Source file src/runtime/mgc.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  // Garbage collector (GC).
     6  //
     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.
    11  //
    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.
    15  //
    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),
    19  // 966-975.
    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.
    23  //
    24  // 1. GC performs sweep termination.
    25  //
    26  //    a. Stop the world. This causes all Ps to reach a GC safe-point.
    27  //
    28  //    b. Sweep any unswept spans. There will only be unswept spans if
    29  //    this GC cycle was forced before the expected time.
    30  //
    31  // 2. GC performs the mark phase.
    32  //
    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.
    38  //
    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.
    45  //
    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.
    51  //
    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).
    55  //
    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.
    60  //
    61  // 3. GC performs mark termination.
    62  //
    63  //    a. Stop the world.
    64  //
    65  //    b. Set gcphase to _GCmarktermination, and disable workers and
    66  //    assists.
    67  //
    68  //    c. Perform housekeeping like flushing mcaches.
    69  //
    70  // 4. GC performs the sweep phase.
    71  //
    72  //    a. Prepare for the sweep phase by setting gcphase to _GCoff,
    73  //    setting up sweep state and disabling the write barrier.
    74  //
    75  //    b. Start the world. From this point on, newly allocated objects
    76  //    are white, and allocating sweeps spans before use if necessary.
    77  //
    78  //    c. GC does concurrent sweeping in the background and in response
    79  //    to allocation. See description below.
    80  //
    81  // 5. When sufficient allocation has taken place, replay the sequence
    82  // starting with 1 above. See discussion of GC rate below.
    83  
    84  // Concurrent sweep.
    85  //
    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".
    90  //
    91  // The background sweeper goroutine simply sweeps spans one-by-one.
    92  //
    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.
   103  //
   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).
   111  
   112  // GC rate.
   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 computed by the gcController.heapGoal method). 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).
   119  
   120  // Oblets
   121  //
   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.
   128  
   129  package runtime
   130  
   131  import (
   132  	"internal/cpu"
   133  	"internal/runtime/atomic"
   134  	"unsafe"
   135  )
   136  
   137  const (
   138  	_DebugGC      = 0
   139  	_FinBlockSize = 4 * 1024
   140  
   141  	// concurrentSweep is a debug flag. Disabling this flag
   142  	// ensures all spans are swept while the world is stopped.
   143  	concurrentSweep = true
   144  
   145  	// debugScanConservative enables debug logging for stack
   146  	// frames that are scanned conservatively.
   147  	debugScanConservative = false
   148  
   149  	// sweepMinHeapDistance is a lower bound on the heap distance
   150  	// (in bytes) reserved for concurrent sweeping between GC
   151  	// cycles.
   152  	sweepMinHeapDistance = 1024 * 1024
   153  )
   154  
   155  // heapObjectsCanMove always returns false in the current garbage collector.
   156  // It exists for go4.org/unsafe/assume-no-moving-gc, which is an
   157  // unfortunate idea that had an even more unfortunate implementation.
   158  // Every time a new Go release happened, the package stopped building,
   159  // and the authors had to add a new file with a new //go:build line, and
   160  // then the entire ecosystem of packages with that as a dependency had to
   161  // explicitly update to the new version. Many packages depend on
   162  // assume-no-moving-gc transitively, through paths like
   163  // inet.af/netaddr -> go4.org/intern -> assume-no-moving-gc.
   164  // This was causing a significant amount of friction around each new
   165  // release, so we added this bool for the package to //go:linkname
   166  // instead. The bool is still unfortunate, but it's not as bad as
   167  // breaking the ecosystem on every new release.
   168  //
   169  // If the Go garbage collector ever does move heap objects, we can set
   170  // this to true to break all the programs using assume-no-moving-gc.
   171  //
   172  //go:linkname heapObjectsCanMove
   173  func heapObjectsCanMove() bool {
   174  	return false
   175  }
   176  
   177  func gcinit() {
   178  	if unsafe.Sizeof(workbuf{}) != _WorkbufSize {
   179  		throw("size of Workbuf is suboptimal")
   180  	}
   181  	// No sweep on the first cycle.
   182  	sweep.active.state.Store(sweepDrainedMask)
   183  
   184  	// Initialize GC pacer state.
   185  	// Use the environment variable GOGC for the initial gcPercent value.
   186  	// Use the environment variable GOMEMLIMIT for the initial memoryLimit value.
   187  	gcController.init(readGOGC(), readGOMEMLIMIT())
   188  
   189  	work.startSema = 1
   190  	work.markDoneSema = 1
   191  	lockInit(&work.sweepWaiters.lock, lockRankSweepWaiters)
   192  	lockInit(&work.assistQueue.lock, lockRankAssistQueue)
   193  	lockInit(&work.wbufSpans.lock, lockRankWbufSpans)
   194  }
   195  
   196  // gcenable is called after the bulk of the runtime initialization,
   197  // just before we're about to start letting user code run.
   198  // It kicks off the background sweeper goroutine, the background
   199  // scavenger goroutine, and enables GC.
   200  func gcenable() {
   201  	// Kick off sweeping and scavenging.
   202  	c := make(chan int, 2)
   203  	go bgsweep(c)
   204  	go bgscavenge(c)
   205  	<-c
   206  	<-c
   207  	memstats.enablegc = true // now that runtime is initialized, GC is okay
   208  }
   209  
   210  // Garbage collector phase.
   211  // Indicates to write barrier and synchronization task to perform.
   212  var gcphase uint32
   213  
   214  // The compiler knows about this variable.
   215  // If you change it, you must change builtin/runtime.go, too.
   216  // If you change the first four bytes, you must also change the write
   217  // barrier insertion code.
   218  //
   219  // writeBarrier should be an internal detail,
   220  // but widely used packages access it using linkname.
   221  // Notable members of the hall of shame include:
   222  //   - github.com/bytedance/sonic
   223  //   - github.com/cloudwego/frugal
   224  //
   225  // Do not remove or change the type signature.
   226  // See go.dev/issue/67401.
   227  //
   228  //go:linkname writeBarrier
   229  var writeBarrier struct {
   230  	enabled bool    // compiler emits a check of this before calling write barrier
   231  	pad     [3]byte // compiler uses 32-bit load for "enabled" field
   232  	alignme uint64  // guarantee alignment so that compiler can use a 32 or 64-bit load
   233  }
   234  
   235  // gcBlackenEnabled is 1 if mutator assists and background mark
   236  // workers are allowed to blacken objects. This must only be set when
   237  // gcphase == _GCmark.
   238  var gcBlackenEnabled uint32
   239  
   240  const (
   241  	_GCoff             = iota // GC not running; sweeping in background, write barrier disabled
   242  	_GCmark                   // GC marking roots and workbufs: allocate black, write barrier ENABLED
   243  	_GCmarktermination        // GC mark termination: allocate black, P's help GC, write barrier ENABLED
   244  )
   245  
   246  //go:nosplit
   247  func setGCPhase(x uint32) {
   248  	atomic.Store(&gcphase, x)
   249  	writeBarrier.enabled = gcphase == _GCmark || gcphase == _GCmarktermination
   250  }
   251  
   252  // gcMarkWorkerMode represents the mode that a concurrent mark worker
   253  // should operate in.
   254  //
   255  // Concurrent marking happens through four different mechanisms. One
   256  // is mutator assists, which happen in response to allocations and are
   257  // not scheduled. The other three are variations in the per-P mark
   258  // workers and are distinguished by gcMarkWorkerMode.
   259  type gcMarkWorkerMode int
   260  
   261  const (
   262  	// gcMarkWorkerNotWorker indicates that the next scheduled G is not
   263  	// starting work and the mode should be ignored.
   264  	gcMarkWorkerNotWorker gcMarkWorkerMode = iota
   265  
   266  	// gcMarkWorkerDedicatedMode indicates that the P of a mark
   267  	// worker is dedicated to running that mark worker. The mark
   268  	// worker should run without preemption.
   269  	gcMarkWorkerDedicatedMode
   270  
   271  	// gcMarkWorkerFractionalMode indicates that a P is currently
   272  	// running the "fractional" mark worker. The fractional worker
   273  	// is necessary when GOMAXPROCS*gcBackgroundUtilization is not
   274  	// an integer and using only dedicated workers would result in
   275  	// utilization too far from the target of gcBackgroundUtilization.
   276  	// The fractional worker should run until it is preempted and
   277  	// will be scheduled to pick up the fractional part of
   278  	// GOMAXPROCS*gcBackgroundUtilization.
   279  	gcMarkWorkerFractionalMode
   280  
   281  	// gcMarkWorkerIdleMode indicates that a P is running the mark
   282  	// worker because it has nothing else to do. The idle worker
   283  	// should run until it is preempted and account its time
   284  	// against gcController.idleMarkTime.
   285  	gcMarkWorkerIdleMode
   286  )
   287  
   288  // gcMarkWorkerModeStrings are the strings labels of gcMarkWorkerModes
   289  // to use in execution traces.
   290  var gcMarkWorkerModeStrings = [...]string{
   291  	"Not worker",
   292  	"GC (dedicated)",
   293  	"GC (fractional)",
   294  	"GC (idle)",
   295  }
   296  
   297  // pollFractionalWorkerExit reports whether a fractional mark worker
   298  // should self-preempt. It assumes it is called from the fractional
   299  // worker.
   300  func pollFractionalWorkerExit() bool {
   301  	// This should be kept in sync with the fractional worker
   302  	// scheduler logic in findRunnableGCWorker.
   303  	now := nanotime()
   304  	delta := now - gcController.markStartTime
   305  	if delta <= 0 {
   306  		return true
   307  	}
   308  	p := getg().m.p.ptr()
   309  	selfTime := p.gcFractionalMarkTime + (now - p.gcMarkWorkerStartTime)
   310  	// Add some slack to the utilization goal so that the
   311  	// fractional worker isn't behind again the instant it exits.
