mono/metadata/sgen-gc.h

   1 /*
   2  * Copyright 2001-2003 Ximian, Inc
   3  * Copyright 2003-2010 Novell, Inc.
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining
   6  * a copy of this software and associated documentation files (the
   7  * "Software"), to deal in the Software without restriction, including
   8  * without limitation the rights to use, copy, modify, merge, publish,
   9  * distribute, sublicense, and/or sell copies of the Software, and to
  10  * permit persons to whom the Software is furnished to do so, subject to
  11  * the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be
  14  * included in all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  17  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  18  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  19  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
  20  * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
  21  * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
  22  * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  23  */
  24 #ifndef __MONO_SGENGC_H__
  25 #define __MONO_SGENGC_H__
  26
  27 /* pthread impl */
  28 #include "config.h"
  29 #include <glib.h>
  30 #include <pthread.h>
  31 #include <signal.h>
  32 #include <mono/utils/mono-compiler.h>
  33 #include <mono/metadata/class-internals.h>
  34 #include <mono/metadata/object-internals.h>
  35
  36 #define SGEN_PARALLEL_MARK
  37
  38 /*
  39  * Turning on heavy statistics will turn off the managed allocator and
  40  * the managed write barrier.
  41  */
  42 //#define HEAVY_STATISTICS
  43
  44 /*
  45  * If this is set, the nursery is aligned to an address aligned to its size, ie.
  46  * a 1MB nursery will be aligned to an address divisible by 1MB. This allows us to
  47  * speed up ptr_in_nursery () checks which are very frequent. This requires the
  48  * nursery size to be a compile time constant.
  49  */
  50 #define SGEN_ALIGN_NURSERY 1
  51
  52 //#define SGEN_BINARY_PROTOCOL
  53
  54 #define SGEN_MAX_DEBUG_LEVEL 2
  55
  56 #define THREAD_HASH_SIZE 11
  57
  58 #define GC_BITS_PER_WORD (sizeof (mword) * 8)
  59
  60 #define ARCH_THREAD_TYPE pthread_t
  61 #define ARCH_GET_THREAD pthread_self
  62 #define ARCH_THREAD_EQUALS(a,b) pthread_equal (a, b)
  63
  64 #if SIZEOF_VOID_P == 4
  65 typedef guint32 mword;
  66 #else
  67 typedef guint64 mword;
  68 #endif
  69
  70 /* for use with write barriers */
  71 typedef struct _RememberedSet RememberedSet;
  72 struct _RememberedSet {
  73         mword *store_next;
  74         mword *end_set;
  75         RememberedSet *next;
  76         mword data [MONO_ZERO_LEN_ARRAY];
  77 };
  78
  79 /* eventually share with MonoThread? */
  80 typedef struct _SgenThreadInfo SgenThreadInfo;
  81
  82 struct _SgenThreadInfo {
  83         SgenThreadInfo *next;
  84         ARCH_THREAD_TYPE id;
  85         unsigned int stop_count; /* to catch duplicate signals */
  86         int signal;
  87         int skip;
  88         volatile int in_critical_region;
  89         void *stack_end;
  90         void *stack_start;
  91         void *stack_start_limit;
  92         char **tlab_next_addr;
  93         char **tlab_start_addr;
  94         char **tlab_temp_end_addr;
  95         char **tlab_real_end_addr;
  96         gpointer **store_remset_buffer_addr;
  97         long *store_remset_buffer_index_addr;
  98         RememberedSet *remset;
  99         gpointer runtime_data;
 100         gpointer stopped_ip;    /* only valid if the thread is stopped */
 101         MonoDomain *stopped_domain; /* ditto */
 102         gpointer *stopped_regs;     /* ditto */
 103 #ifndef HAVE_KW_THREAD
 104         char *tlab_start;
 105         char *tlab_next;
 106         char *tlab_temp_end;
 107         char *tlab_real_end;
 108         gpointer *store_remset_buffer;
 109         long store_remset_buffer_index;
 110 #endif
 111 };
 112
 113 enum {
 114         MEMORY_ROLE_GEN0,
 115         MEMORY_ROLE_GEN1,
 116         MEMORY_ROLE_PINNED,
 117         MEMORY_ROLE_INTERNAL
 118 };
 119
 120 typedef struct _SgenBlock SgenBlock;
 121 struct _SgenBlock {
 122         void *next;
 123         unsigned char role;
 124 };
 125
 126 /*
 127  * The nursery section and the major copying collector's sections use
 128  * this struct.
 129  */
 130 typedef struct _GCMemSection GCMemSection;
 131 struct _GCMemSection {
 132         SgenBlock block;
 133         char *data;
 134         mword size;
 135         /* pointer where more data could be allocated if it fits */
 136         char *next_data;
 137         char *end_data;
 138         /*
 139          * scan starts is an array of pointers to objects equally spaced in the allocation area
 140          * They let use quickly find pinned objects from pinning pointers.
