Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1 | // Copyright 2009 The Go Authors. All rights reserved. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 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 described |
| 10 | // by 1-bit bitmaps in which 0 means uninteresting and 1 means live pointer |
| 11 | // to be visited during GC. The bits in each byte are consumed starting with |
| 12 | // the low bit: 1<<0, 1<<1, and so on. |
| 13 | // |
| 14 | // Heap bitmap |
| 15 | // |
| 16 | // The allocated heap comes from a subset of the memory in the range [start, used), |
| 17 | // where start == mheap_.arena_start and used == mheap_.arena_used. |
| 18 | // The heap bitmap comprises 2 bits for each pointer-sized word in that range, |
| 19 | // stored in bytes indexed backward in memory from start. |
| 20 | // That is, the byte at address start-1 holds the 2-bit entries for the four words |
| 21 | // start through start+3*ptrSize, the byte at start-2 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 holds the same information as in the 1-bit |
| 25 | // bitmaps: 0 means uninteresting and 1 means live pointer to be visited during GC. |
| 26 | // The meaning of the high bit depends on the position of the word being described |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 27 | // in its allocated object. In all words *except* the second word, the |
| 28 | // high bit indicates that the object is still being described. In |
| 29 | // these words, if a bit pair with a high bit 0 is encountered, the |
| 30 | // low bit can also be assumed to be 0, and the object description is |
| 31 | // over. This 00 is called the ``dead'' encoding: it signals that the |
| 32 | // rest of the words in the object are uninteresting to the garbage |
| 33 | // collector. |
| 34 | // |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 35 | // In the second word, the high bit is the GC ``checkmarked'' bit (see below). |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 36 | // |
| 37 | // The 2-bit entries are split when written into the byte, so that the top half |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 38 | // of the byte contains 4 high bits and the bottom half contains 4 low (pointer) |
| 39 | // bits. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 40 | // This form allows a copy from the 1-bit to the 4-bit form to keep the |
| 41 | // pointer bits contiguous, instead of having to space them out. |
| 42 | // |
| 43 | // The code makes use of the fact that the zero value for a heap bitmap |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 44 | // has no live pointer bit set and is (depending on position), not used, |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 45 | // not checkmarked, and is the dead encoding. |
| 46 | // These properties must be preserved when modifying the encoding. |
| 47 | // |
| 48 | // Checkmarks |
| 49 | // |
| 50 | // In a concurrent garbage collector, one worries about failing to mark |
| 51 | // a live object due to mutations without write barriers or bugs in the |
| 52 | // collector implementation. As a sanity check, the GC has a 'checkmark' |
| 53 | // mode that retraverses the object graph with the world stopped, to make |
| 54 | // sure that everything that should be marked is marked. |
| 55 | // In checkmark mode, in the heap bitmap, the high bit of the 2-bit entry |
| 56 | // for the second word of the object holds the checkmark bit. |
| 57 | // When not in checkmark mode, this bit is set to 1. |
| 58 | // |
| 59 | // The smallest possible allocation is 8 bytes. On a 32-bit machine, that |
| 60 | // means every allocated object has two words, so there is room for the |
| 61 | // checkmark bit. On a 64-bit machine, however, the 8-byte allocation is |
| 62 | // just one word, so the second bit pair is not available for encoding the |
| 63 | // checkmark. However, because non-pointer allocations are combined |
| 64 | // into larger 16-byte (maxTinySize) allocations, a plain 8-byte allocation |
| 65 | // must be a pointer, so the type bit in the first word is not actually needed. |
| 66 | // It is still used in general, except in checkmark the type bit is repurposed |
| 67 | // as the checkmark bit and then reinitialized (to 1) as the type bit when |
| 68 | // finished. |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 69 | // |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 70 | |
| 71 | package runtime |
| 72 | |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 73 | import ( |
| 74 | "runtime/internal/atomic" |
| 75 | "runtime/internal/sys" |
| 76 | "unsafe" |
| 77 | ) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 78 | |
| 79 | const ( |
| 80 | bitPointer = 1 << 0 |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 81 | bitScan = 1 << 4 |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 82 | |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 83 | heapBitsShift = 1 // shift offset between successive bitPointer or bitScan entries |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 84 | heapBitmapScale = sys.PtrSize * (8 / 2) // number of data bytes described by one heap bitmap byte |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 85 | |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 86 | // all scan/pointer bits in a byte |
| 87 | bitScanAll = bitScan | bitScan<<heapBitsShift | bitScan<<(2*heapBitsShift) | bitScan<<(3*heapBitsShift) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 88 | bitPointerAll = bitPointer | bitPointer<<heapBitsShift | bitPointer<<(2*heapBitsShift) | bitPointer<<(3*heapBitsShift) |
| 89 | ) |
| 90 | |
| 91 | // addb returns the byte pointer p+n. |
| 92 | //go:nowritebarrier |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 93 | //go:nosplit |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 94 | func addb(p *byte, n uintptr) *byte { |
| 95 | // Note: wrote out full expression instead of calling add(p, n) |
| 96 | // to reduce the number of temporaries generated by the |
| 97 | // compiler for this trivial expression during inlining. |
| 98 | return (*byte)(unsafe.Pointer(uintptr(unsafe.Pointer(p)) + n)) |
| 99 | } |
| 100 | |
| 101 | // subtractb returns the byte pointer p-n. |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 102 | // subtractb is typically used when traversing the pointer tables referred to by hbits |
| 103 | // which are arranged in reverse order. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 104 | //go:nowritebarrier |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 105 | //go:nosplit |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 106 | func subtractb(p *byte, n uintptr) *byte { |
| 107 | // Note: wrote out full expression instead of calling add(p, -n) |
| 108 | // to reduce the number of temporaries generated by the |
| 109 | // compiler for this trivial expression during inlining. |
| 110 | return (*byte)(unsafe.Pointer(uintptr(unsafe.Pointer(p)) - n)) |
| 111 | } |
| 112 | |
| 113 | // add1 returns the byte pointer p+1. |
| 114 | //go:nowritebarrier |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 115 | //go:nosplit |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 116 | func add1(p *byte) *byte { |
| 117 | // Note: wrote out full expression instead of calling addb(p, 1) |
| 118 | // to reduce the number of temporaries generated by the |
| 119 | // compiler for this trivial expression during inlining. |
| 120 | return (*byte)(unsafe.Pointer(uintptr(unsafe.Pointer(p)) + 1)) |
| 121 | } |
| 122 | |
| 123 | // subtract1 returns the byte pointer p-1. |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 124 | // subtract1 is typically used when traversing the pointer tables referred to by hbits |
| 125 | // which are arranged in reverse order. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 126 | //go:nowritebarrier |
| 127 | // |
| 128 | // nosplit because it is used during write barriers and must not be preempted. |
| 129 | //go:nosplit |
| 130 | func subtract1(p *byte) *byte { |
| 131 | // Note: wrote out full expression instead of calling subtractb(p, 1) |
| 132 | // to reduce the number of temporaries generated by the |
| 133 | // compiler for this trivial expression during inlining. |
| 134 | return (*byte)(unsafe.Pointer(uintptr(unsafe.Pointer(p)) - 1)) |
| 135 | } |
| 136 | |
| 137 | // mHeap_MapBits is called each time arena_used is extended. |
| 138 | // It maps any additional bitmap memory needed for the new arena memory. |
| 139 | // It must be called with the expected new value of arena_used, |
| 140 | // *before* h.arena_used has been updated. |
| 141 | // Waiting to update arena_used until after the memory has been mapped |
| 142 | // avoids faults when other threads try access the bitmap immediately |
| 143 | // after observing the change to arena_used. |
| 144 | // |
| 145 | //go:nowritebarrier |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 146 | func (h *mheap) mapBits(arena_used uintptr) { |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 147 | // Caller has added extra mappings to the arena. |
| 148 | // Add extra mappings of bitmap words as needed. |
| 149 | // We allocate extra bitmap pieces in chunks of bitmapChunk. |
| 150 | const bitmapChunk = 8192 |
| 151 | |
| 152 | n := (arena_used - mheap_.arena_start) / heapBitmapScale |
| 153 | n = round(n, bitmapChunk) |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 154 | n = round(n, physPageSize) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 155 | if h.bitmap_mapped >= n { |
| 156 | return |
| 157 | } |
| 158 | |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 159 | sysMap(unsafe.Pointer(h.bitmap-n), n-h.bitmap_mapped, h.arena_reserved, &memstats.gc_sys) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 160 | h.bitmap_mapped = n |
| 161 | } |
| 162 | |
| 163 | // heapBits provides access to the bitmap bits for a single heap word. |
| 164 | // The methods on heapBits take value receivers so that the compiler |
| 165 | // can more easily inline calls to those methods and registerize the |
| 166 | // struct fields independently. |
| 167 | type heapBits struct { |
| 168 | bitp *uint8 |
| 169 | shift uint32 |
| 170 | } |
| 171 | |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 172 | // markBits provides access to the mark bit for an object in the heap. |
| 173 | // bytep points to the byte holding the mark bit. |
| 174 | // mask is a byte with a single bit set that can be &ed with *bytep |
| 175 | // to see if the bit has been set. |
| 176 | // *m.byte&m.mask != 0 indicates the mark bit is set. |
| 177 | // index can be used along with span information to generate |
| 178 | // the address of the object in the heap. |
| 179 | // We maintain one set of mark bits for allocation and one for |
| 180 | // marking purposes. |
| 181 | type markBits struct { |
| 182 | bytep *uint8 |
| 183 | mask uint8 |
| 184 | index uintptr |
| 185 | } |
| 186 | |
| 187 | //go:nosplit |
| 188 | func (s *mspan) allocBitsForIndex(allocBitIndex uintptr) markBits { |
| 189 | whichByte := allocBitIndex / 8 |
| 190 | whichBit := allocBitIndex % 8 |
| 191 | bytePtr := addb(s.allocBits, whichByte) |
| 192 | return markBits{bytePtr, uint8(1 << whichBit), allocBitIndex} |
| 193 | } |
| 194 | |
| 195 | // refillaCache takes 8 bytes s.allocBits starting at whichByte |
| 196 | // and negates them so that ctz (count trailing zeros) instructions |
| 197 | // can be used. It then places these 8 bytes into the cached 64 bit |
| 198 | // s.allocCache. |
| 199 | func (s *mspan) refillAllocCache(whichByte uintptr) { |
| 200 | bytes := (*[8]uint8)(unsafe.Pointer(addb(s.allocBits, whichByte))) |
| 201 | aCache := uint64(0) |
| 202 | aCache |= uint64(bytes[0]) |
| 203 | aCache |= uint64(bytes[1]) << (1 * 8) |
| 204 | aCache |= uint64(bytes[2]) << (2 * 8) |
| 205 | aCache |= uint64(bytes[3]) << (3 * 8) |
| 206 | aCache |= uint64(bytes[4]) << (4 * 8) |
| 207 | aCache |= uint64(bytes[5]) << (5 * 8) |
| 208 | aCache |= uint64(bytes[6]) << (6 * 8) |
| 209 | aCache |= uint64(bytes[7]) << (7 * 8) |
| 210 | s.allocCache = ^aCache |
| 211 | } |
| 212 | |
| 213 | // nextFreeIndex returns the index of the next free object in s at |
| 214 | // or after s.freeindex. |
