[telemetry] Fix CPU metric jiffie overflow correction.

tools/perf/metrics/cpu.py

Index: tools/perf/metrics/cpu.py
diff --git a/tools/perf/metrics/cpu.py b/tools/perf/metrics/cpu.py
index 0709783965d59acb5badafda51a3e278273cf595..a4fd859d332cdb1ca83fc5f6107b766b5cb7f9b1 100644
--- a/tools/perf/metrics/cpu.py
+++ b/tools/perf/metrics/cpu.py
@@ -87,9 +87,9 @@ def _SubtractCpuStats(cpu_stats, start_cpu_stats):
                   start_cpu_stats[process_type]['TotalTime'])
     # Fix overflow for 32-bit jiffie counter, 64-bit counter will not overflow.
     # Linux kernel starts with a value close to an overflow, so correction is
-    # necessary.
+    # necessary. Assume jiffie counter is at 100 Hz.
     if total_time < 0:
-      total_time += 2 ** 32
+      total_time += 2 ** 32 / 100.

[mikecase (-- gone --), 2016/12/06 19:54:15]

do you need to divide by 100 if total_time > 0?

[dtu, 2016/12/06 20:33:58]

No, the log shows that the values in cpu_stats were already divided by 100. So
we just need to divide this correction factor to match the same units.


Locals:
  cpu_process_time : 0.23000000000000004
  cpu_stats        : {'Gpu': {'TotalTime': 5.46, 'CpuProcessTime': 0.68},
'Other': {}, 'Renderer': {'TotalTime': 5.46, 'CpuProcessTime': 0.91}, 'Browser':
{'TotalTime': 5.46, 'CpuProcessTime': 2.51}}
  cpu_usage        : {}
  process_type     : 'Gpu'
  start_cpu_stats  : {'Gpu': {'TotalTime': 42949670.36, 'CpuProcessTime': 0.45},
'Other': {}, 'Renderer': {'TotalTime': 42949670.36, 'CpuProcessTime': 0.63},
'Browser': {'TotalTime': 42949670.36, 'CpuProcessTime': 2.24}}
  total_time       : 4252017631.1

     # Assert that the arguments were given in the correct order.
     assert total_time > 0 and total_time < 2 ** 31, (

[Sami, 2016/12/07 17:20:12]

Should this upper limit be fixed too?

[dtu, 2016/12/08 00:55:29]

Done.

         'Expected total_time > 0, was: %d' % total_time)