   312  	return float64(selfTime)/float64(delta) > 1.2*gcController.fractionalUtilizationGoal
   313  }
   314  
   315  var work workType
   316  
   317  type workType struct {
   318  	full  lfstack          // lock-free list of full blocks workbuf
   319  	_     cpu.CacheLinePad // prevents false-sharing between full and empty
   320  	empty lfstack          // lock-free list of empty blocks workbuf
   321  	_     cpu.CacheLinePad // prevents false-sharing between empty and nproc/nwait
   322  
   323  	wbufSpans struct {
   324  		lock mutex
   325  		// free is a list of spans dedicated to workbufs, but
   326  		// that don't currently contain any workbufs.
   327  		free mSpanList
   328  		// busy is a list of all spans containing workbufs on
   329  		// one of the workbuf lists.
   330  		busy mSpanList
   331  	}
   332  
   333  	// Restore 64-bit alignment on 32-bit.
   334  	_ uint32
   335  
   336  	// bytesMarked is the number of bytes marked this cycle. This
   337  	// includes bytes blackened in scanned objects, noscan objects
   338  	// that go straight to black, and permagrey objects scanned by
   339  	// markroot during the concurrent scan phase. This is updated
   340  	// atomically during the cycle. Updates may be batched
   341  	// arbitrarily, since the value is only read at the end of the
   342  	// cycle.
   343  	//
   344  	// Because of benign races during marking, this number may not
   345  	// be the exact number of marked bytes, but it should be very
   346  	// close.
   347  	//
   348  	// Put this field here because it needs 64-bit atomic access
   349  	// (and thus 8-byte alignment even on 32-bit architectures).
   350  	bytesMarked uint64
   351  
   352  	markrootNext uint32 // next markroot job
   353  	markrootJobs uint32 // number of markroot jobs
   354  
   355  	nproc  uint32
   356  	tstart int64
   357  	nwait  uint32
   358  
   359  	// Number of roots of various root types. Set by gcMarkRootPrepare.
   360  	//
   361  	// nStackRoots == len(stackRoots), but we have nStackRoots for
   362  	// consistency.
   363  	nDataRoots, nBSSRoots, nSpanRoots, nStackRoots int
   364  
   365  	// Base indexes of each root type. Set by gcMarkRootPrepare.
   366  	baseData, baseBSS, baseSpans, baseStacks, baseEnd uint32
   367  
   368  	// stackRoots is a snapshot of all of the Gs that existed
   369  	// before the beginning of concurrent marking. The backing
   370  	// store of this must not be modified because it might be
   371  	// shared with allgs.
   372  	stackRoots []*g
   373  
   374  	// Each type of GC state transition is protected by a lock.
   375  	// Since multiple threads can simultaneously detect the state
   376  	// transition condition, any thread that detects a transition
   377  	// condition must acquire the appropriate transition lock,
   378  	// re-check the transition condition and return if it no
   379  	// longer holds or perform the transition if it does.
   380  	// Likewise, any transition must invalidate the transition
   381  	// condition before releasing the lock. This ensures that each
   382  	// transition is performed by exactly one thread and threads
   383  	// that need the transition to happen block until it has
   384  	// happened.
   385  	//
   386  	// startSema protects the transition from "off" to mark or
   387  	// mark termination.
   388  	startSema uint32
   389  	// markDoneSema protects transitions from mark to mark termination.
   390  	markDoneSema uint32
   391  
   392  	bgMarkDone uint32 // cas to 1 when at a background mark completion point
   393  	// Background mark completion signaling
   394  
   395  	// mode is the concurrency mode of the current GC cycle.
   396  	mode gcMode
   397  
   398  	// userForced indicates the current GC cycle was forced by an
   399  	// explicit user call.
   400  	userForced bool
   401  
   402  	// initialHeapLive is the value of gcController.heapLive at the
   403  	// beginning of this GC cycle.
   404  	initialHeapLive uint64
   405  
   406  	// assistQueue is a queue of assists that are blocked because
   407  	// there was neither enough credit to steal or enough work to
   408  	// do.
   409  	assistQueue struct {
   410  		lock mutex
   411  		q    gQueue
   412  	}
   413  
   414  	// sweepWaiters is a list of blocked goroutines to wake when
   415  	// we transition from mark termination to sweep.
   416  	sweepWaiters struct {
   417  		lock mutex
   418  		list gList
   419  	}
   420  
   421  	// cycles is the number of completed GC cycles, where a GC
   422  	// cycle is sweep termination, mark, mark termination, and
   423  	// sweep. This differs from memstats.numgc, which is
   424  	// incremented at mark termination.
   425  	cycles atomic.Uint32
   426  
   427  	// Timing/utilization stats for this cycle.
   428  	stwprocs, maxprocs                 int32
   429  	tSweepTerm, tMark, tMarkTerm, tEnd int64 // nanotime() of phase start
   430  
   431  	// pauseNS is the total STW time this cycle, measured as the time between
   432  	// when stopping began (just before trying to stop Ps) and just after the
   433  	// world started again.
   434  	pauseNS int64
   435  
   436  	// debug.gctrace heap sizes for this cycle.
   437  	heap0, heap1, heap2 uint64
   438  
   439  	// Cumulative estimated CPU usage.
   440  	cpuStats
   441  }
   442  
   443  // GC runs a garbage collection and blocks the caller until the
   444  // garbage collection is complete. It may also block the entire
   445  // program.
   446  func GC() {
   447  	// We consider a cycle to be: sweep termination, mark, mark
   448  	// termination, and sweep. This function shouldn't return
   449  	// until a full cycle has been completed, from beginning to
   450  	// end. Hence, we always want to finish up the current cycle
   451  	// and start a new one. That means:
   452  	//
   453  	// 1. In sweep termination, mark, or mark termination of cycle
   454  	// N, wait until mark termination N completes and transitions
   455  	// to sweep N.
   456  	//
   457  	// 2. In sweep N, help with sweep N.
   458  	//
   459  	// At this point we can begin a full cycle N+1.
   460  	//
   461  	// 3. Trigger cycle N+1 by starting sweep termination N+1.
   462  	//
   463  	// 4. Wait for mark termination N+1 to complete.
   464  	//
   465  	// 5. Help with sweep N+1 until it's done.
   466  	//
   467  	// This all has to be written to deal with the fact that the
   468  	// GC may move ahead on its own. For example, when we block
   469  	// until mark termination N, we may wake up in cycle N+2.
   470  
   471  	// Wait until the current sweep termination, mark, and mark
   472  	// termination complete.
   473  	n := work.cycles.Load()
   474  	gcWaitOnMark(n)
   475  
   476  	// We're now in sweep N or later. Trigger GC cycle N+1, which
   477  	// will first finish sweep N if necessary and then enter sweep
   478  	// termination N+1.
   479  	gcStart(gcTrigger{kind: gcTriggerCycle, n: n + 1})
   480  
   481  	// Wait for mark termination N+1 to complete.
   482  	gcWaitOnMark(n + 1)
   483  
   484  	// Finish sweep N+1 before returning. We do this both to
   485  	// complete the cycle and because runtime.GC() is often used
   486  	// as part of tests and benchmarks to get the system into a
   487  	// relatively stable and isolated state.
   488  	for work.cycles.Load() == n+1 && sweepone() != ^uintptr(0) {
   489  		Gosched()
   490  	}
   491  
   492  	// Callers may assume that the heap profile reflects the
   493  	// just-completed cycle when this returns (historically this
   494  	// happened because this was a STW GC), but right now the
   495  	// profile still reflects mark termination N, not N+1.
   496  	//
   497  	// As soon as all of the sweep frees from cycle N+1 are done,
   498  	// we can go ahead and publish the heap profile.
   499  	//
   500  	// First, wait for sweeping to finish. (We know there are no
   501  	// more spans on the sweep queue, but we may be concurrently
   502  	// sweeping spans, so we have to wait.)
   503  	for work.cycles.Load() == n+1 && !isSweepDone() {
   504  		Gosched()
   505  	}
   506  
   507  	// Now we're really done with sweeping, so we can publish the
   508  	// stable heap profile. Only do this if we haven't already hit
   509  	// another mark termination.
   510  	mp := acquirem()
   511  	cycle := work.cycles.Load()
   512  	if cycle == n+1 || (gcphase == _GCmark && cycle == n+2) {
   513  		mProf_PostSweep()
   514  	}
   515  	releasem(mp)
   516  }
   517  
   518  // gcWaitOnMark blocks until GC finishes the Nth mark phase. If GC has
   519  // already completed this mark phase, it returns immediately.
   520  func gcWaitOnMark(n uint32) {
   521  	for {
   522  		// Disable phase transitions.
   523  		lock(&work.sweepWaiters.lock)
   524  		nMarks := work.cycles.Load()
   525  		if gcphase != _GCmark {
   526  			// We've already completed this cycle's mark.
   527  			nMarks++
   528  		}
   529  		if nMarks > n {
   530  			// We're done.
   531  			unlock(&work.sweepWaiters.lock)
   532  			return
   533  		}
   534  
   535  		// Wait until sweep termination, mark, and mark
   536  		// termination of cycle N complete.
   537  		work.sweepWaiters.list.push(getg())
   538  		goparkunlock(&work.sweepWaiters.lock, waitReasonWaitForGCCycle, traceBlockUntilGCEnds, 1)
   539  	}
   540  }
   541  
   542  // gcMode indicates how concurrent a GC cycle should be.
