sema.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  // Semaphore implementation exposed to Go.
     6  // Intended use is provide a sleep and wakeup
     7  // primitive that can be used in the contended case
     8  // of other synchronization primitives.
     9  // Thus it targets the same goal as Linux's futex,
    10  // but it has much simpler semantics.
    11  //
    12  // That is, don't think of these as semaphores.
    13  // Think of them as a way to implement sleep and wakeup
    14  // such that every sleep is paired with a single wakeup,
    15  // even if, due to races, the wakeup happens before the sleep.
    16  //
    17  // See Mullender and Cox, ``Semaphores in Plan 9,''
    18  // https://swtch.com/semaphore.pdf
    19  
    20  package runtime
    21  
    22  import (
    23  	"internal/cpu"
    24  	"internal/runtime/atomic"
    25  	"unsafe"
    26  )
    27  
    28  // Asynchronous semaphore for sync.Mutex.
    29  
    30  // A semaRoot holds a balanced tree of sudog with distinct addresses (s.elem).
    31  // Each of those sudog may in turn point (through s.waitlink) to a list
    32  // of other sudogs waiting on the same address.
    33  // The operations on the inner lists of sudogs with the same address
    34  // are all O(1). The scanning of the top-level semaRoot list is O(log n),
    35  // where n is the number of distinct addresses with goroutines blocked
    36  // on them that hash to the given semaRoot.
    37  // See golang.org/issue/17953 for a program that worked badly
    38  // before we introduced the second level of list, and
    39  // BenchmarkSemTable/OneAddrCollision/* for a benchmark that exercises this.
    40  type semaRoot struct {
    41  	lock  mutex
    42  	treap *sudog        // root of balanced tree of unique waiters.
    43  	nwait atomic.Uint32 // Number of waiters. Read w/o the lock.
    44  }
    45  
    46  var semtable semTable
    47  
    48  // Prime to not correlate with any user patterns.
    49  const semTabSize = 251
    50  
    51  type semTable [semTabSize]struct {
    52  	root semaRoot
    53  	pad  [cpu.CacheLinePadSize - unsafe.Sizeof(semaRoot{})]byte
    54  }
    55  
    56  func (t *semTable) rootFor(addr *uint32) *semaRoot {
    57  	return &t[(uintptr(unsafe.Pointer(addr))>>3)%semTabSize].root
    58  }
    59  
    60  // sync_runtime_Semacquire should be an internal detail,
    61  // but widely used packages access it using linkname.
    62  // Notable members of the hall of shame include:
    63  //   - gvisor.dev/gvisor
    64  //   - github.com/sagernet/gvisor
    65  //
    66  // Do not remove or change the type signature.
    67  // See go.dev/issue/67401.
    68  //
    69  //go:linkname sync_runtime_Semacquire sync.runtime_Semacquire
    70  func sync_runtime_Semacquire(addr *uint32) {
    71  	semacquire1(addr, false, semaBlockProfile, 0, waitReasonSemacquire)
    72  }
    73  
    74  //go:linkname poll_runtime_Semacquire internal/poll.runtime_Semacquire
    75  func poll_runtime_Semacquire(addr *uint32) {
    76  	semacquire1(addr, false, semaBlockProfile, 0, waitReasonSemacquire)
    77  }
    78  
    79  // sync_runtime_Semrelease should be an internal detail,
    80  // but widely used packages access it using linkname.
    81  // Notable members of the hall of shame include:
    82  //   - gvisor.dev/gvisor
    83  //   - github.com/sagernet/gvisor
    84  //
    85  // Do not remove or change the type signature.
    86  // See go.dev/issue/67401.
