Source file src/runtime/runtime2.go
1 // Copyright 2009 The Go Authors. All rights reserved. 2 // Use of this source code is governed by a BSD-style 3 // license that can be found in the LICENSE file. 4 5 package runtime 6 7 import ( 8 "internal/goarch" 9 "runtime/internal/atomic" 10 "unsafe" 11 ) 12 13 // defined constants 14 const ( 15 // G status 16 // 17 // Beyond indicating the general state of a G, the G status 18 // acts like a lock on the goroutine's stack (and hence its 19 // ability to execute user code). 20 // 21 // If you add to this list, add to the list 22 // of "okay during garbage collection" status 23 // in mgcmark.go too. 24 // 25 // TODO(austin): The _Gscan bit could be much lighter-weight. 26 // For example, we could choose not to run _Gscanrunnable 27 // goroutines found in the run queue, rather than CAS-looping 28 // until they become _Grunnable. And transitions like 29 // _Gscanwaiting -> _Gscanrunnable are actually okay because 30 // they don't affect stack ownership. 31 32 // _Gidle means this goroutine was just allocated and has not 33 // yet been initialized. 34 _Gidle = iota // 0 35 36 // _Grunnable means this goroutine is on a run queue. It is 37 // not currently executing user code. The stack is not owned. 38 _Grunnable // 1 39 40 // _Grunning means this goroutine may execute user code. The 41 // stack is owned by this goroutine. It is not on a run queue. 42 // It is assigned an M and a P (g.m and g.m.p are valid). 43 _Grunning // 2 44 45 // _Gsyscall means this goroutine is executing a system call. 46 // It is not executing user code. The stack is owned by this 47 // goroutine. It is not on a run queue. It is assigned an M. 48 _Gsyscall // 3 49 50 // _Gwaiting means this goroutine is blocked in the runtime. 51 // It is not executing user code. It is not on a run queue, 52 // but should be recorded somewhere (e.g., a channel wait 53 // queue) so it can be ready()d when necessary. The stack is 54 // not owned *except* that a channel operation may read or 55 // write parts of the stack under the appropriate channel 56 // lock. Otherwise, it is not safe to access the stack after a 57 // goroutine enters _Gwaiting (e.g., it may get moved). 58 _Gwaiting // 4 59 60 // _Gmoribund_unused is currently unused, but hardcoded in gdb 61 // scripts. 62 _Gmoribund_unused // 5 63 64 // _Gdead means this goroutine is currently unused. It may be 65 // just exited, on a free list, or just being initialized. It 66 // is not executing user code. It may or may not have a stack 67 // allocated. The G and its stack (if any) are owned by the M 68 // that is exiting the G or that obtained the G from the free 69 // list. 70 _Gdead // 6 71 72 // _Genqueue_unused is currently unused. 73 _Genqueue_unused // 7 74 75 // _Gcopystack means this goroutine's stack is being moved. It 76 // is not executing user code and is not on a run queue. The 77 // stack is owned by the goroutine that put it in _Gcopystack. 78 _Gcopystack // 8 79 80 // _Gpreempted means this goroutine stopped itself for a 81 // suspendG preemption. It is like _Gwaiting, but nothing is 82 // yet responsible for ready()ing it. Some suspendG must CAS 83 // the status to _Gwaiting to take responsibility for 84 // ready()ing this G. 85 _Gpreempted // 9 86 87 // _Gscan combined with one of the above states other than 88 // _Grunning indicates that GC is scanning the stack. The 89 // goroutine is not executing user code and the stack is owned 90 // by the goroutine that set the _Gscan bit. 91 // 92 // _Gscanrunning is different: it is used to briefly block 93 // state transitions while GC signals the G to scan its own 94 // stack. This is otherwise like _Grunning. 95 // 96 // atomicstatus&~Gscan gives the state the goroutine will 97 // return to when the scan completes. 98 _Gscan = 0x1000 99 _Gscanrunnable = _Gscan + _Grunnable // 0x1001 100 _Gscanrunning = _Gscan + _Grunning // 0x1002 101 _Gscansyscall = _Gscan + _Gsyscall // 0x1003 102 _Gscanwaiting = _Gscan + _Gwaiting // 0x1004 103 _Gscanpreempted = _Gscan + _Gpreempted // 0x1009 104 ) 105 106 const ( 107 // P status 108 109 // _Pidle means a P is not being used to run user code or the 110 // scheduler. Typically, it's on the idle P list and available 111 // to the scheduler, but it may just be transitioning between 112 // other states. 113 // 114 // The P is owned by the idle list or by whatever is 115 // transitioning its state. Its run queue is empty. 116 _Pidle = iota 117 118 // _Prunning means a P is owned by an M and is being used to 119 // run user code or the scheduler. Only the M that owns this P 120 // is allowed to change the P's status from _Prunning. The M 121 // may transition the P to _Pidle (if it has no more work to 122 // do), _Psyscall (when entering a syscall), or _Pgcstop (to 123 // halt for the GC). The M may also hand ownership of the P 124 // off directly to another M (e.g., to schedule a locked G). 125 _Prunning 126 127 // _Psyscall means a P is not running user code. It has 128 // affinity to an M in a syscall but is not owned by it and 129 // may be stolen by another M. This is similar to _Pidle but 130 // uses lightweight transitions and maintains M affinity. 131 // 132 // Leaving _Psyscall must be done with a CAS, either to steal 133 // or retake the P. Note that there's an ABA hazard: even if 134 // an M successfully CASes its original P back to _Prunning 135 // after a syscall, it must understand the P may have been 136 // used by another M in the interim. 