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.
5 // Garbage collector (GC).
7 // The GC runs concurrently with mutator threads, is type accurate (aka precise), allows multiple
8 // GC thread to run in parallel. It is a concurrent mark and sweep that uses a write barrier. It is
9 // non-generational and non-compacting. Allocation is done using size segregated per P allocation
10 // areas to minimize fragmentation while eliminating locks in the common case.
12 // The algorithm decomposes into several steps.
13 // This is a high level description of the algorithm being used. For an overview of GC a good
14 // place to start is Richard Jones' gchandbook.org.
16 // The algorithm's intellectual heritage includes Dijkstra's on-the-fly algorithm, see
17 // Edsger W. Dijkstra, Leslie Lamport, A. J. Martin, C. S. Scholten, and E. F. M. Steffens. 1978.
18 // On-the-fly garbage collection: an exercise in cooperation. Commun. ACM 21, 11 (November 1978),
20 // For journal quality proofs that these steps are complete, correct, and terminate see
21 // Hudson, R., and Moss, J.E.B. Copying Garbage Collection without stopping the world.
22 // Concurrency and Computation: Practice and Experience 15(3-5), 2003.
24 // 1. GC performs sweep termination.
26 // a. Stop the world. This causes all Ps to reach a GC safe-point.
28 // b. Sweep any unswept spans. There will only be unswept spans if
29 // this GC cycle was forced before the expected time.
31 // 2. GC performs the mark phase.
33 // a. Prepare for the mark phase by setting gcphase to _GCmark
34 // (from _GCoff), enabling the write barrier, enabling mutator
35 // assists, and enqueueing root mark jobs. No objects may be
36 // scanned until all Ps have enabled the write barrier, which is
37 // accomplished using STW.
39 // b. Start the world. From this point, GC work is done by mark
40 // workers started by the scheduler and by assists performed as
41 // part of allocation. The write barrier shades both the
42 // overwritten pointer and the new pointer value for any pointer
43 // writes (see mbarrier.go for details). Newly allocated objects
44 // are immediately marked black.
46 // c. GC performs root marking jobs. This includes scanning all
47 // stacks, shading all globals, and shading any heap pointers in
48 // off-heap runtime data structures. Scanning a stack stops a
49 // goroutine, shades any pointers found on its stack, and then
50 // resumes the goroutine.
52 // d. GC drains the work queue of grey objects, scanning each grey
53 // object to black and shading all pointers found in the object
54 // (which in turn may add those pointers to the work queue).
56 // e. Because GC work is spread across local caches, GC uses a
57 // distributed termination algorithm to detect when there are no
58 // more root marking jobs or grey objects (see gcMarkDone). At this
59 // point, GC transitions to mark termination.
61 // 3. GC performs mark termination.
65 // b. Set gcphase to _GCmarktermination, and disable workers and
68 // c. Perform housekeeping like flushing mcaches.
70 // 4. GC performs the sweep phase.
72 // a. Prepare for the sweep phase by setting gcphase to _GCoff,
73 // setting up sweep state and disabling the write barrier.
75 // b. Start the world. From this point on, newly allocated objects
76 // are white, and allocating sweeps spans before use if necessary.
78 // c. GC does concurrent sweeping in the background and in response
79 // to allocation. See description below.
81 // 5. When sufficient allocation has taken place, replay the sequence
82 // starting with 1 above. See discussion of GC rate below.
86 // The sweep phase proceeds concurrently with normal program execution.
87 // The heap is swept span-by-span both lazily (when a goroutine needs another span)
88 // and concurrently in a background goroutine (this helps programs that are not CPU bound).
89 // At the end of STW mark termination all spans are marked as "needs sweeping".
91 // The background sweeper goroutine simply sweeps spans one-by-one.
93 // To avoid requesting more OS memory while there are unswept spans, when a
94 // goroutine needs another span, it first attempts to reclaim that much memory
95 // by sweeping. When a goroutine needs to allocate a new small-object span, it
96 // sweeps small-object spans for the same object size until it frees at least
97 // one object. When a goroutine needs to allocate large-object span from heap,
98 // it sweeps spans until it frees at least that many pages into heap. There is
99 // one case where this may not suffice: if a goroutine sweeps and frees two
100 // nonadjacent one-page spans to the heap, it will allocate a new two-page
101 // span, but there can still be other one-page unswept spans which could be
102 // combined into a two-page span.
104 // It's critical to ensure that no operations proceed on unswept spans (that would corrupt
105 // mark bits in GC bitmap). During GC all mcaches are flushed into the central cache,
106 // so they are empty. When a goroutine grabs a new span into mcache, it sweeps it.
107 // When a goroutine explicitly frees an object or sets a finalizer, it ensures that
108 // the span is swept (either by sweeping it, or by waiting for the concurrent sweep to finish).
109 // The finalizer goroutine is kicked off only when all spans are swept.
110 // When the next GC starts, it sweeps all not-yet-swept spans (if any).
113 // Next GC is after we've allocated an extra amount of memory proportional to
114 // the amount already in use. The proportion is controlled by GOGC environment variable
115 // (100 by default). If GOGC=100 and we're using 4M, we'll GC again when we get to 8M
116 // (this mark is tracked in gcController.heapGoal variable). This keeps the GC cost in
117 // linear proportion to the allocation cost. Adjusting GOGC just changes the linear constant
118 // (and also the amount of extra memory used).
122 // In order to prevent long pauses while scanning large objects and to
123 // improve parallelism, the garbage collector breaks up scan jobs for
124 // objects larger than maxObletBytes into "oblets" of at most
125 // maxObletBytes. When scanning encounters the beginning of a large
126 // object, it scans only the first oblet and enqueues the remaining
127 // oblets as new scan jobs.
133 "runtime/internal/atomic"
139 _ConcurrentSweep = true
140 _FinBlockSize = 4 * 1024
142 // debugScanConservative enables debug logging for stack
143 // frames that are scanned conservatively.
144 debugScanConservative = false
146 // sweepMinHeapDistance is a lower bound on the heap distance
147 // (in bytes) reserved for concurrent sweeping between GC
149 sweepMinHeapDistance = 1024 * 1024
153 if unsafe.Sizeof(workbuf{}) != _WorkbufSize {
154 throw("size of Workbuf is suboptimal")
156 // No sweep on the first cycle.
157 sweep.active.state.Store(sweepDrainedMask)
159 // Initialize GC pacer state.
160 // Use the environment variable GOGC for the initial gcPercent value.
161 gcController.init(readGOGC())
164 work.markDoneSema = 1
165 lockInit(&work.sweepWaiters.lock, lockRankSweepWaiters)
166 lockInit(&work.assistQueue.lock, lockRankAssistQueue)
167 lockInit(&work.wbufSpans.lock, lockRankWbufSpans)
170 // gcenable is called after the bulk of the runtime initialization,
171 // just before we're about to start letting user code run.