    87  //
    88  //go:linkname sync_runtime_Semrelease sync.runtime_Semrelease
    89  func sync_runtime_Semrelease(addr *uint32, handoff bool, skipframes int) {
    90  	semrelease1(addr, handoff, skipframes)
    91  }
    92  
    93  //go:linkname sync_runtime_SemacquireMutex sync.runtime_SemacquireMutex
    94  func sync_runtime_SemacquireMutex(addr *uint32, lifo bool, skipframes int) {
    95  	semacquire1(addr, lifo, semaBlockProfile|semaMutexProfile, skipframes, waitReasonSyncMutexLock)
    96  }
    97  
    98  //go:linkname sync_runtime_SemacquireRWMutexR sync.runtime_SemacquireRWMutexR
    99  func sync_runtime_SemacquireRWMutexR(addr *uint32, lifo bool, skipframes int) {
   100  	semacquire1(addr, lifo, semaBlockProfile|semaMutexProfile, skipframes, waitReasonSyncRWMutexRLock)
   101  }
   102  
   103  //go:linkname sync_runtime_SemacquireRWMutex sync.runtime_SemacquireRWMutex
   104  func sync_runtime_SemacquireRWMutex(addr *uint32, lifo bool, skipframes int) {
   105  	semacquire1(addr, lifo, semaBlockProfile|semaMutexProfile, skipframes, waitReasonSyncRWMutexLock)
   106  }
   107  
   108  //go:linkname poll_runtime_Semrelease internal/poll.runtime_Semrelease
   109  func poll_runtime_Semrelease(addr *uint32) {
   110  	semrelease(addr)
   111  }
   112  
   113  func readyWithTime(s *sudog, traceskip int) {
   114  	if s.releasetime != 0 {
   115  		s.releasetime = cputicks()
   116  	}
   117  	goready(s.g, traceskip)
   118  }
   119  
   120  type semaProfileFlags int
   121  
   122  const (
   123  	semaBlockProfile semaProfileFlags = 1 << iota
   124  	semaMutexProfile
   125  )
   126  
   127  // Called from runtime.
   128  func semacquire(addr *uint32) {
   129  	semacquire1(addr, false, 0, 0, waitReasonSemacquire)
   130  }
   131  
   132  func semacquire1(addr *uint32, lifo bool, profile semaProfileFlags, skipframes int, reason waitReason) {
   133  	gp := getg()
   134  	if gp != gp.m.curg {
   135  		throw("semacquire not on the G stack")
   136  	}
   137  
   138  	// Easy case.
   139  	if cansemacquire(addr) {
   140  		return
   141  	}
   142  
   143  	// Harder case:
   144  	//	increment waiter count
   145  	//	try cansemacquire one more time, return if succeeded
   146  	//	enqueue itself as a waiter
   147  	//	sleep
   148  	//	(waiter descriptor is dequeued by signaler)
   149  	s := acquireSudog()
   150  	root := semtable.rootFor(addr)
   151  	t0 := int64(0)
   152  	s.releasetime = 0
   153  	s.acquiretime = 0
   154  	s.ticket = 0
   155  	if profile&semaBlockProfile != 0 && blockprofilerate > 0 {
   156  		t0 = cputicks()
   157  		s.releasetime = -1
   158  	}
   159  	if profile&semaMutexProfile != 0 && mutexprofilerate > 0 {
   160  		if t0 == 0 {
   161  			t0 = cputicks()
   162  		}
   163  		s.acquiretime = t0
   164  	}
   165  	for {
   166  		lockWithRank(&root.lock, lockRankRoot)
   167  		// Add ourselves to nwait to disable "easy case" in semrelease.
   168  		root.nwait.Add(1)
   169  		// Check cansemacquire to avoid missed wakeup.
   170  		if cansemacquire(addr) {
   171  			root.nwait.Add(-1)
   172  			unlock(&root.lock)
   173  			break
   174  		}
   175  		// Any semrelease after the cansemacquire knows we're waiting
   176  		// (we set nwait above), so go to sleep.