   543  type gcMode int
   544  
   545  const (
   546  	gcBackgroundMode gcMode = iota // concurrent GC and sweep
   547  	gcForceMode                    // stop-the-world GC now, concurrent sweep
   548  	gcForceBlockMode               // stop-the-world GC now and STW sweep (forced by user)
   549  )
   550  
   551  // A gcTrigger is a predicate for starting a GC cycle. Specifically,
   552  // it is an exit condition for the _GCoff phase.
   553  type gcTrigger struct {
   554  	kind gcTriggerKind
   555  	now  int64  // gcTriggerTime: current time
   556  	n    uint32 // gcTriggerCycle: cycle number to start
   557  }
   558  
   559  type gcTriggerKind int
   560  
   561  const (
   562  	// gcTriggerHeap indicates that a cycle should be started when
   563  	// the heap size reaches the trigger heap size computed by the
   564  	// controller.
   565  	gcTriggerHeap gcTriggerKind = iota
   566  
   567  	// gcTriggerTime indicates that a cycle should be started when
   568  	// it's been more than forcegcperiod nanoseconds since the
   569  	// previous GC cycle.
   570  	gcTriggerTime
   571  
   572  	// gcTriggerCycle indicates that a cycle should be started if
   573  	// we have not yet started cycle number gcTrigger.n (relative
   574  	// to work.cycles).
   575  	gcTriggerCycle
   576  )
   577  
   578  // test reports whether the trigger condition is satisfied, meaning
   579  // that the exit condition for the _GCoff phase has been met. The exit
   580  // condition should be tested when allocating.
   581  func (t gcTrigger) test() bool {
   582  	if !memstats.enablegc || panicking.Load() != 0 || gcphase != _GCoff {
   583  		return false
   584  	}
   585  	switch t.kind {
   586  	case gcTriggerHeap:
   587  		trigger, _ := gcController.trigger()
   588  		return gcController.heapLive.Load() >= trigger
   589  	case gcTriggerTime:
   590  		if gcController.gcPercent.Load() < 0 {
   591  			return false
   592  		}
   593  		lastgc := int64(atomic.Load64(&memstats.last_gc_nanotime))
   594  		return lastgc != 0 && t.now-lastgc > forcegcperiod
   595  	case gcTriggerCycle:
   596  		// t.n > work.cycles, but accounting for wraparound.
   597  		return int32(t.n-work.cycles.Load()) > 0
   598  	}
   599  	return true
   600  }
   601  
   602  // gcStart starts the GC. It transitions from _GCoff to _GCmark (if
   603  // debug.gcstoptheworld == 0) or performs all of GC (if
   604  // debug.gcstoptheworld != 0).
   605  //
   606  // This may return without performing this transition in some cases,
   607  // such as when called on a system stack or with locks held.
   608  func gcStart(trigger gcTrigger) {
   609  	// Since this is called from malloc and malloc is called in
   610  	// the guts of a number of libraries that might be holding
   611  	// locks, don't attempt to start GC in non-preemptible or
   612  	// potentially unstable situations.
   613  	mp := acquirem()
   614  	if gp := getg(); gp == mp.g0 || mp.locks > 1 || mp.preemptoff != "" {
   615  		releasem(mp)
   616  		return
   617  	}
   618  	releasem(mp)
   619  	mp = nil
   620  
   621  	// Pick up the remaining unswept/not being swept spans concurrently
   622  	//
   623  	// This shouldn't happen if we're being invoked in background
   624  	// mode since proportional sweep should have just finished
   625  	// sweeping everything, but rounding errors, etc, may leave a
   626  	// few spans unswept. In forced mode, this is necessary since
   627  	// GC can be forced at any point in the sweeping cycle.
   628  	//
   629  	// We check the transition condition continuously here in case
   630  	// this G gets delayed in to the next GC cycle.
   631  	for trigger.test() && sweepone() != ^uintptr(0) {
   632  	}
   633  
   634  	// Perform GC initialization and the sweep termination
   635  	// transition.
   636  	semacquire(&work.startSema)
   637  	// Re-check transition condition under transition lock.
   638  	if !trigger.test() {
   639  		semrelease(&work.startSema)
   640  		return
   641  	}
   642  
   643  	// In gcstoptheworld debug mode, upgrade the mode accordingly.
   644  	// We do this after re-checking the transition condition so
   645  	// that multiple goroutines that detect the heap trigger don't
   646  	// start multiple STW GCs.
   647  	mode := gcBackgroundMode
   648  	if debug.gcstoptheworld == 1 {
   649  		mode = gcForceMode
   650  	} else if debug.gcstoptheworld == 2 {
   651  		mode = gcForceBlockMode
   652  	}
   653  
   654  	// Ok, we're doing it! Stop everybody else
   655  	semacquire(&gcsema)
   656  	semacquire(&worldsema)
   657  
   658  	// For stats, check if this GC was forced by the user.
   659  	// Update it under gcsema to avoid gctrace getting wrong values.
   660  	work.userForced = trigger.kind == gcTriggerCycle
   661  
   662  	trace := traceAcquire()
   663  	if trace.ok() {
   664  		trace.GCStart()
   665  		traceRelease(trace)
   666  	}
   667  
   668  	// Check that all Ps have finished deferred mcache flushes.
   669  	for _, p := range allp {
   670  		if fg := p.mcache.flushGen.Load(); fg != mheap_.sweepgen {
   671  			println("runtime: p", p.id, "flushGen", fg, "!= sweepgen", mheap_.sweepgen)
   672  			throw("p mcache not flushed")
   673  		}
   674  	}
   675  
   676  	gcBgMarkStartWorkers()
   677  
   678  	systemstack(gcResetMarkState)
   679  
   680  	work.stwprocs, work.maxprocs = gomaxprocs, gomaxprocs
   681  	if work.stwprocs > ncpu {
   682  		// This is used to compute CPU time of the STW phases,
   683  		// so it can't be more than ncpu, even if GOMAXPROCS is.
   684  		work.stwprocs = ncpu
   685  	}
   686  	work.heap0 = gcController.heapLive.Load()
   687  	work.pauseNS = 0
   688  	work.mode = mode
   689  
   690  	now := nanotime()
   691  	work.tSweepTerm = now
   692  	var stw worldStop
   693  	systemstack(func() {
   694  		stw = stopTheWorldWithSema(stwGCSweepTerm)
   695  	})
   696  
   697  	// Accumulate fine-grained stopping time.
   698  	work.cpuStats.accumulateGCPauseTime(stw.stoppingCPUTime, 1)
   699  
   700  	// Finish sweep before we start concurrent scan.
   701  	systemstack(func() {
   702  		finishsweep_m()
   703  	})
   704  
   705  	// clearpools before we start the GC. If we wait the memory will not be
   706  	// reclaimed until the next GC cycle.
   707  	clearpools()
   708  
   709  	work.cycles.Add(1)
   710  
   711  	// Assists and workers can start the moment we start
   712  	// the world.
   713  	gcController.startCycle(now, int(gomaxprocs), trigger)
   714  
   715  	// Notify the CPU limiter that assists may begin.
   716  	gcCPULimiter.startGCTransition(true, now)
   717  
   718  	// In STW mode, disable scheduling of user Gs. This may also
   719  	// disable scheduling of this goroutine, so it may block as
   720  	// soon as we start the world again.
   721  	if mode != gcBackgroundMode {
   722  		schedEnableUser(false)
   723  	}
   724  
   725  	// Enter concurrent mark phase and enable
   726  	// write barriers.
   727  	//
   728  	// Because the world is stopped, all Ps will
   729  	// observe that write barriers are enabled by
   730  	// the time we start the world and begin
   731  	// scanning.
   732  	//
   733  	// Write barriers must be enabled before assists are
   734  	// enabled because they must be enabled before
   735  	// any non-leaf heap objects are marked. Since
   736  	// allocations are blocked until assists can
   737  	// happen, we want to enable assists as early as
   738  	// possible.
   739  	setGCPhase(_GCmark)
   740  
   741  	gcBgMarkPrepare() // Must happen before assists are enabled.
   742  	gcMarkRootPrepare()
   743  
   744  	// Mark all active tinyalloc blocks. Since we're
   745  	// allocating from these, they need to be black like
   746  	// other allocations. The alternative is to blacken
   747  	// the tiny block on every allocation from it, which
   748  	// would slow down the tiny allocator.
   749  	gcMarkTinyAllocs()
   750  
   751  	// At this point all Ps have enabled the write
   752  	// barrier, thus maintaining the no white to
   753  	// black invariant. Enable mutator assists to
   754  	// put back-pressure on fast allocating
   755  	// mutators.
   756  	atomic.Store(&gcBlackenEnabled, 1)
   757  
   758  	// In STW mode, we could block the instant systemstack
   759  	// returns, so make sure we're not preemptible.
   760  	mp = acquirem()
   761  
   762  	// Update the CPU stats pause time.
   763  	//
   764  	// Use maxprocs instead of stwprocs here because the total time
   765  	// computed in the CPU stats is based on maxprocs, and we want them
   766  	// to be comparable.
   767  	work.cpuStats.accumulateGCPauseTime(nanotime()-stw.finishedStopping, work.maxprocs)
   768  
   769  	// Concurrent mark.
   770  	systemstack(func() {
   771  		now = startTheWorldWithSema(0, stw)
   772  		work.pauseNS += now - stw.startedStopping
   773  		work.tMark = now
   774  
   775  		// Release the CPU limiter.
   776  		gcCPULimiter.finishGCTransition(now)
   777  	})
   778  
   779  	// Release the world sema before Gosched() in STW mode
   780  	// because we will need to reacquire it later but before
   781  	// this goroutine becomes runnable again, and we could
   782  	// self-deadlock otherwise.
   783  	semrelease(&worldsema)
   784  	releasem(mp)
   785  
   786  	// Make sure we block instead of returning to user code
   787  	// in STW mode.