172 // It kicks off the background sweeper goroutine, the background
173 // scavenger goroutine, and enables GC.
175 // Kick off sweeping and scavenging.
176 c := make(chan int, 2)
181 memstats.enablegc = true // now that runtime is initialized, GC is okay
184 // Garbage collector phase.
185 // Indicates to write barrier and synchronization task to perform.
188 // The compiler knows about this variable.
189 // If you change it, you must change builtin/runtime.go, too.
190 // If you change the first four bytes, you must also change the write
191 // barrier insertion code.
192 var writeBarrier struct {
193 enabled bool // compiler emits a check of this before calling write barrier
194 pad [3]byte // compiler uses 32-bit load for "enabled" field
195 needed bool // whether we need a write barrier for current GC phase
196 cgo bool // whether we need a write barrier for a cgo check
197 alignme uint64 // guarantee alignment so that compiler can use a 32 or 64-bit load
200 // gcBlackenEnabled is 1 if mutator assists and background mark
201 // workers are allowed to blacken objects. This must only be set when
202 // gcphase == _GCmark.
203 var gcBlackenEnabled uint32
206 _GCoff = iota // GC not running; sweeping in background, write barrier disabled
207 _GCmark // GC marking roots and workbufs: allocate black, write barrier ENABLED
208 _GCmarktermination // GC mark termination: allocate black, P's help GC, write barrier ENABLED
212 func setGCPhase(x uint32) {
213 atomic.Store(&gcphase, x)
214 writeBarrier.needed = gcphase == _GCmark || gcphase == _GCmarktermination
215 writeBarrier.enabled = writeBarrier.needed || writeBarrier.cgo
218 // gcMarkWorkerMode represents the mode that a concurrent mark worker
219 // should operate in.
221 // Concurrent marking happens through four different mechanisms. One
222 // is mutator assists, which happen in response to allocations and are
223 // not scheduled. The other three are variations in the per-P mark
224 // workers and are distinguished by gcMarkWorkerMode.
225 type gcMarkWorkerMode int
228 // gcMarkWorkerNotWorker indicates that the next scheduled G is not
229 // starting work and the mode should be ignored.
230 gcMarkWorkerNotWorker gcMarkWorkerMode = iota
232 // gcMarkWorkerDedicatedMode indicates that the P of a mark
233 // worker is dedicated to running that mark worker. The mark
234 // worker should run without preemption.
235 gcMarkWorkerDedicatedMode
237 // gcMarkWorkerFractionalMode indicates that a P is currently
238 // running the "fractional" mark worker. The fractional worker
239 // is necessary when GOMAXPROCS*gcBackgroundUtilization is not
240 // an integer and using only dedicated workers would result in
241 // utilization too far from the target of gcBackgroundUtilization.
242 // The fractional worker should run until it is preempted and
243 // will be scheduled to pick up the fractional part of
244 // GOMAXPROCS*gcBackgroundUtilization.
245 gcMarkWorkerFractionalMode
247 // gcMarkWorkerIdleMode indicates that a P is running the mark
248 // worker because it has nothing else to do. The idle worker
249 // should run until it is preempted and account its time
250 // against gcController.idleMarkTime.
254 // gcMarkWorkerModeStrings are the strings labels of gcMarkWorkerModes
255 // to use in execution traces.
256 var gcMarkWorkerModeStrings = [...]string{
263 // pollFractionalWorkerExit reports whether a fractional mark worker
264 // should self-preempt. It assumes it is called from the fractional
266 func pollFractionalWorkerExit() bool {
267 // This should be kept in sync with the fractional worker
268 // scheduler logic in findRunnableGCWorker.
270 delta := now - gcController.markStartTime
274 p := getg().m.p.ptr()
275 selfTime := p.gcFractionalMarkTime + (now - p.gcMarkWorkerStartTime)
276 // Add some slack to the utilization goal so that the
277 // fractional worker isn't behind again the instant it exits.
278 return float64(selfTime)/float64(delta) > 1.2*gcController.fractionalUtilizationGoal
282 full lfstack // lock-free list of full blocks workbuf
283 empty lfstack // lock-free list of empty blocks workbuf
284 pad0 cpu.CacheLinePad // prevents false-sharing between full/empty and nproc/nwait
288 // free is a list of spans dedicated to workbufs, but
289 // that don't currently contain any workbufs.
291 // busy is a list of all spans containing workbufs on
292 // one of the workbuf lists.
296 // Restore 64-bit alignment on 32-bit.
299 // bytesMarked is the number of bytes marked this cycle. This
300 // includes bytes blackened in scanned objects, noscan objects
301 // that go straight to black, and permagrey objects scanned by
302 // markroot during the concurrent scan phase. This is updated
303 // atomically during the cycle. Updates may be batched
304 // arbitrarily, since the value is only read at the end of the
307 // Because of benign races during marking, this number may not
308 // be the exact number of marked bytes, but it should be very
311 // Put this field here because it needs 64-bit atomic access
312 // (and thus 8-byte alignment even on 32-bit architectures).
315 markrootNext uint32 // next markroot job
316 markrootJobs uint32 // number of markroot jobs
322 // Number of roots of various root types. Set by gcMarkRootPrepare.
323 nDataRoots, nBSSRoots, nSpanRoots, nStackRoots int
325 // Base indexes of each root type. Set by gcMarkRootPrepare.
326 baseData, baseBSS, baseSpans, baseStacks, baseEnd uint32
328 // Each type of GC state transition is protected by a lock.
329 // Since multiple threads can simultaneously detect the state
330 // transition condition, any thread that detects a transition
331 // condition must acquire the appropriate transition lock,
332 // re-check the transition condition and return if it no
333 // longer holds or perform the transition if it does.
334 // Likewise, any transition must invalidate the transition
335 // condition before releasing the lock. This ensures that each
336 // transition is performed by exactly one thread and threads
337 // that need the transition to happen block until it has
340 // startSema protects the transition from "off" to mark or
343 // markDoneSema protects transitions from mark to mark termination.
346 bgMarkReady note // signal background mark worker has started
347 bgMarkDone uint32 // cas to 1 when at a background mark completion point
348 // Background mark completion signaling
350 // mode is the concurrency mode of the current GC cycle.
353 // userForced indicates the current GC cycle was forced by an
354 // explicit user call.
357 // totaltime is the CPU nanoseconds spent in GC since the
358 // program started if debug.gctrace > 0.
361 // initialHeapLive is the value of gcController.heapLive at the
362 // beginning of this GC cycle.
363 initialHeapLive uint64
365 // assistQueue is a queue of assists that are blocked because
366 // there was neither enough credit to steal or enough work to
373 // sweepWaiters is a list of blocked goroutines to wake when
374 // we transition from mark termination to sweep.
