Source file src/runtime/time.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  // Time-related runtime and pieces of package time.
     6  
     7  package runtime
     8  
     9  import (
    10  	"internal/abi"
    11  	"internal/runtime/atomic"
    12  	"runtime/internal/sys"
    13  	"unsafe"
    14  )
    15  
    16  // A timer is a potentially repeating trigger for calling t.f(t.arg, t.seq).
    17  // Timers are allocated by client code, often as part of other data structures.
    18  // Each P has a heap of pointers to timers that it manages.
    19  //
    20  // A timer is expected to be used by only one client goroutine at a time,
    21  // but there will be concurrent access by the P managing that timer.
    22  // Timer accesses are protected by the lock t.mu, with a snapshot of
    23  // t's state bits published in t.astate to enable certain fast paths to make
    24  // decisions about a timer without acquiring the lock.
    25  type timer struct {
    26  	// mu protects reads and writes to all fields, with exceptions noted below.
    27  	mu mutex
    28  
    29  	astate atomic.Uint8 // atomic copy of state bits at last unlock
    30  	state  uint8        // state bits
    31  	isChan bool         // timer has a channel; immutable; can be read without lock
    32  
    33  	blocked uint32 // number of goroutines blocked on timer's channel
    34  
    35  	// Timer wakes up at when, and then at when+period, ... (period > 0 only)
    36  	// each time calling f(arg, seq, delay) in the timer goroutine, so f must be
    37  	// a well-behaved function and not block.
    38  	//
    39  	// The arg and seq are client-specified opaque arguments passed back to f.
    40  	// When used from netpoll, arg and seq have meanings defined by netpoll
    41  	// and are completely opaque to this code; in that context, seq is a sequence
    42  	// number to recognize and squech stale function invocations.
    43  	// When used from package time, arg is a channel (for After, NewTicker)
    44  	// or the function to call (for AfterFunc) and seq is unused (0).
    45  	//
    46  	// Package time does not know about seq, but if this is a channel timer (t.isChan == true),
    47  	// this file uses t.seq as a sequence number to recognize and squelch
    48  	// sends that correspond to an earlier (stale) timer configuration,
    49  	// similar to its use in netpoll. In this usage (that is, when t.isChan == true),
    50  	// writes to seq are protected by both t.mu and t.sendLock,
    51  	// so reads are allowed when holding either of the two mutexes.
    52  	//
    53  	// The delay argument is nanotime() - t.when, meaning the delay in ns between
    54  	// when the timer should have gone off and now. Normally that amount is
    55  	// small enough not to matter, but for channel timers that are fed lazily,
    56  	// the delay can be arbitrarily long; package time subtracts it out to make
    57  	// it look like the send happened earlier than it actually did.
    58  	// (No one looked at the channel since then, or the send would have
    59  	// not happened so late, so no one can tell the difference.)
    60  	when   int64
    61  	period int64
    62  	f      func(arg any, seq uintptr, delay int64)
    63  	arg    any
    64  	seq    uintptr
    65  
    66  	// If non-nil, the timers containing t.
    67  	ts *timers
    68  
    69  	// sendLock protects sends on the timer's channel.
    70  	// Not used for async (pre-Go 1.23) behavior when debug.asynctimerchan.Load() != 0.
    71  	sendLock mutex
    72  
    73  	// isSending is used to handle races between running a
    74  	// channel timer and stopping or resetting the timer.
    75  	// It is used only for channel timers (t.isChan == true).
    76  	// It is not used for tickers.
    77  	// The value is incremented when about to send a value on the channel,
    78  	// and decremented after sending the value.
    79  	// The stop/reset code uses this to detect whether it
    80  	// stopped the channel send.
    81  	//
    82  	// isSending is incremented only when t.mu is held.
    83  	// isSending is decremented only when t.sendLock is held.
    84  	// isSending is read only when both t.mu and t.sendLock are held.
    85  	isSending atomic.Int32
    86  }
    87  
    88  // init initializes a newly allocated timer t.
    89  // Any code that allocates a timer must call t.init before using it.
    90  // The arg and f can be set during init, or they can be nil in init
    91  // and set by a future call to t.modify.
    92  func (t *timer) init(f func(arg any, seq uintptr, delay int64), arg any) {
    93  	lockInit(&t.mu, lockRankTimer)
    94  	t.f = f
    95  	t.arg = arg
    96  }
    97  
    98  // A timers is a per-P set of timers.
    99  type timers struct {
   100  	// mu protects timers; timers are per-P, but the scheduler can
   101  	// access the timers of another P, so we have to lock.
   102  	mu mutex
   103  
   104  	// heap is the set of timers, ordered by heap[i].when.
   105  	// Must hold lock to access.
   106  	heap []timerWhen
   107  
   108  	// len is an atomic copy of len(heap).
   109  	len atomic.Uint32
   110  
   111  	// zombies is the number of timers in the heap
   112  	// that are marked for removal.
   113  	zombies atomic.Int32
   114  
   115  	// raceCtx is the race context used while executing timer functions.
   116  	raceCtx uintptr
   117  
   118  	// minWhenHeap is the minimum heap[i].when value (= heap[0].when).
   119  	// The wakeTime method uses minWhenHeap and minWhenModified
   120  	// to determine the next wake time.
   121  	// If minWhenHeap = 0, it means there are no timers in the heap.
   122  	minWhenHeap atomic.Int64
   123  
   124  	// minWhenModified is a lower bound on the minimum
   125  	// heap[i].when over timers with the timerModified bit set.
   126  	// If minWhenModified = 0, it means there are no timerModified timers in the heap.
   127  	minWhenModified atomic.Int64
   128  }
   129  
   130  type timerWhen struct {
   131  	timer *timer
   132  	when  int64
   133  }
   134  
   135  func (ts *timers) lock() {
   136  	lock(&ts.mu)
   137  }
   138  
   139  func (ts *timers) unlock() {
   140  	// Update atomic copy of len(ts.heap).
   141  	// We only update at unlock so that the len is always
   142  	// the most recent unlocked length, not an ephemeral length.
   143  	// This matters if we lock ts, delete the only timer from the heap,
   144  	// add it back, and unlock. We want ts.len.Load to return 1 the
   145  	// entire time, never 0. This is important for pidleput deciding
   146  	// whether ts is empty.
   147  	ts.len.Store(uint32(len(ts.heap)))
   148  
   149  	unlock(&ts.mu)
   150  }
   151  
   152  // Timer state field.
   153  const (
   154  	// timerHeaped is set when the timer is stored in some P's heap.
   155  	timerHeaped uint8 = 1 << iota
   156  
   157  	// timerModified is set when t.when has been modified
   158  	// but the heap's heap[i].when entry still needs to be updated.
   159  	// That change waits until the heap in which
   160  	// the timer appears can be locked and rearranged.