   177  		root.queue(addr, s, lifo)
   178  		goparkunlock(&root.lock, reason, traceBlockSync, 4+skipframes)
   179  		if s.ticket != 0 || cansemacquire(addr) {
   180  			break
   181  		}
   182  	}
   183  	if s.releasetime > 0 {
   184  		blockevent(s.releasetime-t0, 3+skipframes)
   185  	}
   186  	releaseSudog(s)
   187  }
   188  
   189  func semrelease(addr *uint32) {
   190  	semrelease1(addr, false, 0)
   191  }
   192  
   193  func semrelease1(addr *uint32, handoff bool, skipframes int) {
   194  	root := semtable.rootFor(addr)
   195  	atomic.Xadd(addr, 1)
   196  
   197  	// Easy case: no waiters?
   198  	// This check must happen after the xadd, to avoid a missed wakeup
   199  	// (see loop in semacquire).
   200  	if root.nwait.Load() == 0 {
   201  		return
   202  	}
   203  
   204  	// Harder case: search for a waiter and wake it.
   205  	lockWithRank(&root.lock, lockRankRoot)
   206  	if root.nwait.Load() == 0 {
   207  		// The count is already consumed by another goroutine,
   208  		// so no need to wake up another goroutine.
   209  		unlock(&root.lock)
   210  		return
   211  	}
   212  	s, t0, tailtime := root.dequeue(addr)
   213  	if s != nil {
   214  		root.nwait.Add(-1)
   215  	}
   216  	unlock(&root.lock)
   217  	if s != nil { // May be slow or even yield, so unlock first
   218  		acquiretime := s.acquiretime
   219  		if acquiretime != 0 {
   220  			// Charge contention that this (delayed) unlock caused.
   221  			// If there are N more goroutines waiting beyond the
   222  			// one that's waking up, charge their delay as well, so that
   223  			// contention holding up many goroutines shows up as
   224  			// more costly than contention holding up a single goroutine.
   225  			// It would take O(N) time to calculate how long each goroutine
   226  			// has been waiting, so instead we charge avg(head-wait, tail-wait)*N.
   227  			// head-wait is the longest wait and tail-wait is the shortest.
   228  			// (When we do a lifo insertion, we preserve this property by
   229  			// copying the old head's acquiretime into the inserted new head.
   230  			// In that case the overall average may be slightly high, but that's fine:
   231  			// the average of the ends is only an approximation to the actual
   232  			// average anyway.)
   233  			// The root.dequeue above changed the head and tail acquiretime
   234  			// to the current time, so the next unlock will not re-count this contention.
   235  			dt0 := t0 - acquiretime
   236  			dt := dt0
   237  			if s.waiters != 0 {
   238  				dtail := t0 - tailtime
   239  				dt += (dtail + dt0) / 2 * int64(s.waiters)
   240  			}
   241  			mutexevent(dt, 3+skipframes)
   242  		}
   243  		if s.ticket != 0 {
   244  			throw("corrupted semaphore ticket")
   245  		}
   246  		if handoff && cansemacquire(addr) {
   247  			s.ticket = 1
   248  		}
   249  		readyWithTime(s, 5+skipframes)
   250  		if s.ticket == 1 && getg().m.locks == 0 {
   251  			// Direct G handoff
   252  			// readyWithTime has added the waiter G as runnext in the
   253  			// current P; we now call the scheduler so that we start running
   254  			// the waiter G immediately.
   255  			// Note that waiter inherits our time slice: this is desirable
   256  			// to avoid having a highly contended semaphore hog the P
   257  			// indefinitely. goyield is like Gosched, but it emits a
   258  			// "preempted" trace event instead and, more importantly, puts
   259  			// the current G on the local runq instead of the global one.
   260  			// We only do this in the starving regime (handoff=true), as in
   261  			// the non-starving case it is possible for a different waiter
   262  			// to acquire the semaphore while we are yielding/scheduling,
   263  			// and this would be wasteful. We wait instead to enter starving
   264  			// regime, and then we start to do direct handoffs of ticket and
   265  			// P.
