Source file src/runtime/trace.go

Documentation: runtime

     1  // Copyright 2014 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  // Go execution tracer.
     6  // The tracer captures a wide range of execution events like goroutine
     7  // creation/blocking/unblocking, syscall enter/exit/block, GC-related events,
     8  // changes of heap size, processor start/stop, etc and writes them to a buffer
     9  // in a compact form. A precise nanosecond-precision timestamp and a stack
    10  // trace is captured for most events.
    11  // See https://golang.org/s/go15trace for more info.
    12  
    13  package runtime
    14  
    15  import (
    16  	"runtime/internal/sys"
    17  	"unsafe"
    18  )
    19  
    20  // Event types in the trace, args are given in square brackets.
    21  const (
    22  	traceEvNone              = 0  // unused
    23  	traceEvBatch             = 1  // start of per-P batch of events [pid, timestamp]
    24  	traceEvFrequency         = 2  // contains tracer timer frequency [frequency (ticks per second)]
    25  	traceEvStack             = 3  // stack [stack id, number of PCs, array of {PC, func string ID, file string ID, line}]
    26  	traceEvGomaxprocs        = 4  // current value of GOMAXPROCS [timestamp, GOMAXPROCS, stack id]
    27  	traceEvProcStart         = 5  // start of P [timestamp, thread id]
    28  	traceEvProcStop          = 6  // stop of P [timestamp]
    29  	traceEvGCStart           = 7  // GC start [timestamp, seq, stack id]
    30  	traceEvGCDone            = 8  // GC done [timestamp]
    31  	traceEvGCSTWStart        = 9  // GC STW start [timestamp, kind]
    32  	traceEvGCSTWDone         = 10 // GC STW done [timestamp]
    33  	traceEvGCSweepStart      = 11 // GC sweep start [timestamp, stack id]
    34  	traceEvGCSweepDone       = 12 // GC sweep done [timestamp, swept, reclaimed]
    35  	traceEvGoCreate          = 13 // goroutine creation [timestamp, new goroutine id, new stack id, stack id]
    36  	traceEvGoStart           = 14 // goroutine starts running [timestamp, goroutine id, seq]
    37  	traceEvGoEnd             = 15 // goroutine ends [timestamp]
    38  	traceEvGoStop            = 16 // goroutine stops (like in select{}) [timestamp, stack]
    39  	traceEvGoSched           = 17 // goroutine calls Gosched [timestamp, stack]
    40  	traceEvGoPreempt         = 18 // goroutine is preempted [timestamp, stack]
    41  	traceEvGoSleep           = 19 // goroutine calls Sleep [timestamp, stack]
    42  	traceEvGoBlock           = 20 // goroutine blocks [timestamp, stack]
    43  	traceEvGoUnblock         = 21 // goroutine is unblocked [timestamp, goroutine id, seq, stack]
    44  	traceEvGoBlockSend       = 22 // goroutine blocks on chan send [timestamp, stack]
    45  	traceEvGoBlockRecv       = 23 // goroutine blocks on chan recv [timestamp, stack]
    46  	traceEvGoBlockSelect     = 24 // goroutine blocks on select [timestamp, stack]
    47  	traceEvGoBlockSync       = 25 // goroutine blocks on Mutex/RWMutex [timestamp, stack]
    48  	traceEvGoBlockCond       = 26 // goroutine blocks on Cond [timestamp, stack]
    49  	traceEvGoBlockNet        = 27 // goroutine blocks on network [timestamp, stack]
    50  	traceEvGoSysCall         = 28 // syscall enter [timestamp, stack]
    51  	traceEvGoSysExit         = 29 // syscall exit [timestamp, goroutine id, seq, real timestamp]
    52  	traceEvGoSysBlock        = 30 // syscall blocks [timestamp]
    53  	traceEvGoWaiting         = 31 // denotes that goroutine is blocked when tracing starts [timestamp, goroutine id]
    54  	traceEvGoInSyscall       = 32 // denotes that goroutine is in syscall when tracing starts [timestamp, goroutine id]
    55  	traceEvHeapAlloc         = 33 // memstats.heap_live change [timestamp, heap_alloc]
    56  	traceEvNextGC            = 34 // memstats.next_gc change [timestamp, next_gc]
    57  	traceEvTimerGoroutine    = 35 // denotes timer goroutine [timer goroutine id]
    58  	traceEvFutileWakeup      = 36 // denotes that the previous wakeup of this goroutine was futile [timestamp]
    59  	traceEvString            = 37 // string dictionary entry [ID, length, string]
    60  	traceEvGoStartLocal      = 38 // goroutine starts running on the same P as the last event [timestamp, goroutine id]
    61  	traceEvGoUnblockLocal    = 39 // goroutine is unblocked on the same P as the last event [timestamp, goroutine id, stack]
    62  	traceEvGoSysExitLocal    = 40 // syscall exit on the same P as the last event [timestamp, goroutine id, real timestamp]
    63  	traceEvGoStartLabel      = 41 // goroutine starts running with label [timestamp, goroutine id, seq, label string id]
    64  	traceEvGoBlockGC         = 42 // goroutine blocks on GC assist [timestamp, stack]
    65  	traceEvGCMarkAssistStart = 43 // GC mark assist start [timestamp, stack]
    66  	traceEvGCMarkAssistDone  = 44 // GC mark assist done [timestamp]
    67  	traceEvUserTaskCreate    = 45 // trace.NewContext [timestamp, internal task id, internal parent task id, stack, name string]
    68  	traceEvUserTaskEnd       = 46 // end of a task [timestamp, internal task id, stack]
    69  	traceEvUserRegion        = 47 // trace.WithRegion [timestamp, internal task id, mode(0:start, 1:end), stack, name string]
    70  	traceEvUserLog           = 48 // trace.Log [timestamp, internal task id, key string id, stack, value string]
    71  	traceEvCount             = 49
    72  	// Byte is used but only 6 bits are available for event type.
    73  	// The remaining 2 bits are used to specify the number of arguments.
    74  	// That means, the max event type value is 63.
    75  )
    76  
    77  const (
    78  	// Timestamps in trace are cputicks/traceTickDiv.
    79  	// This makes absolute values of timestamp diffs smaller,
    80  	// and so they are encoded in less number of bytes.
    81  	// 64 on x86 is somewhat arbitrary (one tick is ~20ns on a 3GHz machine).
    82  	// The suggested increment frequency for PowerPC's time base register is
    83  	// 512 MHz according to Power ISA v2.07 section 6.2, so we use 16 on ppc64
    84  	// and ppc64le.
    85  	// Tracing won't work reliably for architectures where cputicks is emulated
    86  	// by nanotime, so the value doesn't matter for those architectures.
    87  	traceTickDiv = 16 + 48*(sys.Goarch386|sys.GoarchAmd64|sys.GoarchAmd64p32)
    88  	// Maximum number of PCs in a single stack trace.
    89  	// Since events contain only stack id rather than whole stack trace,
    90  	// we can allow quite large values here.
