| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | package intelgpu |
| 4 | |
| 5 | import ( |
| 6 | "bufio" |
| 7 | "bytes" |
| 8 | "errors" |
| 9 | "os/exec" |
| 10 | "strconv" |
| 11 | "sync" |
| 12 | "time" |
| 13 | |
| 14 | "github.com/netdata/netdata/go/plugins/logger" |
| 15 | ) |
| 16 | |
| 17 | type intelGpuTop interface { |
| 18 | queryGPUSummaryJson() ([]byte, error) |
| 19 | stop() error |
| 20 | } |
| 21 | |
| 22 | func newIntelGpuTopExec(log *logger.Logger, ndsudoPath string, updateEvery int, device string) (*intelGpuTopExec, error) { |
| 23 | topExec := &intelGpuTopExec{ |
| 24 | Logger: log, |
| 25 | ndsudoPath: ndsudoPath, |
| 26 | updateEvery: updateEvery, |
| 27 | device: device, |
| 28 | firstSampleTimeout: time.Second * 3, |
| 29 | } |
| 30 | |
| 31 | if err := topExec.run(); err != nil { |
| 32 | return nil, err |
| 33 | } |
| 34 | |
| 35 | return topExec, nil |
| 36 | } |
| 37 | |
| 38 | type intelGpuTopExec struct { |
| 39 | *logger.Logger |
| 40 | |
| 41 | ndsudoPath string |
| 42 | updateEvery int |
| 43 | device string |
| 44 | firstSampleTimeout time.Duration |
| 45 | |
| 46 | cmd *exec.Cmd |
| 47 | done chan struct{} |
| 48 | |
| 49 | mux sync.Mutex |
| 50 | lastSample string |
| 51 | } |
| 52 | |
| 53 | func (e *intelGpuTopExec) run() error { |
| 54 | var cmd *exec.Cmd |
| 55 | |
| 56 | if e.device != "" { |
| 57 | cmd = exec.Command(e.ndsudoPath, "igt-device-json", "--interval", e.calcIntervalArg(), "--device", e.device) |
| 58 | } else { |
| 59 | cmd = exec.Command(e.ndsudoPath, "igt-json", "--interval", e.calcIntervalArg()) |
| 60 | } |
| 61 | |
| 62 | e.Debugf("executing '%s'", cmd) |
| 63 | |
| 64 | r, err := cmd.StdoutPipe() |
| 65 | if err != nil { |
| 66 | return err |
| 67 | } |
| 68 | |
| 69 | if err := cmd.Start(); err != nil { |
| 70 | return err |
| 71 | } |
| 72 | |
| 73 | firstSample := make(chan struct{}, 1) |
| 74 | done := make(chan struct{}) |
| 75 | e.cmd = cmd |
| 76 | e.done = done |
| 77 | |
| 78 | go func() { |
| 79 | defer close(done) |
| 80 | sc := bufio.NewScanner(r) |
| 81 | var buf bytes.Buffer |
| 82 | var n int |
| 83 | |
| 84 | for sc.Scan() { |
| 85 | if n++; n > 1000 { |
| 86 | break |
| 87 | } |
| 88 | |
| 89 | text := sc.Text() |
| 90 | |
| 91 | if buf.Len() == 0 && text != "{" || text == "" { |
| 92 | continue |
| 93 | } |
| 94 | |
| 95 | if text == "}," { |
| 96 | text = "}" |
| 97 | } |
| 98 | |
| 99 | buf.WriteString(text + "\n") |
| 100 | |
| 101 | if text[0] == '}' { |
| 102 | e.mux.Lock() |
| 103 | e.lastSample = buf.String() |
| 104 | e.mux.Unlock() |
| 105 | |
| 106 | select { |
| 107 | case firstSample <- struct{}{}: |
| 108 | default: |
| 109 | } |
| 110 | |
| 111 | buf.Reset() |
| 112 | n = 0 |
| 113 | } |
| 114 | } |
| 115 | }() |
| 116 | |
| 117 | select { |
| 118 | case <-e.done: |
| 119 | _ = e.stop() |
| 120 | return errors.New("process exited before the first sample was collected") |
| 121 | case <-time.After(e.firstSampleTimeout): |
| 122 | _ = e.stop() |
| 123 | return errors.New("timed out waiting for first sample") |
| 124 | case <-firstSample: |
| 125 | return nil |
| 126 | } |
| 127 | } |
| 128 | |
| 129 | func (e *intelGpuTopExec) queryGPUSummaryJson() ([]byte, error) { |
| 130 | select { |
| 131 | case <-e.done: |
| 132 | return nil, errors.New("process has already exited") |
| 133 | default: |
| 134 | } |
| 135 | |
| 136 | e.mux.Lock() |
| 137 | defer e.mux.Unlock() |
| 138 | |
| 139 | return []byte(e.lastSample), nil |
| 140 | } |
| 141 | |
| 142 | func (e *intelGpuTopExec) stop() error { |
| 143 | if e.cmd == nil || e.cmd.Process == nil { |
| 144 | return nil |
| 145 | } |
| 146 | |
| 147 | _ = e.cmd.Process.Kill() |
| 148 | _ = e.cmd.Wait() |
| 149 | e.cmd = nil |
| 150 | |
| 151 | select { |
| 152 | case <-e.done: |
| 153 | return nil |
| 154 | case <-time.After(time.Second * 2): |
| 155 | return errors.New("timed out waiting for process to exit") |
| 156 | } |
| 157 | } |
| 158 | |
| 159 | func (e *intelGpuTopExec) calcIntervalArg() string { |
| 160 | // intel_gpu_top appends the end marker ("},\n") of the previous sample to the beginning of the next sample. |
| 161 | // interval must be < than 'firstSampleTimeout' |
| 162 | interval := 900 |
| 163 | if m := min(e.updateEvery, int(e.firstSampleTimeout.Seconds())); m > 1 { |
| 164 | interval = m*1000 - 500 // milliseconds |
| 165 | } |
| 166 | return strconv.Itoa(interval) |
| 167 | } |