   266  			// See issue 33747 for discussion.
   267  			goyield()
   268  		}
   269  	}
   270  }
   271  
   272  func cansemacquire(addr *uint32) bool {
   273  	for {
   274  		v := atomic.Load(addr)
   275  		if v == 0 {
   276  			return false
   277  		}
   278  		if atomic.Cas(addr, v, v-1) {
   279  			return true
   280  		}
   281  	}
   282  }
   283  
   284  // queue adds s to the blocked goroutines in semaRoot.
   285  func (root *semaRoot) queue(addr *uint32, s *sudog, lifo bool) {
   286  	s.g = getg()
   287  	s.elem = unsafe.Pointer(addr)
   288  	s.next = nil
   289  	s.prev = nil
   290  	s.waiters = 0
   291  
   292  	var last *sudog
   293  	pt := &root.treap
   294  	for t := *pt; t != nil; t = *pt {
   295  		if t.elem == unsafe.Pointer(addr) {
   296  			// Already have addr in list.
   297  			if lifo {
   298  				// Substitute s in t's place in treap.
   299  				*pt = s
   300  				s.ticket = t.ticket
   301  				s.acquiretime = t.acquiretime // preserve head acquiretime as oldest time
   302  				s.parent = t.parent
   303  				s.prev = t.prev
   304  				s.next = t.next
   305  				if s.prev != nil {
   306  					s.prev.parent = s
   307  				}
   308  				if s.next != nil {
   309  					s.next.parent = s
   310  				}
   311  				// Add t first in s's wait list.
   312  				s.waitlink = t
   313  				s.waittail = t.waittail
   314  				if s.waittail == nil {
   315  					s.waittail = t
   316  				}
   317  				s.waiters = t.waiters
   318  				if s.waiters+1 != 0 {
   319  					s.waiters++
   320  				}
   321  				t.parent = nil
   322  				t.prev = nil
   323  				t.next = nil
   324  				t.waittail = nil
   325  			} else {
   326  				// Add s to end of t's wait list.
   327  				if t.waittail == nil {
   328  					t.waitlink = s
   329  				} else {
   330  					t.waittail.waitlink = s
   331  				}
   332  				t.waittail = s
   333  				s.waitlink = nil
   334  				if t.waiters+1 != 0 {
   335  					t.waiters++
   336  				}
   337  			}
   338  			return
   339  		}
   340  		last = t
   341  		if uintptr(unsafe.Pointer(addr)) < uintptr(t.elem) {
   342  			pt = &t.prev
   343  		} else {
   344  			pt = &t.next
   345  		}
   346  	}
   347  
   348  	// Add s as new leaf in tree of unique addrs.
   349  	// The balanced tree is a treap using ticket as the random heap priority.
   350  	// That is, it is a binary tree ordered according to the elem addresses,
   351  	// but then among the space of possible binary trees respecting those
   352  	// addresses, it is kept balanced on average by maintaining a heap ordering
   353  	// on the ticket: s.ticket <= both s.prev.ticket and s.next.ticket.
   354  	// https://en.wikipedia.org/wiki/Treap
   355  	// https://faculty.washington.edu/aragon/pubs/rst89.pdf
   356  	//
   357  	// s.ticket compared with zero in couple of places, therefore set lowest bit.
   358  	// It will not affect treap's quality noticeably.
   359  	s.ticket = cheaprand() | 1
   360  	s.parent = last
   361  	*pt = s
   362  
   363  	// Rotate up into tree according to ticket (priority).
   364  	for s.parent != nil && s.parent.ticket > s.ticket {
   365  		if s.parent.prev == s {
   366  			root.rotateRight(s.parent)
   367  		} else {
   368  			if s.parent.next != s {
   369  				panic("semaRoot queue")
   370  			}
   371  			root.rotateLeft(s.parent)
   372  		}
   373  	}
   374  }
   375  
   376  // dequeue searches for and finds the first goroutine
   377  // in semaRoot blocked on addr.