375 sweepWaiters struct {
380 // cycles is the number of completed GC cycles, where a GC
381 // cycle is sweep termination, mark, mark termination, and
382 // sweep. This differs from memstats.numgc, which is
383 // incremented at mark termination.
386 // Timing/utilization stats for this cycle.
387 stwprocs, maxprocs int32
388 tSweepTerm, tMark, tMarkTerm, tEnd int64 // nanotime() of phase start
390 pauseNS int64 // total STW time this cycle
391 pauseStart int64 // nanotime() of last STW
393 // debug.gctrace heap sizes for this cycle.
394 heap0, heap1, heap2, heapGoal uint64
397 // GC runs a garbage collection and blocks the caller until the
398 // garbage collection is complete. It may also block the entire
401 // We consider a cycle to be: sweep termination, mark, mark
402 // termination, and sweep. This function shouldn't return
403 // until a full cycle has been completed, from beginning to
404 // end. Hence, we always want to finish up the current cycle
405 // and start a new one. That means:
407 // 1. In sweep termination, mark, or mark termination of cycle
408 // N, wait until mark termination N completes and transitions
411 // 2. In sweep N, help with sweep N.
413 // At this point we can begin a full cycle N+1.
415 // 3. Trigger cycle N+1 by starting sweep termination N+1.
417 // 4. Wait for mark termination N+1 to complete.
419 // 5. Help with sweep N+1 until it's done.
421 // This all has to be written to deal with the fact that the
422 // GC may move ahead on its own. For example, when we block
423 // until mark termination N, we may wake up in cycle N+2.
425 // Wait until the current sweep termination, mark, and mark
426 // termination complete.
427 n := atomic.Load(&work.cycles)
430 // We're now in sweep N or later. Trigger GC cycle N+1, which
431 // will first finish sweep N if necessary and then enter sweep
433 gcStart(gcTrigger{kind: gcTriggerCycle, n: n + 1})
435 // Wait for mark termination N+1 to complete.
438 // Finish sweep N+1 before returning. We do this both to
439 // complete the cycle and because runtime.GC() is often used
440 // as part of tests and benchmarks to get the system into a
441 // relatively stable and isolated state.
442 for atomic.Load(&work.cycles) == n+1 && sweepone() != ^uintptr(0) {
447 // Callers may assume that the heap profile reflects the
448 // just-completed cycle when this returns (historically this
449 // happened because this was a STW GC), but right now the
450 // profile still reflects mark termination N, not N+1.
452 // As soon as all of the sweep frees from cycle N+1 are done,
453 // we can go ahead and publish the heap profile.
455 // First, wait for sweeping to finish. (We know there are no
456 // more spans on the sweep queue, but we may be concurrently
457 // sweeping spans, so we have to wait.)
458 for atomic.Load(&work.cycles) == n+1 && !isSweepDone() {
462 // Now we're really done with sweeping, so we can publish the
463 // stable heap profile. Only do this if we haven't already hit
464 // another mark termination.
466 cycle := atomic.Load(&work.cycles)
467 if cycle == n+1 || (gcphase == _GCmark && cycle == n+2) {
473 // gcWaitOnMark blocks until GC finishes the Nth mark phase. If GC has
474 // already completed this mark phase, it returns immediately.
475 func gcWaitOnMark(n uint32) {
477 // Disable phase transitions.
478 lock(&work.sweepWaiters.lock)
479 nMarks := atomic.Load(&work.cycles)
480 if gcphase != _GCmark {
481 // We've already completed this cycle's mark.
486 unlock(&work.sweepWaiters.lock)
490 // Wait until sweep termination, mark, and mark
491 // termination of cycle N complete.
492 work.sweepWaiters.list.push(getg())
493 goparkunlock(&work.sweepWaiters.lock, waitReasonWaitForGCCycle, traceEvGoBlock, 1)
497 // gcMode indicates how concurrent a GC cycle should be.
501 gcBackgroundMode gcMode = iota // concurrent GC and sweep
502 gcForceMode // stop-the-world GC now, concurrent sweep
503 gcForceBlockMode // stop-the-world GC now and STW sweep (forced by user)
506 // A gcTrigger is a predicate for starting a GC cycle. Specifically,
507 // it is an exit condition for the _GCoff phase.
508 type gcTrigger struct {
510 now int64 // gcTriggerTime: current time
511 n uint32 // gcTriggerCycle: cycle number to start
514 type gcTriggerKind int
517 // gcTriggerHeap indicates that a cycle should be started when
518 // the heap size reaches the trigger heap size computed by the
520 gcTriggerHeap gcTriggerKind = iota
522 // gcTriggerTime indicates that a cycle should be started when
523 // it's been more than forcegcperiod nanoseconds since the
524 // previous GC cycle.
527 // gcTriggerCycle indicates that a cycle should be started if
528 // we have not yet started cycle number gcTrigger.n (relative
533 // test reports whether the trigger condition is satisfied, meaning
534 // that the exit condition for the _GCoff phase has been met. The exit
535 // condition should be tested when allocating.
536 func (t gcTrigger) test() bool {
537 if !memstats.enablegc || panicking != 0 || gcphase != _GCoff {
542 // Non-atomic access to gcController.heapLive for performance. If
543 // we are going to trigger on this, this thread just
544 // atomically wrote gcController.heapLive anyway and we'll see our
546 return gcController.heapLive >= gcController.trigger
548 if atomic.Loadint32(&gcController.gcPercent) < 0 {
551 lastgc := int64(atomic.Load64(&memstats.last_gc_nanotime))
552 return lastgc != 0 && t.now-lastgc > forcegcperiod
554 // t.n > work.cycles, but accounting for wraparound.
555 return int32(t.n-work.cycles) > 0
560 // gcStart starts the GC. It transitions from _GCoff to _GCmark (if
561 // debug.gcstoptheworld == 0) or performs all of GC (if
562 // debug.gcstoptheworld != 0).
564 // This may return without performing this transition in some cases,
565 // such as when called on a system stack or with locks held.
566 func gcStart(trigger gcTrigger) {
567 // Since this is called from malloc and malloc is called in
568 // the guts of a number of libraries that might be holding
569 // locks, don't attempt to start GC in non-preemptible or
570 // potentially unstable situations.
572 if gp := getg(); gp == mp.g0 || mp.locks > 1 || mp.preemptoff != "" {
579 // Pick up the remaining unswept/not being swept spans concurrently
581 // This shouldn't happen if we're being invoked in background
582 // mode since proportional sweep should have just finished
583 // sweeping everything, but rounding errors, etc, may leave a
584 // few spans unswept. In forced mode, this is necessary since
585 // GC can be forced at any point in the sweeping cycle.
587 // We check the transition condition continuously here in case
588 // this G gets delayed in to the next GC cycle.