| 215 | // There are hardware instructions that can be used to make this |
| 216 | // faster if profiling warrants it. |
| 217 | func (s *mspan) nextFreeIndex() uintptr { |
| 218 | sfreeindex := s.freeindex |
| 219 | snelems := s.nelems |
| 220 | if sfreeindex == snelems { |
| 221 | return sfreeindex |
| 222 | } |
| 223 | if sfreeindex > snelems { |
| 224 | throw("s.freeindex > s.nelems") |
| 225 | } |
| 226 | |
| 227 | aCache := s.allocCache |
| 228 | |
| 229 | bitIndex := sys.Ctz64(aCache) |
| 230 | for bitIndex == 64 { |
| 231 | // Move index to start of next cached bits. |
| 232 | sfreeindex = (sfreeindex + 64) &^ (64 - 1) |
| 233 | if sfreeindex >= snelems { |
| 234 | s.freeindex = snelems |
| 235 | return snelems |
| 236 | } |
| 237 | whichByte := sfreeindex / 8 |
| 238 | // Refill s.allocCache with the next 64 alloc bits. |
| 239 | s.refillAllocCache(whichByte) |
| 240 | aCache = s.allocCache |
| 241 | bitIndex = sys.Ctz64(aCache) |
| 242 | // nothing available in cached bits |
| 243 | // grab the next 8 bytes and try again. |
| 244 | } |
| 245 | result := sfreeindex + uintptr(bitIndex) |
| 246 | if result >= snelems { |
| 247 | s.freeindex = snelems |
| 248 | return snelems |
| 249 | } |
| 250 | |
| 251 | s.allocCache >>= (bitIndex + 1) |
| 252 | sfreeindex = result + 1 |
| 253 | |
| 254 | if sfreeindex%64 == 0 && sfreeindex != snelems { |
| 255 | // We just incremented s.freeindex so it isn't 0. |
| 256 | // As each 1 in s.allocCache was encountered and used for allocation |
| 257 | // it was shifted away. At this point s.allocCache contains all 0s. |
| 258 | // Refill s.allocCache so that it corresponds |
| 259 | // to the bits at s.allocBits starting at s.freeindex. |
| 260 | whichByte := sfreeindex / 8 |
| 261 | s.refillAllocCache(whichByte) |
| 262 | } |
| 263 | s.freeindex = sfreeindex |
| 264 | return result |
| 265 | } |
| 266 | |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 267 | // isFree returns whether the index'th object in s is unallocated. |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 268 | func (s *mspan) isFree(index uintptr) bool { |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 269 | if index < s.freeindex { |
| 270 | return false |
| 271 | } |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 272 | whichByte := index / 8 |
| 273 | whichBit := index % 8 |
| 274 | byteVal := *addb(s.allocBits, whichByte) |
| 275 | return byteVal&uint8(1<<whichBit) == 0 |
| 276 | } |
| 277 | |
| 278 | func (s *mspan) objIndex(p uintptr) uintptr { |
| 279 | byteOffset := p - s.base() |
| 280 | if byteOffset == 0 { |
| 281 | return 0 |
| 282 | } |
| 283 | if s.baseMask != 0 { |
| 284 | // s.baseMask is 0, elemsize is a power of two, so shift by s.divShift |
| 285 | return byteOffset >> s.divShift |
| 286 | } |
| 287 | return uintptr(((uint64(byteOffset) >> s.divShift) * uint64(s.divMul)) >> s.divShift2) |
| 288 | } |
| 289 | |
| 290 | func markBitsForAddr(p uintptr) markBits { |
| 291 | s := spanOf(p) |
| 292 | objIndex := s.objIndex(p) |
| 293 | return s.markBitsForIndex(objIndex) |
| 294 | } |
| 295 | |
| 296 | func (s *mspan) markBitsForIndex(objIndex uintptr) markBits { |
| 297 | whichByte := objIndex / 8 |
| 298 | bitMask := uint8(1 << (objIndex % 8)) // low 3 bits hold the bit index |
| 299 | bytePtr := addb(s.gcmarkBits, whichByte) |
| 300 | return markBits{bytePtr, bitMask, objIndex} |
| 301 | } |
| 302 | |
| 303 | func (s *mspan) markBitsForBase() markBits { |
| 304 | return markBits{s.gcmarkBits, uint8(1), 0} |
| 305 | } |
| 306 | |
| 307 | // isMarked reports whether mark bit m is set. |
| 308 | func (m markBits) isMarked() bool { |
| 309 | return *m.bytep&m.mask != 0 |
| 310 | } |
| 311 | |
| 312 | // setMarked sets the marked bit in the markbits, atomically. Some compilers |
| 313 | // are not able to inline atomic.Or8 function so if it appears as a hot spot consider |
| 314 | // inlining it manually. |
| 315 | func (m markBits) setMarked() { |
| 316 | // Might be racing with other updates, so use atomic update always. |
| 317 | // We used to be clever here and use a non-atomic update in certain |
| 318 | // cases, but it's not worth the risk. |
| 319 | atomic.Or8(m.bytep, m.mask) |
| 320 | } |
| 321 | |
| 322 | // setMarkedNonAtomic sets the marked bit in the markbits, non-atomically. |
| 323 | func (m markBits) setMarkedNonAtomic() { |
| 324 | *m.bytep |= m.mask |
| 325 | } |
| 326 | |
| 327 | // clearMarked clears the marked bit in the markbits, atomically. |
| 328 | func (m markBits) clearMarked() { |
| 329 | // Might be racing with other updates, so use atomic update always. |
| 330 | // We used to be clever here and use a non-atomic update in certain |
| 331 | // cases, but it's not worth the risk. |
| 332 | atomic.And8(m.bytep, ^m.mask) |
| 333 | } |
| 334 | |
| 335 | // clearMarkedNonAtomic clears the marked bit non-atomically. |
| 336 | func (m markBits) clearMarkedNonAtomic() { |
| 337 | *m.bytep ^= m.mask |
| 338 | } |
| 339 | |
| 340 | // markBitsForSpan returns the markBits for the span base address base. |
| 341 | func markBitsForSpan(base uintptr) (mbits markBits) { |
| 342 | if base < mheap_.arena_start || base >= mheap_.arena_used { |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 343 | throw("markBitsForSpan: base out of range") |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 344 | } |
| 345 | mbits = markBitsForAddr(base) |
| 346 | if mbits.mask != 1 { |
| 347 | throw("markBitsForSpan: unaligned start") |
| 348 | } |
| 349 | return mbits |
| 350 | } |
| 351 | |
| 352 | // advance advances the markBits to the next object in the span. |
| 353 | func (m *markBits) advance() { |
| 354 | if m.mask == 1<<7 { |
| 355 | m.bytep = (*uint8)(unsafe.Pointer(uintptr(unsafe.Pointer(m.bytep)) + 1)) |
| 356 | m.mask = 1 |
| 357 | } else { |
| 358 | m.mask = m.mask << 1 |
| 359 | } |
| 360 | m.index++ |
| 361 | } |
| 362 | |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 363 | // heapBitsForAddr returns the heapBits for the address addr. |
| 364 | // The caller must have already checked that addr is in the range [mheap_.arena_start, mheap_.arena_used). |
| 365 | // |
| 366 | // nosplit because it is used during write barriers and must not be preempted. |
| 367 | //go:nosplit |
| 368 | func heapBitsForAddr(addr uintptr) heapBits { |
| 369 | // 2 bits per work, 4 pairs per byte, and a mask is hard coded. |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 370 | off := (addr - mheap_.arena_start) / sys.PtrSize |
| 371 | return heapBits{(*uint8)(unsafe.Pointer(mheap_.bitmap - off/4 - 1)), uint32(off & 3)} |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 372 | } |
| 373 | |
| 374 | // heapBitsForSpan returns the heapBits for the span base address base. |
| 375 | func heapBitsForSpan(base uintptr) (hbits heapBits) { |
| 376 | if base < mheap_.arena_start || base >= mheap_.arena_used { |
| 377 | throw("heapBitsForSpan: base out of range") |
| 378 | } |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 379 | return heapBitsForAddr(base) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 380 | } |
| 381 | |
| 382 | // heapBitsForObject returns the base address for the heap object |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 383 | // containing the address p, the heapBits for base, |
| 384 | // the object's span, and of the index of the object in s. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 385 | // If p does not point into a heap object, |
| 386 | // return base == 0 |
| 387 | // otherwise return the base of the object. |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 388 | // |
| 389 | // refBase and refOff optionally give the base address of the object |
| 390 | // in which the pointer p was found and the byte offset at which it |
| 391 | // was found. These are used for error reporting. |
| 392 | func heapBitsForObject(p, refBase, refOff uintptr) (base uintptr, hbits heapBits, s *mspan, objIndex uintptr) { |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 393 | arenaStart := mheap_.arena_start |
| 394 | if p < arenaStart || p >= mheap_.arena_used { |
| 395 | return |
| 396 | } |
| 397 | off := p - arenaStart |
| 398 | idx := off >> _PageShift |
| 399 | // p points into the heap, but possibly to the middle of an object. |
| 400 | // Consult the span table to find the block beginning. |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 401 | s = mheap_.spans[idx] |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 402 | if s == nil || p < s.base() || p >= s.limit || s.state != mSpanInUse { |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 403 | if s == nil || s.state == _MSpanStack { |
| 404 | // If s is nil, the virtual address has never been part of the heap. |
| 405 | // This pointer may be to some mmap'd region, so we allow it. |
| 406 | // Pointers into stacks are also ok, the runtime manages these explicitly. |
| 407 | return |
| 408 | } |
| 409 | |
| 410 | // The following ensures that we are rigorous about what data |
| 411 | // structures hold valid pointers. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 412 | if debug.invalidptr != 0 { |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 413 | // Typically this indicates an incorrect use |
| 414 | // of unsafe or cgo to store a bad pointer in |
| 415 | // the Go heap. It may also indicate a runtime |
| 416 | // bug. |
| 417 | // |
| 418 | // TODO(austin): We could be more aggressive |
| 419 | // and detect pointers to unallocated objects |
| 420 | // in allocated spans. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 421 | printlock() |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 422 | print("runtime: pointer ", hex(p)) |
| 423 | if s.state != mSpanInUse { |
| 424 | print(" to unallocated span") |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 425 | } else { |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 426 | print(" to unused region of span") |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 427 | } |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 428 | print(" idx=", hex(idx), " span.base()=", hex(s.base()), " span.limit=", hex(s.limit), " span.state=", s.state, "\n") |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 429 | if refBase != 0 { |
| 430 | print("runtime: found in object at *(", hex(refBase), "+", hex(refOff), ")\n") |
| 431 | gcDumpObject("object", refBase, refOff) |
| 432 | } |
| 433 | throw("found bad pointer in Go heap (incorrect use of unsafe or cgo?)") |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 434 | } |
| 435 | return |
| 436 | } |
| 437 | // If this span holds object of a power of 2 size, just mask off the bits to |
| 438 | // the interior of the object. Otherwise use the size to get the base. |
| 439 | if s.baseMask != 0 { |
| 440 | // optimize for power of 2 sized objects. |
| 441 | base = s.base() |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 442 | base = base + (p-base)&uintptr(s.baseMask) |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 443 | objIndex = (base - s.base()) >> s.divShift |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 444 | // base = p & s.baseMask is faster for small spans, |
| 445 | // but doesn't work for large spans. |
| 446 | // Overall, it's faster to use the more general computation above. |
| 447 | } else { |
| 448 | base = s.base() |
| 449 | if p-base >= s.elemsize { |
| 450 | // n := (p - base) / s.elemsize, using division by multiplication |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 451 | objIndex = uintptr(p-base) >> s.divShift * uintptr(s.divMul) >> s.divShift2 |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 452 | base += objIndex * s.elemsize |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 453 | } |
| 454 | } |
| 455 | // Now that we know the actual base, compute heapBits to return to caller. |
| 456 | hbits = heapBitsForAddr(base) |
| 457 | return |
| 458 | } |
| 459 | |
| 460 | // prefetch the bits. |
| 461 | func (h heapBits) prefetch() { |
| 462 | prefetchnta(uintptr(unsafe.Pointer((h.bitp)))) |
| 463 | } |
| 464 | |
| 465 | // next returns the heapBits describing the next pointer-sized word in memory. |
| 466 | // That is, if h describes address p, h.next() describes p+ptrSize. |
| 467 | // Note that next does not modify h. The caller must record the result. |