    91  	traceStackSize = 128
    92  	// Identifier of a fake P that is used when we trace without a real P.
    93  	traceGlobProc = -1
    94  	// Maximum number of bytes to encode uint64 in base-128.
    95  	traceBytesPerNumber = 10
    96  	// Shift of the number of arguments in the first event byte.
    97  	traceArgCountShift = 6
    98  	// Flag passed to traceGoPark to denote that the previous wakeup of this
    99  	// goroutine was futile. For example, a goroutine was unblocked on a mutex,
   100  	// but another goroutine got ahead and acquired the mutex before the first
   101  	// goroutine is scheduled, so the first goroutine has to block again.
   102  	// Such wakeups happen on buffered channels and sync.Mutex,
   103  	// but are generally not interesting for end user.
   104  	traceFutileWakeup byte = 128
   105  )
   106  
   107  // trace is global tracing context.
   108  var trace struct {
   109  	lock          mutex       // protects the following members
   110  	lockOwner     *g          // to avoid deadlocks during recursive lock locks
   111  	enabled       bool        // when set runtime traces events
   112  	shutdown      bool        // set when we are waiting for trace reader to finish after setting enabled to false
   113  	headerWritten bool        // whether ReadTrace has emitted trace header
   114  	footerWritten bool        // whether ReadTrace has emitted trace footer
   115  	shutdownSema  uint32      // used to wait for ReadTrace completion
   116  	seqStart      uint64      // sequence number when tracing was started
   117  	ticksStart    int64       // cputicks when tracing was started
   118  	ticksEnd      int64       // cputicks when tracing was stopped
   119  	timeStart     int64       // nanotime when tracing was started
   120  	timeEnd       int64       // nanotime when tracing was stopped
   121  	seqGC         uint64      // GC start/done sequencer
   122  	reading       traceBufPtr // buffer currently handed off to user
   123  	empty         traceBufPtr // stack of empty buffers
   124  	fullHead      traceBufPtr // queue of full buffers
   125  	fullTail      traceBufPtr
   126  	reader        guintptr        // goroutine that called ReadTrace, or nil
   127  	stackTab      traceStackTable // maps stack traces to unique ids
   128  
   129  	// Dictionary for traceEvString.
   130  	//
   131  	// TODO: central lock to access the map is not ideal.
   132  	//   option: pre-assign ids to all user annotation region names and tags
   133  	//   option: per-P cache
   134  	//   option: sync.Map like data structure
   135  	stringsLock mutex
   136  	strings     map[string]uint64
   137  	stringSeq   uint64
   138  
   139  	// markWorkerLabels maps gcMarkWorkerMode to string ID.
   140  	markWorkerLabels [len(gcMarkWorkerModeStrings)]uint64
   141  
   142  	bufLock mutex       // protects buf
   143  	buf     traceBufPtr // global trace buffer, used when running without a p
   144  }
   145  
   146  // traceBufHeader is per-P tracing buffer.
   147  type traceBufHeader struct {
   148  	link      traceBufPtr             // in trace.empty/full
   149  	lastTicks uint64                  // when we wrote the last event
   150  	pos       int                     // next write offset in arr
   151  	stk       [traceStackSize]uintptr // scratch buffer for traceback
   152  }
   153  
   154  // traceBuf is per-P tracing buffer.
   155  //
   156  //go:notinheap
   157  type traceBuf struct {
   158  	traceBufHeader
   159  	arr [64<<10 - unsafe.Sizeof(traceBufHeader{})]byte // underlying buffer for traceBufHeader.buf
   160  }
   161  
   162  // traceBufPtr is a *traceBuf that is not traced by the garbage
   163  // collector and doesn't have write barriers. traceBufs are not
   164  // allocated from the GC'd heap, so this is safe, and are often
   165  // manipulated in contexts where write barriers are not allowed, so
   166  // this is necessary.
   167  //
   168  // TODO: Since traceBuf is now go:notinheap, this isn't necessary.
   169  type traceBufPtr uintptr
   170  
   171  func (tp traceBufPtr) ptr() *traceBuf   { return (*traceBuf)(unsafe.Pointer(tp)) }
   172  func (tp *traceBufPtr) set(b *traceBuf) { *tp = traceBufPtr(unsafe.Pointer(b)) }
   173  func traceBufPtrOf(b *traceBuf) traceBufPtr {
   174  	return traceBufPtr(unsafe.Pointer(b))
   175  }
   176  
   177  // StartTrace enables tracing for the current process.
   178  // While tracing, the data will be buffered and available via ReadTrace.
   179  // StartTrace returns an error if tracing is already enabled.
   180  // Most clients should use the runtime/trace package or the testing package's
   181  // -test.trace flag instead of calling StartTrace directly.
   182  func StartTrace() error {
   183  	// Stop the world, so that we can take a consistent snapshot
   184  	// of all goroutines at the beginning of the trace.
   185  	stopTheWorld("start tracing")
   186  
   187  	// We are in stop-the-world, but syscalls can finish and write to trace concurrently.
   188  	// Exitsyscall could check trace.enabled long before and then suddenly wake up
   189  	// and decide to write to trace at a random point in time.
   190  	// However, such syscall will use the global trace.buf buffer, because we've
   191  	// acquired all p's by doing stop-the-world. So this protects us from such races.
   192  	lock(&trace.bufLock)
   193  
   194  	if trace.enabled || trace.shutdown {
   195  		unlock(&trace.bufLock)
   196  		startTheWorld()
   197  		return errorString("tracing is already enabled")
   198  	}
   199  
   200  	// Can't set trace.enabled yet. While the world is stopped, exitsyscall could
   201  	// already emit a delayed event (see exitTicks in exitsyscall) if we set trace.enabled here.
   202  	// That would lead to an inconsistent trace:
   203  	// - either GoSysExit appears before EvGoInSyscall,
   204  	// - or GoSysExit appears for a goroutine for which we don't emit EvGoInSyscall below.
   205  	// To instruct traceEvent that it must not ignore events below, we set startingtrace.
   206  	// trace.enabled is set afterwards once we have emitted all preliminary events.
   207  	_g_ := getg()
   208  	_g_.m.startingtrace = true
   209  
   210  	// Obtain current stack ID to use in all traceEvGoCreate events below.
   211  	mp := acquirem()
   212  	stkBuf := make([]uintptr, traceStackSize)
   213  	stackID := traceStackID(mp, stkBuf, 2)
   214  	releasem(mp)
   215  
   216  	for _, gp := range allgs {
   217  		status := readgstatus(gp)
   218  		if status != _Gdead {
   219  			gp.traceseq = 0
   220  			gp.tracelastp = getg().m.p
   221  			// +PCQuantum because traceFrameForPC expects return PCs and subtracts PCQuantum.
   222  			id := trace.stackTab.put([]uintptr{gp.startpc + sys.PCQuantum})
   223  			traceEvent(traceEvGoCreate, -1, uint64(gp.goid), uint64(id), stackID)
   224  		}
   225  		if status == _Gwaiting {
   226  			// traceEvGoWaiting is implied to have seq=1.