589 for trigger.test() && sweepone() != ^uintptr(0) {
593 // Perform GC initialization and the sweep termination
595 semacquire(&work.startSema)
596 // Re-check transition condition under transition lock.
598 semrelease(&work.startSema)
602 // For stats, check if this GC was forced by the user.
603 work.userForced = trigger.kind == gcTriggerCycle
605 // In gcstoptheworld debug mode, upgrade the mode accordingly.
606 // We do this after re-checking the transition condition so
607 // that multiple goroutines that detect the heap trigger don't
608 // start multiple STW GCs.
609 mode := gcBackgroundMode
610 if debug.gcstoptheworld == 1 {
612 } else if debug.gcstoptheworld == 2 {
613 mode = gcForceBlockMode
616 // Ok, we're doing it! Stop everybody else
618 semacquire(&worldsema)
624 // Check that all Ps have finished deferred mcache flushes.
625 for _, p := range allp {
626 if fg := atomic.Load(&p.mcache.flushGen); fg != mheap_.sweepgen {
627 println("runtime: p", p.id, "flushGen", fg, "!= sweepgen", mheap_.sweepgen)
628 throw("p mcache not flushed")
632 gcBgMarkStartWorkers()
634 systemstack(gcResetMarkState)
636 work.stwprocs, work.maxprocs = gomaxprocs, gomaxprocs
637 if work.stwprocs > ncpu {
638 // This is used to compute CPU time of the STW phases,
639 // so it can't be more than ncpu, even if GOMAXPROCS is.
642 work.heap0 = atomic.Load64(&gcController.heapLive)
647 work.tSweepTerm = now
648 work.pauseStart = now
652 systemstack(stopTheWorldWithSema)
653 // Finish sweep before we start concurrent scan.
658 // clearpools before we start the GC. If we wait they memory will not be
659 // reclaimed until the next GC cycle.
664 // Assists and workers can start the moment we start
666 gcController.startCycle(now)
667 work.heapGoal = gcController.heapGoal
669 // In STW mode, disable scheduling of user Gs. This may also
670 // disable scheduling of this goroutine, so it may block as
671 // soon as we start the world again.
672 if mode != gcBackgroundMode {
673 schedEnableUser(false)
676 // Enter concurrent mark phase and enable
679 // Because the world is stopped, all Ps will
680 // observe that write barriers are enabled by
681 // the time we start the world and begin
684 // Write barriers must be enabled before assists are
685 // enabled because they must be enabled before
686 // any non-leaf heap objects are marked. Since
687 // allocations are blocked until assists can
688 // happen, we want enable assists as early as
692 gcBgMarkPrepare() // Must happen before assist enable.
695 // Mark all active tinyalloc blocks. Since we're
696 // allocating from these, they need to be black like
697 // other allocations. The alternative is to blacken
698 // the tiny block on every allocation from it, which
699 // would slow down the tiny allocator.
702 // At this point all Ps have enabled the write
703 // barrier, thus maintaining the no white to
704 // black invariant. Enable mutator assists to
705 // put back-pressure on fast allocating
707 atomic.Store(&gcBlackenEnabled, 1)
709 // In STW mode, we could block the instant systemstack
710 // returns, so make sure we're not preemptible.
715 now = startTheWorldWithSema(trace.enabled)
716 work.pauseNS += now - work.pauseStart
718 memstats.gcPauseDist.record(now - work.pauseStart)
721 // Release the world sema before Gosched() in STW mode
722 // because we will need to reacquire it later but before
723 // this goroutine becomes runnable again, and we could
724 // self-deadlock otherwise.
725 semrelease(&worldsema)
728 // Make sure we block instead of returning to user code
730 if mode != gcBackgroundMode {
734 semrelease(&work.startSema)
737 // gcMarkDoneFlushed counts the number of P's with flushed work.
739 // Ideally this would be a captured local in gcMarkDone, but forEachP
740 // escapes its callback closure, so it can't capture anything.
742 // This is protected by markDoneSema.
743 var gcMarkDoneFlushed uint32
745 // gcMarkDone transitions the GC from mark to mark termination if all
746 // reachable objects have been marked (that is, there are no grey
747 // objects and can be no more in the future). Otherwise, it flushes
748 // all local work to the global queues where it can be discovered by
751 // This should be called when all local mark work has been drained and
752 // there are no remaining workers. Specifically, when
754 // work.nwait == work.nproc && !gcMarkWorkAvailable(p)
756 // The calling context must be preemptible.
758 // Flushing local work is important because idle Ps may have local
759 // work queued. This is the only way to make that work visible and
760 // drive GC to completion.
762 // It is explicitly okay to have write barriers in this function. If
763 // it does transition to mark termination, then all reachable objects
764 // have been marked, so the write barrier cannot shade any more
767 // Ensure only one thread is running the ragged barrier at a
769 semacquire(&work.markDoneSema)
772 // Re-check transition condition under transition lock.
774 // It's critical that this checks the global work queues are
775 // empty before performing the ragged barrier. Otherwise,
776 // there could be global work that a P could take after the P
777 // has passed the ragged barrier.
778 if !(gcphase == _GCmark && work.nwait == work.nproc && !gcMarkWorkAvailable(nil)) {
779 semrelease(&work.markDoneSema)
783 // forEachP needs worldsema to execute, and we'll need it to
784 // stop the world later, so acquire worldsema now.
785 semacquire(&worldsema)
787 // Flush all local buffers and collect flushedWork flags.
788 gcMarkDoneFlushed = 0
791 // Mark the user stack as preemptible so that it may be scanned.
792 // Otherwise, our attempt to force all P's to a safepoint could
793 // result in a deadlock as we attempt to preempt a worker that's
794 // trying to preempt us (e.g. for a stack scan).
795 casgstatus(gp, _Grunning, _Gwaiting)
796 forEachP(func(_p_ *p) {
797 // Flush the write barrier buffer, since this may add
798 // work to the gcWork.
801 // Flush the gcWork, since this may create global work
802 // and set the flushedWork flag.
804 // TODO(austin): Break up these workbufs to
805 // better distribute work.
807 // Collect the flushedWork flag.
808 if _p_.gcw.flushedWork {
809 atomic.Xadd(&gcMarkDoneFlushed, 1)
810 _p_.gcw.flushedWork = false
813 casgstatus(gp, _Gwaiting, _Grunning)
816 if gcMarkDoneFlushed != 0 {
817 // More grey objects were discovered since the
818 // previous termination check, so there may be more
819 // work to do. Keep going. It's possible the
820 // transition condition became true again during the
821 // ragged barrier, so re-check it.
822 semrelease(&worldsema)
826 // There was no global work, no local work, and no Ps
827 // communicated work since we took markDoneSema. Therefore
828 // there are no grey objects and no more objects can be
829 // shaded. Transition to mark termination.
