3 * Simple generational GC.
6 * Paolo Molaro (lupus@ximian.com)
7 * Rodrigo Kumpera (kumpera@gmail.com)
9 * Copyright 2005-2011 Novell, Inc (http://www.novell.com)
10 * Copyright 2011 Xamarin Inc (http://www.xamarin.com)
12 * Thread start/stop adapted from Boehm's GC:
13 * Copyright (c) 1994 by Xerox Corporation. All rights reserved.
14 * Copyright (c) 1996 by Silicon Graphics. All rights reserved.
15 * Copyright (c) 1998 by Fergus Henderson. All rights reserved.
16 * Copyright (c) 2000-2004 by Hewlett-Packard Company. All rights reserved.
17 * Copyright 2001-2003 Ximian, Inc
18 * Copyright 2003-2010 Novell, Inc.
19 * Copyright 2011 Xamarin, Inc.
20 * Copyright (C) 2012 Xamarin Inc
22 * Licensed under the MIT license. See LICENSE file in the project root for full license information.
24 * Important: allocation provides always zeroed memory, having to do
25 * a memset after allocation is deadly for performance.
26 * Memory usage at startup is currently as follows:
28 * 64 KB internal space
30 * We should provide a small memory config with half the sizes
32 * We currently try to make as few mono assumptions as possible:
33 * 1) 2-word header with no GC pointers in it (first vtable, second to store the
35 * 2) gc descriptor is the second word in the vtable (first word in the class)
36 * 3) 8 byte alignment is the minimum and enough (not true for special structures (SIMD), FIXME)
37 * 4) there is a function to get an object's size and the number of
38 * elements in an array.
39 * 5) we know the special way bounds are allocated for complex arrays
40 * 6) we know about proxies and how to treat them when domains are unloaded
42 * Always try to keep stack usage to a minimum: no recursive behaviour
43 * and no large stack allocs.
45 * General description.
46 * Objects are initially allocated in a nursery using a fast bump-pointer technique.
47 * When the nursery is full we start a nursery collection: this is performed with a
49 * When the old generation is full we start a copying GC of the old generation as well:
50 * this will be changed to mark&sweep with copying when fragmentation becomes to severe
51 * in the future. Maybe we'll even do both during the same collection like IMMIX.
53 * The things that complicate this description are:
54 * *) pinned objects: we can't move them so we need to keep track of them
55 * *) no precise info of the thread stacks and registers: we need to be able to
56 * quickly find the objects that may be referenced conservatively and pin them
57 * (this makes the first issues more important)
58 * *) large objects are too expensive to be dealt with using copying GC: we handle them
59 * with mark/sweep during major collections
60 * *) some objects need to not move even if they are small (interned strings, Type handles):
61 * we use mark/sweep for them, too: they are not allocated in the nursery, but inside
62 * PinnedChunks regions
68 *) we could have a function pointer in MonoClass to implement
69 customized write barriers for value types
71 *) investigate the stuff needed to advance a thread to a GC-safe
72 point (single-stepping, read from unmapped memory etc) and implement it.
73 This would enable us to inline allocations and write barriers, for example,
74 or at least parts of them, like the write barrier checks.
75 We may need this also for handling precise info on stacks, even simple things
76 as having uninitialized data on the stack and having to wait for the prolog
77 to zero it. Not an issue for the last frame that we scan conservatively.
78 We could always not trust the value in the slots anyway.
80 *) modify the jit to save info about references in stack locations:
81 this can be done just for locals as a start, so that at least
82 part of the stack is handled precisely.
84 *) test/fix endianess issues
86 *) Implement a card table as the write barrier instead of remembered
87 sets? Card tables are not easy to implement with our current
88 memory layout. We have several different kinds of major heap
89 objects: Small objects in regular blocks, small objects in pinned
90 chunks and LOS objects. If we just have a pointer we have no way
91 to tell which kind of object it points into, therefore we cannot
92 know where its card table is. The least we have to do to make
93 this happen is to get rid of write barriers for indirect stores.
96 *) Get rid of write barriers for indirect stores. We can do this by
97 telling the GC to wbarrier-register an object once we do an ldloca
98 or ldelema on it, and to unregister it once it's not used anymore
99 (it can only travel downwards on the stack). The problem with
100 unregistering is that it needs to happen eventually no matter
101 what, even if exceptions are thrown, the thread aborts, etc.
102 Rodrigo suggested that we could do only the registering part and
103 let the collector find out (pessimistically) when it's safe to
104 unregister, namely when the stack pointer of the thread that
105 registered the object is higher than it was when the registering
106 happened. This might make for a good first implementation to get
107 some data on performance.
109 *) Some sort of blacklist support? Blacklists is a concept from the
110 Boehm GC: if during a conservative scan we find pointers to an
111 area which we might use as heap, we mark that area as unusable, so
112 pointer retention by random pinning pointers is reduced.
114 *) experiment with max small object size (very small right now - 2kb,
115 because it's tied to the max freelist size)
117 *) add an option to mmap the whole heap in one chunk: it makes for many
118 simplifications in the checks (put the nursery at the top and just use a single
119 check for inclusion/exclusion): the issue this has is that on 32 bit systems it's
120 not flexible (too much of the address space may be used by default or we can't
121 increase the heap as needed) and we'd need a race-free mechanism to return memory
122 back to the system (mprotect(PROT_NONE) will still keep the memory allocated if it
123 was written to, munmap is needed, but the following mmap may not find the same segment
126 *) memzero the major fragments after restarting the world and optionally a smaller
129 *) investigate having fragment zeroing threads
131 *) separate locks for finalization and other minor stuff to reduce
134 *) try a different copying order to improve memory locality
136 *) a thread abort after a store but before the write barrier will
137 prevent the write barrier from executing
139 *) specialized dynamically generated markers/copiers
141 *) Dynamically adjust TLAB size to the number of threads. If we have
142 too many threads that do allocation, we might need smaller TLABs,
143 and we might get better performance with larger TLABs if we only
144 have a handful of threads. We could sum up the space left in all
145 assigned TLABs and if that's more than some percentage of the
146 nursery size, reduce the TLAB size.
148 *) Explore placing unreachable objects on unused nursery memory.
149 Instead of memset'ng a region to zero, place an int[] covering it.
150 A good place to start is add_nursery_frag. The tricky thing here is
151 placing those objects atomically outside of a collection.
153 *) Allocation should use asymmetric Dekker synchronization:
154 http://blogs.oracle.com/dave/resource/Asymmetric-Dekker-Synchronization.txt
155 This should help weak consistency archs.
162 #define _XOPEN_SOURCE
163 #define _DARWIN_C_SOURCE
169 #ifdef HAVE_PTHREAD_H
172 #ifdef HAVE_PTHREAD_NP_H
173 #include <pthread_np.h>
181 #include "mono/sgen/sgen-gc.h"
182 #include "mono/sgen/sgen-cardtable.h"
183 #include "mono/sgen/sgen-protocol.h"
184 #include "mono/sgen/sgen-memory-governor.h"
185 #include "mono/sgen/sgen-hash-table.h"
186 #include "mono/sgen/sgen-pinning.h"
187 #include "mono/sgen/sgen-workers.h"
188 #include "mono/sgen/sgen-client.h"
189 #include "mono/sgen/sgen-pointer-queue.h"
190 #include "mono/sgen/gc-internal-agnostic.h"
191 #include "mono/utils/mono-proclib.h"
192 #include "mono/utils/mono-memory-model.h"
193 #include "mono/utils/hazard-pointer.h"
195 #include <mono/utils/memcheck.h>
196 #include <mono/utils/mono-mmap-internals.h>
198 #undef pthread_create
200 #undef pthread_detach
203 * ######################################################################
204 * ######## Types and constants used by the GC.
205 * ######################################################################
208 /* 0 means not initialized, 1 is initialized, -1 means in progress */
209 static int gc_initialized = 0;
210 /* If set, check if we need to do something every X allocations */
211 gboolean has_per_allocation_action;
212 /* If set, do a heap check every X allocation */
213 guint32 verify_before_allocs = 0;
214 /* If set, do a minor collection before every X allocation */
215 guint32 collect_before_allocs = 0;
216 /* If set, do a whole heap check before each collection */
217 static gboolean whole_heap_check_before_collection = FALSE;
218 /* If set, do a remset consistency check at various opportunities */
219 static gboolean remset_consistency_checks = FALSE;
220 /* If set, do a mod union consistency check before each finishing collection pause */
221 static gboolean mod_union_consistency_check = FALSE;
222 /* If set, check whether mark bits are consistent after major collections */
223 static gboolean check_mark_bits_after_major_collection = FALSE;
224 /* If set, check that all nursery objects are pinned/not pinned, depending on context */
225 static gboolean check_nursery_objects_pinned = FALSE;
226 /* If set, do a few checks when the concurrent collector is used */
227 static gboolean do_concurrent_checks = FALSE;
228 /* If set, do a plausibility check on the scan_starts before and after
230 static gboolean do_scan_starts_check = FALSE;
232 static gboolean disable_minor_collections = FALSE;
233 static gboolean disable_major_collections = FALSE;
234 static gboolean do_verify_nursery = FALSE;
235 static gboolean do_dump_nursery_content = FALSE;
236 static gboolean enable_nursery_canaries = FALSE;
238 static gboolean precleaning_enabled = TRUE;
239 static gboolean dynamic_nursery = FALSE;
240 static size_t min_nursery_size = 0;
241 static size_t max_nursery_size = 0;
243 #ifdef HEAVY_STATISTICS
244 guint64 stat_objects_alloced_degraded = 0;
245 guint64 stat_bytes_alloced_degraded = 0;
247 guint64 stat_copy_object_called_nursery = 0;
248 guint64 stat_objects_copied_nursery = 0;
249 guint64 stat_copy_object_called_major = 0;
250 guint64 stat_objects_copied_major = 0;
252 guint64 stat_scan_object_called_nursery = 0;
253 guint64 stat_scan_object_called_major = 0;
255 guint64 stat_slots_allocated_in_vain;
257 guint64 stat_nursery_copy_object_failed_from_space = 0;
258 guint64 stat_nursery_copy_object_failed_forwarded = 0;
259 guint64 stat_nursery_copy_object_failed_pinned = 0;
260 guint64 stat_nursery_copy_object_failed_to_space = 0;
262 static guint64 stat_wbarrier_add_to_global_remset = 0;
263 static guint64 stat_wbarrier_arrayref_copy = 0;
264 static guint64 stat_wbarrier_generic_store = 0;
265 static guint64 stat_wbarrier_generic_store_atomic = 0;
266 static guint64 stat_wbarrier_set_root = 0;
269 static guint64 stat_pinned_objects = 0;
271 static guint64 time_minor_pre_collection_fragment_clear = 0;
272 static guint64 time_minor_pinning = 0;
273 static guint64 time_minor_scan_remsets = 0;
274 static guint64 time_minor_scan_major_blocks = 0;
275 static guint64 time_minor_scan_los = 0;
276 static guint64 time_minor_scan_pinned = 0;
277 static guint64 time_minor_scan_roots = 0;
278 static guint64 time_minor_finish_gray_stack = 0;
279 static guint64 time_minor_fragment_creation = 0;
281 static guint64 time_major_pre_collection_fragment_clear = 0;
282 static guint64 time_major_pinning = 0;
283 static guint64 time_major_scan_pinned = 0;
284 static guint64 time_major_scan_roots = 0;
285 static guint64 time_major_scan_mod_union = 0;
286 static guint64 time_major_finish_gray_stack = 0;
287 static guint64 time_major_free_bigobjs = 0;
288 static guint64 time_major_los_sweep = 0;
289 static guint64 time_major_sweep = 0;
290 static guint64 time_major_fragment_creation = 0;
292 static guint64 time_max = 0;
294 static int sgen_max_pause_time = SGEN_DEFAULT_MAX_PAUSE_TIME;
295 static float sgen_max_pause_margin = SGEN_DEFAULT_MAX_PAUSE_MARGIN;
297 static SGEN_TV_DECLARE (time_major_conc_collection_start);
298 static SGEN_TV_DECLARE (time_major_conc_collection_end);
300 int gc_debug_level = 0;
302 static char* gc_params_options;
303 static char* gc_debug_options;
307 mono_gc_flush_info (void)
309 fflush (gc_debug_file);
313 #define TV_DECLARE SGEN_TV_DECLARE
314 #define TV_GETTIME SGEN_TV_GETTIME
315 #define TV_ELAPSED SGEN_TV_ELAPSED
317 static SGEN_TV_DECLARE (sgen_init_timestamp);
319 NurseryClearPolicy nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
321 #define object_is_forwarded SGEN_OBJECT_IS_FORWARDED
322 #define object_is_pinned SGEN_OBJECT_IS_PINNED
323 #define pin_object SGEN_PIN_OBJECT
325 #define ptr_in_nursery sgen_ptr_in_nursery
327 #define LOAD_VTABLE SGEN_LOAD_VTABLE
330 nursery_canaries_enabled (void)
332 return enable_nursery_canaries;
335 #define safe_object_get_size sgen_safe_object_get_size
337 #if defined(HAVE_CONC_GC_AS_DEFAULT)
338 /* Use concurrent major on deskstop platforms */
339 #define DEFAULT_MAJOR SGEN_MAJOR_CONCURRENT
341 #define DEFAULT_MAJOR SGEN_MAJOR_SERIAL
347 SGEN_MAJOR_CONCURRENT,
348 SGEN_MAJOR_CONCURRENT_PARALLEL
354 SGEN_MINOR_SIMPLE_PARALLEL,
361 SGEN_MODE_THROUGHPUT,
366 * ######################################################################
367 * ######## Global data.
368 * ######################################################################
370 MonoCoopMutex gc_mutex;
372 #define SCAN_START_SIZE SGEN_SCAN_START_SIZE
374 size_t degraded_mode = 0;
376 static mword bytes_pinned_from_failed_allocation = 0;
378 GCMemSection *nursery_section = NULL;
379 static volatile mword lowest_heap_address = ~(mword)0;
380 static volatile mword highest_heap_address = 0;
382 MonoCoopMutex sgen_interruption_mutex;
384 int current_collection_generation = -1;
385 volatile gboolean concurrent_collection_in_progress = FALSE;
387 /* objects that are ready to be finalized */
388 static SgenPointerQueue fin_ready_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
389 static SgenPointerQueue critical_fin_queue = SGEN_POINTER_QUEUE_INIT (INTERNAL_MEM_FINALIZE_READY);
391 /* registered roots: the key to the hash is the root start address */
393 * Different kinds of roots are kept separate to speed up pin_from_roots () for example.
395 SgenHashTable roots_hash [ROOT_TYPE_NUM] = {
396 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
397 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL),
398 SGEN_HASH_TABLE_INIT (INTERNAL_MEM_ROOTS_TABLE, INTERNAL_MEM_ROOT_RECORD, sizeof (RootRecord), sgen_aligned_addr_hash, NULL)
400 static mword roots_size = 0; /* amount of memory in the root set */
402 /* The size of a TLAB */
403 /* The bigger the value, the less often we have to go to the slow path to allocate a new
404 * one, but the more space is wasted by threads not allocating much memory.
406 * FIXME: Make this self-tuning for each thread.
408 guint32 tlab_size = (1024 * 4);
410 #define MAX_SMALL_OBJ_SIZE SGEN_MAX_SMALL_OBJ_SIZE
412 #define ALLOC_ALIGN SGEN_ALLOC_ALIGN
414 #define ALIGN_UP SGEN_ALIGN_UP
416 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
417 MonoNativeThreadId main_gc_thread = NULL;
420 /*Object was pinned during the current collection*/
421 static mword objects_pinned;
424 * ######################################################################
425 * ######## Macros and function declarations.
426 * ######################################################################
429 /* forward declarations */
430 static void scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx);
432 static void pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx);
433 static void finish_gray_stack (int generation, ScanCopyContext ctx);
436 SgenMajorCollector major_collector;
437 SgenMinorCollector sgen_minor_collector;
439 static SgenRememberedSet remset;
442 * The gray queue a worker job must use. If we're not parallel or
443 * concurrent, we use the main gray queue.