   161  	// timerModified is only set when timerHeaped is also set.
   162  	timerModified
   163  
   164  	// timerZombie is set when the timer has been stopped
   165  	// but is still present in some P's heap.
   166  	// Only set when timerHeaped is also set.
   167  	// It is possible for timerModified and timerZombie to both
   168  	// be set, meaning that the timer was modified and then stopped.
   169  	// A timer sending to a channel may be placed in timerZombie
   170  	// to take it out of the heap even though the timer is not stopped,
   171  	// as long as nothing is reading from the channel.
   172  	timerZombie
   173  )
   174  
   175  // timerDebug enables printing a textual debug trace of all timer operations to stderr.
   176  const timerDebug = false
   177  
   178  func (t *timer) trace(op string) {
   179  	if timerDebug {
   180  		t.trace1(op)
   181  	}
   182  }
   183  
   184  func (t *timer) trace1(op string) {
   185  	if !timerDebug {
   186  		return
   187  	}
   188  	bits := [4]string{"h", "m", "z", "c"}
   189  	for i := range 3 {
   190  		if t.state&(1<<i) == 0 {
   191  			bits[i] = "-"
   192  		}
   193  	}
   194  	if !t.isChan {
   195  		bits[3] = "-"
   196  	}
   197  	print("T ", t, " ", bits[0], bits[1], bits[2], bits[3], " b=", t.blocked, " ", op, "\n")
   198  }
   199  
   200  func (ts *timers) trace(op string) {
   201  	if timerDebug {
   202  		println("TS", ts, op)
   203  	}
   204  }
   205  
   206  // lock locks the timer, allowing reading or writing any of the timer fields.
   207  func (t *timer) lock() {
   208  	lock(&t.mu)
   209  	t.trace("lock")
   210  }
   211  
   212  // unlock updates t.astate and unlocks the timer.
   213  func (t *timer) unlock() {
   214  	t.trace("unlock")
   215  	// Let heap fast paths know whether heap[i].when is accurate.
   216  	// Also let maybeRunChan know whether channel is in heap.
   217  	t.astate.Store(t.state)
   218  	unlock(&t.mu)
   219  }
   220  
   221  // hchan returns the channel in t.arg.
   222  // t must be a timer with a channel.
   223  func (t *timer) hchan() *hchan {
   224  	if !t.isChan {
   225  		badTimer()
   226  	}
   227  	// Note: t.arg is a chan time.Time,
   228  	// and runtime cannot refer to that type,
   229  	// so we cannot use a type assertion.
   230  	return (*hchan)(efaceOf(&t.arg).data)
   231  }
   232  
   233  // updateHeap updates t as directed by t.state, updating t.state
   234  // and returning a bool indicating whether the state (and ts.heap[0].when) changed.
   235  // The caller must hold t's lock, or the world can be stopped instead.
   236  // The timer set t.ts must be non-nil and locked, t must be t.ts.heap[0], and updateHeap
   237  // takes care of moving t within the timers heap to preserve the heap invariants.
   238  // If ts == nil, then t must not be in a heap (or is in a heap that is
   239  // temporarily not maintaining its invariant, such as during timers.adjust).
   240  func (t *timer) updateHeap() (updated bool) {
   241  	assertWorldStoppedOrLockHeld(&t.mu)
   242  	t.trace("updateHeap")
   243  	ts := t.ts
   244  	if ts == nil || t != ts.heap[0].timer {
   245  		badTimer()
   246  	}
   247  	assertLockHeld(&ts.mu)
   248  	if t.state&timerZombie != 0 {
   249  		// Take timer out of heap.
   250  		t.state &^= timerHeaped | timerZombie | timerModified
   251  		ts.zombies.Add(-1)
   252  		ts.deleteMin()
   253  		return true
   254  	}
   255  
   256  	if t.state&timerModified != 0 {
   257  		// Update ts.heap[0].when and move within heap.
   258  		t.state &^= timerModified
   259  		ts.heap[0].when = t.when
   260  		ts.siftDown(0)
   261  		ts.updateMinWhenHeap()
   262  		return true
   263  	}
   264  
   265  	return false
   266  }
   267  
   268  // maxWhen is the maximum value for timer's when field.
   269  const maxWhen = 1<<63 - 1
   270  
   271  // verifyTimers can be set to true to add debugging checks that the
   272  // timer heaps are valid.
   273  const verifyTimers = false
   274  
   275  // Package time APIs.
   276  // Godoc uses the comments in package time, not these.
   277  
   278  // time.now is implemented in assembly.
   279  
   280  // timeSleep puts the current goroutine to sleep for at least ns nanoseconds.
   281  //
   282  //go:linkname timeSleep time.Sleep
   283  func timeSleep(ns int64) {
   284  	if ns <= 0 {
   285  		return
   286  	}
   287  
   288  	gp := getg()
   289  	t := gp.timer
   290  	if t == nil {
   291  		t = new(timer)
   292  		t.init(goroutineReady, gp)
   293  		gp.timer = t
   294  	}
   295  	when := nanotime() + ns
   296  	if when < 0 { // check for overflow.
   297  		when = maxWhen
   298  	}
   299  	gp.sleepWhen = when
   300  	gopark(resetForSleep, nil, waitReasonSleep, traceBlockSleep, 1)
   301  }
   302  
   303  // resetForSleep is called after the goroutine is parked for timeSleep.
   304  // We can't call timer.reset in timeSleep itself because if this is a short
   305  // sleep and there are many goroutines then the P can wind up running the
   306  // timer function, goroutineReady, before the goroutine has been parked.
   307  func resetForSleep(gp *g, _ unsafe.Pointer) bool {
   308  	gp.timer.reset(gp.sleepWhen, 0)
   309  	return true
   310  }
   311  
   312  // A timeTimer is a runtime-allocated time.Timer or time.Ticker
   313  // with the additional runtime state following it.
   314  // The runtime state is inaccessible to package time.
   315  type timeTimer struct {
   316  	c    unsafe.Pointer // <-chan time.Time
   317  	init bool
   318  	timer
   319  }
   320  
   321  // newTimer allocates and returns a new time.Timer or time.Ticker (same layout)
   322  // with the given parameters.
   323  //
   324  //go:linkname newTimer time.newTimer
   325  func newTimer(when, period int64, f func(arg any, seq uintptr, delay int64), arg any, c *hchan) *timeTimer {
   326  	t := new(timeTimer)
   327  	t.timer.init(nil, nil)
   328  	t.trace("new")
   329  	if raceenabled {
   330  		racerelease(unsafe.Pointer(&t.timer))
   331  	}
   332  	if c != nil {
   333  		lockInit(&t.sendLock, lockRankTimerSend)
   334  		t.isChan = true
   335  		c.timer = &t.timer
   336  		if c.dataqsiz == 0 {
   337  			throw("invalid timer channel: no capacity")
   338  		}
   339  	}
   340  	t.modify(when, period, f, arg, 0)
   341  	t.init = true
   342  	return t
   343  }
   344  
   345  // stopTimer stops a timer.