832 work.pauseStart = now
833 getg().m.preemptoff = "gcing"
837 systemstack(stopTheWorldWithSema)
838 // The gcphase is _GCmark, it will transition to _GCmarktermination
839 // below. The important thing is that the wb remains active until
840 // all marking is complete. This includes writes made by the GC.
842 // There is sometimes work left over when we enter mark termination due
843 // to write barriers performed after the completion barrier above.
844 // Detect this and resume concurrent mark. This is obviously
847 // See issue #27993 for details.
849 // Switch to the system stack to call wbBufFlush1, though in this case
850 // it doesn't matter because we're non-preemptible anyway.
853 for _, p := range allp {
862 getg().m.preemptoff = ""
864 now := startTheWorldWithSema(true)
865 work.pauseNS += now - work.pauseStart
866 memstats.gcPauseDist.record(now - work.pauseStart)
868 semrelease(&worldsema)
872 // Disable assists and background workers. We must do
873 // this before waking blocked assists.
874 atomic.Store(&gcBlackenEnabled, 0)
876 // Wake all blocked assists. These will run when we
877 // start the world again.
880 // Likewise, release the transition lock. Blocked
881 // workers and assists will run when we start the
883 semrelease(&work.markDoneSema)
885 // In STW mode, re-enable user goroutines. These will be
886 // queued to run after we start the world.
887 schedEnableUser(true)
889 // endCycle depends on all gcWork cache stats being flushed.
890 // The termination algorithm above ensured that up to
891 // allocations since the ragged barrier.
892 nextTriggerRatio := gcController.endCycle(work.userForced)
894 // Perform mark termination. This will restart the world.
895 gcMarkTermination(nextTriggerRatio)
898 // World must be stopped and mark assists and background workers must be
900 func gcMarkTermination(nextTriggerRatio float64) {
901 // Start marktermination (write barrier remains enabled for now).
902 setGCPhase(_GCmarktermination)
904 work.heap1 = gcController.heapLive
905 startTime := nanotime()
908 mp.preemptoff = "gcing"
912 casgstatus(gp, _Grunning, _Gwaiting)
913 gp.waitreason = waitReasonGarbageCollection
915 // Run gc on the g0 stack. We do this so that the g stack
916 // we're currently running on will no longer change. Cuts
917 // the root set down a bit (g0 stacks are not scanned, and
918 // we don't need to scan gc's internal state). We also
919 // need to switch to g0 so we can shrink the stack.
922 // Must return immediately.
923 // The outer function's stack may have moved
924 // during gcMark (it shrinks stacks, including the
925 // outer function's stack), so we must not refer
926 // to any of its variables. Return back to the
927 // non-system stack to pick up the new addresses
928 // before continuing.
932 work.heap2 = work.bytesMarked
933 if debug.gccheckmark > 0 {
934 // Run a full non-parallel, stop-the-world
935 // mark using checkmark bits, to check that we
936 // didn't forget to mark anything during the
937 // concurrent mark process.
940 gcw := &getg().m.p.ptr().gcw
942 wbBufFlush1(getg().m.p.ptr())
947 // marking is complete so we can turn the write barrier off
953 casgstatus(gp, _Gwaiting, _Grunning)
962 if gcphase != _GCoff {
963 throw("gc done but gcphase != _GCoff")
966 // Record heap_inuse for scavenger.
967 memstats.last_heap_inuse = memstats.heap_inuse
969 // Update GC trigger and pacing for the next cycle.
970 gcController.commit(nextTriggerRatio)
971 gcPaceSweeper(gcController.trigger)
972 gcPaceScavenger(gcController.heapGoal, gcController.lastHeapGoal)
974 // Update timing memstats
976 sec, nsec, _ := time_now()
977 unixNow := sec*1e9 + int64(nsec)
978 work.pauseNS += now - work.pauseStart
980 memstats.gcPauseDist.record(now - work.pauseStart)
981 atomic.Store64(&memstats.last_gc_unix, uint64(unixNow)) // must be Unix time to make sense to user
982 atomic.Store64(&memstats.last_gc_nanotime, uint64(now)) // monotonic time for us
983 memstats.pause_ns[memstats.numgc%uint32(len(memstats.pause_ns))] = uint64(work.pauseNS)
984 memstats.pause_end[memstats.numgc%uint32(len(memstats.pause_end))] = uint64(unixNow)
985 memstats.pause_total_ns += uint64(work.pauseNS)
987 // Update work.totaltime.
988 sweepTermCpu := int64(work.stwprocs) * (work.tMark - work.tSweepTerm)
989 // We report idle marking time below, but omit it from the
990 // overall utilization here since it's "free".
991 markCpu := gcController.assistTime + gcController.dedicatedMarkTime + gcController.fractionalMarkTime
992 markTermCpu := int64(work.stwprocs) * (work.tEnd - work.tMarkTerm)
993 cycleCpu := sweepTermCpu + markCpu + markTermCpu
994 work.totaltime += cycleCpu
996 // Compute overall GC CPU utilization.
997 totalCpu := sched.totaltime + (now-sched.procresizetime)*int64(gomaxprocs)
998 memstats.gc_cpu_fraction = float64(work.totaltime) / float64(totalCpu)
1000 // Reset sweep state.
1002 sweep.npausesweep = 0
1004 if work.userForced {
1005 memstats.numforcedgc++
1008 // Bump GC cycle count and wake goroutines waiting on sweep.
1009 lock(&work.sweepWaiters.lock)
1011 injectglist(&work.sweepWaiters.list)
1012 unlock(&work.sweepWaiters.lock)
1014 // Finish the current heap profiling cycle and start a new
1015 // heap profiling cycle. We do this before starting the world
1016 // so events don't leak into the wrong cycle.
1019 // There may be stale spans in mcaches that need to be swept.
1020 // Those aren't tracked in any sweep lists, so we need to
1021 // count them against sweep completion until we ensure all
1022 // those spans have been forced out.
1023 sl := sweep.active.begin()
1025 throw("failed to set sweep barrier")
1028 systemstack(func() { startTheWorldWithSema(true) })
1030 // Flush the heap profile so we can start a new cycle next GC.
1031 // This is relatively expensive, so we don't do it with the
1035 // Prepare workbufs for freeing by the sweeper. We do this
1036 // asynchronously because it can take non-trivial time.
1037 prepareFreeWorkbufs()
1039 // Free stack spans. This must be done between GC cycles.
1040 systemstack(freeStackSpans)
1042 // Ensure all mcaches are flushed. Each P will flush its own
1043 // mcache before allocating, but idle Ps may not. Since this
1044 // is necessary to sweep all spans, we need to ensure all
1045 // mcaches are flushed before we start the next GC cycle.
1046 systemstack(func() {
1047 forEachP(func(_p_ *p) {
1048 _p_.mcache.prepareForSweep()
1051 // Now that we've swept stale spans in mcaches, they don't
1052 // count against unswept spans.