445 static SgenGrayQueue*
446 sgen_workers_get_job_gray_queue (WorkerData *worker_data, SgenGrayQueue *default_gray_queue)
449 return &worker_data->private_gray_queue;
450 SGEN_ASSERT (0, default_gray_queue, "Why don't we have a default gray queue when we're not running in a worker thread?");
451 return default_gray_queue;
455 gray_queue_redirect (SgenGrayQueue *queue)
457 sgen_workers_take_from_queue (queue);
461 sgen_scan_area_with_callback (char *start, char *end, IterateObjectCallbackFunc callback, void *data, gboolean allow_flags, gboolean fail_on_canaries)
463 while (start < end) {
467 if (!*(void**)start) {
468 start += sizeof (void*); /* should be ALLOC_ALIGN, really */
473 if (!(obj = (char *)SGEN_OBJECT_IS_FORWARDED (start)))
479 if (!sgen_client_object_is_array_fill ((GCObject*)obj)) {
480 CHECK_CANARY_FOR_OBJECT ((GCObject*)obj, fail_on_canaries);
481 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
482 callback ((GCObject*)obj, size, data);
483 CANARIFY_SIZE (size);
485 size = ALIGN_UP (safe_object_get_size ((GCObject*)obj));
493 * sgen_add_to_global_remset:
495 * The global remset contains locations which point into newspace after
496 * a minor collection. This can happen if the objects they point to are pinned.
498 * LOCKING: If called from a parallel collector, the global remset
499 * lock must be held. For serial collectors that is not necessary.
502 sgen_add_to_global_remset (gpointer ptr, GCObject *obj)
504 SGEN_ASSERT (5, sgen_ptr_in_nursery (obj), "Target pointer of global remset must be in the nursery");
506 HEAVY_STAT (++stat_wbarrier_add_to_global_remset);
508 if (!major_collector.is_concurrent) {
509 SGEN_ASSERT (5, current_collection_generation != -1, "Global remsets can only be added during collections");
511 if (current_collection_generation == -1)
512 SGEN_ASSERT (5, sgen_concurrent_collection_in_progress (), "Global remsets outside of collection pauses can only be added by the concurrent collector");
515 if (!object_is_pinned (obj))
516 SGEN_ASSERT (5, sgen_minor_collector.is_split || sgen_concurrent_collection_in_progress (), "Non-pinned objects can only remain in nursery if it is a split nursery");
517 else if (sgen_cement_lookup_or_register (obj))
520 remset.record_pointer (ptr);
522 sgen_pin_stats_register_global_remset (obj);
524 SGEN_LOG (8, "Adding global remset for %p", ptr);
525 binary_protocol_global_remset (ptr, obj, (gpointer)SGEN_LOAD_VTABLE (obj));
529 * sgen_drain_gray_stack:
531 * Scan objects in the gray stack until the stack is empty. This should be called
532 * frequently after each object is copied, to achieve better locality and cache
537 sgen_drain_gray_stack (ScanCopyContext ctx)
539 SGEN_ASSERT (0, ctx.ops->drain_gray_stack, "Why do we have a scan/copy context with a missing drain gray stack function?");
541 return ctx.ops->drain_gray_stack (ctx.queue);
545 * Addresses in the pin queue are already sorted. This function finds
546 * the object header for each address and pins the object. The
547 * addresses must be inside the nursery section. The (start of the)
548 * address array is overwritten with the addresses of the actually
549 * pinned objects. Return the number of pinned objects.
552 pin_objects_from_nursery_pin_queue (gboolean do_scan_objects, ScanCopyContext ctx)
554 GCMemSection *section = nursery_section;
555 void **start = sgen_pinning_get_entry (section->pin_queue_first_entry);
556 void **end = sgen_pinning_get_entry (section->pin_queue_last_entry);
557 void *start_nursery = section->data;
558 void *end_nursery = section->end_data;
563 void *pinning_front = start_nursery;
565 void **definitely_pinned = start;
566 ScanObjectFunc scan_func = ctx.ops->scan_object;
567 SgenGrayQueue *queue = ctx.queue;
569 sgen_nursery_allocator_prepare_for_pinning ();
571 while (start < end) {
572 GCObject *obj_to_pin = NULL;
573 size_t obj_to_pin_size = 0;
578 SGEN_ASSERT (0, addr >= start_nursery && addr < end_nursery, "Potential pinning address out of range");
579 SGEN_ASSERT (0, addr >= last, "Pin queue not sorted");
586 SGEN_LOG (5, "Considering pinning addr %p", addr);
587 /* We've already processed everything up to pinning_front. */
588 if (addr < pinning_front) {
594 * Find the closest scan start <= addr. We might search backward in the
595 * scan_starts array because entries might be NULL. In the worst case we
596 * start at start_nursery.
598 idx = ((char*)addr - (char*)section->data) / SCAN_START_SIZE;
599 SGEN_ASSERT (0, idx < section->num_scan_start, "Scan start index out of range");
600 search_start = (void*)section->scan_starts [idx];
601 if (!search_start || search_start > addr) {
604 search_start = section->scan_starts [idx];
605 if (search_start && search_start <= addr)
608 if (!search_start || search_start > addr)
609 search_start = start_nursery;
613 * If the pinning front is closer than the scan start we found, start
614 * searching at the front.
616 if (search_start < pinning_front)
617 search_start = pinning_front;
620 * Now addr should be in an object a short distance from search_start.
622 * search_start must point to zeroed mem or point to an object.
625 size_t obj_size, canarified_obj_size;
628 if (!*(void**)search_start) {
629 search_start = (void*)ALIGN_UP ((mword)search_start + sizeof (gpointer));
630 /* The loop condition makes sure we don't overrun addr. */
634 canarified_obj_size = obj_size = ALIGN_UP (safe_object_get_size ((GCObject*)search_start));
637 * Filler arrays are marked by an invalid sync word. We don't
638 * consider them for pinning. They are not delimited by canaries,
641 if (!sgen_client_object_is_array_fill ((GCObject*)search_start)) {
642 CHECK_CANARY_FOR_OBJECT (search_start, TRUE);
643 CANARIFY_SIZE (canarified_obj_size);
645 if (addr >= search_start && (char*)addr < (char*)search_start + obj_size) {
646 /* This is the object we're looking for. */
647 obj_to_pin = (GCObject*)search_start;
648 obj_to_pin_size = canarified_obj_size;
653 /* Skip to the next object */
654 search_start = (void*)((char*)search_start + canarified_obj_size);
655 } while (search_start <= addr);
657 /* We've searched past the address we were looking for. */
659 pinning_front = search_start;
660 goto next_pin_queue_entry;
664 * We've found an object to pin. It might still be a dummy array, but we
665 * can advance the pinning front in any case.
667 pinning_front = (char*)obj_to_pin + obj_to_pin_size;
670 * If this is a dummy array marking the beginning of a nursery
671 * fragment, we don't pin it.
673 if (sgen_client_object_is_array_fill (obj_to_pin))
674 goto next_pin_queue_entry;
677 * Finally - pin the object!
679 desc = sgen_obj_get_descriptor_safe (obj_to_pin);
680 if (do_scan_objects) {
681 scan_func (obj_to_pin, desc, queue);
683 SGEN_LOG (4, "Pinned object %p, vtable %p (%s), count %d\n",
684 obj_to_pin, *(void**)obj_to_pin, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj_to_pin)), count);
685 binary_protocol_pin (obj_to_pin,
686 (gpointer)LOAD_VTABLE (obj_to_pin),
687 safe_object_get_size (obj_to_pin));
689 pin_object (obj_to_pin);
690 GRAY_OBJECT_ENQUEUE_SERIAL (queue, obj_to_pin, desc);
691 sgen_pin_stats_register_object (obj_to_pin, GENERATION_NURSERY);
692 definitely_pinned [count] = obj_to_pin;
695 if (concurrent_collection_in_progress)
696 sgen_pinning_register_pinned_in_nursery (obj_to_pin);
698 next_pin_queue_entry:
702 sgen_client_nursery_objects_pinned (definitely_pinned, count);
703 stat_pinned_objects += count;
708 pin_objects_in_nursery (gboolean do_scan_objects, ScanCopyContext ctx)
712 if (nursery_section->pin_queue_first_entry == nursery_section->pin_queue_last_entry)
715 reduced_to = pin_objects_from_nursery_pin_queue (do_scan_objects, ctx);
716 nursery_section->pin_queue_last_entry = nursery_section->pin_queue_first_entry + reduced_to;
720 * This function is only ever called (via `collector_pin_object()` in `sgen-copy-object.h`)
721 * when we can't promote an object because we're out of memory.
724 sgen_pin_object (GCObject *object, SgenGrayQueue *queue)
726 SGEN_ASSERT (0, sgen_ptr_in_nursery (object), "We're only supposed to use this for pinning nursery objects when out of memory.");
729 * All pinned objects are assumed to have been staged, so we need to stage as well.
730 * Also, the count of staged objects shows that "late pinning" happened.
732 sgen_pin_stage_ptr (object);
734 SGEN_PIN_OBJECT (object);
735 binary_protocol_pin (object, (gpointer)LOAD_VTABLE (object), safe_object_get_size (object));
738 sgen_pin_stats_register_object (object, GENERATION_NURSERY);
740 GRAY_OBJECT_ENQUEUE_SERIAL (queue, object, sgen_obj_get_descriptor_safe (object));
743 /* Sort the addresses in array in increasing order.
744 * Done using a by-the book heap sort. Which has decent and stable performance, is pretty cache efficient.
747 sgen_sort_addresses (void **array, size_t size)
752 for (i = 1; i < size; ++i) {
755 size_t parent = (child - 1) / 2;
757 if (array [parent] >= array [child])
760 tmp = array [parent];
761 array [parent] = array [child];
768 for (i = size - 1; i > 0; --i) {
771 array [i] = array [0];
777 while (root * 2 + 1 <= end) {
778 size_t child = root * 2 + 1;
780 if (child < end && array [child] < array [child + 1])
782 if (array [root] >= array [child])
786 array [root] = array [child];
795 * Scan the memory between start and end and queue values which could be pointers
796 * to the area between start_nursery and end_nursery for later consideration.
797 * Typically used for thread stacks.
800 sgen_conservatively_pin_objects_from (void **start, void **end, void *start_nursery, void *end_nursery, int pin_type)
804 SGEN_ASSERT (0, ((mword)start & (SIZEOF_VOID_P - 1)) == 0, "Why are we scanning for references in unaligned memory ?");
806 #if defined(VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE) && !defined(_WIN64)
807 VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE (start, (char*)end - (char*)start);
810 while (start < end) {
812 * *start can point to the middle of an object
813 * note: should we handle pointing at the end of an object?
814 * pinning in C# code disallows pointing at the end of an object
815 * but there is some small chance that an optimizing C compiler
816 * may keep the only reference to an object by pointing
817 * at the end of it. We ignore this small chance for now.
818 * Pointers to the end of an object are indistinguishable
819 * from pointers to the start of the next object in memory
820 * so if we allow that we'd need to pin two objects...
821 * We queue the pointer in an array, the
822 * array will then be sorted and uniqued. This way
823 * we can coalesce several pinning pointers and it should
824 * be faster since we'd do a memory scan with increasing
825 * addresses. Note: we can align the address to the allocation
826 * alignment, so the unique process is more effective.
828 mword addr = (mword)*start;
829 addr &= ~(ALLOC_ALIGN - 1);
830 if (addr >= (mword)start_nursery && addr < (mword)end_nursery) {
831 SGEN_LOG (6, "Pinning address %p from %p", (void*)addr, start);
832 sgen_pin_stage_ptr ((void*)addr);
833 binary_protocol_pin_stage (start, (void*)addr);
834 sgen_pin_stats_register_address ((char*)addr, pin_type);
840 SGEN_LOG (7, "found %d potential pinned heap pointers", count);
844 * The first thing we do in a collection is to identify pinned objects.
845 * This function considers all the areas of memory that need to be
846 * conservatively scanned.
849 pin_from_roots (void *start_nursery, void *end_nursery, ScanCopyContext ctx)
853 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);
854 /* objects pinned from the API are inside these roots */
855 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_PINNED], void **, start_root, RootRecord *, root) {
856 SGEN_LOG (6, "Pinned roots %p-%p", start_root, root->end_root);
857 sgen_conservatively_pin_objects_from (start_root, (void**)root->end_root, start_nursery, end_nursery, PIN_TYPE_OTHER);
858 } SGEN_HASH_TABLE_FOREACH_END;
859 /* now deal with the thread stacks
860 * in the future we should be able to conservatively scan only:
861 * *) the cpu registers
862 * *) the unmanaged stack frames
863 * *) the _last_ managed stack frame
864 * *) pointers slots in managed frames
866 sgen_client_scan_thread_data (start_nursery, end_nursery, FALSE, ctx);
870 single_arg_user_copy_or_mark (GCObject **obj, void *gc_data)
872 ScanCopyContext *ctx = (ScanCopyContext *)gc_data;
873 ctx->ops->copy_or_mark_object (obj, ctx->queue);
877 * The memory area from start_root to end_root contains pointers to objects.
878 * Their position is precisely described by @desc (this means that the pointer
879 * can be either NULL or the pointer to the start of an object).
880 * This functions copies them to to_space updates them.
882 * This function is not thread-safe!
885 precisely_scan_objects_from (void** start_root, void** end_root, char* n_start, char *n_end, SgenDescriptor desc, ScanCopyContext ctx)
887 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
888 ScanPtrFieldFunc scan_field_func = ctx.ops->scan_ptr_field;
889 SgenGrayQueue *queue = ctx.queue;
891 switch (desc & ROOT_DESC_TYPE_MASK) {
892 case ROOT_DESC_BITMAP:
893 desc >>= ROOT_DESC_TYPE_SHIFT;
895 if ((desc & 1) && *start_root) {
896 copy_func ((GCObject**)start_root, queue);
897 SGEN_LOG (9, "Overwrote root at %p with %p", start_root, *start_root);
903 case ROOT_DESC_COMPLEX: {
904 gsize *bitmap_data = (gsize *)sgen_get_complex_descriptor_bitmap (desc);
905 gsize bwords = (*bitmap_data) - 1;
906 void **start_run = start_root;
908 while (bwords-- > 0) {
909 gsize bmap = *bitmap_data++;
910 void **objptr = start_run;
912 if ((bmap & 1) && *objptr) {
913 copy_func ((GCObject**)objptr, queue);
914 SGEN_LOG (9, "Overwrote root at %p with %p", objptr, *objptr);
919 start_run += GC_BITS_PER_WORD;
923 case ROOT_DESC_VECTOR: {
926 for (p = start_root; p < end_root; p++) {
928 scan_field_func (NULL, (GCObject**)p, queue);
932 case ROOT_DESC_USER: {
933 SgenUserRootMarkFunc marker = sgen_get_user_descriptor_func (desc);
934 marker (start_root, single_arg_user_copy_or_mark, &ctx);
937 case ROOT_DESC_RUN_LEN:
938 g_assert_not_reached ();
940 g_assert_not_reached ();
945 reset_heap_boundaries (void)
947 lowest_heap_address = ~(mword)0;
948 highest_heap_address = 0;
952 sgen_update_heap_boundaries (mword low, mword high)
957 old = lowest_heap_address;
960 } while (SGEN_CAS_PTR ((gpointer*)&lowest_heap_address, (gpointer)low, (gpointer)old) != (gpointer)old);
963 old = highest_heap_address;
966 } while (SGEN_CAS_PTR ((gpointer*)&highest_heap_address, (gpointer)high, (gpointer)old) != (gpointer)old);
970 * Allocate and setup the data structures needed to be able to allocate objects
971 * in the nursery. The nursery is stored in nursery_section.
