks `sgen_have_pending_finalizers()` to
2676 * determine whether it can exit quickly. The latter must therefore only return FALSE if
2677 * all finalizers have really finished running.
2679 * `sgen_gc_invoke_finalizers()` first dequeues a finalizable object, and then finalizes it.
2680 * This means that just checking whether the queues are empty leaves the possibility that an
2681 * object might have been dequeued but not yet finalized. That's why we need the additional
2682 * flag `pending_unqueued_finalizer`.
2685 static volatile gboolean pending_unqueued_finalizer = FALSE;
2686 volatile gboolean sgen_suspend_finalizers = FALSE;
2689 sgen_set_suspend_finalizers (void)
2691 sgen_suspend_finalizers = TRUE;
2695 sgen_gc_invoke_finalizers (void)
2699 g_assert (!pending_unqueued_finalizer);
2701 /* FIXME: batch to reduce lock contention */
2702 while (sgen_have_pending_finalizers ()) {
2708 * We need to set `pending_unqueued_finalizer` before dequeing the
2709 * finalizable object.
2711 if (!sgen_pointer_queue_is_empty (&fin_ready_queue)) {
2712 pending_unqueued_finalizer = TRUE;
2713 mono_memory_write_barrier ();
2714 obj = (GCObject *)sgen_pointer_queue_pop (&fin_ready_queue);
2715 } else if (!sgen_pointer_queue_is_empty (&critical_fin_queue)) {
2716 pending_unqueued_finalizer = TRUE;
2717 mono_memory_write_barrier ();
2718 obj = (GCObject *)sgen_pointer_queue_pop (&critical_fin_queue);
2724 SGEN_LOG (7, "Finalizing object %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
2732 /* the object is on the stack so it is pinned */
2733 /*g_print ("Calling finalizer for object: %p (%s)\n", obj, sgen_client_object_safe_name (obj));*/
2734 sgen_client_run_finalize (obj);
2737 if (pending_unqueued_finalizer) {
2738 mono_memory_write_barrier ();
2739 pending_unqueued_finalizer = FALSE;
2746 sgen_have_pending_finalizers (void)
2748 if (sgen_suspend_finalizers)
2750 return pending_unqueued_finalizer || !sgen_pointer_queue_is_empty (&fin_ready_queue) || !sgen_pointer_queue_is_empty (&critical_fin_queue);
2754 * ######################################################################
2755 * ######## registered roots support
2756 * ######################################################################
2760 * We do not coalesce roots.
2763 sgen_register_root (char *start, size_t size, SgenDescriptor descr, int root_type, int source, const char *msg)
2765 RootRecord new_root;
2768 for (i = 0; i < ROOT_TYPE_NUM; ++i) {
2769 RootRecord *root = (RootRecord *)sgen_hash_table_lookup (&roots_hash [i], start);
2770 /* we allow changing the size and the descriptor (for thread statics etc) */
2772 size_t old_size = root->end_root - start;
2773 root->end_root = start + size;
2774 SGEN_ASSERT (0, !!root->root_desc == !!descr, "Can't change whether a root is precise or conservative.");
2775 SGEN_ASSERT (0, root->source == source, "Can't change a root's source identifier.");
2776 SGEN_ASSERT (0, !!root->msg == !!msg, "Can't change a root's message.");
2777 root->root_desc = descr;
2779 roots_size -= old_size;
2785 new_root.end_root = start + size;
2786 new_root.root_desc = descr;
2787 new_root.source = source;
2790 sgen_hash_table_replace (&roots_hash [root_type], start, &new_root, NULL);
2793 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);
2800 sgen_deregister_root (char* addr)
2806 for (root_type = 0; root_type < ROOT_TYPE_NUM; ++root_type) {
2807 if (sgen_hash_table_remove (&roots_hash [root_type], addr, &root))
2808 roots_size -= (root.end_root - addr);
2814 sgen_wbroots_iterate_live_block_ranges (sgen_cardtable_block_callback cb)
2818 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_WBARRIER], void **, start_root, RootRecord *, root) {
2819 cb ((mword)start_root, (mword)root->end_root - (mword)start_root);
2820 } SGEN_HASH_TABLE_FOREACH_END;
2823 /* Root equivalent of sgen_client_cardtable_scan_object */
2825 sgen_wbroot_scan_card_table (void** start_root, mword size, ScanCopyContext ctx)
2827 ScanPtrFieldFunc scan_field_func = ctx.ops->scan_ptr_field;
2828 guint8 *card_data = sgen_card_table_get_card_scan_address ((mword)start_root);
2829 guint8 *card_base = card_data;
2830 mword card_count = sgen_card_table_number_of_cards_in_range ((mword)start_root, size);
2831 guint8 *card_data_end = card_data + card_count;
2832 mword extra_idx = 0;
