github.com/comwrg/go/src@v0.0.0-20220319063731-c238d0440370/runtime/mbitmap.go (about) 1 // Copyright 2009 The Go Authors. All rights reserved. 2 // Use of this source code is governed by a BSD-style 3 // license that can be found in the LICENSE file. 4 5 // Garbage collector: type and heap bitmaps. 6 // 7 // Stack, data, and bss bitmaps 8 // 9 // Stack frames and global variables in the data and bss sections are 10 // described by bitmaps with 1 bit per pointer-sized word. A "1" bit 11 // means the word is a live pointer to be visited by the GC (referred to 12 // as "pointer"). A "0" bit means the word should be ignored by GC 13 // (referred to as "scalar", though it could be a dead pointer value). 14 // 15 // Heap bitmap 16 // 17 // The heap bitmap comprises 2 bits for each pointer-sized word in the heap, 18 // stored in the heapArena metadata backing each heap arena. 19 // That is, if ha is the heapArena for the arena starting a start, 20 // then ha.bitmap[0] holds the 2-bit entries for the four words start 21 // through start+3*ptrSize, ha.bitmap[1] holds the entries for 22 // start+4*ptrSize through start+7*ptrSize, and so on. 23 // 24 // In each 2-bit entry, the lower bit is a pointer/scalar bit, just 25 // like in the stack/data bitmaps described above. The upper bit 26 // indicates scan/dead: a "1" value ("scan") indicates that there may 27 // be pointers in later words of the allocation, and a "0" value 28 // ("dead") indicates there are no more pointers in the allocation. If 29 // the upper bit is 0, the lower bit must also be 0, and this 30 // indicates scanning can ignore the rest of the allocation. 31 // 32 // The 2-bit entries are split when written into the byte, so that the top half 33 // of the byte contains 4 high (scan) bits and the bottom half contains 4 low 34 // (pointer) bits. This form allows a copy from the 1-bit to the 4-bit form to 35 // keep the pointer bits contiguous, instead of having to space them out. 36 // 37 // The code makes use of the fact that the zero value for a heap 38 // bitmap means scalar/dead. This property must be preserved when 39 // modifying the encoding. 40 // 41 // The bitmap for noscan spans is not maintained. Code must ensure 42 // that an object is scannable before consulting its bitmap by 43 // checking either the noscan bit in the span or by consulting its 44 // type's information. 45 46 package runtime 47 48 import ( 49 "runtime/internal/atomic" 50 "runtime/internal/sys" 51 "unsafe" 52 ) 53 54 const ( 55 bitPointer = 1 << 0 56 bitScan = 1 << 4 57 58 heapBitsShift = 1 // shift offset between successive bitPointer or bitScan entries 59 wordsPerBitmapByte = 8 / 2 // heap words described by one bitmap byte 60 61 // all scan/pointer bits in a byte 62 bitScanAll = bitScan | bitScan<<heapBitsShift | bitScan<<(2*heapBitsShift) | bitScan<<(3*heapBitsShift) 63 bitPointerAll = bitPointer | bitPointer<<heapBitsShift | bitPointer<<(2*heapBitsShift) | bitPointer<<(3*heapBitsShift) 64 ) 65 66 // addb returns the byte pointer p+n. 67 //go:nowritebarrier 68 //go:nosplit 69 func addb(p *byte, n uintptr) *byte { 70 // Note: wrote out full expression instead of calling add(p, n) 71 // to reduce the number of temporaries generated by the 72 // compiler for this trivial expression during inlining. 73 return (*byte)(unsafe.Pointer(uintptr(unsafe.Pointer(p)) + n)) 74 } 75 76 // subtractb returns the byte pointer p-n. 77 //go:nowritebarrier 78 //go:nosplit 79 func subtractb(p *byte, n uintptr) *byte { 80 // Note: wrote out full expression instead of calling add(p, -n) 81 // to reduce the number of temporaries generated by the 82 // compiler for this trivial expression during inlining. 83 return (*byte)(unsafe.Pointer(uintptr(unsafe.Pointer(p)) - n)) 84 } 85 86 // add1 returns the byte pointer p+1. 87 //go:nowritebarrier 88 //go:nosplit 89 func add1(p *byte) *byte { 90 // Note: wrote out full expression instead of calling addb(p, 1) 91 // to reduce the number of temporaries generated by the 92 // compiler for this trivial expression during inlining. 93 return (*byte)(unsafe.Pointer(uintptr(unsafe.Pointer(p)) + 1)) 94 } 95 96 // subtract1 returns the byte pointer p-1. 97 //go:nowritebarrier 98 // 99 // nosplit because it is used during write barriers and must not be preempted. 100 //go:nosplit 101 func subtract1(p *byte) *byte { 102 // Note: wrote out full expression instead of calling subtractb(p, 1) 103 // to reduce the number of temporaries generated by the 104 // compiler for this trivial expression during inlining. 105 return (*byte)(unsafe.Pointer(uintptr(unsafe.Pointer(p)) - 1)) 106 } 107 108 // heapBits provides access to the bitmap bits for a single heap word. 109 // The methods on heapBits take value receivers so that the compiler 110 // can more easily inline calls to those methods and registerize the 111 // struct fields independently. 112 type heapBits struct { 113 bitp *uint8 114 shift uint32 115 arena uint32 // Index of heap arena containing bitp 116 last *uint8 // Last byte arena's bitmap 117 } 118 119 // Make the compiler check that heapBits.arena is large enough to hold 120 // the maximum arena frame number. 121 var _ = heapBits{arena: (1<<heapAddrBits)/heapArenaBytes - 1} 122 123 // markBits provides access to the mark bit for an object in the heap. 124 // bytep points to the byte holding the mark bit. 125 // mask is a byte with a single bit set that can be &ed with *bytep 126 // to see if the bit has been set. 127 // *m.byte&m.mask != 0 indicates the mark bit is set. 128 // index can be used along with span information to generate 129 // the address of the object in the heap. 130 // We maintain one set of mark bits for allocation and one for 131 // marking purposes. 132 type markBits struct { 133 bytep *uint8 134 mask uint8 135 index uintptr 136 } 137 138 //go:nosplit 139 func (s *mspan) allocBitsForIndex(allocBitIndex uintptr) markBits { 140 bytep, mask := s.allocBits.bitp(allocBitIndex) 141 return markBits{bytep, mask, allocBitIndex} 142 } 143 144 // refillAllocCache takes 8 bytes s.allocBits starting at whichByte 145 // and negates them so that ctz (count trailing zeros) instructions 146 // can be used. It then places these 8 bytes into the cached 64 bit 147 // s.allocCache. 148 func (s *mspan) refillAllocCache(whichByte uintptr) { 149 bytes := (*[8]uint8)(unsafe.Pointer(s.allocBits.bytep(whichByte))) 150 aCache := uint64(0) 151 aCache |= uint64(bytes[0]) 152 aCache |= uint64(bytes[1]) << (1 * 8) 153 aCache |= uint64(bytes[2]) << (2 * 8) 154 aCache |= uint64(bytes[3]) << (3 * 8) 155 aCache |= uint64(bytes[4]) << (4 * 8) 156 aCache |= uint64(bytes[5]) << (5 * 8) 157 aCache |= uint64(bytes[6]) << (6 * 8) 158 aCache |= uint64(bytes[7]) << (7 * 8) 159 s.allocCache = ^aCache 160 } 161 162 // nextFreeIndex returns the index of the next free object in s at 163 // or after s.freeindex. 164 // There are hardware instructions that can be used to make this 165 // faster if profiling warrants it. 166 func (s *mspan) nextFreeIndex() uintptr { 167 sfreeindex := s.freeindex 168 snelems := s.nelems 169 if sfreeindex == snelems { 170 return sfreeindex 171 } 172 if sfreeindex > snelems { 173 throw("s.freeindex > s.nelems") 174 } 175 176 aCache := s.allocCache 177 178 bitIndex := sys.Ctz64(aCache) 179 for bitIndex == 64 { 180 // Move index to start of next cached bits. 181 sfreeindex = (sfreeindex + 64) &^ (64 - 1) 182 if sfreeindex >= snelems { 183 s.freeindex = snelems 184 return snelems 185 } 186 whichByte := sfreeindex / 8 187 // Refill s.allocCache with the next 64 alloc bits. 188 s.refillAllocCache(whichByte) 189 aCache = s.allocCache 190 bitIndex = sys.Ctz64(aCache) 191 // nothing available in cached bits 192 // grab the next 8 bytes and try again. 193 } 194 result := sfreeindex + uintptr(bitIndex) 195 if result >= snelems { 196 s.freeindex = snelems 197 return snelems 198 } 199 200 s.allocCache >>= uint(bitIndex + 1) 201 sfreeindex = result + 1 202 203 if sfreeindex%64 == 0 && sfreeindex != snelems { 204 // We just incremented s.freeindex so it isn't 0. 205 // As each 1 in s.allocCache was encountered and used for allocation 206 // it was shifted away. At this point s.allocCache contains all 0s. 207 // Refill s.allocCache so that it corresponds 208 // to the bits at s.allocBits starting at s.freeindex. 209 whichByte := sfreeindex / 8 210 s.refillAllocCache(whichByte) 211 } 212 s.freeindex = sfreeindex 213 return result 214 } 215 216 // isFree reports whether the index'th object in s is unallocated. 217 // 218 // The caller must ensure s.state is mSpanInUse, and there must have 219 // been no preemption points since ensuring this (which could allow a 220 // GC transition, which would allow the state to change). 221 func (s *mspan) isFree(index uintptr) bool { 222 if index < s.freeindex { 223 return false 224 } 225 bytep, mask := s.allocBits.bitp(index) 226 return *bytep&mask == 0 227 } 228 229 // divideByElemSize returns n/s.elemsize. 