   378  // If the sudog was being profiled, dequeue returns the time
   379  // at which it was woken up as now. Otherwise now is 0.
   380  // If there are additional entries in the wait list, dequeue
   381  // returns tailtime set to the last entry's acquiretime.
   382  // Otherwise tailtime is found.acquiretime.
   383  func (root *semaRoot) dequeue(addr *uint32) (found *sudog, now, tailtime int64) {
   384  	ps := &root.treap
   385  	s := *ps
   386  	for ; s != nil; s = *ps {
   387  		if s.elem == unsafe.Pointer(addr) {
   388  			goto Found
   389  		}
   390  		if uintptr(unsafe.Pointer(addr)) < uintptr(s.elem) {
   391  			ps = &s.prev
   392  		} else {
   393  			ps = &s.next
   394  		}
   395  	}
   396  	return nil, 0, 0
   397  
   398  Found:
   399  	now = int64(0)
   400  	if s.acquiretime != 0 {
   401  		now = cputicks()
   402  	}
   403  	if t := s.waitlink; t != nil {
   404  		// Substitute t, also waiting on addr, for s in root tree of unique addrs.
   405  		*ps = t
   406  		t.ticket = s.ticket
   407  		t.parent = s.parent
   408  		t.prev = s.prev
   409  		if t.prev != nil {
   410  			t.prev.parent = t
   411  		}
   412  		t.next = s.next
   413  		if t.next != nil {
   414  			t.next.parent = t
   415  		}
   416  		if t.waitlink != nil {
   417  			t.waittail = s.waittail
   418  		} else {
   419  			t.waittail = nil
   420  		}
   421  		t.waiters = s.waiters
   422  		if t.waiters > 1 {
   423  			t.waiters--
   424  		}
   425  		// Set head and tail acquire time to 'now',
   426  		// because the caller will take care of charging
   427  		// the delays before now for all entries in the list.
   428  		t.acquiretime = now
   429  		tailtime = s.waittail.acquiretime
   430  		s.waittail.acquiretime = now
   431  		s.waitlink = nil
   432  		s.waittail = nil
   433  	} else {
   434  		// Rotate s down to be leaf of tree for removal, respecting priorities.
   435  		for s.next != nil || s.prev != nil {
   436  			if s.next == nil || s.prev != nil && s.prev.ticket < s.next.ticket {
   437  				root.rotateRight(s)
   438  			} else {
   439  				root.rotateLeft(s)
   440  			}
   441  		}
   442  		// Remove s, now a leaf.
   443  		if s.parent != nil {
   444  			if s.parent.prev == s {
   445  				s.parent.prev = nil
   446  			} else {
   447  				s.parent.next = nil
   448  			}
   449  		} else {
   450  			root.treap = nil
   451  		}
   452  		tailtime = s.acquiretime
   453  	}
   454  	s.parent = nil
   455  	s.elem = nil
   456  	s.next = nil
   457  	s.prev = nil
   458  	s.ticket = 0
   459  	return s, now, tailtime
   460  }
   461  
   462  // rotateLeft rotates the tree rooted at node x.
   463  // turning (x a (y b c)) into (y (x a b) c).
   464  func (root *semaRoot) rotateLeft(x *sudog) {
   465  	// p -> (x a (y b c))
   466  	p := x.parent
   467  	y := x.next
   468  	b := y.prev
   469  
   470  	y.prev = x
   471  	x.parent = y
   472  	x.next = b
   473  	if b != nil {
   474  		b.parent = x
   475  	}
   476  
   477  	y.parent = p
   478  	if p == nil {
   479  		root.treap = y
   480  	} else if p.prev == x {
   481  		p.prev = y
   482  	} else {
   483  		if p.next != x {
   484  			throw("semaRoot rotateLeft")
   485  		}
   486  		p.next = y
   487  	}
   488  }
   489  
   490  // rotateRight rotates the tree rooted at node y.
