libgo/go/runtime/cpuprof.go

   1 // Copyright 2011 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 // CPU profiling.
   6 //
   7 // The signal handler for the profiling clock tick adds a new stack trace
   8 // to a log of recent traces. The log is read by a user goroutine that
   9 // turns it into formatted profile data. If the reader does not keep up
  10 // with the log, those writes will be recorded as a count of lost records.
  11 // The actual profile buffer is in profbuf.go.
  12
  13 package runtime
  14
  15 import (
  16         "runtime/internal/atomic"
  17         "runtime/internal/sys"
  18         "unsafe"
  19 )
  20
  21 const maxCPUProfStack = 64
  22
  23 type cpuProfile struct {
  24         lock mutex
  25         on   bool     // profiling is on
  26         log  *profBuf // profile events written here
  27
  28         // extra holds extra stacks accumulated in addNonGo
  29         // corresponding to profiling signals arriving on
  30         // non-Go-created threads. Those stacks are written
  31         // to log the next time a normal Go thread gets the
  32         // signal handler.
  33         // Assuming the stacks are 2 words each (we don't get
  34         // a full traceback from those threads), plus one word
  35         // size for framing, 100 Hz profiling would generate
  36         // 300 words per second.
  37         // Hopefully a normal Go thread will get the profiling
  38         // signal at least once every few seconds.
  39         extra     [1000]uintptr
  40         numExtra  int
  41         lostExtra uint64 // count of frames lost because extra is full
  42 }
  43
  44 var cpuprof cpuProfile
  45
  46 // SetCPUProfileRate sets the CPU profiling rate to hz samples per second.
  47 // If hz <= 0, SetCPUProfileRate turns off profiling.
  48 // If the profiler is on, the rate cannot be changed without first turning it off.
  49 //
  50 // Most clients should use the runtime/pprof package or
  51 // the testing package's -test.cpuprofile flag instead of calling
  52 // SetCPUProfileRate directly.
  53 func SetCPUProfileRate(hz int) {
  54         // Clamp hz to something reasonable.
  55         if hz < 0 {
  56                 hz = 0
  57         }
  58         if hz > 1000000 {
  59                 hz = 1000000
  60         }
  61
  62         lock(&cpuprof.lock)
  63         if hz > 0 {
  64                 if cpuprof.on || cpuprof.log != nil {
  65                         print("runtime: cannot set cpu profile rate until previous profile has finished.\n")
  66                         unlock(&cpuprof.lock)
  67                         return
  68                 }
  69
  70                 cpuprof.on = true
  71                 cpuprof.log = newProfBuf(1, 1<<17, 1<<14)
  72                 hdr := [1]uint64{uint64(hz)}
  73                 cpuprof.log.write(nil, nanotime(), hdr[:], nil)
  74                 setcpuprofilerate(int32(hz))
  75         } else if cpuprof.on {
  76                 setcpuprofilerate(0)
  77                 cpuprof.on = false
  78                 cpuprof.addExtra()
  79                 cpuprof.log.close()
  80         }
  81         unlock(&cpuprof.lock)
  82 }
  83
  84 // add adds the stack trace to the profile.
  85 // It is called from signal handlers and other limited environments
  86 // and cannot allocate memory or acquire locks that might be
  87 // held at the time of the signal, nor can it use substantial amounts
  88 // of stack.
  89 //go:nowritebarrierrec
  90 func (p *cpuProfile) add(gp *g, stk []uintptr) {
  91         // Simple cas-lock to coordinate with setcpuprofilerate.
  92         for !atomic.Cas(&prof.signalLock, 0, 1) {
  93                 osyield()
  94         }
  95
  96         if prof.hz != 0 { // implies cpuprof.log != nil
  97                 if p.numExtra > 0 || p.lostExtra > 0 {
  98                         p.addExtra()
  99                 }
 100                 hdr := [1]uint64{1}
 101                 // Note: write "knows" that the argument is &gp.labels,
 102                 // because otherwise its write barrier behavior may not
 103                 // be correct. See the long comment there before
 104                 // changing the argument here.
 105                 cpuprof.log.write(&gp.labels, nanotime(), hdr[:], stk)
 106         }
 107
 108         atomic.Store(&prof.signalLock, 0)
 109 }
 110
 111 // addNonGo adds the non-Go stack trace to the profile.
 112 // It is called from a non-Go thread, so we cannot use much stack at all,
 113 // nor do anything that needs a g or an m.
 114 // In particular, we can't call cpuprof.log.write.