| 468 | // |
| 469 | // nosplit because it is used during write barriers and must not be preempted. |
| 470 | //go:nosplit |
| 471 | func (h heapBits) next() heapBits { |
| 472 | if h.shift < 3*heapBitsShift { |
| 473 | return heapBits{h.bitp, h.shift + heapBitsShift} |
| 474 | } |
| 475 | return heapBits{subtract1(h.bitp), 0} |
| 476 | } |
| 477 | |
| 478 | // forward returns the heapBits describing n pointer-sized words ahead of h in memory. |
| 479 | // That is, if h describes address p, h.forward(n) describes p+n*ptrSize. |
| 480 | // h.forward(1) is equivalent to h.next(), just slower. |
| 481 | // Note that forward does not modify h. The caller must record the result. |
| 482 | // bits returns the heap bits for the current word. |
| 483 | func (h heapBits) forward(n uintptr) heapBits { |
| 484 | n += uintptr(h.shift) / heapBitsShift |
| 485 | return heapBits{subtractb(h.bitp, n/4), uint32(n%4) * heapBitsShift} |
| 486 | } |
| 487 | |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 488 | // The caller can test morePointers and isPointer by &-ing with bitScan and bitPointer. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 489 | // The result includes in its higher bits the bits for subsequent words |
| 490 | // described by the same bitmap byte. |
| 491 | func (h heapBits) bits() uint32 { |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 492 | // The (shift & 31) eliminates a test and conditional branch |
| 493 | // from the generated code. |
| 494 | return uint32(*h.bitp) >> (h.shift & 31) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 495 | } |
| 496 | |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 497 | // morePointers returns true if this word and all remaining words in this object |
| 498 | // are scalars. |
| 499 | // h must not describe the second word of the object. |
| 500 | func (h heapBits) morePointers() bool { |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 501 | return h.bits()&bitScan != 0 |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 502 | } |
| 503 | |
| 504 | // isPointer reports whether the heap bits describe a pointer word. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 505 | // |
| 506 | // nosplit because it is used during write barriers and must not be preempted. |
| 507 | //go:nosplit |
| 508 | func (h heapBits) isPointer() bool { |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 509 | return h.bits()&bitPointer != 0 |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 510 | } |
| 511 | |
| 512 | // hasPointers reports whether the given object has any pointers. |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 513 | // It must be told how large the object at h is for efficiency. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 514 | // h must describe the initial word of the object. |
| 515 | func (h heapBits) hasPointers(size uintptr) bool { |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 516 | if size == sys.PtrSize { // 1-word objects are always pointers |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 517 | return true |
| 518 | } |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 519 | return (*h.bitp>>h.shift)&bitScan != 0 |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 520 | } |
| 521 | |
| 522 | // isCheckmarked reports whether the heap bits have the checkmarked bit set. |
| 523 | // It must be told how large the object at h is, because the encoding of the |
| 524 | // checkmark bit varies by size. |
| 525 | // h must describe the initial word of the object. |
| 526 | func (h heapBits) isCheckmarked(size uintptr) bool { |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 527 | if size == sys.PtrSize { |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 528 | return (*h.bitp>>h.shift)&bitPointer != 0 |
| 529 | } |
| 530 | // All multiword objects are 2-word aligned, |
| 531 | // so we know that the initial word's 2-bit pair |
| 532 | // and the second word's 2-bit pair are in the |
| 533 | // same heap bitmap byte, *h.bitp. |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 534 | return (*h.bitp>>(heapBitsShift+h.shift))&bitScan != 0 |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 535 | } |
| 536 | |
| 537 | // setCheckmarked sets the checkmarked bit. |
| 538 | // It must be told how large the object at h is, because the encoding of the |
| 539 | // checkmark bit varies by size. |
| 540 | // h must describe the initial word of the object. |
| 541 | func (h heapBits) setCheckmarked(size uintptr) { |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 542 | if size == sys.PtrSize { |
| 543 | atomic.Or8(h.bitp, bitPointer<<h.shift) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 544 | return |
| 545 | } |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 546 | atomic.Or8(h.bitp, bitScan<<(heapBitsShift+h.shift)) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 547 | } |
| 548 | |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 549 | // bulkBarrierPreWrite executes writebarrierptr_prewrite1 |
| 550 | // for every pointer slot in the memory range [src, src+size), |
| 551 | // using pointer/scalar information from [dst, dst+size). |
| 552 | // This executes the write barriers necessary before a memmove. |
| 553 | // src, dst, and size must be pointer-aligned. |
| 554 | // The range [dst, dst+size) must lie within a single object. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 555 | // |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 556 | // As a special case, src == 0 indicates that this is being used for a |
| 557 | // memclr. bulkBarrierPreWrite will pass 0 for the src of each write |
| 558 | // barrier. |
| 559 | // |
| 560 | // Callers should call bulkBarrierPreWrite immediately before |
| 561 | // calling memmove(dst, src, size). This function is marked nosplit |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 562 | // to avoid being preempted; the GC must not stop the goroutine |
| 563 | // between the memmove and the execution of the barriers. |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 564 | // The caller is also responsible for cgo pointer checks if this |
| 565 | // may be writing Go pointers into non-Go memory. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 566 | // |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 567 | // The pointer bitmap is not maintained for allocations containing |
| 568 | // no pointers at all; any caller of bulkBarrierPreWrite must first |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 569 | // make sure the underlying allocation contains pointers, usually |
| 570 | // by checking typ.kind&kindNoPointers. |
| 571 | // |
| 572 | //go:nosplit |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 573 | func bulkBarrierPreWrite(dst, src, size uintptr) { |
| 574 | if (dst|src|size)&(sys.PtrSize-1) != 0 { |
| 575 | throw("bulkBarrierPreWrite: unaligned arguments") |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 576 | } |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 577 | if !writeBarrier.needed { |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 578 | return |
| 579 | } |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 580 | if !inheap(dst) { |
| 581 | // If dst is on the stack and in a higher frame than the |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 582 | // caller, we either need to execute write barriers on |
| 583 | // it (which is what happens for normal stack writes |
| 584 | // through pointers to higher frames), or we need to |
| 585 | // force the mark termination stack scan to scan the |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 586 | // frame containing dst. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 587 | // |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 588 | // Executing write barriers on dst is complicated in the |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 589 | // general case because we either need to unwind the |
| 590 | // stack to get the stack map, or we need the type's |
| 591 | // bitmap, which may be a GC program. |
| 592 | // |
| 593 | // Hence, we opt for forcing the re-scan to scan the |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 594 | // frame containing dst, which we can do by simply |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 595 | // unwinding the stack barriers between the current SP |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 596 | // and dst's frame. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 597 | gp := getg().m.curg |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 598 | if gp != nil && gp.stack.lo <= dst && dst < gp.stack.hi { |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 599 | // Run on the system stack to give it more |
| 600 | // stack space. |
| 601 | systemstack(func() { |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 602 | gcUnwindBarriers(gp, dst) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 603 | }) |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 604 | return |
| 605 | } |
| 606 | |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 607 | // If dst is a global, use the data or BSS bitmaps to |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 608 | // execute write barriers. |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 609 | for _, datap := range activeModules() { |
| 610 | if datap.data <= dst && dst < datap.edata { |
| 611 | bulkBarrierBitmap(dst, src, size, dst-datap.data, datap.gcdatamask.bytedata) |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 612 | return |
| 613 | } |
| 614 | } |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 615 | for _, datap := range activeModules() { |
| 616 | if datap.bss <= dst && dst < datap.ebss { |
| 617 | bulkBarrierBitmap(dst, src, size, dst-datap.bss, datap.gcbssmask.bytedata) |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 618 | return |
| 619 | } |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 620 | } |
| 621 | return |
| 622 | } |
| 623 | |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 624 | h := heapBitsForAddr(dst) |
| 625 | if src == 0 { |
| 626 | for i := uintptr(0); i < size; i += sys.PtrSize { |
| 627 | if h.isPointer() { |
| 628 | dstx := (*uintptr)(unsafe.Pointer(dst + i)) |
| 629 | writebarrierptr_prewrite1(dstx, 0) |
| 630 | } |
| 631 | h = h.next() |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 632 | } |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 633 | } else { |
| 634 | for i := uintptr(0); i < size; i += sys.PtrSize { |
| 635 | if h.isPointer() { |
| 636 | dstx := (*uintptr)(unsafe.Pointer(dst + i)) |
| 637 | srcx := (*uintptr)(unsafe.Pointer(src + i)) |
| 638 | writebarrierptr_prewrite1(dstx, *srcx) |
| 639 | } |
| 640 | h = h.next() |
| 641 | } |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 642 | } |
| 643 | } |
| 644 | |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 645 | // bulkBarrierBitmap executes write barriers for copying from [src, |
| 646 | // src+size) to [dst, dst+size) using a 1-bit pointer bitmap. src is |
| 647 | // assumed to start maskOffset bytes into the data covered by the |
| 648 | // bitmap in bits (which may not be a multiple of 8). |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 649 | // |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 650 | // This is used by bulkBarrierPreWrite for writes to data and BSS. |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 651 | // |
| 652 | //go:nosplit |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 653 | func bulkBarrierBitmap(dst, src, size, maskOffset uintptr, bits *uint8) { |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 654 | word := maskOffset / sys.PtrSize |
| 655 | bits = addb(bits, word/8) |
| 656 | mask := uint8(1) << (word % 8) |
| 657 | |
| 658 | for i := uintptr(0); i < size; i += sys.PtrSize { |
| 659 | if mask == 0 { |
| 660 | bits = addb(bits, 1) |
| 661 | if *bits == 0 { |
| 662 | // Skip 8 words. |
| 663 | i += 7 * sys.PtrSize |
| 664 | continue |
| 665 | } |
| 666 | mask = 1 |
| 667 | } |
| 668 | if *bits&mask != 0 { |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 669 | dstx := (*uintptr)(unsafe.Pointer(dst + i)) |
| 670 | if src == 0 { |