 141          */
 142         char **scan_starts;
 143         /* in major collections indexes in the pin_queue for objects that pin this section */
 144         void **pin_queue_start;
 145         int pin_queue_num_entries;
 146         unsigned short num_scan_start;
 147         gboolean is_to_space;
 148 };
 149
 150 #define SGEN_SIZEOF_GC_MEM_SECTION      ((sizeof (GCMemSection) + 7) & ~7)
 151
 152 /*
 153  * to quickly find the head of an object pinned by a conservative
 154  * address we keep track of the objects allocated for each
 155  * SGEN_SCAN_START_SIZE memory chunk in the nursery or other memory
 156  * sections. Larger values have less memory overhead and bigger
 157  * runtime cost. 4-8 KB are reasonable values.
 158  */
 159 #define SGEN_SCAN_START_SIZE (4096*2)
 160
 161 /*
 162  * Objects bigger then this go into the large object space.  This size
 163  * has a few constraints.  It must fit into the major heap, which in
 164  * the case of the copying collector means that it must fit into a
 165  * pinned chunk.  It must also play well with the GC descriptors, some
 166  * of which (DESC_TYPE_RUN_LENGTH, DESC_TYPE_SMALL_BITMAP) encode the
 167  * object size.
 168  */
 169 #define SGEN_MAX_SMALL_OBJ_SIZE 8000
 170
 171 /* This is also the MAJOR_SECTION_SIZE for the copying major
 172    collector */
 173 #define SGEN_PINNED_CHUNK_SIZE  (128 * 1024)
 174
 175 #define SGEN_PINNED_CHUNK_FOR_PTR(o)    ((SgenBlock*)(((mword)(o)) & ~(SGEN_PINNED_CHUNK_SIZE - 1)))
 176
 177 typedef struct _SgenPinnedChunk SgenPinnedChunk;
 178
 179 #ifdef __APPLE__
 180 const static int suspend_signal_num = SIGXFSZ;
 181 #else
 182 const static int suspend_signal_num = SIGPWR;
 183 #endif
 184 const static int restart_signal_num = SIGXCPU;
 185
 186 /*
 187  * Recursion is not allowed for the thread lock.
 188  */
 189 #define LOCK_DECLARE(name) pthread_mutex_t name = PTHREAD_MUTEX_INITIALIZER
 190 #define LOCK_INIT(name)
 191 #define LOCK_GC pthread_mutex_lock (&gc_mutex)
 192 #define UNLOCK_GC pthread_mutex_unlock (&gc_mutex)
 193 #define LOCK_INTERRUPTION pthread_mutex_lock (&interruption_mutex)
 194 #define UNLOCK_INTERRUPTION pthread_mutex_unlock (&interruption_mutex)
 195
 196 #ifdef SGEN_PARALLEL_MARK
 197 #define LOCK_GLOBAL_REMSET pthread_mutex_lock (&global_remset_mutex)
 198 #define UNLOCK_GLOBAL_REMSET pthread_mutex_unlock (&global_remset_mutex)
 199 #else
 200 #define LOCK_GLOBAL_REMSET
 201 #define UNLOCK_GLOBAL_REMSET
 202 #endif
 203
 204 #ifdef SGEN_PARALLEL_MARK
 205 #define SGEN_CAS_PTR    InterlockedCompareExchangePointer
 206 #define SGEN_ATOMIC_ADD(x,i)    do {                                    \
 207                 int __old_x;                                            \
 208                 do {                                                    \
 209                         __old_x = (x);                                  \
 210                 } while (InterlockedCompareExchange (&(x), __old_x + (i), __old_x) != __old_x); \
 211         } while (0)
 212 #else
 213 #define SGEN_CAS_PTR(p,n,c)     ((*(void**)(p) == (void*)(c)) ? (*(void**)(p) = (void*)(n), (void*)(c)) : (*(void**)(p)))
 214 #define SGEN_ATOMIC_ADD(x,i)    ((x) += (i))
 215 #endif
 216
 217 /* non-pthread will need to provide their own version of start/stop */
 218 #define USE_SIGNAL_BASED_START_STOP_WORLD 1
 219 /* we intercept pthread_create calls to know which threads exist */
 220 #define USE_PTHREAD_INTERCEPT 1
 221
 222 #ifdef HEAVY_STATISTICS
 223 #define HEAVY_STAT(x)   x
 224
 225 extern long long stat_objects_alloced_degraded;
 226 extern long long stat_bytes_alloced_degraded;
 227 extern long long stat_copy_object_called_major;
 228 extern long long stat_objects_copied_major;
 229 #else
 230 #define HEAVY_STAT(x)
 231 #endif
 232
 233 #define SGEN_ALLOC_ALIGN                8
 234 #define SGEN_ALLOC_ALIGN_BITS   3
 235
 236 #define SGEN_ALIGN_UP(s)                (((s)+(SGEN_ALLOC_ALIGN-1)) & ~(SGEN_ALLOC_ALIGN-1))
 237
 238 #ifdef SGEN_ALIGN_NURSERY
 239 #define SGEN_PTR_IN_NURSERY(p,bits,start,end)   (((mword)(p) & ~((1 << (bits)) - 1)) == (mword)(start))
 240 #else
 241 #define SGEN_PTR_IN_NURSERY(p,bits,start,end)   ((char*)(p) >= (start) && (char*)(p) < (end))
 242 #endif
 243
 244 /* Structure that corresponds to a MonoVTable: desc is a mword so requires
 245  * no cast from a pointer to an integer
 246  */
 247 typedef struct {
 248         MonoClass *klass;
 249         mword desc;
 250 } GCVTable;
 251
 252 /* these bits are set in the object vtable: we could merge them since an object can be
 253  * either pinned or forwarded but not both.