2833 char *obj_start = sgen_card_table_align_pointer (start_root);
2834 char *obj_end = (char*)start_root + size;
2835 #ifdef SGEN_HAVE_OVERLAPPING_CARDS
2836 guint8 *overflow_scan_end = NULL;
2839 #ifdef SGEN_HAVE_OVERLAPPING_CARDS
2840 /*Check for overflow and if so, setup to scan in two steps*/
2841 if (card_data_end >= SGEN_SHADOW_CARDTABLE_END) {
2842 overflow_scan_end = sgen_shadow_cardtable + (card_data_end - SGEN_SHADOW_CARDTABLE_END);
2843 card_data_end = SGEN_SHADOW_CARDTABLE_END;
2849 card_data = sgen_find_next_card (card_data, card_data_end);
2851 for (; card_data < card_data_end; card_data = sgen_find_next_card (card_data + 1, card_data_end)) {
2852 size_t idx = (card_data - card_base) + extra_idx;
2853 char *start = (char*)(obj_start + idx * CARD_SIZE_IN_BYTES);
2854 char *card_end = start + CARD_SIZE_IN_BYTES;
2855 char *elem = start, *first_elem = start;
2858 * Don't clean first and last card on 32bit systems since they
2859 * may also be part from other roots.
2861 if (card_data != card_base && card_data != (card_data_end - 1))
2862 sgen_card_table_prepare_card_for_scanning (card_data);
2864 card_end = MIN (card_end, obj_end);
2866 if (elem < (char*)start_root)
2867 first_elem = elem = (char*)start_root;
2869 for (; elem < card_end; elem += SIZEOF_VOID_P) {
2870 if (*(GCObject**)elem)
2871 scan_field_func (NULL, (GCObject**)elem, ctx.queue);
2874 binary_protocol_card_scan (first_elem, elem - first_elem);
2877 #ifdef SGEN_HAVE_OVERLAPPING_CARDS
2878 if (overflow_scan_end) {
2879 extra_idx = card_data - card_base;
2880 card_base = card_data = sgen_shadow_cardtable;
2881 card_data_end = overflow_scan_end;
2882 overflow_scan_end = NULL;
2889 sgen_wbroots_scan_card_table (ScanCopyContext ctx)
2894 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_WBARRIER], void **, start_root, RootRecord *, root) {
2895 SGEN_ASSERT (0, (root->root_desc & ROOT_DESC_TYPE_MASK) == ROOT_DESC_VECTOR, "Unsupported root type");
2897 sgen_wbroot_scan_card_table (start_root, (mword)root->end_root - (mword)start_root, ctx);
2898 } SGEN_HASH_TABLE_FOREACH_END;
2902 * ######################################################################
2903 * ######## Thread handling (stop/start code)
2904 * ######################################################################
2908 sgen_get_current_collection_generation (void)
2910 return current_collection_generation;
2914 sgen_thread_attach (SgenThreadInfo* info)
2916 info->tlab_start = info->tlab_next = info->tlab_temp_end = info->tlab_real_end = NULL;
2918 sgen_client_thread_attach (info);
2924 sgen_thread_detach_with_lock (SgenThreadInfo *p)
2926 sgen_client_thread_detach_with_lock (p);
2930 * ######################################################################
2931 * ######## Write barriers
2932 * ######################################################################
2936 * Note: the write barriers first do the needed GC work and then do the actual store:
2937 * this way the value is visible to the conservative GC scan after the write barrier
2938 * itself. If a GC interrupts the barrier in the middle, value will be kept alive by
2939 * the conservative scan, otherwise by the remembered set scan.
2943 * mono_gc_wbarrier_arrayref_copy:
2946 mono_gc_wbarrier_arrayref_copy (gpointer dest_ptr, gpointer src_ptr, int count)
2948 HEAVY_STAT (++stat_wbarrier_arrayref_copy);
2949 /*This check can be done without taking a lock since dest_ptr array is pinned*/
2950 if (ptr_in_nursery (dest_ptr) || count <= 0) {
2951 mono_gc_memmove_aligned (dest_ptr, src_ptr, count * sizeof (gpointer));
2955 #ifdef SGEN_HEAVY_BINARY_PROTOCOL
2956 if (binary_protocol_is_heavy_enabled ()) {
2958 for (i = 0; i < count; ++i) {
2959 gpointer dest = (gpointer*)dest_ptr + i;
2960 gpointer obj = *((gpointer*)src_ptr + i);
2962 binary_protocol_wbarrier (dest, obj, (gpointer)LOAD_VTABLE (obj));
2967 remset.wbarrier_arrayref_copy (dest_ptr, src_ptr, count);
2971 * mono_gc_wbarrier_generic_nostore:
2974 mono_gc_wbarrier_generic_nostore (gpointer ptr)
2978 HEAVY_STAT (++stat_wbarrier_generic_store);
2980 sgen_client_wbarrier_generic_nostore_check (ptr);
2982 obj = *(gpointer*)ptr;
2984 binary_protocol_wbarrier (ptr, obj, (gpointer)LOAD_VTABLE (obj));
2987 * We need to record old->old pointer locations for the
2988 * concurrent collector.