230 // n must be within [0, s.npages*_PageSize), 231 // or may be exactly s.npages*_PageSize 232 // if s.elemsize is from sizeclasses.go. 233 func (s *mspan) divideByElemSize(n uintptr) uintptr { 234 const doubleCheck = false 235 236 // See explanation in mksizeclasses.go's computeDivMagic. 237 q := uintptr((uint64(n) * uint64(s.divMul)) >> 32) 238 239 if doubleCheck && q != n/s.elemsize { 240 println(n, "/", s.elemsize, "should be", n/s.elemsize, "but got", q) 241 throw("bad magic division") 242 } 243 return q 244 } 245 246 func (s *mspan) objIndex(p uintptr) uintptr { 247 return s.divideByElemSize(p - s.base()) 248 } 249 250 func markBitsForAddr(p uintptr) markBits { 251 s := spanOf(p) 252 objIndex := s.objIndex(p) 253 return s.markBitsForIndex(objIndex) 254 } 255 256 func (s *mspan) markBitsForIndex(objIndex uintptr) markBits { 257 bytep, mask := s.gcmarkBits.bitp(objIndex) 258 return markBits{bytep, mask, objIndex} 259 } 260 261 func (s *mspan) markBitsForBase() markBits { 262 return markBits{(*uint8)(s.gcmarkBits), uint8(1), 0} 263 } 264 265 // isMarked reports whether mark bit m is set. 266 func (m markBits) isMarked() bool { 267 return *m.bytep&m.mask != 0 268 } 269 270 // setMarked sets the marked bit in the markbits, atomically. 271 func (m markBits) setMarked() { 272 // Might be racing with other updates, so use atomic update always. 273 // We used to be clever here and use a non-atomic update in certain 274 // cases, but it's not worth the risk. 275 atomic.Or8(m.bytep, m.mask) 276 } 277 278 // setMarkedNonAtomic sets the marked bit in the markbits, non-atomically. 279 func (m markBits) setMarkedNonAtomic() { 280 *m.bytep |= m.mask 281 } 282 283 // clearMarked clears the marked bit in the markbits, atomically. 284 func (m markBits) clearMarked() { 285 // Might be racing with other updates, so use atomic update always. 286 // We used to be clever here and use a non-atomic update in certain 287 // cases, but it's not worth the risk. 288 atomic.And8(m.bytep, ^m.mask) 289 } 290 291 // markBitsForSpan returns the markBits for the span base address base. 292 func markBitsForSpan(base uintptr) (mbits markBits) { 293 mbits = markBitsForAddr(base) 294 if mbits.mask != 1 { 295 throw("markBitsForSpan: unaligned start") 296 } 297 return mbits 298 } 299 300 // advance advances the markBits to the next object in the span. 301 func (m *markBits) advance() { 302 if m.mask == 1<<7 { 303 m.bytep = (*uint8)(unsafe.Pointer(uintptr(unsafe.Pointer(m.bytep)) + 1)) 304 m.mask = 1 305 } else { 306 m.mask = m.mask << 1 307 } 308 m.index++ 309 } 310 311 // heapBitsForAddr returns the heapBits for the address addr. 312 // The caller must ensure addr is in an allocated span. 313 // In particular, be careful not to point past the end of an object. 314 // 315 // nosplit because it is used during write barriers and must not be preempted. 316 //go:nosplit 317 func heapBitsForAddr(addr uintptr) (h heapBits) { 318 // 2 bits per word, 4 pairs per byte, and a mask is hard coded. 319 arena := arenaIndex(addr) 320 ha := mheap_.arenas[arena.l1()][arena.l2()] 321 // The compiler uses a load for nil checking ha, but in this 322 // case we'll almost never hit that cache line again, so it 323 // makes more sense to do a value check. 324 if ha == nil { 325 // addr is not in the heap. Return nil heapBits, which 326 // we expect to crash in the caller. 327 return 328 } 329 h.bitp = &ha.bitmap[(addr/(sys.PtrSize*4))%heapArenaBitmapBytes] 330 h.shift = uint32((addr / sys.PtrSize) & 3) 331 h.arena = uint32(arena) 332 h.last = &ha.bitmap[len(ha.bitmap)-1] 333 return 334 } 335 336 // clobberdeadPtr is a special value that is used by the compiler to 337 // clobber dead stack slots, when -clobberdead flag is set. 338 const clobberdeadPtr = uintptr(0xdeaddead | 0xdeaddead<<((^uintptr(0)>>63)*32)) 339 340 // badPointer throws bad pointer in heap panic. 341 func badPointer(s *mspan, p, refBase, refOff uintptr) { 342 // Typically this indicates an incorrect use 343 // of unsafe or cgo to store a bad pointer in 344 // the Go heap. It may also indicate a runtime 345 // bug. 346 // 347 // TODO(austin): We could be more aggressive 348 // and detect pointers to unallocated objects 349 // in allocated spans. 350 printlock() 351 print("runtime: pointer ", hex(p)) 352 if s != nil { 353 state := s.state.get() 354 if state != mSpanInUse { 355 print(" to unallocated span") 356 } else { 357 print(" to unused region of span") 358 } 359 print(" span.base()=", hex(s.base()), " span.limit=", hex(s.limit), " span.state=", state) 360 } 361 print("\n") 362 if refBase != 0 { 363 print("runtime: found in object at *(", hex(refBase), "+", hex(refOff), ")\n") 364 gcDumpObject("object", refBase, refOff) 365 } 366 getg().m.traceback = 2 367 throw("found bad pointer in Go heap (incorrect use of unsafe or cgo?)") 368 } 369 370 // findObject returns the base address for the heap object containing 371 // the address p, the object's span, and the index of the object in s. 372 // If p does not point into a heap object, it returns base == 0. 373 // 374 // If p points is an invalid heap pointer and debug.invalidptr != 0, 375 // findObject panics. 376 // 377 // refBase and refOff optionally give the base address of the object 378 // in which the pointer p was found and the byte offset at which it 379 // was found. These are used for error reporting. 380 // 381 // It is nosplit so it is safe for p to be a pointer to the current goroutine's stack. 382 // Since p is a uintptr, it would not be adjusted if the stack were to move. 383 //go:nosplit 384 func findObject(p, refBase, refOff uintptr) (base uintptr, s *mspan, objIndex uintptr) { 385 s = spanOf(p) 386 // If s is nil, the virtual address has never been part of the heap. 387 // This pointer may be to some mmap'd region, so we allow it. 388 if s == nil { 389 if GOARCH == "amd64" && p == clobberdeadPtr && debug.invalidptr != 0 { 390 // Crash if clobberdeadPtr is seen. Only on AMD64 for now, as 391 // it is the only platform where compiler's clobberdead mode is 392 // implemented. On AMD64 clobberdeadPtr cannot be a valid address. 393 badPointer(s, p, refBase, refOff) 394 } 395 return 396 } 397 // If p is a bad pointer, it may not be in s's bounds. 398 // 399 // Check s.state to synchronize with span initialization 400 // before checking other fields. See also spanOfHeap. 401 if state := s.state.get(); state != mSpanInUse || p < s.base() || p >= s.limit { 402 // Pointers into stacks are also ok, the runtime manages these explicitly. 403 if state == mSpanManual { 404 return 405 } 406 // The following ensures that we are rigorous about what data 407 // structures hold valid pointers. 408 if debug.invalidptr != 0 { 409 badPointer(s, p, refBase, refOff) 410 } 411 return 412 } 413 414 objIndex = s.objIndex(p) 415 base = s.base() + objIndex*s.elemsize 416 return 417 } 418 419 // next returns the heapBits describing the next pointer-sized word in memory. 420 // That is, if h describes address p, h.next() describes p+ptrSize. 421 // Note that next does not modify h. The caller must record the result. 422 // 423 // nosplit because it is used during write barriers and must not be preempted. 424 //go:nosplit 425 func (h heapBits) next() heapBits { 426 if h.shift < 3*heapBitsShift { 427 h.shift += heapBitsShift 428 } else if h.bitp != h.last { 429 h.bitp, h.shift = add1(h.bitp), 0 430 } else { 431 // Move to the next arena. 432 return h.nextArena() 433 } 434 return h 435 } 436 437 // nextArena advances h to the beginning of the next heap arena. 438 // 439 // This is a slow-path helper to next. gc's inliner knows that 440 // heapBits.next can be inlined even though it calls this. This is 441 // marked noinline so it doesn't get inlined into next and cause next 442 // to be too big to inline. 443 // 444 //go:nosplit 445 //go:noinline 446 func (h heapBits) nextArena() heapBits { 447 h.arena++ 448 ai := arenaIdx(h.arena) 449 l2 := mheap_.arenas[ai.l1()] 450 if l2 == nil { 451 // We just passed the end of the object, which 452 // was also the end of the heap. Poison h. It 453 // should never be dereferenced at this point. 454 return heapBits{} 455 } 456 ha := l2[ai.l2()] 457 if ha == nil { 458 return heapBits{} 459 } 460 h.bitp, h.shift = &ha.bitmap[0], 0 461 h.last = &ha.bitmap[len(ha.bitmap)-1] 462 return h 463 } 464 465 // forward returns the heapBits describing n pointer-sized words ahead of h in memory. 466 // That is, if h describes address p, h.forward(n) describes p+n*ptrSize. 467 // h.forward(1) is equivalent to h.next(), just slower. 468 // Note that forward does not modify h. The caller must record the result. 469 // bits returns the heap bits for the current word. 470 //go:nosplit 471 func (h heapBits) forward(n uintptr) heapBits { 472 n += uintptr(h.shift) / heapBitsShift 473 nbitp := uintptr(unsafe.Pointer(h.bitp)) + n/4 474 h.shift = uint32(n%4) * heapBitsShift 475 if nbitp <= uintptr(unsafe.Pointer(h.last)) { 476 h.bitp = (*uint8)(unsafe.Pointer(nbitp)) 477 return h 478 } 479 480 // We're in a new heap arena. 