| 671 | writebarrierptr_prewrite1(dstx, 0) |
| 672 | } else { |
| 673 | srcx := (*uintptr)(unsafe.Pointer(src + i)) |
| 674 | writebarrierptr_prewrite1(dstx, *srcx) |
| 675 | } |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 676 | } |
| 677 | mask <<= 1 |
| 678 | } |
| 679 | } |
| 680 | |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 681 | // typeBitsBulkBarrier executes writebarrierptr_prewrite for every |
| 682 | // pointer that would be copied from [src, src+size) to [dst, |
| 683 | // dst+size) by a memmove using the type bitmap to locate those |
| 684 | // pointer slots. |
| 685 | // |
| 686 | // The type typ must correspond exactly to [src, src+size) and [dst, dst+size). |
| 687 | // dst, src, and size must be pointer-aligned. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 688 | // The type typ must have a plain bitmap, not a GC program. |
| 689 | // The only use of this function is in channel sends, and the |
| 690 | // 64 kB channel element limit takes care of this for us. |
| 691 | // |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 692 | // Must not be preempted because it typically runs right before memmove, |
| 693 | // and the GC must observe them as an atomic action. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 694 | // |
| 695 | //go:nosplit |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 696 | func typeBitsBulkBarrier(typ *_type, dst, src, size uintptr) { |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 697 | if typ == nil { |
| 698 | throw("runtime: typeBitsBulkBarrier without type") |
| 699 | } |
| 700 | if typ.size != size { |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 701 | println("runtime: typeBitsBulkBarrier with type ", typ.string(), " of size ", typ.size, " but memory size", size) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 702 | throw("runtime: invalid typeBitsBulkBarrier") |
| 703 | } |
| 704 | if typ.kind&kindGCProg != 0 { |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 705 | println("runtime: typeBitsBulkBarrier with type ", typ.string(), " with GC prog") |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 706 | throw("runtime: invalid typeBitsBulkBarrier") |
| 707 | } |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 708 | if !writeBarrier.needed { |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 709 | return |
| 710 | } |
| 711 | ptrmask := typ.gcdata |
| 712 | var bits uint32 |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 713 | for i := uintptr(0); i < typ.ptrdata; i += sys.PtrSize { |
| 714 | if i&(sys.PtrSize*8-1) == 0 { |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 715 | bits = uint32(*ptrmask) |
| 716 | ptrmask = addb(ptrmask, 1) |
| 717 | } else { |
| 718 | bits = bits >> 1 |
| 719 | } |
| 720 | if bits&1 != 0 { |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 721 | dstx := (*uintptr)(unsafe.Pointer(dst + i)) |
| 722 | srcx := (*uintptr)(unsafe.Pointer(src + i)) |
| 723 | writebarrierptr_prewrite(dstx, *srcx) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 724 | } |
| 725 | } |
| 726 | } |
| 727 | |
| 728 | // The methods operating on spans all require that h has been returned |
| 729 | // by heapBitsForSpan and that size, n, total are the span layout description |
| 730 | // returned by the mspan's layout method. |
| 731 | // If total > size*n, it means that there is extra leftover memory in the span, |
| 732 | // usually due to rounding. |
| 733 | // |
| 734 | // TODO(rsc): Perhaps introduce a different heapBitsSpan type. |
| 735 | |
| 736 | // initSpan initializes the heap bitmap for a span. |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 737 | // It clears all checkmark bits. |
| 738 | // If this is a span of pointer-sized objects, it initializes all |
| 739 | // words to pointer/scan. |
| 740 | // Otherwise, it initializes all words to scalar/dead. |
| 741 | func (h heapBits) initSpan(s *mspan) { |
| 742 | size, n, total := s.layout() |
| 743 | |
| 744 | // Init the markbit structures |
| 745 | s.freeindex = 0 |
| 746 | s.allocCache = ^uint64(0) // all 1s indicating all free. |
| 747 | s.nelems = n |
| 748 | s.allocBits = nil |
| 749 | s.gcmarkBits = nil |
| 750 | s.gcmarkBits = newMarkBits(s.nelems) |
| 751 | s.allocBits = newAllocBits(s.nelems) |
| 752 | |
| 753 | // Clear bits corresponding to objects. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 754 | if total%heapBitmapScale != 0 { |
| 755 | throw("initSpan: unaligned length") |
| 756 | } |
| 757 | nbyte := total / heapBitmapScale |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 758 | if sys.PtrSize == 8 && size == sys.PtrSize { |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 759 | end := h.bitp |
| 760 | bitp := subtractb(end, nbyte-1) |
| 761 | for { |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 762 | *bitp = bitPointerAll | bitScanAll |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 763 | if bitp == end { |
| 764 | break |
| 765 | } |
| 766 | bitp = add1(bitp) |
| 767 | } |
| 768 | return |
| 769 | } |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 770 | memclrNoHeapPointers(unsafe.Pointer(subtractb(h.bitp, nbyte-1)), nbyte) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 771 | } |
| 772 | |
| 773 | // initCheckmarkSpan initializes a span for being checkmarked. |
| 774 | // It clears the checkmark bits, which are set to 1 in normal operation. |
| 775 | func (h heapBits) initCheckmarkSpan(size, n, total uintptr) { |
| 776 | // The ptrSize == 8 is a compile-time constant false on 32-bit and eliminates this code entirely. |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 777 | if sys.PtrSize == 8 && size == sys.PtrSize { |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 778 | // Checkmark bit is type bit, bottom bit of every 2-bit entry. |
| 779 | // Only possible on 64-bit system, since minimum size is 8. |
| 780 | // Must clear type bit (checkmark bit) of every word. |
| 781 | // The type bit is the lower of every two-bit pair. |
| 782 | bitp := h.bitp |
| 783 | for i := uintptr(0); i < n; i += 4 { |
| 784 | *bitp &^= bitPointerAll |
| 785 | bitp = subtract1(bitp) |
| 786 | } |
| 787 | return |
| 788 | } |
| 789 | for i := uintptr(0); i < n; i++ { |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 790 | *h.bitp &^= bitScan << (heapBitsShift + h.shift) |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 791 | h = h.forward(size / sys.PtrSize) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 792 | } |
| 793 | } |
| 794 | |
| 795 | // clearCheckmarkSpan undoes all the checkmarking in a span. |
| 796 | // The actual checkmark bits are ignored, so the only work to do |
| 797 | // is to fix the pointer bits. (Pointer bits are ignored by scanobject |
| 798 | // but consulted by typedmemmove.) |
| 799 | func (h heapBits) clearCheckmarkSpan(size, n, total uintptr) { |
| 800 | // The ptrSize == 8 is a compile-time constant false on 32-bit and eliminates this code entirely. |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 801 | if sys.PtrSize == 8 && size == sys.PtrSize { |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 802 | // Checkmark bit is type bit, bottom bit of every 2-bit entry. |
| 803 | // Only possible on 64-bit system, since minimum size is 8. |
| 804 | // Must clear type bit (checkmark bit) of every word. |
| 805 | // The type bit is the lower of every two-bit pair. |
| 806 | bitp := h.bitp |
| 807 | for i := uintptr(0); i < n; i += 4 { |
| 808 | *bitp |= bitPointerAll |
| 809 | bitp = subtract1(bitp) |
| 810 | } |
| 811 | } |
| 812 | } |
| 813 | |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 814 | // oneBitCount is indexed by byte and produces the |
| 815 | // number of 1 bits in that byte. For example 128 has 1 bit set |
| 816 | // and oneBitCount[128] will holds 1. |
| 817 | var oneBitCount = [256]uint8{ |
| 818 | 0, 1, 1, 2, 1, 2, 2, 3, |
| 819 | 1, 2, 2, 3, 2, 3, 3, 4, |
| 820 | 1, 2, 2, 3, 2, 3, 3, 4, |
| 821 | 2, 3, 3, 4, 3, 4, 4, 5, |
| 822 | 1, 2, 2, 3, 2, 3, 3, 4, |
| 823 | 2, 3, 3, 4, 3, 4, 4, 5, |
| 824 | 2, 3, 3, 4, 3, 4, 4, 5, |
| 825 | 3, 4, 4, 5, 4, 5, 5, 6, |
| 826 | 1, 2, 2, 3, 2, 3, 3, 4, |
| 827 | 2, 3, 3, 4, 3, 4, 4, 5, |
| 828 | 2, 3, 3, 4, 3, 4, 4, 5, |
| 829 | 3, 4, 4, 5, 4, 5, 5, 6, |
| 830 | 2, 3, 3, 4, 3, 4, 4, 5, |
| 831 | 3, 4, 4, 5, 4, 5, 5, 6, |
| 832 | 3, 4, 4, 5, 4, 5, 5, 6, |
| 833 | 4, 5, 5, 6, 5, 6, 6, 7, |
| 834 | 1, 2, 2, 3, 2, 3, 3, 4, |
| 835 | 2, 3, 3, 4, 3, 4, 4, 5, |
| 836 | 2, 3, 3, 4, 3, 4, 4, 5, |
| 837 | 3, 4, 4, 5, 4, 5, 5, 6, |
| 838 | 2, 3, 3, 4, 3, 4, 4, 5, |
| 839 | 3, 4, 4, 5, 4, 5, 5, 6, |
| 840 | 3, 4, 4, 5, 4, 5, 5, 6, |
| 841 | 4, 5, 5, 6, 5, 6, 6, 7, |
| 842 | 2, 3, 3, 4, 3, 4, 4, 5, |
| 843 | 3, 4, 4, 5, 4, 5, 5, 6, |
| 844 | 3, 4, 4, 5, 4, 5, 5, 6, |
| 845 | 4, 5, 5, 6, 5, 6, 6, 7, |
| 846 | 3, 4, 4, 5, 4, 5, 5, 6, |
| 847 | 4, 5, 5, 6, 5, 6, 6, 7, |
| 848 | 4, 5, 5, 6, 5, 6, 6, 7, |
| 849 | 5, 6, 6, 7, 6, 7, 7, 8} |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 850 | |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 851 | // countFree runs through the mark bits in a span and counts the number of free objects |
| 852 | // in the span. |
| 853 | // TODO:(rlh) Use popcount intrinsic. |
| 854 | func (s *mspan) countFree() int { |
| 855 | count := 0 |
| 856 | maxIndex := s.nelems / 8 |
| 857 | for i := uintptr(0); i < maxIndex; i++ { |
| 858 | mrkBits := *addb(s.gcmarkBits, i) |
| 859 | count += int(oneBitCount[mrkBits]) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 860 | } |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 861 | if bitsInLastByte := s.nelems % 8; bitsInLastByte != 0 { |
| 862 | mrkBits := *addb(s.gcmarkBits, maxIndex) |
| 863 | mask := uint8((1 << bitsInLastByte) - 1) |
| 864 | bits := mrkBits & mask |
| 865 | count += int(oneBitCount[bits]) |
| 866 | } |
| 867 | return int(s.nelems) - count |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 868 | } |
| 869 | |
| 870 | // heapBitsSetType records that the new allocation [x, x+size) |
| 871 | // holds in [x, x+dataSize) one or more values of type typ. |
| 872 | // (The number of values is given by dataSize / typ.size.) |
| 873 | // If dataSize < size, the fragment [x+dataSize, x+size) is |
| 874 | // recorded as non-pointer data. |
| 875 | // It is known that the type has pointers somewhere; |
| 876 | // malloc does not call heapBitsSetType when there are no pointers, |
| 877 | // because all free objects are marked as noscan during |
| 878 | // heapBitsSweepSpan. |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 879 | // |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 880 | // There can only be one allocation from a given span active at a time, |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 881 | // and the bitmap for a span always falls on byte boundaries, |
| 882 | // so there are no write-write races for access to the heap bitmap. |
| 883 | // Hence, heapBitsSetType can access the bitmap without atomics. |
| 884 | // |
| 885 | // There can be read-write races between heapBitsSetType and things |
| 886 | // that read the heap bitmap like scanobject. However, since |
| 887 | // heapBitsSetType is only used for objects that have not yet been |
| 888 | // made reachable, readers will ignore bits being modified by this |
| 889 | // function. This does mean this function cannot transiently modify |
| 890 | // bits that belong to neighboring objects. Also, on weakly-ordered |
| 891 | // machines, callers must execute a store/store (publication) barrier |
| 892 | // between calling this function and making the object reachable. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 893 | func heapBitsSetType(x, size, dataSize uintptr, typ *_type) { |
| 894 | const doubleCheck = false // slow but helpful; enable to test modifications to this code |
| 895 | |
| 896 | // dataSize is always size rounded up to the next malloc size class, |
| 897 | // except in the case of allocating a defer block, in which case |
| 898 | // size is sizeof(_defer{}) (at least 6 words) and dataSize may be |
| 899 | // arbitrarily larger. |
| 900 | // |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 901 | // The checks for size == sys.PtrSize and size == 2*sys.PtrSize can therefore |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 902 | // assume that dataSize == size without checking it explicitly. |