1053 sweep.active.end(sl)
1055 // Print gctrace before dropping worldsema. As soon as we drop
1056 // worldsema another cycle could start and smash the stats
1057 // we're trying to print.
1058 if debug.gctrace > 0 {
1059 util := int(memstats.gc_cpu_fraction * 100)
1063 print("gc ", memstats.numgc,
1064 " @", string(itoaDiv(sbuf[:], uint64(work.tSweepTerm-runtimeInitTime)/1e6, 3)), "s ",
1066 prev := work.tSweepTerm
1067 for i, ns := range []int64{work.tMark, work.tMarkTerm, work.tEnd} {
1071 print(string(fmtNSAsMS(sbuf[:], uint64(ns-prev))))
1074 print(" ms clock, ")
1075 for i, ns := range []int64{sweepTermCpu, gcController.assistTime, gcController.dedicatedMarkTime + gcController.fractionalMarkTime, gcController.idleMarkTime, markTermCpu} {
1076 if i == 2 || i == 3 {
1077 // Separate mark time components with /.
1082 print(string(fmtNSAsMS(sbuf[:], uint64(ns))))
1085 work.heap0>>20, "->", work.heap1>>20, "->", work.heap2>>20, " MB, ",
1086 work.heapGoal>>20, " MB goal, ",
1087 work.maxprocs, " P")
1088 if work.userForced {
1095 semrelease(&worldsema)
1097 // Careful: another GC cycle may start now.
1102 // now that gc is done, kick off finalizer thread if needed
1103 if !concurrentSweep {
1104 // give the queued finalizers, if any, a chance to run
1109 // gcBgMarkStartWorkers prepares background mark worker goroutines. These
1110 // goroutines will not run until the mark phase, but they must be started while
1111 // the work is not stopped and from a regular G stack. The caller must hold
1113 func gcBgMarkStartWorkers() {
1114 // Background marking is performed by per-P G's. Ensure that each P has
1115 // a background GC G.
1117 // Worker Gs don't exit if gomaxprocs is reduced. If it is raised
1118 // again, we can reuse the old workers; no need to create new workers.
1119 for gcBgMarkWorkerCount < gomaxprocs {
1122 notetsleepg(&work.bgMarkReady, -1)
1123 noteclear(&work.bgMarkReady)
1124 // The worker is now guaranteed to be added to the pool before
1125 // its P's next findRunnableGCWorker.
1127 gcBgMarkWorkerCount++
1131 // gcBgMarkPrepare sets up state for background marking.
1132 // Mutator assists must not yet be enabled.
1133 func gcBgMarkPrepare() {
1134 // Background marking will stop when the work queues are empty
1135 // and there are no more workers (note that, since this is
1136 // concurrent, this may be a transient state, but mark
1137 // termination will clean it up). Between background workers
1138 // and assists, we don't really know how many workers there
1139 // will be, so we pretend to have an arbitrarily large number
1140 // of workers, almost all of which are "waiting". While a
1141 // worker is working it decrements nwait. If nproc == nwait,
1142 // there are no workers.
1143 work.nproc = ^uint32(0)
1144 work.nwait = ^uint32(0)
1147 // gcBgMarkWorker is an entry in the gcBgMarkWorkerPool. It points to a single
1148 // gcBgMarkWorker goroutine.
1149 type gcBgMarkWorkerNode struct {
1150 // Unused workers are managed in a lock-free stack. This field must be first.
1153 // The g of this worker.
1156 // Release this m on park. This is used to communicate with the unlock
1157 // function, which cannot access the G's stack. It is unused outside of
1158 // gcBgMarkWorker().
1162 func gcBgMarkWorker() {
1165 // We pass node to a gopark unlock function, so it can't be on
1166 // the stack (see gopark). Prevent deadlock from recursively
1167 // starting GC by disabling preemption.
1168 gp.m.preemptoff = "GC worker init"
1169 node := new(gcBgMarkWorkerNode)
1170 gp.m.preemptoff = ""
1174 node.m.set(acquirem())
1175 notewakeup(&work.bgMarkReady)
1176 // After this point, the background mark worker is generally scheduled
1177 // cooperatively by gcController.findRunnableGCWorker. While performing
1178 // work on the P, preemption is disabled because we are working on
1179 // P-local work buffers. When the preempt flag is set, this puts itself
1180 // into _Gwaiting to be woken up by gcController.findRunnableGCWorker
1181 // at the appropriate time.
1183 // When preemption is enabled (e.g., while in gcMarkDone), this worker
1184 // may be preempted and schedule as a _Grunnable G from a runq. That is
1185 // fine; it will eventually gopark again for further scheduling via
1186 // findRunnableGCWorker.
1188 // Since we disable preemption before notifying bgMarkReady, we
1189 // guarantee that this G will be in the worker pool for the next
1190 // findRunnableGCWorker. This isn't strictly necessary, but it reduces
1191 // latency between _GCmark starting and the workers starting.
1194 // Go to sleep until woken by
1195 // gcController.findRunnableGCWorker.
1196 gopark(func(g *g, nodep unsafe.Pointer) bool {
1197 node := (*gcBgMarkWorkerNode)(nodep)
1199 if mp := node.m.ptr(); mp != nil {
1200 // The worker G is no longer running; release
1203 // N.B. it is _safe_ to release the M as soon
1204 // as we are no longer performing P-local mark
1207 // However, since we cooperatively stop work
1208 // when gp.preempt is set, if we releasem in
1209 // the loop then the following call to gopark
1210 // would immediately preempt the G. This is
1211 // also safe, but inefficient: the G must
1212 // schedule again only to enter gopark and park
1213 // again. Thus, we defer the release until
1214 // after parking the G.
1218 // Release this G to the pool.
1219 gcBgMarkWorkerPool.push(&node.node)
1220 // Note that at this point, the G may immediately be
1221 // rescheduled and may be running.
1223 }, unsafe.Pointer(node), waitReasonGCWorkerIdle, traceEvGoBlock, 0)
1225 // Preemption must not occur here, or another G might see
1226 // p.gcMarkWorkerMode.
1228 // Disable preemption so we can use the gcw. If the
1229 // scheduler wants to preempt us, we'll stop draining,
1230 // dispose the gcw, and then preempt.
1231 node.m.set(acquirem())
1232 pp := gp.m.p.ptr() // P can't change with preemption disabled.