481 past := nbitp - (uintptr(unsafe.Pointer(h.last)) + 1) 482 h.arena += 1 + uint32(past/heapArenaBitmapBytes) 483 ai := arenaIdx(h.arena) 484 if l2 := mheap_.arenas[ai.l1()]; l2 != nil && l2[ai.l2()] != nil { 485 a := l2[ai.l2()] 486 h.bitp = &a.bitmap[past%heapArenaBitmapBytes] 487 h.last = &a.bitmap[len(a.bitmap)-1] 488 } else { 489 h.bitp, h.last = nil, nil 490 } 491 return h 492 } 493 494 // forwardOrBoundary is like forward, but stops at boundaries between 495 // contiguous sections of the bitmap. It returns the number of words 496 // advanced over, which will be <= n. 497 func (h heapBits) forwardOrBoundary(n uintptr) (heapBits, uintptr) { 498 maxn := 4 * ((uintptr(unsafe.Pointer(h.last)) + 1) - uintptr(unsafe.Pointer(h.bitp))) 499 if n > maxn { 500 n = maxn 501 } 502 return h.forward(n), n 503 } 504 505 // The caller can test morePointers and isPointer by &-ing with bitScan and bitPointer. 506 // The result includes in its higher bits the bits for subsequent words 507 // described by the same bitmap byte. 508 // 509 // nosplit because it is used during write barriers and must not be preempted. 510 //go:nosplit 511 func (h heapBits) bits() uint32 { 512 // The (shift & 31) eliminates a test and conditional branch 513 // from the generated code. 514 return uint32(*h.bitp) >> (h.shift & 31) 515 } 516 517 // morePointers reports whether this word and all remaining words in this object 518 // are scalars. 519 // h must not describe the second word of the object. 520 func (h heapBits) morePointers() bool { 521 return h.bits()&bitScan != 0 522 } 523 524 // isPointer reports whether the heap bits describe a pointer word. 525 // 526 // nosplit because it is used during write barriers and must not be preempted. 527 //go:nosplit 528 func (h heapBits) isPointer() bool { 529 return h.bits()&bitPointer != 0 530 } 531 532 // bulkBarrierPreWrite executes a write barrier 533 // for every pointer slot in the memory range [src, src+size), 534 // using pointer/scalar information from [dst, dst+size). 535 // This executes the write barriers necessary before a memmove. 536 // src, dst, and size must be pointer-aligned. 537 // The range [dst, dst+size) must lie within a single object. 538 // It does not perform the actual writes. 539 // 540 // As a special case, src == 0 indicates that this is being used for a 541 // memclr. bulkBarrierPreWrite will pass 0 for the src of each write 542 // barrier. 543 // 544 // Callers should call bulkBarrierPreWrite immediately before 545 // calling memmove(dst, src, size). This function is marked nosplit 546 // to avoid being preempted; the GC must not stop the goroutine 547 // between the memmove and the execution of the barriers. 548 // The caller is also responsible for cgo pointer checks if this 549 // may be writing Go pointers into non-Go memory. 550 // 551 // The pointer bitmap is not maintained for allocations containing 552 // no pointers at all; any caller of bulkBarrierPreWrite must first 553 // make sure the underlying allocation contains pointers, usually 554 // by checking typ.ptrdata. 555 // 556 // Callers must perform cgo checks if writeBarrier.cgo. 557 // 558 //go:nosplit 559 func bulkBarrierPreWrite(dst, src, size uintptr) { 560 if (dst|src|size)&(sys.PtrSize-1) != 0 { 561 throw("bulkBarrierPreWrite: unaligned arguments") 562 } 563 if !writeBarrier.needed { 564 return 565 } 566 if s := spanOf(dst); s == nil { 567 // If dst is a global, use the data or BSS bitmaps to 568 // execute write barriers. 569 for _, datap := range activeModules() { 570 if datap.data <= dst && dst < datap.edata { 571 bulkBarrierBitmap(dst, src, size, dst-datap.data, datap.gcdatamask.bytedata) 572 return 573 } 574 } 575 for _, datap := range activeModules() { 576 if datap.bss <= dst && dst < datap.ebss { 577 bulkBarrierBitmap(dst, src, size, dst-datap.bss, datap.gcbssmask.bytedata) 578 return 579 } 580 } 581 return 582 } else if s.state.get() != mSpanInUse || dst < s.base() || s.limit <= dst { 583 // dst was heap memory at some point, but isn't now. 584 // It can't be a global. It must be either our stack, 585 // or in the case of direct channel sends, it could be 586 // another stack. Either way, no need for barriers. 587 // This will also catch if dst is in a freed span, 588 // though that should never have. 589 return 590 } 591 592 buf := &getg().m.p.ptr().wbBuf 593 h := heapBitsForAddr(dst) 594 if src == 0 { 595 for i := uintptr(0); i < size; i += sys.PtrSize { 596 if h.isPointer() { 597 dstx := (*uintptr)(unsafe.Pointer(dst + i)) 598 if !buf.putFast(*dstx, 0) { 599 wbBufFlush(nil, 0) 600 } 601 } 602 h = h.next() 603 } 604 } else { 605 for i := uintptr(0); i < size; i += sys.PtrSize { 606 if h.isPointer() { 607 dstx := (*uintptr)(unsafe.Pointer(dst + i)) 608 srcx := (*uintptr)(unsafe.Pointer(src + i)) 609 if !buf.putFast(*dstx, *srcx) { 610 wbBufFlush(nil, 0) 611 } 612 } 613 h = h.next() 614 } 615 } 616 } 617 618 // bulkBarrierPreWriteSrcOnly is like bulkBarrierPreWrite but 619 // does not execute write barriers for [dst, dst+size). 620 // 621 // In addition to the requirements of bulkBarrierPreWrite 622 // callers need to ensure [dst, dst+size) is zeroed. 623 // 624 // This is used for special cases where e.g. dst was just 625 // created and zeroed with malloc. 626 //go:nosplit 627 func bulkBarrierPreWriteSrcOnly(dst, src, size uintptr) { 628 if (dst|src|size)&(sys.PtrSize-1) != 0 { 629 throw("bulkBarrierPreWrite: unaligned arguments") 630 } 631 if !writeBarrier.needed { 632 return 633 } 634 buf := &getg().m.p.ptr().wbBuf 635 h := heapBitsForAddr(dst) 636 for i := uintptr(0); i < size; i += sys.PtrSize { 637 if h.isPointer() { 638 srcx := (*uintptr)(unsafe.Pointer(src + i)) 639 if !buf.putFast(0, *srcx) { 640 wbBufFlush(nil, 0) 641 } 642 } 643 h = h.next() 644 } 645 } 646 647 // bulkBarrierBitmap executes write barriers for copying from [src, 648 // src+size) to [dst, dst+size) using a 1-bit pointer bitmap. src is 649 // assumed to start maskOffset bytes into the data covered by the 650 // bitmap in bits (which may not be a multiple of 8). 651 // 652 // This is used by bulkBarrierPreWrite for writes to data and BSS. 653 // 654 //go:nosplit 655 func bulkBarrierBitmap(dst, src, size, maskOffset uintptr, bits *uint8) { 656 word := maskOffset / sys.PtrSize 657 bits = addb(bits, word/8) 658 mask := uint8(1) << (word % 8) 659 660 buf := &getg().m.p.ptr().wbBuf 661 for i := uintptr(0); i < size; i += sys.PtrSize { 662 if mask == 0 { 663 bits = addb(bits, 1) 664 if *bits == 0 { 665 // Skip 8 words. 666 i += 7 * sys.PtrSize 667 continue 668 } 669 mask = 1 670 } 671 if *bits&mask != 0 { 672 dstx := (*uintptr)(unsafe.Pointer(dst + i)) 673 if src == 0 { 674 if !buf.putFast(*dstx, 0) { 675 wbBufFlush(nil, 0) 676 } 677 } else { 678 srcx := (*uintptr)(unsafe.Pointer(src + i)) 679 if !buf.putFast(*dstx, *srcx) { 680 wbBufFlush(nil, 0) 681 } 682 } 683 } 684 mask <<= 1 685 } 686 } 687 688 // typeBitsBulkBarrier executes a write barrier for every 689 // pointer that would be copied from [src, src+size) to [dst, 690 // dst+size) by a memmove using the type bitmap to locate those 691 // pointer slots. 692 // 693 // The type typ must correspond exactly to [src, src+size) and [dst, dst+size). 694 // dst, src, and size must be pointer-aligned. 695 // The type typ must have a plain bitmap, not a GC program. 696 // The only use of this function is in channel sends, and the 697 // 64 kB channel element limit takes care of this for us. 698 // 699 // Must not be preempted because it typically runs right before memmove, 700 // and the GC must observe them as an atomic action. 701 // 702 // Callers must perform cgo checks if writeBarrier.cgo. 703 // 704 //go:nosplit 705 func typeBitsBulkBarrier(typ *_type, dst, src, size uintptr) { 706 if typ == nil { 707 throw("runtime: typeBitsBulkBarrier without type") 708 } 709 if typ.size != size { 710 println("runtime: typeBitsBulkBarrier with type ", typ.string(), " of size ", typ.size, " but memory size", size) 711 throw("runtime: invalid typeBitsBulkBarrier") 712 } 713 if typ.kind&kindGCProg != 0 { 714 println("runtime: typeBitsBulkBarrier with type ", typ.string(), " with GC prog") 715 throw("runtime: invalid typeBitsBulkBarrier") 716 } 717 if !writeBarrier.needed { 718 return 719 } 720 ptrmask := typ.gcdata 721 buf := &getg().m.p.ptr().wbBuf 722 var bits uint32 723 for i := uintptr(0); i < typ.ptrdata; i += sys.PtrSize { 724 if i&(sys.PtrSize*8-1) == 0 { 725 bits = uint32(*ptrmask) 726 ptrmask = addb(ptrmask, 1) 727 } else { 728 bits = bits >> 1 729 } 730 if bits&1 != 0 { 731 dstx := (*uintptr)(unsafe.Pointer(dst + i)) 732 srcx := (*uintptr)(unsafe.Pointer(src + i)) 733 if !buf.putFast(*dstx, *srcx) { 734 wbBufFlush(nil, 0) 735 } 736 } 737 } 738 } 739 740 // The methods operating on spans all require that h has been returned 741 // by heapBitsForSpan and that size, n, total are the span layout description 742 // returned by the mspan's layout method. 743 // If total > size*n, it means that there is extra leftover memory in the span, 744 // usually due to rounding. 