   788  	if mode != gcBackgroundMode {
   789  		Gosched()
   790  	}
   791  
   792  	semrelease(&work.startSema)
   793  }
   794  
   795  // gcMarkDoneFlushed counts the number of P's with flushed work.
   796  //
   797  // Ideally this would be a captured local in gcMarkDone, but forEachP
   798  // escapes its callback closure, so it can't capture anything.
   799  //
   800  // This is protected by markDoneSema.
   801  var gcMarkDoneFlushed uint32
   802  
   803  // gcMarkDone transitions the GC from mark to mark termination if all
   804  // reachable objects have been marked (that is, there are no grey
   805  // objects and can be no more in the future). Otherwise, it flushes
   806  // all local work to the global queues where it can be discovered by
   807  // other workers.
   808  //
   809  // This should be called when all local mark work has been drained and
   810  // there are no remaining workers. Specifically, when
   811  //
   812  //	work.nwait == work.nproc && !gcMarkWorkAvailable(p)
   813  //
   814  // The calling context must be preemptible.
   815  //
   816  // Flushing local work is important because idle Ps may have local
   817  // work queued. This is the only way to make that work visible and
   818  // drive GC to completion.
   819  //
   820  // It is explicitly okay to have write barriers in this function. If
   821  // it does transition to mark termination, then all reachable objects
   822  // have been marked, so the write barrier cannot shade any more
   823  // objects.
   824  func gcMarkDone() {
   825  	// Ensure only one thread is running the ragged barrier at a
   826  	// time.
   827  	semacquire(&work.markDoneSema)
   828  
   829  top:
   830  	// Re-check transition condition under transition lock.
   831  	//
   832  	// It's critical that this checks the global work queues are
   833  	// empty before performing the ragged barrier. Otherwise,
   834  	// there could be global work that a P could take after the P
   835  	// has passed the ragged barrier.
   836  	if !(gcphase == _GCmark && work.nwait == work.nproc && !gcMarkWorkAvailable(nil)) {
   837  		semrelease(&work.markDoneSema)
   838  		return
   839  	}
   840  
   841  	// forEachP needs worldsema to execute, and we'll need it to
   842  	// stop the world later, so acquire worldsema now.
   843  	semacquire(&worldsema)
   844  
   845  	// Flush all local buffers and collect flushedWork flags.
   846  	gcMarkDoneFlushed = 0
   847  	forEachP(waitReasonGCMarkTermination, func(pp *p) {
   848  		// Flush the write barrier buffer, since this may add
   849  		// work to the gcWork.
   850  		wbBufFlush1(pp)
   851  
   852  		// Flush the gcWork, since this may create global work
   853  		// and set the flushedWork flag.
   854  		//
   855  		// TODO(austin): Break up these workbufs to
   856  		// better distribute work.
   857  		pp.gcw.dispose()
   858  		// Collect the flushedWork flag.
   859  		if pp.gcw.flushedWork {
   860  			atomic.Xadd(&gcMarkDoneFlushed, 1)
   861  			pp.gcw.flushedWork = false
   862  		}
   863  	})
   864  
   865  	if gcMarkDoneFlushed != 0 {
   866  		// More grey objects were discovered since the
   867  		// previous termination check, so there may be more
   868  		// work to do. Keep going. It's possible the
   869  		// transition condition became true again during the
   870  		// ragged barrier, so re-check it.
   871  		semrelease(&worldsema)
   872  		goto top
   873  	}
   874  
   875  	// There was no global work, no local work, and no Ps
   876  	// communicated work since we took markDoneSema. Therefore
   877  	// there are no grey objects and no more objects can be
   878  	// shaded. Transition to mark termination.
   879  	now := nanotime()
   880  	work.tMarkTerm = now
   881  	getg().m.preemptoff = "gcing"
   882  	var stw worldStop
   883  	systemstack(func() {
   884  		stw = stopTheWorldWithSema(stwGCMarkTerm)
   885  	})
   886  	// The gcphase is _GCmark, it will transition to _GCmarktermination
   887  	// below. The important thing is that the wb remains active until
   888  	// all marking is complete. This includes writes made by the GC.
   889  
   890  	// Accumulate fine-grained stopping time.
   891  	work.cpuStats.accumulateGCPauseTime(stw.stoppingCPUTime, 1)
   892  
   893  	// There is sometimes work left over when we enter mark termination due
   894  	// to write barriers performed after the completion barrier above.
   895  	// Detect this and resume concurrent mark. This is obviously
   896  	// unfortunate.
   897  	//
   898  	// See issue #27993 for details.
   899  	//
   900  	// Switch to the system stack to call wbBufFlush1, though in this case
   901  	// it doesn't matter because we're non-preemptible anyway.
   902  	restart := false
   903  	systemstack(func() {
   904  		for _, p := range allp {
   905  			wbBufFlush1(p)
   906  			if !p.gcw.empty() {
   907  				restart = true
   908  				break
   909  			}
   910  		}
   911  	})
   912  	if restart {
   913  		getg().m.preemptoff = ""
   914  		systemstack(func() {
   915  			// Accumulate the time we were stopped before we had to start again.
   916  			work.cpuStats.accumulateGCPauseTime(nanotime()-stw.finishedStopping, work.maxprocs)
   917  
   918  			// Start the world again.
   919  			now := startTheWorldWithSema(0, stw)
   920  			work.pauseNS += now - stw.startedStopping
   921  		})
   922  		semrelease(&worldsema)
   923  		goto top
   924  	}
   925  
   926  	gcComputeStartingStackSize()
   927  
   928  	// Disable assists and background workers. We must do
   929  	// this before waking blocked assists.
   930  	atomic.Store(&gcBlackenEnabled, 0)
   931  
   932  	// Notify the CPU limiter that GC assists will now cease.
   933  	gcCPULimiter.startGCTransition(false, now)
   934  
   935  	// Wake all blocked assists. These will run when we
   936  	// start the world again.
   937  	gcWakeAllAssists()
   938  
   939  	// Likewise, release the transition lock. Blocked
   940  	// workers and assists will run when we start the
   941  	// world again.
   942  	semrelease(&work.markDoneSema)
   943  
   944  	// In STW mode, re-enable user goroutines. These will be
   945  	// queued to run after we start the world.
   946  	schedEnableUser(true)
   947  
   948  	// endCycle depends on all gcWork cache stats being flushed.
   949  	// The termination algorithm above ensured that up to
   950  	// allocations since the ragged barrier.
   951  	gcController.endCycle(now, int(gomaxprocs), work.userForced)
   952  
   953  	// Perform mark termination. This will restart the world.
   954  	gcMarkTermination(stw)
   955  }
   956  
   957  // World must be stopped and mark assists and background workers must be
   958  // disabled.
   959  func gcMarkTermination(stw worldStop) {
   960  	// Start marktermination (write barrier remains enabled for now).
   961  	setGCPhase(_GCmarktermination)
   962  
   963  	work.heap1 = gcController.heapLive.Load()
   964  	startTime := nanotime()
   965  
   966  	mp := acquirem()
   967  	mp.preemptoff = "gcing"
   968  	mp.traceback = 2
   969  	curgp := mp.curg
   970  	// N.B. The execution tracer is not aware of this status
   971  	// transition and handles it specially based on the
   972  	// wait reason.
   973  	casGToWaitingForGC(curgp, _Grunning, waitReasonGarbageCollection)
   974  
   975  	// Run gc on the g0 stack. We do this so that the g stack
   976  	// we're currently running on will no longer change. Cuts
   977  	// the root set down a bit (g0 stacks are not scanned, and
   978  	// we don't need to scan gc's internal state).  We also
   979  	// need to switch to g0 so we can shrink the stack.
   980  	systemstack(func() {
   981  		gcMark(startTime)
   982  		// Must return immediately.
   983  		// The outer function's stack may have moved
   984  		// during gcMark (it shrinks stacks, including the
   985  		// outer function's stack), so we must not refer
   986  		// to any of its variables. Return back to the
   987  		// non-system stack to pick up the new addresses
   988  		// before continuing.
   989  	})
   990  
   991  	var stwSwept bool
   992  	systemstack(func() {
   993  		work.heap2 = work.bytesMarked
   994  		if debug.gccheckmark > 0 {
   995  			// Run a full non-parallel, stop-the-world
   996  			// mark using checkmark bits, to check that we
   997  			// didn't forget to mark anything during the
   998  			// concurrent mark process.
   999  			startCheckmarks()
  1000  			gcResetMarkState()
  1001  			gcw := &getg().m.p.ptr().gcw
  1002  			gcDrain(gcw, 0)
  1003  			wbBufFlush1(getg().m.p.ptr())
  1004  			gcw.dispose()
  1005  			endCheckmarks()
  1006  		}
  1007  
  1008  		// marking is complete so we can turn the write barrier off
  1009  		setGCPhase(_GCoff)
  1010  		stwSwept = gcSweep(work.mode)
  1011  	})
  1012  
  1013  	mp.traceback = 0
  1014  	casgstatus(curgp, _Gwaiting, _Grunning)
  1015  
  1016  	trace := traceAcquire()
  1017  	if trace.ok() {
  1018  		trace.GCDone()
  1019  		traceRelease(trace)
  1020  	}
  1021  
  1022  	// all done
  1023  	mp.preemptoff = ""
  1024  
  1025  	if gcphase != _GCoff {
  1026  		throw("gc done but gcphase != _GCoff")
  1027  	}
  1028  
  1029  	// Record heapInUse for scavenger.
  1030  	memstats.lastHeapInUse = gcController.heapInUse.load()
  1031  
  1032  	// Update GC trigger and pacing, as well as downstream consumers
  1033  	// of this pacing information, for the next cycle.