   346  // It reports whether t was stopped before being run.
   347  //
   348  //go:linkname stopTimer time.stopTimer
   349  func stopTimer(t *timeTimer) bool {
   350  	return t.stop()
   351  }
   352  
   353  // resetTimer resets an inactive timer, adding it to the timer heap.
   354  //
   355  // Reports whether the timer was modified before it was run.
   356  //
   357  //go:linkname resetTimer time.resetTimer
   358  func resetTimer(t *timeTimer, when, period int64) bool {
   359  	if raceenabled {
   360  		racerelease(unsafe.Pointer(&t.timer))
   361  	}
   362  	return t.reset(when, period)
   363  }
   364  
   365  // Go runtime.
   366  
   367  // Ready the goroutine arg.
   368  func goroutineReady(arg any, _ uintptr, _ int64) {
   369  	goready(arg.(*g), 0)
   370  }
   371  
   372  // addHeap adds t to the timers heap.
   373  // The caller must hold ts.lock or the world must be stopped.
   374  // The caller must also have checked that t belongs in the heap.
   375  // Callers that are not sure can call t.maybeAdd instead,
   376  // but note that maybeAdd has different locking requirements.
   377  func (ts *timers) addHeap(t *timer) {
   378  	assertWorldStoppedOrLockHeld(&ts.mu)
   379  	// Timers rely on the network poller, so make sure the poller
   380  	// has started.
   381  	if netpollInited.Load() == 0 {
   382  		netpollGenericInit()
   383  	}
   384  
   385  	if t.ts != nil {
   386  		throw("ts set in timer")
   387  	}
   388  	t.ts = ts
   389  	ts.heap = append(ts.heap, timerWhen{t, t.when})
   390  	ts.siftUp(len(ts.heap) - 1)
   391  	if t == ts.heap[0].timer {
   392  		ts.updateMinWhenHeap()
   393  	}
   394  }
   395  
   396  // maybeRunAsync checks whether t needs to be triggered and runs it if so.
   397  // The caller is responsible for locking the timer and for checking that we
   398  // are running timers in async mode. If the timer needs to be run,
   399  // maybeRunAsync will unlock and re-lock it.
   400  // The timer is always locked on return.
   401  func (t *timer) maybeRunAsync() {
   402  	assertLockHeld(&t.mu)
   403  	if t.state&timerHeaped == 0 && t.isChan && t.when > 0 {
   404  		// If timer should have triggered already (but nothing looked at it yet),
   405  		// trigger now, so that a receive after the stop sees the "old" value
   406  		// that should be there.
   407  		// (It is possible to have t.blocked > 0 if there is a racing receive
   408  		// in blockTimerChan, but timerHeaped not being set means
   409  		// it hasn't run t.maybeAdd yet; in that case, running the
   410  		// timer ourselves now is fine.)
   411  		if now := nanotime(); t.when <= now {
   412  			systemstack(func() {
   413  				t.unlockAndRun(now) // resets t.when
   414  			})
   415  			t.lock()
   416  		}
   417  	}
   418  }
   419  
   420  // stop stops the timer t. It may be on some other P, so we can't
   421  // actually remove it from the timers heap. We can only mark it as stopped.
   422  // It will be removed in due course by the P whose heap it is on.
   423  // Reports whether the timer was stopped before it was run.
   424  func (t *timer) stop() bool {
   425  	async := debug.asynctimerchan.Load() != 0
   426  	if !async && t.isChan {
   427  		lock(&t.sendLock)
   428  	}
   429  
   430  	t.lock()
   431  	t.trace("stop")
   432  	if async {
   433  		t.maybeRunAsync()
   434  	}
   435  	if t.state&timerHeaped != 0 {
   436  		t.state |= timerModified
   437  		if t.state&timerZombie == 0 {
   438  			t.state |= timerZombie
   439  			t.ts.zombies.Add(1)
   440  		}
   441  	}
   442  	pending := t.when > 0
   443  	t.when = 0
   444  
   445  	if !async && t.isChan {
   446  		// Stop any future sends with stale values.
   447  		// See timer.unlockAndRun.
   448  		t.seq++
   449  
   450  		// If there is currently a send in progress,
   451  		// incrementing seq is going to prevent that
   452  		// send from actually happening. That means
   453  		// that we should return true: the timer was
   454  		// stopped, even though t.when may be zero.
   455  		if t.period == 0 && t.isSending.Load() > 0 {
   456  			pending = true
   457  		}
   458  	}
   459  	t.unlock()
   460  	if !async && t.isChan {
   461  		unlock(&t.sendLock)
   462  		if timerchandrain(t.hchan()) {
   463  			pending = true
   464  		}
   465  	}
   466  
   467  	return pending
   468  }
   469  
   470  // deleteMin removes timer 0 from ts.
   471  // ts must be locked.
   472  func (ts *timers) deleteMin() {
   473  	assertLockHeld(&ts.mu)
   474  	t := ts.heap[0].timer
   475  	if t.ts != ts {
   476  		throw("wrong timers")
   477  	}
   478  	t.ts = nil
   479  	last := len(ts.heap) - 1
   480  	if last > 0 {
   481  		ts.heap[0] = ts.heap[last]
   482  	}
   483  	ts.heap[last] = timerWhen{}
   484  	ts.heap = ts.heap[:last]
   485  	if last > 0 {
   486  		ts.siftDown(0)
   487  	}
   488  	ts.updateMinWhenHeap()
   489  	if last == 0 {
   490  		// If there are no timers, then clearly there are no timerModified timers.
   491  		ts.minWhenModified.Store(0)
   492  	}
   493  }
   494  
   495  // modify modifies an existing timer.
   496  // This is called by the netpoll code or time.Ticker.Reset or time.Timer.Reset.
   497  // Reports whether the timer was modified before it was run.
   498  // If f == nil, then t.f, t.arg, and t.seq are not modified.
   499  func (t *timer) modify(when, period int64, f func(arg any, seq uintptr, delay int64), arg any, seq uintptr) bool {
   500  	if when <= 0 {
   501  		throw("timer when must be positive")
   502  	}
   503  	if period < 0 {
   504  		throw("timer period must be non-negative")
   505  	}
   506  	async := debug.asynctimerchan.Load() != 0
   507  
   508  	if !async && t.isChan {
   509  		lock(&t.sendLock)
   510  	}
   511  
   512  	t.lock()
   513  	if async {
   514  		t.maybeRunAsync()
   515  	}
   516  	t.trace("modify")
   517  	oldPeriod := t.period
   518  	t.period = period
   519  	if f != nil {
   520  		t.f = f
   521  		t.arg = arg
   522  		t.seq = seq
   523  	}
   524  
   525  	wake := false
   526  	pending := t.when > 0
   527  	t.when = when
   528  	if t.state&timerHeaped != 0 {
   529  		t.state |= timerModified
   530  		if t.state&timerZombie != 0 {
   531  			// In the heap but marked for removal (by a Stop).