| 903 | |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 904 | if sys.PtrSize == 8 && size == sys.PtrSize { |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 905 | // It's one word and it has pointers, it must be a pointer. |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 906 | // Since all allocated one-word objects are pointers |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 907 | // (non-pointers are aggregated into tinySize allocations), |
| 908 | // initSpan sets the pointer bits for us. Nothing to do here. |
| 909 | if doubleCheck { |
| 910 | h := heapBitsForAddr(x) |
| 911 | if !h.isPointer() { |
| 912 | throw("heapBitsSetType: pointer bit missing") |
| 913 | } |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 914 | if !h.morePointers() { |
| 915 | throw("heapBitsSetType: scan bit missing") |
| 916 | } |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 917 | } |
| 918 | return |
| 919 | } |
| 920 | |
| 921 | h := heapBitsForAddr(x) |
| 922 | ptrmask := typ.gcdata // start of 1-bit pointer mask (or GC program, handled below) |
| 923 | |
| 924 | // Heap bitmap bits for 2-word object are only 4 bits, |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 925 | // so also shared with objects next to it. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 926 | // This is called out as a special case primarily for 32-bit systems, |
| 927 | // so that on 32-bit systems the code below can assume all objects |
| 928 | // are 4-word aligned (because they're all 16-byte aligned). |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 929 | if size == 2*sys.PtrSize { |
| 930 | if typ.size == sys.PtrSize { |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 931 | // We're allocating a block big enough to hold two pointers. |
| 932 | // On 64-bit, that means the actual object must be two pointers, |
| 933 | // or else we'd have used the one-pointer-sized block. |
| 934 | // On 32-bit, however, this is the 8-byte block, the smallest one. |
| 935 | // So it could be that we're allocating one pointer and this was |
| 936 | // just the smallest block available. Distinguish by checking dataSize. |
| 937 | // (In general the number of instances of typ being allocated is |
| 938 | // dataSize/typ.size.) |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 939 | if sys.PtrSize == 4 && dataSize == sys.PtrSize { |
| 940 | // 1 pointer object. On 32-bit machines clear the bit for the |
| 941 | // unused second word. |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 942 | *h.bitp &^= (bitPointer | bitScan | ((bitPointer | bitScan) << heapBitsShift)) << h.shift |
| 943 | *h.bitp |= (bitPointer | bitScan) << h.shift |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 944 | } else { |
| 945 | // 2-element slice of pointer. |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 946 | *h.bitp |= (bitPointer | bitScan | bitPointer<<heapBitsShift) << h.shift |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 947 | } |
| 948 | return |
| 949 | } |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 950 | // Otherwise typ.size must be 2*sys.PtrSize, |
| 951 | // and typ.kind&kindGCProg == 0. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 952 | if doubleCheck { |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 953 | if typ.size != 2*sys.PtrSize || typ.kind&kindGCProg != 0 { |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 954 | print("runtime: heapBitsSetType size=", size, " but typ.size=", typ.size, " gcprog=", typ.kind&kindGCProg != 0, "\n") |
| 955 | throw("heapBitsSetType") |
| 956 | } |
| 957 | } |
| 958 | b := uint32(*ptrmask) |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 959 | hb := (b & 3) | bitScan |
| 960 | // bitPointer == 1, bitScan is 1 << 4, heapBitsShift is 1. |
| 961 | // 110011 is shifted h.shift and complemented. |
| 962 | // This clears out the bits that are about to be |
| 963 | // ored into *h.hbitp in the next instructions. |
| 964 | *h.bitp &^= (bitPointer | bitScan | ((bitPointer | bitScan) << heapBitsShift)) << h.shift |
| 965 | *h.bitp |= uint8(hb << h.shift) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 966 | return |
| 967 | } |
| 968 | |
| 969 | // Copy from 1-bit ptrmask into 2-bit bitmap. |
| 970 | // The basic approach is to use a single uintptr as a bit buffer, |
| 971 | // alternating between reloading the buffer and writing bitmap bytes. |
| 972 | // In general, one load can supply two bitmap byte writes. |
| 973 | // This is a lot of lines of code, but it compiles into relatively few |
| 974 | // machine instructions. |
| 975 | |
| 976 | var ( |
| 977 | // Ptrmask input. |
| 978 | p *byte // last ptrmask byte read |
| 979 | b uintptr // ptrmask bits already loaded |
| 980 | nb uintptr // number of bits in b at next read |
| 981 | endp *byte // final ptrmask byte to read (then repeat) |
| 982 | endnb uintptr // number of valid bits in *endp |
| 983 | pbits uintptr // alternate source of bits |
| 984 | |
| 985 | // Heap bitmap output. |
| 986 | w uintptr // words processed |
| 987 | nw uintptr // number of words to process |
| 988 | hbitp *byte // next heap bitmap byte to write |
| 989 | hb uintptr // bits being prepared for *hbitp |
| 990 | ) |
| 991 | |
| 992 | hbitp = h.bitp |
| 993 | |
| 994 | // Handle GC program. Delayed until this part of the code |
| 995 | // so that we can use the same double-checking mechanism |
| 996 | // as the 1-bit case. Nothing above could have encountered |
| 997 | // GC programs: the cases were all too small. |
| 998 | if typ.kind&kindGCProg != 0 { |
| 999 | heapBitsSetTypeGCProg(h, typ.ptrdata, typ.size, dataSize, size, addb(typ.gcdata, 4)) |
| 1000 | if doubleCheck { |
| 1001 | // Double-check the heap bits written by GC program |
| 1002 | // by running the GC program to create a 1-bit pointer mask |
| 1003 | // and then jumping to the double-check code below. |
| 1004 | // This doesn't catch bugs shared between the 1-bit and 4-bit |
| 1005 | // GC program execution, but it does catch mistakes specific |
| 1006 | // to just one of those and bugs in heapBitsSetTypeGCProg's |
| 1007 | // implementation of arrays. |
| 1008 | lock(&debugPtrmask.lock) |
| 1009 | if debugPtrmask.data == nil { |
| 1010 | debugPtrmask.data = (*byte)(persistentalloc(1<<20, 1, &memstats.other_sys)) |
| 1011 | } |
| 1012 | ptrmask = debugPtrmask.data |
| 1013 | runGCProg(addb(typ.gcdata, 4), nil, ptrmask, 1) |
| 1014 | goto Phase4 |
| 1015 | } |
| 1016 | return |
| 1017 | } |
| 1018 | |
| 1019 | // Note about sizes: |
| 1020 | // |
| 1021 | // typ.size is the number of words in the object, |
| 1022 | // and typ.ptrdata is the number of words in the prefix |
| 1023 | // of the object that contains pointers. That is, the final |
| 1024 | // typ.size - typ.ptrdata words contain no pointers. |
| 1025 | // This allows optimization of a common pattern where |
| 1026 | // an object has a small header followed by a large scalar |
| 1027 | // buffer. If we know the pointers are over, we don't have |
| 1028 | // to scan the buffer's heap bitmap at all. |
| 1029 | // The 1-bit ptrmasks are sized to contain only bits for |
| 1030 | // the typ.ptrdata prefix, zero padded out to a full byte |
| 1031 | // of bitmap. This code sets nw (below) so that heap bitmap |
| 1032 | // bits are only written for the typ.ptrdata prefix; if there is |
| 1033 | // more room in the allocated object, the next heap bitmap |
| 1034 | // entry is a 00, indicating that there are no more pointers |
| 1035 | // to scan. So only the ptrmask for the ptrdata bytes is needed. |
| 1036 | // |
| 1037 | // Replicated copies are not as nice: if there is an array of |
| 1038 | // objects with scalar tails, all but the last tail does have to |
| 1039 | // be initialized, because there is no way to say "skip forward". |
| 1040 | // However, because of the possibility of a repeated type with |
| 1041 | // size not a multiple of 4 pointers (one heap bitmap byte), |
| 1042 | // the code already must handle the last ptrmask byte specially |
| 1043 | // by treating it as containing only the bits for endnb pointers, |
| 1044 | // where endnb <= 4. We represent large scalar tails that must |
| 1045 | // be expanded in the replication by setting endnb larger than 4. |
| 1046 | // This will have the effect of reading many bits out of b, |
| 1047 | // but once the real bits are shifted out, b will supply as many |
| 1048 | // zero bits as we try to read, which is exactly what we need. |
| 1049 | |
| 1050 | p = ptrmask |
| 1051 | if typ.size < dataSize { |
| 1052 | // Filling in bits for an array of typ. |
| 1053 | // Set up for repetition of ptrmask during main loop. |
| 1054 | // Note that ptrmask describes only a prefix of |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1055 | const maxBits = sys.PtrSize*8 - 7 |
| 1056 | if typ.ptrdata/sys.PtrSize <= maxBits { |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1057 | // Entire ptrmask fits in uintptr with room for a byte fragment. |
| 1058 | // Load into pbits and never read from ptrmask again. |
| 1059 | // This is especially important when the ptrmask has |
| 1060 | // fewer than 8 bits in it; otherwise the reload in the middle |
| 1061 | // of the Phase 2 loop would itself need to loop to gather |
| 1062 | // at least 8 bits. |
| 1063 | |
| 1064 | // Accumulate ptrmask into b. |
| 1065 | // ptrmask is sized to describe only typ.ptrdata, but we record |
| 1066 | // it as describing typ.size bytes, since all the high bits are zero. |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1067 | nb = typ.ptrdata / sys.PtrSize |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1068 | for i := uintptr(0); i < nb; i += 8 { |
| 1069 | b |= uintptr(*p) << i |
| 1070 | p = add1(p) |
| 1071 | } |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1072 | nb = typ.size / sys.PtrSize |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1073 | |
| 1074 | // Replicate ptrmask to fill entire pbits uintptr. |
| 1075 | // Doubling and truncating is fewer steps than |
| 1076 | // iterating by nb each time. (nb could be 1.) |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1077 | // Since we loaded typ.ptrdata/sys.PtrSize bits |
| 1078 | // but are pretending to have typ.size/sys.PtrSize, |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1079 | // there might be no replication necessary/possible. |
| 1080 | pbits = b |
| 1081 | endnb = nb |
| 1082 | if nb+nb <= maxBits { |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1083 | for endnb <= sys.PtrSize*8 { |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1084 | pbits |= pbits << endnb |
| 1085 | endnb += endnb |
| 1086 | } |
| 1087 | // Truncate to a multiple of original ptrmask. |
| 1088 | endnb = maxBits / nb * nb |
| 1089 | pbits &= 1<<endnb - 1 |
| 1090 | b = pbits |
| 1091 | nb = endnb |
| 1092 | } |
| 1093 | |
| 1094 | // Clear p and endp as sentinel for using pbits. |
| 1095 | // Checked during Phase 2 loop. |
| 1096 | p = nil |
| 1097 | endp = nil |
| 1098 | } else { |
| 1099 | // Ptrmask is larger. Read it multiple times. |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1100 | n := (typ.ptrdata/sys.PtrSize+7)/8 - 1 |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1101 | endp = addb(ptrmask, n) |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1102 | endnb = typ.size/sys.PtrSize - n*8 |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1103 | } |
| 1104 | } |
| 1105 | if p != nil { |
| 1106 | b = uintptr(*p) |
| 1107 | p = add1(p) |
| 1108 | nb = 8 |
| 1109 | } |
| 1110 | |
| 1111 | if typ.size == dataSize { |
| 1112 | // Single entry: can stop once we reach the non-pointer data. |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1113 | nw = typ.ptrdata / sys.PtrSize |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1114 | } else { |
| 1115 | // Repeated instances of typ in an array. |
| 1116 | // Have to process first N-1 entries in full, but can stop |
| 1117 | // once we reach the non-pointer data in the final entry. |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1118 | nw = ((dataSize/typ.size-1)*typ.size + typ.ptrdata) / sys.PtrSize |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1119 | } |
