Re: [PATCH RESEND] kprobe: Use percpu counter to collect nhit statistics

2016-02-09 Thread Ingo Molnar
* Steven Rostedt wrote: > On Thu, 4 Feb 2016 10:24:12 + > 平松雅巳 / HIRAMATU，MASAMI wrote: > > > From: Martin KaFai Lau [mailto:ka...@fb.com] > > > > > >When doing ebpf+kprobe on some hot TCP functions (e.g. > > >tcp_rcv_established), the kprobe_dispatcher shows up in 'perf report'. > > > > >

Re: [PATCH RESEND] kprobe: Use percpu counter to collect nhit statistics

2016-02-04 Thread Steven Rostedt
On Thu, 4 Feb 2016 10:24:12 + 平松雅巳 / HIRAMATU，MASAMI wrote: > From: Martin KaFai Lau [mailto:ka...@fb.com] > > > >When doing ebpf+kprobe on some hot TCP functions (e.g. > >tcp_rcv_established), the kprobe_dispatcher shows up in 'perf report'. > > > >In kprobe_dispatcher(), there is a lot of c

RE: [PATCH RESEND] kprobe: Use percpu counter to collect nhit statistics

2016-02-04 Thread 平松雅巳 / HIRAMATU，MASAMI
From: Martin KaFai Lau [mailto:ka...@fb.com] > >When doing ebpf+kprobe on some hot TCP functions (e.g. >tcp_rcv_established), the kprobe_dispatcher shows up in 'perf report'. > >In kprobe_dispatcher(), there is a lot of cache bouncing >in 'tk->nhit++'. 'tk->nhit' and 'tk->tp.flags' also share >the

[PATCH RESEND] kprobe: Use percpu counter to collect nhit statistics

2016-02-03 Thread Martin KaFai Lau
When doing ebpf+kprobe on some hot TCP functions (e.g. tcp_rcv_established), the kprobe_dispatcher shows up in 'perf report'. In kprobe_dispatcher(), there is a lot of cache bouncing in 'tk->nhit++'. 'tk->nhit' and 'tk->tp.flags' also share the same cacheline. perf report (cycles:pp): 8.30% ipv