745 // 746 // TODO(rsc): Perhaps introduce a different heapBitsSpan type. 747 748 // initSpan initializes the heap bitmap for a span. 749 // If this is a span of pointer-sized objects, it initializes all 750 // words to pointer/scan. 751 // Otherwise, it initializes all words to scalar/dead. 752 func (h heapBits) initSpan(s *mspan) { 753 // Clear bits corresponding to objects. 754 nw := (s.npages << _PageShift) / sys.PtrSize 755 if nw%wordsPerBitmapByte != 0 { 756 throw("initSpan: unaligned length") 757 } 758 if h.shift != 0 { 759 throw("initSpan: unaligned base") 760 } 761 isPtrs := sys.PtrSize == 8 && s.elemsize == sys.PtrSize 762 for nw > 0 { 763 hNext, anw := h.forwardOrBoundary(nw) 764 nbyte := anw / wordsPerBitmapByte 765 if isPtrs { 766 bitp := h.bitp 767 for i := uintptr(0); i < nbyte; i++ { 768 *bitp = bitPointerAll | bitScanAll 769 bitp = add1(bitp) 770 } 771 } else { 772 memclrNoHeapPointers(unsafe.Pointer(h.bitp), nbyte) 773 } 774 h = hNext 775 nw -= anw 776 } 777 } 778 779 // countAlloc returns the number of objects allocated in span s by 780 // scanning the allocation bitmap. 781 func (s *mspan) countAlloc() int { 782 count := 0 783 bytes := divRoundUp(s.nelems, 8) 784 // Iterate over each 8-byte chunk and count allocations 785 // with an intrinsic. Note that newMarkBits guarantees that 786 // gcmarkBits will be 8-byte aligned, so we don't have to 787 // worry about edge cases, irrelevant bits will simply be zero. 788 for i := uintptr(0); i < bytes; i += 8 { 789 // Extract 64 bits from the byte pointer and get a OnesCount. 790 // Note that the unsafe cast here doesn't preserve endianness, 791 // but that's OK. We only care about how many bits are 1, not 792 // about the order we discover them in. 793 mrkBits := *(*uint64)(unsafe.Pointer(s.gcmarkBits.bytep(i))) 794 count += sys.OnesCount64(mrkBits) 795 } 796 return count 797 } 798 799 // heapBitsSetType records that the new allocation [x, x+size) 800 // holds in [x, x+dataSize) one or more values of type typ. 801 // (The number of values is given by dataSize / typ.size.) 802 // If dataSize < size, the fragment [x+dataSize, x+size) is 803 // recorded as non-pointer data. 804 // It is known that the type has pointers somewhere; 805 // malloc does not call heapBitsSetType when there are no pointers, 806 // because all free objects are marked as noscan during 807 // heapBitsSweepSpan. 808 // 809 // There can only be one allocation from a given span active at a time, 810 // and the bitmap for a span always falls on byte boundaries, 811 // so there are no write-write races for access to the heap bitmap. 812 // Hence, heapBitsSetType can access the bitmap without atomics. 813 // 814 // There can be read-write races between heapBitsSetType and things 815 // that read the heap bitmap like scanobject. However, since 816 // heapBitsSetType is only used for objects that have not yet been 817 // made reachable, readers will ignore bits being modified by this 818 // function. This does mean this function cannot transiently modify 819 // bits that belong to neighboring objects. Also, on weakly-ordered 820 // machines, callers must execute a store/store (publication) barrier 821 // between calling this function and making the object reachable. 822 func heapBitsSetType(x, size, dataSize uintptr, typ *_type) { 823 const doubleCheck = false // slow but helpful; enable to test modifications to this code 824 825 const ( 826 mask1 = bitPointer | bitScan // 00010001 827 mask2 = bitPointer | bitScan | mask1<<heapBitsShift // 00110011 828 mask3 = bitPointer | bitScan | mask2<<heapBitsShift // 01110111 829 ) 830 831 // dataSize is always size rounded up to the next malloc size class, 832 // except in the case of allocating a defer block, in which case 833 // size is sizeof(_defer{}) (at least 6 words) and dataSize may be 834 // arbitrarily larger. 835 // 836 // The checks for size == sys.PtrSize and size == 2*sys.PtrSize can therefore 837 // assume that dataSize == size without checking it explicitly. 838 839 if sys.PtrSize == 8 && size == sys.PtrSize { 840 // It's one word and it has pointers, it must be a pointer. 841 // Since all allocated one-word objects are pointers 842 // (non-pointers are aggregated into tinySize allocations), 843 // initSpan sets the pointer bits for us. Nothing to do here. 844 if doubleCheck { 845 h := heapBitsForAddr(x) 846 if !h.isPointer() { 847 throw("heapBitsSetType: pointer bit missing") 848 } 849 if !h.morePointers() { 850 throw("heapBitsSetType: scan bit missing") 851 } 852 } 853 return 854 } 855 856 h := heapBitsForAddr(x) 857 ptrmask := typ.gcdata // start of 1-bit pointer mask (or GC program, handled below) 858 859 // 2-word objects only have 4 bitmap bits and 3-word objects only have 6 bitmap bits. 860 // Therefore, these objects share a heap bitmap byte with the objects next to them. 861 // These are called out as a special case primarily so the code below can assume all 862 // objects are at least 4 words long and that their bitmaps start either at the beginning 863 // of a bitmap byte, or half-way in (h.shift of 0 and 2 respectively). 864 865 if size == 2*sys.PtrSize { 866 if typ.size == sys.PtrSize { 867 // We're allocating a block big enough to hold two pointers. 868 // On 64-bit, that means the actual object must be two pointers, 869 // or else we'd have used the one-pointer-sized block. 870 // On 32-bit, however, this is the 8-byte block, the smallest one. 871 // So it could be that we're allocating one pointer and this was 872 // just the smallest block available. Distinguish by checking dataSize. 873 // (In general the number of instances of typ being allocated is 874 // dataSize/typ.size.) 875 if sys.PtrSize == 4 && dataSize == sys.PtrSize { 876 // 1 pointer object. On 32-bit machines clear the bit for the 877 // unused second word. 878 *h.bitp &^= (bitPointer | bitScan | (bitPointer|bitScan)<<heapBitsShift) << h.shift 879 *h.bitp |= (bitPointer | bitScan) << h.shift 880 } else { 881 // 2-element array of pointer. 882 *h.bitp |= (bitPointer | bitScan | (bitPointer|bitScan)<<heapBitsShift) << h.shift 883 } 884 return 885 } 886 // Otherwise typ.size must be 2*sys.PtrSize, 887 // and typ.kind&kindGCProg == 0. 888 if doubleCheck { 889 if typ.size != 2*sys.PtrSize || typ.kind&kindGCProg != 0 { 890 print("runtime: heapBitsSetType size=", size, " but typ.size=", typ.size, " gcprog=", typ.kind&kindGCProg != 0, "\n") 891 throw("heapBitsSetType") 892 } 893 } 894 b := uint32(*ptrmask) 895 hb := b & 3 896 hb |= bitScanAll & ((bitScan << (typ.ptrdata / sys.PtrSize)) - 1) 897 // Clear the bits for this object so we can set the 898 // appropriate ones. 899 *h.bitp &^= (bitPointer | bitScan | ((bitPointer | bitScan) << heapBitsShift)) << h.shift 900 *h.bitp |= uint8(hb << h.shift) 901 return 902 } else if size == 3*sys.PtrSize { 903 b := uint8(*ptrmask) 904 if doubleCheck { 905 if b == 0 { 906 println("runtime: invalid type ", typ.string()) 907 throw("heapBitsSetType: called with non-pointer type") 908 } 909 if sys.PtrSize != 8 { 910 throw("heapBitsSetType: unexpected 3 pointer wide size class on 32 bit") 911 } 912 if typ.kind&kindGCProg != 0 { 913 throw("heapBitsSetType: unexpected GC prog for 3 pointer wide size class") 914 } 915 if typ.size == 2*sys.PtrSize { 916 print("runtime: heapBitsSetType size=", size, " but typ.size=", typ.size, "\n") 917 throw("heapBitsSetType: inconsistent object sizes") 918 } 919 } 920 if typ.size == sys.PtrSize { 921 // The type contains a pointer otherwise heapBitsSetType wouldn't have been called. 922 // Since the type is only 1 pointer wide and contains a pointer, its gcdata must be exactly 1. 923 if doubleCheck && *typ.gcdata != 1 { 924 print("runtime: heapBitsSetType size=", size, " typ.size=", typ.size, "but *typ.gcdata", *typ.gcdata, "\n") 925 throw("heapBitsSetType: unexpected gcdata for 1 pointer wide type size in 3 pointer wide size class") 926 } 927 // 3 element array of pointers. Unrolling ptrmask 3 times into p yields 00000111. 928 b = 7 929 } 930 931 hb := b & 7 932 // Set bitScan bits for all pointers. 933 hb |= hb << wordsPerBitmapByte 934 // First bitScan bit is always set since the type contains pointers. 935 hb |= bitScan 936 // Second bitScan bit needs to also be set if the third bitScan bit is set. 937 hb |= hb & (bitScan << (2 * heapBitsShift)) >> 1 938 939 // For h.shift > 1 heap bits cross a byte boundary and need to be written part 940 // to h.bitp and part to the next h.bitp. 941 switch h.shift { 942 case 0: 943 *h.bitp &^= mask3 << 0 944 *h.bitp |= hb << 0 945 case 1: 946 *h.bitp &^= mask3 << 1 947 *h.bitp |= hb << 1 948 case 2: 949 *h.bitp &^= mask2 << 2 950 *h.bitp |= (hb & mask2) << 2 951 // Two words written to the first byte. 952 // Advance two words to get to the next byte. 953 h = h.next().next() 954 *h.bitp &^= mask1 955 *h.bitp |= (hb >> 2) & mask1 956 case 3: 957 *h.bitp &^= mask1 << 3 958 *h.bitp |= (hb & mask1) << 3 959 // One word written to the first byte. 960 // Advance one word to get to the next byte. 961 h = h.next() 962 *h.bitp &^= mask2 963 *h.bitp |= (hb >> 1) & mask2 964 } 965 return 966 } 967 968 // Copy from 1-bit ptrmask into 2-bit bitmap. 