| 1120 | if nw == 0 { |
| 1121 | // No pointers! Caller was supposed to check. |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1122 | println("runtime: invalid type ", typ.string()) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1123 | throw("heapBitsSetType: called with non-pointer type") |
| 1124 | return |
| 1125 | } |
| 1126 | if nw < 2 { |
| 1127 | // Must write at least 2 words, because the "no scan" |
| 1128 | // encoding doesn't take effect until the third word. |
| 1129 | nw = 2 |
| 1130 | } |
| 1131 | |
| 1132 | // Phase 1: Special case for leading byte (shift==0) or half-byte (shift==4). |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1133 | // The leading byte is special because it contains the bits for word 1, |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1134 | // which does not have the scan bit set. |
| 1135 | // The leading half-byte is special because it's a half a byte, |
| 1136 | // so we have to be careful with the bits already there. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1137 | switch { |
| 1138 | default: |
| 1139 | throw("heapBitsSetType: unexpected shift") |
| 1140 | |
| 1141 | case h.shift == 0: |
| 1142 | // Ptrmask and heap bitmap are aligned. |
| 1143 | // Handle first byte of bitmap specially. |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1144 | // |
| 1145 | // The first byte we write out covers the first four |
| 1146 | // words of the object. The scan/dead bit on the first |
| 1147 | // word must be set to scan since there are pointers |
| 1148 | // somewhere in the object. The scan/dead bit on the |
| 1149 | // second word is the checkmark, so we don't set it. |
| 1150 | // In all following words, we set the scan/dead |
| 1151 | // appropriately to indicate that the object contains |
| 1152 | // to the next 2-bit entry in the bitmap. |
| 1153 | // |
| 1154 | // TODO: It doesn't matter if we set the checkmark, so |
| 1155 | // maybe this case isn't needed any more. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1156 | hb = b & bitPointerAll |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1157 | hb |= bitScan | bitScan<<(2*heapBitsShift) | bitScan<<(3*heapBitsShift) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1158 | if w += 4; w >= nw { |
| 1159 | goto Phase3 |
| 1160 | } |
| 1161 | *hbitp = uint8(hb) |
| 1162 | hbitp = subtract1(hbitp) |
| 1163 | b >>= 4 |
| 1164 | nb -= 4 |
| 1165 | |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1166 | case sys.PtrSize == 8 && h.shift == 2: |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1167 | // Ptrmask and heap bitmap are misaligned. |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1168 | // The bits for the first two words are in a byte shared |
| 1169 | // with another object, so we must be careful with the bits |
| 1170 | // already there. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1171 | // We took care of 1-word and 2-word objects above, |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1172 | // so this is at least a 6-word object. |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1173 | hb = (b & (bitPointer | bitPointer<<heapBitsShift)) << (2 * heapBitsShift) |
| 1174 | // This is not noscan, so set the scan bit in the |
| 1175 | // first word. |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1176 | hb |= bitScan << (2 * heapBitsShift) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1177 | b >>= 2 |
| 1178 | nb -= 2 |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1179 | // Note: no bitScan for second word because that's |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1180 | // the checkmark. |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1181 | *hbitp &^= uint8((bitPointer | bitScan | (bitPointer << heapBitsShift)) << (2 * heapBitsShift)) |
| 1182 | *hbitp |= uint8(hb) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1183 | hbitp = subtract1(hbitp) |
| 1184 | if w += 2; w >= nw { |
| 1185 | // We know that there is more data, because we handled 2-word objects above. |
| 1186 | // This must be at least a 6-word object. If we're out of pointer words, |
| 1187 | // mark no scan in next bitmap byte and finish. |
| 1188 | hb = 0 |
| 1189 | w += 4 |
| 1190 | goto Phase3 |
| 1191 | } |
| 1192 | } |
| 1193 | |
| 1194 | // Phase 2: Full bytes in bitmap, up to but not including write to last byte (full or partial) in bitmap. |
| 1195 | // The loop computes the bits for that last write but does not execute the write; |
| 1196 | // it leaves the bits in hb for processing by phase 3. |
| 1197 | // To avoid repeated adjustment of nb, we subtract out the 4 bits we're going to |
| 1198 | // use in the first half of the loop right now, and then we only adjust nb explicitly |
| 1199 | // if the 8 bits used by each iteration isn't balanced by 8 bits loaded mid-loop. |
| 1200 | nb -= 4 |
| 1201 | for { |
| 1202 | // Emit bitmap byte. |
| 1203 | // b has at least nb+4 bits, with one exception: |
| 1204 | // if w+4 >= nw, then b has only nw-w bits, |
| 1205 | // but we'll stop at the break and then truncate |
| 1206 | // appropriately in Phase 3. |
| 1207 | hb = b & bitPointerAll |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1208 | hb |= bitScanAll |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1209 | if w += 4; w >= nw { |
| 1210 | break |
| 1211 | } |
| 1212 | *hbitp = uint8(hb) |
| 1213 | hbitp = subtract1(hbitp) |
| 1214 | b >>= 4 |
| 1215 | |
| 1216 | // Load more bits. b has nb right now. |
| 1217 | if p != endp { |
| 1218 | // Fast path: keep reading from ptrmask. |
| 1219 | // nb unmodified: we just loaded 8 bits, |
| 1220 | // and the next iteration will consume 8 bits, |
| 1221 | // leaving us with the same nb the next time we're here. |
| 1222 | if nb < 8 { |
| 1223 | b |= uintptr(*p) << nb |
| 1224 | p = add1(p) |
| 1225 | } else { |
| 1226 | // Reduce the number of bits in b. |
| 1227 | // This is important if we skipped |
| 1228 | // over a scalar tail, since nb could |
| 1229 | // be larger than the bit width of b. |
| 1230 | nb -= 8 |
| 1231 | } |
| 1232 | } else if p == nil { |
| 1233 | // Almost as fast path: track bit count and refill from pbits. |
| 1234 | // For short repetitions. |
| 1235 | if nb < 8 { |
| 1236 | b |= pbits << nb |
| 1237 | nb += endnb |
| 1238 | } |
| 1239 | nb -= 8 // for next iteration |
| 1240 | } else { |
| 1241 | // Slow path: reached end of ptrmask. |
| 1242 | // Process final partial byte and rewind to start. |
| 1243 | b |= uintptr(*p) << nb |
| 1244 | nb += endnb |
| 1245 | if nb < 8 { |
| 1246 | b |= uintptr(*ptrmask) << nb |
| 1247 | p = add1(ptrmask) |
| 1248 | } else { |
| 1249 | nb -= 8 |
| 1250 | p = ptrmask |
| 1251 | } |
| 1252 | } |
| 1253 | |
| 1254 | // Emit bitmap byte. |
| 1255 | hb = b & bitPointerAll |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1256 | hb |= bitScanAll |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1257 | if w += 4; w >= nw { |
| 1258 | break |
| 1259 | } |
| 1260 | *hbitp = uint8(hb) |
| 1261 | hbitp = subtract1(hbitp) |
| 1262 | b >>= 4 |
| 1263 | } |
| 1264 | |
| 1265 | Phase3: |
| 1266 | // Phase 3: Write last byte or partial byte and zero the rest of the bitmap entries. |
| 1267 | if w > nw { |
| 1268 | // Counting the 4 entries in hb not yet written to memory, |
| 1269 | // there are more entries than possible pointer slots. |
| 1270 | // Discard the excess entries (can't be more than 3). |
| 1271 | mask := uintptr(1)<<(4-(w-nw)) - 1 |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1272 | hb &= mask | mask<<4 // apply mask to both pointer bits and scan bits |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1273 | } |
| 1274 | |
| 1275 | // Change nw from counting possibly-pointer words to total words in allocation. |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1276 | nw = size / sys.PtrSize |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1277 | |
| 1278 | // Write whole bitmap bytes. |
| 1279 | // The first is hb, the rest are zero. |
| 1280 | if w <= nw { |
| 1281 | *hbitp = uint8(hb) |
| 1282 | hbitp = subtract1(hbitp) |
| 1283 | hb = 0 // for possible final half-byte below |
| 1284 | for w += 4; w <= nw; w += 4 { |
| 1285 | *hbitp = 0 |
| 1286 | hbitp = subtract1(hbitp) |
| 1287 | } |
| 1288 | } |
| 1289 | |
| 1290 | // Write final partial bitmap byte if any. |
| 1291 | // We know w > nw, or else we'd still be in the loop above. |
| 1292 | // It can be bigger only due to the 4 entries in hb that it counts. |
| 1293 | // If w == nw+4 then there's nothing left to do: we wrote all nw entries |
| 1294 | // and can discard the 4 sitting in hb. |
| 1295 | // But if w == nw+2, we need to write first two in hb. |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1296 | // The byte is shared with the next object, so be careful with |
| 1297 | // existing bits. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1298 | if w == nw+2 { |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1299 | *hbitp = *hbitp&^(bitPointer|bitScan|(bitPointer|bitScan)<<heapBitsShift) | uint8(hb) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1300 | } |
| 1301 | |
| 1302 | Phase4: |
| 1303 | // Phase 4: all done, but perhaps double check. |
| 1304 | if doubleCheck { |
| 1305 | end := heapBitsForAddr(x + size) |
| 1306 | if typ.kind&kindGCProg == 0 && (hbitp != end.bitp || (w == nw+2) != (end.shift == 2)) { |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1307 | println("ended at wrong bitmap byte for", typ.string(), "x", dataSize/typ.size) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1308 | print("typ.size=", typ.size, " typ.ptrdata=", typ.ptrdata, " dataSize=", dataSize, " size=", size, "\n") |
| 1309 | print("w=", w, " nw=", nw, " b=", hex(b), " nb=", nb, " hb=", hex(hb), "\n") |
| 1310 | h0 := heapBitsForAddr(x) |
| 1311 | print("initial bits h0.bitp=", h0.bitp, " h0.shift=", h0.shift, "\n") |
| 1312 | print("ended at hbitp=", hbitp, " but next starts at bitp=", end.bitp, " shift=", end.shift, "\n") |
| 1313 | throw("bad heapBitsSetType") |
| 1314 | } |
| 1315 | |
| 1316 | // Double-check that bits to be written were written correctly. |
| 1317 | // Does not check that other bits were not written, unfortunately. |
| 1318 | h := heapBitsForAddr(x) |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1319 | nptr := typ.ptrdata / sys.PtrSize |
| 1320 | ndata := typ.size / sys.PtrSize |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1321 | count := dataSize / typ.size |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1322 | totalptr := ((count-1)*typ.size + typ.ptrdata) / sys.PtrSize |
| 1323 | for i := uintptr(0); i < size/sys.PtrSize; i++ { |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1324 | j := i % ndata |
| 1325 | var have, want uint8 |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1326 | have = (*h.bitp >> h.shift) & (bitPointer | bitScan) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1327 | if i >= totalptr { |
| 1328 | want = 0 // deadmarker |
| 1329 | if typ.kind&kindGCProg != 0 && i < (totalptr+3)/4*4 { |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1330 | want = bitScan |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1331 | } |
| 1332 | } else { |
| 1333 | if j < nptr && (*addb(ptrmask, j/8)>>(j%8))&1 != 0 { |
| 1334 | want |= bitPointer |
| 1335 | } |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1336 | if i != 1 { |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1337 | want |= bitScan |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1338 | } else { |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1339 | have &^= bitScan |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1340 | } |
| 1341 | } |
| 1342 | if have != want { |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1343 | println("mismatch writing bits for", typ.string(), "x", dataSize/typ.size) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1344 | print("typ.size=", typ.size, " typ.ptrdata=", typ.ptrdata, " dataSize=", dataSize, " size=", size, "\n") |
| 1345 | print("kindGCProg=", typ.kind&kindGCProg != 0, "\n") |
| 1346 | print("w=", w, " nw=", nw, " b=", hex(b), " nb=", nb, " hb=", hex(hb), "\n") |
| 1347 | h0 := heapBitsForAddr(x) |