   532  			// Unmark it, since it has been Reset and will be running again.
   533  			t.ts.zombies.Add(-1)
   534  			t.state &^= timerZombie
   535  		}
   536  		// The corresponding heap[i].when is updated later.
   537  		// See comment in type timer above and in timers.adjust below.
   538  		if min := t.ts.minWhenModified.Load(); min == 0 || when < min {
   539  			wake = true
   540  			// Force timerModified bit out to t.astate before updating t.minWhenModified,
   541  			// to synchronize with t.ts.adjust. See comment in adjust.
   542  			t.astate.Store(t.state)
   543  			t.ts.updateMinWhenModified(when)
   544  		}
   545  	}
   546  
   547  	add := t.needsAdd()
   548  
   549  	if !async && t.isChan {
   550  		// Stop any future sends with stale values.
   551  		// See timer.unlockAndRun.
   552  		t.seq++
   553  
   554  		// If there is currently a send in progress,
   555  		// incrementing seq is going to prevent that
   556  		// send from actually happening. That means
   557  		// that we should return true: the timer was
   558  		// stopped, even though t.when may be zero.
   559  		if oldPeriod == 0 && t.isSending.Load() > 0 {
   560  			pending = true
   561  		}
   562  	}
   563  	t.unlock()
   564  	if !async && t.isChan {
   565  		if timerchandrain(t.hchan()) {
   566  			pending = true
   567  		}
   568  		unlock(&t.sendLock)
   569  	}
   570  
   571  	if add {
   572  		t.maybeAdd()
   573  	}
   574  	if wake {
   575  		wakeNetPoller(when)
   576  	}
   577  
   578  	return pending
   579  }
   580  
   581  // needsAdd reports whether t needs to be added to a timers heap.
   582  // t must be locked.
   583  func (t *timer) needsAdd() bool {
   584  	assertLockHeld(&t.mu)
   585  	need := t.state&timerHeaped == 0 && t.when > 0 && (!t.isChan || t.blocked > 0)
   586  	if need {
   587  		t.trace("needsAdd+")
   588  	} else {
   589  		t.trace("needsAdd-")
   590  	}
   591  	return need
   592  }
   593  
   594  // maybeAdd adds t to the local timers heap if it needs to be in a heap.
   595  // The caller must not hold t's lock nor any timers heap lock.
   596  // The caller probably just unlocked t, but that lock must be dropped
   597  // in order to acquire a ts.lock, to avoid lock inversions.
   598  // (timers.adjust holds ts.lock while acquiring each t's lock,
   599  // so we cannot hold any t's lock while acquiring ts.lock).
   600  //
   601  // Strictly speaking it *might* be okay to hold t.lock and
   602  // acquire ts.lock at the same time, because we know that
   603  // t is not in any ts.heap, so nothing holding a ts.lock would
   604  // be acquiring the t.lock at the same time, meaning there
   605  // isn't a possible deadlock. But it is easier and safer not to be
   606  // too clever and respect the static ordering.
   607  // (If we don't, we have to change the static lock checking of t and ts.)
   608  //
   609  // Concurrent calls to time.Timer.Reset or blockTimerChan
   610  // may result in concurrent calls to t.maybeAdd,
   611  // so we cannot assume that t is not in a heap on entry to t.maybeAdd.
   612  func (t *timer) maybeAdd() {
   613  	// Note: Not holding any locks on entry to t.maybeAdd,
   614  	// so the current g can be rescheduled to a different M and P
   615  	// at any time, including between the ts := assignment and the
   616  	// call to ts.lock. If a reschedule happened then, we would be
   617  	// adding t to some other P's timers, perhaps even a P that the scheduler
   618  	// has marked as idle with no timers, in which case the timer could
   619  	// go unnoticed until long after t.when.
   620  	// Calling acquirem instead of using getg().m makes sure that
   621  	// we end up locking and inserting into the current P's timers.
   622  	mp := acquirem()
   623  	ts := &mp.p.ptr().timers
   624  	ts.lock()
   625  	ts.cleanHead()
   626  	t.lock()
   627  	t.trace("maybeAdd")
   628  	when := int64(0)
   629  	wake := false
   630  	if t.needsAdd() {
   631  		t.state |= timerHeaped
   632  		when = t.when
   633  		wakeTime := ts.wakeTime()
   634  		wake = wakeTime == 0 || when < wakeTime
   635  		ts.addHeap(t)
   636  	}
   637  	t.unlock()
   638  	ts.unlock()
   639  	releasem(mp)
   640  	if wake {
   641  		wakeNetPoller(when)
   642  	}
   643  }
   644  
   645  // reset resets the time when a timer should fire.
   646  // If used for an inactive timer, the timer will become active.
   647  // Reports whether the timer was active and was stopped.
   648  func (t *timer) reset(when, period int64) bool {
   649  	return t.modify(when, period, nil, nil, 0)
   650  }
   651  
   652  // cleanHead cleans up the head of the timer queue. This speeds up
   653  // programs that create and delete timers; leaving them in the heap
   654  // slows down heap operations.
   655  // The caller must have locked ts.
   656  func (ts *timers) cleanHead() {
   657  	ts.trace("cleanHead")
   658  	assertLockHeld(&ts.mu)
   659  	gp := getg()
   660  	for {
   661  		if len(ts.heap) == 0 {
   662  			return
   663  		}
   664  
   665  		// This loop can theoretically run for a while, and because
   666  		// it is holding timersLock it cannot be preempted.
   667  		// If someone is trying to preempt us, just return.
   668  		// We can clean the timers later.
   669  		if gp.preemptStop {
   670  			return
   671  		}
   672  
   673  		// Delete zombies from tail of heap. It requires no heap adjustments at all,
   674  		// and doing so increases the chances that when we swap out a zombie
   675  		// in heap[0] for the tail of the heap, we'll get a non-zombie timer,
   676  		// shortening this loop.
   677  		n := len(ts.heap)
   678  		if t := ts.heap[n-1].timer; t.astate.Load()&timerZombie != 0 {
   679  			t.lock()
   680  			if t.state&timerZombie != 0 {
   681  				t.state &^= timerHeaped | timerZombie | timerModified
   682  				t.ts = nil
   683  				ts.zombies.Add(-1)
   684  				ts.heap[n-1] = timerWhen{}
   685  				ts.heap = ts.heap[:n-1]
   686  			}
   687  			t.unlock()
   688  			continue
   689  		}
   690  
   691  		t := ts.heap[0].timer
   692  		if t.ts != ts {
   693  			throw("bad ts")
   694  		}
   695  
   696  		if t.astate.Load()&(timerModified|timerZombie) == 0 {
   697  			// Fast path: head of timers does not need adjustment.