974 alloc_nursery (gboolean dynamic, size_t min_size, size_t max_size)
981 min_size = SGEN_DEFAULT_NURSERY_MIN_SIZE;
983 max_size = SGEN_DEFAULT_NURSERY_MAX_SIZE;
985 SGEN_ASSERT (0, min_size == max_size, "We can't have nursery ranges for static configuration.");
987 min_size = max_size = SGEN_DEFAULT_NURSERY_SIZE;
990 SGEN_ASSERT (0, !nursery_section, "Why are we allocating the nursery twice?");
991 SGEN_LOG (2, "Allocating nursery size: %zu, initial %zu", max_size, min_size);
993 /* FIXME: handle OOM */
994 nursery_section = (GCMemSection *)sgen_alloc_internal (INTERNAL_MEM_SECTION);
996 /* If there isn't enough space even for the nursery we should simply abort. */
997 g_assert (sgen_memgov_try_alloc_space (max_size, SPACE_NURSERY));
1000 * The nursery section range represents the memory section where objects
1001 * can be found. This is used when iterating for objects in the nursery,
1002 * pinning etc. sgen_nursery_max_size represents the total allocated space
1003 * for the nursery. sgen_nursery_size represents the current size of the
1004 * nursery and it is used for allocation limits, heuristics etc. The
1005 * nursery section is not always identical to the current nursery size
1006 * because it can contain pinned objects from when the nursery was larger.
1008 * sgen_nursery_size <= nursery_section size <= sgen_nursery_max_size
1010 data = (char *)major_collector.alloc_heap (max_size, max_size);
1011 sgen_update_heap_boundaries ((mword)data, (mword)(data + max_size));
1012 nursery_section->data = data;
1013 nursery_section->end_data = data + min_size;
1014 scan_starts = (max_size + SCAN_START_SIZE - 1) / SCAN_START_SIZE;
1015 nursery_section->scan_starts = (char **)sgen_alloc_internal_dynamic (sizeof (char*) * scan_starts, INTERNAL_MEM_SCAN_STARTS, TRUE);
1016 nursery_section->num_scan_start = scan_starts;
1018 sgen_nursery_allocator_set_nursery_bounds (data, min_size, max_size);
1022 mono_gc_get_logfile (void)
1024 return gc_debug_file;
1028 mono_gc_params_set (const char* options)
1030 if (gc_params_options)
1031 g_free (gc_params_options);
1033 gc_params_options = g_strdup (options);
1037 mono_gc_debug_set (const char* options)
1039 if (gc_debug_options)
1040 g_free (gc_debug_options);
1042 gc_debug_options = g_strdup (options);
1046 scan_finalizer_entries (SgenPointerQueue *fin_queue, ScanCopyContext ctx)
1048 CopyOrMarkObjectFunc copy_func = ctx.ops->copy_or_mark_object;
1049 SgenGrayQueue *queue = ctx.queue;
1052 for (i = 0; i < fin_queue->next_slot; ++i) {
1053 GCObject *obj = (GCObject *)fin_queue->data [i];
1056 SGEN_LOG (5, "Scan of fin ready object: %p (%s)\n", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
1057 copy_func ((GCObject**)&fin_queue->data [i], queue);
1062 generation_name (int generation)
1064 switch (generation) {
1065 case GENERATION_NURSERY: return "nursery";
1066 case GENERATION_OLD: return "old";
1067 default: g_assert_not_reached ();
1072 sgen_generation_name (int generation)
1074 return generation_name (generation);
1078 finish_gray_stack (int generation, ScanCopyContext ctx)
1082 int done_with_ephemerons, ephemeron_rounds = 0;
1083 char *start_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_start () : NULL;
1084 char *end_addr = generation == GENERATION_NURSERY ? sgen_get_nursery_end () : (char*)-1;
1085 SgenGrayQueue *queue = ctx.queue;
1087 binary_protocol_finish_gray_stack_start (sgen_timestamp (), generation);
1089 * We copied all the reachable objects. Now it's the time to copy
1090 * the objects that were not referenced by the roots, but by the copied objects.
1091 * we built a stack of objects pointed to by gray_start: they are
1092 * additional roots and we may add more items as we go.
1093 * We loop until gray_start == gray_objects which means no more objects have
1094 * been added. Note this is iterative: no recursion is involved.
1095 * We need to walk the LO list as well in search of marked big objects
1096 * (use a flag since this is needed only on major collections). We need to loop
1097 * here as well, so keep a counter of marked LO (increasing it in copy_object).
1098 * To achieve better cache locality and cache usage, we drain the gray stack
1099 * frequently, after each object is copied, and just finish the work here.
1101 sgen_drain_gray_stack (ctx);
1103 SGEN_LOG (2, "%s generation done", generation_name (generation));
1106 Reset bridge data, we might have lingering data from a previous collection if this is a major
1107 collection trigged by minor overflow.
1109 We must reset the gathered bridges since their original block might be evacuated due to major
1110 fragmentation in the meanwhile and the bridge code should not have to deal with that.
1112 if (sgen_client_bridge_need_processing ())
1113 sgen_client_bridge_reset_data ();
1116 * Mark all strong toggleref objects. This must be done before we walk ephemerons or finalizers
1117 * to ensure they see the full set of live objects.
1119 sgen_client_mark_togglerefs (start_addr, end_addr, ctx);
1122 * Walk the ephemeron tables marking all values with reachable keys. This must be completely done
1123 * before processing finalizable objects and non-tracking weak links to avoid finalizing/clearing
1124 * objects that are in fact reachable.
1126 done_with_ephemerons = 0;
1128 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1129 sgen_drain_gray_stack (ctx);
1131 } while (!done_with_ephemerons);
1133 if (sgen_client_bridge_need_processing ()) {
1134 /*Make sure the gray stack is empty before we process bridge objects so we get liveness right*/
1135 sgen_drain_gray_stack (ctx);
1136 sgen_collect_bridge_objects (generation, ctx);
1137 if (generation == GENERATION_OLD)
1138 sgen_collect_bridge_objects (GENERATION_NURSERY, ctx);
1141 Do the first bridge step here, as the collector liveness state will become useless after that.
1143 An important optimization is to only proccess the possibly dead part of the object graph and skip
1144 over all live objects as we transitively know everything they point must be alive too.
1146 The above invariant is completely wrong if we let the gray queue be drained and mark/copy everything.
1148 This has the unfortunate side effect of making overflow collections perform the first step twice, but
1149 given we now have heuristics that perform major GC in anticipation of minor overflows this should not
1152 sgen_client_bridge_processing_stw_step ();
1156 Make sure we drain the gray stack before processing disappearing links and finalizers.
1157 If we don't make sure it is empty we might wrongly see a live object as dead.
1159 sgen_drain_gray_stack (ctx);
1162 We must clear weak links that don't track resurrection before processing object ready for
1163 finalization so they can be cleared before that.
1165 sgen_null_link_in_range (generation, ctx, FALSE);
1166 if (generation == GENERATION_OLD)
1167 sgen_null_link_in_range (GENERATION_NURSERY, ctx, FALSE);
1170 /* walk the finalization queue and move also the objects that need to be
1171 * finalized: use the finalized objects as new roots so the objects they depend
1172 * on are also not reclaimed. As with the roots above, only objects in the nursery
1173 * are marked/copied.
1175 sgen_finalize_in_range (generation, ctx);
1176 if (generation == GENERATION_OLD)
1177 sgen_finalize_in_range (GENERATION_NURSERY, ctx);
1178 /* drain the new stack that might have been created */
1179 SGEN_LOG (6, "Precise scan of gray area post fin");
1180 sgen_drain_gray_stack (ctx);
1183 * This must be done again after processing finalizable objects since CWL slots are cleared only after the key is finalized.
1185 done_with_ephemerons = 0;
1187 done_with_ephemerons = sgen_client_mark_ephemerons (ctx);
1188 sgen_drain_gray_stack (ctx);
1190 } while (!done_with_ephemerons);
1192 sgen_client_clear_unreachable_ephemerons (ctx);
1195 * We clear togglerefs only after all possible chances of revival are done.
1196 * This is semantically more inline with what users expect and it allows for
1197 * user finalizers to correctly interact with TR objects.
1199 sgen_client_clear_togglerefs (start_addr, end_addr, ctx);
1202 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);
1205 * handle disappearing links
1206 * Note we do this after checking the finalization queue because if an object
1207 * survives (at least long enough to be finalized) we don't clear the link.
1208 * This also deals with a possible issue with the monitor reclamation: with the Boehm
1209 * GC a finalized object my lose the monitor because it is cleared before the finalizer is
1212 g_assert (sgen_gray_object_queue_is_empty (queue));
1214 sgen_null_link_in_range (generation, ctx, TRUE);
1215 if (generation == GENERATION_OLD)
1216 sgen_null_link_in_range (GENERATION_NURSERY, ctx, TRUE);
1217 if (sgen_gray_object_queue_is_empty (queue))
1219 sgen_drain_gray_stack (ctx);
1222 g_assert (sgen_gray_object_queue_is_empty (queue));
1224 binary_protocol_finish_gray_stack_end (sgen_timestamp (), generation);
1228 sgen_check_section_scan_starts (GCMemSection *section)
1231 for (i = 0; i < section->num_scan_start; ++i) {
1232 if (section->scan_starts [i]) {
1233 mword size = safe_object_get_size ((GCObject*) section->scan_starts [i]);
1234 SGEN_ASSERT (0, size >= SGEN_CLIENT_MINIMUM_OBJECT_SIZE && size <= MAX_SMALL_OBJ_SIZE, "Weird object size at scan starts.");
1240 check_scan_starts (void)
1242 if (!do_scan_starts_check)
1244 sgen_check_section_scan_starts (nursery_section);
1245 major_collector.check_scan_starts ();
1249 scan_from_registered_roots (char *addr_start, char *addr_end, int root_type, ScanCopyContext ctx)
1253 SGEN_HASH_TABLE_FOREACH (&roots_hash [root_type], void **, start_root, RootRecord *, root) {
1254 SGEN_LOG (6, "Precise root scan %p-%p (desc: %p)", start_root, root->end_root, (void*)root->root_desc);
1255 precisely_scan_objects_from (start_root, (void**)root->end_root, addr_start, addr_end, root->root_desc, ctx);
1256 } SGEN_HASH_TABLE_FOREACH_END;
1262 static gboolean inited = FALSE;
1267 mono_counters_register ("Collection max time", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME | MONO_COUNTER_MONOTONIC, &time_max);
1269 mono_counters_register ("Minor fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pre_collection_fragment_clear);
1270 mono_counters_register ("Minor pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_pinning);
1271 mono_counters_register ("Minor scan remembered set", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_remsets);
1272 mono_counters_register ("Minor scan major blocks", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_major_blocks);
1273 mono_counters_register ("Minor scan los", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_los);
1274 mono_counters_register ("Minor scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_pinned);
1275 mono_counters_register ("Minor scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_scan_roots);
1276 mono_counters_register ("Minor fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_minor_fragment_creation);
1278 mono_counters_register ("Major fragment clear", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pre_collection_fragment_clear);
1279 mono_counters_register ("Major pinning", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_pinning);
1280 mono_counters_register ("Major scan pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_pinned);
1281 mono_counters_register ("Major scan roots", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_roots);
1282 mono_counters_register ("Major scan mod union", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_scan_mod_union);
1283 mono_counters_register ("Major finish gray stack", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_finish_gray_stack);
1284 mono_counters_register ("Major free big objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_free_bigobjs);
1285 mono_counters_register ("Major LOS sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_los_sweep);
1286 mono_counters_register ("Major sweep", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_sweep);
1287 mono_counters_register ("Major fragment creation", MONO_COUNTER_GC | MONO_COUNTER_ULONG | MONO_COUNTER_TIME, &time_major_fragment_creation);
1289 mono_counters_register ("Number of pinned objects", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_pinned_objects);
1291 #ifdef HEAVY_STATISTICS
1292 mono_counters_register ("WBarrier remember pointer", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_add_to_global_remset);
1293 mono_counters_register ("WBarrier arrayref copy", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_arrayref_copy);
1294 mono_counters_register ("WBarrier generic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store);
1295 mono_counters_register ("WBarrier generic atomic store called", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_generic_store_atomic);
1296 mono_counters_register ("WBarrier set root", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_wbarrier_set_root);
1298 mono_counters_register ("# objects allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_alloced_degraded);
1299 mono_counters_register ("bytes allocated degraded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_bytes_alloced_degraded);
1301 mono_counters_register ("# copy_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_nursery);
1302 mono_counters_register ("# objects copied (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_nursery);
1303 mono_counters_register ("# copy_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_copy_object_called_major);
1304 mono_counters_register ("# objects copied (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_objects_copied_major);
1306 mono_counters_register ("# scan_object() called (nursery)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_nursery);
1307 mono_counters_register ("# scan_object() called (major)", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_scan_object_called_major);
1309 mono_counters_register ("Slots allocated in vain", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_slots_allocated_in_vain);
1311 mono_counters_register ("# nursery copy_object() failed from space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_from_space);
1312 mono_counters_register ("# nursery copy_object() failed forwarded", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_forwarded);
1313 mono_counters_register ("# nursery copy_object() failed pinned", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_pinned);
1314 mono_counters_register ("# nursery copy_object() failed to space", MONO_COUNTER_GC | MONO_COUNTER_ULONG, &stat_nursery_copy_object_failed_to_space);
1316 sgen_nursery_allocator_init_heavy_stats ();
1324 reset_pinned_from_failed_allocation (void)
1326 bytes_pinned_from_failed_allocation = 0;
1330 sgen_set_pinned_from_failed_allocation (mword objsize)
1332 bytes_pinned_from_failed_allocation += objsize;
1336 sgen_collection_is_concurrent (void)
1338 switch (current_collection_generation) {
1339 case GENERATION_NURSERY:
1341 case GENERATION_OLD:
1342 return concurrent_collection_in_progress;
1344 g_error ("Invalid current generation %d", current_collection_generation);
1350 sgen_concurrent_collection_in_progress (void)
1352 return concurrent_collection_in_progress;
1356 SgenThreadPoolJob job;
1357 SgenObjectOperations *ops;
1358 SgenGrayQueue *gc_thread_gray_queue;
1363 int job_index, job_split_count;
1366 static ScanCopyContext
1367 scan_copy_context_for_scan_job (void *worker_data_untyped, ScanJob *job)
1369 WorkerData *worker_data = (WorkerData *)worker_data_untyped;
1373 * For jobs enqueued on workers we set the ops at job runtime in order
1374 * to be able to profit from on the fly optimized object ops or other
1375 * object ops changes, like forced concurrent finish.