   227  			gp.traceseq++
   228  			traceEvent(traceEvGoWaiting, -1, uint64(gp.goid))
   229  		}
   230  		if status == _Gsyscall {
   231  			gp.traceseq++
   232  			traceEvent(traceEvGoInSyscall, -1, uint64(gp.goid))
   233  		} else {
   234  			gp.sysblocktraced = false
   235  		}
   236  	}
   237  	traceProcStart()
   238  	traceGoStart()
   239  	// Note: ticksStart needs to be set after we emit traceEvGoInSyscall events.
   240  	// If we do it the other way around, it is possible that exitsyscall will
   241  	// query sysexitticks after ticksStart but before traceEvGoInSyscall timestamp.
   242  	// It will lead to a false conclusion that cputicks is broken.
   243  	trace.ticksStart = cputicks()
   244  	trace.timeStart = nanotime()
   245  	trace.headerWritten = false
   246  	trace.footerWritten = false
   247  
   248  	// string to id mapping
   249  	//  0 : reserved for an empty string
   250  	//  remaining: other strings registered by traceString
   251  	trace.stringSeq = 0
   252  	trace.strings = make(map[string]uint64)
   253  
   254  	trace.seqGC = 0
   255  	_g_.m.startingtrace = false
   256  	trace.enabled = true
   257  
   258  	// Register runtime goroutine labels.
   259  	_, pid, bufp := traceAcquireBuffer()
   260  	for i, label := range gcMarkWorkerModeStrings[:] {
   261  		trace.markWorkerLabels[i], bufp = traceString(bufp, pid, label)
   262  	}
   263  	traceReleaseBuffer(pid)
   264  
   265  	unlock(&trace.bufLock)
   266  
   267  	startTheWorld()
   268  	return nil
   269  }
   270  
   271  // StopTrace stops tracing, if it was previously enabled.
   272  // StopTrace only returns after all the reads for the trace have completed.
   273  func StopTrace() {
   274  	// Stop the world so that we can collect the trace buffers from all p's below,
   275  	// and also to avoid races with traceEvent.
   276  	stopTheWorld("stop tracing")
   277  
   278  	// See the comment in StartTrace.
   279  	lock(&trace.bufLock)
   280  
   281  	if !trace.enabled {
   282  		unlock(&trace.bufLock)
   283  		startTheWorld()
   284  		return
   285  	}
   286  
   287  	traceGoSched()
   288  
   289  	// Loop over all allocated Ps because dead Ps may still have
   290  	// trace buffers.
   291  	for _, p := range allp[:cap(allp)] {
   292  		buf := p.tracebuf
   293  		if buf != 0 {
   294  			traceFullQueue(buf)
   295  			p.tracebuf = 0
   296  		}
   297  	}
   298  	if trace.buf != 0 {
   299  		buf := trace.buf
   300  		trace.buf = 0
   301  		if buf.ptr().pos != 0 {
   302  			traceFullQueue(buf)
   303  		}
   304  	}
   305  
   306  	for {
   307  		trace.ticksEnd = cputicks()
   308  		trace.timeEnd = nanotime()
   309  		// Windows time can tick only every 15ms, wait for at least one tick.
   310  		if trace.timeEnd != trace.timeStart {
   311  			break
   312  		}
   313  		osyield()
   314  	}
   315  
   316  	trace.enabled = false
   317  	trace.shutdown = true
   318  	unlock(&trace.bufLock)
   319  
   320  	startTheWorld()
   321  
   322  	// The world is started but we've set trace.shutdown, so new tracing can't start.
   323  	// Wait for the trace reader to flush pending buffers and stop.
   324  	semacquire(&trace.shutdownSema)
   325  	if raceenabled {
   326  		raceacquire(unsafe.Pointer(&trace.shutdownSema))
   327  	}
   328  
   329  	// The lock protects us from races with StartTrace/StopTrace because they do stop-the-world.
   330  	lock(&trace.lock)
   331  	for _, p := range allp[:cap(allp)] {
   332  		if p.tracebuf != 0 {
   333  			throw("trace: non-empty trace buffer in proc")
   334  		}
   335  	}
   336  	if trace.buf != 0 {
   337  		throw("trace: non-empty global trace buffer")
   338  	}
   339  	if trace.fullHead != 0 || trace.fullTail != 0 {
   340  		throw("trace: non-empty full trace buffer")
   341  	}
   342  	if trace.reading != 0 || trace.reader != 0 {
   343  		throw("trace: reading after shutdown")
   344  	}
   345  	for trace.empty != 0 {
   346  		buf := trace.empty
   347  		trace.empty = buf.ptr().link
   348  		sysFree(unsafe.Pointer(buf), unsafe.Sizeof(*buf.ptr()), &memstats.other_sys)
   349  	}
   350  	trace.strings = nil
   351  	trace.shutdown = false
   352  	unlock(&trace.lock)
   353  }
   354  
   355  // ReadTrace returns the next chunk of binary tracing data, blocking until data
   356  // is available. If tracing is turned off and all the data accumulated while it
   357  // was on has been returned, ReadTrace returns nil. The caller must copy the
   358  // returned data before calling ReadTrace again.
   359  // ReadTrace must be called from one goroutine at a time.
   360  func ReadTrace() []byte {
   361  	// This function may need to lock trace.lock recursively
   362  	// (goparkunlock -> traceGoPark -> traceEvent -> traceFlush).
   363  	// To allow this we use trace.lockOwner.
   364  	// Also this function must not allocate while holding trace.lock:
   365  	// allocation can call heap allocate, which will try to emit a trace
   366  	// event while holding heap lock.
   367  	lock(&trace.lock)
   368  	trace.lockOwner = getg()
   369  
   370  	if trace.reader != 0 {
   371  		// More than one goroutine reads trace. This is bad.
   372  		// But we rather do not crash the program because of tracing,
   373  		// because tracing can be enabled at runtime on prod servers.
   374  		trace.lockOwner = nil
   375  		unlock(&trace.lock)
   376  		println("runtime: ReadTrace called from multiple goroutines simultaneously")
   377  		return nil
   378  	}
   379  	// Recycle the old buffer.
   380  	if buf := trace.reading; buf != 0 {
   381  		buf.ptr().link = trace.empty
   382  		trace.empty = buf
   383  		trace.reading = 0
   384  	}
   385  	// Write trace header.
   386  	if !trace.headerWritten {
   387  		trace.headerWritten = true
   388  		trace.lockOwner = nil
   389  		unlock(&trace.lock)
   390  		return []byte("go 1.11 trace\x00\x00\x00")
   391  	}
   392  	// Wait for new data.
   393  	if trace.fullHead == 0 && !trace.shutdown {
   394  		trace.reader.set(getg())
   395  		goparkunlock(&trace.lock, waitReasonTraceReaderBlocked, traceEvGoBlock, 2)
   396  		lock(&trace.lock)
   397  	}
   398  	// Write a buffer.