  1034  	systemstack(gcControllerCommit)
  1035  
  1036  	// Update timing memstats
  1037  	now := nanotime()
  1038  	sec, nsec, _ := time_now()
  1039  	unixNow := sec*1e9 + int64(nsec)
  1040  	work.pauseNS += now - stw.startedStopping
  1041  	work.tEnd = now
  1042  	atomic.Store64(&memstats.last_gc_unix, uint64(unixNow)) // must be Unix time to make sense to user
  1043  	atomic.Store64(&memstats.last_gc_nanotime, uint64(now)) // monotonic time for us
  1044  	memstats.pause_ns[memstats.numgc%uint32(len(memstats.pause_ns))] = uint64(work.pauseNS)
  1045  	memstats.pause_end[memstats.numgc%uint32(len(memstats.pause_end))] = uint64(unixNow)
  1046  	memstats.pause_total_ns += uint64(work.pauseNS)
  1047  
  1048  	// Accumulate CPU stats.
  1049  	//
  1050  	// Use maxprocs instead of stwprocs for GC pause time because the total time
  1051  	// computed in the CPU stats is based on maxprocs, and we want them to be
  1052  	// comparable.
  1053  	//
  1054  	// Pass gcMarkPhase=true to accumulate so we can get all the latest GC CPU stats
  1055  	// in there too.
  1056  	work.cpuStats.accumulateGCPauseTime(now-stw.finishedStopping, work.maxprocs)
  1057  	work.cpuStats.accumulate(now, true)
  1058  
  1059  	// Compute overall GC CPU utilization.
  1060  	// Omit idle marking time from the overall utilization here since it's "free".
  1061  	memstats.gc_cpu_fraction = float64(work.cpuStats.GCTotalTime-work.cpuStats.GCIdleTime) / float64(work.cpuStats.TotalTime)
  1062  
  1063  	// Reset assist time and background time stats.
  1064  	//
  1065  	// Do this now, instead of at the start of the next GC cycle, because
  1066  	// these two may keep accumulating even if the GC is not active.
  1067  	scavenge.assistTime.Store(0)
  1068  	scavenge.backgroundTime.Store(0)
  1069  
  1070  	// Reset idle time stat.
  1071  	sched.idleTime.Store(0)
  1072  
  1073  	if work.userForced {
  1074  		memstats.numforcedgc++
  1075  	}
  1076  
  1077  	// Bump GC cycle count and wake goroutines waiting on sweep.
  1078  	lock(&work.sweepWaiters.lock)
  1079  	memstats.numgc++
  1080  	injectglist(&work.sweepWaiters.list)
  1081  	unlock(&work.sweepWaiters.lock)
  1082  
  1083  	// Increment the scavenge generation now.
  1084  	//
  1085  	// This moment represents peak heap in use because we're
  1086  	// about to start sweeping.
  1087  	mheap_.pages.scav.index.nextGen()
  1088  
  1089  	// Release the CPU limiter.
  1090  	gcCPULimiter.finishGCTransition(now)
  1091  
  1092  	// Finish the current heap profiling cycle and start a new
  1093  	// heap profiling cycle. We do this before starting the world
  1094  	// so events don't leak into the wrong cycle.
  1095  	mProf_NextCycle()
  1096  
  1097  	// There may be stale spans in mcaches that need to be swept.
  1098  	// Those aren't tracked in any sweep lists, so we need to
  1099  	// count them against sweep completion until we ensure all
  1100  	// those spans have been forced out.
  1101  	//
  1102  	// If gcSweep fully swept the heap (for example if the sweep
  1103  	// is not concurrent due to a GODEBUG setting), then we expect
  1104  	// the sweepLocker to be invalid, since sweeping is done.
  1105  	//
  1106  	// N.B. Below we might duplicate some work from gcSweep; this is
  1107  	// fine as all that work is idempotent within a GC cycle, and
  1108  	// we're still holding worldsema so a new cycle can't start.
  1109  	sl := sweep.active.begin()
  1110  	if !stwSwept && !sl.valid {
  1111  		throw("failed to set sweep barrier")
  1112  	} else if stwSwept && sl.valid {
  1113  		throw("non-concurrent sweep failed to drain all sweep queues")
  1114  	}
  1115  
  1116  	systemstack(func() {
  1117  		// The memstats updated above must be updated with the world
  1118  		// stopped to ensure consistency of some values, such as
  1119  		// sched.idleTime and sched.totaltime. memstats also include
  1120  		// the pause time (work,pauseNS), forcing computation of the
  1121  		// total pause time before the pause actually ends.
  1122  		//
  1123  		// Here we reuse the same now for start the world so that the
  1124  		// time added to /sched/pauses/total/gc:seconds will be
  1125  		// consistent with the value in memstats.
  1126  		startTheWorldWithSema(now, stw)
  1127  	})
  1128  
  1129  	// Flush the heap profile so we can start a new cycle next GC.
  1130  	// This is relatively expensive, so we don't do it with the
  1131  	// world stopped.
  1132  	mProf_Flush()
  1133  
  1134  	// Prepare workbufs for freeing by the sweeper. We do this
  1135  	// asynchronously because it can take non-trivial time.
  1136  	prepareFreeWorkbufs()
  1137  
  1138  	// Free stack spans. This must be done between GC cycles.
  1139  	systemstack(freeStackSpans)
  1140  
  1141  	// Ensure all mcaches are flushed. Each P will flush its own
  1142  	// mcache before allocating, but idle Ps may not. Since this
  1143  	// is necessary to sweep all spans, we need to ensure all
  1144  	// mcaches are flushed before we start the next GC cycle.
  1145  	//
  1146  	// While we're here, flush the page cache for idle Ps to avoid
  1147  	// having pages get stuck on them. These pages are hidden from
  1148  	// the scavenger, so in small idle heaps a significant amount
  1149  	// of additional memory might be held onto.
  1150  	//
  1151  	// Also, flush the pinner cache, to avoid leaking that memory
  1152  	// indefinitely.
  1153  	forEachP(waitReasonFlushProcCaches, func(pp *p) {
  1154  		pp.mcache.prepareForSweep()
  1155  		if pp.status == _Pidle {
  1156  			systemstack(func() {
  1157  				lock(&mheap_.lock)
  1158  				pp.pcache.flush(&mheap_.pages)
  1159  				unlock(&mheap_.lock)
  1160  			})
  1161  		}
  1162  		pp.pinnerCache = nil
  1163  	})
  1164  	if sl.valid {
  1165  		// Now that we've swept stale spans in mcaches, they don't
  1166  		// count against unswept spans.
  1167  		//
  1168  		// Note: this sweepLocker may not be valid if sweeping had
  1169  		// already completed during the STW. See the corresponding
  1170  		// begin() call that produced sl.
  1171  		sweep.active.end(sl)
  1172  	}
  1173  
  1174  	// Print gctrace before dropping worldsema. As soon as we drop
  1175  	// worldsema another cycle could start and smash the stats
  1176  	// we're trying to print.
  1177  	if debug.gctrace > 0 {
  1178  		util := int(memstats.gc_cpu_fraction * 100)
  1179  
  1180  		var sbuf [24]byte
  1181  		printlock()
  1182  		print("gc ", memstats.numgc,
  1183  			" @", string(itoaDiv(sbuf[:], uint64(work.tSweepTerm-runtimeInitTime)/1e6, 3)), "s ",
  1184  			util, "%: ")
  1185  		prev := work.tSweepTerm
  1186  		for i, ns := range []int64{work.tMark, work.tMarkTerm, work.tEnd} {
  1187  			if i != 0 {
  1188  				print("+")
  1189  			}
  1190  			print(string(fmtNSAsMS(sbuf[:], uint64(ns-prev))))
  1191  			prev = ns
  1192  		}
  1193  		print(" ms clock, ")
  1194  		for i, ns := range []int64{
  1195  			int64(work.stwprocs) * (work.tMark - work.tSweepTerm),
  1196  			gcController.assistTime.Load(),
  1197  			gcController.dedicatedMarkTime.Load() + gcController.fractionalMarkTime.Load(),
  1198  			gcController.idleMarkTime.Load(),
  1199  			int64(work.stwprocs) * (work.tEnd - work.tMarkTerm),
  1200  		} {
  1201  			if i == 2 || i == 3 {
  1202  				// Separate mark time components with /.
  1203  				print("/")
  1204  			} else if i != 0 {
  1205  				print("+")
  1206  			}
  1207  			print(string(fmtNSAsMS(sbuf[:], uint64(ns))))
  1208  		}
  1209  		print(" ms cpu, ",
  1210  			work.heap0>>20, "->", work.heap1>>20, "->", work.heap2>>20, " MB, ",
  1211  			gcController.lastHeapGoal>>20, " MB goal, ",
  1212  			gcController.lastStackScan.Load()>>20, " MB stacks, ",
  1213  			gcController.globalsScan.Load()>>20, " MB globals, ",
  1214  			work.maxprocs, " P")
  1215  		if work.userForced {
  1216  			print(" (forced)")
  1217  		}
  1218  		print("\n")
  1219  		printunlock()
  1220  	}
  1221  
  1222  	// Set any arena chunks that were deferred to fault.
  1223  	lock(&userArenaState.lock)
  1224  	faultList := userArenaState.fault
  1225  	userArenaState.fault = nil
  1226  	unlock(&userArenaState.lock)
  1227  	for _, lc := range faultList {
  1228  		lc.mspan.setUserArenaChunkToFault()
  1229  	}
  1230  
  1231  	// Enable huge pages on some metadata if we cross a heap threshold.