2990 if (!ptr_in_nursery (obj) && !concurrent_collection_in_progress) {
2991 SGEN_LOG (8, "Skipping remset at %p", ptr);
2995 SGEN_LOG (8, "Adding remset at %p", ptr);
2997 remset.wbarrier_generic_nostore (ptr);
3001 * mono_gc_wbarrier_generic_store:
3004 mono_gc_wbarrier_generic_store (gpointer ptr, GCObject* value)
3006 SGEN_LOG (8, "Wbarrier store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
3007 SGEN_UPDATE_REFERENCE_ALLOW_NULL (ptr, value);
3008 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
3009 mono_gc_wbarrier_generic_nostore (ptr);
3010 sgen_dummy_use (value);
3014 * mono_gc_wbarrier_generic_store_atomic:
3015 * Same as \c mono_gc_wbarrier_generic_store but performs the store
3016 * as an atomic operation with release semantics.
3019 mono_gc_wbarrier_generic_store_atomic (gpointer ptr, GCObject *value)
3021 HEAVY_STAT (++stat_wbarrier_generic_store_atomic);
3023 SGEN_LOG (8, "Wbarrier atomic store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
3025 InterlockedWritePointer ((volatile gpointer *)ptr, value);
3027 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
3028 mono_gc_wbarrier_generic_nostore (ptr);
3030 sgen_dummy_use (value);
3034 sgen_wbarrier_range_copy (gpointer _dest, gpointer _src, int size)
3036 remset.wbarrier_range_copy (_dest,_src, size);
3040 * ######################################################################
3041 * ######## Other mono public interface functions.
3042 * ######################################################################
3046 sgen_gc_collect (int generation)
3051 sgen_perform_collection (0, generation, "user request", TRUE, TRUE);
3056 sgen_gc_collection_count (int generation)
3058 return InterlockedRead (generation == GENERATION_NURSERY ? &gc_stats.minor_gc_count : &gc_stats.major_gc_count);
3062 sgen_gc_get_used_size (void)
3066 tot = los_memory_usage;
3067 tot += nursery_section->end_data - nursery_section->data;
3068 tot += major_collector.get_used_size ();
3069 /* FIXME: account for pinned objects */
3075 sgen_env_var_error (const char *env_var, const char *fallback, const char *description_format, ...)
3079 va_start (ap, description_format);
3081 fprintf (stderr, "Warning: In environment variable `%s': ", env_var);
3082 vfprintf (stderr, description_format, ap);
3084 fprintf (stderr, " - %s", fallback);
3085 fprintf (stderr, "\n");
3091 parse_double_in_interval (const char *env_var, const char *opt_name, const char *opt, double min, double max, double *result)
3094 double val = strtod (opt, &endptr);
3095 if (endptr == opt) {
3096 sgen_env_var_error (env_var, "Using default value.", "`%s` must be a number.", opt_name);
3099 else if (val < min || val > max) {
3100 sgen_env_var_error (env_var, "Using default value.", "`%s` must be between %.2f - %.2f.", opt_name, min, max);
3108 parse_sgen_minor (const char *opt)
3111 return SGEN_MINOR_DEFAULT;
3113 if (!strcmp (opt, "simple")) {
3114 return SGEN_MINOR_SIMPLE;
3115 } else if (!strcmp (opt, "simple-par")) {
3116 return SGEN_MINOR_SIMPLE_PARALLEL;
3117 } else if (!strcmp (opt, "split")) {
3118 return SGEN_MINOR_SPLIT;