 115 // Instead, we copy the stack into cpuprof.extra,
 116 // which will be drained the next time a Go thread
 117 // gets the signal handling event.
 118 //go:nosplit
 119 //go:nowritebarrierrec
 120 func (p *cpuProfile) addNonGo(stk []uintptr) {
 121         // Simple cas-lock to coordinate with SetCPUProfileRate.
 122         // (Other calls to add or addNonGo should be blocked out
 123         // by the fact that only one SIGPROF can be handled by the
 124         // process at a time. If not, this lock will serialize those too.)
 125         for !atomic.Cas(&prof.signalLock, 0, 1) {
 126                 osyield()
 127         }
 128
 129         if cpuprof.numExtra+1+len(stk) < len(cpuprof.extra) {
 130                 i := cpuprof.numExtra
 131                 cpuprof.extra[i] = uintptr(1 + len(stk))
 132                 copy(cpuprof.extra[i+1:], stk)
 133                 cpuprof.numExtra += 1 + len(stk)
 134         } else {
 135                 cpuprof.lostExtra++
 136         }
 137
 138         atomic.Store(&prof.signalLock, 0)
 139 }
 140
 141 // addExtra adds the "extra" profiling events,
 142 // queued by addNonGo, to the profile log.
 143 // addExtra is called either from a signal handler on a Go thread
 144 // or from an ordinary goroutine; either way it can use stack
 145 // and has a g. The world may be stopped, though.
 146 func (p *cpuProfile) addExtra() {
 147         // Copy accumulated non-Go profile events.
 148         hdr := [1]uint64{1}
 149         for i := 0; i < p.numExtra; {
 150                 p.log.write(nil, 0, hdr[:], p.extra[i+1:i+int(p.extra[i])])
 151                 i += int(p.extra[i])
 152         }
 153         p.numExtra = 0
 154
 155         // Report any lost events.
 156         if p.lostExtra > 0 {
 157                 hdr := [1]uint64{p.lostExtra}
 158                 lostStk := [2]uintptr{
 159                         funcPC(_LostExternalCode) + sys.PCQuantum,
 160                         funcPC(_ExternalCode) + sys.PCQuantum,
 161                 }
 162                 cpuprof.log.write(nil, 0, hdr[:], lostStk[:])
 163                 p.lostExtra = 0
 164         }
 165 }
 166
 167 func (p *cpuProfile) addLostAtomic64(count uint64) {
 168         hdr := [1]uint64{count}
 169         lostStk := [2]uintptr{
 170                 funcPC(_LostSIGPROFDuringAtomic64) + sys.PCQuantum,
 171                 funcPC(_System) + sys.PCQuantum,
 172         }
 173         cpuprof.log.write(nil, 0, hdr[:], lostStk[:])
 174 }
 175
 176 // CPUProfile panics.
 177 // It formerly provided raw access to chunks of
 178 // a pprof-format profile generated by the runtime.
 179 // The details of generating that format have changed,
 180 // so this functionality has been removed.
 181 //
 182 // Deprecated: use the runtime/pprof package,
 183 // or the handlers in the net/http/pprof package,
 184 // or the testing package's -test.cpuprofile flag instead.
 185 func CPUProfile() []byte {
 186         panic("CPUProfile no longer available")
 187 }
 188
 189 //go:linkname runtime_pprof_runtime_cyclesPerSecond runtime_pprof.runtime_cyclesPerSecond
 190 func runtime_pprof_runtime_cyclesPerSecond() int64 {
 191         return tickspersecond()
 192 }
 193
 194 // readProfile, provided to runtime/pprof, returns the next chunk of
 195 // binary CPU profiling stack trace data, blocking until data is available.
 196 // If profiling is turned off and all the profile data accumulated while it was
 197 // on has been returned, readProfile returns eof=true.
 198 // The caller must save the returned data and tags before calling readProfile again.
 199 //
 200 //go:linkname runtime_pprof_readProfile runtime_pprof.readProfile
 201 func runtime_pprof_readProfile() ([]uint64, []unsafe.Pointer, bool) {
 202         lock(&cpuprof.lock)
 203         log := cpuprof.log
 204         unlock(&cpuprof.lock)
 205         data, tags, eof := log.read(profBufBlocking)
 206         if len(data) == 0 && eof {
 207                 lock(&cpuprof.lock)
 208                 cpuprof.log = nil
 209                 unlock(&cpuprof.lock)
 210         }
 211         return data, tags, eof
 212 }