137 _Psyscall 138 139 // _Pgcstop means a P is halted for STW and owned by the M 140 // that stopped the world. The M that stopped the world 141 // continues to use its P, even in _Pgcstop. Transitioning 142 // from _Prunning to _Pgcstop causes an M to release its P and 143 // park. 144 // 145 // The P retains its run queue and startTheWorld will restart 146 // the scheduler on Ps with non-empty run queues. 147 _Pgcstop 148 149 // _Pdead means a P is no longer used (GOMAXPROCS shrank). We 150 // reuse Ps if GOMAXPROCS increases. A dead P is mostly 151 // stripped of its resources, though a few things remain 152 // (e.g., trace buffers). 153 _Pdead 154 ) 155 156 // Mutual exclusion locks. In the uncontended case, 157 // as fast as spin locks (just a few user-level instructions), 158 // but on the contention path they sleep in the kernel. 159 // A zeroed Mutex is unlocked (no need to initialize each lock). 160 // Initialization is helpful for static lock ranking, but not required. 161 type mutex struct { 162 // Empty struct if lock ranking is disabled, otherwise includes the lock rank 163 lockRankStruct 164 // Futex-based impl treats it as uint32 key, 165 // while sema-based impl as M* waitm. 166 // Used to be a union, but unions break precise GC. 167 key uintptr 168 } 169 170 // sleep and wakeup on one-time events. 171 // before any calls to notesleep or notewakeup, 172 // must call noteclear to initialize the Note. 173 // then, exactly one thread can call notesleep 174 // and exactly one thread can call notewakeup (once). 175 // once notewakeup has been called, the notesleep 176 // will return. future notesleep will return immediately. 177 // subsequent noteclear must be called only after 178 // previous notesleep has returned, e.g. it's disallowed 179 // to call noteclear straight after notewakeup. 180 // 181 // notetsleep is like notesleep but wakes up after 182 // a given number of nanoseconds even if the event 183 // has not yet happened. if a goroutine uses notetsleep to 184 // wake up early, it must wait to call noteclear until it 185 // can be sure that no other goroutine is calling 186 // notewakeup. 187 // 188 // notesleep/notetsleep are generally called on g0, 189 // notetsleepg is similar to notetsleep but is called on user g. 190 type note struct { 191 // Futex-based impl treats it as uint32 key, 192 // while sema-based impl as M* waitm. 193 // Used to be a union, but unions break precise GC. 194 key uintptr 195 } 196 197 type funcval struct { 198 fn uintptr 199 // variable-size, fn-specific data here 200 } 201 202 type iface struct { 203 tab *itab 204 data unsafe.Pointer 205 } 206 207 type eface struct { 208 _type *_type 209 data unsafe.Pointer 210 } 211 212 func efaceOf(ep *any) *eface { 213 return (*eface)(unsafe.Pointer(ep)) 214 } 215 216 // The guintptr, muintptr, and puintptr are all used to bypass write barriers. 217 // It is particularly important to avoid write barriers when the current P has 218 // been released, because the GC thinks the world is stopped, and an 219 // unexpected write barrier would not be synchronized with the GC, 220 // which can lead to a half-executed write barrier that has marked the object 221 // but not queued it. If the GC skips the object and completes before the 222 // queuing can occur, it will incorrectly free the object. 223 // 224 // We tried using special assignment functions invoked only when not 225 // holding a running P, but then some updates to a particular memory 226 // word went through write barriers and some did not. This breaks the 227 // write barrier shadow checking mode, and it is also scary: better to have 228 // a word that is completely ignored by the GC than to have one for which 229 // only a few updates are ignored. 230 // 231 // Gs and Ps are always reachable via true pointers in the 232 // allgs and allp lists or (during allocation before they reach those lists) 233 // from stack variables. 234 // 235 // Ms are always reachable via true pointers either from allm or 236 // freem. Unlike Gs and Ps we do free Ms, so it's important that 237 // nothing ever hold an muintptr across a safe point. 238 239 // A guintptr holds a goroutine pointer, but typed as a uintptr 240 // to bypass write barriers. It is used in the Gobuf goroutine state 241 // and in scheduling lists that are manipulated without a P. 242 // 243 // The Gobuf.g goroutine pointer is almost always updated by assembly code. 244 // In one of the few places it is updated by Go code - func save - it must be 245 // treated as a uintptr to avoid a write barrier being emitted at a bad time. 246 // Instead of figuring out how to emit the write barriers missing in the 247 // assembly manipulation, we change the type of the field to uintptr, 248 // so that it does not require write barriers at all. 249 // 250 // Goroutine structs are published in the allg list and never freed. 251 // That will keep the goroutine structs from being collected. 252 // There is never a time that Gobuf.g's contain the only references 253 // to a goroutine: the publishing of the goroutine in allg comes first. 254 // Goroutine pointers are also kept in non-GC-visible places like TLS, 255 // so I can't see them ever moving. If we did want to start moving data 256 // in the GC, we'd need to allocate the goroutine structs from an 257 // alternate arena. Using guintptr doesn't make that problem any worse. 258 // Note that pollDesc.rg, pollDesc.wg also store g in uintptr form, 259 // so they would need to be updated too if g's start moving. 