1377 SGEN_ASSERT (0, sgen_workers_is_worker_thread (mono_native_thread_id_get ()), "We need a context for the scan job");
1378 job->ops = sgen_workers_get_idle_func_object_ops ();
1381 return CONTEXT_FROM_OBJECT_OPERATIONS (job->ops, sgen_workers_get_job_gray_queue (worker_data, job->gc_thread_gray_queue));
1389 } ScanFromRegisteredRootsJob;
1392 job_scan_from_registered_roots (void *worker_data_untyped, SgenThreadPoolJob *job)
1394 ScanFromRegisteredRootsJob *job_data = (ScanFromRegisteredRootsJob*)job;
1395 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1397 scan_from_registered_roots (job_data->heap_start, job_data->heap_end, job_data->root_type, ctx);
1404 } ScanThreadDataJob;
1407 job_scan_thread_data (void *worker_data_untyped, SgenThreadPoolJob *job)
1409 ScanThreadDataJob *job_data = (ScanThreadDataJob*)job;
1410 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1412 sgen_client_scan_thread_data (job_data->heap_start, job_data->heap_end, TRUE, ctx);
1417 SgenPointerQueue *queue;
1418 } ScanFinalizerEntriesJob;
1421 job_scan_finalizer_entries (void *worker_data_untyped, SgenThreadPoolJob *job)
1423 ScanFinalizerEntriesJob *job_data = (ScanFinalizerEntriesJob*)job;
1424 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, &job_data->scan_job);
1426 scan_finalizer_entries (job_data->queue, ctx);
1430 job_scan_wbroots (void *worker_data_untyped, SgenThreadPoolJob *job)
1432 ScanJob *job_data = (ScanJob*)job;
1433 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, job_data);
1435 sgen_wbroots_scan_card_table (ctx);
1439 job_scan_major_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1441 SGEN_TV_DECLARE (atv);
1442 SGEN_TV_DECLARE (btv);
1443 ParallelScanJob *job_data = (ParallelScanJob*)job;
1444 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1446 SGEN_TV_GETTIME (atv);
1447 major_collector.scan_card_table (CARDTABLE_SCAN_GLOBAL, ctx, job_data->job_index, job_data->job_split_count);
1448 SGEN_TV_GETTIME (btv);
1449 time_minor_scan_major_blocks += SGEN_TV_ELAPSED (atv, btv);
1453 job_scan_los_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1455 SGEN_TV_DECLARE (atv);
1456 SGEN_TV_DECLARE (btv);
1457 ParallelScanJob *job_data = (ParallelScanJob*)job;
1458 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1460 SGEN_TV_GETTIME (atv);
1461 sgen_los_scan_card_table (CARDTABLE_SCAN_GLOBAL, ctx, job_data->job_index, job_data->job_split_count);
1462 SGEN_TV_GETTIME (btv);
1463 time_minor_scan_los += SGEN_TV_ELAPSED (atv, btv);
1467 job_scan_major_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1469 ParallelScanJob *job_data = (ParallelScanJob*)job;
1470 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1472 g_assert (concurrent_collection_in_progress);
1473 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx, job_data->job_index, job_data->job_split_count);
1477 job_scan_los_mod_union_card_table (void *worker_data_untyped, SgenThreadPoolJob *job)
1479 ParallelScanJob *job_data = (ParallelScanJob*)job;
1480 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1482 g_assert (concurrent_collection_in_progress);
1483 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION, ctx, job_data->job_index, job_data->job_split_count);
1487 job_major_mod_union_preclean (void *worker_data_untyped, SgenThreadPoolJob *job)
1489 ParallelScanJob *job_data = (ParallelScanJob*)job;
1490 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, (ScanJob*)job_data);
1492 g_assert (concurrent_collection_in_progress);
1494 major_collector.scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx, job_data->job_index, job_data->job_split_count);
1498 job_los_mod_union_preclean (void *worker_data_untyped, SgenThreadPoolJob *job)
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);
1505 sgen_los_scan_card_table (CARDTABLE_SCAN_MOD_UNION_PRECLEAN, ctx, job_data->job_index, job_data->job_split_count);
1509 job_scan_last_pinned (void *worker_data_untyped, SgenThreadPoolJob *job)
1511 ScanJob *job_data = (ScanJob*)job;
1512 ScanCopyContext ctx = scan_copy_context_for_scan_job (worker_data_untyped, job_data);
1514 g_assert (concurrent_collection_in_progress);
1516 sgen_scan_pin_queue_objects (ctx);
1520 workers_finish_callback (void)
1522 ParallelScanJob *psj;
1524 int split_count = sgen_workers_get_job_split_count ();
1526 /* Mod union preclean jobs */
1527 for (i = 0; i < split_count; i++) {
1528 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("preclean major mod union cardtable", job_major_mod_union_preclean, sizeof (ParallelScanJob));
1529 psj->scan_job.gc_thread_gray_queue = NULL;
1531 psj->job_split_count = split_count;
1532 sgen_workers_enqueue_job (&psj->scan_job.job, TRUE);
1535 for (i = 0; i < split_count; i++) {
1536 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("preclean los mod union cardtable", job_los_mod_union_preclean, sizeof (ParallelScanJob));
1537 psj->scan_job.gc_thread_gray_queue = NULL;
1539 psj->job_split_count = split_count;
1540 sgen_workers_enqueue_job (&psj->scan_job.job, TRUE);
1543 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan last pinned", job_scan_last_pinned, sizeof (ScanJob));
1544 sj->gc_thread_gray_queue = NULL;
1545 sgen_workers_enqueue_job (&sj->job, TRUE);
1549 init_gray_queue (SgenGrayQueue *gc_thread_gray_queue, gboolean use_workers)
1552 sgen_workers_init_distribute_gray_queue ();
1553 sgen_gray_object_queue_init (gc_thread_gray_queue, NULL, TRUE);
1557 enqueue_scan_remembered_set_jobs (SgenGrayQueue *gc_thread_gray_queue, SgenObjectOperations *ops, gboolean enqueue)
1559 int i, split_count = sgen_workers_get_job_split_count ();
1562 sj = (ScanJob*)sgen_thread_pool_job_alloc ("scan wbroots", job_scan_wbroots, sizeof (ScanJob));
1564 sj->gc_thread_gray_queue = gc_thread_gray_queue;
1565 sgen_workers_enqueue_job (&sj->job, enqueue);
1567 for (i = 0; i < split_count; i++) {
1568 ParallelScanJob *psj;
1570 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("scan major remsets", job_scan_major_card_table, sizeof (ParallelScanJob));
1571 psj->scan_job.ops = ops;
1572 psj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1574 psj->job_split_count = split_count;
1575 sgen_workers_enqueue_job (&psj->scan_job.job, enqueue);
1577 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("scan LOS remsets", job_scan_los_card_table, sizeof (ParallelScanJob));
1578 psj->scan_job.ops = ops;
1579 psj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1581 psj->job_split_count = split_count;
1582 sgen_workers_enqueue_job (&psj->scan_job.job, enqueue);
1587 enqueue_scan_from_roots_jobs (SgenGrayQueue *gc_thread_gray_queue, char *heap_start, char *heap_end, SgenObjectOperations *ops, gboolean enqueue)
1589 ScanFromRegisteredRootsJob *scrrj;
1590 ScanThreadDataJob *stdj;
1591 ScanFinalizerEntriesJob *sfej;
1593 /* registered roots, this includes static fields */
1595 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots normal", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1596 scrrj->scan_job.ops = ops;
1597 scrrj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1598 scrrj->heap_start = heap_start;
1599 scrrj->heap_end = heap_end;
1600 scrrj->root_type = ROOT_TYPE_NORMAL;
1601 sgen_workers_enqueue_job (&scrrj->scan_job.job, enqueue);
1603 if (current_collection_generation == GENERATION_OLD) {
1604 /* During minors we scan the cardtable for these roots instead */
1605 scrrj = (ScanFromRegisteredRootsJob*)sgen_thread_pool_job_alloc ("scan from registered roots wbarrier", job_scan_from_registered_roots, sizeof (ScanFromRegisteredRootsJob));
1606 scrrj->scan_job.ops = ops;
1607 scrrj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1608 scrrj->heap_start = heap_start;
1609 scrrj->heap_end = heap_end;
1610 scrrj->root_type = ROOT_TYPE_WBARRIER;
1611 sgen_workers_enqueue_job (&scrrj->scan_job.job, enqueue);
1616 stdj = (ScanThreadDataJob*)sgen_thread_pool_job_alloc ("scan thread data", job_scan_thread_data, sizeof (ScanThreadDataJob));
1617 stdj->scan_job.ops = ops;
1618 stdj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1619 stdj->heap_start = heap_start;
1620 stdj->heap_end = heap_end;
1621 sgen_workers_enqueue_job (&stdj->scan_job.job, enqueue);
1623 /* Scan the list of objects ready for finalization. */
1625 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1626 sfej->scan_job.ops = ops;
1627 sfej->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1628 sfej->queue = &fin_ready_queue;
1629 sgen_workers_enqueue_job (&sfej->scan_job.job, enqueue);
1631 sfej = (ScanFinalizerEntriesJob*)sgen_thread_pool_job_alloc ("scan critical finalizer entries", job_scan_finalizer_entries, sizeof (ScanFinalizerEntriesJob));
1632 sfej->scan_job.ops = ops;
1633 sfej->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
1634 sfej->queue = &critical_fin_queue;
1635 sgen_workers_enqueue_job (&sfej->scan_job.job, enqueue);
1639 * Perform a nursery collection.
1641 * Return whether any objects were late-pinned due to being out of memory.
1644 collect_nursery (const char *reason, gboolean is_overflow, SgenGrayQueue *unpin_queue)
1646 gboolean needs_major, is_parallel = FALSE;
1647 mword fragment_total;
1648 SgenGrayQueue gc_thread_gray_queue;
1649 SgenObjectOperations *object_ops_nopar, *object_ops_par = NULL;
1650 ScanCopyContext ctx;
1653 SGEN_TV_DECLARE (last_minor_collection_start_tv);
1654 SGEN_TV_DECLARE (last_minor_collection_end_tv);
1656 if (disable_minor_collections)
1659 TV_GETTIME (last_minor_collection_start_tv);
1660 atv = last_minor_collection_start_tv;
1662 binary_protocol_collection_begin (gc_stats.minor_gc_count, GENERATION_NURSERY);
1664 if (sgen_concurrent_collection_in_progress ()) {
1665 /* FIXME Support parallel nursery collections with concurrent major */
1666 object_ops_nopar = &sgen_minor_collector.serial_ops_with_concurrent_major;
1668 object_ops_nopar = &sgen_minor_collector.serial_ops;
1669 if (sgen_minor_collector.is_parallel && sgen_nursery_size >= SGEN_PARALLEL_MINOR_MIN_NURSERY_SIZE) {
1670 object_ops_par = &sgen_minor_collector.parallel_ops;
1675 if (do_verify_nursery || do_dump_nursery_content)
1676 sgen_debug_verify_nursery (do_dump_nursery_content);
1678 current_collection_generation = GENERATION_NURSERY;
1680 SGEN_ASSERT (0, !sgen_collection_is_concurrent (), "Why is the nursery collection concurrent?");
1682 reset_pinned_from_failed_allocation ();
1684 check_scan_starts ();
1686 sgen_nursery_alloc_prepare_for_minor ();
1691 SGEN_LOG (1, "Start nursery collection %d %p-%p, size: %d", gc_stats.minor_gc_count, nursery_section->data, nursery_section->end_data, (int)(nursery_section->end_data - nursery_section->data));
1693 /* world must be stopped already */
1695 time_minor_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1697 sgen_client_pre_collection_checks ();
1699 major_collector.start_nursery_collection ();
1701 sgen_memgov_minor_collection_start ();
1703 init_gray_queue (&gc_thread_gray_queue, is_parallel);
1704 ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops_nopar, &gc_thread_gray_queue);
1706 gc_stats.minor_gc_count ++;
1708 sgen_process_fin_stage_entries ();
1710 /* pin from pinned handles */
1711 sgen_init_pinning ();
1712 if (concurrent_collection_in_progress)
1713 sgen_init_pinning_for_conc ();
1714 sgen_client_binary_protocol_mark_start (GENERATION_NURSERY);
1715 pin_from_roots (nursery_section->data, nursery_section->end_data, ctx);
1716 /* pin cemented objects */
1717 sgen_pin_cemented_objects ();
1718 /* identify pinned objects */
1719 sgen_optimize_pin_queue ();
1720 sgen_pinning_setup_section (nursery_section);
1722 pin_objects_in_nursery (FALSE, ctx);
1723 sgen_pinning_trim_queue_to_section (nursery_section);
1724 if (concurrent_collection_in_progress)
1725 sgen_finish_pinning_for_conc ();
1727 if (remset_consistency_checks)
1728 sgen_check_remset_consistency ();
1730 if (whole_heap_check_before_collection) {
1731 sgen_clear_nursery_fragments ();
1732 sgen_check_whole_heap (FALSE);
1736 time_minor_pinning += TV_ELAPSED (btv, atv);
1737 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), (long long)TV_ELAPSED (btv, atv));
1738 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1740 remset.start_scan_remsets ();
1742 enqueue_scan_remembered_set_jobs (&gc_thread_gray_queue, is_parallel ? NULL : object_ops_nopar, is_parallel);
1744 /* we don't have complete write barrier yet, so we scan all the old generation sections */
1746 time_minor_scan_remsets += TV_ELAPSED (atv, btv);
1747 SGEN_LOG (2, "Old generation scan: %lld usecs", (long long)TV_ELAPSED (atv, btv));
1749 sgen_pin_stats_report ();
1751 /* FIXME: Why do we do this at this specific, seemingly random, point? */
1752 sgen_client_collecting_minor (&fin_ready_queue, &critical_fin_queue);
1755 time_minor_scan_pinned += TV_ELAPSED (btv, atv);
1757 enqueue_scan_from_roots_jobs (&gc_thread_gray_queue, nursery_section->data, nursery_section->end_data, is_parallel ? NULL : object_ops_nopar, is_parallel);
1760 gray_queue_redirect (&gc_thread_gray_queue);
1761 sgen_workers_start_all_workers (object_ops_nopar, object_ops_par, NULL);
1762 sgen_workers_join ();
1766 time_minor_scan_roots += TV_ELAPSED (atv, btv);
1768 finish_gray_stack (GENERATION_NURSERY, ctx);
1771 time_minor_finish_gray_stack += TV_ELAPSED (btv, atv);
1772 sgen_client_binary_protocol_mark_end (GENERATION_NURSERY);
1774 if (objects_pinned) {
1775 sgen_optimize_pin_queue ();
1776 sgen_pinning_setup_section (nursery_section);
1780 * This is the latest point at which we can do this check, because
1781 * sgen_build_nursery_fragments() unpins nursery objects again.
1783 if (remset_consistency_checks)
1784 sgen_check_remset_consistency ();
1787 if (sgen_max_pause_time) {
1791 duration = (int)(TV_ELAPSED (last_minor_collection_start_tv, btv) / 10000);
1792 if (duration > (sgen_max_pause_time * sgen_max_pause_margin))
1793 sgen_resize_nursery (TRUE);
1795 sgen_resize_nursery (FALSE);
1797 sgen_resize_nursery (FALSE);
1800 /* walk the pin_queue, build up the fragment list of free memory, unmark
1801 * pinned objects as we go, memzero() the empty fragments so they are ready for the
1804 sgen_client_binary_protocol_reclaim_start (GENERATION_NURSERY);
1805 fragment_total = sgen_build_nursery_fragments (nursery_section, unpin_queue);
1806 if (!fragment_total)
1809 /* Clear TLABs for all threads */
1810 sgen_clear_tlabs ();
1812 sgen_client_binary_protocol_reclaim_end (GENERATION_NURSERY);
1814 time_minor_fragment_creation += TV_ELAPSED (atv, btv);
1815 SGEN_LOG (2, "Fragment creation: %lld usecs, %lu bytes available", (long long)TV_ELAPSED (atv, btv), (unsigned long)fragment_total);
1817 if (remset_consistency_checks)
1818 sgen_check_major_refs ();
1820 major_collector.finish_nursery_collection ();
1822 TV_GETTIME (last_minor_collection_end_tv);
1823 gc_stats.minor_gc_time += TV_ELAPSED (last_minor_collection_start_tv, last_minor_collection_end_tv);
1825 sgen_debug_dump_heap ("minor", gc_stats.minor_gc_count - 1, NULL);
1827 /* prepare the pin queue for the next collection */
1828 sgen_finish_pinning ();
1829 if (sgen_have_pending_finalizers ()) {
1830 SGEN_LOG (4, "Finalizer-thread wakeup");
1831 sgen_client_finalize_notify ();
1833 sgen_pin_stats_reset ();
1834 /* clear cemented hash */
1835 sgen_cement_clear_below_threshold ();
1837 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
1839 check_scan_starts ();
1841 binary_protocol_flush_buffers (FALSE);
1843 sgen_memgov_minor_collection_end (reason, is_overflow);
1845 /*objects are late pinned because of lack of memory, so a major is a good call*/
1846 needs_major = objects_pinned > 0;
1847 current_collection_generation = -1;
1850 binary_protocol_collection_end (gc_stats.minor_gc_count - 1, GENERATION_NURSERY, 0, 0);
1852 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1853 sgen_check_nursery_objects_pinned (unpin_queue != NULL);
1859 COPY_OR_MARK_FROM_ROOTS_SERIAL,
1860 COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT,
1861 COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT
1862 } CopyOrMarkFromRootsMode;
1865 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)
1870 /* FIXME: only use these values for the precise scan
1871 * note that to_space pointers should be excluded anyway...