   399  	if trace.fullHead != 0 {
   400  		buf := traceFullDequeue()
   401  		trace.reading = buf
   402  		trace.lockOwner = nil
   403  		unlock(&trace.lock)
   404  		return buf.ptr().arr[:buf.ptr().pos]
   405  	}
   406  	// Write footer with timer frequency.
   407  	if !trace.footerWritten {
   408  		trace.footerWritten = true
   409  		// Use float64 because (trace.ticksEnd - trace.ticksStart) * 1e9 can overflow int64.
   410  		freq := float64(trace.ticksEnd-trace.ticksStart) * 1e9 / float64(trace.timeEnd-trace.timeStart) / traceTickDiv
   411  		trace.lockOwner = nil
   412  		unlock(&trace.lock)
   413  		var data []byte
   414  		data = append(data, traceEvFrequency|0<<traceArgCountShift)
   415  		data = traceAppend(data, uint64(freq))
   416  		for i := range timers {
   417  			tb := &timers[i]
   418  			if tb.gp != nil {
   419  				data = append(data, traceEvTimerGoroutine|0<<traceArgCountShift)
   420  				data = traceAppend(data, uint64(tb.gp.goid))
   421  			}
   422  		}
   423  		// This will emit a bunch of full buffers, we will pick them up
   424  		// on the next iteration.
   425  		trace.stackTab.dump()
   426  		return data
   427  	}
   428  	// Done.
   429  	if trace.shutdown {
   430  		trace.lockOwner = nil
   431  		unlock(&trace.lock)
   432  		if raceenabled {
   433  			// Model synchronization on trace.shutdownSema, which race
   434  			// detector does not see. This is required to avoid false
   435  			// race reports on writer passed to trace.Start.
   436  			racerelease(unsafe.Pointer(&trace.shutdownSema))
   437  		}
   438  		// trace.enabled is already reset, so can call traceable functions.
   439  		semrelease(&trace.shutdownSema)
   440  		return nil
   441  	}
   442  	// Also bad, but see the comment above.
   443  	trace.lockOwner = nil
   444  	unlock(&trace.lock)
   445  	println("runtime: spurious wakeup of trace reader")
   446  	return nil
   447  }
   448  
   449  // traceReader returns the trace reader that should be woken up, if any.
   450  func traceReader() *g {
   451  	if trace.reader == 0 || (trace.fullHead == 0 && !trace.shutdown) {
   452  		return nil
   453  	}
   454  	lock(&trace.lock)
   455  	if trace.reader == 0 || (trace.fullHead == 0 && !trace.shutdown) {
   456  		unlock(&trace.lock)
   457  		return nil
   458  	}
   459  	gp := trace.reader.ptr()
   460  	trace.reader.set(nil)
   461  	unlock(&trace.lock)
   462  	return gp
   463  }
   464  
   465  // traceProcFree frees trace buffer associated with pp.
   466  func traceProcFree(pp *p) {
   467  	buf := pp.tracebuf
   468  	pp.tracebuf = 0
   469  	if buf == 0 {
   470  		return
   471  	}
   472  	lock(&trace.lock)
   473  	traceFullQueue(buf)
   474  	unlock(&trace.lock)
   475  }
   476  
   477  // traceFullQueue queues buf into queue of full buffers.
   478  func traceFullQueue(buf traceBufPtr) {
   479  	buf.ptr().link = 0
   480  	if trace.fullHead == 0 {
   481  		trace.fullHead = buf
   482  	} else {
   483  		trace.fullTail.ptr().link = buf
   484  	}
   485  	trace.fullTail = buf
   486  }
   487  
   488  // traceFullDequeue dequeues from queue of full buffers.
   489  func traceFullDequeue() traceBufPtr {
   490  	buf := trace.fullHead
   491  	if buf == 0 {
   492  		return 0
   493  	}
   494  	trace.fullHead = buf.ptr().link
   495  	if trace.fullHead == 0 {
   496  		trace.fullTail = 0
   497  	}
   498  	buf.ptr().link = 0
   499  	return buf
   500  }
   501  
   502  // traceEvent writes a single event to trace buffer, flushing the buffer if necessary.
   503  // ev is event type.
   504  // If skip > 0, write current stack id as the last argument (skipping skip top frames).
   505  // If skip = 0, this event type should contain a stack, but we don't want
   506  // to collect and remember it for this particular call.
   507  func traceEvent(ev byte, skip int, args ...uint64) {
   508  	mp, pid, bufp := traceAcquireBuffer()
   509  	// Double-check trace.enabled now that we've done m.locks++ and acquired bufLock.
   510  	// This protects from races between traceEvent and StartTrace/StopTrace.
   511  
   512  	// The caller checked that trace.enabled == true, but trace.enabled might have been
   513  	// turned off between the check and now. Check again. traceLockBuffer did mp.locks++,
   514  	// StopTrace does stopTheWorld, and stopTheWorld waits for mp.locks to go back to zero,
   515  	// so if we see trace.enabled == true now, we know it's true for the rest of the function.
   516  	// Exitsyscall can run even during stopTheWorld. The race with StartTrace/StopTrace
   517  	// during tracing in exitsyscall is resolved by locking trace.bufLock in traceLockBuffer.
   518  	//
   519  	// Note trace_userTaskCreate runs the same check.
   520  	if !trace.enabled && !mp.startingtrace {
   521  		traceReleaseBuffer(pid)
   522  		return
   523  	}
   524  
   525  	if skip > 0 {
   526  		if getg() == mp.curg {
   527  			skip++ // +1 because stack is captured in traceEventLocked.
   528  		}
   529  	}
   530  	traceEventLocked(0, mp, pid, bufp, ev, skip, args...)
   531  	traceReleaseBuffer(pid)
   532  }
   533  
   534  func traceEventLocked(extraBytes int, mp *m, pid int32, bufp *traceBufPtr, ev byte, skip int, args ...uint64) {
   535  	buf := (*bufp).ptr()
   536  	// TODO: test on non-zero extraBytes param.
   537  	maxSize := 2 + 5*traceBytesPerNumber + extraBytes // event type, length, sequence, timestamp, stack id and two add params
   538  	if buf == nil || len(buf.arr)-buf.pos < maxSize {
   539  		buf = traceFlush(traceBufPtrOf(buf), pid).ptr()
   540  		(*bufp).set(buf)
   541  	}
   542  
   543  	ticks := uint64(cputicks()) / traceTickDiv
   544  	tickDiff := ticks - buf.lastTicks
   545  	buf.lastTicks = ticks
   546  	narg := byte(len(args))
   547  	if skip >= 0 {
   548  		narg++
   549  	}
   550  	// We have only 2 bits for number of arguments.
   551  	// If number is >= 3, then the event type is followed by event length in bytes.
   552  	if narg > 3 {
   553  		narg = 3
   554  	}
   555  	startPos := buf.pos
   556  	buf.byte(ev | narg<<traceArgCountShift)
   557  	var lenp *byte
   558  	if narg == 3 {
   559  		// Reserve the byte for length assuming that length < 128.