260 type guintptr uintptr 261 262 //go:nosplit 263 func (gp guintptr) ptr() *g { return (*g)(unsafe.Pointer(gp)) } 264 265 //go:nosplit 266 func (gp *guintptr) set(g *g) { *gp = guintptr(unsafe.Pointer(g)) } 267 268 //go:nosplit 269 func (gp *guintptr) cas(old, new guintptr) bool { 270 return atomic.Casuintptr((*uintptr)(unsafe.Pointer(gp)), uintptr(old), uintptr(new)) 271 } 272 273 // setGNoWB performs *gp = new without a write barrier. 274 // For times when it's impractical to use a guintptr. 275 //go:nosplit 276 //go:nowritebarrier 277 func setGNoWB(gp **g, new *g) { 278 (*guintptr)(unsafe.Pointer(gp)).set(new) 279 } 280 281 type puintptr uintptr 282 283 //go:nosplit 284 func (pp puintptr) ptr() *p { return (*p)(unsafe.Pointer(pp)) } 285 286 //go:nosplit 287 func (pp *puintptr) set(p *p) { *pp = puintptr(unsafe.Pointer(p)) } 288 289 // muintptr is a *m that is not tracked by the garbage collector. 290 // 291 // Because we do free Ms, there are some additional constrains on 292 // muintptrs: 293 // 294 // 1. Never hold an muintptr locally across a safe point. 295 // 296 // 2. Any muintptr in the heap must be owned by the M itself so it can 297 // ensure it is not in use when the last true *m is released. 298 type muintptr uintptr 299 300 //go:nosplit 301 func (mp muintptr) ptr() *m { return (*m)(unsafe.Pointer(mp)) } 302 303 //go:nosplit 304 func (mp *muintptr) set(m *m) { *mp = muintptr(unsafe.Pointer(m)) } 305 306 // setMNoWB performs *mp = new without a write barrier. 307 // For times when it's impractical to use an muintptr. 308 //go:nosplit 309 //go:nowritebarrier 310 func setMNoWB(mp **m, new *m) { 311 (*muintptr)(unsafe.Pointer(mp)).set(new) 312 } 313 314 type gobuf struct { 315 // The offsets of sp, pc, and g are known to (hard-coded in) libmach. 316 // 317 // ctxt is unusual with respect to GC: it may be a 318 // heap-allocated funcval, so GC needs to track it, but it 319 // needs to be set and cleared from assembly, where it's 320 // difficult to have write barriers. However, ctxt is really a 321 // saved, live register, and we only ever exchange it between 322 // the real register and the gobuf. Hence, we treat it as a 323 // root during stack scanning, which means assembly that saves 324 // and restores it doesn't need write barriers. It's still 325 // typed as a pointer so that any other writes from Go get 326 // write barriers. 327 sp uintptr 328 pc uintptr 329 g guintptr 330 ctxt unsafe.Pointer 331 ret uintptr 332 lr uintptr 333 bp uintptr // for framepointer-enabled architectures 334 } 335 336 // sudog represents a g in a wait list, such as for sending/receiving 337 // on a channel. 338 // 339 // sudog is necessary because the g ↔ synchronization object relation 340 // is many-to-many. A g can be on many wait lists, so there may be 341 // many sudogs for one g; and many gs may be waiting on the same 342 // synchronization object, so there may be many sudogs for one object. 343 // 344 // sudogs are allocated from a special pool. Use acquireSudog and 345 // releaseSudog to allocate and free them. 346 type sudog struct { 347 // The following fields are protected by the hchan.lock of the 348 // channel this sudog is blocking on. shrinkstack depends on 349 // this for sudogs involved in channel ops. 350 351 g *g 352 353 next *sudog 354 prev *sudog 355 elem unsafe.Pointer // data element (may point to stack) 356 357 // The following fields are never accessed concurrently. 358 // For channels, waitlink is only accessed by g. 359 // For semaphores, all fields (including the ones above) 360 // are only accessed when holding a semaRoot lock. 361 362 acquiretime int64 363 releasetime int64 364 ticket uint32 365 366 // isSelect indicates g is participating in a select, so 367 // g.selectDone must be CAS'd to win the wake-up race. 368 isSelect bool 369 370 // success indicates whether communication over channel c 371 // succeeded. It is true if the goroutine was awoken because a 372 // value was delivered over channel c, and false if awoken 373 // because c was closed. 374 success bool 375 376 parent *sudog // semaRoot binary tree 377 waitlink *sudog // g.waiting list or semaRoot 378 waittail *sudog // semaRoot 379 c *hchan // channel 380 } 381 382 type libcall struct { 383 fn uintptr 384 n uintptr // number of parameters 385 args uintptr // parameters 386 r1 uintptr // return values 387 r2 uintptr 388 err uintptr // error number 389 } 390 391 // Stack describes a Go execution stack. 392 // The bounds of the stack are exactly [lo, hi), 393 // with no implicit data structures on either side. 394 type stack struct { 395 lo uintptr 396 hi uintptr 397 } 398 399 // heldLockInfo gives info on a held lock and the rank of that lock 400 type heldLockInfo struct { 401 lockAddr uintptr 402 rank lockRank 403 } 404 405 type g struct { 406 // Stack parameters. 407 // stack describes the actual stack memory: [stack.lo, stack.hi). 408 // stackguard0 is the stack pointer compared in the Go stack growth prologue. 409 // It is stack.lo+StackGuard normally, but can be StackPreempt to trigger a preemption. 410 // stackguard1 is the stack pointer compared in the C stack growth prologue. 411 // It is stack.lo+StackGuard on g0 and gsignal stacks. 412 // It is ~0 on other goroutine stacks, to trigger a call to morestackc (and crash). 