3120 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default instead.", "Unknown minor collector `%s'.", opt);
3121 return SGEN_MINOR_DEFAULT;
3126 parse_sgen_major (const char *opt)
3129 return SGEN_MAJOR_DEFAULT;
3131 if (!strcmp (opt, "marksweep")) {
3132 return SGEN_MAJOR_SERIAL;
3133 } else if (!strcmp (opt, "marksweep-conc")) {
3134 return SGEN_MAJOR_CONCURRENT;
3135 } else if (!strcmp (opt, "marksweep-conc-par")) {
3136 return SGEN_MAJOR_CONCURRENT_PARALLEL;
3138 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default instead.", "Unknown major collector `%s'.", opt);
3139 return SGEN_MAJOR_DEFAULT;
3145 parse_sgen_mode (const char *opt)
3148 return SGEN_MODE_NONE;
3150 if (!strcmp (opt, "balanced")) {
3151 return SGEN_MODE_BALANCED;
3152 } else if (!strcmp (opt, "throughput")) {
3153 return SGEN_MODE_THROUGHPUT;
3154 } else if (!strcmp (opt, "pause") || g_str_has_prefix (opt, "pause:")) {
3155 return SGEN_MODE_PAUSE;
3157 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default configurations.", "Unknown mode `%s'.", opt);
3158 return SGEN_MODE_NONE;
3163 init_sgen_minor (SgenMinor minor)
3166 case SGEN_MINOR_DEFAULT:
3167 case SGEN_MINOR_SIMPLE:
3168 sgen_simple_nursery_init (&sgen_minor_collector, FALSE);
3170 case SGEN_MINOR_SIMPLE_PARALLEL:
3171 sgen_simple_nursery_init (&sgen_minor_collector, TRUE);
3173 case SGEN_MINOR_SPLIT:
3174 sgen_split_nursery_init (&sgen_minor_collector);
3177 g_assert_not_reached ();
3182 init_sgen_major (SgenMajor major)
3184 if (major == SGEN_MAJOR_DEFAULT)
3185 major = DEFAULT_MAJOR;
3188 case SGEN_MAJOR_SERIAL:
3189 sgen_marksweep_init (&major_collector);
3191 case SGEN_MAJOR_CONCURRENT:
3192 sgen_marksweep_conc_init (&major_collector);
3194 case SGEN_MAJOR_CONCURRENT_PARALLEL:
3195 sgen_marksweep_conc_par_init (&major_collector);
3198 g_assert_not_reached ();
3203 * If sgen mode is set, major/minor configuration is fixed. The other gc_params
3204 * are parsed and processed after major/minor initialization, so it can potentially
3205 * override some knobs set by the sgen mode. We can consider locking out additional
3206 * configurations when gc_modes are used.
3209 init_sgen_mode (SgenMode mode)
3211 SgenMinor minor = SGEN_MINOR_DEFAULT;
3212 SgenMajor major = SGEN_MAJOR_DEFAULT;
3215 case SGEN_MODE_BALANCED:
3217 * Use a dynamic parallel nursery with a major concurrent collector.
3218 * This uses the default values for max pause time and nursery size.
3220 minor = SGEN_MINOR_SIMPLE;
3221 major = SGEN_MAJOR_CONCURRENT;
3222 dynamic_nursery = TRUE;
3224 case SGEN_MODE_THROUGHPUT:
3226 * Use concurrent major to let the mutator do more work. Use a larger
3227 * nursery, without pause time constraints, in order to collect more
3228 * objects in parallel and avoid repetitive collection tasks (pinning,
3229 * root scanning etc)
3231 minor = SGEN_MINOR_SIMPLE_PARALLEL;
3232 major = SGEN_MAJOR_CONCURRENT;
3233 dynamic_nursery = TRUE;
3234 sgen_max_pause_time = 0;
3236 case SGEN_MODE_PAUSE:
3238 * Use concurrent major and dynamic nursery with a more
3239 * aggressive shrinking relative to pause times.