969 // The basic approach is to use a single uintptr as a bit buffer, 970 // alternating between reloading the buffer and writing bitmap bytes. 971 // In general, one load can supply two bitmap byte writes. 972 // This is a lot of lines of code, but it compiles into relatively few 973 // machine instructions. 974 975 outOfPlace := false 976 if arenaIndex(x+size-1) != arenaIdx(h.arena) || (doubleCheck && fastrand()%2 == 0) { 977 // This object spans heap arenas, so the bitmap may be 978 // discontiguous. Unroll it into the object instead 979 // and then copy it out. 980 // 981 // In doubleCheck mode, we randomly do this anyway to 982 // stress test the bitmap copying path. 983 outOfPlace = true 984 h.bitp = (*uint8)(unsafe.Pointer(x)) 985 h.last = nil 986 } 987 988 var ( 989 // Ptrmask input. 990 p *byte // last ptrmask byte read 991 b uintptr // ptrmask bits already loaded 992 nb uintptr // number of bits in b at next read 993 endp *byte // final ptrmask byte to read (then repeat) 994 endnb uintptr // number of valid bits in *endp 995 pbits uintptr // alternate source of bits 996 997 // Heap bitmap output. 998 w uintptr // words processed 999 nw uintptr // number of words to process 1000 hbitp *byte // next heap bitmap byte to write 1001 hb uintptr // bits being prepared for *hbitp 1002 ) 1003 1004 hbitp = h.bitp 1005 1006 // Handle GC program. Delayed until this part of the code 1007 // so that we can use the same double-checking mechanism 1008 // as the 1-bit case. Nothing above could have encountered 1009 // GC programs: the cases were all too small. 1010 if typ.kind&kindGCProg != 0 { 1011 heapBitsSetTypeGCProg(h, typ.ptrdata, typ.size, dataSize, size, addb(typ.gcdata, 4)) 1012 if doubleCheck { 1013 // Double-check the heap bits written by GC program 1014 // by running the GC program to create a 1-bit pointer mask 1015 // and then jumping to the double-check code below. 1016 // This doesn't catch bugs shared between the 1-bit and 4-bit 1017 // GC program execution, but it does catch mistakes specific 1018 // to just one of those and bugs in heapBitsSetTypeGCProg's 1019 // implementation of arrays. 1020 lock(&debugPtrmask.lock) 1021 if debugPtrmask.data == nil { 1022 debugPtrmask.data = (*byte)(persistentalloc(1<<20, 1, &memstats.other_sys)) 1023 } 1024 ptrmask = debugPtrmask.data 1025 runGCProg(addb(typ.gcdata, 4), nil, ptrmask, 1) 1026 } 1027 goto Phase4 1028 } 1029 1030 // Note about sizes: 1031 // 1032 // typ.size is the number of words in the object, 1033 // and typ.ptrdata is the number of words in the prefix 1034 // of the object that contains pointers. That is, the final 1035 // typ.size - typ.ptrdata words contain no pointers. 1036 // This allows optimization of a common pattern where 1037 // an object has a small header followed by a large scalar 1038 // buffer. If we know the pointers are over, we don't have 1039 // to scan the buffer's heap bitmap at all. 1040 // The 1-bit ptrmasks are sized to contain only bits for 1041 // the typ.ptrdata prefix, zero padded out to a full byte 1042 // of bitmap. This code sets nw (below) so that heap bitmap 1043 // bits are only written for the typ.ptrdata prefix; if there is 1044 // more room in the allocated object, the next heap bitmap 1045 // entry is a 00, indicating that there are no more pointers 1046 // to scan. So only the ptrmask for the ptrdata bytes is needed. 1047 // 1048 // Replicated copies are not as nice: if there is an array of 1049 // objects with scalar tails, all but the last tail does have to 1050 // be initialized, because there is no way to say "skip forward". 1051 // However, because of the possibility of a repeated type with 1052 // size not a multiple of 4 pointers (one heap bitmap byte), 1053 // the code already must handle the last ptrmask byte specially 1054 // by treating it as containing only the bits for endnb pointers, 1055 // where endnb <= 4. We represent large scalar tails that must 1056 // be expanded in the replication by setting endnb larger than 4. 1057 // This will have the effect of reading many bits out of b, 1058 // but once the real bits are shifted out, b will supply as many 1059 // zero bits as we try to read, which is exactly what we need. 1060 1061 p = ptrmask 1062 if typ.size < dataSize { 1063 // Filling in bits for an array of typ. 1064 // Set up for repetition of ptrmask during main loop. 1065 // Note that ptrmask describes only a prefix of 1066 const maxBits = sys.PtrSize*8 - 7 1067 if typ.ptrdata/sys.PtrSize <= maxBits { 1068 // Entire ptrmask fits in uintptr with room for a byte fragment. 1069 // Load into pbits and never read from ptrmask again. 1070 // This is especially important when the ptrmask has 1071 // fewer than 8 bits in it; otherwise the reload in the middle 1072 // of the Phase 2 loop would itself need to loop to gather 1073 // at least 8 bits. 1074 1075 // Accumulate ptrmask into b. 1076 // ptrmask is sized to describe only typ.ptrdata, but we record 1077 // it as describing typ.size bytes, since all the high bits are zero. 1078 nb = typ.ptrdata / sys.PtrSize 1079 for i := uintptr(0); i < nb; i += 8 { 1080 b |= uintptr(*p) << i 1081 p = add1(p) 1082 } 1083 nb = typ.size / sys.PtrSize 1084 1085 // Replicate ptrmask to fill entire pbits uintptr. 1086 // Doubling and truncating is fewer steps than 1087 // iterating by nb each time. (nb could be 1.) 1088 // Since we loaded typ.ptrdata/sys.PtrSize bits 1089 // but are pretending to have typ.size/sys.PtrSize, 1090 // there might be no replication necessary/possible. 1091 pbits = b 1092 endnb = nb 1093 if nb+nb <= maxBits { 1094 for endnb <= sys.PtrSize*8 { 1095 pbits |= pbits << endnb 1096 endnb += endnb 1097 } 1098 // Truncate to a multiple of original ptrmask. 1099 // Because nb+nb <= maxBits, nb fits in a byte. 1100 // Byte division is cheaper than uintptr division. 1101 endnb = uintptr(maxBits/byte(nb)) * nb 1102 pbits &= 1<<endnb - 1 1103 b = pbits 1104 nb = endnb 1105 } 1106 1107 // Clear p and endp as sentinel for using pbits. 1108 // Checked during Phase 2 loop. 1109 p = nil 1110 endp = nil 1111 } else { 1112 // Ptrmask is larger. Read it multiple times. 1113 n := (typ.ptrdata/sys.PtrSize+7)/8 - 1 1114 endp = addb(ptrmask, n) 1115 endnb = typ.size/sys.PtrSize - n*8 1116 } 1117 } 1118 if p != nil { 1119 b = uintptr(*p) 1120 p = add1(p) 1121 nb = 8 1122 } 1123 1124 if typ.size == dataSize { 1125 // Single entry: can stop once we reach the non-pointer data. 1126 nw = typ.ptrdata / sys.PtrSize 1127 } else { 1128 // Repeated instances of typ in an array. 1129 // Have to process first N-1 entries in full, but can stop 1130 // once we reach the non-pointer data in the final entry. 1131 nw = ((dataSize/typ.size-1)*typ.size + typ.ptrdata) / sys.PtrSize 1132 } 1133 if nw == 0 { 1134 // No pointers! Caller was supposed to check. 1135 println("runtime: invalid type ", typ.string()) 1136 throw("heapBitsSetType: called with non-pointer type") 1137 return 1138 } 1139 1140 // Phase 1: Special case for leading byte (shift==0) or half-byte (shift==2). 1141 // The leading byte is special because it contains the bits for word 1, 1142 // which does not have the scan bit set. 1143 // The leading half-byte is special because it's a half a byte, 1144 // so we have to be careful with the bits already there. 1145 switch { 1146 default: 1147 throw("heapBitsSetType: unexpected shift") 1148 1149 case h.shift == 0: 1150 // Ptrmask and heap bitmap are aligned. 1151 // 1152 // This is a fast path for small objects. 1153 // 1154 // The first byte we write out covers the first four 1155 // words of the object. The scan/dead bit on the first 1156 // word must be set to scan since there are pointers 1157 // somewhere in the object. 1158 // In all following words, we set the scan/dead 1159 // appropriately to indicate that the object continues 1160 // to the next 2-bit entry in the bitmap. 1161 // 1162 // We set four bits at a time here, but if the object 1163 // is fewer than four words, phase 3 will clear 1164 // unnecessary bits. 1165 hb = b & bitPointerAll 1166 hb |= bitScanAll 1167 if w += 4; w >= nw { 1168 goto Phase3 1169 } 1170 *hbitp = uint8(hb) 1171 hbitp = add1(hbitp) 1172 b >>= 4 1173 nb -= 4 1174 1175 case h.shift == 2: 1176 // Ptrmask and heap bitmap are misaligned. 1177 // 1178 // On 32 bit architectures only the 6-word object that corresponds 1179 // to a 24 bytes size class can start with h.shift of 2 here since 1180 // all other non 16 byte aligned size classes have been handled by 1181 // special code paths at the beginning of heapBitsSetType on 32 bit. 1182 // 1183 // Many size classes are only 16 byte aligned. On 64 bit architectures 1184 // this results in a heap bitmap position starting with a h.shift of 2. 1185 // 1186 // The bits for the first two words are in a byte shared 1187 // with another object, so we must be careful with the bits 1188 // already there. 