413 stack stack // offset known to runtime/cgo 414 stackguard0 uintptr // offset known to liblink 415 stackguard1 uintptr // offset known to liblink 416 417 _panic *_panic // innermost panic - offset known to liblink 418 _defer *_defer // innermost defer 419 m *m // current m; offset known to arm liblink 420 sched gobuf 421 syscallsp uintptr // if status==Gsyscall, syscallsp = sched.sp to use during gc 422 syscallpc uintptr // if status==Gsyscall, syscallpc = sched.pc to use during gc 423 stktopsp uintptr // expected sp at top of stack, to check in traceback 424 // param is a generic pointer parameter field used to pass 425 // values in particular contexts where other storage for the 426 // parameter would be difficult to find. It is currently used 427 // in three ways: 428 // 1. When a channel operation wakes up a blocked goroutine, it sets param to 429 // point to the sudog of the completed blocking operation. 430 // 2. By gcAssistAlloc1 to signal back to its caller that the goroutine completed 431 // the GC cycle. It is unsafe to do so in any other way, because the goroutine's 432 // stack may have moved in the meantime. 433 // 3. By debugCallWrap to pass parameters to a new goroutine because allocating a 434 // closure in the runtime is forbidden. 435 param unsafe.Pointer 436 atomicstatus uint32 437 stackLock uint32 // sigprof/scang lock; TODO: fold in to atomicstatus 438 goid int64 439 schedlink guintptr 440 waitsince int64 // approx time when the g become blocked 441 waitreason waitReason // if status==Gwaiting 442 443 preempt bool // preemption signal, duplicates stackguard0 = stackpreempt 444 preemptStop bool // transition to _Gpreempted on preemption; otherwise, just deschedule 445 preemptShrink bool // shrink stack at synchronous safe point 446 447 // asyncSafePoint is set if g is stopped at an asynchronous 448 // safe point. This means there are frames on the stack 449 // without precise pointer information. 450 asyncSafePoint bool 451 452 paniconfault bool // panic (instead of crash) on unexpected fault address 453 gcscandone bool // g has scanned stack; protected by _Gscan bit in status 454 throwsplit bool // must not split stack 455 // activeStackChans indicates that there are unlocked channels 456 // pointing into this goroutine's stack. If true, stack 457 // copying needs to acquire channel locks to protect these 458 // areas of the stack. 459 activeStackChans bool 460 // parkingOnChan indicates that the goroutine is about to 461 // park on a chansend or chanrecv. Used to signal an unsafe point 462 // for stack shrinking. It's a boolean value, but is updated atomically. 463 parkingOnChan uint8 464 465 raceignore int8 // ignore race detection events 466 sysblocktraced bool // StartTrace has emitted EvGoInSyscall about this goroutine 467 tracking bool // whether we're tracking this G for sched latency statistics 468 trackingSeq uint8 // used to decide whether to track this G 469 runnableStamp int64 // timestamp of when the G last became runnable, only used when tracking 470 runnableTime int64 // the amount of time spent runnable, cleared when running, only used when tracking 471 sysexitticks int64 // cputicks when syscall has returned (for tracing) 472 traceseq uint64 // trace event sequencer 473 tracelastp puintptr // last P emitted an event for this goroutine 474 lockedm muintptr 475 sig uint32 476 writebuf []byte 477 sigcode0 uintptr 478 sigcode1 uintptr 479 sigpc uintptr 480 gopc uintptr // pc of go statement that created this goroutine 481 ancestors *[]ancestorInfo // ancestor information goroutine(s) that created this goroutine (only used if debug.tracebackancestors) 482 startpc uintptr // pc of goroutine function 483 racectx uintptr 484 waiting *sudog // sudog structures this g is waiting on (that have a valid elem ptr); in lock order 485 cgoCtxt []uintptr // cgo traceback context 486 labels unsafe.Pointer // profiler labels 487 timer *timer // cached timer for time.Sleep 488 selectDone uint32 // are we participating in a select and did someone win the race? 489 490 // Per-G GC state 491 492 // gcAssistBytes is this G's GC assist credit in terms of 493 // bytes allocated. If this is positive, then the G has credit 494 // to allocate gcAssistBytes bytes without assisting. If this 495 // is negative, then the G must correct this by performing 496 // scan work. We track this in bytes to make it fast to update 497 // and check for debt in the malloc hot path. The assist ratio 498 // determines how this corresponds to scan work debt. 499 gcAssistBytes int64 500 } 501 502 // gTrackingPeriod is the number of transitions out of _Grunning between 503 // latency tracking runs. 504 const gTrackingPeriod = 8 505 506 const ( 507 // tlsSlots is the number of pointer-sized slots reserved for TLS on some platforms, 508 // like Windows. 509 tlsSlots = 6 510 tlsSize = tlsSlots * goarch.PtrSize 511 ) 512 513 type m struct { 514 g0 *g // goroutine with scheduling stack 515 morebuf gobuf // gobuf arg to morestack 516 divmod uint32 // div/mod denominator for arm - known to liblink 517 _ uint32 // align next field to 8 bytes 518 519 // Fields not known to debuggers. 520 procid uint64 // for debuggers, but offset not hard-coded 521 gsignal *g // signal-handling g 522 goSigStack gsignalStack // Go-allocated signal handling stack 523 sigmask sigset // storage for saved signal mask 524 tls [tlsSlots]uintptr // thread-local storage (for x86 extern register) 525 mstartfn func() 526 curg *g // current running goroutine 527 caughtsig guintptr // goroutine running during fatal signal 528 p puintptr // attached p for executing go code (nil if not executing go code) 529 nextp puintptr 530 oldp puintptr // the p that was attached before executing a syscall 531 id int64 532 mallocing int32 533 throwing int32 534 preemptoff string // if != "", keep curg running on this m 535 locks int32 536 dying int32 537 profilehz int32 538 spinning bool // m is out of work and is actively looking for work 539 blocked bool // m is blocked on a note 540 newSigstack bool // minit on C thread called sigaltstack 541 printlock int8 542 incgo bool // m is executing a cgo call 543 freeWait uint32 // if == 0, safe to free g0 and delete m (atomic) 544 fastrand uint64 545 needextram bool 546 traceback uint8 547 ncgocall uint64 // number of cgo calls in total 548 ncgo int32 // number of cgo calls currently in progress 549 cgoCallersUse uint32 // if non-zero, cgoCallers in use temporarily 550 cgoCallers *cgoCallers // cgo traceback if crashing in cgo call 551 park note 552 alllink *m // on allm 553 schedlink muintptr 554 lockedg guintptr 555 createstack [32]uintptr // stack that created this thread. 