Messages in this thread | | | Date | Thu, 24 Aug 2017 13:51:59 -0700 | From | Andi Kleen <> | Subject | Re: [PATCH] perf stat: Add event-interval-print option |
| |
On Thu, Aug 24, 2017 at 04:39:02PM -0400, Aaron Lindsay wrote: > Does anyone have any feedback on this patch or the idea in general? > > Also, (for our bookkeeping - not trying to rush things) is there any > chance this will still make it in for the 4.14 merge window, or is it > 4.15 material at the earliest?
You can already do that with leader sampling through perf record
e.g.
% perf record -e '{cpu/cpu-cycles,period=10000/,branches,branch-misses}:S' ...
% perf script swapper 0 [005] 3905226.804037: 10210 cpu/cpu-cycles,period=10000/: ffffffff9e87643c intel_idle ([kernel.kallsyms]) swapper 0 [005] 3905226.804037: 345 branches: ffffffff9e87643c intel_idle ([kernel.kallsyms]) swapper 0 [005] 3905226.804037: 58 branch-misses: ffffffff9e87643c intel_idle ([kernel.kallsyms]) :16797 16797 [004] 3905226.804038: 10160 cpu/cpu-cycles,period=10000/: ffffffffc07fdd10 kvm_fetch_guest_virt ([kvm]) :16797 16797 [004] 3905226.804038: 515 branches: ffffffffc07fdd10 kvm_fetch_guest_virt ([kvm]) :16797 16797 [004] 3905226.804038: 19 branch-misses: ffffffffc07fdd10 kvm_fetch_guest_virt ([kvm])
Yes it's a useful technique.
-Andi
| |