   560  		buf.varint(0)
   561  		lenp = &buf.arr[buf.pos-1]
   562  	}
   563  	buf.varint(tickDiff)
   564  	for _, a := range args {
   565  		buf.varint(a)
   566  	}
   567  	if skip == 0 {
   568  		buf.varint(0)
   569  	} else if skip > 0 {
   570  		buf.varint(traceStackID(mp, buf.stk[:], skip))
   571  	}
   572  	evSize := buf.pos - startPos
   573  	if evSize > maxSize {
   574  		throw("invalid length of trace event")
   575  	}
   576  	if lenp != nil {
   577  		// Fill in actual length.
   578  		*lenp = byte(evSize - 2)
   579  	}
   580  }
   581  
   582  func traceStackID(mp *m, buf []uintptr, skip int) uint64 {
   583  	_g_ := getg()
   584  	gp := mp.curg
   585  	var nstk int
   586  	if gp == _g_ {
   587  		nstk = callers(skip+1, buf[:])
   588  	} else if gp != nil {
   589  		gp = mp.curg
   590  		nstk = gcallers(gp, skip, buf[:])
   591  	}
   592  	if nstk > 0 {
   593  		nstk-- // skip runtime.goexit
   594  	}
   595  	if nstk > 0 && gp.goid == 1 {
   596  		nstk-- // skip runtime.main
   597  	}
   598  	id := trace.stackTab.put(buf[:nstk])
   599  	return uint64(id)
   600  }
   601  
   602  // traceAcquireBuffer returns trace buffer to use and, if necessary, locks it.
   603  func traceAcquireBuffer() (mp *m, pid int32, bufp *traceBufPtr) {
   604  	mp = acquirem()
   605  	if p := mp.p.ptr(); p != nil {
   606  		return mp, p.id, &p.tracebuf
   607  	}
   608  	lock(&trace.bufLock)
   609  	return mp, traceGlobProc, &trace.buf
   610  }
   611  
   612  // traceReleaseBuffer releases a buffer previously acquired with traceAcquireBuffer.
   613  func traceReleaseBuffer(pid int32) {
   614  	if pid == traceGlobProc {
   615  		unlock(&trace.bufLock)
   616  	}
   617  	releasem(getg().m)
   618  }
   619  
   620  // traceFlush puts buf onto stack of full buffers and returns an empty buffer.
   621  func traceFlush(buf traceBufPtr, pid int32) traceBufPtr {
   622  	owner := trace.lockOwner
   623  	dolock := owner == nil || owner != getg().m.curg
   624  	if dolock {
   625  		lock(&trace.lock)
   626  	}
   627  	if buf != 0 {
   628  		traceFullQueue(buf)
   629  	}
   630  	if trace.empty != 0 {
   631  		buf = trace.empty
   632  		trace.empty = buf.ptr().link
   633  	} else {
   634  		buf = traceBufPtr(sysAlloc(unsafe.Sizeof(traceBuf{}), &memstats.other_sys))
   635  		if buf == 0 {
   636  			throw("trace: out of memory")
   637  		}
   638  	}
   639  	bufp := buf.ptr()
   640  	bufp.link.set(nil)
   641  	bufp.pos = 0
   642  
   643  	// initialize the buffer for a new batch
   644  	ticks := uint64(cputicks()) / traceTickDiv
   645  	bufp.lastTicks = ticks
   646  	bufp.byte(traceEvBatch | 1<<traceArgCountShift)
   647  	bufp.varint(uint64(pid))
   648  	bufp.varint(ticks)
   649  
   650  	if dolock {
   651  		unlock(&trace.lock)
   652  	}
   653  	return buf
   654  }
   655  
   656  // traceString adds a string to the trace.strings and returns the id.
   657  func traceString(bufp *traceBufPtr, pid int32, s string) (uint64, *traceBufPtr) {
   658  	if s == "" {
   659  		return 0, bufp
   660  	}
   661  
   662  	lock(&trace.stringsLock)
   663  	if raceenabled {
   664  		// raceacquire is necessary because the map access
   665  		// below is race annotated.
   666  		raceacquire(unsafe.Pointer(&trace.stringsLock))
   667  	}
   668  
   669  	if id, ok := trace.strings[s]; ok {
   670  		if raceenabled {
   671  			racerelease(unsafe.Pointer(&trace.stringsLock))
   672  		}
   673  		unlock(&trace.stringsLock)
   674  
   675  		return id, bufp
   676  	}
   677  
   678  	trace.stringSeq++
   679  	id := trace.stringSeq
   680  	trace.strings[s] = id
   681  
   682  	if raceenabled {
   683  		racerelease(unsafe.Pointer(&trace.stringsLock))
   684  	}
   685  	unlock(&trace.stringsLock)
   686  
   687  	// memory allocation in above may trigger tracing and
   688  	// cause *bufp changes. Following code now works with *bufp,
   689  	// so there must be no memory allocation or any activities
   690  	// that causes tracing after this point.
   691  
   692  	buf := (*bufp).ptr()
   693  	size := 1 + 2*traceBytesPerNumber + len(s)
   694  	if buf == nil || len(buf.arr)-buf.pos < size {
   695  		buf = traceFlush(traceBufPtrOf(buf), pid).ptr()
   696  		(*bufp).set(buf)
   697  	}
   698  	buf.byte(traceEvString)
   699  	buf.varint(id)
   700  
   701  	// double-check the string and the length can fit.
   702  	// Otherwise, truncate the string.
   703  	slen := len(s)
   704  	if room := len(buf.arr) - buf.pos; room < slen+traceBytesPerNumber {
   705  		slen = room
   706  	}
   707  
   708  	buf.varint(uint64(slen))
   709  	buf.pos += copy(buf.arr[buf.pos:], s[:slen])
   710  
   711  	(*bufp).set(buf)
   712  	return id, bufp
   713  }
   714  
   715  // traceAppend appends v to buf in little-endian-base-128 encoding.
   716  func traceAppend(buf []byte, v uint64) []byte {
   717  	for ; v >= 0x80; v >>= 7 {
   718  		buf = append(buf, 0x80|byte(v))
   719  	}
   720  	buf = append(buf, byte(v))
   721  	return buf
   722  }
   723  
   724  // varint appends v to buf in little-endian-base-128 encoding.
   725  func (buf *traceBuf) varint(v uint64) {
   726  	pos := buf.pos
   727  	for ; v >= 0x80; v >>= 7 {
   728  		buf.arr[pos] = 0x80 | byte(v)
   729  		pos++
   730  	}
   731  	buf.arr[pos] = byte(v)
   732  	pos++
   733  	buf.pos = pos
   734  }
   735  
   736  // byte appends v to buf.
   737  func (buf *traceBuf) byte(v byte) {
   738  	buf.arr[buf.pos] = v
   739  	buf.pos++
   740  }
   741  
   742  // traceStackTable maps stack traces (arrays of PC's) to unique uint32 ids.