556 lockedExt uint32 // tracking for external LockOSThread 557 lockedInt uint32 // tracking for internal lockOSThread 558 nextwaitm muintptr // next m waiting for lock 559 waitunlockf func(*g, unsafe.Pointer) bool 560 waitlock unsafe.Pointer 561 waittraceev byte 562 waittraceskip int 563 startingtrace bool 564 syscalltick uint32 565 freelink *m // on sched.freem 566 567 // these are here because they are too large to be on the stack 568 // of low-level NOSPLIT functions. 569 libcall libcall 570 libcallpc uintptr // for cpu profiler 571 libcallsp uintptr 572 libcallg guintptr 573 syscall libcall // stores syscall parameters on windows 574 575 vdsoSP uintptr // SP for traceback while in VDSO call (0 if not in call) 576 vdsoPC uintptr // PC for traceback while in VDSO call 577 578 // preemptGen counts the number of completed preemption 579 // signals. This is used to detect when a preemption is 580 // requested, but fails. Accessed atomically. 581 preemptGen uint32 582 583 // Whether this is a pending preemption signal on this M. 584 // Accessed atomically. 585 signalPending uint32 586 587 dlogPerM 588 589 mOS 590 591 // Up to 10 locks held by this m, maintained by the lock ranking code. 592 locksHeldLen int 593 locksHeld [10]heldLockInfo 594 } 595 596 type p struct { 597 id int32 598 status uint32 // one of pidle/prunning/... 599 link puintptr 600 schedtick uint32 // incremented on every scheduler call 601 syscalltick uint32 // incremented on every system call 602 sysmontick sysmontick // last tick observed by sysmon 603 m muintptr // back-link to associated m (nil if idle) 604 mcache *mcache 605 pcache pageCache 606 raceprocctx uintptr 607 608 deferpool []*_defer // pool of available defer structs (see panic.go) 609 deferpoolbuf [32]*_defer 610 611 // Cache of goroutine ids, amortizes accesses to runtime·sched.goidgen. 612 goidcache uint64 613 goidcacheend uint64 614 615 // Queue of runnable goroutines. Accessed without lock. 616 runqhead uint32 617 runqtail uint32 618 runq [256]guintptr 619 // runnext, if non-nil, is a runnable G that was ready'd by 620 // the current G and should be run next instead of what's in 621 // runq if there's time remaining in the running G's time 622 // slice. It will inherit the time left in the current time 623 // slice. If a set of goroutines is locked in a 624 // communicate-and-wait pattern, this schedules that set as a 625 // unit and eliminates the (potentially large) scheduling 626 // latency that otherwise arises from adding the ready'd 627 // goroutines to the end of the run queue. 628 // 629 // Note that while other P's may atomically CAS this to zero, 630 // only the owner P can CAS it to a valid G. 631 runnext guintptr 632 633 // Available G's (status == Gdead) 634 gFree struct { 635 gList 636 n int32 637 } 638 639 sudogcache []*sudog 640 sudogbuf [128]*sudog 641 642 // Cache of mspan objects from the heap. 643 mspancache struct { 644 // We need an explicit length here because this field is used 645 // in allocation codepaths where write barriers are not allowed, 646 // and eliminating the write barrier/keeping it eliminated from 647 // slice updates is tricky, moreso than just managing the length 648 // ourselves. 649 len int 650 buf [128]*mspan 651 } 652 653 tracebuf traceBufPtr 654 655 // traceSweep indicates the sweep events should be traced. 656 // This is used to defer the sweep start event until a span 657 // has actually been swept. 658 traceSweep bool 659 // traceSwept and traceReclaimed track the number of bytes 660 // swept and reclaimed by sweeping in the current sweep loop. 661 traceSwept, traceReclaimed uintptr 662 663 palloc persistentAlloc // per-P to avoid mutex 664 665 _ uint32 // Alignment for atomic fields below 666 667 // The when field of the first entry on the timer heap. 668 // This is updated using atomic functions. 669 // This is 0 if the timer heap is empty. 670 timer0When uint64 671 672 // The earliest known nextwhen field of a timer with 673 // timerModifiedEarlier status. Because the timer may have been 674 // modified again, there need not be any timer with this value. 675 // This is updated using atomic functions. 676 // This is 0 if there are no timerModifiedEarlier timers. 677 timerModifiedEarliest uint64 678 679 // Per-P GC state 680 gcAssistTime int64 // Nanoseconds in assistAlloc 681 gcFractionalMarkTime int64 // Nanoseconds in fractional mark worker (atomic) 682 683 // gcMarkWorkerMode is the mode for the next mark worker to run in. 684 // That is, this is used to communicate with the worker goroutine 685 // selected for immediate execution by 686 // gcController.findRunnableGCWorker. When scheduling other goroutines, 687 // this field must be set to gcMarkWorkerNotWorker. 688 gcMarkWorkerMode gcMarkWorkerMode 689 // gcMarkWorkerStartTime is the nanotime() at which the most recent 690 // mark worker started. 