On Fri, 14 Jul 2017, Alexander Shishkin wrote: > Vince Weaver <vincent.wea...@maine.edu> writes: > > > I was tracking down some regressions in my perf_event_test testsuite. > > Some of the tests broke in the 4.11-rc1 timeframe. > > > > I've bisected one of them, this report is about > > tests/overflow/simul_oneshot_group_overflow > > This test creates an event group containing two sampling events, set > > to overflow to a signal handler (which disables and then refreshes the > > event). > > > > On a good kernel you get the following: > > Event perf::instructions with period 1000000 > > Event perf::instructions with period 2000000 > > fd 3 overflows: 946 (perf::instructions/1000000) > > fd 4 overflows: 473 (perf::instructions/2000000) > > Ending counts: > > Count 0: 946379875 > > Count 1: 946365218 > > > > With the broken kernels you get: > > Event perf::instructions with period 1000000 > > Event perf::instructions with period 2000000 > > fd 3 overflows: 938 (perf::instructions/1000000) > > fd 4 overflows: 318 (perf::instructions/2000000) > > Ending counts: > > Count 0: 946373080 > > Count 1: 653373058 > > I'm not sure I'm seeing it (granted, it's a friday evening): is it the > difference in overflow counts?
It's two things. It's created an grouped event, with the two events both perf::instructions. 1. The total count at the end should be the same for both (on the failing kernels it is not) 2. The overflow count for both events should be roughly total_events/sample_freq. (on the failing kernels it is not) > Also, are they cpu or task bound? The open looks like this: perf_event_open(&pe,0,-1,-1,0); On the failing case, the group leader is pinned. The source code for the test is here: https://github.com/deater/perf_event_tests/blob/master/tests/overflow/simul_oneshot_group_overflow.c Vince