1189 // 1190 // We took care of 1-word, 2-word, and 3-word objects above, 1191 // so this is at least a 6-word object. 1192 hb = (b & (bitPointer | bitPointer<<heapBitsShift)) << (2 * heapBitsShift) 1193 hb |= bitScan << (2 * heapBitsShift) 1194 if nw > 1 { 1195 hb |= bitScan << (3 * heapBitsShift) 1196 } 1197 b >>= 2 1198 nb -= 2 1199 *hbitp &^= uint8((bitPointer | bitScan | ((bitPointer | bitScan) << heapBitsShift)) << (2 * heapBitsShift)) 1200 *hbitp |= uint8(hb) 1201 hbitp = add1(hbitp) 1202 if w += 2; w >= nw { 1203 // We know that there is more data, because we handled 2-word and 3-word objects above. 1204 // This must be at least a 6-word object. If we're out of pointer words, 1205 // mark no scan in next bitmap byte and finish. 1206 hb = 0 1207 w += 4 1208 goto Phase3 1209 } 1210 } 1211 1212 // Phase 2: Full bytes in bitmap, up to but not including write to last byte (full or partial) in bitmap. 1213 // The loop computes the bits for that last write but does not execute the write; 1214 // it leaves the bits in hb for processing by phase 3. 1215 // To avoid repeated adjustment of nb, we subtract out the 4 bits we're going to 1216 // use in the first half of the loop right now, and then we only adjust nb explicitly 1217 // if the 8 bits used by each iteration isn't balanced by 8 bits loaded mid-loop. 1218 nb -= 4 1219 for { 1220 // Emit bitmap byte. 1221 // b has at least nb+4 bits, with one exception: 1222 // if w+4 >= nw, then b has only nw-w bits, 1223 // but we'll stop at the break and then truncate 1224 // appropriately in Phase 3. 1225 hb = b & bitPointerAll 1226 hb |= bitScanAll 1227 if w += 4; w >= nw { 1228 break 1229 } 1230 *hbitp = uint8(hb) 1231 hbitp = add1(hbitp) 1232 b >>= 4 1233 1234 // Load more bits. b has nb right now. 1235 if p != endp { 1236 // Fast path: keep reading from ptrmask. 1237 // nb unmodified: we just loaded 8 bits, 1238 // and the next iteration will consume 8 bits, 1239 // leaving us with the same nb the next time we're here. 1240 if nb < 8 { 1241 b |= uintptr(*p) << nb 1242 p = add1(p) 1243 } else { 1244 // Reduce the number of bits in b. 1245 // This is important if we skipped 1246 // over a scalar tail, since nb could 1247 // be larger than the bit width of b. 1248 nb -= 8 1249 } 1250 } else if p == nil { 1251 // Almost as fast path: track bit count and refill from pbits. 1252 // For short repetitions. 1253 if nb < 8 { 1254 b |= pbits << nb 1255 nb += endnb 1256 } 1257 nb -= 8 // for next iteration 1258 } else { 1259 // Slow path: reached end of ptrmask. 1260 // Process final partial byte and rewind to start. 1261 b |= uintptr(*p) << nb 1262 nb += endnb 1263 if nb < 8 { 1264 b |= uintptr(*ptrmask) << nb 1265 p = add1(ptrmask) 1266 } else { 1267 nb -= 8 1268 p = ptrmask 1269 } 1270 } 1271 1272 // Emit bitmap byte. 1273 hb = b & bitPointerAll 1274 hb |= bitScanAll 1275 if w += 4; w >= nw { 1276 break 1277 } 1278 *hbitp = uint8(hb) 1279 hbitp = add1(hbitp) 1280 b >>= 4 1281 } 1282 1283 Phase3: 1284 // Phase 3: Write last byte or partial byte and zero the rest of the bitmap entries. 1285 if w > nw { 1286 // Counting the 4 entries in hb not yet written to memory, 1287 // there are more entries than possible pointer slots. 1288 // Discard the excess entries (can't be more than 3). 1289 mask := uintptr(1)<<(4-(w-nw)) - 1 1290 hb &= mask | mask<<4 // apply mask to both pointer bits and scan bits 1291 } 1292 1293 // Change nw from counting possibly-pointer words to total words in allocation. 1294 nw = size / sys.PtrSize 1295 1296 // Write whole bitmap bytes. 1297 // The first is hb, the rest are zero. 1298 if w <= nw { 1299 *hbitp = uint8(hb) 1300 hbitp = add1(hbitp) 1301 hb = 0 // for possible final half-byte below 1302 for w += 4; w <= nw; w += 4 { 1303 *hbitp = 0 1304 hbitp = add1(hbitp) 1305 } 1306 } 1307 1308 // Write final partial bitmap byte if any. 1309 // We know w > nw, or else we'd still be in the loop above. 1310 // It can be bigger only due to the 4 entries in hb that it counts. 1311 // If w == nw+4 then there's nothing left to do: we wrote all nw entries 1312 // and can discard the 4 sitting in hb. 1313 // But if w == nw+2, we need to write first two in hb. 1314 // The byte is shared with the next object, so be careful with 1315 // existing bits. 1316 if w == nw+2 { 1317 *hbitp = *hbitp&^(bitPointer|bitScan|(bitPointer|bitScan)<<heapBitsShift) | uint8(hb) 1318 } 1319 1320 Phase4: 1321 // Phase 4: Copy unrolled bitmap to per-arena bitmaps, if necessary. 1322 if outOfPlace { 1323 // TODO: We could probably make this faster by 1324 // handling [x+dataSize, x+size) specially. 1325 h := heapBitsForAddr(x) 1326 // cnw is the number of heap words, or bit pairs 1327 // remaining (like nw above). 1328 cnw := size / sys.PtrSize 1329 src := (*uint8)(unsafe.Pointer(x)) 1330 // We know the first and last byte of the bitmap are 1331 // not the same, but it's still possible for small 1332 // objects span arenas, so it may share bitmap bytes 1333 // with neighboring objects. 1334 // 1335 // Handle the first byte specially if it's shared. See 1336 // Phase 1 for why this is the only special case we need. 1337 if doubleCheck { 1338 if !(h.shift == 0 || h.shift == 2) { 1339 print("x=", x, " size=", size, " cnw=", h.shift, "\n") 1340 throw("bad start shift") 1341 } 1342 } 1343 if h.shift == 2 { 1344 *h.bitp = *h.bitp&^((bitPointer|bitScan|(bitPointer|bitScan)<<heapBitsShift)<<(2*heapBitsShift)) | *src 1345 h = h.next().next() 1346 cnw -= 2 1347 src = addb(src, 1) 1348 } 1349 // We're now byte aligned. Copy out to per-arena 1350 // bitmaps until the last byte (which may again be 1351 // partial). 1352 for cnw >= 4 { 1353 // This loop processes four words at a time, 1354 // so round cnw down accordingly. 1355 hNext, words := h.forwardOrBoundary(cnw / 4 * 4) 1356 1357 // n is the number of bitmap bytes to copy. 1358 n := words / 4 1359 memmove(unsafe.Pointer(h.bitp), unsafe.Pointer(src), n) 1360 cnw -= words 1361 h = hNext 1362 src = addb(src, n) 1363 } 1364 if doubleCheck && h.shift != 0 { 1365 print("cnw=", cnw, " h.shift=", h.shift, "\n") 1366 throw("bad shift after block copy") 1367 } 1368 // Handle the last byte if it's shared. 1369 if cnw == 2 { 1370 *h.bitp = *h.bitp&^(bitPointer|bitScan|(bitPointer|bitScan)<<heapBitsShift) | *src 1371 src = addb(src, 1) 1372 h = h.next().next() 1373 } 1374 if doubleCheck { 1375 if uintptr(unsafe.Pointer(src)) > x+size { 1376 throw("copy exceeded object size") 1377 } 1378 if !(cnw == 0 || cnw == 2) { 1379 print("x=", x, " size=", size, " cnw=", cnw, "\n") 1380 throw("bad number of remaining words") 1381 } 1382 // Set up hbitp so doubleCheck code below can check it. 1383 hbitp = h.bitp 1384 } 1385 // Zero the object where we wrote the bitmap. 1386 memclrNoHeapPointers(unsafe.Pointer(x), uintptr(unsafe.Pointer(src))-x) 1387 } 1388 1389 // Double check the whole bitmap. 1390 if doubleCheck { 1391 // x+size may not point to the heap, so back up one 1392 // word and then advance it the way we do above. 1393 end := heapBitsForAddr(x + size - sys.PtrSize) 1394 if outOfPlace { 1395 // In out-of-place copying, we just advance 1396 // using next. 1397 end = end.next() 1398 } else { 1399 // Don't use next because that may advance to 1400 // the next arena and the in-place logic 1401 // doesn't do that. 1402 end.shift += heapBitsShift 1403 if end.shift == 4*heapBitsShift { 1404 end.bitp, end.shift = add1(end.bitp), 0 1405 } 1406 } 1407 if typ.kind&kindGCProg == 0 && (hbitp != end.bitp || (w == nw+2) != (end.shift == 2)) { 1408 println("ended at wrong bitmap byte for", typ.string(), "x", dataSize/typ.size) 1409 print("typ.size=", typ.size, " typ.ptrdata=", typ.ptrdata, " dataSize=", dataSize, " size=", size, "\n") 1410 print("w=", w, " nw=", nw, " b=", hex(b), " nb=", nb, " hb=", hex(hb), "\n") 1411 h0 := heapBitsForAddr(x) 1412 print("initial bits h0.bitp=", h0.bitp, " h0.shift=", h0.shift, "\n") 1413 print("ended at hbitp=", hbitp, " but next starts at bitp=", end.bitp, " shift=", end.shift, "\n") 1414 throw("bad heapBitsSetType") 1415 } 1416 1417 // Double-check that bits to be written were written correctly. 1418 // Does not check that other bits were not written, unfortunately. 1419 h := heapBitsForAddr(x) 1420 nptr := typ.ptrdata / sys.PtrSize 1421 ndata := typ.size / sys.PtrSize 1422 count := dataSize / typ.size 1423 totalptr := ((count-1)*typ.size + typ.ptrdata) / sys.PtrSize 1424 for i := uintptr(0); i < size/sys.PtrSize; i++ { 1425 j := i % ndata 1426 var have, want uint8 1427 have = (*h.bitp >> h.shift) & (bitPointer | bitScan) 1428 if i >= totalptr { 1429 if typ.kind&kindGCProg != 0 && i < (totalptr+3)/4*4 { 1430 // heapBitsSetTypeGCProg always fills 1431 // in full nibbles of bitScan. 