   491  // turning (y (x a b) c) into (x a (y b c)).
   492  func (root *semaRoot) rotateRight(y *sudog) {
   493  	// p -> (y (x a b) c)
   494  	p := y.parent
   495  	x := y.prev
   496  	b := x.next
   497  
   498  	x.next = y
   499  	y.parent = x
   500  	y.prev = b
   501  	if b != nil {
   502  		b.parent = y
   503  	}
   504  
   505  	x.parent = p
   506  	if p == nil {
   507  		root.treap = x
   508  	} else if p.prev == y {
   509  		p.prev = x
   510  	} else {
   511  		if p.next != y {
   512  			throw("semaRoot rotateRight")
   513  		}
   514  		p.next = x
   515  	}
   516  }
   517  
   518  // notifyList is a ticket-based notification list used to implement sync.Cond.
   519  //
   520  // It must be kept in sync with the sync package.
   521  type notifyList struct {
   522  	// wait is the ticket number of the next waiter. It is atomically
   523  	// incremented outside the lock.
   524  	wait atomic.Uint32
   525  
   526  	// notify is the ticket number of the next waiter to be notified. It can
   527  	// be read outside the lock, but is only written to with lock held.
   528  	//
   529  	// Both wait & notify can wrap around, and such cases will be correctly
   530  	// handled as long as their "unwrapped" difference is bounded by 2^31.
   531  	// For this not to be the case, we'd need to have 2^31+ goroutines
   532  	// blocked on the same condvar, which is currently not possible.
   533  	notify uint32
   534  
   535  	// List of parked waiters.
   536  	lock mutex
   537  	head *sudog
   538  	tail *sudog
   539  }
   540  
   541  // less checks if a < b, considering a & b running counts that may overflow the
   542  // 32-bit range, and that their "unwrapped" difference is always less than 2^31.
   543  func less(a, b uint32) bool {
   544  	return int32(a-b) < 0
   545  }
   546  
   547  // notifyListAdd adds the caller to a notify list such that it can receive
   548  // notifications. The caller must eventually call notifyListWait to wait for
   549  // such a notification, passing the returned ticket number.
   550  //
   551  //go:linkname notifyListAdd sync.runtime_notifyListAdd
   552  func notifyListAdd(l *notifyList) uint32 {
   553  	// This may be called concurrently, for example, when called from
   554  	// sync.Cond.Wait while holding a RWMutex in read mode.
   555  	return l.wait.Add(1) - 1
   556  }
   557  
   558  // notifyListWait waits for a notification. If one has been sent since
   559  // notifyListAdd was called, it returns immediately. Otherwise, it blocks.
   560  //
   561  //go:linkname notifyListWait sync.runtime_notifyListWait
   562  func notifyListWait(l *notifyList, t uint32) {
   563  	lockWithRank(&l.lock, lockRankNotifyList)
   564  
   565  	// Return right away if this ticket has already been notified.
   566  	if less(t, l.notify) {
   567  		unlock(&l.lock)
   568  		return
   569  	}
   570  
   571  	// Enqueue itself.
   572  	s := acquireSudog()
   573  	s.g = getg()
   574  	s.ticket = t
   575  	s.releasetime = 0
   576  	t0 := int64(0)
   577  	if blockprofilerate > 0 {
   578  		t0 = cputicks()
   579  		s.releasetime = -1
   580  	}
   581  	if l.tail == nil {
   582  		l.head = s
   583  	} else {
   584  		l.tail.next = s
   585  	}
   586  	l.tail = s
   587  	goparkunlock(&l.lock, waitReasonSyncCondWait, traceBlockCondWait, 3)
   588  	if t0 != 0 {
   589  		blockevent(s.releasetime-t0, 2)
   590  	}
   591  	releaseSudog(s)
   592  }
   593  
   594  // notifyListNotifyAll notifies all entries in the list.