 254  * We store them in the vtable slot because the bits are used in the sync block for
 255  * other purposes: if we merge them and alloc the sync blocks aligned to 8 bytes, we can change
 256  * this and use bit 3 in the syncblock (with the lower two bits both set for forwarded, that
 257  * would be an invalid combination for the monitor and hash code).
 258  * The values are already shifted.
 259  * The forwarding address is stored in the sync block.
 260  */
 261 #define SGEN_FORWARDED_BIT 1
 262 #define SGEN_PINNED_BIT 2
 263 #define SGEN_VTABLE_BITS_MASK 0x3
 264
 265 /* returns NULL if not forwarded, or the forwarded address */
 266 #define SGEN_OBJECT_IS_FORWARDED(obj) (((mword*)(obj))[0] & SGEN_FORWARDED_BIT ? (void*)(((mword*)(obj))[0] & ~SGEN_VTABLE_BITS_MASK) : NULL)
 267 #define SGEN_OBJECT_IS_PINNED(obj) (((mword*)(obj))[0] & SGEN_PINNED_BIT)
 268
 269 /* set the forwarded address fw_addr for object obj */
 270 #define SGEN_FORWARD_OBJECT(obj,fw_addr) do {                           \
 271                 ((mword*)(obj))[0] = (mword)(fw_addr) | SGEN_FORWARDED_BIT; \
 272         } while (0)
 273 #define SGEN_PIN_OBJECT(obj) do {       \
 274                 ((mword*)(obj))[0] |= SGEN_PINNED_BIT;  \
 275         } while (0)
 276 #define SGEN_UNPIN_OBJECT(obj) do {     \
 277                 ((mword*)(obj))[0] &= ~SGEN_PINNED_BIT; \
 278         } while (0)
 279
 280 /*
 281  * Since we set bits in the vtable, use the macro to load it from the pointer to
 282  * an object that is potentially pinned.
 283  */
 284 #define SGEN_LOAD_VTABLE(addr) ((*(mword*)(addr)) & ~SGEN_VTABLE_BITS_MASK)
 285
 286 /*
 287  * ######################################################################
 288  * ########  GC descriptors
 289  * ######################################################################
 290  * Used to quickly get the info the GC needs about an object: size and
 291  * where the references are held.
 292  */
 293 #define OBJECT_HEADER_WORDS (sizeof(MonoObject)/sizeof(gpointer))
 294 #define LOW_TYPE_BITS 3
 295 #define SMALL_BITMAP_SHIFT 16
 296 #define SMALL_BITMAP_SIZE (GC_BITS_PER_WORD - SMALL_BITMAP_SHIFT)
 297 #define VECTOR_INFO_SHIFT 14
 298 #define VECTOR_ELSIZE_SHIFT 3
 299 #define LARGE_BITMAP_SIZE (GC_BITS_PER_WORD - LOW_TYPE_BITS)
 300 #define MAX_ELEMENT_SIZE 0x3ff
 301 #define VECTOR_SUBTYPE_PTRFREE (DESC_TYPE_V_PTRFREE << VECTOR_INFO_SHIFT)
 302 #define VECTOR_SUBTYPE_REFS    (DESC_TYPE_V_REFS << VECTOR_INFO_SHIFT)
 303 #define VECTOR_SUBTYPE_RUN_LEN (DESC_TYPE_V_RUN_LEN << VECTOR_INFO_SHIFT)
 304 #define VECTOR_SUBTYPE_BITMAP  (DESC_TYPE_V_BITMAP << VECTOR_INFO_SHIFT)
 305
 306 /* objects are aligned to 8 bytes boundaries
 307  * A descriptor is a pointer in MonoVTable, so 32 or 64 bits of size.
 308  * The low 3 bits define the type of the descriptor. The other bits
 309  * depend on the type.
 310  * As a general rule the 13 remaining low bits define the size, either
 311  * of the whole object or of the elements in the arrays. While for objects
 312  * the size is already in bytes, for arrays we need to shift, because
 313  * array elements might be smaller than 8 bytes. In case of arrays, we
 314  * use two bits to describe what the additional high bits represents,
 315  * so the default behaviour can handle element sizes less than 2048 bytes.