   698  			return
   699  		}
   700  
   701  		t.lock()
   702  		updated := t.updateHeap()
   703  		t.unlock()
   704  		if !updated {
   705  			// Head of timers does not need adjustment.
   706  			return
   707  		}
   708  	}
   709  }
   710  
   711  // take moves any timers from src into ts
   712  // and then clears the timer state from src,
   713  // because src is being destroyed.
   714  // The caller must not have locked either timers.
   715  // For now this is only called when the world is stopped.
   716  func (ts *timers) take(src *timers) {
   717  	ts.trace("take")
   718  	assertWorldStopped()
   719  	if len(src.heap) > 0 {
   720  		// The world is stopped, so we ignore the locking of ts and src here.
   721  		// That would introduce a sched < timers lock ordering,
   722  		// which we'd rather avoid in the static ranking.
   723  		for _, tw := range src.heap {
   724  			t := tw.timer
   725  			t.ts = nil
   726  			if t.state&timerZombie != 0 {
   727  				t.state &^= timerHeaped | timerZombie | timerModified
   728  			} else {
   729  				t.state &^= timerModified
   730  				ts.addHeap(t)
   731  			}
   732  		}
   733  		src.heap = nil
   734  		src.zombies.Store(0)
   735  		src.minWhenHeap.Store(0)
   736  		src.minWhenModified.Store(0)
   737  		src.len.Store(0)
   738  		ts.len.Store(uint32(len(ts.heap)))
   739  	}
   740  }
   741  
   742  // adjust looks through the timers in ts.heap for
   743  // any timers that have been modified to run earlier, and puts them in
   744  // the correct place in the heap. While looking for those timers,
   745  // it also moves timers that have been modified to run later,
   746  // and removes deleted timers. The caller must have locked ts.
   747  func (ts *timers) adjust(now int64, force bool) {
   748  	ts.trace("adjust")
   749  	assertLockHeld(&ts.mu)
   750  	// If we haven't yet reached the time of the earliest modified
   751  	// timer, don't do anything. This speeds up programs that adjust
   752  	// a lot of timers back and forth if the timers rarely expire.
   753  	// We'll postpone looking through all the adjusted timers until
   754  	// one would actually expire.
   755  	if !force {
   756  		first := ts.minWhenModified.Load()
   757  		if first == 0 || first > now {
   758  			if verifyTimers {
   759  				ts.verify()
   760  			}
   761  			return
   762  		}
   763  	}
   764  
   765  	// minWhenModified is a lower bound on the earliest t.when
   766  	// among the timerModified timers. We want to make it more precise:
   767  	// we are going to scan the heap and clean out all the timerModified bits,
   768  	// at which point minWhenModified can be set to 0 (indicating none at all).
   769  	//
   770  	// Other P's can be calling ts.wakeTime concurrently, and we'd like to
   771  	// keep ts.wakeTime returning an accurate value throughout this entire process.
   772  	//
   773  	// Setting minWhenModified = 0 *before* the scan could make wakeTime
   774  	// return an incorrect value: if minWhenModified < minWhenHeap, then clearing
   775  	// it to 0 will make wakeTime return minWhenHeap (too late) until the scan finishes.
   776  	// To avoid that, we want to set minWhenModified to 0 *after* the scan.
   777  	//
   778  	// Setting minWhenModified = 0 *after* the scan could result in missing
   779  	// concurrent timer modifications in other goroutines; those will lock
   780  	// the specific timer, set the timerModified bit, and set t.when.
   781  	// To avoid that, we want to set minWhenModified to 0 *before* the scan.
   782  	//
   783  	// The way out of this dilemma is to preserve wakeTime a different way.
   784  	// wakeTime is min(minWhenHeap, minWhenModified), and minWhenHeap
   785  	// is protected by ts.lock, which we hold, so we can modify it however we like
   786  	// in service of keeping wakeTime accurate.
   787  	//
   788  	// So we can:
   789  	//
   790  	//	1. Set minWhenHeap = min(minWhenHeap, minWhenModified)
   791  	//	2. Set minWhenModified = 0
   792  	//	   (Other goroutines may modify timers and update minWhenModified now.)
   793  	//	3. Scan timers
   794  	//	4. Set minWhenHeap = heap[0].when
   795  	//
   796  	// That order preserves a correct value of wakeTime throughout the entire
   797  	// operation:
   798  	// Step 1 “locks in” an accurate wakeTime even with minWhenModified cleared.
   799  	// Step 2 makes sure concurrent t.when updates are not lost during the scan.
   800  	// Step 3 processes all modified timer values, justifying minWhenModified = 0.
   801  	// Step 4 corrects minWhenHeap to a precise value.
   802  	//
   803  	// The wakeTime method implementation reads minWhenModified *before* minWhenHeap,
   804  	// so that if the minWhenModified is observed to be 0, that means the minWhenHeap that
   805  	// follows will include the information that was zeroed out of it.
   806  	//
   807  	// Originally Step 3 locked every timer, which made sure any timer update that was
   808  	// already in progress during Steps 1+2 completed and was observed by Step 3.
   809  	// All that locking was too expensive, so now we do an atomic load of t.astate to
   810  	// decide whether we need to do a full lock. To make sure that we still observe any
   811  	// timer update already in progress during Steps 1+2, t.modify sets timerModified
   812  	// in t.astate *before* calling t.updateMinWhenModified. That ensures that the
   813  	// overwrite in Step 2 cannot lose an update: if it does overwrite an update, Step 3
   814  	// will see the timerModified and do a full lock.
   815  	ts.minWhenHeap.Store(ts.wakeTime())
   816  	ts.minWhenModified.Store(0)
   817  
   818  	changed := false
   819  	for i := 0; i < len(ts.heap); i++ {
   820  		tw := &ts.heap[i]
   821  		t := tw.timer
   822  		if t.ts != ts {
   823  			throw("bad ts")
   824  		}
   825  
   826  		if t.astate.Load()&(timerModified|timerZombie) == 0 {
   827  			// Does not need adjustment.
   828  			continue
   829  		}
   830  
   831  		t.lock()
   832  		switch {
   833  		case t.state&timerHeaped == 0:
   834  			badTimer()
   835  
   836  		case t.state&timerZombie != 0:
   837  			ts.zombies.Add(-1)
   838  			t.state &^= timerHeaped | timerZombie | timerModified
   839  			n := len(ts.heap)
   840  			ts.heap[i] = ts.heap[n-1]
   841  			ts.heap[n-1] = timerWhen{}
   842  			ts.heap = ts.heap[:n-1]
   843  			t.ts = nil
   844  			i--
   845  			changed = true
   846  
   847  		case t.state&timerModified != 0:
   848  			tw.when = t.when
   849  			t.state &^= timerModified
   850  			changed = true
   851  		}
   852  		t.unlock()
   853  	}
   854  
   855  	if changed {
   856  		ts.initHeap()
   857  	}
   858  	ts.updateMinWhenHeap()
   859  
   860  	if verifyTimers {
   861  		ts.verify()
   862  	}
   863  }
   864  
   865  // wakeTime looks at ts's timers and returns the time when we
   866  // should wake up the netpoller. It returns 0 if there are no timers.