1873 char *heap_start = NULL;
1874 char *heap_end = (char*)-1;
1875 ScanCopyContext ctx = CONTEXT_FROM_OBJECT_OPERATIONS (object_ops_nopar, gc_thread_gray_queue);
1876 gboolean concurrent = mode != COPY_OR_MARK_FROM_ROOTS_SERIAL;
1878 SGEN_ASSERT (0, !!concurrent == !!concurrent_collection_in_progress, "We've been called with the wrong mode.");
1880 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1881 /*This cleans up unused fragments */
1882 sgen_nursery_allocator_prepare_for_pinning ();
1884 if (do_concurrent_checks)
1885 sgen_debug_check_nursery_is_clean ();
1887 /* The concurrent collector doesn't touch the nursery. */
1888 sgen_nursery_alloc_prepare_for_major ();
1893 /* Pinning depends on this */
1894 sgen_clear_nursery_fragments ();
1896 if (whole_heap_check_before_collection)
1897 sgen_check_whole_heap (TRUE);
1900 time_major_pre_collection_fragment_clear += TV_ELAPSED (atv, btv);
1904 sgen_client_pre_collection_checks ();
1906 if (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1907 /* Remsets are not useful for a major collection */
1908 remset.clear_cards ();
1911 sgen_process_fin_stage_entries ();
1914 sgen_init_pinning ();
1915 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT)
1916 sgen_init_pinning_for_conc ();
1917 SGEN_LOG (6, "Collecting pinned addresses");
1918 pin_from_roots ((void*)lowest_heap_address, (void*)highest_heap_address, ctx);
1919 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1920 /* Pin cemented objects that were forced */
1921 sgen_pin_cemented_objects ();
1923 sgen_optimize_pin_queue ();
1924 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
1926 * Cemented objects that are in the pinned list will be marked. When
1927 * marking concurrently we won't mark mod-union cards for these objects.
1928 * Instead they will remain cemented until the next major collection,
1929 * when we will recheck if they are still pinned in the roots.
1931 sgen_cement_force_pinned ();
1934 sgen_client_collecting_major_1 ();
1937 * pin_queue now contains all candidate pointers, sorted and
1938 * uniqued. We must do two passes now to figure out which
1939 * objects are pinned.
1941 * The first is to find within the pin_queue the area for each
1942 * section. This requires that the pin_queue be sorted. We
1943 * also process the LOS objects and pinned chunks here.
1945 * The second, destructive, pass is to reduce the section
1946 * areas to pointers to the actually pinned objects.
1948 SGEN_LOG (6, "Pinning from sections");
1949 /* first pass for the sections */
1950 sgen_find_section_pin_queue_start_end (nursery_section);
1951 /* identify possible pointers to the insize of large objects */
1952 SGEN_LOG (6, "Pinning from large objects");
1953 for (bigobj = los_object_list; bigobj; bigobj = bigobj->next) {
1955 if (sgen_find_optimized_pin_queue_area ((char*)bigobj->data, (char*)bigobj->data + sgen_los_object_size (bigobj), &dummy, &dummy)) {
1956 binary_protocol_pin (bigobj->data, (gpointer)LOAD_VTABLE (bigobj->data), safe_object_get_size (bigobj->data));
1958 if (sgen_los_object_is_pinned (bigobj->data)) {
1959 SGEN_ASSERT (0, mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, "LOS objects can only be pinned here after concurrent marking.");
1962 sgen_los_pin_object (bigobj->data);
1963 if (SGEN_OBJECT_HAS_REFERENCES (bigobj->data))
1964 GRAY_OBJECT_ENQUEUE_SERIAL (gc_thread_gray_queue, bigobj->data, sgen_obj_get_descriptor ((GCObject*)bigobj->data));
1965 sgen_pin_stats_register_object (bigobj->data, GENERATION_OLD);
1966 SGEN_LOG (6, "Marked large object %p (%s) size: %lu from roots", bigobj->data,
1967 sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (bigobj->data)),
1968 (unsigned long)sgen_los_object_size (bigobj));
1970 sgen_client_pinned_los_object (bigobj->data);
1974 pin_objects_in_nursery (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT, ctx);
1975 if (check_nursery_objects_pinned && !sgen_minor_collector.is_split)
1976 sgen_check_nursery_objects_pinned (mode != COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT);
1978 major_collector.pin_objects (gc_thread_gray_queue);
1979 if (old_next_pin_slot)
1980 *old_next_pin_slot = sgen_get_pinned_count ();
1983 time_major_pinning += TV_ELAPSED (atv, btv);
1984 SGEN_LOG (2, "Finding pinned pointers: %zd in %lld usecs", sgen_get_pinned_count (), (long long)TV_ELAPSED (atv, btv));
1985 SGEN_LOG (4, "Start scan with %zd pinned objects", sgen_get_pinned_count ());
1987 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT)
1988 sgen_finish_pinning_for_conc ();
1990 major_collector.init_to_space ();
1992 SGEN_ASSERT (0, sgen_workers_all_done (), "Why are the workers not done when we start or finish a major collection?");
1993 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
1994 if (object_ops_par != NULL)
1995 sgen_workers_set_num_active_workers (0);
1996 if (sgen_workers_have_idle_work ()) {
1998 * We force the finish of the worker with the new object ops context
1999 * which can also do copying. We need to have finished pinning.
2001 sgen_workers_start_all_workers (object_ops_nopar, object_ops_par, NULL);
2003 sgen_workers_join ();
2007 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
2008 main_gc_thread = mono_native_thread_self ();
2011 sgen_client_collecting_major_2 ();
2014 time_major_scan_pinned += TV_ELAPSED (btv, atv);
2016 sgen_client_collecting_major_3 (&fin_ready_queue, &critical_fin_queue);
2018 enqueue_scan_from_roots_jobs (gc_thread_gray_queue, heap_start, heap_end, object_ops_nopar, FALSE);
2021 time_major_scan_roots += TV_ELAPSED (atv, btv);
2024 * We start the concurrent worker after pinning and after we scanned the roots
2025 * in order to make sure that the worker does not finish before handling all
2028 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
2029 sgen_workers_set_num_active_workers (1);
2030 gray_queue_redirect (gc_thread_gray_queue);
2031 if (precleaning_enabled) {
2032 sgen_workers_start_all_workers (object_ops_nopar, object_ops_par, workers_finish_callback);
2034 sgen_workers_start_all_workers (object_ops_nopar, object_ops_par, NULL);
2038 if (mode == COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT) {
2039 int i, split_count = sgen_workers_get_job_split_count ();
2040 gboolean parallel = object_ops_par != NULL;
2042 /* If we're not parallel we finish the collection on the gc thread */
2044 gray_queue_redirect (gc_thread_gray_queue);
2046 /* Mod union card table */
2047 for (i = 0; i < split_count; i++) {
2048 ParallelScanJob *psj;
2050 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("scan mod union cardtable", job_scan_major_mod_union_card_table, sizeof (ParallelScanJob));
2051 psj->scan_job.ops = parallel ? NULL : object_ops_nopar;
2052 psj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
2054 psj->job_split_count = split_count;
2055 sgen_workers_enqueue_job (&psj->scan_job.job, parallel);
2057 psj = (ParallelScanJob*)sgen_thread_pool_job_alloc ("scan LOS mod union cardtable", job_scan_los_mod_union_card_table, sizeof (ParallelScanJob));
2058 psj->scan_job.ops = parallel ? NULL : object_ops_nopar;
2059 psj->scan_job.gc_thread_gray_queue = gc_thread_gray_queue;
2061 psj->job_split_count = split_count;
2062 sgen_workers_enqueue_job (&psj->scan_job.job, parallel);
2067 * If we enqueue a job while workers are running we need to sgen_workers_ensure_awake
2068 * in order to make sure that we are running the idle func and draining all worker
2069 * gray queues. The operation of starting workers implies this, so we start them after
2070 * in order to avoid doing this operation twice. The workers will drain the main gray
2071 * stack that contained roots and pinned objects and also scan the mod union card
2074 sgen_workers_start_all_workers (object_ops_nopar, object_ops_par, NULL);
2075 sgen_workers_join ();
2079 sgen_pin_stats_report ();
2081 if (mode == COPY_OR_MARK_FROM_ROOTS_START_CONCURRENT) {
2082 sgen_finish_pinning ();
2084 sgen_pin_stats_reset ();
2086 if (do_concurrent_checks)
2087 sgen_debug_check_nursery_is_clean ();
2092 major_start_collection (SgenGrayQueue *gc_thread_gray_queue, const char *reason, gboolean concurrent, size_t *old_next_pin_slot)
2094 SgenObjectOperations *object_ops_nopar, *object_ops_par = NULL;
2096 binary_protocol_collection_begin (gc_stats.major_gc_count, GENERATION_OLD);
2098 current_collection_generation = GENERATION_OLD;
2100 sgen_workers_assert_gray_queue_is_empty ();
2103 sgen_cement_reset ();
2106 g_assert (major_collector.is_concurrent);
2107 concurrent_collection_in_progress = TRUE;
2109 object_ops_nopar = &major_collector.major_ops_concurrent_start;
2110 if (major_collector.is_parallel)
2111 object_ops_par = &major_collector.major_ops_conc_par_start;
2114 object_ops_nopar = &major_collector.major_ops_serial;
2117 reset_pinned_from_failed_allocation ();
2119 sgen_memgov_major_collection_start (concurrent, reason);
2121 //count_ref_nonref_objs ();
2122 //consistency_check ();
2124 check_scan_starts ();
2127 SGEN_LOG (1, "Start major collection %d", gc_stats.major_gc_count);
2128 gc_stats.major_gc_count ++;
2130 if (major_collector.start_major_collection)
2131 major_collector.start_major_collection ();
2133 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);
2137 major_finish_collection (SgenGrayQueue *gc_thread_gray_queue, const char *reason, gboolean is_overflow, size_t old_next_pin_slot, gboolean forced)
2139 ScannedObjectCounts counts;
2140 SgenObjectOperations *object_ops_nopar;
2141 mword fragment_total;
2147 if (concurrent_collection_in_progress) {
2148 SgenObjectOperations *object_ops_par = NULL;
2150 object_ops_nopar = &major_collector.major_ops_concurrent_finish;
2151 if (major_collector.is_parallel)
2152 object_ops_par = &major_collector.major_ops_conc_par_finish;
2154 major_copy_or_mark_from_roots (gc_thread_gray_queue, NULL, COPY_OR_MARK_FROM_ROOTS_FINISH_CONCURRENT, object_ops_nopar, object_ops_par);
2156 #ifdef SGEN_DEBUG_INTERNAL_ALLOC
2157 main_gc_thread = NULL;
2160 object_ops_nopar = &major_collector.major_ops_serial;
2163 sgen_workers_assert_gray_queue_is_empty ();
2165 finish_gray_stack (GENERATION_OLD, CONTEXT_FROM_OBJECT_OPERATIONS (object_ops_nopar, gc_thread_gray_queue));
2167 time_major_finish_gray_stack += TV_ELAPSED (btv, atv);
2169 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after joining");
2171 if (objects_pinned) {
2172 g_assert (!concurrent_collection_in_progress);
2175 * This is slow, but we just OOM'd.
2177 * See comment at `sgen_pin_queue_clear_discarded_entries` for how the pin
2178 * queue is laid out at this point.
2180 sgen_pin_queue_clear_discarded_entries (nursery_section, old_next_pin_slot);
2182 * We need to reestablish all pinned nursery objects in the pin queue
2183 * because they're needed for fragment creation. Unpinning happens by
2184 * walking the whole queue, so it's not necessary to reestablish where major
2185 * heap block pins are - all we care is that they're still in there
2188 sgen_optimize_pin_queue ();
2189 sgen_find_section_pin_queue_start_end (nursery_section);
2193 reset_heap_boundaries ();
2194 sgen_update_heap_boundaries ((mword)sgen_get_nursery_start (), (mword)sgen_get_nursery_end ());
2196 /* walk the pin_queue, build up the fragment list of free memory, unmark
2197 * pinned objects as we go, memzero() the empty fragments so they are ready for the
2200 fragment_total = sgen_build_nursery_fragments (nursery_section, NULL);
2201 if (!fragment_total)
2203 SGEN_LOG (4, "Free space in nursery after major %ld", (long)fragment_total);
2205 if (do_concurrent_checks && concurrent_collection_in_progress)
2206 sgen_debug_check_nursery_is_clean ();
2208 /* prepare the pin queue for the next collection */
2209 sgen_finish_pinning ();
2211 /* Clear TLABs for all threads */
2212 sgen_clear_tlabs ();
2214 sgen_pin_stats_reset ();
2216 sgen_cement_clear_below_threshold ();
2218 if (check_mark_bits_after_major_collection)
2219 sgen_check_heap_marked (concurrent_collection_in_progress);
2222 time_major_fragment_creation += TV_ELAPSED (atv, btv);
2224 binary_protocol_sweep_begin (GENERATION_OLD, !major_collector.sweeps_lazily);
2225 sgen_memgov_major_pre_sweep ();
2228 time_major_free_bigobjs += TV_ELAPSED (btv, atv);
2233 time_major_los_sweep += TV_ELAPSED (atv, btv);
2235 major_collector.sweep ();
2237 binary_protocol_sweep_end (GENERATION_OLD, !major_collector.sweeps_lazily);
2240 time_major_sweep += TV_ELAPSED (btv, atv);
2242 sgen_debug_dump_heap ("major", gc_stats.major_gc_count - 1, reason);
2244 if (sgen_have_pending_finalizers ()) {
2245 SGEN_LOG (4, "Finalizer-thread wakeup");
2246 sgen_client_finalize_notify ();
2249 sgen_memgov_major_collection_end (forced, concurrent_collection_in_progress, reason, is_overflow);
2250 current_collection_generation = -1;
2252 memset (&counts, 0, sizeof (ScannedObjectCounts));
2253 major_collector.finish_major_collection (&counts);
2255 sgen_workers_assert_gray_queue_is_empty ();
2257 SGEN_ASSERT (0, sgen_workers_all_done (), "Can't have workers working after major collection has finished");
2258 if (concurrent_collection_in_progress)
2259 concurrent_collection_in_progress = FALSE;
2261 check_scan_starts ();
2263 binary_protocol_flush_buffers (FALSE);
2265 //consistency_check ();
2267 binary_protocol_collection_end (gc_stats.major_gc_count - 1, GENERATION_OLD, counts.num_scanned_objects, counts.num_unique_scanned_objects);
2271 major_do_collection (const char *reason, gboolean is_overflow, gboolean forced)
2273 TV_DECLARE (time_start);
2274 TV_DECLARE (time_end);
2275 size_t old_next_pin_slot;
2276 SgenGrayQueue gc_thread_gray_queue;
2278 if (disable_major_collections)
2281 if (major_collector.get_and_reset_num_major_objects_marked) {
2282 long long num_marked = major_collector.get_and_reset_num_major_objects_marked ();
2283 g_assert (!num_marked);
2286 /* world must be stopped already */
2287 TV_GETTIME (time_start);
2289 init_gray_queue (&gc_thread_gray_queue, FALSE);
2290 major_start_collection (&gc_thread_gray_queue, reason, FALSE, &old_next_pin_slot);
2291 major_finish_collection (&gc_thread_gray_queue, reason, is_overflow, old_next_pin_slot, forced);
2292 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2294 TV_GETTIME (time_end);
2295 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2297 /* FIXME: also report this to the user, preferably in gc-end. */
2298 if (major_collector.get_and_reset_num_major_objects_marked)
2299 major_collector.get_and_reset_num_major_objects_marked ();
2301 return bytes_pinned_from_failed_allocation > 0;
2305 major_start_concurrent_collection (const char *reason)
2307 TV_DECLARE (time_start);
2308 TV_DECLARE (time_end);
2309 long long num_objects_marked;
2310 SgenGrayQueue gc_thread_gray_queue;
2312 if (disable_major_collections)
2315 TV_GETTIME (time_start);
2316 SGEN_TV_GETTIME (time_major_conc_collection_start);
2318 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2319 g_assert (num_objects_marked == 0);
2321 binary_protocol_concurrent_start ();
2323 init_gray_queue (&gc_thread_gray_queue, TRUE);
2324 // FIXME: store reason and pass it when finishing
2325 major_start_collection (&gc_thread_gray_queue, reason, TRUE, NULL);
2326 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2328 num_objects_marked = major_collector.get_and_reset_num_major_objects_marked ();
2330 TV_GETTIME (time_end);
2331 gc_stats.major_gc_time += TV_ELAPSED (time_start, time_end);
2333 current_collection_generation = -1;
2337 * Returns whether the major collection has finished.