 316  * The high 16 bits, if 0 it means the object is pointer-free.
 317  * This design should make it easy and fast to skip over ptr-free data.
 318  * The first 4 types should cover >95% of the objects.
 319  * Note that since the size of objects is limited to 64K, larger objects
 320  * will be allocated in the large object heap.
 321  * If we want 4-bytes alignment, we need to put vector and small bitmap
 322  * inside complex.
 323  */
 324 enum {
 325         /*
 326          * We don't use 0 so that 0 isn't a valid GC descriptor.  No
 327          * deep reason for this other than to be able to identify a
 328          * non-inited descriptor for debugging.
 329          *
 330          * If an object contains no references, its GC descriptor is
 331          * always DESC_TYPE_RUN_LENGTH, without a size, no exceptions.
 332          * This is so that we can quickly check for that in
 333          * copy_object_no_checks(), without having to fetch the
 334          * object's class.
 335          */
 336         DESC_TYPE_RUN_LENGTH = 1, /* 15 bits aligned byte size | 1-3 (offset, numptr) bytes tuples */
 337         DESC_TYPE_SMALL_BITMAP, /* 15 bits aligned byte size | 16-48 bit bitmap */
 338         DESC_TYPE_COMPLEX,      /* index for bitmap into complex_descriptors */
 339         DESC_TYPE_VECTOR,       /* 10 bits element size | 1 bit array | 2 bits desc | element desc */
 340         DESC_TYPE_ARRAY,        /* 10 bits element size | 1 bit array | 2 bits desc | element desc */
 341         DESC_TYPE_LARGE_BITMAP, /* | 29-61 bitmap bits */
 342         DESC_TYPE_COMPLEX_ARR,  /* index for bitmap into complex_descriptors */
 343         /* subtypes for arrays and vectors */
 344         DESC_TYPE_V_PTRFREE = 0,/* there are no refs: keep first so it has a zero value  */
 345         DESC_TYPE_V_REFS,       /* all the array elements are refs */
 346         DESC_TYPE_V_RUN_LEN,    /* elements are run-length encoded as DESC_TYPE_RUN_LENGTH */
 347         DESC_TYPE_V_BITMAP      /* elements are as the bitmap in DESC_TYPE_SMALL_BITMAP */
 348 };
 349
 350 #define SGEN_VTABLE_HAS_REFERENCES(vt)  (((MonoVTable*)(vt))->gc_descr != (void*)DESC_TYPE_RUN_LENGTH)
 351
 352 /* helper macros to scan and traverse objects, macros because we resue them in many functions */
 353 #define OBJ_RUN_LEN_SIZE(size,desc,obj) do { \
 354                 (size) = ((desc) & 0xfff8) >> 1;        \
 355     } while (0)
 356
 357 #define OBJ_BITMAP_SIZE(size,desc,obj) do { \
 358                 (size) = ((desc) & 0xfff8) >> 1;        \
 359     } while (0)
 360
 361 //#define PREFETCH(addr) __asm__ __volatile__ ("     prefetchnta     %0": : "m"(*(char *)(addr)))
 362 #define PREFETCH(addr)
 363
 364 /* code using these macros must define a HANDLE_PTR(ptr) macro that does the work */
 365 #define OBJ_RUN_LEN_FOREACH_PTR(desc,obj)       do {    \
 366                 if ((desc) & 0xffff0000) {      \
 367                         /* there are pointers */        \
 368                         void **_objptr_end;     \
 369                         void **_objptr = (void**)(obj); \
 370                         _objptr += ((desc) >> 16) & 0xff;       \
 371                         _objptr_end = _objptr + (((desc) >> 24) & 0xff);        \
 372                         while (_objptr < _objptr_end) { \
 373                                 HANDLE_PTR (_objptr, (obj));    \
 374                                 _objptr++;      \
 375                         }       \
 376                 }       \
 377         } while (0)
 378
 379 /* a bitmap desc means that there are pointer references or we'd have
 380  * choosen run-length, instead: add an assert to check.