| 1348 | print("initial bits h0.bitp=", h0.bitp, " h0.shift=", h0.shift, "\n") |
| 1349 | print("current bits h.bitp=", h.bitp, " h.shift=", h.shift, " *h.bitp=", hex(*h.bitp), "\n") |
| 1350 | print("ptrmask=", ptrmask, " p=", p, " endp=", endp, " endnb=", endnb, " pbits=", hex(pbits), " b=", hex(b), " nb=", nb, "\n") |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1351 | println("at word", i, "offset", i*sys.PtrSize, "have", have, "want", want) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1352 | if typ.kind&kindGCProg != 0 { |
| 1353 | println("GC program:") |
| 1354 | dumpGCProg(addb(typ.gcdata, 4)) |
| 1355 | } |
| 1356 | throw("bad heapBitsSetType") |
| 1357 | } |
| 1358 | h = h.next() |
| 1359 | } |
| 1360 | if ptrmask == debugPtrmask.data { |
| 1361 | unlock(&debugPtrmask.lock) |
| 1362 | } |
| 1363 | } |
| 1364 | } |
| 1365 | |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1366 | // heapBitsSetTypeNoScan marks x as noscan by setting the first word |
| 1367 | // of x in the heap bitmap to scalar/dead. |
| 1368 | func heapBitsSetTypeNoScan(x uintptr) { |
| 1369 | h := heapBitsForAddr(uintptr(x)) |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1370 | *h.bitp &^= (bitPointer | bitScan) << h.shift |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1371 | } |
| 1372 | |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1373 | var debugPtrmask struct { |
| 1374 | lock mutex |
| 1375 | data *byte |
| 1376 | } |
| 1377 | |
| 1378 | // heapBitsSetTypeGCProg implements heapBitsSetType using a GC program. |
| 1379 | // progSize is the size of the memory described by the program. |
| 1380 | // elemSize is the size of the element that the GC program describes (a prefix of). |
| 1381 | // dataSize is the total size of the intended data, a multiple of elemSize. |
| 1382 | // allocSize is the total size of the allocated memory. |
| 1383 | // |
| 1384 | // GC programs are only used for large allocations. |
| 1385 | // heapBitsSetType requires that allocSize is a multiple of 4 words, |
| 1386 | // so that the relevant bitmap bytes are not shared with surrounding |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1387 | // objects. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1388 | func heapBitsSetTypeGCProg(h heapBits, progSize, elemSize, dataSize, allocSize uintptr, prog *byte) { |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1389 | if sys.PtrSize == 8 && allocSize%(4*sys.PtrSize) != 0 { |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1390 | // Alignment will be wrong. |
| 1391 | throw("heapBitsSetTypeGCProg: small allocation") |
| 1392 | } |
| 1393 | var totalBits uintptr |
| 1394 | if elemSize == dataSize { |
| 1395 | totalBits = runGCProg(prog, nil, h.bitp, 2) |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1396 | if totalBits*sys.PtrSize != progSize { |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1397 | println("runtime: heapBitsSetTypeGCProg: total bits", totalBits, "but progSize", progSize) |
| 1398 | throw("heapBitsSetTypeGCProg: unexpected bit count") |
| 1399 | } |
| 1400 | } else { |
| 1401 | count := dataSize / elemSize |
| 1402 | |
| 1403 | // Piece together program trailer to run after prog that does: |
| 1404 | // literal(0) |
| 1405 | // repeat(1, elemSize-progSize-1) // zeros to fill element size |
| 1406 | // repeat(elemSize, count-1) // repeat that element for count |
| 1407 | // This zero-pads the data remaining in the first element and then |
| 1408 | // repeats that first element to fill the array. |
| 1409 | var trailer [40]byte // 3 varints (max 10 each) + some bytes |
| 1410 | i := 0 |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1411 | if n := elemSize/sys.PtrSize - progSize/sys.PtrSize; n > 0 { |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1412 | // literal(0) |
| 1413 | trailer[i] = 0x01 |
| 1414 | i++ |
| 1415 | trailer[i] = 0 |
| 1416 | i++ |
| 1417 | if n > 1 { |
| 1418 | // repeat(1, n-1) |
| 1419 | trailer[i] = 0x81 |
| 1420 | i++ |
| 1421 | n-- |
| 1422 | for ; n >= 0x80; n >>= 7 { |
| 1423 | trailer[i] = byte(n | 0x80) |
| 1424 | i++ |
| 1425 | } |
| 1426 | trailer[i] = byte(n) |
| 1427 | i++ |
| 1428 | } |
| 1429 | } |
| 1430 | // repeat(elemSize/ptrSize, count-1) |
| 1431 | trailer[i] = 0x80 |
| 1432 | i++ |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1433 | n := elemSize / sys.PtrSize |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1434 | for ; n >= 0x80; n >>= 7 { |
| 1435 | trailer[i] = byte(n | 0x80) |
| 1436 | i++ |
| 1437 | } |
| 1438 | trailer[i] = byte(n) |
| 1439 | i++ |
| 1440 | n = count - 1 |
| 1441 | for ; n >= 0x80; n >>= 7 { |
| 1442 | trailer[i] = byte(n | 0x80) |
| 1443 | i++ |
| 1444 | } |
| 1445 | trailer[i] = byte(n) |
| 1446 | i++ |
| 1447 | trailer[i] = 0 |
| 1448 | i++ |
| 1449 | |
| 1450 | runGCProg(prog, &trailer[0], h.bitp, 2) |
| 1451 | |
| 1452 | // Even though we filled in the full array just now, |
| 1453 | // record that we only filled in up to the ptrdata of the |
| 1454 | // last element. This will cause the code below to |
| 1455 | // memclr the dead section of the final array element, |
| 1456 | // so that scanobject can stop early in the final element. |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1457 | totalBits = (elemSize*(count-1) + progSize) / sys.PtrSize |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1458 | } |
| 1459 | endProg := unsafe.Pointer(subtractb(h.bitp, (totalBits+3)/4)) |
| 1460 | endAlloc := unsafe.Pointer(subtractb(h.bitp, allocSize/heapBitmapScale)) |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1461 | memclrNoHeapPointers(add(endAlloc, 1), uintptr(endProg)-uintptr(endAlloc)) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1462 | } |
| 1463 | |
| 1464 | // progToPointerMask returns the 1-bit pointer mask output by the GC program prog. |
| 1465 | // size the size of the region described by prog, in bytes. |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1466 | // The resulting bitvector will have no more than size/sys.PtrSize bits. |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1467 | func progToPointerMask(prog *byte, size uintptr) bitvector { |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1468 | n := (size/sys.PtrSize + 7) / 8 |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1469 | x := (*[1 << 30]byte)(persistentalloc(n+1, 1, &memstats.buckhash_sys))[:n+1] |
| 1470 | x[len(x)-1] = 0xa1 // overflow check sentinel |
| 1471 | n = runGCProg(prog, nil, &x[0], 1) |
| 1472 | if x[len(x)-1] != 0xa1 { |
| 1473 | throw("progToPointerMask: overflow") |
| 1474 | } |
| 1475 | return bitvector{int32(n), &x[0]} |
| 1476 | } |
| 1477 | |
| 1478 | // Packed GC pointer bitmaps, aka GC programs. |
| 1479 | // |
| 1480 | // For large types containing arrays, the type information has a |
| 1481 | // natural repetition that can be encoded to save space in the |
| 1482 | // binary and in the memory representation of the type information. |
| 1483 | // |
| 1484 | // The encoding is a simple Lempel-Ziv style bytecode machine |
| 1485 | // with the following instructions: |
| 1486 | // |
| 1487 | // 00000000: stop |
| 1488 | // 0nnnnnnn: emit n bits copied from the next (n+7)/8 bytes |
| 1489 | // 10000000 n c: repeat the previous n bits c times; n, c are varints |
| 1490 | // 1nnnnnnn c: repeat the previous n bits c times; c is a varint |
| 1491 | |
| 1492 | // runGCProg executes the GC program prog, and then trailer if non-nil, |
| 1493 | // writing to dst with entries of the given size. |
| 1494 | // If size == 1, dst is a 1-bit pointer mask laid out moving forward from dst. |
| 1495 | // If size == 2, dst is the 2-bit heap bitmap, and writes move backward |
| 1496 | // starting at dst (because the heap bitmap does). In this case, the caller guarantees |
| 1497 | // that only whole bytes in dst need to be written. |
| 1498 | // |
| 1499 | // runGCProg returns the number of 1- or 2-bit entries written to memory. |
| 1500 | func runGCProg(prog, trailer, dst *byte, size int) uintptr { |
| 1501 | dstStart := dst |
| 1502 | |
| 1503 | // Bits waiting to be written to memory. |
| 1504 | var bits uintptr |
| 1505 | var nbits uintptr |
| 1506 | |
| 1507 | p := prog |
| 1508 | Run: |
| 1509 | for { |
| 1510 | // Flush accumulated full bytes. |
| 1511 | // The rest of the loop assumes that nbits <= 7. |
| 1512 | for ; nbits >= 8; nbits -= 8 { |
| 1513 | if size == 1 { |
| 1514 | *dst = uint8(bits) |
| 1515 | dst = add1(dst) |
| 1516 | bits >>= 8 |
| 1517 | } else { |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1518 | v := bits&bitPointerAll | bitScanAll |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1519 | *dst = uint8(v) |
| 1520 | dst = subtract1(dst) |
| 1521 | bits >>= 4 |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1522 | v = bits&bitPointerAll | bitScanAll |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1523 | *dst = uint8(v) |
| 1524 | dst = subtract1(dst) |
| 1525 | bits >>= 4 |
| 1526 | } |
| 1527 | } |
| 1528 | |
| 1529 | // Process one instruction. |
| 1530 | inst := uintptr(*p) |
| 1531 | p = add1(p) |
| 1532 | n := inst & 0x7F |
| 1533 | if inst&0x80 == 0 { |
| 1534 | // Literal bits; n == 0 means end of program. |
| 1535 | if n == 0 { |
| 1536 | // Program is over; continue in trailer if present. |
| 1537 | if trailer != nil { |
| 1538 | //println("trailer") |
| 1539 | p = trailer |
| 1540 | trailer = nil |
| 1541 | continue |
| 1542 | } |
| 1543 | //println("done") |
| 1544 | break Run |
| 1545 | } |
| 1546 | //println("lit", n, dst) |
| 1547 | nbyte := n / 8 |
| 1548 | for i := uintptr(0); i < nbyte; i++ { |
| 1549 | bits |= uintptr(*p) << nbits |
| 1550 | p = add1(p) |
| 1551 | if size == 1 { |
| 1552 | *dst = uint8(bits) |
| 1553 | dst = add1(dst) |
| 1554 | bits >>= 8 |
| 1555 | } else { |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1556 | v := bits&0xf | bitScanAll |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1557 | *dst = uint8(v) |
| 1558 | dst = subtract1(dst) |
| 1559 | bits >>= 4 |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1560 | v = bits&0xf | bitScanAll |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1561 | *dst = uint8(v) |
| 1562 | dst = subtract1(dst) |
| 1563 | bits >>= 4 |
| 1564 | } |
| 1565 | } |
| 1566 | if n %= 8; n > 0 { |
| 1567 | bits |= uintptr(*p) << nbits |
| 1568 | p = add1(p) |
| 1569 | nbits += n |
| 1570 | } |
| 1571 | continue Run |
| 1572 | } |
| 1573 | |
| 1574 | // Repeat. If n == 0, it is encoded in a varint in the next bytes. |
| 1575 | if n == 0 { |
| 1576 | for off := uint(0); ; off += 7 { |
| 1577 | x := uintptr(*p) |
| 1578 | p = add1(p) |
| 1579 | n |= (x & 0x7F) << off |
| 1580 | if x&0x80 == 0 { |
| 1581 | break |
| 1582 | } |
| 1583 | } |
| 1584 | } |
| 1585 | |
| 1586 | // Count is encoded in a varint in the next bytes. |
| 1587 | c := uintptr(0) |
| 1588 | for off := uint(0); ; off += 7 { |
| 1589 | x := uintptr(*p) |
| 1590 | p = add1(p) |
| 1591 | c |= (x & 0x7F) << off |
| 1592 | if x&0x80 == 0 { |
| 1593 | break |
| 1594 | } |
| 1595 | } |
| 1596 | c *= n // now total number of bits to copy |
| 1597 | |
| 1598 | // If the number of bits being repeated is small, load them |
| 1599 | // into a register and use that register for the entire loop |
| 1600 | // instead of repeatedly reading from memory. |
| 1601 | // Handling fewer than 8 bits here makes the general loop simpler. |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1602 | // The cutoff is sys.PtrSize*8 - 7 to guarantee that when we add |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1603 | // the pattern to a bit buffer holding at most 7 bits (a partial byte) |
| 1604 | // it will not overflow. |
| 1605 | src := dst |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1606 | const maxBits = sys.PtrSize*8 - 7 |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1607 | if n <= maxBits { |
| 1608 | // Start with bits in output buffer. |
| 1609 | pattern := bits |
| 1610 | npattern := nbits |
| 1611 | |
| 1612 | // If we need more bits, fetch them from memory. |
| 1613 | if size == 1 { |
| 1614 | src = subtract1(src) |
| 1615 | for npattern < n { |
| 1616 | pattern <<= 8 |
| 1617 | pattern |= uintptr(*src) |
| 1618 | src = subtract1(src) |
| 1619 | npattern += 8 |
| 1620 | } |
| 1621 | } else { |