2340 major_should_finish_concurrent_collection (void)
2342 return sgen_workers_all_done ();
2346 major_update_concurrent_collection (void)
2348 TV_DECLARE (total_start);
2349 TV_DECLARE (total_end);
2351 TV_GETTIME (total_start);
2353 binary_protocol_concurrent_update ();
2355 major_collector.update_cardtable_mod_union ();
2356 sgen_los_update_cardtable_mod_union ();
2358 TV_GETTIME (total_end);
2359 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end);
2363 major_finish_concurrent_collection (gboolean forced)
2365 SgenGrayQueue gc_thread_gray_queue;
2366 TV_DECLARE (total_start);
2367 TV_DECLARE (total_end);
2369 TV_GETTIME (total_start);
2371 binary_protocol_concurrent_finish ();
2374 * We need to stop all workers since we're updating the cardtable below.
2375 * The workers will be resumed with a finishing pause context to avoid
2376 * additional cardtable and object scanning.
2378 sgen_workers_stop_all_workers ();
2380 SGEN_TV_GETTIME (time_major_conc_collection_end);
2381 gc_stats.major_gc_time_concurrent += SGEN_TV_ELAPSED (time_major_conc_collection_start, time_major_conc_collection_end);
2383 major_collector.update_cardtable_mod_union ();
2384 sgen_los_update_cardtable_mod_union ();
2386 if (mod_union_consistency_check)
2387 sgen_check_mod_union_consistency ();
2389 current_collection_generation = GENERATION_OLD;
2390 sgen_cement_reset ();
2391 init_gray_queue (&gc_thread_gray_queue, FALSE);
2392 major_finish_collection (&gc_thread_gray_queue, "finishing", FALSE, -1, forced);
2393 sgen_gray_object_queue_dispose (&gc_thread_gray_queue);
2395 TV_GETTIME (total_end);
2396 gc_stats.major_gc_time += TV_ELAPSED (total_start, total_end);
2398 current_collection_generation = -1;
2402 * Ensure an allocation request for @size will succeed by freeing enough memory.
2404 * LOCKING: The GC lock MUST be held.
2407 sgen_ensure_free_space (size_t size, int generation)
2409 int generation_to_collect = -1;
2410 const char *reason = NULL;
2412 if (generation == GENERATION_OLD) {
2413 if (sgen_need_major_collection (size)) {
2414 reason = "LOS overflow";
2415 generation_to_collect = GENERATION_OLD;
2418 if (degraded_mode) {
2419 if (sgen_need_major_collection (size)) {
2420 reason = "Degraded mode overflow";
2421 generation_to_collect = GENERATION_OLD;
2423 } else if (sgen_need_major_collection (size)) {
2424 reason = concurrent_collection_in_progress ? "Forced finish concurrent collection" : "Minor allowance";
2425 generation_to_collect = GENERATION_OLD;
2427 generation_to_collect = GENERATION_NURSERY;
2428 reason = "Nursery full";
2432 if (generation_to_collect == -1) {
2433 if (concurrent_collection_in_progress && sgen_workers_all_done ()) {
2434 generation_to_collect = GENERATION_OLD;
2435 reason = "Finish concurrent collection";
2439 if (generation_to_collect == -1)
2441 sgen_perform_collection (size, generation_to_collect, reason, FALSE, TRUE);
2445 * LOCKING: Assumes the GC lock is held.
2448 sgen_perform_collection (size_t requested_size, int generation_to_collect, const char *reason, gboolean wait_to_finish, gboolean stw)
2450 TV_DECLARE (gc_total_start);
2451 TV_DECLARE (gc_total_end);
2452 int overflow_generation_to_collect = -1;
2453 int oldest_generation_collected = generation_to_collect;
2454 const char *overflow_reason = NULL;
2455 gboolean finish_concurrent = concurrent_collection_in_progress && (major_should_finish_concurrent_collection () || generation_to_collect == GENERATION_OLD);
2457 binary_protocol_collection_requested (generation_to_collect, requested_size, wait_to_finish ? 1 : 0);
2459 SGEN_ASSERT (0, generation_to_collect == GENERATION_NURSERY || generation_to_collect == GENERATION_OLD, "What generation is this?");
2462 sgen_stop_world (generation_to_collect);
2464 SGEN_ASSERT (0, sgen_is_world_stopped (), "We can only collect if the world is stopped");
2467 TV_GETTIME (gc_total_start);
2469 // FIXME: extract overflow reason
2470 // FIXME: minor overflow for concurrent case
2471 if (generation_to_collect == GENERATION_NURSERY && !finish_concurrent) {
2472 if (concurrent_collection_in_progress)
2473 major_update_concurrent_collection ();
2475 if (collect_nursery (reason, FALSE, NULL) && !concurrent_collection_in_progress) {
2476 overflow_generation_to_collect = GENERATION_OLD;
2477 overflow_reason = "Minor overflow";
2479 } else if (finish_concurrent) {
2480 major_finish_concurrent_collection (wait_to_finish);
2481 oldest_generation_collected = GENERATION_OLD;
2483 SGEN_ASSERT (0, generation_to_collect == GENERATION_OLD, "We should have handled nursery collections above");
2484 if (major_collector.is_concurrent && !wait_to_finish) {
2485 collect_nursery ("Concurrent start", FALSE, NULL);
2486 major_start_concurrent_collection (reason);
2487 oldest_generation_collected = GENERATION_NURSERY;
2488 } else if (major_do_collection (reason, FALSE, wait_to_finish)) {
2489 overflow_generation_to_collect = GENERATION_NURSERY;
2490 overflow_reason = "Excessive pinning";
2494 if (overflow_generation_to_collect != -1) {
2495 SGEN_ASSERT (0, !concurrent_collection_in_progress, "We don't yet support overflow collections with the concurrent collector");
2498 * We need to do an overflow collection, either because we ran out of memory
2499 * or the nursery is fully pinned.
2502 if (overflow_generation_to_collect == GENERATION_NURSERY)
2503 collect_nursery (overflow_reason, TRUE, NULL);
2505 major_do_collection (overflow_reason, TRUE, wait_to_finish);
2507 oldest_generation_collected = MAX (oldest_generation_collected, overflow_generation_to_collect);
2510 SGEN_LOG (2, "Heap size: %lu, LOS size: %lu", (unsigned long)sgen_gc_get_total_heap_allocation (), (unsigned long)los_memory_usage);
2512 /* this also sets the proper pointers for the next allocation */
2513 if (generation_to_collect == GENERATION_NURSERY && !sgen_can_alloc_size (requested_size)) {
2514 /* TypeBuilder and MonoMethod are killing mcs with fragmentation */
2515 SGEN_LOG (1, "nursery collection didn't find enough room for %zd alloc (%zd pinned)", requested_size, sgen_get_pinned_count ());
2516 sgen_dump_pin_queue ();
2520 TV_GETTIME (gc_total_end);
2521 time_max = MAX (time_max, TV_ELAPSED (gc_total_start, gc_total_end));
2524 sgen_restart_world (oldest_generation_collected);
2528 * ######################################################################
2529 * ######## Memory allocation from the OS
2530 * ######################################################################
2531 * This section of code deals with getting memory from the OS and
2532 * allocating memory for GC-internal data structures.
2533 * Internal memory can be handled with a freelist for small objects.
2539 G_GNUC_UNUSED static void
2540 report_internal_mem_usage (void)
2542 printf ("Internal memory usage:\n");
2543 sgen_report_internal_mem_usage ();
2544 printf ("Pinned memory usage:\n");
2545 major_collector.report_pinned_memory_usage ();
2549 * ######################################################################
2550 * ######## Finalization support
2551 * ######################################################################
2555 * If the object has been forwarded it means it's still referenced from a root.
2556 * If it is pinned it's still alive as well.
2557 * A LOS object is only alive if we have pinned it.
2558 * Return TRUE if @obj is ready to be finalized.
2560 static inline gboolean
2561 sgen_is_object_alive (GCObject *object)
2563 if (ptr_in_nursery (object))
2564 return sgen_nursery_is_object_alive (object);
2566 return sgen_major_is_object_alive (object);
2570 * This function returns true if @object is either alive and belongs to the
2571 * current collection - major collections are full heap, so old gen objects
2572 * are never alive during a minor collection.
2575 sgen_is_object_alive_and_on_current_collection (GCObject *object)
2577 if (ptr_in_nursery (object))
2578 return sgen_nursery_is_object_alive (object);
2580 if (current_collection_generation == GENERATION_NURSERY)
2583 return sgen_major_is_object_alive (object);
2588 sgen_gc_is_object_ready_for_finalization (GCObject *object)
2590 return !sgen_is_object_alive (object);
2594 sgen_queue_finalization_entry (GCObject *obj)
2596 gboolean critical = sgen_client_object_has_critical_finalizer (obj);
2598 sgen_pointer_queue_add (critical ? &critical_fin_queue : &fin_ready_queue, obj);
2600 sgen_client_object_queued_for_finalization (obj);
2604 sgen_object_is_live (GCObject *obj)
2606 return sgen_is_object_alive_and_on_current_collection (obj);
2610 * `System.GC.WaitForPendingFinalizers` first checks `sgen_have_pending_finalizers()` to
2611 * determine whether it can exit quickly. The latter must therefore only return FALSE if
2612 * all finalizers have really finished running.
2614 * `sgen_gc_invoke_finalizers()` first dequeues a finalizable object, and then finalizes it.
2615 * This means that just checking whether the queues are empty leaves the possibility that an
2616 * object might have been dequeued but not yet finalized. That's why we need the additional
2617 * flag `pending_unqueued_finalizer`.
2620 static volatile gboolean pending_unqueued_finalizer = FALSE;
2621 volatile gboolean sgen_suspend_finalizers = FALSE;
2624 sgen_set_suspend_finalizers (void)
2626 sgen_suspend_finalizers = TRUE;
2630 sgen_gc_invoke_finalizers (void)
2634 g_assert (!pending_unqueued_finalizer);
2636 /* FIXME: batch to reduce lock contention */
2637 while (sgen_have_pending_finalizers ()) {
2643 * We need to set `pending_unqueued_finalizer` before dequeing the
2644 * finalizable object.
2646 if (!sgen_pointer_queue_is_empty (&fin_ready_queue)) {
2647 pending_unqueued_finalizer = TRUE;
2648 mono_memory_write_barrier ();
2649 obj = (GCObject *)sgen_pointer_queue_pop (&fin_ready_queue);
2650 } else if (!sgen_pointer_queue_is_empty (&critical_fin_queue)) {
2651 pending_unqueued_finalizer = TRUE;
2652 mono_memory_write_barrier ();
2653 obj = (GCObject *)sgen_pointer_queue_pop (&critical_fin_queue);
2659 SGEN_LOG (7, "Finalizing object %p (%s)", obj, sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (obj)));
2667 /* the object is on the stack so it is pinned */
2668 /*g_print ("Calling finalizer for object: %p (%s)\n", obj, sgen_client_object_safe_name (obj));*/
2669 sgen_client_run_finalize (obj);
2672 if (pending_unqueued_finalizer) {
2673 mono_memory_write_barrier ();
2674 pending_unqueued_finalizer = FALSE;
2681 sgen_have_pending_finalizers (void)
2683 if (sgen_suspend_finalizers)
2685 return pending_unqueued_finalizer || !sgen_pointer_queue_is_empty (&fin_ready_queue) || !sgen_pointer_queue_is_empty (&critical_fin_queue);
2689 * ######################################################################
2690 * ######## registered roots support
2691 * ######################################################################
2695 * We do not coalesce roots.
2698 sgen_register_root (char *start, size_t size, SgenDescriptor descr, int root_type, int source, const char *msg)
2700 RootRecord new_root;
2703 for (i = 0; i < ROOT_TYPE_NUM; ++i) {
2704 RootRecord *root = (RootRecord *)sgen_hash_table_lookup (&roots_hash [i], start);
2705 /* we allow changing the size and the descriptor (for thread statics etc) */
2707 size_t old_size = root->end_root - start;
2708 root->end_root = start + size;
2709 SGEN_ASSERT (0, !!root->root_desc == !!descr, "Can't change whether a root is precise or conservative.");
2710 SGEN_ASSERT (0, root->source == source, "Can't change a root's source identifier.");
2711 SGEN_ASSERT (0, !!root->msg == !!msg, "Can't change a root's message.");
2712 root->root_desc = descr;
2714 roots_size -= old_size;
2720 new_root.end_root = start + size;
2721 new_root.root_desc = descr;
2722 new_root.source = source;
2725 sgen_hash_table_replace (&roots_hash [root_type], start, &new_root, NULL);
2728 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);
2735 sgen_deregister_root (char* addr)
2741 for (root_type = 0; root_type < ROOT_TYPE_NUM; ++root_type) {
2742 if (sgen_hash_table_remove (&roots_hash [root_type], addr, &root))
2743 roots_size -= (root.end_root - addr);
2749 sgen_wbroots_iterate_live_block_ranges (sgen_cardtable_block_callback cb)
2753 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_WBARRIER], void **, start_root, RootRecord *, root) {
2754 cb ((mword)start_root, (mword)root->end_root - (mword)start_root);
2755 } SGEN_HASH_TABLE_FOREACH_END;
2758 /* Root equivalent of sgen_client_cardtable_scan_object */
2760 sgen_wbroot_scan_card_table (void** start_root, mword size, ScanCopyContext ctx)
2762 ScanPtrFieldFunc scan_field_func = ctx.ops->scan_ptr_field;
2763 guint8 *card_data = sgen_card_table_get_card_scan_address ((mword)start_root);
2764 guint8 *card_base = card_data;
2765 mword card_count = sgen_card_table_number_of_cards_in_range ((mword)start_root, size);
2766 guint8 *card_data_end = card_data + card_count;
2767 mword extra_idx = 0;
2768 char *obj_start = sgen_card_table_align_pointer (start_root);
2769 char *obj_end = (char*)start_root + size;
2770 #ifdef SGEN_HAVE_OVERLAPPING_CARDS
2771 guint8 *overflow_scan_end = NULL;
2774 #ifdef SGEN_HAVE_OVERLAPPING_CARDS
2775 /*Check for overflow and if so, setup to scan in two steps*/
2776 if (card_data_end >= SGEN_SHADOW_CARDTABLE_END) {
2777 overflow_scan_end = sgen_shadow_cardtable + (card_data_end - SGEN_SHADOW_CARDTABLE_END);
2778 card_data_end = SGEN_SHADOW_CARDTABLE_END;
2784 card_data = sgen_find_next_card (card_data, card_data_end);
2786 for (; card_data < card_data_end; card_data = sgen_find_next_card (card_data + 1, card_data_end)) {
2787 size_t idx = (card_data - card_base) + extra_idx;
2788 char *start = (char*)(obj_start + idx * CARD_SIZE_IN_BYTES);
2789 char *card_end = start + CARD_SIZE_IN_BYTES;
2790 char *elem = start, *first_elem = start;
2793 * Don't clean first and last card on 32bit systems since they
2794 * may also be part from other roots.