   743  // It is lock-free for reading.
   744  type traceStackTable struct {
   745  	lock mutex
   746  	seq  uint32
   747  	mem  traceAlloc
   748  	tab  [1 << 13]traceStackPtr
   749  }
   750  
   751  // traceStack is a single stack in traceStackTable.
   752  type traceStack struct {
   753  	link traceStackPtr
   754  	hash uintptr
   755  	id   uint32
   756  	n    int
   757  	stk  [0]uintptr // real type [n]uintptr
   758  }
   759  
   760  type traceStackPtr uintptr
   761  
   762  func (tp traceStackPtr) ptr() *traceStack { return (*traceStack)(unsafe.Pointer(tp)) }
   763  
   764  // stack returns slice of PCs.
   765  func (ts *traceStack) stack() []uintptr {
   766  	return (*[traceStackSize]uintptr)(unsafe.Pointer(&ts.stk))[:ts.n]
   767  }
   768  
   769  // put returns a unique id for the stack trace pcs and caches it in the table,
   770  // if it sees the trace for the first time.
   771  func (tab *traceStackTable) put(pcs []uintptr) uint32 {
   772  	if len(pcs) == 0 {
   773  		return 0
   774  	}
   775  	hash := memhash(unsafe.Pointer(&pcs[0]), 0, uintptr(len(pcs))*unsafe.Sizeof(pcs[0]))
   776  	// First, search the hashtable w/o the mutex.
   777  	if id := tab.find(pcs, hash); id != 0 {
   778  		return id
   779  	}
   780  	// Now, double check under the mutex.
   781  	lock(&tab.lock)
   782  	if id := tab.find(pcs, hash); id != 0 {
   783  		unlock(&tab.lock)
   784  		return id
   785  	}
   786  	// Create new record.
   787  	tab.seq++
   788  	stk := tab.newStack(len(pcs))
   789  	stk.hash = hash
   790  	stk.id = tab.seq
   791  	stk.n = len(pcs)
   792  	stkpc := stk.stack()
   793  	for i, pc := range pcs {
   794  		stkpc[i] = pc
   795  	}
   796  	part := int(hash % uintptr(len(tab.tab)))
   797  	stk.link = tab.tab[part]
   798  	atomicstorep(unsafe.Pointer(&tab.tab[part]), unsafe.Pointer(stk))
   799  	unlock(&tab.lock)
   800  	return stk.id
   801  }
   802  
   803  // find checks if the stack trace pcs is already present in the table.
   804  func (tab *traceStackTable) find(pcs []uintptr, hash uintptr) uint32 {
   805  	part := int(hash % uintptr(len(tab.tab)))
   806  Search:
   807  	for stk := tab.tab[part].ptr(); stk != nil; stk = stk.link.ptr() {
   808  		if stk.hash == hash && stk.n == len(pcs) {
   809  			for i, stkpc := range stk.stack() {
   810  				if stkpc != pcs[i] {
   811  					continue Search
   812  				}
   813  			}
   814  			return stk.id
   815  		}
   816  	}
   817  	return 0
   818  }
   819  
   820  // newStack allocates a new stack of size n.
   821  func (tab *traceStackTable) newStack(n int) *traceStack {
   822  	return (*traceStack)(tab.mem.alloc(unsafe.Sizeof(traceStack{}) + uintptr(n)*sys.PtrSize))
   823  }
   824  
   825  // allFrames returns all of the Frames corresponding to pcs.
   826  func allFrames(pcs []uintptr) []Frame {
   827  	frames := make([]Frame, 0, len(pcs))
   828  	ci := CallersFrames(pcs)
   829  	for {
   830  		f, more := ci.Next()
   831  		frames = append(frames, f)
   832  		if !more {
   833  			return frames
   834  		}
   835  	}
   836  }
   837  
   838  // dump writes all previously cached stacks to trace buffers,
   839  // releases all memory and resets state.
   840  func (tab *traceStackTable) dump() {
   841  	var tmp [(2 + 4*traceStackSize) * traceBytesPerNumber]byte
   842  	bufp := traceFlush(0, 0)
   843  	for _, stk := range tab.tab {
   844  		stk := stk.ptr()
   845  		for ; stk != nil; stk = stk.link.ptr() {
   846  			tmpbuf := tmp[:0]
   847  			tmpbuf = traceAppend(tmpbuf, uint64(stk.id))
   848  			frames := allFrames(stk.stack())
   849  			tmpbuf = traceAppend(tmpbuf, uint64(len(frames)))
   850  			for _, f := range frames {
   851  				var frame traceFrame
   852  				frame, bufp = traceFrameForPC(bufp, 0, f)
   853  				tmpbuf = traceAppend(tmpbuf, uint64(f.PC))
   854  				tmpbuf = traceAppend(tmpbuf, uint64(frame.funcID))
   855  				tmpbuf = traceAppend(tmpbuf, uint64(frame.fileID))
   856  				tmpbuf = traceAppend(tmpbuf, uint64(frame.line))
   857  			}
   858  			// Now copy to the buffer.
   859  			size := 1 + traceBytesPerNumber + len(tmpbuf)
   860  			if buf := bufp.ptr(); len(buf.arr)-buf.pos < size {
   861  				bufp = traceFlush(bufp, 0)
   862  			}
   863  			buf := bufp.ptr()
   864  			buf.byte(traceEvStack | 3<<traceArgCountShift)
   865  			buf.varint(uint64(len(tmpbuf)))
   866  			buf.pos += copy(buf.arr[buf.pos:], tmpbuf)
   867  		}
   868  	}
   869  
   870  	lock(&trace.lock)
   871  	traceFullQueue(bufp)
   872  	unlock(&trace.lock)
   873  
   874  	tab.mem.drop()
   875  	*tab = traceStackTable{}
   876  }
   877  
   878  type traceFrame struct {
   879  	funcID uint64
   880  	fileID uint64
   881  	line   uint64
   882  }
   883  
   884  // traceFrameForPC records the frame information.
   885  // It may allocate memory.
   886  func traceFrameForPC(buf traceBufPtr, pid int32, f Frame) (traceFrame, traceBufPtr) {
   887  	bufp := &buf
   888  	var frame traceFrame
   889  
   890  	fn := f.Function
   891  	const maxLen = 1 << 10
   892  	if len(fn) > maxLen {
   893  		fn = fn[len(fn)-maxLen:]
   894  	}
   895  	frame.funcID, bufp = traceString(bufp, pid, fn)
   896  	frame.line = uint64(f.Line)
   897  	file := f.File
   898  	if len(file) > maxLen {
   899  		file = file[len(file)-maxLen:]
   900  	}
   901  	frame.fileID, bufp = traceString(bufp, pid, file)
   902  	return frame, (*bufp)
   903  }
   904  
   905  // traceAlloc is a non-thread-safe region allocator.
   906  // It holds a linked list of traceAllocBlock.
   907  type traceAlloc struct {
   908  	head traceAllocBlockPtr
   909  	off  uintptr
   910  }
   911  
   912  // traceAllocBlock is a block in traceAlloc.