3241 minor = SGEN_MINOR_SIMPLE_PARALLEL;
3242 major = SGEN_MAJOR_CONCURRENT;
3243 dynamic_nursery = TRUE;
3244 sgen_max_pause_margin = SGEN_PAUSE_MODE_MAX_PAUSE_MARGIN;
3247 g_assert_not_reached ();
3250 init_sgen_minor (minor);
3251 init_sgen_major (major);
3259 SgenMajor sgen_major = SGEN_MAJOR_DEFAULT;
3260 SgenMinor sgen_minor = SGEN_MINOR_DEFAULT;
3261 SgenMode sgen_mode = SGEN_MODE_NONE;
3262 char *params_opts = NULL;
3263 char *debug_opts = NULL;
3264 size_t max_heap = 0;
3265 size_t soft_limit = 0;
3267 gboolean debug_print_allowance = FALSE;
3268 double allowance_ratio = 0, save_target = 0;
3269 gboolean cement_enabled = TRUE;
3272 result = InterlockedCompareExchange (&gc_initialized, -1, 0);
3275 /* already inited */
3278 /* being inited by another thread */
3279 mono_thread_info_usleep (1000);
3282 /* we will init it */
3285 g_assert_not_reached ();
3287 } while (result != 0);
3289 SGEN_TV_GETTIME (sgen_init_timestamp);
3291 #ifdef SGEN_WITHOUT_MONO
3292 mono_thread_smr_init ();
3295 mono_coop_mutex_init (&gc_mutex);
3297 gc_debug_file = stderr;
3299 mono_coop_mutex_init (&sgen_interruption_mutex);
3301 if ((env = g_getenv (MONO_GC_PARAMS_NAME)) || gc_params_options) {
3302 params_opts = g_strdup_printf ("%s,%s", gc_params_options ? gc_params_options : "", env ? env : "");
3307 opts = g_strsplit (params_opts, ",", -1);
3308 for (ptr = opts; *ptr; ++ptr) {
3310 if (g_str_has_prefix (opt, "major=")) {
3311 opt = strchr (opt, '=') + 1;
3312 sgen_major = parse_sgen_major (opt);
3313 } else if (g_str_has_prefix (opt, "minor=")) {
3314 opt = strchr (opt, '=') + 1;
3315 sgen_minor = parse_sgen_minor (opt);
3316 } else if (g_str_has_prefix (opt, "mode=")) {
3317 opt = strchr (opt, '=') + 1;
3318 sgen_mode = parse_sgen_mode (opt);
3326 sgen_init_internal_allocator ();
3327 sgen_init_nursery_allocator ();
3328 sgen_init_fin_weak_hash ();
3329 sgen_init_hash_table ();
3330 sgen_init_descriptors ();
3331 sgen_init_gray_queues ();
3332 sgen_init_allocator ();
3333 sgen_init_gchandles ();
3335 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_SECTION, SGEN_SIZEOF_GC_MEM_SECTION);
3336 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_GRAY_QUEUE, sizeof (GrayQueueSection));
3338 sgen_client_init ();
3340 if (sgen_mode != SGEN_MODE_NONE) {
3341 if (sgen_minor != SGEN_MINOR_DEFAULT || sgen_major != SGEN_MAJOR_DEFAULT)
3342 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Ignoring major/minor configuration", "Major/minor configurations cannot be used with sgen modes");
3343 init_sgen_mode (sgen_mode);
3345 init_sgen_minor (sgen_minor);
3346 init_sgen_major (sgen_major);
3350 gboolean usage_printed = FALSE;
3352 for (ptr = opts; *ptr; ++ptr) {
3354 if (!strcmp (opt, ""))
3356 if (g_str_has_prefix (opt, "major="))
3358 if (g_str_has_prefix (opt, "minor="))
3360 if (g_str_has_prefix (opt, "mode=")) {
3361 if (g_str_has_prefix (opt, "mode=pause:")) {
3362 char *str_pause = strchr (opt, ':') + 1;
3363 int pause = atoi (str_pause);
3365 sgen_max_pause_time = pause;
3367 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default", "Invalid maximum pause time for `pause` sgen mode");
3371 if (g_str_has_prefix (opt, "max-heap-size=")) {
3372 size_t page_size = mono_pagesize ();
3373 size_t max_heap_candidate = 0;
3374 opt = strchr (opt, '=') + 1;
3375 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &max_heap_candidate)) {
3376 max_heap = (max_heap_candidate + page_size - 1) & ~(size_t)(page_size - 1);
3377 if (max_heap != max_heap_candidate)
3378 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Rounding up.", "`max-heap-size` size must be a multiple of %d.", page_size);
3380 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`max-heap-size` must be an integer.");
3384 if (g_str_has_prefix (opt, "soft-heap-limit=")) {
3385 opt = strchr (opt, '=') + 1;
3386 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &soft_limit)) {
3387 if (soft_limit <= 0) {
3388 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be positive.");
3392 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be an integer.");
3396 if (g_str_has_prefix (opt, "nursery-size=")) {
3398 opt = strchr (opt, '=') + 1;
3399 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &val)) {