2796 if (card_data != card_base && card_data != (card_data_end - 1))
2797 sgen_card_table_prepare_card_for_scanning (card_data);
2799 card_end = MIN (card_end, obj_end);
2801 if (elem < (char*)start_root)
2802 first_elem = elem = (char*)start_root;
2804 for (; elem < card_end; elem += SIZEOF_VOID_P) {
2805 if (*(GCObject**)elem)
2806 scan_field_func (NULL, (GCObject**)elem, ctx.queue);
2809 binary_protocol_card_scan (first_elem, elem - first_elem);
2812 #ifdef SGEN_HAVE_OVERLAPPING_CARDS
2813 if (overflow_scan_end) {
2814 extra_idx = card_data - card_base;
2815 card_base = card_data = sgen_shadow_cardtable;
2816 card_data_end = overflow_scan_end;
2817 overflow_scan_end = NULL;
2824 sgen_wbroots_scan_card_table (ScanCopyContext ctx)
2829 SGEN_HASH_TABLE_FOREACH (&roots_hash [ROOT_TYPE_WBARRIER], void **, start_root, RootRecord *, root) {
2830 SGEN_ASSERT (0, (root->root_desc & ROOT_DESC_TYPE_MASK) == ROOT_DESC_VECTOR, "Unsupported root type");
2832 sgen_wbroot_scan_card_table (start_root, (mword)root->end_root - (mword)start_root, ctx);
2833 } SGEN_HASH_TABLE_FOREACH_END;
2837 * ######################################################################
2838 * ######## Thread handling (stop/start code)
2839 * ######################################################################
2843 sgen_get_current_collection_generation (void)
2845 return current_collection_generation;
2849 sgen_thread_attach (SgenThreadInfo* info)
2851 info->tlab_start = info->tlab_next = info->tlab_temp_end = info->tlab_real_end = NULL;
2853 sgen_client_thread_attach (info);
2859 sgen_thread_detach_with_lock (SgenThreadInfo *p)
2861 sgen_client_thread_detach_with_lock (p);
2865 * ######################################################################
2866 * ######## Write barriers
2867 * ######################################################################
2871 * Note: the write barriers first do the needed GC work and then do the actual store:
2872 * this way the value is visible to the conservative GC scan after the write barrier
2873 * itself. If a GC interrupts the barrier in the middle, value will be kept alive by
2874 * the conservative scan, otherwise by the remembered set scan.
2878 * mono_gc_wbarrier_arrayref_copy:
2881 mono_gc_wbarrier_arrayref_copy (gpointer dest_ptr, gpointer src_ptr, int count)
2883 HEAVY_STAT (++stat_wbarrier_arrayref_copy);
2884 /*This check can be done without taking a lock since dest_ptr array is pinned*/
2885 if (ptr_in_nursery (dest_ptr) || count <= 0) {
2886 mono_gc_memmove_aligned (dest_ptr, src_ptr, count * sizeof (gpointer));
2890 #ifdef SGEN_HEAVY_BINARY_PROTOCOL
2891 if (binary_protocol_is_heavy_enabled ()) {
2893 for (i = 0; i < count; ++i) {
2894 gpointer dest = (gpointer*)dest_ptr + i;
2895 gpointer obj = *((gpointer*)src_ptr + i);
2897 binary_protocol_wbarrier (dest, obj, (gpointer)LOAD_VTABLE (obj));
2902 remset.wbarrier_arrayref_copy (dest_ptr, src_ptr, count);
2906 * mono_gc_wbarrier_generic_nostore:
2909 mono_gc_wbarrier_generic_nostore (gpointer ptr)
2913 HEAVY_STAT (++stat_wbarrier_generic_store);
2915 sgen_client_wbarrier_generic_nostore_check (ptr);
2917 obj = *(gpointer*)ptr;
2919 binary_protocol_wbarrier (ptr, obj, (gpointer)LOAD_VTABLE (obj));
2922 * We need to record old->old pointer locations for the
2923 * concurrent collector.
2925 if (!ptr_in_nursery (obj) && !concurrent_collection_in_progress) {
2926 SGEN_LOG (8, "Skipping remset at %p", ptr);
2930 SGEN_LOG (8, "Adding remset at %p", ptr);
2932 remset.wbarrier_generic_nostore (ptr);
2936 * mono_gc_wbarrier_generic_store:
2939 mono_gc_wbarrier_generic_store (gpointer ptr, GCObject* value)
2941 SGEN_LOG (8, "Wbarrier store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2942 SGEN_UPDATE_REFERENCE_ALLOW_NULL (ptr, value);
2943 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2944 mono_gc_wbarrier_generic_nostore (ptr);
2945 sgen_dummy_use (value);
2949 * mono_gc_wbarrier_generic_store_atomic:
2950 * Same as \c mono_gc_wbarrier_generic_store but performs the store
2951 * as an atomic operation with release semantics.
2954 mono_gc_wbarrier_generic_store_atomic (gpointer ptr, GCObject *value)
2956 HEAVY_STAT (++stat_wbarrier_generic_store_atomic);
2958 SGEN_LOG (8, "Wbarrier atomic store at %p to %p (%s)", ptr, value, value ? sgen_client_vtable_get_name (SGEN_LOAD_VTABLE (value)) : "null");
2960 InterlockedWritePointer ((volatile gpointer *)ptr, value);
2962 if (ptr_in_nursery (value) || concurrent_collection_in_progress)
2963 mono_gc_wbarrier_generic_nostore (ptr);
2965 sgen_dummy_use (value);
2969 sgen_wbarrier_range_copy (gpointer _dest, gpointer _src, int size)
2971 remset.wbarrier_range_copy (_dest,_src, size);
2975 * ######################################################################
2976 * ######## Other mono public interface functions.
2977 * ######################################################################
2981 sgen_gc_collect (int generation)
2986 sgen_perform_collection (0, generation, "user request", TRUE, TRUE);
2991 sgen_gc_collection_count (int generation)
2993 if (generation == 0)
2994 return gc_stats.minor_gc_count;
2995 return gc_stats.major_gc_count;
2999 sgen_gc_get_used_size (void)
3003 tot = los_memory_usage;
3004 tot += nursery_section->end_data - nursery_section->data;
3005 tot += major_collector.get_used_size ();
3006 /* FIXME: account for pinned objects */
3012 sgen_env_var_error (const char *env_var, const char *fallback, const char *description_format, ...)
3016 va_start (ap, description_format);
3018 fprintf (stderr, "Warning: In environment variable `%s': ", env_var);
3019 vfprintf (stderr, description_format, ap);
3021 fprintf (stderr, " - %s", fallback);
3022 fprintf (stderr, "\n");
3028 parse_double_in_interval (const char *env_var, const char *opt_name, const char *opt, double min, double max, double *result)
3031 double val = strtod (opt, &endptr);
3032 if (endptr == opt) {
3033 sgen_env_var_error (env_var, "Using default value.", "`%s` must be a number.", opt_name);
3036 else if (val < min || val > max) {
3037 sgen_env_var_error (env_var, "Using default value.", "`%s` must be between %.2f - %.2f.", opt_name, min, max);
3045 parse_sgen_minor (const char *opt)
3048 return SGEN_MINOR_DEFAULT;
3050 if (!strcmp (opt, "simple")) {
3051 return SGEN_MINOR_SIMPLE;
3052 } else if (!strcmp (opt, "simple-par")) {
3053 return SGEN_MINOR_SIMPLE_PARALLEL;
3054 } else if (!strcmp (opt, "split")) {
3055 return SGEN_MINOR_SPLIT;
3057 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default instead.", "Unknown minor collector `%s'.", opt);
3058 return SGEN_MINOR_DEFAULT;
3063 parse_sgen_major (const char *opt)
3066 return SGEN_MAJOR_DEFAULT;
3068 if (!strcmp (opt, "marksweep")) {
3069 return SGEN_MAJOR_SERIAL;
3070 } else if (!strcmp (opt, "marksweep-conc")) {
3071 return SGEN_MAJOR_CONCURRENT;
3072 } else if (!strcmp (opt, "marksweep-conc-par")) {
3073 return SGEN_MAJOR_CONCURRENT_PARALLEL;
3075 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default instead.", "Unknown major collector `%s'.", opt);
3076 return SGEN_MAJOR_DEFAULT;
3082 parse_sgen_mode (const char *opt)
3085 return SGEN_MODE_NONE;
3087 if (!strcmp (opt, "balanced")) {
3088 return SGEN_MODE_BALANCED;
3089 } else if (!strcmp (opt, "throughput")) {
3090 return SGEN_MODE_THROUGHPUT;
3091 } else if (!strcmp (opt, "pause") || g_str_has_prefix (opt, "pause:")) {
3092 return SGEN_MODE_PAUSE;
3094 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default configurations.", "Unknown mode `%s'.", opt);
3095 return SGEN_MODE_NONE;
3100 init_sgen_minor (SgenMinor minor)
3103 case SGEN_MINOR_DEFAULT:
3104 case SGEN_MINOR_SIMPLE:
3105 sgen_simple_nursery_init (&sgen_minor_collector, FALSE);
3107 case SGEN_MINOR_SIMPLE_PARALLEL:
3108 sgen_simple_nursery_init (&sgen_minor_collector, TRUE);
3110 case SGEN_MINOR_SPLIT:
3111 sgen_split_nursery_init (&sgen_minor_collector);
3114 g_assert_not_reached ();
3119 init_sgen_major (SgenMajor major)
3121 if (major == SGEN_MAJOR_DEFAULT)
3122 major = DEFAULT_MAJOR;
3125 case SGEN_MAJOR_SERIAL:
3126 sgen_marksweep_init (&major_collector);
3128 case SGEN_MAJOR_CONCURRENT:
3129 sgen_marksweep_conc_init (&major_collector);
3131 case SGEN_MAJOR_CONCURRENT_PARALLEL:
3132 sgen_marksweep_conc_par_init (&major_collector);
3135 g_assert_not_reached ();
3140 * If sgen mode is set, major/minor configuration is fixed. The other gc_params
3141 * are parsed and processed after major/minor initialization, so it can potentially
3142 * override some knobs set by the sgen mode. We can consider locking out additional
3143 * configurations when gc_modes are used.
3146 init_sgen_mode (SgenMode mode)
3148 SgenMinor minor = SGEN_MINOR_DEFAULT;
3149 SgenMajor major = SGEN_MAJOR_DEFAULT;
3152 case SGEN_MODE_BALANCED:
3154 * Use a dynamic parallel nursery with a major concurrent collector.
3155 * This uses the default values for max pause time and nursery size.
3157 minor = SGEN_MINOR_SIMPLE;
3158 major = SGEN_MAJOR_CONCURRENT;
3159 dynamic_nursery = TRUE;
3161 case SGEN_MODE_THROUGHPUT:
3163 * Use concurrent major to let the mutator do more work. Use a larger
3164 * nursery, without pause time constraints, in order to collect more
3165 * objects in parallel and avoid repetitive collection tasks (pinning,
3166 * root scanning etc)
3168 minor = SGEN_MINOR_SIMPLE_PARALLEL;
3169 major = SGEN_MAJOR_CONCURRENT;
3170 dynamic_nursery = TRUE;
3171 sgen_max_pause_time = 0;
3173 case SGEN_MODE_PAUSE:
3175 * Use concurrent major and dynamic nursery with a more
3176 * aggressive shrinking relative to pause times.