   913  //
   914  // traceAllocBlock is allocated from non-GC'd memory, so it must not
   915  // contain heap pointers. Writes to pointers to traceAllocBlocks do
   916  // not need write barriers.
   917  //
   918  //go:notinheap
   919  type traceAllocBlock struct {
   920  	next traceAllocBlockPtr
   921  	data [64<<10 - sys.PtrSize]byte
   922  }
   923  
   924  // TODO: Since traceAllocBlock is now go:notinheap, this isn't necessary.
   925  type traceAllocBlockPtr uintptr
   926  
   927  func (p traceAllocBlockPtr) ptr() *traceAllocBlock   { return (*traceAllocBlock)(unsafe.Pointer(p)) }
   928  func (p *traceAllocBlockPtr) set(x *traceAllocBlock) { *p = traceAllocBlockPtr(unsafe.Pointer(x)) }
   929  
   930  // alloc allocates n-byte block.
   931  func (a *traceAlloc) alloc(n uintptr) unsafe.Pointer {
   932  	n = round(n, sys.PtrSize)
   933  	if a.head == 0 || a.off+n > uintptr(len(a.head.ptr().data)) {
   934  		if n > uintptr(len(a.head.ptr().data)) {
   935  			throw("trace: alloc too large")
   936  		}
   937  		block := (*traceAllocBlock)(sysAlloc(unsafe.Sizeof(traceAllocBlock{}), &memstats.other_sys))
   938  		if block == nil {
   939  			throw("trace: out of memory")
   940  		}
   941  		block.next.set(a.head.ptr())
   942  		a.head.set(block)
   943  		a.off = 0
   944  	}
   945  	p := &a.head.ptr().data[a.off]
   946  	a.off += n
   947  	return unsafe.Pointer(p)
   948  }
   949  
   950  // drop frees all previously allocated memory and resets the allocator.
   951  func (a *traceAlloc) drop() {
   952  	for a.head != 0 {
   953  		block := a.head.ptr()
   954  		a.head.set(block.next.ptr())
   955  		sysFree(unsafe.Pointer(block), unsafe.Sizeof(traceAllocBlock{}), &memstats.other_sys)
   956  	}
   957  }
   958  
   959  // The following functions write specific events to trace.
   960  
   961  func traceGomaxprocs(procs int32) {
   962  	traceEvent(traceEvGomaxprocs, 1, uint64(procs))
   963  }
   964  
   965  func traceProcStart() {
   966  	traceEvent(traceEvProcStart, -1, uint64(getg().m.id))
   967  }
   968  
   969  func traceProcStop(pp *p) {
   970  	// Sysmon and stopTheWorld can stop Ps blocked in syscalls,
   971  	// to handle this we temporary employ the P.
   972  	mp := acquirem()
   973  	oldp := mp.p
   974  	mp.p.set(pp)
   975  	traceEvent(traceEvProcStop, -1)
   976  	mp.p = oldp
   977  	releasem(mp)
   978  }
   979  
   980  func traceGCStart() {
   981  	traceEvent(traceEvGCStart, 3, trace.seqGC)
   982  	trace.seqGC++
   983  }
   984  
   985  func traceGCDone() {
   986  	traceEvent(traceEvGCDone, -1)
   987  }
   988  
   989  func traceGCSTWStart(kind int) {
   990  	traceEvent(traceEvGCSTWStart, -1, uint64(kind))
   991  }
   992  
   993  func traceGCSTWDone() {
   994  	traceEvent(traceEvGCSTWDone, -1)
   995  }
   996  
   997  // traceGCSweepStart prepares to trace a sweep loop. This does not
   998  // emit any events until traceGCSweepSpan is called.
   999  //
  1000  // traceGCSweepStart must be paired with traceGCSweepDone and there
  1001  // must be no preemption points between these two calls.
  1002  func traceGCSweepStart() {
  1003  	// Delay the actual GCSweepStart event until the first span
  1004  	// sweep. If we don't sweep anything, don't emit any events.
  1005  	_p_ := getg().m.p.ptr()
  1006  	if _p_.traceSweep {
  1007  		throw("double traceGCSweepStart")
  1008  	}
  1009  	_p_.traceSweep, _p_.traceSwept, _p_.traceReclaimed = true, 0, 0
  1010  }
  1011  
  1012  // traceGCSweepSpan traces the sweep of a single page.
  1013  //
  1014  // This may be called outside a traceGCSweepStart/traceGCSweepDone
  1015  // pair; however, it will not emit any trace events in this case.
  1016  func traceGCSweepSpan(bytesSwept uintptr) {
  1017  	_p_ := getg().m.p.ptr()
  1018  	if _p_.traceSweep {
  1019  		if _p_.traceSwept == 0 {
  1020  			traceEvent(traceEvGCSweepStart, 1)
  1021  		}
  1022  		_p_.traceSwept += bytesSwept
  1023  	}
  1024  }
  1025  
  1026  func traceGCSweepDone() {
  1027  	_p_ := getg().m.p.ptr()
  1028  	if !_p_.traceSweep {
  1029  		throw("missing traceGCSweepStart")
  1030  	}
  1031  	if _p_.traceSwept != 0 {
  1032  		traceEvent(traceEvGCSweepDone, -1, uint64(_p_.traceSwept), uint64(_p_.traceReclaimed))
  1033  	}
  1034  	_p_.traceSweep = false
  1035  }
  1036  
  1037  func traceGCMarkAssistStart() {
  1038  	traceEvent(traceEvGCMarkAssistStart, 1)
  1039  }
  1040  
  1041  func traceGCMarkAssistDone() {
  1042  	traceEvent(traceEvGCMarkAssistDone, -1)
  1043  }
  1044  
  1045  func traceGoCreate(newg *g, pc uintptr) {
  1046  	newg.traceseq = 0
  1047  	newg.tracelastp = getg().m.p
  1048  	// +PCQuantum because traceFrameForPC expects return PCs and subtracts PCQuantum.