  1232  	if gcController.heapGoal() > minHeapForMetadataHugePages {
  1233  		systemstack(func() {
  1234  			mheap_.enableMetadataHugePages()
  1235  		})
  1236  	}
  1237  
  1238  	semrelease(&worldsema)
  1239  	semrelease(&gcsema)
  1240  	// Careful: another GC cycle may start now.
  1241  
  1242  	releasem(mp)
  1243  	mp = nil
  1244  
  1245  	// now that gc is done, kick off finalizer thread if needed
  1246  	if !concurrentSweep {
  1247  		// give the queued finalizers, if any, a chance to run
  1248  		Gosched()
  1249  	}
  1250  }
  1251  
  1252  // gcBgMarkStartWorkers prepares background mark worker goroutines. These
  1253  // goroutines will not run until the mark phase, but they must be started while
  1254  // the work is not stopped and from a regular G stack. The caller must hold
  1255  // worldsema.
  1256  func gcBgMarkStartWorkers() {
  1257  	// Background marking is performed by per-P G's. Ensure that each P has
  1258  	// a background GC G.
  1259  	//
  1260  	// Worker Gs don't exit if gomaxprocs is reduced. If it is raised
  1261  	// again, we can reuse the old workers; no need to create new workers.
  1262  	if gcBgMarkWorkerCount >= gomaxprocs {
  1263  		return
  1264  	}
  1265  
  1266  	// Increment mp.locks when allocating. We are called within gcStart,
  1267  	// and thus must not trigger another gcStart via an allocation. gcStart
  1268  	// bails when allocating with locks held, so simulate that for these
  1269  	// allocations.
  1270  	//
  1271  	// TODO(prattmic): cleanup gcStart to use a more explicit "in gcStart"
  1272  	// check for bailing.
  1273  	mp := acquirem()
  1274  	ready := make(chan struct{}, 1)
  1275  	releasem(mp)
  1276  
  1277  	for gcBgMarkWorkerCount < gomaxprocs {
  1278  		mp := acquirem() // See above, we allocate a closure here.
  1279  		go gcBgMarkWorker(ready)
  1280  		releasem(mp)
  1281  
  1282  		// N.B. we intentionally wait on each goroutine individually
  1283  		// rather than starting all in a batch and then waiting once
  1284  		// afterwards. By running one goroutine at a time, we can take
  1285  		// advantage of runnext to bounce back and forth between
  1286  		// workers and this goroutine. In an overloaded application,
  1287  		// this can reduce GC start latency by prioritizing these
  1288  		// goroutines rather than waiting on the end of the run queue.
  1289  		<-ready
  1290  		// The worker is now guaranteed to be added to the pool before
  1291  		// its P's next findRunnableGCWorker.
  1292  
  1293  		gcBgMarkWorkerCount++
  1294  	}
  1295  }
  1296  
  1297  // gcBgMarkPrepare sets up state for background marking.
  1298  // Mutator assists must not yet be enabled.
  1299  func gcBgMarkPrepare() {
  1300  	// Background marking will stop when the work queues are empty
  1301  	// and there are no more workers (note that, since this is
  1302  	// concurrent, this may be a transient state, but mark
  1303  	// termination will clean it up). Between background workers
  1304  	// and assists, we don't really know how many workers there
  1305  	// will be, so we pretend to have an arbitrarily large number
  1306  	// of workers, almost all of which are "waiting". While a
  1307  	// worker is working it decrements nwait. If nproc == nwait,
  1308  	// there are no workers.
  1309  	work.nproc = ^uint32(0)
  1310  	work.nwait = ^uint32(0)
  1311  }
  1312  
  1313  // gcBgMarkWorkerNode is an entry in the gcBgMarkWorkerPool. It points to a single
  1314  // gcBgMarkWorker goroutine.
  1315  type gcBgMarkWorkerNode struct {
  1316  	// Unused workers are managed in a lock-free stack. This field must be first.
  1317  	node lfnode
  1318  
  1319  	// The g of this worker.
  1320  	gp guintptr
  1321  
  1322  	// Release this m on park. This is used to communicate with the unlock
  1323  	// function, which cannot access the G's stack. It is unused outside of
  1324  	// gcBgMarkWorker().
  1325  	m muintptr
  1326  }
  1327  
  1328  func gcBgMarkWorker(ready chan struct{}) {
  1329  	gp := getg()
  1330  
  1331  	// We pass node to a gopark unlock function, so it can't be on
  1332  	// the stack (see gopark). Prevent deadlock from recursively
  1333  	// starting GC by disabling preemption.
  1334  	gp.m.preemptoff = "GC worker init"
  1335  	node := new(gcBgMarkWorkerNode)
  1336  	gp.m.preemptoff = ""
  1337  
  1338  	node.gp.set(gp)
  1339  
  1340  	node.m.set(acquirem())
  1341  
  1342  	ready <- struct{}{}
  1343  	// After this point, the background mark worker is generally scheduled
  1344  	// cooperatively by gcController.findRunnableGCWorker. While performing
  1345  	// work on the P, preemption is disabled because we are working on
  1346  	// P-local work buffers. When the preempt flag is set, this puts itself
  1347  	// into _Gwaiting to be woken up by gcController.findRunnableGCWorker
  1348  	// at the appropriate time.
  1349  	//
  1350  	// When preemption is enabled (e.g., while in gcMarkDone), this worker
  1351  	// may be preempted and schedule as a _Grunnable G from a runq. That is
  1352  	// fine; it will eventually gopark again for further scheduling via
  1353  	// findRunnableGCWorker.
  1354  	//
  1355  	// Since we disable preemption before notifying ready, we guarantee that
  1356  	// this G will be in the worker pool for the next findRunnableGCWorker.
  1357  	// This isn't strictly necessary, but it reduces latency between
  1358  	// _GCmark starting and the workers starting.
  1359  
  1360  	for {
  1361  		// Go to sleep until woken by
  1362  		// gcController.findRunnableGCWorker.
  1363  		gopark(func(g *g, nodep unsafe.Pointer) bool {
  1364  			node := (*gcBgMarkWorkerNode)(nodep)
  1365  
  1366  			if mp := node.m.ptr(); mp != nil {
  1367  				// The worker G is no longer running; release
  1368  				// the M.
  1369  				//
  1370  				// N.B. it is _safe_ to release the M as soon
  1371  				// as we are no longer performing P-local mark
  1372  				// work.
  1373  				//
  1374  				// However, since we cooperatively stop work
  1375  				// when gp.preempt is set, if we releasem in
  1376  				// the loop then the following call to gopark
  1377  				// would immediately preempt the G. This is
  1378  				// also safe, but inefficient: the G must
  1379  				// schedule again only to enter gopark and park
  1380  				// again. Thus, we defer the release until
  1381  				// after parking the G.
  1382  				releasem(mp)
  1383  			}
  1384  
  1385  			// Release this G to the pool.
  1386  			gcBgMarkWorkerPool.push(&node.node)
  1387  			// Note that at this point, the G may immediately be
  1388  			// rescheduled and may be running.
  1389  			return true
  1390  		}, unsafe.Pointer(node), waitReasonGCWorkerIdle, traceBlockSystemGoroutine, 0)
  1391  
  1392  		// Preemption must not occur here, or another G might see
  1393  		// p.gcMarkWorkerMode.
  1394  
  1395  		// Disable preemption so we can use the gcw. If the
  1396  		// scheduler wants to preempt us, we'll stop draining,
  1397  		// dispose the gcw, and then preempt.
  1398  		node.m.set(acquirem())
  1399  		pp := gp.m.p.ptr() // P can't change with preemption disabled.
  1400  
  1401  		if gcBlackenEnabled == 0 {
  1402  			println("worker mode", pp.gcMarkWorkerMode)
  1403  			throw("gcBgMarkWorker: blackening not enabled")
  1404  		}
  1405  
  1406  		if pp.gcMarkWorkerMode == gcMarkWorkerNotWorker {
  1407  			throw("gcBgMarkWorker: mode not set")
  1408  		}
  1409  
  1410  		startTime := nanotime()
  1411  		pp.gcMarkWorkerStartTime = startTime
  1412  		var trackLimiterEvent bool
  1413  		if pp.gcMarkWorkerMode == gcMarkWorkerIdleMode {
  1414  			trackLimiterEvent = pp.limiterEvent.start(limiterEventIdleMarkWork, startTime)
  1415  		}
  1416  
  1417  		decnwait := atomic.Xadd(&work.nwait, -1)
  1418  		if decnwait == work.nproc {
  1419  			println("runtime: work.nwait=", decnwait, "work.nproc=", work.nproc)
  1420  			throw("work.nwait was > work.nproc")
  1421  		}
  1422  
  1423  		systemstack(func() {
  1424  			// Mark our goroutine preemptible so its stack
  1425  			// can be scanned. This lets two mark workers
  1426  			// scan each other (otherwise, they would
  1427  			// deadlock). We must not modify anything on
  1428  			// the G stack. However, stack shrinking is
  1429  			// disabled for mark workers, so it is safe to
  1430  			// read from the G stack.
  1431  			//
  1432  			// N.B. The execution tracer is not aware of this status
  1433  			// transition and handles it specially based on the
  1434  			// wait reason.
  1435  			casGToWaitingForGC(gp, _Grunning, waitReasonGCWorkerActive)
  1436  			switch pp.gcMarkWorkerMode {
  1437  			default:
  1438  				throw("gcBgMarkWorker: unexpected gcMarkWorkerMode")
  1439  			case gcMarkWorkerDedicatedMode:
  1440  				gcDrainMarkWorkerDedicated(&pp.gcw, true)
  1441  				if gp.preempt {
  1442  					// We were preempted. This is
  1443  					// a useful signal to kick
  1444  					// everything out of the run
  1445  					// queue so it can run
  1446  					// somewhere else.