 381  */
 382 #define OBJ_BITMAP_FOREACH_PTR(desc,obj)        do {    \
 383                 /* there are pointers */        \
 384                 void **_objptr = (void**)(obj); \
 385                 gsize _bmap = (desc) >> 16;     \
 386                 _objptr += OBJECT_HEADER_WORDS; \
 387                 while (_bmap) { \
 388                         if ((_bmap & 1)) {      \
 389                                 HANDLE_PTR (_objptr, (obj));    \
 390                         }       \
 391                         _bmap >>= 1;    \
 392                         ++_objptr;      \
 393                 }       \
 394         } while (0)
 395
 396 #define OBJ_LARGE_BITMAP_FOREACH_PTR(vt,obj)    do {    \
 397                 /* there are pointers */        \
 398                 void **_objptr = (void**)(obj); \
 399                 gsize _bmap = (vt)->desc >> LOW_TYPE_BITS;      \
 400                 _objptr += OBJECT_HEADER_WORDS; \
 401                 while (_bmap) { \
 402                         if ((_bmap & 1)) {      \
 403                                 HANDLE_PTR (_objptr, (obj));    \
 404                         }       \
 405                         _bmap >>= 1;    \
 406                         ++_objptr;      \
 407                 }       \
 408         } while (0)
 409
 410 gsize* mono_sgen_get_complex_descriptor (GCVTable *vt) MONO_INTERNAL;
 411
 412 #define OBJ_COMPLEX_FOREACH_PTR(vt,obj) do {    \
 413                 /* there are pointers */        \
 414                 void **_objptr = (void**)(obj); \
 415                 gsize *bitmap_data = mono_sgen_get_complex_descriptor ((vt)); \
 416                 int bwords = (*bitmap_data) - 1;        \
 417                 void **start_run = _objptr;     \
 418                 bitmap_data++;  \
 419                 if (0) {        \
 420                         MonoObject *myobj = (MonoObject*)obj;   \
 421                         g_print ("found %d at %p (0x%zx): %s.%s\n", bwords, (obj), (vt)->desc, myobj->vtable->klass->name_space, myobj->vtable->klass->name);   \
 422                 }       \
 423                 while (bwords-- > 0) {  \
 424                         gsize _bmap = *bitmap_data++;   \
 425                         _objptr = start_run;    \
 426                         /*g_print ("bitmap: 0x%x/%d at %p\n", _bmap, bwords, _objptr);*/        \
 427                         while (_bmap) { \
 428                                 if ((_bmap & 1)) {      \
 429                                         HANDLE_PTR (_objptr, (obj));    \
 430                                 }       \
 431                                 _bmap >>= 1;    \
 432                                 ++_objptr;      \
 433                         }       \
 434                         start_run += GC_BITS_PER_WORD;  \
 435                 }       \
 436         } while (0)
 437
 438 /* this one is untested */
 439 #define OBJ_COMPLEX_ARR_FOREACH_PTR(vt,obj)     do {    \
 440                 /* there are pointers */        \
 441                 gsize *mbitmap_data = mono_sgen_get_complex_descriptor ((vt)); \
 442                 int mbwords = (*mbitmap_data++) - 1;    \
 443                 int el_size = mono_array_element_size (vt->klass);      \
 444                 char *e_start = (char*)(obj) +  G_STRUCT_OFFSET (MonoArray, vector);    \
 445                 char *e_end = e_start + el_size * mono_array_length_fast ((MonoArray*)(obj));   \
 446                 if (0)                                                  \
 447                         g_print ("found %d at %p (0x%zx): %s.%s\n", mbwords, (obj), (vt)->desc, vt->klass->name_space, vt->klass->name); \
 448                 while (e_start < e_end) {       \
 449                         void **_objptr = (void**)e_start;       \
 450                         gsize *bitmap_data = mbitmap_data;      \
 451                         unsigned int bwords = mbwords;  \
 452                         while (bwords-- > 0) {  \
 453                                 gsize _bmap = *bitmap_data++;   \
 454                                 void **start_run = _objptr;     \
 455                                 /*g_print ("bitmap: 0x%x\n", _bmap);*/  \
 456                                 while (_bmap) { \
 457                                         if ((_bmap & 1)) {      \
 458                                                 HANDLE_PTR (_objptr, (obj));    \
 459                                         }       \
 460                                         _bmap >>= 1;    \
 461                                         ++_objptr;      \
 462                                 }       \
 463                                 _objptr = start_run + GC_BITS_PER_WORD; \
 464                         }       \
 465                         e_start += el_size;     \
 466                 }       \
 467         } while (0)
 468