3400 if ((val & (val - 1))) {
3401 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be a power of two.");
3405 if (val < SGEN_MAX_NURSERY_WASTE) {
3406 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.",
3407 "`nursery-size` must be at least %d bytes.", SGEN_MAX_NURSERY_WASTE);
3411 min_nursery_size = max_nursery_size = val;
3412 dynamic_nursery = FALSE;
3414 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be an integer.");
3419 if (g_str_has_prefix (opt, "save-target-ratio=")) {
3421 opt = strchr (opt, '=') + 1;
3422 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "save-target-ratio", opt,
3423 SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO, &val)) {
3428 if (g_str_has_prefix (opt, "default-allowance-ratio=")) {
3430 opt = strchr (opt, '=') + 1;
3431 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "default-allowance-ratio", opt,
3432 SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, SGEN_MAX_ALLOWANCE_NURSERY_SIZE_RATIO, &val)) {
3433 allowance_ratio = val;
3438 if (!strcmp (opt, "cementing")) {
3439 cement_enabled = TRUE;
3442 if (!strcmp (opt, "no-cementing")) {
3443 cement_enabled = FALSE;
3447 if (!strcmp (opt, "precleaning")) {
3448 precleaning_enabled = TRUE;
3451 if (!strcmp (opt, "no-precleaning")) {
3452 precleaning_enabled = FALSE;
3456 if (!strcmp (opt, "dynamic-nursery")) {
3457 if (sgen_minor_collector.is_split)
3458 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.",
3459 "dynamic-nursery not supported with split-nursery.");
3461 dynamic_nursery = TRUE;
3464 if (!strcmp (opt, "no-dynamic-nursery")) {
3465 dynamic_nursery = FALSE;
3469 if (major_collector.handle_gc_param && major_collector.handle_gc_param (opt))
3472 if (sgen_minor_collector.handle_gc_param && sgen_minor_collector.handle_gc_param (opt))
3475 if (sgen_client_handle_gc_param (opt))
3478 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Ignoring.", "Unknown option `%s`.", opt);
3483 fprintf (stderr, "\n%s must be a comma-delimited list of one or more of the following:\n", MONO_GC_PARAMS_NAME);
3484 fprintf (stderr, " max-heap-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
3485 fprintf (stderr, " soft-heap-limit=n (where N is an integer, possibly with a k, m or a g suffix)\n");
3486 fprintf (stderr, " mode=MODE (where MODE is 'balanced', 'throughput' or 'pause[:N]' and N is maximum pause in milliseconds)\n");
3487 fprintf (stderr, " nursery-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
3488 fprintf (stderr, " major=COLLECTOR (where COLLECTOR is `marksweep', `marksweep-conc', `marksweep-par')\n");
3489 fprintf (stderr, " minor=COLLECTOR (where COLLECTOR is `simple' or `split')\n");
3490 fprintf (stderr, " wbarrier=WBARRIER (where WBARRIER is `remset' or `cardtable')\n");
3491 fprintf (stderr, " [no-]cementing\n");
3492 fprintf (stderr, " [no-]dynamic-nursery\n");
3493 if (major_collector.print_gc_param_usage)
3494 major_collector.print_gc_param_usage ();
3495 if (sgen_minor_collector.print_gc_param_usage)
3496 sgen_minor_collector.print_gc_param_usage ();
3497 sgen_client_print_gc_params_usage ();
3498 fprintf (stderr, " Experimental options:\n");
3499 fprintf (stderr, " save-target-ratio=R (where R must be between %.2f - %.2f).\n", SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO);
3500 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);
3501 fprintf (stderr, "\n");
3503 usage_printed = TRUE;
3509 g_free (params_opts);
3511 alloc_nursery (dynamic_nursery, min_nursery_size, max_nursery_size);
3513 sgen_pinning_init ();
3514 sgen_cement_init (cement_enabled);
3516 if ((env = g_getenv (MONO_GC_DEBUG_NAME)) || gc_debug_options) {
3517 debug_opts = g_strdup_printf ("%s,%s", gc_debug_options ? gc_debug_options : "", env ? env : "");
3522 gboolean usage_printed = FALSE;
3524 opts = g_strsplit (debug_opts, ",", -1);
3525 for (ptr = opts; ptr && *ptr; ptr ++) {
3527 if (!strcmp (opt, ""))
3529 if (opt [0] >= '0' && opt [0] <= '9') {
3530 gc_debug_level = atoi (opt);
3535 char *rf = g_strdup_printf ("%s.%d", opt, mono_process_current_pid ());
3536 gc_debug_file = fopen (rf, "wb");
3538 gc_debug_file = stderr;
3541 } else if (!strcmp (opt, "print-allowance")) {
3542 debug_print_allowance = TRUE;