1432 want = bitScan 1433 } 1434 } else { 1435 if j < nptr && (*addb(ptrmask, j/8)>>(j%8))&1 != 0 { 1436 want |= bitPointer 1437 } 1438 want |= bitScan 1439 } 1440 if have != want { 1441 println("mismatch writing bits for", typ.string(), "x", dataSize/typ.size) 1442 print("typ.size=", typ.size, " typ.ptrdata=", typ.ptrdata, " dataSize=", dataSize, " size=", size, "\n") 1443 print("kindGCProg=", typ.kind&kindGCProg != 0, " outOfPlace=", outOfPlace, "\n") 1444 print("w=", w, " nw=", nw, " b=", hex(b), " nb=", nb, " hb=", hex(hb), "\n") 1445 h0 := heapBitsForAddr(x) 1446 print("initial bits h0.bitp=", h0.bitp, " h0.shift=", h0.shift, "\n") 1447 print("current bits h.bitp=", h.bitp, " h.shift=", h.shift, " *h.bitp=", hex(*h.bitp), "\n") 1448 print("ptrmask=", ptrmask, " p=", p, " endp=", endp, " endnb=", endnb, " pbits=", hex(pbits), " b=", hex(b), " nb=", nb, "\n") 1449 println("at word", i, "offset", i*sys.PtrSize, "have", hex(have), "want", hex(want)) 1450 if typ.kind&kindGCProg != 0 { 1451 println("GC program:") 1452 dumpGCProg(addb(typ.gcdata, 4)) 1453 } 1454 throw("bad heapBitsSetType") 1455 } 1456 h = h.next() 1457 } 1458 if ptrmask == debugPtrmask.data { 1459 unlock(&debugPtrmask.lock) 1460 } 1461 } 1462 } 1463 1464 var debugPtrmask struct { 1465 lock mutex 1466 data *byte 1467 } 1468 1469 // heapBitsSetTypeGCProg implements heapBitsSetType using a GC program. 1470 // progSize is the size of the memory described by the program. 1471 // elemSize is the size of the element that the GC program describes (a prefix of). 1472 // dataSize is the total size of the intended data, a multiple of elemSize. 1473 // allocSize is the total size of the allocated memory. 1474 // 1475 // GC programs are only used for large allocations. 1476 // heapBitsSetType requires that allocSize is a multiple of 4 words, 1477 // so that the relevant bitmap bytes are not shared with surrounding 1478 // objects. 1479 func heapBitsSetTypeGCProg(h heapBits, progSize, elemSize, dataSize, allocSize uintptr, prog *byte) { 1480 if sys.PtrSize == 8 && allocSize%(4*sys.PtrSize) != 0 { 1481 // Alignment will be wrong. 1482 throw("heapBitsSetTypeGCProg: small allocation") 1483 } 1484 var totalBits uintptr 1485 if elemSize == dataSize { 1486 totalBits = runGCProg(prog, nil, h.bitp, 2) 1487 if totalBits*sys.PtrSize != progSize { 1488 println("runtime: heapBitsSetTypeGCProg: total bits", totalBits, "but progSize", progSize) 1489 throw("heapBitsSetTypeGCProg: unexpected bit count") 1490 } 1491 } else { 1492 count := dataSize / elemSize 1493 1494 // Piece together program trailer to run after prog that does: 1495 // literal(0) 1496 // repeat(1, elemSize-progSize-1) // zeros to fill element size 1497 // repeat(elemSize, count-1) // repeat that element for count 1498 // This zero-pads the data remaining in the first element and then 1499 // repeats that first element to fill the array. 1500 var trailer [40]byte // 3 varints (max 10 each) + some bytes 1501 i := 0 1502 if n := elemSize/sys.PtrSize - progSize/sys.PtrSize; n > 0 { 1503 // literal(0) 1504 trailer[i] = 0x01 1505 i++ 1506 trailer[i] = 0 1507 i++ 1508 if n > 1 { 1509 // repeat(1, n-1) 1510 trailer[i] = 0x81 1511 i++ 1512 n-- 1513 for ; n >= 0x80; n >>= 7 { 1514 trailer[i] = byte(n | 0x80) 1515 i++ 1516 } 1517 trailer[i] = byte(n) 1518 i++ 1519 } 1520 } 1521 // repeat(elemSize/ptrSize, count-1) 1522 trailer[i] = 0x80 1523 i++ 1524 n := elemSize / sys.PtrSize 1525 for ; n >= 0x80; n >>= 7 { 1526 trailer[i] = byte(n | 0x80) 1527 i++ 1528 } 1529 trailer[i] = byte(n) 1530 i++ 1531 n = count - 1 1532 for ; n >= 0x80; n >>= 7 { 1533 trailer[i] = byte(n | 0x80) 1534 i++ 1535 } 1536 trailer[i] = byte(n) 1537 i++ 1538 trailer[i] = 0 1539 i++ 1540 1541 runGCProg(prog, &trailer[0], h.bitp, 2) 1542 1543 // Even though we filled in the full array just now, 1544 // record that we only filled in up to the ptrdata of the 1545 // last element. This will cause the code below to 1546 // memclr the dead section of the final array element, 1547 // so that scanobject can stop early in the final element. 1548 totalBits = (elemSize*(count-1) + progSize) / sys.PtrSize 1549 } 1550 endProg := unsafe.Pointer(addb(h.bitp, (totalBits+3)/4)) 1551 endAlloc := unsafe.Pointer(addb(h.bitp, allocSize/sys.PtrSize/wordsPerBitmapByte)) 1552 memclrNoHeapPointers(endProg, uintptr(endAlloc)-uintptr(endProg)) 1553 } 1554 1555 // progToPointerMask returns the 1-bit pointer mask output by the GC program prog. 1556 // size the size of the region described by prog, in bytes. 1557 // The resulting bitvector will have no more than size/sys.PtrSize bits. 1558 func progToPointerMask(prog *byte, size uintptr) bitvector { 1559 n := (size/sys.PtrSize + 7) / 8 1560 x := (*[1 << 30]byte)(persistentalloc(n+1, 1, &memstats.buckhash_sys))[:n+1] 1561 x[len(x)-1] = 0xa1 // overflow check sentinel 1562 n = runGCProg(prog, nil, &x[0], 1) 1563 if x[len(x)-1] != 0xa1 { 1564 throw("progToPointerMask: overflow") 1565 } 1566 return bitvector{int32(n), &x[0]} 1567 } 1568 1569 // Packed GC pointer bitmaps, aka GC programs. 1570 // 1571 // For large types containing arrays, the type information has a 1572 // natural repetition that can be encoded to save space in the 1573 // binary and in the memory representation of the type information. 1574 // 1575 // The encoding is a simple Lempel-Ziv style bytecode machine 1576 // with the following instructions: 1577 // 1578 // 00000000: stop 1579 // 0nnnnnnn: emit n bits copied from the next (n+7)/8 bytes 1580 // 10000000 n c: repeat the previous n bits c times; n, c are varints 1581 // 1nnnnnnn c: repeat the previous n bits c times; c is a varint 1582 1583 // runGCProg executes the GC program prog, and then trailer if non-nil, 1584 // writing to dst with entries of the given size. 1585 // If size == 1, dst is a 1-bit pointer mask laid out moving forward from dst. 1586 // If size == 2, dst is the 2-bit heap bitmap, and writes move backward 1587 // starting at dst (because the heap bitmap does). In this case, the caller guarantees 1588 // that only whole bytes in dst need to be written. 1589 // 1590 // runGCProg returns the number of 1- or 2-bit entries written to memory. 1591 func runGCProg(prog, trailer, dst *byte, size int) uintptr { 1592 dstStart := dst 1593 1594 // Bits waiting to be written to memory. 1595 var bits uintptr 1596 var nbits uintptr 1597 1598 p := prog 1599 Run: 1600 for { 1601 // Flush accumulated full bytes. 1602 // The rest of the loop assumes that nbits <= 7. 1603 for ; nbits >= 8; nbits -= 8 { 1604 if size == 1 { 1605 *dst = uint8(bits) 1606 dst = add1(dst) 1607 bits >>= 8 1608 } else { 1609 v := bits&bitPointerAll | bitScanAll 1610 *dst = uint8(v) 1611 dst = add1(dst) 1612 bits >>= 4 1613 v = bits&bitPointerAll | bitScanAll 1614 *dst = uint8(v) 1615 dst = add1(dst) 1616 bits >>= 4 1617 } 1618 } 1619 1620 // Process one instruction. 1621 inst := uintptr(*p) 1622 p = add1(p) 1623 n := inst & 0x7F 1624 if inst&0x80 == 0 { 1625 // Literal bits; n == 0 means end of program. 1626 if n == 0 { 1627 // Program is over; continue in trailer if present. 1628 if trailer != nil { 1629 p = trailer 1630 trailer = nil 1631 continue 1632 } 1633 break Run 1634 } 1635 nbyte := n / 8 1636 for i := uintptr(0); i < nbyte; i++ { 1637 bits |= uintptr(*p) << nbits 1638 p = add1(p) 1639 if size == 1 { 1640 *dst = uint8(bits) 1641 dst = add1(dst) 1642 bits >>= 8 1643 } else { 1644 v := bits&0xf | bitScanAll 1645 *dst = uint8(v) 1646 dst = add1(dst) 1647 bits >>= 4 1648 v = bits&0xf | bitScanAll 1649 *dst = uint8(v) 1650 dst = add1(dst) 1651 bits >>= 4 1652 } 1653 } 1654 if n %= 8; n > 0 { 1655 bits |= uintptr(*p) << nbits 1656 p = add1(p) 1657 nbits += n 1658 } 1659 continue Run 1660 } 1661 1662 // Repeat. If n == 0, it is encoded in a varint in the next bytes. 1663 if n == 0 { 1664 for off := uint(0); ; off += 7 { 1665 x := uintptr(*p) 1666 p = add1(p) 1667 n |= (x & 0x7F) << off 1668 if x&0x80 == 0 { 1669 break 1670 } 1671 } 1672 } 1673 1674 // Count is encoded in a varint in the next bytes. 1675 c := uintptr(0) 1676 for off := uint(0); ; off += 7 { 1677 x := uintptr(*p) 1678 p = add1(p) 1679 c |= (x & 0x7F) << off 1680 if x&0x80 == 0 { 1681 break 1682 } 1683 } 1684 c *= n // now total number of bits to copy 1685 1686 // If the number of bits being repeated is small, load them 1687 // into a register and use that register for the entire loop 1688 // instead of repeatedly reading from memory. 1689 // Handling fewer than 8 bits here makes the general loop simpler. 1690 // The cutoff is sys.PtrSize*8 - 7 to guarantee that when we add 1691 // the pattern to a bit buffer holding at most 7 bits (a partial byte) 1692 // it will not overflow. 1693 src := dst 1694 const maxBits = sys.PtrSize*8 - 7 1695 if n <= maxBits { 1696 // Start with bits in output buffer. 1697 pattern := bits 1698 npattern := nbits 1699 1700 // If we need more bits, fetch them from memory. 1701 if size == 1 { 1702 src = subtract1(src) 1703 for npattern < n { 1704 pattern <<= 8 1705 pattern |= uintptr(*src) 1706 src = subtract1(src) 1707 npattern += 8 1708 } 1709 } else { 1710 src = subtract1(src) 1711 for npattern < n { 1712 pattern <<= 4 1713 pattern |= uintptr(*src) & 0xf 1714 src = subtract1(src) 1715 npattern += 4 1716 } 1717 } 1718 1719 // We started with the whole bit output buffer, 1720 // and then we loaded bits from whole bytes. 