   867  // This function is invoked when dropping a P, so it must run without
   868  // any write barriers.
   869  //
   870  //go:nowritebarrierrec
   871  func (ts *timers) wakeTime() int64 {
   872  	// Note that the order of these two loads matters:
   873  	// adjust updates minWhen to make it safe to clear minNextWhen.
   874  	// We read minWhen after reading minNextWhen so that
   875  	// if we see a cleared minNextWhen, we are guaranteed to see
   876  	// the updated minWhen.
   877  	nextWhen := ts.minWhenModified.Load()
   878  	when := ts.minWhenHeap.Load()
   879  	if when == 0 || (nextWhen != 0 && nextWhen < when) {
   880  		when = nextWhen
   881  	}
   882  	return when
   883  }
   884  
   885  // check runs any timers in ts that are ready.
   886  // If now is not 0 it is the current time.
   887  // It returns the passed time or the current time if now was passed as 0.
   888  // and the time when the next timer should run or 0 if there is no next timer,
   889  // and reports whether it ran any timers.
   890  // If the time when the next timer should run is not 0,
   891  // it is always larger than the returned time.
   892  // We pass now in and out to avoid extra calls of nanotime.
   893  //
   894  //go:yeswritebarrierrec
   895  func (ts *timers) check(now int64) (rnow, pollUntil int64, ran bool) {
   896  	ts.trace("check")
   897  	// If it's not yet time for the first timer, or the first adjusted
   898  	// timer, then there is nothing to do.
   899  	next := ts.wakeTime()
   900  	if next == 0 {
   901  		// No timers to run or adjust.
   902  		return now, 0, false
   903  	}
   904  
   905  	if now == 0 {
   906  		now = nanotime()
   907  	}
   908  
   909  	// If this is the local P, and there are a lot of deleted timers,
   910  	// clear them out. We only do this for the local P to reduce
   911  	// lock contention on timersLock.
   912  	zombies := ts.zombies.Load()
   913  	if zombies < 0 {
   914  		badTimer()
   915  	}
   916  	force := ts == &getg().m.p.ptr().timers && int(zombies) > int(ts.len.Load())/4
   917  
   918  	if now < next && !force {
   919  		// Next timer is not ready to run, and we don't need to clear deleted timers.
   920  		return now, next, false
   921  	}
   922  
   923  	ts.lock()
   924  	if len(ts.heap) > 0 {
   925  		ts.adjust(now, false)
   926  		for len(ts.heap) > 0 {
   927  			// Note that runtimer may temporarily unlock ts.
   928  			if tw := ts.run(now); tw != 0 {
   929  				if tw > 0 {
   930  					pollUntil = tw
   931  				}
   932  				break
   933  			}
   934  			ran = true
   935  		}
   936  
   937  		// Note: Delaying the forced adjustment until after the ts.run
   938  		// (as opposed to calling ts.adjust(now, force) above)
   939  		// is significantly faster under contention, such as in
   940  		// package time's BenchmarkTimerAdjust10000,
   941  		// though we do not fully understand why.
   942  		force = ts == &getg().m.p.ptr().timers && int(ts.zombies.Load()) > int(ts.len.Load())/4
   943  		if force {
   944  			ts.adjust(now, true)
   945  		}
   946  	}
   947  	ts.unlock()
   948  
   949  	return now, pollUntil, ran
   950  }
   951  
   952  // run examines the first timer in ts. If it is ready based on now,
   953  // it runs the timer and removes or updates it.
   954  // Returns 0 if it ran a timer, -1 if there are no more timers, or the time
   955  // when the first timer should run.
   956  // The caller must have locked ts.
   957  // If a timer is run, this will temporarily unlock ts.
   958  //
   959  //go:systemstack
   960  func (ts *timers) run(now int64) int64 {
   961  	ts.trace("run")
   962  	assertLockHeld(&ts.mu)
   963  Redo:
   964  	if len(ts.heap) == 0 {
   965  		return -1
   966  	}
   967  	tw := ts.heap[0]
   968  	t := tw.timer
   969  	if t.ts != ts {
   970  		throw("bad ts")
   971  	}
   972  
   973  	if t.astate.Load()&(timerModified|timerZombie) == 0 && tw.when > now {
   974  		// Fast path: not ready to run.
   975  		return tw.when
   976  	}
   977  
   978  	t.lock()
   979  	if t.updateHeap() {
   980  		t.unlock()
   981  		goto Redo
   982  	}
   983  
   984  	if t.state&timerHeaped == 0 || t.state&timerModified != 0 {
   985  		badTimer()
   986  	}
   987  
   988  	if t.when > now {
   989  		// Not ready to run.
   990  		t.unlock()
   991  		return t.when
   992  	}
   993  
   994  	t.unlockAndRun(now)
   995  	assertLockHeld(&ts.mu) // t is unlocked now, but not ts
   996  	return 0
   997  }
   998  
   999  // unlockAndRun unlocks and runs the timer t (which must be locked).
  1000  // If t is in a timer set (t.ts != nil), the caller must also have locked the timer set,
  1001  // and this call will temporarily unlock the timer set while running the timer function.
  1002  // unlockAndRun returns with t unlocked and t.ts (re-)locked.
  1003  //
  1004  //go:systemstack
  1005  func (t *timer) unlockAndRun(now int64) {
  1006  	t.trace("unlockAndRun")
  1007  	assertLockHeld(&t.mu)
  1008  	if t.ts != nil {
  1009  		assertLockHeld(&t.ts.mu)
  1010  	}
  1011  	if raceenabled {
  1012  		// Note that we are running on a system stack,
  1013  		// so there is no chance of getg().m being reassigned
  1014  		// out from under us while this function executes.
  1015  		tsLocal := &getg().m.p.ptr().timers
  1016  		if tsLocal.raceCtx == 0 {
  1017  			tsLocal.raceCtx = racegostart(abi.FuncPCABIInternal((*timers).run) + sys.PCQuantum)
  1018  		}
  1019  		raceacquirectx(tsLocal.raceCtx, unsafe.Pointer(t))
  1020  	}
  1021  
  1022  	if t.state&(timerModified|timerZombie) != 0 {
  1023  		badTimer()
  1024  	}
  1025  
  1026  	f := t.f
  1027  	arg := t.arg
  1028  	seq := t.seq
  1029  	var next int64
  1030  	delay := now - t.when
  1031  	if t.period > 0 {
  1032  		// Leave in heap but adjust next time to fire.