   595  //
   596  //go:linkname notifyListNotifyAll sync.runtime_notifyListNotifyAll
   597  func notifyListNotifyAll(l *notifyList) {
   598  	// Fast-path: if there are no new waiters since the last notification
   599  	// we don't need to acquire the lock.
   600  	if l.wait.Load() == atomic.Load(&l.notify) {
   601  		return
   602  	}
   603  
   604  	// Pull the list out into a local variable, waiters will be readied
   605  	// outside the lock.
   606  	lockWithRank(&l.lock, lockRankNotifyList)
   607  	s := l.head
   608  	l.head = nil
   609  	l.tail = nil
   610  
   611  	// Update the next ticket to be notified. We can set it to the current
   612  	// value of wait because any previous waiters are already in the list
   613  	// or will notice that they have already been notified when trying to
   614  	// add themselves to the list.
   615  	atomic.Store(&l.notify, l.wait.Load())
   616  	unlock(&l.lock)
   617  
   618  	// Go through the local list and ready all waiters.
   619  	for s != nil {
   620  		next := s.next
   621  		s.next = nil
   622  		readyWithTime(s, 4)
   623  		s = next
   624  	}
   625  }
   626  
   627  // notifyListNotifyOne notifies one entry in the list.
   628  //
   629  //go:linkname notifyListNotifyOne sync.runtime_notifyListNotifyOne
   630  func notifyListNotifyOne(l *notifyList) {
   631  	// Fast-path: if there are no new waiters since the last notification
   632  	// we don't need to acquire the lock at all.
   633  	if l.wait.Load() == atomic.Load(&l.notify) {
   634  		return
   635  	}
   636  
   637  	lockWithRank(&l.lock, lockRankNotifyList)
   638  
   639  	// Re-check under the lock if we need to do anything.
   640  	t := l.notify
   641  	if t == l.wait.Load() {
   642  		unlock(&l.lock)
   643  		return
   644  	}
   645  
   646  	// Update the next notify ticket number.
   647  	atomic.Store(&l.notify, t+1)
   648  
   649  	// Try to find the g that needs to be notified.
   650  	// If it hasn't made it to the list yet we won't find it,
   651  	// but it won't park itself once it sees the new notify number.
   652  	//
   653  	// This scan looks linear but essentially always stops quickly.
   654  	// Because g's queue separately from taking numbers,
   655  	// there may be minor reorderings in the list, but we
   656  	// expect the g we're looking for to be near the front.
   657  	// The g has others in front of it on the list only to the
   658  	// extent that it lost the race, so the iteration will not
   659  	// be too long. This applies even when the g is missing:
   660  	// it hasn't yet gotten to sleep and has lost the race to
   661  	// the (few) other g's that we find on the list.
   662  	for p, s := (*sudog)(nil), l.head; s != nil; p, s = s, s.next {
   663  		if s.ticket == t {
   664  			n := s.next
   665  			if p != nil {
   666  				p.next = n
   667  			} else {
   668  				l.head = n
   669  			}
   670  			if n == nil {
   671  				l.tail = p
   672  			}
   673  			unlock(&l.lock)
   674  			s.next = nil
   675  			readyWithTime(s, 4)
   676  			return
   677  		}
   678  	}
   679  	unlock(&l.lock)
   680  }
   681  
   682  //go:linkname notifyListCheck sync.runtime_notifyListCheck
   683  func notifyListCheck(sz uintptr) {
   684  	if sz != unsafe.Sizeof(notifyList{}) {
   685  		print("runtime: bad notifyList size - sync=", sz, " runtime=", unsafe.Sizeof(notifyList{}), "\n")
   686  		throw("bad notifyList size")
   687  	}
   688  }
   689  
   690  //go:linkname sync_nanotime sync.runtime_nanotime
   691  func sync_nanotime() int64 {
   692  	return nanotime()
   693  }
   694
View as plain text