  1049  	id := trace.stackTab.put([]uintptr{pc + sys.PCQuantum})
  1050  	traceEvent(traceEvGoCreate, 2, uint64(newg.goid), uint64(id))
  1051  }
  1052  
  1053  func traceGoStart() {
  1054  	_g_ := getg().m.curg
  1055  	_p_ := _g_.m.p
  1056  	_g_.traceseq++
  1057  	if _g_ == _p_.ptr().gcBgMarkWorker.ptr() {
  1058  		traceEvent(traceEvGoStartLabel, -1, uint64(_g_.goid), _g_.traceseq, trace.markWorkerLabels[_p_.ptr().gcMarkWorkerMode])
  1059  	} else if _g_.tracelastp == _p_ {
  1060  		traceEvent(traceEvGoStartLocal, -1, uint64(_g_.goid))
  1061  	} else {
  1062  		_g_.tracelastp = _p_
  1063  		traceEvent(traceEvGoStart, -1, uint64(_g_.goid), _g_.traceseq)
  1064  	}
  1065  }
  1066  
  1067  func traceGoEnd() {
  1068  	traceEvent(traceEvGoEnd, -1)
  1069  }
  1070  
  1071  func traceGoSched() {
  1072  	_g_ := getg()
  1073  	_g_.tracelastp = _g_.m.p
  1074  	traceEvent(traceEvGoSched, 1)
  1075  }
  1076  
  1077  func traceGoPreempt() {
  1078  	_g_ := getg()
  1079  	_g_.tracelastp = _g_.m.p
  1080  	traceEvent(traceEvGoPreempt, 1)
  1081  }
  1082  
  1083  func traceGoPark(traceEv byte, skip int) {
  1084  	if traceEv&traceFutileWakeup != 0 {
  1085  		traceEvent(traceEvFutileWakeup, -1)
  1086  	}
  1087  	traceEvent(traceEv & ^traceFutileWakeup, skip)
  1088  }
  1089  
  1090  func traceGoUnpark(gp *g, skip int) {
  1091  	_p_ := getg().m.p
  1092  	gp.traceseq++
  1093  	if gp.tracelastp == _p_ {
  1094  		traceEvent(traceEvGoUnblockLocal, skip, uint64(gp.goid))
  1095  	} else {
  1096  		gp.tracelastp = _p_
  1097  		traceEvent(traceEvGoUnblock, skip, uint64(gp.goid), gp.traceseq)
  1098  	}
  1099  }
  1100  
  1101  func traceGoSysCall() {
  1102  	traceEvent(traceEvGoSysCall, 1)
  1103  }
  1104  
  1105  func traceGoSysExit(ts int64) {
  1106  	if ts != 0 && ts < trace.ticksStart {
  1107  		// There is a race between the code that initializes sysexitticks
  1108  		// (in exitsyscall, which runs without a P, and therefore is not
  1109  		// stopped with the rest of the world) and the code that initializes
  1110  		// a new trace. The recorded sysexitticks must therefore be treated
  1111  		// as "best effort". If they are valid for this trace, then great,
  1112  		// use them for greater accuracy. But if they're not valid for this
  1113  		// trace, assume that the trace was started after the actual syscall
  1114  		// exit (but before we actually managed to start the goroutine,
  1115  		// aka right now), and assign a fresh time stamp to keep the log consistent.
  1116  		ts = 0
  1117  	}
  1118  	_g_ := getg().m.curg
  1119  	_g_.traceseq++
  1120  	_g_.tracelastp = _g_.m.p
  1121  	traceEvent(traceEvGoSysExit, -1, uint64(_g_.goid), _g_.traceseq, uint64(ts)/traceTickDiv)
  1122  }
  1123  
  1124  func traceGoSysBlock(pp *p) {
  1125  	// Sysmon and stopTheWorld can declare syscalls running on remote Ps as blocked,
  1126  	// to handle this we temporary employ the P.
  1127  	mp := acquirem()
  1128  	oldp := mp.p
  1129  	mp.p.set(pp)
  1130  	traceEvent(traceEvGoSysBlock, -1)
  1131  	mp.p = oldp
  1132  	releasem(mp)
  1133  }
  1134  
  1135  func traceHeapAlloc() {
  1136  	traceEvent(traceEvHeapAlloc, -1, memstats.heap_live)
  1137  }
  1138  
  1139  func traceNextGC() {
  1140  	if memstats.next_gc == ^uint64(0) {
  1141  		// Heap-based triggering is disabled.
  1142  		traceEvent(traceEvNextGC, -1, 0)
  1143  	} else {
  1144  		traceEvent(traceEvNextGC, -1, memstats.next_gc)
  1145  	}
  1146  }
  1147  
  1148  // To access runtime functions from runtime/trace.
  1149  // See runtime/trace/annotation.go
  1150  
  1151  //go:linkname trace_userTaskCreate runtime/trace.userTaskCreate
  1152  func trace_userTaskCreate(id, parentID uint64, taskType string) {
  1153  	if !trace.enabled {
  1154  		return
  1155  	}
  1156  
  1157  	// Same as in traceEvent.
  1158  	mp, pid, bufp := traceAcquireBuffer()
  1159  	if !trace.enabled && !mp.startingtrace {
  1160  		traceReleaseBuffer(pid)
  1161  		return
  1162  	}
  1163  
  1164  	typeStringID, bufp := traceString(bufp, pid, taskType)
  1165  	traceEventLocked(0, mp, pid, bufp, traceEvUserTaskCreate, 3, id, parentID, typeStringID)
  1166  	traceReleaseBuffer(pid)
  1167  }
  1168  
  1169  //go:linkname trace_userTaskEnd runtime/trace.userTaskEnd
  1170  func trace_userTaskEnd(id uint64) {
  1171  	traceEvent(traceEvUserTaskEnd, 2, id)
  1172  }
  1173  
  1174  //go:linkname trace_userRegion runtime/trace.userRegion
  1175  func trace_userRegion(id, mode uint64, name string) {
  1176  	if !trace.enabled {
  1177  		return
  1178  	}
  1179  
  1180  	mp, pid, bufp := traceAcquireBuffer()
  1181  	if !trace.enabled && !mp.startingtrace {
  1182  		traceReleaseBuffer(pid)
  1183  		return
  1184  	}
  1185  
  1186  	nameStringID, bufp := traceString(bufp, pid, name)
  1187  	traceEventLocked(0, mp, pid, bufp, traceEvUserRegion, 3, id, mode, nameStringID)
  1188  	traceReleaseBuffer(pid)
  1189  }
  1190  
  1191  //go:linkname trace_userLog runtime/trace.userLog
  1192  func trace_userLog(id uint64, category, message string) {
  1193  	if !trace.enabled {
  1194  		return
  1195  	}
  1196  
  1197  	mp, pid, bufp := traceAcquireBuffer()
  1198  	if !trace.enabled && !mp.startingtrace {
  1199  		traceReleaseBuffer(pid)
  1200  		return
  1201  	}
  1202  
  1203  	categoryID, bufp := traceString(bufp, pid, category)
  1204  
  1205  	extraSpace := traceBytesPerNumber + len(message) // extraSpace for the value string
  1206  	traceEventLocked(extraSpace, mp, pid, bufp, traceEvUserLog, 3, id, categoryID)
  1207  	// traceEventLocked reserved extra space for val and len(val)
  1208  	// in buf, so buf now has room for the following.
  1209  	buf := (*bufp).ptr()
  1210  
  1211  	// double-check the message and its length can fit.
  1212  	// Otherwise, truncate the message.
  1213  	slen := len(message)
  1214  	if room := len(buf.arr) - buf.pos; room < slen+traceBytesPerNumber {
  1215  		slen = room
  1216  	}
  1217  	buf.varint(uint64(slen))
  1218  	buf.pos += copy(buf.arr[buf.pos:], message[:slen])
  1219  
  1220  	traceReleaseBuffer(pid)
  1221  }
  1222  

View as plain text