  1033  		next = t.when + t.period*(1+delay/t.period)
  1034  		if next < 0 { // check for overflow.
  1035  			next = maxWhen
  1036  		}
  1037  	} else {
  1038  		next = 0
  1039  	}
  1040  	ts := t.ts
  1041  	t.when = next
  1042  	if t.state&timerHeaped != 0 {
  1043  		t.state |= timerModified
  1044  		if next == 0 {
  1045  			t.state |= timerZombie
  1046  			t.ts.zombies.Add(1)
  1047  		}
  1048  		t.updateHeap()
  1049  	}
  1050  
  1051  	async := debug.asynctimerchan.Load() != 0
  1052  	if !async && t.isChan && t.period == 0 {
  1053  		// Tell Stop/Reset that we are sending a value.
  1054  		if t.isSending.Add(1) < 0 {
  1055  			throw("too many concurrent timer firings")
  1056  		}
  1057  	}
  1058  
  1059  	t.unlock()
  1060  
  1061  	if raceenabled {
  1062  		// Temporarily use the current P's racectx for g0.
  1063  		gp := getg()
  1064  		if gp.racectx != 0 {
  1065  			throw("unexpected racectx")
  1066  		}
  1067  		gp.racectx = gp.m.p.ptr().timers.raceCtx
  1068  	}
  1069  
  1070  	if ts != nil {
  1071  		ts.unlock()
  1072  	}
  1073  
  1074  	if !async && t.isChan {
  1075  		// For a timer channel, we want to make sure that no stale sends
  1076  		// happen after a t.stop or t.modify, but we cannot hold t.mu
  1077  		// during the actual send (which f does) due to lock ordering.
  1078  		// It can happen that we are holding t's lock above, we decide
  1079  		// it's time to send a time value (by calling f), grab the parameters,
  1080  		// unlock above, and then a t.stop or t.modify changes the timer
  1081  		// and returns. At that point, the send needs not to happen after all.
  1082  		// The way we arrange for it not to happen is that t.stop and t.modify
  1083  		// both increment t.seq while holding both t.mu and t.sendLock.
  1084  		// We copied the seq value above while holding t.mu.
  1085  		// Now we can acquire t.sendLock (which will be held across the send)
  1086  		// and double-check that t.seq is still the seq value we saw above.
  1087  		// If not, the timer has been updated and we should skip the send.
  1088  		// We skip the send by reassigning f to a no-op function.
  1089  		//
  1090  		// The isSending field tells t.stop or t.modify that we have
  1091  		// started to send the value. That lets them correctly return
  1092  		// true meaning that no value was sent.
  1093  		lock(&t.sendLock)
  1094  
  1095  		if t.period == 0 {
  1096  			// We are committed to possibly sending a value
  1097  			// based on seq, so no need to keep telling
  1098  			// stop/modify that we are sending.
  1099  			if t.isSending.Add(-1) < 0 {
  1100  				throw("mismatched isSending updates")
  1101  			}
  1102  		}
  1103  
  1104  		if t.seq != seq {
  1105  			f = func(any, uintptr, int64) {}
  1106  		}
  1107  	}
  1108  
  1109  	f(arg, seq, delay)
  1110  
  1111  	if !async && t.isChan {
  1112  		unlock(&t.sendLock)
  1113  	}
  1114  
  1115  	if ts != nil {
  1116  		ts.lock()
  1117  	}
  1118  
  1119  	if raceenabled {
  1120  		gp := getg()
  1121  		gp.racectx = 0
  1122  	}
  1123  }
  1124  
  1125  // verifyTimerHeap verifies that the timers is in a valid state.
  1126  // This is only for debugging, and is only called if verifyTimers is true.
  1127  // The caller must have locked ts.
  1128  func (ts *timers) verify() {
  1129  	assertLockHeld(&ts.mu)
  1130  	for i, tw := range ts.heap {
  1131  		if i == 0 {
  1132  			// First timer has no parent.
  1133  			continue
  1134  		}
  1135  
  1136  		// The heap is timerHeapN-ary. See siftupTimer and siftdownTimer.
  1137  		p := int(uint(i-1) / timerHeapN)
  1138  		if tw.when < ts.heap[p].when {
  1139  			print("bad timer heap at ", i, ": ", p, ": ", ts.heap[p].when, ", ", i, ": ", tw.when, "\n")
  1140  			throw("bad timer heap")
  1141  		}
  1142  	}
  1143  	if n := int(ts.len.Load()); len(ts.heap) != n {
  1144  		println("timer heap len", len(ts.heap), "!= atomic len", n)
  1145  		throw("bad timer heap len")
  1146  	}
  1147  }
  1148  
  1149  // updateMinWhenHeap sets ts.minWhenHeap to ts.heap[0].when.
  1150  // The caller must have locked ts or the world must be stopped.
  1151  func (ts *timers) updateMinWhenHeap() {
  1152  	assertWorldStoppedOrLockHeld(&ts.mu)
  1153  	if len(ts.heap) == 0 {
  1154  		ts.minWhenHeap.Store(0)
  1155  	} else {
  1156  		ts.minWhenHeap.Store(ts.heap[0].when)
  1157  	}
  1158  }
  1159  
  1160  // updateMinWhenModified updates ts.minWhenModified to be <= when.
  1161  // ts need not be (and usually is not) locked.
  1162  func (ts *timers) updateMinWhenModified(when int64) {
  1163  	for {
  1164  		old := ts.minWhenModified.Load()
  1165  		if old != 0 && old < when {
  1166  			return
  1167  		}
  1168  		if ts.minWhenModified.CompareAndSwap(old, when) {
  1169  			return
  1170  		}
  1171  	}
  1172  }
  1173  
  1174  // timeSleepUntil returns the time when the next timer should fire. Returns
  1175  // maxWhen if there are no timers.
  1176  // This is only called by sysmon and checkdead.
  1177  func timeSleepUntil() int64 {
  1178  	next := int64(maxWhen)
  1179  
  1180  	// Prevent allp slice changes. This is like retake.
  1181  	lock(&allpLock)
  1182  	for _, pp := range allp {
  1183  		if pp == nil {
  1184  			// This can happen if procresize has grown
  1185  			// allp but not yet created new Ps.
  1186  			continue
  1187  		}
  1188  
  1189  		if w := pp.timers.wakeTime(); w != 0 {
  1190  			next = min(next, w)
  1191  		}
  1192  	}
  1193  	unlock(&allpLock)
  1194  
  1195  	return next
  1196  }
  1197  
  1198  const timerHeapN = 4
  1199  
  1200  // Heap maintenance algorithms.
  1201  // These algorithms check for slice index errors manually.