  1447  					if drainQ, n := runqdrain(pp); n > 0 {
  1448  						lock(&sched.lock)
  1449  						globrunqputbatch(&drainQ, int32(n))
  1450  						unlock(&sched.lock)
  1451  					}
  1452  				}
  1453  				// Go back to draining, this time
  1454  				// without preemption.
  1455  				gcDrainMarkWorkerDedicated(&pp.gcw, false)
  1456  			case gcMarkWorkerFractionalMode:
  1457  				gcDrainMarkWorkerFractional(&pp.gcw)
  1458  			case gcMarkWorkerIdleMode:
  1459  				gcDrainMarkWorkerIdle(&pp.gcw)
  1460  			}
  1461  			casgstatus(gp, _Gwaiting, _Grunning)
  1462  		})
  1463  
  1464  		// Account for time and mark us as stopped.
  1465  		now := nanotime()
  1466  		duration := now - startTime
  1467  		gcController.markWorkerStop(pp.gcMarkWorkerMode, duration)
  1468  		if trackLimiterEvent {
  1469  			pp.limiterEvent.stop(limiterEventIdleMarkWork, now)
  1470  		}
  1471  		if pp.gcMarkWorkerMode == gcMarkWorkerFractionalMode {
  1472  			atomic.Xaddint64(&pp.gcFractionalMarkTime, duration)
  1473  		}
  1474  
  1475  		// Was this the last worker and did we run out
  1476  		// of work?
  1477  		incnwait := atomic.Xadd(&work.nwait, +1)
  1478  		if incnwait > work.nproc {
  1479  			println("runtime: p.gcMarkWorkerMode=", pp.gcMarkWorkerMode,
  1480  				"work.nwait=", incnwait, "work.nproc=", work.nproc)
  1481  			throw("work.nwait > work.nproc")
  1482  		}
  1483  
  1484  		// We'll releasem after this point and thus this P may run
  1485  		// something else. We must clear the worker mode to avoid
  1486  		// attributing the mode to a different (non-worker) G in
  1487  		// traceGoStart.
  1488  		pp.gcMarkWorkerMode = gcMarkWorkerNotWorker
  1489  
  1490  		// If this worker reached a background mark completion
  1491  		// point, signal the main GC goroutine.
  1492  		if incnwait == work.nproc && !gcMarkWorkAvailable(nil) {
  1493  			// We don't need the P-local buffers here, allow
  1494  			// preemption because we may schedule like a regular
  1495  			// goroutine in gcMarkDone (block on locks, etc).
  1496  			releasem(node.m.ptr())
  1497  			node.m.set(nil)
  1498  
  1499  			gcMarkDone()
  1500  		}
  1501  	}
  1502  }
  1503  
  1504  // gcMarkWorkAvailable reports whether executing a mark worker
  1505  // on p is potentially useful. p may be nil, in which case it only
  1506  // checks the global sources of work.
  1507  func gcMarkWorkAvailable(p *p) bool {
  1508  	if p != nil && !p.gcw.empty() {
  1509  		return true
  1510  	}
  1511  	if !work.full.empty() {
  1512  		return true // global work available
  1513  	}
  1514  	if work.markrootNext < work.markrootJobs {
  1515  		return true // root scan work available
  1516  	}
  1517  	return false
  1518  }
  1519  
  1520  // gcMark runs the mark (or, for concurrent GC, mark termination)
  1521  // All gcWork caches must be empty.
  1522  // STW is in effect at this point.
  1523  func gcMark(startTime int64) {
  1524  	if gcphase != _GCmarktermination {
  1525  		throw("in gcMark expecting to see gcphase as _GCmarktermination")
  1526  	}
  1527  	work.tstart = startTime
  1528  
  1529  	// Check that there's no marking work remaining.
  1530  	if work.full != 0 || work.markrootNext < work.markrootJobs {
  1531  		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")
  1532  		panic("non-empty mark queue after concurrent mark")
  1533  	}
  1534  
  1535  	if debug.gccheckmark > 0 {
  1536  		// This is expensive when there's a large number of
  1537  		// Gs, so only do it if checkmark is also enabled.
  1538  		gcMarkRootCheck()
  1539  	}
  1540  
  1541  	// Drop allg snapshot. allgs may have grown, in which case
  1542  	// this is the only reference to the old backing store and
  1543  	// there's no need to keep it around.
  1544  	work.stackRoots = nil
  1545  
  1546  	// Clear out buffers and double-check that all gcWork caches
  1547  	// are empty. This should be ensured by gcMarkDone before we
  1548  	// enter mark termination.
  1549  	//
  1550  	// TODO: We could clear out buffers just before mark if this
  1551  	// has a non-negligible impact on STW time.
  1552  	for _, p := range allp {
  1553  		// The write barrier may have buffered pointers since
  1554  		// the gcMarkDone barrier. However, since the barrier
  1555  		// ensured all reachable objects were marked, all of
  1556  		// these must be pointers to black objects. Hence we
  1557  		// can just discard the write barrier buffer.
  1558  		if debug.gccheckmark > 0 {
  1559  			// For debugging, flush the buffer and make
  1560  			// sure it really was all marked.
  1561  			wbBufFlush1(p)
  1562  		} else {
  1563  			p.wbBuf.reset()
  1564  		}
  1565  
  1566  		gcw := &p.gcw
  1567  		if !gcw.empty() {
  1568  			printlock()
  1569  			print("runtime: P ", p.id, " flushedWork ", gcw.flushedWork)
  1570  			if gcw.wbuf1 == nil {
  1571  				print(" wbuf1=<nil>")
  1572  			} else {
  1573  				print(" wbuf1.n=", gcw.wbuf1.nobj)
  1574  			}
  1575  			if gcw.wbuf2 == nil {
  1576  				print(" wbuf2=<nil>")
  1577  			} else {
  1578  				print(" wbuf2.n=", gcw.wbuf2.nobj)
  1579  			}
  1580  			print("\n")
  1581  			throw("P has cached GC work at end of mark termination")
  1582  		}
  1583  		// There may still be cached empty buffers, which we
  1584  		// need to flush since we're going to free them. Also,
  1585  		// there may be non-zero stats because we allocated
  1586  		// black after the gcMarkDone barrier.
  1587  		gcw.dispose()
  1588  	}
  1589  
  1590  	// Flush scanAlloc from each mcache since we're about to modify
  1591  	// heapScan directly. If we were to flush this later, then scanAlloc
  1592  	// might have incorrect information.
  1593  	//
  1594  	// Note that it's not important to retain this information; we know
  1595  	// exactly what heapScan is at this point via scanWork.
  1596  	for _, p := range allp {
  1597  		c := p.mcache
  1598  		if c == nil {
  1599  			continue
  1600  		}
  1601  		c.scanAlloc = 0
  1602  	}
  1603  
  1604  	// Reset controller state.
  1605  	gcController.resetLive(work.bytesMarked)
  1606  }
  1607  
  1608  // gcSweep must be called on the system stack because it acquires the heap
  1609  // lock. See mheap for details.
  1610  //
  1611  // Returns true if the heap was fully swept by this function.
  1612  //
  1613  // The world must be stopped.
  1614  //
  1615  //go:systemstack
  1616  func gcSweep(mode gcMode) bool {
  1617  	assertWorldStopped()
  1618  
  1619  	if gcphase != _GCoff {
  1620  		throw("gcSweep being done but phase is not GCoff")
  1621  	}
  1622  
  1623  	lock(&mheap_.lock)
  1624  	mheap_.sweepgen += 2
  1625  	sweep.active.reset()
  1626  	mheap_.pagesSwept.Store(0)
  1627  	mheap_.sweepArenas = mheap_.allArenas
  1628  	mheap_.reclaimIndex.Store(0)
  1629  	mheap_.reclaimCredit.Store(0)
  1630  	unlock(&mheap_.lock)
  1631  
  1632  	sweep.centralIndex.clear()
  1633  
  1634  	if !concurrentSweep || mode == gcForceBlockMode {
  1635  		// Special case synchronous sweep.
  1636  		// Record that no proportional sweeping has to happen.
  1637  		lock(&mheap_.lock)
  1638  		mheap_.sweepPagesPerByte = 0
  1639  		unlock(&mheap_.lock)
  1640  		// Flush all mcaches.
  1641  		for _, pp := range allp {
  1642  			pp.mcache.prepareForSweep()
  1643  		}
  1644  		// Sweep all spans eagerly.
  1645  		for sweepone() != ^uintptr(0) {
  1646  		}
  1647  		// Free workbufs eagerly.
  1648  		prepareFreeWorkbufs()
  1649  		for freeSomeWbufs(false) {
  1650  		}
  1651  		// All "free" events for this mark/sweep cycle have
  1652  		// now happened, so we can make this profile cycle
  1653  		// available immediately.
  1654  		mProf_NextCycle()
  1655  		mProf_Flush()
  1656  		return true
  1657  	}
  1658  
  1659  	// Background sweep.
  1660  	lock(&sweep.lock)
  1661  	if sweep.parked {
  1662  		sweep.parked = false
  1663  		ready(sweep.g, 0, true)
  1664  	}
  1665  	unlock(&sweep.lock)
  1666  	return false
  1667  }
  1668  
  1669  // gcResetMarkState resets global state prior to marking (concurrent
  1670  // or STW) and resets the stack scan state of all Gs.
  1671  //
  1672  // This is safe to do without the world stopped because any Gs created
  1673  // during or after this will start out in the reset state.
  1674  //
  1675  // gcResetMarkState must be called on the system stack because it acquires
  1676  // the heap lock. See mheap for details.
  1677  //
  1678  //go:systemstack
  1679  func gcResetMarkState() {
  1680  	// This may be called during a concurrent phase, so lock to make sure
  1681  	// allgs doesn't change.