| 1622 | src = add1(src) |
| 1623 | for npattern < n { |
| 1624 | pattern <<= 4 |
| 1625 | pattern |= uintptr(*src) & 0xf |
| 1626 | src = add1(src) |
| 1627 | npattern += 4 |
| 1628 | } |
| 1629 | } |
| 1630 | |
| 1631 | // We started with the whole bit output buffer, |
| 1632 | // and then we loaded bits from whole bytes. |
| 1633 | // Either way, we might now have too many instead of too few. |
| 1634 | // Discard the extra. |
| 1635 | if npattern > n { |
| 1636 | pattern >>= npattern - n |
| 1637 | npattern = n |
| 1638 | } |
| 1639 | |
| 1640 | // Replicate pattern to at most maxBits. |
| 1641 | if npattern == 1 { |
| 1642 | // One bit being repeated. |
| 1643 | // If the bit is 1, make the pattern all 1s. |
| 1644 | // If the bit is 0, the pattern is already all 0s, |
| 1645 | // but we can claim that the number of bits |
| 1646 | // in the word is equal to the number we need (c), |
| 1647 | // because right shift of bits will zero fill. |
| 1648 | if pattern == 1 { |
| 1649 | pattern = 1<<maxBits - 1 |
| 1650 | npattern = maxBits |
| 1651 | } else { |
| 1652 | npattern = c |
| 1653 | } |
| 1654 | } else { |
| 1655 | b := pattern |
| 1656 | nb := npattern |
| 1657 | if nb+nb <= maxBits { |
| 1658 | // Double pattern until the whole uintptr is filled. |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1659 | for nb <= sys.PtrSize*8 { |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1660 | b |= b << nb |
| 1661 | nb += nb |
| 1662 | } |
| 1663 | // Trim away incomplete copy of original pattern in high bits. |
| 1664 | // TODO(rsc): Replace with table lookup or loop on systems without divide? |
| 1665 | nb = maxBits / npattern * npattern |
| 1666 | b &= 1<<nb - 1 |
| 1667 | pattern = b |
| 1668 | npattern = nb |
| 1669 | } |
| 1670 | } |
| 1671 | |
| 1672 | // Add pattern to bit buffer and flush bit buffer, c/npattern times. |
| 1673 | // Since pattern contains >8 bits, there will be full bytes to flush |
| 1674 | // on each iteration. |
| 1675 | for ; c >= npattern; c -= npattern { |
| 1676 | bits |= pattern << nbits |
| 1677 | nbits += npattern |
| 1678 | if size == 1 { |
| 1679 | for nbits >= 8 { |
| 1680 | *dst = uint8(bits) |
| 1681 | dst = add1(dst) |
| 1682 | bits >>= 8 |
| 1683 | nbits -= 8 |
| 1684 | } |
| 1685 | } else { |
| 1686 | for nbits >= 4 { |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1687 | *dst = uint8(bits&0xf | bitScanAll) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1688 | dst = subtract1(dst) |
| 1689 | bits >>= 4 |
| 1690 | nbits -= 4 |
| 1691 | } |
| 1692 | } |
| 1693 | } |
| 1694 | |
| 1695 | // Add final fragment to bit buffer. |
| 1696 | if c > 0 { |
| 1697 | pattern &= 1<<c - 1 |
| 1698 | bits |= pattern << nbits |
| 1699 | nbits += c |
| 1700 | } |
| 1701 | continue Run |
| 1702 | } |
| 1703 | |
| 1704 | // Repeat; n too large to fit in a register. |
| 1705 | // Since nbits <= 7, we know the first few bytes of repeated data |
| 1706 | // are already written to memory. |
| 1707 | off := n - nbits // n > nbits because n > maxBits and nbits <= 7 |
| 1708 | if size == 1 { |
| 1709 | // Leading src fragment. |
| 1710 | src = subtractb(src, (off+7)/8) |
| 1711 | if frag := off & 7; frag != 0 { |
| 1712 | bits |= uintptr(*src) >> (8 - frag) << nbits |
| 1713 | src = add1(src) |
| 1714 | nbits += frag |
| 1715 | c -= frag |
| 1716 | } |
| 1717 | // Main loop: load one byte, write another. |
| 1718 | // The bits are rotating through the bit buffer. |
| 1719 | for i := c / 8; i > 0; i-- { |
| 1720 | bits |= uintptr(*src) << nbits |
| 1721 | src = add1(src) |
| 1722 | *dst = uint8(bits) |
| 1723 | dst = add1(dst) |
| 1724 | bits >>= 8 |
| 1725 | } |
| 1726 | // Final src fragment. |
| 1727 | if c %= 8; c > 0 { |
| 1728 | bits |= (uintptr(*src) & (1<<c - 1)) << nbits |
| 1729 | nbits += c |
| 1730 | } |
| 1731 | } else { |
| 1732 | // Leading src fragment. |
| 1733 | src = addb(src, (off+3)/4) |
| 1734 | if frag := off & 3; frag != 0 { |
| 1735 | bits |= (uintptr(*src) & 0xf) >> (4 - frag) << nbits |
| 1736 | src = subtract1(src) |
| 1737 | nbits += frag |
| 1738 | c -= frag |
| 1739 | } |
| 1740 | // Main loop: load one byte, write another. |
| 1741 | // The bits are rotating through the bit buffer. |
| 1742 | for i := c / 4; i > 0; i-- { |
| 1743 | bits |= (uintptr(*src) & 0xf) << nbits |
| 1744 | src = subtract1(src) |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1745 | *dst = uint8(bits&0xf | bitScanAll) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1746 | dst = subtract1(dst) |
| 1747 | bits >>= 4 |
| 1748 | } |
| 1749 | // Final src fragment. |
| 1750 | if c %= 4; c > 0 { |
| 1751 | bits |= (uintptr(*src) & (1<<c - 1)) << nbits |
| 1752 | nbits += c |
| 1753 | } |
| 1754 | } |
| 1755 | } |
| 1756 | |
| 1757 | // Write any final bits out, using full-byte writes, even for the final byte. |
| 1758 | var totalBits uintptr |
| 1759 | if size == 1 { |
| 1760 | totalBits = (uintptr(unsafe.Pointer(dst))-uintptr(unsafe.Pointer(dstStart)))*8 + nbits |
| 1761 | nbits += -nbits & 7 |
| 1762 | for ; nbits > 0; nbits -= 8 { |
| 1763 | *dst = uint8(bits) |
| 1764 | dst = add1(dst) |
| 1765 | bits >>= 8 |
| 1766 | } |
| 1767 | } else { |
| 1768 | totalBits = (uintptr(unsafe.Pointer(dstStart))-uintptr(unsafe.Pointer(dst)))*4 + nbits |
| 1769 | nbits += -nbits & 3 |
| 1770 | for ; nbits > 0; nbits -= 4 { |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1771 | v := bits&0xf | bitScanAll |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1772 | *dst = uint8(v) |
| 1773 | dst = subtract1(dst) |
| 1774 | bits >>= 4 |
| 1775 | } |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1776 | } |
| 1777 | return totalBits |
| 1778 | } |
| 1779 | |
| 1780 | func dumpGCProg(p *byte) { |
| 1781 | nptr := 0 |
| 1782 | for { |
| 1783 | x := *p |
| 1784 | p = add1(p) |
| 1785 | if x == 0 { |
| 1786 | print("\t", nptr, " end\n") |
| 1787 | break |
| 1788 | } |
| 1789 | if x&0x80 == 0 { |
| 1790 | print("\t", nptr, " lit ", x, ":") |
| 1791 | n := int(x+7) / 8 |
| 1792 | for i := 0; i < n; i++ { |
| 1793 | print(" ", hex(*p)) |
| 1794 | p = add1(p) |
| 1795 | } |
| 1796 | print("\n") |
| 1797 | nptr += int(x) |
| 1798 | } else { |
| 1799 | nbit := int(x &^ 0x80) |
| 1800 | if nbit == 0 { |
| 1801 | for nb := uint(0); ; nb += 7 { |
| 1802 | x := *p |
| 1803 | p = add1(p) |
| 1804 | nbit |= int(x&0x7f) << nb |
| 1805 | if x&0x80 == 0 { |
| 1806 | break |
| 1807 | } |
| 1808 | } |
| 1809 | } |
| 1810 | count := 0 |
| 1811 | for nb := uint(0); ; nb += 7 { |
| 1812 | x := *p |
| 1813 | p = add1(p) |
| 1814 | count |= int(x&0x7f) << nb |
| 1815 | if x&0x80 == 0 { |
| 1816 | break |
| 1817 | } |
| 1818 | } |
| 1819 | print("\t", nptr, " repeat ", nbit, " × ", count, "\n") |
| 1820 | nptr += nbit * count |
| 1821 | } |
| 1822 | } |
| 1823 | } |
| 1824 | |
| 1825 | // Testing. |
| 1826 | |
| 1827 | func getgcmaskcb(frame *stkframe, ctxt unsafe.Pointer) bool { |
| 1828 | target := (*stkframe)(ctxt) |
| 1829 | if frame.sp <= target.sp && target.sp < frame.varp { |
| 1830 | *target = *frame |
| 1831 | return false |
| 1832 | } |
| 1833 | return true |
| 1834 | } |
| 1835 | |
| 1836 | // gcbits returns the GC type info for x, for testing. |
| 1837 | // The result is the bitmap entries (0 or 1), one entry per byte. |
| 1838 | //go:linkname reflect_gcbits reflect.gcbits |
| 1839 | func reflect_gcbits(x interface{}) []byte { |
| 1840 | ret := getgcmask(x) |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1841 | typ := (*ptrtype)(unsafe.Pointer(efaceOf(&x)._type)).elem |
| 1842 | nptr := typ.ptrdata / sys.PtrSize |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1843 | for uintptr(len(ret)) > nptr && ret[len(ret)-1] == 0 { |
| 1844 | ret = ret[:len(ret)-1] |
| 1845 | } |
| 1846 | return ret |
| 1847 | } |
| 1848 | |
| 1849 | // Returns GC type info for object p for testing. |
| 1850 | func getgcmask(ep interface{}) (mask []byte) { |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1851 | e := *efaceOf(&ep) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1852 | p := e.data |
| 1853 | t := e._type |
| 1854 | // data or bss |
Dan Willemsen | ebae302 | 2017-01-13 23:01:08 -0800 | [diff] [blame] | 1855 | for _, datap := range activeModules() { |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1856 | // data |
| 1857 | if datap.data <= uintptr(p) && uintptr(p) < datap.edata { |
| 1858 | bitmap := datap.gcdatamask.bytedata |
| 1859 | n := (*ptrtype)(unsafe.Pointer(t)).elem.size |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1860 | mask = make([]byte, n/sys.PtrSize) |
| 1861 | for i := uintptr(0); i < n; i += sys.PtrSize { |
| 1862 | off := (uintptr(p) + i - datap.data) / sys.PtrSize |
| 1863 | mask[i/sys.PtrSize] = (*addb(bitmap, off/8) >> (off % 8)) & 1 |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1864 | } |
| 1865 | return |
| 1866 | } |
| 1867 | |
| 1868 | // bss |
| 1869 | if datap.bss <= uintptr(p) && uintptr(p) < datap.ebss { |
| 1870 | bitmap := datap.gcbssmask.bytedata |
| 1871 | n := (*ptrtype)(unsafe.Pointer(t)).elem.size |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1872 | mask = make([]byte, n/sys.PtrSize) |
| 1873 | for i := uintptr(0); i < n; i += sys.PtrSize { |
| 1874 | off := (uintptr(p) + i - datap.bss) / sys.PtrSize |
| 1875 | mask[i/sys.PtrSize] = (*addb(bitmap, off/8) >> (off % 8)) & 1 |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1876 | } |
| 1877 | return |
| 1878 | } |
| 1879 | } |
| 1880 | |
| 1881 | // heap |
| 1882 | var n uintptr |
| 1883 | var base uintptr |
| 1884 | if mlookup(uintptr(p), &base, &n, nil) != 0 { |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1885 | mask = make([]byte, n/sys.PtrSize) |
| 1886 | for i := uintptr(0); i < n; i += sys.PtrSize { |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1887 | hbits := heapBitsForAddr(base + i) |
| 1888 | if hbits.isPointer() { |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1889 | mask[i/sys.PtrSize] = 1 |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1890 | } |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1891 | if i != 1*sys.PtrSize && !hbits.morePointers() { |
| 1892 | mask = mask[:i/sys.PtrSize] |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1893 | break |
| 1894 | } |
| 1895 | } |
| 1896 | return |
| 1897 | } |
| 1898 | |
| 1899 | // stack |
| 1900 | if _g_ := getg(); _g_.m.curg.stack.lo <= uintptr(p) && uintptr(p) < _g_.m.curg.stack.hi { |
| 1901 | var frame stkframe |
| 1902 | frame.sp = uintptr(p) |
| 1903 | _g_ := getg() |
| 1904 | gentraceback(_g_.m.curg.sched.pc, _g_.m.curg.sched.sp, 0, _g_.m.curg, 0, nil, 1000, getgcmaskcb, noescape(unsafe.Pointer(&frame)), 0) |
| 1905 | if frame.fn != nil { |
| 1906 | f := frame.fn |
| 1907 | targetpc := frame.continpc |
| 1908 | if targetpc == 0 { |
| 1909 | return |
| 1910 | } |
| 1911 | if targetpc != f.entry { |
| 1912 | targetpc-- |
| 1913 | } |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1914 | pcdata := pcdatavalue(f, _PCDATA_StackMapIndex, targetpc, nil) |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1915 | if pcdata == -1 { |
| 1916 | return |
| 1917 | } |
| 1918 | stkmap := (*stackmap)(funcdata(f, _FUNCDATA_LocalsPointerMaps)) |
| 1919 | if stkmap == nil || stkmap.n <= 0 { |
| 1920 | return |
| 1921 | } |
| 1922 | bv := stackmapdata(stkmap, pcdata) |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1923 | size := uintptr(bv.n) * sys.PtrSize |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1924 | n := (*ptrtype)(unsafe.Pointer(t)).elem.size |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1925 | mask = make([]byte, n/sys.PtrSize) |
| 1926 | for i := uintptr(0); i < n; i += sys.PtrSize { |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1927 | bitmap := bv.bytedata |
Dan Willemsen | 38f2dba | 2016-07-08 14:54:35 -0700 | [diff] [blame] | 1928 | off := (uintptr(p) + i - frame.varp + size) / sys.PtrSize |
| 1929 | mask[i/sys.PtrSize] = (*addb(bitmap, off/8) >> (off % 8)) & 1 |
Dan Willemsen | 09eb3b1 | 2015-09-16 14:34:17 -0700 | [diff] [blame] | 1930 | } |
| 1931 | } |
| 1932 | return |
| 1933 | } |
| 1934 | |
| 1935 | // otherwise, not something the GC knows about. |
| 1936 | // possibly read-only data, like malloc(0). |
| 1937 | // must not have pointers |
| 1938 | return |
| 1939 | } |