1234 if gcBlackenEnabled == 0 {
1235 println("worker mode", pp.gcMarkWorkerMode)
1236 throw("gcBgMarkWorker: blackening not enabled")
1239 if pp.gcMarkWorkerMode == gcMarkWorkerNotWorker {
1240 throw("gcBgMarkWorker: mode not set")
1243 startTime := nanotime()
1244 pp.gcMarkWorkerStartTime = startTime
1246 decnwait := atomic.Xadd(&work.nwait, -1)
1247 if decnwait == work.nproc {
1248 println("runtime: work.nwait=", decnwait, "work.nproc=", work.nproc)
1249 throw("work.nwait was > work.nproc")
1252 systemstack(func() {
1253 // Mark our goroutine preemptible so its stack
1254 // can be scanned. This lets two mark workers
1255 // scan each other (otherwise, they would
1256 // deadlock). We must not modify anything on
1257 // the G stack. However, stack shrinking is
1258 // disabled for mark workers, so it is safe to
1259 // read from the G stack.
1260 casgstatus(gp, _Grunning, _Gwaiting)
1261 switch pp.gcMarkWorkerMode {
1263 throw("gcBgMarkWorker: unexpected gcMarkWorkerMode")
1264 case gcMarkWorkerDedicatedMode:
1265 gcDrain(&pp.gcw, gcDrainUntilPreempt|gcDrainFlushBgCredit)
1267 // We were preempted. This is
1268 // a useful signal to kick
1269 // everything out of the run
1270 // queue so it can run
1272 if drainQ, n := runqdrain(pp); n > 0 {
1274 globrunqputbatch(&drainQ, int32(n))
1278 // Go back to draining, this time
1279 // without preemption.
1280 gcDrain(&pp.gcw, gcDrainFlushBgCredit)
1281 case gcMarkWorkerFractionalMode:
1282 gcDrain(&pp.gcw, gcDrainFractional|gcDrainUntilPreempt|gcDrainFlushBgCredit)
1283 case gcMarkWorkerIdleMode:
1284 gcDrain(&pp.gcw, gcDrainIdle|gcDrainUntilPreempt|gcDrainFlushBgCredit)
1286 casgstatus(gp, _Gwaiting, _Grunning)
1289 // Account for time.
1290 duration := nanotime() - startTime
1291 gcController.logWorkTime(pp.gcMarkWorkerMode, duration)
1292 if pp.gcMarkWorkerMode == gcMarkWorkerFractionalMode {
1293 atomic.Xaddint64(&pp.gcFractionalMarkTime, duration)
1296 // Was this the last worker and did we run out
1298 incnwait := atomic.Xadd(&work.nwait, +1)
1299 if incnwait > work.nproc {
1300 println("runtime: p.gcMarkWorkerMode=", pp.gcMarkWorkerMode,
1301 "work.nwait=", incnwait, "work.nproc=", work.nproc)
1302 throw("work.nwait > work.nproc")
1305 // We'll releasem after this point and thus this P may run
1306 // something else. We must clear the worker mode to avoid
1307 // attributing the mode to a different (non-worker) G in
1309 pp.gcMarkWorkerMode = gcMarkWorkerNotWorker
1311 // If this worker reached a background mark completion
1312 // point, signal the main GC goroutine.
1313 if incnwait == work.nproc && !gcMarkWorkAvailable(nil) {
1314 // We don't need the P-local buffers here, allow
1315 // preemption becuse we may schedule like a regular
1316 // goroutine in gcMarkDone (block on locks, etc).
1317 releasem(node.m.ptr())
1325 // gcMarkWorkAvailable reports whether executing a mark worker
1326 // on p is potentially useful. p may be nil, in which case it only
1327 // checks the global sources of work.
1328 func gcMarkWorkAvailable(p *p) bool {
1329 if p != nil && !p.gcw.empty() {
1332 if !work.full.empty() {
1333 return true // global work available
1335 if work.markrootNext < work.markrootJobs {
1336 return true // root scan work available
1341 // gcMark runs the mark (or, for concurrent GC, mark termination)
1342 // All gcWork caches must be empty.
1343 // STW is in effect at this point.
1344 func gcMark(startTime int64) {
1345 if debug.allocfreetrace > 0 {
1349 if gcphase != _GCmarktermination {
1350 throw("in gcMark expecting to see gcphase as _GCmarktermination")
1352 work.tstart = startTime
1354 // Check that there's no marking work remaining.
1355 if work.full != 0 || work.markrootNext < work.markrootJobs {
1356 print("runtime: full=", hex(work.full), " next=", work.markrootNext, " jobs=", work.markrootJobs, " nDataRoots=", work.nDataRoots, " nBSSRoots=", work.nBSSRoots, " nSpanRoots=", work.nSpanRoots, " nStackRoots=", work.nStackRoots, "\n")
1357 panic("non-empty mark queue after concurrent mark")
1360 if debug.gccheckmark > 0 {
1361 // This is expensive when there's a large number of
1362 // Gs, so only do it if checkmark is also enabled.
1366 throw("work.full != 0")
1369 // Clear out buffers and double-check that all gcWork caches
1370 // are empty. This should be ensured by gcMarkDone before we
1371 // enter mark termination.
1373 // TODO: We could clear out buffers just before mark if this
1374 // has a non-negligible impact on STW time.
1375 for _, p := range allp {
1376 // The write barrier may have buffered pointers since
1377 // the gcMarkDone barrier. However, since the barrier
1378 // ensured all reachable objects were marked, all of
1379 // these must be pointers to black objects. Hence we
1380 // can just discard the write barrier buffer.
1381 if debug.gccheckmark > 0 {
1382 // For debugging, flush the buffer and make
1383 // sure it really was all marked.
1392 print("runtime: P ", p.id, " flushedWork ", gcw.flushedWork)
1393 if gcw.wbuf1 == nil {
1394 print(" wbuf1=<nil>")
1396 print(" wbuf1.n=", gcw.wbuf1.nobj)
1398 if gcw.wbuf2 == nil {
1399 print(" wbuf2=<nil>")
1401 print(" wbuf2.n=", gcw.wbuf2.nobj)
1404 throw("P has cached GC work at end of mark termination")
1406 // There may still be cached empty buffers, which we
1407 // need to flush since we're going to free them. Also,
1408 // there may be non-zero stats because we allocated
1409 // black after the gcMarkDone barrier.
1413 // Flush scanAlloc from each mcache since we're about to modify
1414 // heapScan directly. If we were to flush this later, then scanAlloc
1415 // might have incorrect information.
1417 // Note that it's not important to retain this information; we know
1418 // exactly what heapScan is at this point via scanWork.
1419 for _, p := range allp {
1427 // Reset controller state.
1428 gcController.resetLive(work.bytesMarked)
1431 // gcSweep must be called on the system stack because it acquires the heap
1432 // lock. See mheap for details.
1434 // The world must be stopped.