  1682  	forEachG(func(gp *g) {
  1683  		gp.gcscandone = false // set to true in gcphasework
  1684  		gp.gcAssistBytes = 0
  1685  	})
  1686  
  1687  	// Clear page marks. This is just 1MB per 64GB of heap, so the
  1688  	// time here is pretty trivial.
  1689  	lock(&mheap_.lock)
  1690  	arenas := mheap_.allArenas
  1691  	unlock(&mheap_.lock)
  1692  	for _, ai := range arenas {
  1693  		ha := mheap_.arenas[ai.l1()][ai.l2()]
  1694  		clear(ha.pageMarks[:])
  1695  	}
  1696  
  1697  	work.bytesMarked = 0
  1698  	work.initialHeapLive = gcController.heapLive.Load()
  1699  }
  1700  
  1701  // Hooks for other packages
  1702  
  1703  var poolcleanup func()
  1704  var boringCaches []unsafe.Pointer  // for crypto/internal/boring
  1705  var uniqueMapCleanup chan struct{} // for unique
  1706  
  1707  // sync_runtime_registerPoolCleanup should be an internal detail,
  1708  // but widely used packages access it using linkname.
  1709  // Notable members of the hall of shame include:
  1710  //   - github.com/bytedance/gopkg
  1711  //   - github.com/songzhibin97/gkit
  1712  //
  1713  // Do not remove or change the type signature.
  1714  // See go.dev/issue/67401.
  1715  //
  1716  //go:linkname sync_runtime_registerPoolCleanup sync.runtime_registerPoolCleanup
  1717  func sync_runtime_registerPoolCleanup(f func()) {
  1718  	poolcleanup = f
  1719  }
  1720  
  1721  //go:linkname boring_registerCache crypto/internal/boring/bcache.registerCache
  1722  func boring_registerCache(p unsafe.Pointer) {
  1723  	boringCaches = append(boringCaches, p)
  1724  }
  1725  
  1726  //go:linkname unique_runtime_registerUniqueMapCleanup unique.runtime_registerUniqueMapCleanup
  1727  func unique_runtime_registerUniqueMapCleanup(f func()) {
  1728  	// Start the goroutine in the runtime so it's counted as a system goroutine.
  1729  	uniqueMapCleanup = make(chan struct{}, 1)
  1730  	go func(cleanup func()) {
  1731  		for {
  1732  			<-uniqueMapCleanup
  1733  			cleanup()
  1734  		}
  1735  	}(f)
  1736  }
  1737  
  1738  func clearpools() {
  1739  	// clear sync.Pools
  1740  	if poolcleanup != nil {
  1741  		poolcleanup()
  1742  	}
  1743  
  1744  	// clear boringcrypto caches
  1745  	for _, p := range boringCaches {
  1746  		atomicstorep(p, nil)
  1747  	}
  1748  
  1749  	// clear unique maps
  1750  	if uniqueMapCleanup != nil {
  1751  		select {
  1752  		case uniqueMapCleanup <- struct{}{}:
  1753  		default:
  1754  		}
  1755  	}
  1756  
  1757  	// Clear central sudog cache.
  1758  	// Leave per-P caches alone, they have strictly bounded size.
  1759  	// Disconnect cached list before dropping it on the floor,
  1760  	// so that a dangling ref to one entry does not pin all of them.
  1761  	lock(&sched.sudoglock)
  1762  	var sg, sgnext *sudog
  1763  	for sg = sched.sudogcache; sg != nil; sg = sgnext {
  1764  		sgnext = sg.next
  1765  		sg.next = nil
  1766  	}
  1767  	sched.sudogcache = nil
  1768  	unlock(&sched.sudoglock)
  1769  
  1770  	// Clear central defer pool.
  1771  	// Leave per-P pools alone, they have strictly bounded size.
  1772  	lock(&sched.deferlock)
  1773  	// disconnect cached list before dropping it on the floor,
  1774  	// so that a dangling ref to one entry does not pin all of them.
  1775  	var d, dlink *_defer
  1776  	for d = sched.deferpool; d != nil; d = dlink {
  1777  		dlink = d.link
  1778  		d.link = nil
  1779  	}
  1780  	sched.deferpool = nil
  1781  	unlock(&sched.deferlock)
  1782  }
  1783  
  1784  // Timing
  1785  
  1786  // itoaDiv formats val/(10**dec) into buf.
  1787  func itoaDiv(buf []byte, val uint64, dec int) []byte {
  1788  	i := len(buf) - 1
  1789  	idec := i - dec
  1790  	for val >= 10 || i >= idec {
  1791  		buf[i] = byte(val%10 + '0')
  1792  		i--
  1793  		if i == idec {
  1794  			buf[i] = '.'
  1795  			i--
  1796  		}
  1797  		val /= 10
  1798  	}
  1799  	buf[i] = byte(val + '0')
  1800  	return buf[i:]
  1801  }
  1802  
  1803  // fmtNSAsMS nicely formats ns nanoseconds as milliseconds.
  1804  func fmtNSAsMS(buf []byte, ns uint64) []byte {
  1805  	if ns >= 10e6 {
  1806  		// Format as whole milliseconds.
  1807  		return itoaDiv(buf, ns/1e6, 0)
  1808  	}
  1809  	// Format two digits of precision, with at most three decimal places.
  1810  	x := ns / 1e3
  1811  	if x == 0 {
  1812  		buf[0] = '0'
  1813  		return buf[:1]
  1814  	}
  1815  	dec := 3
  1816  	for x >= 100 {
  1817  		x /= 10
  1818  		dec--
  1819  	}
  1820  	return itoaDiv(buf, x, dec)
  1821  }
  1822  
  1823  // Helpers for testing GC.
  1824  
  1825  // gcTestMoveStackOnNextCall causes the stack to be moved on a call
  1826  // immediately following the call to this. It may not work correctly
  1827  // if any other work appears after this call (such as returning).
  1828  // Typically the following call should be marked go:noinline so it
  1829  // performs a stack check.
  1830  //
  1831  // In rare cases this may not cause the stack to move, specifically if
  1832  // there's a preemption between this call and the next.
  1833  func gcTestMoveStackOnNextCall() {
  1834  	gp := getg()
  1835  	gp.stackguard0 = stackForceMove
  1836  }
  1837  
  1838  // gcTestIsReachable performs a GC and returns a bit set where bit i
  1839  // is set if ptrs[i] is reachable.
  1840  func gcTestIsReachable(ptrs ...unsafe.Pointer) (mask uint64) {
  1841  	// This takes the pointers as unsafe.Pointers in order to keep
  1842  	// them live long enough for us to attach specials. After
  1843  	// that, we drop our references to them.
  1844  
  1845  	if len(ptrs) > 64 {
  1846  		panic("too many pointers for uint64 mask")
  1847  	}
  1848  
  1849  	// Block GC while we attach specials and drop our references
  1850  	// to ptrs. Otherwise, if a GC is in progress, it could mark
  1851  	// them reachable via this function before we have a chance to
  1852  	// drop them.
  1853  	semacquire(&gcsema)
  1854  
  1855  	// Create reachability specials for ptrs.
  1856  	specials := make([]*specialReachable, len(ptrs))
  1857  	for i, p := range ptrs {
  1858  		lock(&mheap_.speciallock)
  1859  		s := (*specialReachable)(mheap_.specialReachableAlloc.alloc())
  1860  		unlock(&mheap_.speciallock)
  1861  		s.special.kind = _KindSpecialReachable
  1862  		if !addspecial(p, &s.special) {
  1863  			throw("already have a reachable special (duplicate pointer?)")
  1864  		}
  1865  		specials[i] = s
  1866  		// Make sure we don't retain ptrs.
  1867  		ptrs[i] = nil
  1868  	}
  1869  
  1870  	semrelease(&gcsema)
  1871  
  1872  	// Force a full GC and sweep.
  1873  	GC()
  1874  
  1875  	// Process specials.
  1876  	for i, s := range specials {
  1877  		if !s.done {
  1878  			printlock()
  1879  			println("runtime: object", i, "was not swept")
  1880  			throw("IsReachable failed")
  1881  		}
  1882  		if s.reachable {
  1883  			mask |= 1 << i
  1884  		}
  1885  		lock(&mheap_.speciallock)
  1886  		mheap_.specialReachableAlloc.free(unsafe.Pointer(s))
  1887  		unlock(&mheap_.speciallock)
  1888  	}
  1889  
  1890  	return mask
  1891  }
  1892  
  1893  // gcTestPointerClass returns the category of what p points to, one of:
  1894  // "heap", "stack", "data", "bss", "other". This is useful for checking
  1895  // that a test is doing what it's intended to do.
  1896  //
  1897  // This is nosplit simply to avoid extra pointer shuffling that may
  1898  // complicate a test.
  1899  //
  1900  //go:nosplit
  1901  func gcTestPointerClass(p unsafe.Pointer) string {
  1902  	p2 := uintptr(noescape(p))
  1903  	gp := getg()
  1904  	if gp.stack.lo <= p2 && p2 < gp.stack.hi {
  1905  		return "stack"
  1906  	}
  1907  	if base, _, _ := findObject(p2, 0, 0); base != 0 {
  1908  		return "heap"
  1909  	}
  1910  	for _, datap := range activeModules() {
  1911  		if datap.data <= p2 && p2 < datap.edata || datap.noptrdata <= p2 && p2 < datap.enoptrdata {
  1912  			return "data"
  1913  		}
  1914  		if datap.bss <= p2 && p2 < datap.ebss || datap.noptrbss <= p2 && p2 <= datap.enoptrbss {
  1915  			return "bss"
  1916  		}
  1917  	}
  1918  	KeepAlive(p)
  1919  	return "other"
  1920  }
  1921  

View as plain text