 469 #define OBJ_VECTOR_FOREACH_PTR(vt,obj)  do {    \
 470                 /* note: 0xffffc000 excludes DESC_TYPE_V_PTRFREE */     \
 471                 if ((vt)->desc & 0xffffc000) {  \
 472                         int el_size = ((vt)->desc >> 3) & MAX_ELEMENT_SIZE;     \
 473                         /* there are pointers */        \
 474                         int etype = (vt)->desc & 0xc000;        \
 475                         if (etype == (DESC_TYPE_V_REFS << 14)) {        \
 476                                 void **p = (void**)((char*)(obj) + G_STRUCT_OFFSET (MonoArray, vector));        \
 477                                 void **end_refs = (void**)((char*)p + el_size * mono_array_length_fast ((MonoArray*)(obj)));    \
 478                                 /* Note: this code can handle also arrays of struct with only references in them */     \
 479                                 while (p < end_refs) {  \
 480                                         HANDLE_PTR (p, (obj));  \
 481                                         ++p;    \
 482                                 }       \
 483                         } else if (etype == DESC_TYPE_V_RUN_LEN << 14) {        \
 484                                 int offset = ((vt)->desc >> 16) & 0xff; \
 485                                 int num_refs = ((vt)->desc >> 24) & 0xff;       \
 486                                 char *e_start = (char*)(obj) + G_STRUCT_OFFSET (MonoArray, vector);     \
 487                                 char *e_end = e_start + el_size * mono_array_length_fast ((MonoArray*)(obj));   \
 488                                 while (e_start < e_end) {       \
 489                                         void **p = (void**)e_start;     \
 490                                         int i;  \
 491                                         p += offset;    \
 492                                         for (i = 0; i < num_refs; ++i) {        \
 493                                                 HANDLE_PTR (p + i, (obj));      \
 494                                         }       \
 495                                         e_start += el_size;     \
 496                                 }       \
 497                         } else if (etype == DESC_TYPE_V_BITMAP << 14) { \
 498                                 char *e_start = (char*)(obj) +  G_STRUCT_OFFSET (MonoArray, vector);    \
 499                                 char *e_end = e_start + el_size * mono_array_length_fast ((MonoArray*)(obj));   \
 500                                 while (e_start < e_end) {       \
 501                                         void **p = (void**)e_start;     \
 502                                         gsize _bmap = (vt)->desc >> 16; \
 503                                         /* Note: there is no object header here to skip */      \
 504                                         while (_bmap) { \
 505                                                 if ((_bmap & 1)) {      \
 506                                                         HANDLE_PTR (p, (obj));  \
 507                                                 }       \
 508                                                 _bmap >>= 1;    \
 509                                                 ++p;    \
 510                                         }       \
 511                                         e_start += el_size;     \
 512                                 }       \
 513                         }       \
 514                 }       \
 515         } while (0)
 516
 517 typedef struct _SgenInternalAllocator SgenInternalAllocator;
 518
 519 #define SGEN_GRAY_QUEUE_SECTION_SIZE    (128 - 3)
 520
 521 /*
 522  * This is a stack now instead of a queue, so the most recently added items are removed
 523  * first, improving cache locality, and keeping the stack size manageable.
 524  */
 525 typedef struct _GrayQueueSection GrayQueueSection;
 526 struct _GrayQueueSection {
 527         int end;
 528         GrayQueueSection *next;
 529         char *objects [SGEN_GRAY_QUEUE_SECTION_SIZE];
 530 };
 531
 532 typedef struct _SgenGrayQueue SgenGrayQueue;
 533
 534 typedef void (*GrayQueueAllocPrepareFunc) (SgenGrayQueue*);
 535
 536 struct _SgenGrayQueue {
 537         SgenInternalAllocator *allocator;
 538         GrayQueueSection *first;
 539         GrayQueueSection *free_list;
 540         int balance;
 541         GrayQueueAllocPrepareFunc alloc_prepare_func;
 542         void *alloc_prepare_data;
 543 };
 544
 545 #if SGEN_MAX_DEBUG_LEVEL >= 9
 546 #define GRAY_OBJECT_ENQUEUE gray_object_enqueue
 547 #define GRAY_OBJECT_DEQUEUE(queue,o) ((o) = gray_object_dequeue ((queue)))
 548 #else
 549 #define GRAY_OBJECT_ENQUEUE(queue,o) do {                               \
 550                 if (G_UNLIKELY (!(queue)->first || (queue)->first->end == SGEN_GRAY_QUEUE_SECTION_SIZE)) \