3543 } else if (!strcmp (opt, "print-pinning")) {
3544 sgen_pin_stats_enable ();
3545 } else if (!strcmp (opt, "verify-before-allocs")) {
3546 verify_before_allocs = 1;
3547 has_per_allocation_action = TRUE;
3548 } else if (g_str_has_prefix (opt, "max-valloc-size=")) {
3549 size_t max_valloc_size;
3550 char *arg = strchr (opt, '=') + 1;
3551 if (*opt && mono_gc_parse_environment_string_extract_number (arg, &max_valloc_size)) {
3552 mono_valloc_set_limit (max_valloc_size);
3554 sgen_env_var_error (MONO_GC_DEBUG_NAME, NULL, "`max-valloc-size` must be an integer.");
3557 } else if (g_str_has_prefix (opt, "verify-before-allocs=")) {
3558 char *arg = strchr (opt, '=') + 1;
3559 verify_before_allocs = atoi (arg);
3560 has_per_allocation_action = TRUE;
3561 } else if (!strcmp (opt, "collect-before-allocs")) {
3562 collect_before_allocs = 1;
3563 has_per_allocation_action = TRUE;
3564 } else if (g_str_has_prefix (opt, "collect-before-allocs=")) {
3565 char *arg = strchr (opt, '=') + 1;
3566 has_per_allocation_action = TRUE;
3567 collect_before_allocs = atoi (arg);
3568 } else if (!strcmp (opt, "verify-before-collections")) {
3569 whole_heap_check_before_collection = TRUE;
3570 } else if (!strcmp (opt, "check-remset-consistency")) {
3571 remset_consistency_checks = TRUE;
3572 nursery_clear_policy = CLEAR_AT_GC;
3573 } else if (!strcmp (opt, "mod-union-consistency-check")) {
3574 if (!major_collector.is_concurrent) {
3575 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`mod-union-consistency-check` only works with concurrent major collector.");
3578 mod_union_consistency_check = TRUE;
3579 } else if (!strcmp (opt, "check-mark-bits")) {
3580 check_mark_bits_after_major_collection = TRUE;
3581 } else if (!strcmp (opt, "check-nursery-pinned")) {
3582 check_nursery_objects_pinned = TRUE;
3583 } else if (!strcmp (opt, "clear-at-gc")) {
3584 nursery_clear_policy = CLEAR_AT_GC;
3585 } else if (!strcmp (opt, "clear-nursery-at-gc")) {
3586 nursery_clear_policy = CLEAR_AT_GC;
3587 } else if (!strcmp (opt, "clear-at-tlab-creation")) {
3588 nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
3589 } else if (!strcmp (opt, "debug-clear-at-tlab-creation")) {
3590 nursery_clear_policy = CLEAR_AT_TLAB_CREATION_DEBUG;
3591 } else if (!strcmp (opt, "check-scan-starts")) {
3592 do_scan_starts_check = TRUE;
3593 } else if (!strcmp (opt, "verify-nursery-at-minor-gc")) {
3594 do_verify_nursery = TRUE;
3595 } else if (!strcmp (opt, "check-concurrent")) {
3596 if (!major_collector.is_concurrent) {
3597 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`check-concurrent` only works with concurrent major collectors.");
3600 nursery_clear_policy = CLEAR_AT_GC;
3601 do_concurrent_checks = TRUE;
3602 } else if (!strcmp (opt, "dump-nursery-at-minor-gc")) {
3603 do_dump_nursery_content = TRUE;
3604 } else if (!strcmp (opt, "disable-minor")) {
3605 disable_minor_collections = TRUE;
3606 } else if (!strcmp (opt, "disable-major")) {
3607 disable_major_collections = TRUE;
3608 } else if (g_str_has_prefix (opt, "heap-dump=")) {
3609 char *filename = strchr (opt, '=') + 1;
3610 nursery_clear_policy = CLEAR_AT_GC;
3611 sgen_debug_enable_heap_dump (filename);
3612 } else if (g_str_has_prefix (opt, "binary-protocol=")) {
3613 char *filename = strchr (opt, '=') + 1;
3614 char *colon = strrchr (filename, ':');
3617 if (!mono_gc_parse_environment_string_extract_number (colon + 1, &limit)) {
3618 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring limit.", "Binary protocol file size limit must be an integer.");
3623 binary_protocol_init (filename, (long long)limit);
3624 } else if (!strcmp (opt, "nursery-canaries")) {
3625 do_verify_nursery = TRUE;
3626 enable_nursery_canaries = TRUE;
3627 } else if (!sgen_client_handle_gc_debug (opt)) {
3628 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "Unknown option `%s`.", opt);
3633 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);
3634 fprintf (stderr, "Valid <option>s are:\n");
3635 fprintf (stderr, " collect-before-allocs[=<n>]\n");
3636 fprintf (stderr, " verify-before-allocs[=<n>]\n");
3637 fprintf (stderr, " max-valloc-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
3638 fprintf (stderr, " check-remset-consistency\n");