691 gcMarkWorkerStartTime int64 692 693 // gcw is this P's GC work buffer cache. The work buffer is 694 // filled by write barriers, drained by mutator assists, and 695 // disposed on certain GC state transitions. 696 gcw gcWork 697 698 // wbBuf is this P's GC write barrier buffer. 699 // 700 // TODO: Consider caching this in the running G. 701 wbBuf wbBuf 702 703 runSafePointFn uint32 // if 1, run sched.safePointFn at next safe point 704 705 // statsSeq is a counter indicating whether this P is currently 706 // writing any stats. Its value is even when not, odd when it is. 707 statsSeq uint32 708 709 // Lock for timers. We normally access the timers while running 710 // on this P, but the scheduler can also do it from a different P. 711 timersLock mutex 712 713 // Actions to take at some time. This is used to implement the 714 // standard library's time package. 715 // Must hold timersLock to access. 716 timers []*timer 717 718 // Number of timers in P's heap. 719 // Modified using atomic instructions. 720 numTimers uint32 721 722 // Number of timerDeleted timers in P's heap. 723 // Modified using atomic instructions. 724 deletedTimers uint32 725 726 // Race context used while executing timer functions. 727 timerRaceCtx uintptr 728 729 // scannableStackSizeDelta accumulates the amount of stack space held by 730 // live goroutines (i.e. those eligible for stack scanning). 731 // Flushed to gcController.scannableStackSize once scannableStackSizeSlack 732 // or -scannableStackSizeSlack is reached. 733 scannableStackSizeDelta int64 734 735 // preempt is set to indicate that this P should be enter the 736 // scheduler ASAP (regardless of what G is running on it). 737 preempt bool 738 739 // Padding is no longer needed. False sharing is now not a worry because p is large enough 740 // that its size class is an integer multiple of the cache line size (for any of our architectures). 741 } 742 743 type schedt struct { 744 // accessed atomically. keep at top to ensure alignment on 32-bit systems. 745 goidgen uint64 746 lastpoll uint64 // time of last network poll, 0 if currently polling 747 pollUntil uint64 // time to which current poll is sleeping 748 749 lock mutex 750 751 // When increasing nmidle, nmidlelocked, nmsys, or nmfreed, be 752 // sure to call checkdead(). 753 754 midle muintptr // idle m's waiting for work 755 nmidle int32 // number of idle m's waiting for work 756 nmidlelocked int32 // number of locked m's waiting for work 757 mnext int64 // number of m's that have been created and next M ID 758 maxmcount int32 // maximum number of m's allowed (or die) 759 nmsys int32 // number of system m's not counted for deadlock 760 nmfreed int64 // cumulative number of freed m's 761 762 ngsys uint32 // number of system goroutines; updated atomically 763 764 pidle puintptr // idle p's 765 npidle uint32 766 nmspinning uint32 // See "Worker thread parking/unparking" comment in proc.go. 767 768 // Global runnable queue. 769 runq gQueue 770 runqsize int32 771 772 // disable controls selective disabling of the scheduler. 773 // 774 // Use schedEnableUser to control this. 775 // 776 // disable is protected by sched.lock. 777 disable struct { 778 // user disables scheduling of user goroutines. 779 user bool 780 runnable gQueue // pending runnable Gs 781 n int32 // length of runnable 782 } 783 784 // Global cache of dead G's. 785 gFree struct { 786 lock mutex 787 stack gList // Gs with stacks 788 noStack gList // Gs without stacks 789 n int32 790 } 791 792 // Central cache of sudog structs. 793 sudoglock mutex 794 sudogcache *sudog 795 796 // Central pool of available defer structs. 797 deferlock mutex 798 deferpool *_defer 799 800 // freem is the list of m's waiting to be freed when their 801 // m.exited is set. Linked through m.freelink. 802 freem *m 803 804 gcwaiting uint32 // gc is waiting to run 805 stopwait int32 806 stopnote note 807 sysmonwait uint32 808 sysmonnote note 809 810 // safepointFn should be called on each P at the next GC 811 // safepoint if p.runSafePointFn is set. 812 safePointFn func(*p) 813 safePointWait int32 814 safePointNote note 815 816 profilehz int32 // cpu profiling rate 817 818 procresizetime int64 // nanotime() of last change to gomaxprocs 819 totaltime int64 // ∫gomaxprocs dt up to procresizetime 820 821 // sysmonlock protects sysmon's actions on the runtime. 822 // 823 // Acquire and hold this mutex to block sysmon from interacting 824 // with the rest of the runtime. 825 sysmonlock mutex 826 827 // timeToRun is a distribution of scheduling latencies, defined 828 // as the sum of time a G spends in the _Grunnable state before 829 // it transitions to _Grunning. 830 // 831 // timeToRun is protected by sched.lock. 832 timeToRun timeHistogram 833 } 834 835 // Values for the flags field of a sigTabT. 836 const ( 837 _SigNotify = 1 << iota // let signal.Notify have signal, even if from kernel 838 _SigKill // if signal.Notify doesn't take it, exit quietly 839 _SigThrow // if signal.Notify doesn't take it, exit loudly 840 _SigPanic // if the signal is from the kernel, panic 841 _SigDefault // if the signal isn't explicitly requested, don't monitor it 842 _SigGoExit // cause all runtime procs to exit (only used on Plan 9). 843 _SigSetStack // Don't explicitly install handler, but add SA_ONSTACK to existing libc handler 844 _SigUnblock // always unblock; see blockableSig 845 _SigIgn // _SIG_DFL action is to ignore the signal 846 ) 847 848 // Layout of in-memory per-function information prepared by linker 849 // See https://golang.org/s/go12symtab. 