1721 // Either way, we might now have too many instead of too few. 1722 // Discard the extra. 1723 if npattern > n { 1724 pattern >>= npattern - n 1725 npattern = n 1726 } 1727 1728 // Replicate pattern to at most maxBits. 1729 if npattern == 1 { 1730 // One bit being repeated. 1731 // If the bit is 1, make the pattern all 1s. 1732 // If the bit is 0, the pattern is already all 0s, 1733 // but we can claim that the number of bits 1734 // in the word is equal to the number we need (c), 1735 // because right shift of bits will zero fill. 1736 if pattern == 1 { 1737 pattern = 1<<maxBits - 1 1738 npattern = maxBits 1739 } else { 1740 npattern = c 1741 } 1742 } else { 1743 b := pattern 1744 nb := npattern 1745 if nb+nb <= maxBits { 1746 // Double pattern until the whole uintptr is filled. 1747 for nb <= sys.PtrSize*8 { 1748 b |= b << nb 1749 nb += nb 1750 } 1751 // Trim away incomplete copy of original pattern in high bits. 1752 // TODO(rsc): Replace with table lookup or loop on systems without divide? 1753 nb = maxBits / npattern * npattern 1754 b &= 1<<nb - 1 1755 pattern = b 1756 npattern = nb 1757 } 1758 } 1759 1760 // Add pattern to bit buffer and flush bit buffer, c/npattern times. 1761 // Since pattern contains >8 bits, there will be full bytes to flush 1762 // on each iteration. 1763 for ; c >= npattern; c -= npattern { 1764 bits |= pattern << nbits 1765 nbits += npattern 1766 if size == 1 { 1767 for nbits >= 8 { 1768 *dst = uint8(bits) 1769 dst = add1(dst) 1770 bits >>= 8 1771 nbits -= 8 1772 } 1773 } else { 1774 for nbits >= 4 { 1775 *dst = uint8(bits&0xf | bitScanAll) 1776 dst = add1(dst) 1777 bits >>= 4 1778 nbits -= 4 1779 } 1780 } 1781 } 1782 1783 // Add final fragment to bit buffer. 1784 if c > 0 { 1785 pattern &= 1<<c - 1 1786 bits |= pattern << nbits 1787 nbits += c 1788 } 1789 continue Run 1790 } 1791 1792 // Repeat; n too large to fit in a register. 1793 // Since nbits <= 7, we know the first few bytes of repeated data 1794 // are already written to memory. 1795 off := n - nbits // n > nbits because n > maxBits and nbits <= 7 1796 if size == 1 { 1797 // Leading src fragment. 1798 src = subtractb(src, (off+7)/8) 1799 if frag := off & 7; frag != 0 { 1800 bits |= uintptr(*src) >> (8 - frag) << nbits 1801 src = add1(src) 1802 nbits += frag 1803 c -= frag 1804 } 1805 // Main loop: load one byte, write another. 1806 // The bits are rotating through the bit buffer. 1807 for i := c / 8; i > 0; i-- { 1808 bits |= uintptr(*src) << nbits 1809 src = add1(src) 1810 *dst = uint8(bits) 1811 dst = add1(dst) 1812 bits >>= 8 1813 } 1814 // Final src fragment. 1815 if c %= 8; c > 0 { 1816 bits |= (uintptr(*src) & (1<<c - 1)) << nbits 1817 nbits += c 1818 } 1819 } else { 1820 // Leading src fragment. 1821 src = subtractb(src, (off+3)/4) 1822 if frag := off & 3; frag != 0 { 1823 bits |= (uintptr(*src) & 0xf) >> (4 - frag) << nbits 1824 src = add1(src) 1825 nbits += frag 1826 c -= frag 1827 } 1828 // Main loop: load one byte, write another. 1829 // The bits are rotating through the bit buffer. 1830 for i := c / 4; i > 0; i-- { 1831 bits |= (uintptr(*src) & 0xf) << nbits 1832 src = add1(src) 1833 *dst = uint8(bits&0xf | bitScanAll) 1834 dst = add1(dst) 1835 bits >>= 4 1836 } 1837 // Final src fragment. 1838 if c %= 4; c > 0 { 1839 bits |= (uintptr(*src) & (1<<c - 1)) << nbits 1840 nbits += c 1841 } 1842 } 1843 } 1844 1845 // Write any final bits out, using full-byte writes, even for the final byte. 1846 var totalBits uintptr 1847 if size == 1 { 1848 totalBits = (uintptr(unsafe.Pointer(dst))-uintptr(unsafe.Pointer(dstStart)))*8 + nbits 1849 nbits += -nbits & 7 1850 for ; nbits > 0; nbits -= 8 { 1851 *dst = uint8(bits) 1852 dst = add1(dst) 1853 bits >>= 8 1854 } 1855 } else { 1856 totalBits = (uintptr(unsafe.Pointer(dst))-uintptr(unsafe.Pointer(dstStart)))*4 + nbits 1857 nbits += -nbits & 3 1858 for ; nbits > 0; nbits -= 4 { 1859 v := bits&0xf | bitScanAll 1860 *dst = uint8(v) 1861 dst = add1(dst) 1862 bits >>= 4 1863 } 1864 } 1865 return totalBits 1866 } 1867 1868 // materializeGCProg allocates space for the (1-bit) pointer bitmask 1869 // for an object of size ptrdata. Then it fills that space with the 1870 // pointer bitmask specified by the program prog. 1871 // The bitmask starts at s.startAddr. 1872 // The result must be deallocated with dematerializeGCProg. 1873 func materializeGCProg(ptrdata uintptr, prog *byte) *mspan { 1874 // Each word of ptrdata needs one bit in the bitmap. 1875 bitmapBytes := divRoundUp(ptrdata, 8*sys.PtrSize) 1876 // Compute the number of pages needed for bitmapBytes. 1877 pages := divRoundUp(bitmapBytes, pageSize) 1878 s := mheap_.allocManual(pages, spanAllocPtrScalarBits) 1879 runGCProg(addb(prog, 4), nil, (*byte)(unsafe.Pointer(s.startAddr)), 1) 1880 return s 1881 } 1882 func dematerializeGCProg(s *mspan) { 1883 mheap_.freeManual(s, spanAllocPtrScalarBits) 1884 } 1885 1886 func dumpGCProg(p *byte) { 1887 nptr := 0 1888 for { 1889 x := *p 1890 p = add1(p) 1891 if x == 0 { 1892 print("\t", nptr, " end\n") 1893 break 1894 } 1895 if x&0x80 == 0 { 1896 print("\t", nptr, " lit ", x, ":") 1897 n := int(x+7) / 8 1898 for i := 0; i < n; i++ { 1899 print(" ", hex(*p)) 1900 p = add1(p) 1901 } 1902 print("\n") 1903 nptr += int(x) 1904 } else { 1905 nbit := int(x &^ 0x80) 1906 if nbit == 0 { 1907 for nb := uint(0); ; nb += 7 { 1908 x := *p 1909 p = add1(p) 1910 nbit |= int(x&0x7f) << nb 1911 if x&0x80 == 0 { 1912 break 1913 } 1914 } 1915 } 1916 count := 0 1917 for nb := uint(0); ; nb += 7 { 1918 x := *p 1919 p = add1(p) 1920 count |= int(x&0x7f) << nb 1921 if x&0x80 == 0 { 1922 break 1923 } 1924 } 1925 print("\t", nptr, " repeat ", nbit, " × ", count, "\n") 1926 nptr += nbit * count 1927 } 1928 } 1929 } 1930 1931 // Testing. 1932 1933 func getgcmaskcb(frame *stkframe, ctxt unsafe.Pointer) bool { 1934 target := (*stkframe)(ctxt) 1935 if frame.sp <= target.sp && target.sp < frame.varp { 1936 *target = *frame 1937 return false 1938 } 1939 return true 1940 } 1941 1942 // gcbits returns the GC type info for x, for testing. 1943 // The result is the bitmap entries (0 or 1), one entry per byte. 1944 //go:linkname reflect_gcbits reflect.gcbits 1945 func reflect_gcbits(x interface{}) []byte { 1946 ret := getgcmask(x) 1947 typ := (*ptrtype)(unsafe.Pointer(efaceOf(&x)._type)).elem 1948 nptr := typ.ptrdata / sys.PtrSize 1949 for uintptr(len(ret)) > nptr && ret[len(ret)-1] == 0 { 1950 ret = ret[:len(ret)-1] 1951 } 1952 return ret 1953 } 1954 1955 // Returns GC type info for the pointer stored in ep for testing. 1956 // If ep points to the stack, only static live information will be returned 1957 // (i.e. not for objects which are only dynamically live stack objects). 1958 func getgcmask(ep interface{}) (mask []byte) { 1959 e := *efaceOf(&ep) 1960 p := e.data 1961 t := e._type 1962 // data or bss 1963 for _, datap := range activeModules() { 1964 // data 1965 if datap.data <= uintptr(p) && uintptr(p) < datap.edata { 1966 bitmap := datap.gcdatamask.bytedata 1967 n := (*ptrtype)(unsafe.Pointer(t)).elem.size 1968 mask = make([]byte, n/sys.PtrSize) 1969 for i := uintptr(0); i < n; i += sys.PtrSize { 1970 off := (uintptr(p) + i - datap.data) / sys.PtrSize 1971 mask[i/sys.PtrSize] = (*addb(bitmap, off/8) >> (off % 8)) & 1 1972 } 1973 return 1974 } 1975 1976 // bss 1977 if datap.bss <= uintptr(p) && uintptr(p) < datap.ebss { 1978 bitmap := datap.gcbssmask.bytedata 1979 n := (*ptrtype)(unsafe.Pointer(t)).elem.size 1980 mask = make([]byte, n/sys.PtrSize) 1981 for i := uintptr(0); i < n; i += sys.PtrSize { 1982 off := (uintptr(p) + i - datap.bss) / sys.PtrSize 1983 mask[i/sys.PtrSize] = (*addb(bitmap, off/8) >> (off % 8)) & 1 1984 } 1985 return 1986 } 1987 } 1988 1989 // heap 1990 if base, s, _ := findObject(uintptr(p), 0, 0); base != 0 { 1991 hbits := heapBitsForAddr(base) 1992 n := s.elemsize 1993 mask = make([]byte, n/sys.PtrSize) 1994 for i := uintptr(0); i < n; i += sys.PtrSize { 1995 if hbits.isPointer() { 1996 mask[i/sys.PtrSize] = 1 1997 } 1998 if !hbits.morePointers() { 1999 mask = mask[:i/sys.PtrSize] 2000 break 2001 } 2002 hbits = hbits.next() 2003 } 2004 return 2005 } 2006 2007 // stack 2008 if _g_ := getg(); _g_.m.curg.stack.lo <= uintptr(p) && uintptr(p) < _g_.m.curg.stack.hi { 2009 var frame stkframe 2010 frame.sp = uintptr(p) 2011 _g_ := getg() 2012 gentraceback(_g_.m.curg.sched.pc, _g_.m.curg.sched.sp, 0, _g_.m.curg, 0, nil, 1000, getgcmaskcb, noescape(unsafe.Pointer(&frame)), 0) 2013 if frame.fn.valid() { 2014 locals, _, _ := getStackMap(&frame, nil, false) 2015 if locals.n == 0 { 2016 return 2017 } 2018 size := uintptr(locals.n) * sys.PtrSize 2019 n := (*ptrtype)(unsafe.Pointer(t)).elem.size 2020 mask = make([]byte, n/sys.PtrSize) 2021 for i := uintptr(0); i < n; i += sys.PtrSize { 2022 off := (uintptr(p) + i - frame.varp + size) / sys.PtrSize 2023 mask[i/sys.PtrSize] = locals.ptrbit(off) 2024 } 2025 } 2026 return 2027 } 2028 2029 // otherwise, not something the GC knows about. 2030 // possibly read-only data, like malloc(0). 2031 // must not have pointers 2032 return 2033 }