1437 func gcSweep(mode gcMode) {
1438 assertWorldStopped()
1440 if gcphase != _GCoff {
1441 throw("gcSweep being done but phase is not GCoff")
1445 mheap_.sweepgen += 2
1446 sweep.active.reset()
1447 mheap_.pagesSwept.Store(0)
1448 mheap_.sweepArenas = mheap_.allArenas
1449 mheap_.reclaimIndex.Store(0)
1450 mheap_.reclaimCredit.Store(0)
1451 unlock(&mheap_.lock)
1453 sweep.centralIndex.clear()
1455 if !_ConcurrentSweep || mode == gcForceBlockMode {
1456 // Special case synchronous sweep.
1457 // Record that no proportional sweeping has to happen.
1459 mheap_.sweepPagesPerByte = 0
1460 unlock(&mheap_.lock)
1461 // Sweep all spans eagerly.
1462 for sweepone() != ^uintptr(0) {
1465 // Free workbufs eagerly.
1466 prepareFreeWorkbufs()
1467 for freeSomeWbufs(false) {
1469 // All "free" events for this mark/sweep cycle have
1470 // now happened, so we can make this profile cycle
1471 // available immediately.
1477 // Background sweep.
1480 sweep.parked = false
1481 ready(sweep.g, 0, true)
1486 // gcResetMarkState resets global state prior to marking (concurrent
1487 // or STW) and resets the stack scan state of all Gs.
1489 // This is safe to do without the world stopped because any Gs created
1490 // during or after this will start out in the reset state.
1492 // gcResetMarkState must be called on the system stack because it acquires
1493 // the heap lock. See mheap for details.
1496 func gcResetMarkState() {
1497 // This may be called during a concurrent phase, so lock to make sure
1498 // allgs doesn't change.
1499 forEachG(func(gp *g) {
1500 gp.gcscandone = false // set to true in gcphasework
1501 gp.gcAssistBytes = 0
1504 // Clear page marks. This is just 1MB per 64GB of heap, so the
1505 // time here is pretty trivial.
1507 arenas := mheap_.allArenas
1508 unlock(&mheap_.lock)
1509 for _, ai := range arenas {
1510 ha := mheap_.arenas[ai.l1()][ai.l2()]
1511 for i := range ha.pageMarks {
1516 work.bytesMarked = 0
1517 work.initialHeapLive = atomic.Load64(&gcController.heapLive)
1520 // Hooks for other packages
1522 var poolcleanup func()
1524 //go:linkname sync_runtime_registerPoolCleanup sync.runtime_registerPoolCleanup
1525 func sync_runtime_registerPoolCleanup(f func()) {
1531 if poolcleanup != nil {
1535 // Clear central sudog cache.
1536 // Leave per-P caches alone, they have strictly bounded size.
1537 // Disconnect cached list before dropping it on the floor,
1538 // so that a dangling ref to one entry does not pin all of them.
1539 lock(&sched.sudoglock)
1540 var sg, sgnext *sudog
1541 for sg = sched.sudogcache; sg != nil; sg = sgnext {
1545 sched.sudogcache = nil
1546 unlock(&sched.sudoglock)
1548 // Clear central defer pool.
1549 // Leave per-P pools alone, they have strictly bounded size.
1550 lock(&sched.deferlock)
1551 // disconnect cached list before dropping it on the floor,
1552 // so that a dangling ref to one entry does not pin all of them.
1553 var d, dlink *_defer
1554 for d = sched.deferpool; d != nil; d = dlink {
1558 sched.deferpool = nil
1559 unlock(&sched.deferlock)
1564 // itoaDiv formats val/(10**dec) into buf.
1565 func itoaDiv(buf []byte, val uint64, dec int) []byte {
1568 for val >= 10 || i >= idec {
1569 buf[i] = byte(val%10 + '0')
1577 buf[i] = byte(val + '0')
1581 // fmtNSAsMS nicely formats ns nanoseconds as milliseconds.
1582 func fmtNSAsMS(buf []byte, ns uint64) []byte {
1584 // Format as whole milliseconds.
1585 return itoaDiv(buf, ns/1e6, 0)
1587 // Format two digits of precision, with at most three decimal places.
1598 return itoaDiv(buf, x, dec)
1601 // Helpers for testing GC.
1603 // gcTestMoveStackOnNextCall causes the stack to be moved on a call
1604 // immediately following the call to this. It may not work correctly
1605 // if any other work appears after this call (such as returning).
1606 // Typically the following call should be marked go:noinline so it
1607 // performs a stack check.
1609 // In rare cases this may not cause the stack to move, specifically if
1610 // there's a preemption between this call and the next.
1611 func gcTestMoveStackOnNextCall() {
1613 gp.stackguard0 = stackForceMove
1616 // gcTestIsReachable performs a GC and returns a bit set where bit i
1617 // is set if ptrs[i] is reachable.
1618 func gcTestIsReachable(ptrs ...unsafe.Pointer) (mask uint64) {
1619 // This takes the pointers as unsafe.Pointers in order to keep
1620 // them live long enough for us to attach specials. After
1621 // that, we drop our references to them.
1624 panic("too many pointers for uint64 mask")
1627 // Block GC while we attach specials and drop our references
1628 // to ptrs. Otherwise, if a GC is in progress, it could mark
1629 // them reachable via this function before we have a chance to
1633 // Create reachability specials for ptrs.
1634 specials := make([]*specialReachable, len(ptrs))
1635 for i, p := range ptrs {
1636 lock(&mheap_.speciallock)
1637 s := (*specialReachable)(mheap_.specialReachableAlloc.alloc())
1638 unlock(&mheap_.speciallock)
1639 s.special.kind = _KindSpecialReachable
1640 if !addspecial(p, &s.special) {
1641 throw("already have a reachable special (duplicate pointer?)")
1644 // Make sure we don't retain ptrs.
1650 // Force a full GC and sweep.
1653 // Process specials.
1654 for i, s := range specials {
1657 println("runtime: object", i, "was not swept")
1658 throw("IsReachable failed")
1663 lock(&mheap_.speciallock)
1664 mheap_.specialReachableAlloc.free(unsafe.Pointer(s))
1665 unlock(&mheap_.speciallock)
1671 // gcTestPointerClass returns the category of what p points to, one of:
1672 // "heap", "stack", "data", "bss", "other". This is useful for checking
1673 // that a test is doing what it's intended to do.
1675 // This is nosplit simply to avoid extra pointer shuffling that may
1676 // complicate a test.
1679 func gcTestPointerClass(p unsafe.Pointer) string {
1680 p2 := uintptr(noescape(p))
1682 if gp.stack.lo <= p2 && p2 < gp.stack.hi {
1685 if base, _, _ := findObject(p2, 0, 0); base != 0 {
1688 for _, datap := range activeModules() {
1689 if datap.data <= p2 && p2 < datap.edata || datap.noptrdata <= p2 && p2 < datap.enoptrdata {
1692 if datap.bss <= p2 && p2 < datap.ebss || datap.noptrbss <= p2 && p2 <= datap.enoptrbss {