850 // Keep in sync with linker (../cmd/link/internal/ld/pcln.go:/pclntab) 851 // and with package debug/gosym and with symtab.go in package runtime. 852 type _func struct { 853 entryoff uint32 // start pc, as offset from moduledata.text/pcHeader.textStart 854 nameoff int32 // function name 855 856 args int32 // in/out args size 857 deferreturn uint32 // offset of start of a deferreturn call instruction from entry, if any. 858 859 pcsp uint32 860 pcfile uint32 861 pcln uint32 862 npcdata uint32 863 cuOffset uint32 // runtime.cutab offset of this function's CU 864 funcID funcID // set for certain special runtime functions 865 flag funcFlag 866 _ [1]byte // pad 867 nfuncdata uint8 // must be last, must end on a uint32-aligned boundary 868 } 869 870 // Pseudo-Func that is returned for PCs that occur in inlined code. 871 // A *Func can be either a *_func or a *funcinl, and they are distinguished 872 // by the first uintptr. 873 type funcinl struct { 874 ones uint32 // set to ^0 to distinguish from _func 875 entry uintptr // entry of the real (the "outermost") frame 876 name string 877 file string 878 line int 879 } 880 881 // layout of Itab known to compilers 882 // allocated in non-garbage-collected memory 883 // Needs to be in sync with 884 // ../cmd/compile/internal/reflectdata/reflect.go:/^func.WriteTabs. 885 type itab struct { 886 inter *interfacetype 887 _type *_type 888 hash uint32 // copy of _type.hash. Used for type switches. 889 _ [4]byte 890 fun [1]uintptr // variable sized. fun[0]==0 means _type does not implement inter. 891 } 892 893 // Lock-free stack node. 894 // Also known to export_test.go. 895 type lfnode struct { 896 next uint64 897 pushcnt uintptr 898 } 899 900 type forcegcstate struct { 901 lock mutex 902 g *g 903 idle uint32 904 } 905 906 // extendRandom extends the random numbers in r[:n] to the whole slice r. 907 // Treats n<0 as n==0. 908 func extendRandom(r []byte, n int) { 909 if n < 0 { 910 n = 0 911 } 912 for n < len(r) { 913 // Extend random bits using hash function & time seed 914 w := n 915 if w > 16 { 916 w = 16 917 } 918 h := memhash(unsafe.Pointer(&r[n-w]), uintptr(nanotime()), uintptr(w)) 919 for i := 0; i < goarch.PtrSize && n < len(r); i++ { 920 r[n] = byte(h) 921 n++ 922 h >>= 8 923 } 924 } 925 } 926 927 // A _defer holds an entry on the list of deferred calls. 928 // If you add a field here, add code to clear it in deferProcStack. 929 // This struct must match the code in cmd/compile/internal/ssagen/ssa.go:deferstruct 930 // and cmd/compile/internal/ssagen/ssa.go:(*state).call. 931 // Some defers will be allocated on the stack and some on the heap. 932 // All defers are logically part of the stack, so write barriers to 933 // initialize them are not required. All defers must be manually scanned, 934 // and for heap defers, marked. 935 type _defer struct { 936 started bool 937 heap bool 938 // openDefer indicates that this _defer is for a frame with open-coded 939 // defers. We have only one defer record for the entire frame (which may 940 // currently have 0, 1, or more defers active). 941 openDefer bool 942 sp uintptr // sp at time of defer 943 pc uintptr // pc at time of defer 944 fn func() // can be nil for open-coded defers 945 _panic *_panic // panic that is running defer 946 link *_defer // next defer on G; can point to either heap or stack! 947 948 // If openDefer is true, the fields below record values about the stack 949 // frame and associated function that has the open-coded defer(s). sp 950 // above will be the sp for the frame, and pc will be address of the 951 // deferreturn call in the function. 952 fd unsafe.Pointer // funcdata for the function associated with the frame 953 varp uintptr // value of varp for the stack frame 954 // framepc is the current pc associated with the stack frame. Together, 955 // with sp above (which is the sp associated with the stack frame), 956 // framepc/sp can be used as pc/sp pair to continue a stack trace via 957 // gentraceback(). 958 framepc uintptr 959 } 960 961 // A _panic holds information about an active panic. 962 // 963 // A _panic value must only ever live on the stack. 964 // 965 // The argp and link fields are stack pointers, but don't need special 966 // handling during stack growth: because they are pointer-typed and 967 // _panic values only live on the stack, regular stack pointer 968 // adjustment takes care of them. 969 type _panic struct { 970 argp unsafe.Pointer // pointer to arguments of deferred call run during panic; cannot move - known to liblink 971 arg any // argument to panic 972 link *_panic // link to earlier panic 973 pc uintptr // where to return to in runtime if this panic is bypassed 974 sp unsafe.Pointer // where to return to in runtime if this panic is bypassed 975 recovered bool // whether this panic is over 976 aborted bool // the panic was aborted 977 goexit bool 978 } 979 980 // stack traces 981 type stkframe struct { 982 fn funcInfo // function being run 983 pc uintptr // program counter within fn 984 continpc uintptr // program counter where execution can continue, or 0 if not 985 lr uintptr // program counter at caller aka link register 986 sp uintptr // stack pointer at pc 987 fp uintptr // stack pointer at caller aka frame pointer 988 varp uintptr // top of local variables 989 argp uintptr // pointer to function arguments 990 arglen uintptr // number of bytes at argp 991 argmap *bitvector // force use of this argmap 992 } 993 994 // ancestorInfo records details of where a goroutine was started. 