  1202  // Slice index error can happen if the program is using racy
  1203  // access to timers. We don't want to panic here, because
  1204  // it will cause the program to crash with a mysterious
  1205  // "panic holding locks" message. Instead, we panic while not
  1206  // holding a lock.
  1207  
  1208  // siftUp puts the timer at position i in the right place
  1209  // in the heap by moving it up toward the top of the heap.
  1210  func (ts *timers) siftUp(i int) {
  1211  	heap := ts.heap
  1212  	if i >= len(heap) {
  1213  		badTimer()
  1214  	}
  1215  	tw := heap[i]
  1216  	when := tw.when
  1217  	if when <= 0 {
  1218  		badTimer()
  1219  	}
  1220  	for i > 0 {
  1221  		p := int(uint(i-1) / timerHeapN) // parent
  1222  		if when >= heap[p].when {
  1223  			break
  1224  		}
  1225  		heap[i] = heap[p]
  1226  		i = p
  1227  	}
  1228  	if heap[i].timer != tw.timer {
  1229  		heap[i] = tw
  1230  	}
  1231  }
  1232  
  1233  // siftDown puts the timer at position i in the right place
  1234  // in the heap by moving it down toward the bottom of the heap.
  1235  func (ts *timers) siftDown(i int) {
  1236  	heap := ts.heap
  1237  	n := len(heap)
  1238  	if i >= n {
  1239  		badTimer()
  1240  	}
  1241  	if i*timerHeapN+1 >= n {
  1242  		return
  1243  	}
  1244  	tw := heap[i]
  1245  	when := tw.when
  1246  	if when <= 0 {
  1247  		badTimer()
  1248  	}
  1249  	for {
  1250  		leftChild := i*timerHeapN + 1
  1251  		if leftChild >= n {
  1252  			break
  1253  		}
  1254  		w := when
  1255  		c := -1
  1256  		for j, tw := range heap[leftChild:min(leftChild+timerHeapN, n)] {
  1257  			if tw.when < w {
  1258  				w = tw.when
  1259  				c = leftChild + j
  1260  			}
  1261  		}
  1262  		if c < 0 {
  1263  			break
  1264  		}
  1265  		heap[i] = heap[c]
  1266  		i = c
  1267  	}
  1268  	if heap[i].timer != tw.timer {
  1269  		heap[i] = tw
  1270  	}
  1271  }
  1272  
  1273  // initHeap reestablishes the heap order in the slice ts.heap.
  1274  // It takes O(n) time for n=len(ts.heap), not the O(n log n) of n repeated add operations.
  1275  func (ts *timers) initHeap() {
  1276  	// Last possible element that needs sifting down is parent of last element;
  1277  	// last element is len(t)-1; parent of last element is (len(t)-1-1)/timerHeapN.
  1278  	if len(ts.heap) <= 1 {
  1279  		return
  1280  	}
  1281  	for i := int(uint(len(ts.heap)-1-1) / timerHeapN); i >= 0; i-- {
  1282  		ts.siftDown(i)
  1283  	}
  1284  }
  1285  
  1286  // badTimer is called if the timer data structures have been corrupted,
  1287  // presumably due to racy use by the program. We panic here rather than
  1288  // panicking due to invalid slice access while holding locks.
  1289  // See issue #25686.
  1290  func badTimer() {
  1291  	throw("timer data corruption")
  1292  }
  1293  
  1294  // Timer channels.
  1295  
  1296  // maybeRunChan checks whether the timer needs to run
  1297  // to send a value to its associated channel. If so, it does.
  1298  // The timer must not be locked.
  1299  func (t *timer) maybeRunChan() {
  1300  	if t.astate.Load()&timerHeaped != 0 {
  1301  		// If the timer is in the heap, the ordinary timer code
  1302  		// is in charge of sending when appropriate.
  1303  		return
  1304  	}
  1305  
  1306  	t.lock()
  1307  	now := nanotime()
  1308  	if t.state&timerHeaped != 0 || t.when == 0 || t.when > now {
  1309  		t.trace("maybeRunChan-")
  1310  		// Timer in the heap, or not running at all, or not triggered.
  1311  		t.unlock()
  1312  		return
  1313  	}
  1314  	t.trace("maybeRunChan+")
  1315  	systemstack(func() {
  1316  		t.unlockAndRun(now)
  1317  	})
  1318  }
  1319  
  1320  // blockTimerChan is called when a channel op has decided to block on c.
  1321  // The caller holds the channel lock for c and possibly other channels.
  1322  // blockTimerChan makes sure that c is in a timer heap,
  1323  // adding it if needed.
  1324  func blockTimerChan(c *hchan) {
  1325  	t := c.timer
  1326  	t.lock()
  1327  	t.trace("blockTimerChan")
  1328  	if !t.isChan {
  1329  		badTimer()
  1330  	}
  1331  
  1332  	t.blocked++
  1333  
  1334  	// If this is the first enqueue after a recent dequeue,
  1335  	// the timer may still be in the heap but marked as a zombie.
  1336  	// Unmark it in this case, if the timer is still pending.
  1337  	if t.state&timerHeaped != 0 && t.state&timerZombie != 0 && t.when > 0 {
  1338  		t.state &^= timerZombie
  1339  		t.ts.zombies.Add(-1)
  1340  	}
  1341  
  1342  	// t.maybeAdd must be called with t unlocked,
  1343  	// because it needs to lock t.ts before t.
  1344  	// Then it will do nothing if t.needsAdd(state) is false.
  1345  	// Check that now before the unlock,
  1346  	// avoiding the extra lock-lock-unlock-unlock
  1347  	// inside maybeAdd when t does not need to be added.
  1348  	add := t.needsAdd()
  1349  	t.unlock()
  1350  	if add {
  1351  		t.maybeAdd()
  1352  	}
  1353  }
  1354  
  1355  // unblockTimerChan is called when a channel op that was blocked on c
  1356  // is no longer blocked. Every call to blockTimerChan must be paired with
  1357  // a call to unblockTimerChan.
  1358  // The caller holds the channel lock for c and possibly other channels.
  1359  // unblockTimerChan removes c from the timer heap when nothing is
  1360  // blocked on it anymore.
  1361  func unblockTimerChan(c *hchan) {
  1362  	t := c.timer
  1363  	t.lock()
  1364  	t.trace("unblockTimerChan")
  1365  	if !t.isChan || t.blocked == 0 {
  1366  		badTimer()
  1367  	}
  1368  	t.blocked--
  1369  	if t.blocked == 0 && t.state&timerHeaped != 0 && t.state&timerZombie == 0 {
  1370  		// Last goroutine that was blocked on this timer.
  1371  		// Mark for removal from heap but do not clear t.when,
  1372  		// so that we know what time it is still meant to trigger.
  1373  		t.state |= timerZombie
  1374  		t.ts.zombies.Add(1)
  1375  	}
  1376  	t.unlock()
  1377  }
  1378  

View as plain text