 551                         mono_sgen_gray_object_enqueue ((queue), (o));   \
 552                 else                                                    \
 553                         (queue)->first->objects [(queue)->first->end++] = (o); \
 554         } while (0)
 555 #define GRAY_OBJECT_DEQUEUE(queue,o) do {                               \
 556                 if (!(queue)->first)                                    \
 557                         (o) = NULL;                                     \
 558                 else if (G_UNLIKELY ((queue)->first->end == 1))         \
 559                         (o) = mono_sgen_gray_object_dequeue ((queue));          \
 560                 else                                                    \
 561                         (o) = (queue)->first->objects [--(queue)->first->end]; \
 562         } while (0)
 563 #endif
 564
 565 void mono_sgen_gray_object_enqueue (SgenGrayQueue *queue, char *obj) MONO_INTERNAL;
 566 char* mono_sgen_gray_object_dequeue (SgenGrayQueue *queue) MONO_INTERNAL;
 567
 568 typedef void (*IterateObjectCallbackFunc) (char*, size_t, void*);
 569
 570 void* mono_sgen_alloc_os_memory (size_t size, int activate) MONO_INTERNAL;
 571 void* mono_sgen_alloc_os_memory_aligned (mword size, mword alignment, gboolean activate) MONO_INTERNAL;
 572 void mono_sgen_free_os_memory (void *addr, size_t size) MONO_INTERNAL;
 573
 574 int mono_sgen_thread_handshake (int signum) MONO_INTERNAL;
 575 SgenThreadInfo* mono_sgen_thread_info_lookup (ARCH_THREAD_TYPE id) MONO_INTERNAL;
 576 SgenThreadInfo** mono_sgen_get_thread_table (void) MONO_INTERNAL;
 577 void mono_sgen_wait_for_suspend_ack (int count) MONO_INTERNAL;
 578
 579 gboolean mono_sgen_is_worker_thread (pthread_t thread) MONO_INTERNAL;
 580
 581 void mono_sgen_update_heap_boundaries (mword low, mword high) MONO_INTERNAL;
 582
 583 void mono_sgen_register_major_sections_alloced (int num_sections) MONO_INTERNAL;
 584 mword mono_sgen_get_minor_collection_allowance (void) MONO_INTERNAL;
 585
 586 void mono_sgen_scan_area_with_callback (char *start, char *end, IterateObjectCallbackFunc callback, void *data) MONO_INTERNAL;
 587 void mono_sgen_check_section_scan_starts (GCMemSection *section) MONO_INTERNAL;
 588
 589 /* Keep in sync with mono_sgen_dump_internal_mem_usage() in dump_heap()! */
 590 enum {
 591         INTERNAL_MEM_MANAGED,
 592         INTERNAL_MEM_PIN_QUEUE,
 593         INTERNAL_MEM_FRAGMENT,
 594         INTERNAL_MEM_SECTION,
 595         INTERNAL_MEM_SCAN_STARTS,
 596         INTERNAL_MEM_FIN_TABLE,
 597         INTERNAL_MEM_FINALIZE_ENTRY,
 598         INTERNAL_MEM_DISLINK_TABLE,
 599         INTERNAL_MEM_DISLINK,
 600         INTERNAL_MEM_ROOTS_TABLE,
 601         INTERNAL_MEM_ROOT_RECORD,
 602         INTERNAL_MEM_STATISTICS,
 603         INTERNAL_MEM_REMSET,
 604         INTERNAL_MEM_GRAY_QUEUE,
 605         INTERNAL_MEM_STORE_REMSET,
 606         INTERNAL_MEM_MS_TABLES,
 607         INTERNAL_MEM_MS_BLOCK_INFO,
 608         INTERNAL_MEM_EPHEMERON_LINK,
 609         INTERNAL_MEM_WORKER_DATA,
 610         INTERNAL_MEM_MAX
 611 };
 612
 613 #define SGEN_INTERNAL_FREELIST_NUM_SLOTS        30
 614
 615 struct _SgenInternalAllocator {
 616         SgenPinnedChunk *chunk_list;
 617         SgenPinnedChunk *free_lists [SGEN_INTERNAL_FREELIST_NUM_SLOTS];
 618         void *delayed_free_lists [SGEN_INTERNAL_FREELIST_NUM_SLOTS];
 619         long small_internal_mem_bytes [INTERNAL_MEM_MAX];
 620 };
 621
 622 void mono_sgen_init_internal_allocator (void) MONO_INTERNAL;
 623
 624 SgenInternalAllocator* mono_sgen_get_unmanaged_allocator (void) MONO_INTERNAL;
 625
 626 const char* mono_sgen_internal_mem_type_name (int type) MONO_INTERNAL;
 627 void mono_sgen_report_internal_mem_usage (void) MONO_INTERNAL;
 628 void mono_sgen_report_internal_mem_usage_full (SgenInternalAllocator *alc) MONO_INTERNAL;
 629 void mono_sgen_dump_internal_mem_usage (FILE *heap_dump_file) MONO_INTERNAL;
 630 void mono_sgen_dump_section (GCMemSection *section, const char *type) MONO_INTERNAL;
 631 void mono_sgen_dump_occupied (char *start, char *end, char *section_start) MONO_INTERNAL;
 632
 633 void mono_sgen_register_moved_object (void *obj, void *destination) MONO_INTERNAL;
 634
 635 void mono_sgen_register_fixed_internal_mem_type (int type, size_t size) MONO_INTERNAL;
 636
 637 void* mono_sgen_alloc_internal (int type) MONO_INTERNAL;
 638 void mono_sgen_free_internal (void *addr, int type) MONO_INTERNAL;
 639
 640 void* mono_sgen_alloc_internal_dynamic (size_t size, int type) MONO_INTERNAL;
 641 void mono_sgen_free_internal_dynamic (void *addr, size_t size, int type) MONO_INTERNAL;
 642
 643 void* mono_sgen_alloc_internal_fixed (SgenInternalAllocator *allocator, int type) MONO_INTERNAL;
 644 void mono_sgen_free_internal_fixed (SgenInternalAllocator *allocator, void *addr, int type) MONO_INTERNAL;
 645