995 type ancestorInfo struct { 996 pcs []uintptr // pcs from the stack of this goroutine 997 goid int64 // goroutine id of this goroutine; original goroutine possibly dead 998 gopc uintptr // pc of go statement that created this goroutine 999 } 1000 1001 const ( 1002 _TraceRuntimeFrames = 1 << iota // include frames for internal runtime functions. 1003 _TraceTrap // the initial PC, SP are from a trap, not a return PC from a call 1004 _TraceJumpStack // if traceback is on a systemstack, resume trace at g that called into it 1005 ) 1006 1007 // The maximum number of frames we print for a traceback 1008 const _TracebackMaxFrames = 100 1009 1010 // A waitReason explains why a goroutine has been stopped. 1011 // See gopark. Do not re-use waitReasons, add new ones. 1012 type waitReason uint8 1013 1014 const ( 1015 waitReasonZero waitReason = iota // "" 1016 waitReasonGCAssistMarking // "GC assist marking" 1017 waitReasonIOWait // "IO wait" 1018 waitReasonChanReceiveNilChan // "chan receive (nil chan)" 1019 waitReasonChanSendNilChan // "chan send (nil chan)" 1020 waitReasonDumpingHeap // "dumping heap" 1021 waitReasonGarbageCollection // "garbage collection" 1022 waitReasonGarbageCollectionScan // "garbage collection scan" 1023 waitReasonPanicWait // "panicwait" 1024 waitReasonSelect // "select" 1025 waitReasonSelectNoCases // "select (no cases)" 1026 waitReasonGCAssistWait // "GC assist wait" 1027 waitReasonGCSweepWait // "GC sweep wait" 1028 waitReasonGCScavengeWait // "GC scavenge wait" 1029 waitReasonChanReceive // "chan receive" 1030 waitReasonChanSend // "chan send" 1031 waitReasonFinalizerWait // "finalizer wait" 1032 waitReasonForceGCIdle // "force gc (idle)" 1033 waitReasonSemacquire // "semacquire" 1034 waitReasonSleep // "sleep" 1035 waitReasonSyncCondWait // "sync.Cond.Wait" 1036 waitReasonTimerGoroutineIdle // "timer goroutine (idle)" 1037 waitReasonTraceReaderBlocked // "trace reader (blocked)" 1038 waitReasonWaitForGCCycle // "wait for GC cycle" 1039 waitReasonGCWorkerIdle // "GC worker (idle)" 1040 waitReasonPreempted // "preempted" 1041 waitReasonDebugCall // "debug call" 1042 ) 1043 1044 var waitReasonStrings = [...]string{ 1045 waitReasonZero: "", 1046 waitReasonGCAssistMarking: "GC assist marking", 1047 waitReasonIOWait: "IO wait", 1048 waitReasonChanReceiveNilChan: "chan receive (nil chan)", 1049 waitReasonChanSendNilChan: "chan send (nil chan)", 1050 waitReasonDumpingHeap: "dumping heap", 1051 waitReasonGarbageCollection: "garbage collection", 1052 waitReasonGarbageCollectionScan: "garbage collection scan", 1053 waitReasonPanicWait: "panicwait", 1054 waitReasonSelect: "select", 1055 waitReasonSelectNoCases: "select (no cases)", 1056 waitReasonGCAssistWait: "GC assist wait", 1057 waitReasonGCSweepWait: "GC sweep wait", 1058 waitReasonGCScavengeWait: "GC scavenge wait", 1059 waitReasonChanReceive: "chan receive", 1060 waitReasonChanSend: "chan send", 1061 waitReasonFinalizerWait: "finalizer wait", 1062 waitReasonForceGCIdle: "force gc (idle)", 1063 waitReasonSemacquire: "semacquire", 1064 waitReasonSleep: "sleep", 1065 waitReasonSyncCondWait: "sync.Cond.Wait", 1066 waitReasonTimerGoroutineIdle: "timer goroutine (idle)", 1067 waitReasonTraceReaderBlocked: "trace reader (blocked)", 1068 waitReasonWaitForGCCycle: "wait for GC cycle", 1069 waitReasonGCWorkerIdle: "GC worker (idle)", 1070 waitReasonPreempted: "preempted", 1071 waitReasonDebugCall: "debug call", 1072 } 1073 1074 func (w waitReason) String() string { 1075 if w < 0 || w >= waitReason(len(waitReasonStrings)) { 1076 return "unknown wait reason" 1077 } 1078 return waitReasonStrings[w] 1079 } 1080 1081 var ( 1082 allm *m 1083 gomaxprocs int32 1084 ncpu int32 1085 forcegc forcegcstate 1086 sched schedt 1087 newprocs int32 1088 1089 // allpLock protects P-less reads and size changes of allp, idlepMask, 1090 // and timerpMask, and all writes to allp. 1091 allpLock mutex 1092 // len(allp) == gomaxprocs; may change at safe points, otherwise 1093 // immutable. 1094 allp []*p 1095 // Bitmask of Ps in _Pidle list, one bit per P. Reads and writes must 1096 // be atomic. Length may change at safe points. 1097 // 1098 // Each P must update only its own bit. In order to maintain 1099 // consistency, a P going idle must the idle mask simultaneously with 1100 // updates to the idle P list under the sched.lock, otherwise a racing 1101 // pidleget may clear the mask before pidleput sets the mask, 1102 // corrupting the bitmap. 1103 // 1104 // N.B., procresize takes ownership of all Ps in stopTheWorldWithSema. 1105 idlepMask pMask 1106 // Bitmask of Ps that may have a timer, one bit per P. Reads and writes 1107 // must be atomic. Length may change at safe points. 1108 timerpMask pMask 1109 1110 // Pool of GC parked background workers. Entries are type 1111 // *gcBgMarkWorkerNode. 1112 gcBgMarkWorkerPool lfstack 1113 1114 // Total number of gcBgMarkWorker goroutines. Protected by worldsema. 1115 gcBgMarkWorkerCount int32 1116 1117 // Information about what cpu features are available. 1118 // Packages outside the runtime should not use these 1119 // as they are not an external api. 1120 // Set on startup in asm_{386,amd64}.s 1121 processorVersionInfo uint32 1122 isIntel bool 1123 1124 goarm uint8 // set by cmd/link on arm systems 1125 ) 1126 1127 // Set by the linker so the runtime can determine the buildmode. 1128 var ( 1129 islibrary bool // -buildmode=c-shared 1130 isarchive bool // -buildmode=c-archive 1131 ) 1132 1133 // Must agree with internal/buildcfg.Experiment.FramePointer. 1134 const framepointer_enabled = GOARCH == "amd64" || GOARCH == "arm64" 1135