3639 fprintf (stderr, " check-mark-bits\n");
3640 fprintf (stderr, " check-nursery-pinned\n");
3641 fprintf (stderr, " verify-before-collections\n");
3642 fprintf (stderr, " verify-nursery-at-minor-gc\n");
3643 fprintf (stderr, " dump-nursery-at-minor-gc\n");
3644 fprintf (stderr, " disable-minor\n");
3645 fprintf (stderr, " disable-major\n");
3646 fprintf (stderr, " check-concurrent\n");
3647 fprintf (stderr, " clear-[nursery-]at-gc\n");
3648 fprintf (stderr, " clear-at-tlab-creation\n");
3649 fprintf (stderr, " debug-clear-at-tlab-creation\n");
3650 fprintf (stderr, " check-scan-starts\n");
3651 fprintf (stderr, " print-allowance\n");
3652 fprintf (stderr, " print-pinning\n");
3653 fprintf (stderr, " heap-dump=<filename>\n");
3654 fprintf (stderr, " binary-protocol=<filename>[:<file-size-limit>]\n");
3655 fprintf (stderr, " nursery-canaries\n");
3656 sgen_client_print_gc_debug_usage ();
3657 fprintf (stderr, "\n");
3659 usage_printed = TRUE;
3666 g_free (debug_opts);
3668 if (check_mark_bits_after_major_collection)
3669 nursery_clear_policy = CLEAR_AT_GC;
3671 if (major_collector.post_param_init)
3672 major_collector.post_param_init (&major_collector);
3674 sgen_thread_pool_start ();
3676 sgen_memgov_init (max_heap, soft_limit, debug_print_allowance, allowance_ratio, save_target);
3678 memset (&remset, 0, sizeof (remset));
3680 sgen_card_table_init (&remset);
3682 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");
3686 sgen_init_bridge ();
3690 sgen_gc_initialized ()
3692 return gc_initialized > 0;
3696 sgen_get_nursery_clear_policy (void)
3698 return nursery_clear_policy;
3704 mono_coop_mutex_lock (&gc_mutex);
3708 sgen_gc_unlock (void)
3710 mono_coop_mutex_unlock (&gc_mutex);
3714 sgen_major_collector_iterate_live_block_ranges (sgen_cardtable_block_callback callback)
3716 major_collector.iterate_live_block_ranges (callback);
3720 sgen_major_collector_iterate_block_ranges (sgen_cardtable_block_callback callback)
3722 major_collector.iterate_block_ranges (callback);
3726 sgen_get_major_collector (void)
3728 return &major_collector;
3732 sgen_get_minor_collector (void)
3734 return &sgen_minor_collector;
3738 sgen_get_remset (void)
3744 count_cards (long long *major_total, long long *major_marked, long long *los_total, long long *los_marked)
3746 sgen_get_major_collector ()->count_cards (major_total, major_marked);
3747 sgen_los_count_cards (los_total, los_marked);
3750 static gboolean world_is_stopped = FALSE;
3752 /* LOCKING: assumes the GC lock is held */
3754 sgen_stop_world (int generation)
3756 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3758 SGEN_ASSERT (0, !world_is_stopped, "Why are we stopping a stopped world?");
3760 binary_protocol_world_stopping (generation, sgen_timestamp (), (gpointer) (gsize) mono_native_thread_id_get ());
3762 sgen_client_stop_world (generation);
3764 world_is_stopped = TRUE;
3766 if (binary_protocol_is_heavy_enabled ())
3767 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3768 binary_protocol_world_stopped (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3771 /* LOCKING: assumes the GC lock is held */
3773 sgen_restart_world (int generation)
3775 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3778 SGEN_ASSERT (0, world_is_stopped, "Why are we restarting a running world?");
3780 if (binary_protocol_is_heavy_enabled ())
3781 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3782 binary_protocol_world_restarting (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3784 world_is_stopped = FALSE;
3786 sgen_client_restart_world (generation, &stw_time);
3788 binary_protocol_world_restarted (generation, sgen_timestamp ());
3790 if (sgen_client_bridge_need_processing ())
3791 sgen_client_bridge_processing_finish (generation);
3793 sgen_memgov_collection_end (generation, stw_time);
3797 sgen_is_world_stopped (void)
3799 return world_is_stopped;
3803 sgen_check_whole_heap_stw (void)
3805 sgen_stop_world (0);
3806 sgen_clear_nursery_fragments ();
3807 sgen_check_whole_heap (TRUE);
3808 sgen_restart_world (0);
3812 sgen_timestamp (void)
3814 SGEN_TV_DECLARE (timestamp);
3815 SGEN_TV_GETTIME (timestamp);
3816 return SGEN_TV_ELAPSED (sgen_init_timestamp, timestamp);
3819 #endif /* HAVE_SGEN_GC */