 646 void* mono_sgen_alloc_internal_full (SgenInternalAllocator *allocator, size_t size, int type) MONO_INTERNAL;
 647 void mono_sgen_free_internal_full (SgenInternalAllocator *allocator, void *addr, size_t size, int type) MONO_INTERNAL;
 648
 649 void mono_sgen_free_internal_delayed (void *addr, int type, SgenInternalAllocator *thread_allocator) MONO_INTERNAL;
 650
 651 void mono_sgen_debug_printf (int level, const char *format, ...) MONO_INTERNAL;
 652
 653 void mono_sgen_internal_scan_objects (SgenInternalAllocator *alc, IterateObjectCallbackFunc callback, void *callback_data) MONO_INTERNAL;
 654 void mono_sgen_internal_scan_pinned_objects (SgenInternalAllocator *alc, IterateObjectCallbackFunc callback, void *callback_data) MONO_INTERNAL;
 655
 656 void** mono_sgen_find_optimized_pin_queue_area (void *start, void *end, int *num) MONO_INTERNAL;
 657 void mono_sgen_find_section_pin_queue_start_end (GCMemSection *section) MONO_INTERNAL;
 658 void mono_sgen_pin_objects_in_section (GCMemSection *section, SgenGrayQueue *queue) MONO_INTERNAL;
 659
 660 void mono_sgen_pin_stats_register_object (char *obj, size_t size);
 661
 662 void mono_sgen_add_to_global_remset (gpointer ptr) MONO_INTERNAL;
 663
 664 typedef struct _SgenMajorCollector SgenMajorCollector;
 665 struct _SgenMajorCollector {
 666         size_t section_size;
 667
 668         gboolean (*is_object_live) (char *obj);
 669         void* (*alloc_small_pinned_obj) (size_t size, gboolean has_references);
 670         void* (*alloc_degraded) (MonoVTable *vtable, size_t size);
 671         void (*copy_or_mark_object) (void **obj_slot, SgenGrayQueue *queue);
 672         void (*minor_scan_object) (char *start, SgenGrayQueue *queue);
 673         char* (*minor_scan_vtype) (char *start, mword desc, char* from_start, char* from_end, SgenGrayQueue *queue);
 674         void (*major_scan_object) (char *start, SgenGrayQueue *queue);
 675         void (*copy_object) (void **obj_slot, SgenGrayQueue *queue);
 676         void* (*alloc_object) (int size, gboolean has_references);
 677         void (*free_pinned_object) (char *obj, size_t size);
 678         void (*iterate_objects) (gboolean non_pinned, gboolean pinned, IterateObjectCallbackFunc callback, void *data);
 679         void (*free_non_pinned_object) (char *obj, size_t size);
 680         void (*find_pin_queue_start_ends) (SgenGrayQueue *queue);
 681         void (*pin_objects) (SgenGrayQueue *queue);
 682         void (*init_to_space) (void);
 683         void (*sweep) (void);
 684         void (*check_scan_starts) (void);
 685         void (*dump_heap) (FILE *heap_dump_file);
 686         gint64 (*get_used_size) (void);
 687         void (*start_nursery_collection) (void);
 688         void (*finish_nursery_collection) (void);
 689         void (*finish_major_collection) (void);
 690         gboolean (*ptr_is_in_non_pinned_space) (char *ptr);
 691         gboolean (*obj_is_from_pinned_alloc) (char *obj);
 692         void (*report_pinned_memory_usage) (void);
 693         int (*get_num_major_sections) (void);
 694 };
 695
 696 void mono_sgen_marksweep_init (SgenMajorCollector *collector, int nursery_bits, char *nursery_start, char *nursery_end) MONO_INTERNAL;
 697 void mono_sgen_copying_init (SgenMajorCollector *collector, int the_nursery_bits, char *the_nursery_start, char *the_nursery_end) MONO_INTERNAL;
 698
 699 /*
 700  * This function can be called on an object whose first word, the
 701  * vtable field, is not intact.  This is necessary for the parallel
 702  * collector.
 703  */
 704 static inline guint
 705 mono_sgen_par_object_get_size (MonoVTable *vtable, MonoObject* o)
 706 {
 707         MonoClass *klass = vtable->klass;
 708         /*
 709          * We depend on mono_string_length_fast and
 710          * mono_array_length_fast not using the object's vtable.
 711          */
 712         if (klass == mono_defaults.string_class) {
 713                 return sizeof (MonoString) + 2 * mono_string_length_fast ((MonoString*) o) + 2;
 714         } else if (klass->rank) {
 715                 MonoArray *array = (MonoArray*)o;
 716                 size_t size = sizeof (MonoArray) + klass->sizes.element_size * mono_array_length_fast (array);
 717                 if (G_UNLIKELY (array->bounds)) {
 718                         size += sizeof (mono_array_size_t) - 1;
 719                         size &= ~(sizeof (mono_array_size_t) - 1);
 720                         size += sizeof (MonoArrayBounds) * klass->rank;
 721                 }
 722                 return size;
 723         } else {
 724                 /* from a created object: the class must be inited already */
 725                 return klass->instance_size;
 726         }
 727 }
 728
 729 #define mono_sgen_safe_object_get_size(o)               mono_sgen_par_object_get_size ((MonoVTable*)SGEN_LOAD_VTABLE ((o)), (o))
 730
 731 #endif /* __MONO_SGENGC_H__ */