3177 * FIXME use parallel minors
3179 minor = SGEN_MINOR_SIMPLE;
3180 major = SGEN_MAJOR_CONCURRENT;
3181 dynamic_nursery = TRUE;
3182 sgen_max_pause_margin = SGEN_PAUSE_MODE_MAX_PAUSE_MARGIN;
3185 g_assert_not_reached ();
3188 init_sgen_minor (minor);
3189 init_sgen_major (major);
3197 SgenMajor sgen_major = SGEN_MAJOR_DEFAULT;
3198 SgenMinor sgen_minor = SGEN_MINOR_DEFAULT;
3199 SgenMode sgen_mode = SGEN_MODE_NONE;
3200 char *params_opts = NULL;
3201 char *debug_opts = NULL;
3202 size_t max_heap = 0;
3203 size_t soft_limit = 0;
3205 gboolean debug_print_allowance = FALSE;
3206 double allowance_ratio = 0, save_target = 0;
3207 gboolean cement_enabled = TRUE;
3210 result = InterlockedCompareExchange (&gc_initialized, -1, 0);
3213 /* already inited */
3216 /* being inited by another thread */
3217 mono_thread_info_usleep (1000);
3220 /* we will init it */
3223 g_assert_not_reached ();
3225 } while (result != 0);
3227 SGEN_TV_GETTIME (sgen_init_timestamp);
3229 #ifdef SGEN_WITHOUT_MONO
3230 mono_thread_smr_init ();
3233 mono_coop_mutex_init (&gc_mutex);
3235 gc_debug_file = stderr;
3237 mono_coop_mutex_init (&sgen_interruption_mutex);
3239 if ((env = g_getenv (MONO_GC_PARAMS_NAME)) || gc_params_options) {
3240 params_opts = g_strdup_printf ("%s,%s", gc_params_options ? gc_params_options : "", env ? env : "");
3245 opts = g_strsplit (params_opts, ",", -1);
3246 for (ptr = opts; *ptr; ++ptr) {
3248 if (g_str_has_prefix (opt, "major=")) {
3249 opt = strchr (opt, '=') + 1;
3250 sgen_major = parse_sgen_major (opt);
3251 } else if (g_str_has_prefix (opt, "minor=")) {
3252 opt = strchr (opt, '=') + 1;
3253 sgen_minor = parse_sgen_minor (opt);
3254 } else if (g_str_has_prefix (opt, "mode=")) {
3255 opt = strchr (opt, '=') + 1;
3256 sgen_mode = parse_sgen_mode (opt);
3264 sgen_init_internal_allocator ();
3265 sgen_init_nursery_allocator ();
3266 sgen_init_fin_weak_hash ();
3267 sgen_init_hash_table ();
3268 sgen_init_descriptors ();
3269 sgen_init_gray_queues ();
3270 sgen_init_allocator ();
3271 sgen_init_gchandles ();
3273 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_SECTION, SGEN_SIZEOF_GC_MEM_SECTION);
3274 sgen_register_fixed_internal_mem_type (INTERNAL_MEM_GRAY_QUEUE, sizeof (GrayQueueSection));
3276 sgen_client_init ();
3278 if (sgen_mode != SGEN_MODE_NONE) {
3279 if (sgen_minor != SGEN_MINOR_DEFAULT || sgen_major != SGEN_MAJOR_DEFAULT)
3280 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Ignoring major/minor configuration", "Major/minor configurations cannot be used with sgen modes");
3281 init_sgen_mode (sgen_mode);
3283 init_sgen_minor (sgen_minor);
3284 init_sgen_major (sgen_major);
3288 gboolean usage_printed = FALSE;
3290 for (ptr = opts; *ptr; ++ptr) {
3292 if (!strcmp (opt, ""))
3294 if (g_str_has_prefix (opt, "major="))
3296 if (g_str_has_prefix (opt, "minor="))
3298 if (g_str_has_prefix (opt, "mode=")) {
3299 if (g_str_has_prefix (opt, "mode=pause:")) {
3300 char *str_pause = strchr (opt, ':') + 1;
3301 int pause = atoi (str_pause);
3303 sgen_max_pause_time = pause;
3305 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default", "Invalid maximum pause time for `pause` sgen mode");
3309 if (g_str_has_prefix (opt, "max-heap-size=")) {
3310 size_t page_size = mono_pagesize ();
3311 size_t max_heap_candidate = 0;
3312 opt = strchr (opt, '=') + 1;
3313 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &max_heap_candidate)) {
3314 max_heap = (max_heap_candidate + page_size - 1) & ~(size_t)(page_size - 1);
3315 if (max_heap != max_heap_candidate)
3316 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Rounding up.", "`max-heap-size` size must be a multiple of %d.", page_size);
3318 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`max-heap-size` must be an integer.");
3322 if (g_str_has_prefix (opt, "soft-heap-limit=")) {
3323 opt = strchr (opt, '=') + 1;
3324 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &soft_limit)) {
3325 if (soft_limit <= 0) {
3326 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be positive.");
3330 sgen_env_var_error (MONO_GC_PARAMS_NAME, NULL, "`soft-heap-limit` must be an integer.");
3334 if (g_str_has_prefix (opt, "nursery-size=")) {
3336 opt = strchr (opt, '=') + 1;
3337 if (*opt && mono_gc_parse_environment_string_extract_number (opt, &val)) {
3338 if ((val & (val - 1))) {
3339 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be a power of two.");
3343 if (val < SGEN_MAX_NURSERY_WASTE) {
3344 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.",
3345 "`nursery-size` must be at least %d bytes.", SGEN_MAX_NURSERY_WASTE);
3349 min_nursery_size = max_nursery_size = val;
3350 dynamic_nursery = FALSE;
3352 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.", "`nursery-size` must be an integer.");
3357 if (g_str_has_prefix (opt, "save-target-ratio=")) {
3359 opt = strchr (opt, '=') + 1;
3360 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "save-target-ratio", opt,
3361 SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO, &val)) {
3366 if (g_str_has_prefix (opt, "default-allowance-ratio=")) {
3368 opt = strchr (opt, '=') + 1;
3369 if (parse_double_in_interval (MONO_GC_PARAMS_NAME, "default-allowance-ratio", opt,
3370 SGEN_MIN_ALLOWANCE_NURSERY_SIZE_RATIO, SGEN_MAX_ALLOWANCE_NURSERY_SIZE_RATIO, &val)) {
3371 allowance_ratio = val;
3376 if (!strcmp (opt, "cementing")) {
3377 cement_enabled = TRUE;
3380 if (!strcmp (opt, "no-cementing")) {
3381 cement_enabled = FALSE;
3385 if (!strcmp (opt, "precleaning")) {
3386 precleaning_enabled = TRUE;
3389 if (!strcmp (opt, "no-precleaning")) {
3390 precleaning_enabled = FALSE;
3394 if (!strcmp (opt, "dynamic-nursery")) {
3395 if (sgen_minor_collector.is_split)
3396 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Using default value.",
3397 "dynamic-nursery not supported with split-nursery.");
3399 dynamic_nursery = TRUE;
3402 if (!strcmp (opt, "no-dynamic-nursery")) {
3403 dynamic_nursery = FALSE;
3407 if (major_collector.handle_gc_param && major_collector.handle_gc_param (opt))
3410 if (sgen_minor_collector.handle_gc_param && sgen_minor_collector.handle_gc_param (opt))
3413 if (sgen_client_handle_gc_param (opt))
3416 sgen_env_var_error (MONO_GC_PARAMS_NAME, "Ignoring.", "Unknown option `%s`.", opt);
3421 fprintf (stderr, "\n%s must be a comma-delimited list of one or more of the following:\n", MONO_GC_PARAMS_NAME);
3422 fprintf (stderr, " max-heap-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
3423 fprintf (stderr, " soft-heap-limit=n (where N is an integer, possibly with a k, m or a g suffix)\n");
3424 fprintf (stderr, " mode=MODE (where MODE is 'balanced', 'throughput' or 'pause[:N]' and N is maximum pause in milliseconds)\n");
3425 fprintf (stderr, " nursery-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
3426 fprintf (stderr, " major=COLLECTOR (where COLLECTOR is `marksweep', `marksweep-conc', `marksweep-par')\n");
3427 fprintf (stderr, " minor=COLLECTOR (where COLLECTOR is `simple' or `split')\n");
3428 fprintf (stderr, " wbarrier=WBARRIER (where WBARRIER is `remset' or `cardtable')\n");
3429 fprintf (stderr, " [no-]cementing\n");
3430 fprintf (stderr, " [no-]dynamic-nursery\n");
3431 if (major_collector.print_gc_param_usage)
3432 major_collector.print_gc_param_usage ();
3433 if (sgen_minor_collector.print_gc_param_usage)
3434 sgen_minor_collector.print_gc_param_usage ();
3435 sgen_client_print_gc_params_usage ();
3436 fprintf (stderr, " Experimental options:\n");
3437 fprintf (stderr, " save-target-ratio=R (where R must be between %.2f - %.2f).\n", SGEN_MIN_SAVE_TARGET_RATIO, SGEN_MAX_SAVE_TARGET_RATIO);
3438 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);
3439 fprintf (stderr, "\n");
3441 usage_printed = TRUE;
3447 g_free (params_opts);
3449 alloc_nursery (dynamic_nursery, min_nursery_size, max_nursery_size);
3451 sgen_pinning_init ();
3452 sgen_cement_init (cement_enabled);
3454 if ((env = g_getenv (MONO_GC_DEBUG_NAME)) || gc_debug_options) {
3455 debug_opts = g_strdup_printf ("%s,%s", gc_debug_options ? gc_debug_options : "", env ? env : "");
3460 gboolean usage_printed = FALSE;
3462 opts = g_strsplit (debug_opts, ",", -1);
3463 for (ptr = opts; ptr && *ptr; ptr ++) {
3465 if (!strcmp (opt, ""))
3467 if (opt [0] >= '0' && opt [0] <= '9') {
3468 gc_debug_level = atoi (opt);
3473 char *rf = g_strdup_printf ("%s.%d", opt, mono_process_current_pid ());
3474 gc_debug_file = fopen (rf, "wb");
3476 gc_debug_file = stderr;
3479 } else if (!strcmp (opt, "print-allowance")) {
3480 debug_print_allowance = TRUE;
3481 } else if (!strcmp (opt, "print-pinning")) {
3482 sgen_pin_stats_enable ();
3483 } else if (!strcmp (opt, "verify-before-allocs")) {
3484 verify_before_allocs = 1;
3485 has_per_allocation_action = TRUE;
3486 } else if (g_str_has_prefix (opt, "max-valloc-size=")) {
3487 size_t max_valloc_size;
3488 char *arg = strchr (opt, '=') + 1;
3489 if (*opt && mono_gc_parse_environment_string_extract_number (arg, &max_valloc_size)) {
3490 mono_valloc_set_limit (max_valloc_size);
3492 sgen_env_var_error (MONO_GC_DEBUG_NAME, NULL, "`max-valloc-size` must be an integer.");
3495 } else if (g_str_has_prefix (opt, "verify-before-allocs=")) {
3496 char *arg = strchr (opt, '=') + 1;
3497 verify_before_allocs = atoi (arg);
3498 has_per_allocation_action = TRUE;
3499 } else if (!strcmp (opt, "collect-before-allocs")) {
3500 collect_before_allocs = 1;
3501 has_per_allocation_action = TRUE;
3502 } else if (g_str_has_prefix (opt, "collect-before-allocs=")) {
3503 char *arg = strchr (opt, '=') + 1;
3504 has_per_allocation_action = TRUE;
3505 collect_before_allocs = atoi (arg);
3506 } else if (!strcmp (opt, "verify-before-collections")) {
3507 whole_heap_check_before_collection = TRUE;
3508 } else if (!strcmp (opt, "check-remset-consistency")) {
3509 remset_consistency_checks = TRUE;
3510 nursery_clear_policy = CLEAR_AT_GC;
3511 } else if (!strcmp (opt, "mod-union-consistency-check")) {
3512 if (!major_collector.is_concurrent) {
3513 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`mod-union-consistency-check` only works with concurrent major collector.");
3516 mod_union_consistency_check = TRUE;
3517 } else if (!strcmp (opt, "check-mark-bits")) {
3518 check_mark_bits_after_major_collection = TRUE;
3519 } else if (!strcmp (opt, "check-nursery-pinned")) {
3520 check_nursery_objects_pinned = TRUE;
3521 } else if (!strcmp (opt, "clear-at-gc")) {
3522 nursery_clear_policy = CLEAR_AT_GC;
3523 } else if (!strcmp (opt, "clear-nursery-at-gc")) {
3524 nursery_clear_policy = CLEAR_AT_GC;
3525 } else if (!strcmp (opt, "clear-at-tlab-creation")) {
3526 nursery_clear_policy = CLEAR_AT_TLAB_CREATION;
3527 } else if (!strcmp (opt, "debug-clear-at-tlab-creation")) {
3528 nursery_clear_policy = CLEAR_AT_TLAB_CREATION_DEBUG;
3529 } else if (!strcmp (opt, "check-scan-starts")) {
3530 do_scan_starts_check = TRUE;
3531 } else if (!strcmp (opt, "verify-nursery-at-minor-gc")) {
3532 do_verify_nursery = TRUE;
3533 } else if (!strcmp (opt, "check-concurrent")) {
3534 if (!major_collector.is_concurrent) {
3535 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "`check-concurrent` only works with concurrent major collectors.");
3538 nursery_clear_policy = CLEAR_AT_GC;
3539 do_concurrent_checks = TRUE;
3540 } else if (!strcmp (opt, "dump-nursery-at-minor-gc")) {
3541 do_dump_nursery_content = TRUE;
3542 } else if (!strcmp (opt, "disable-minor")) {
3543 disable_minor_collections = TRUE;
3544 } else if (!strcmp (opt, "disable-major")) {
3545 disable_major_collections = TRUE;
3546 } else if (g_str_has_prefix (opt, "heap-dump=")) {
3547 char *filename = strchr (opt, '=') + 1;
3548 nursery_clear_policy = CLEAR_AT_GC;
3549 sgen_debug_enable_heap_dump (filename);
3550 } else if (g_str_has_prefix (opt, "binary-protocol=")) {
3551 char *filename = strchr (opt, '=') + 1;
3552 char *colon = strrchr (filename, ':');
3555 if (!mono_gc_parse_environment_string_extract_number (colon + 1, &limit)) {
3556 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring limit.", "Binary protocol file size limit must be an integer.");
3561 binary_protocol_init (filename, (long long)limit);
3562 } else if (!strcmp (opt, "nursery-canaries")) {
3563 do_verify_nursery = TRUE;
3564 enable_nursery_canaries = TRUE;
3565 } else if (!sgen_client_handle_gc_debug (opt)) {
3566 sgen_env_var_error (MONO_GC_DEBUG_NAME, "Ignoring.", "Unknown option `%s`.", opt);
3571 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);
3572 fprintf (stderr, "Valid <option>s are:\n");
3573 fprintf (stderr, " collect-before-allocs[=<n>]\n");
3574 fprintf (stderr, " verify-before-allocs[=<n>]\n");
3575 fprintf (stderr, " max-valloc-size=N (where N is an integer, possibly with a k, m or a g suffix)\n");
3576 fprintf (stderr, " check-remset-consistency\n");
3577 fprintf (stderr, " check-mark-bits\n");
3578 fprintf (stderr, " check-nursery-pinned\n");
3579 fprintf (stderr, " verify-before-collections\n");
3580 fprintf (stderr, " verify-nursery-at-minor-gc\n");
3581 fprintf (stderr, " dump-nursery-at-minor-gc\n");
3582 fprintf (stderr, " disable-minor\n");
3583 fprintf (stderr, " disable-major\n");
3584 fprintf (stderr, " check-concurrent\n");
3585 fprintf (stderr, " clear-[nursery-]at-gc\n");
3586 fprintf (stderr, " clear-at-tlab-creation\n");
3587 fprintf (stderr, " debug-clear-at-tlab-creation\n");
3588 fprintf (stderr, " check-scan-starts\n");
3589 fprintf (stderr, " print-allowance\n");
3590 fprintf (stderr, " print-pinning\n");
3591 fprintf (stderr, " heap-dump=<filename>\n");
3592 fprintf (stderr, " binary-protocol=<filename>[:<file-size-limit>]\n");
3593 fprintf (stderr, " nursery-canaries\n");
3594 sgen_client_print_gc_debug_usage ();
3595 fprintf (stderr, "\n");
3597 usage_printed = TRUE;
3604 g_free (debug_opts);
3606 if (check_mark_bits_after_major_collection)
3607 nursery_clear_policy = CLEAR_AT_GC;
3609 if (major_collector.post_param_init)
3610 major_collector.post_param_init (&major_collector);
3612 if (major_collector.is_concurrent || sgen_minor_collector.is_parallel) {
3613 int num_workers = 1;
3614 if (major_collector.is_parallel || sgen_minor_collector.is_parallel) {
3615 num_workers = mono_cpu_count ();
3616 if (num_workers <= 1) {
3618 major_collector.is_parallel = FALSE;
3619 sgen_minor_collector.is_parallel = FALSE;
3622 if (major_collector.is_concurrent || sgen_minor_collector.is_parallel)
3623 sgen_workers_init (num_workers, (SgenWorkerCallback) major_collector.worker_init_cb);
3626 sgen_memgov_init (max_heap, soft_limit, debug_print_allowance, allowance_ratio, save_target);
3628 memset (&remset, 0, sizeof (remset));
3630 sgen_card_table_init (&remset);
3632 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");
3636 sgen_init_bridge ();
3640 sgen_gc_initialized ()
3642 return gc_initialized > 0;
3646 sgen_get_nursery_clear_policy (void)
3648 return nursery_clear_policy;
3654 mono_coop_mutex_lock (&gc_mutex);
3658 sgen_gc_unlock (void)
3660 mono_coop_mutex_unlock (&gc_mutex);
3664 sgen_major_collector_iterate_live_block_ranges (sgen_cardtable_block_callback callback)
3666 major_collector.iterate_live_block_ranges (callback);
3670 sgen_major_collector_iterate_block_ranges (sgen_cardtable_block_callback callback)
3672 major_collector.iterate_block_ranges (callback);
3676 sgen_get_major_collector (void)
3678 return &major_collector;
3682 sgen_get_minor_collector (void)
3684 return &sgen_minor_collector;
3688 sgen_get_remset (void)
3694 count_cards (long long *major_total, long long *major_marked, long long *los_total, long long *los_marked)
3696 sgen_get_major_collector ()->count_cards (major_total, major_marked);
3697 sgen_los_count_cards (los_total, los_marked);
3700 static gboolean world_is_stopped = FALSE;
3702 /* LOCKING: assumes the GC lock is held */
3704 sgen_stop_world (int generation)
3706 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3708 SGEN_ASSERT (0, !world_is_stopped, "Why are we stopping a stopped world?");
3710 binary_protocol_world_stopping (generation, sgen_timestamp (), (gpointer) (gsize) mono_native_thread_id_get ());
3712 sgen_client_stop_world (generation);
3714 world_is_stopped = TRUE;
3716 if (binary_protocol_is_heavy_enabled ())
3717 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3718 binary_protocol_world_stopped (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3721 /* LOCKING: assumes the GC lock is held */
3723 sgen_restart_world (int generation)
3725 long long major_total = -1, major_marked = -1, los_total = -1, los_marked = -1;
3728 SGEN_ASSERT (0, world_is_stopped, "Why are we restarting a running world?");
3730 if (binary_protocol_is_heavy_enabled ())
3731 count_cards (&major_total, &major_marked, &los_total, &los_marked);
3732 binary_protocol_world_restarting (generation, sgen_timestamp (), major_total, major_marked, los_total, los_marked);
3734 world_is_stopped = FALSE;
3736 sgen_client_restart_world (generation, &stw_time);
3738 binary_protocol_world_restarted (generation, sgen_timestamp ());
3740 if (sgen_client_bridge_need_processing ())
3741 sgen_client_bridge_processing_finish (generation);
3743 sgen_memgov_collection_end (generation, stw_time);
3747 sgen_is_world_stopped (void)
3749 return world_is_stopped;
3753 sgen_check_whole_heap_stw (void)
3755 sgen_stop_world (0);
3756 sgen_clear_nursery_fragments ();
3757 sgen_check_whole_heap (TRUE);
3758 sgen_restart_world (0);
3762 sgen_timestamp (void)
3764 SGEN_TV_DECLARE (timestamp);
3765 SGEN_TV_GETTIME (timestamp);
3766 return SGEN_TV_ELAPSED (sgen_init_timestamp, timestamp);
3769 #endif /* HAVE_SGEN_GC */