From: Xiao Guangrong<xiaoguangrong@xxxxxxxxxxxxxxxxxx> This patchset introduces a perf-based tool (perf kvm stat record/report) which can analysis kvm events more smartly. This is a presentation on 2012 Japan LinuxCon: http://events.linuxfoundation.org/images/stories/pdf/lcjp2012_guangrong.pdf You can get more detail from it. Any question/comment please feel free let us know. Patch 1 and patch 3 can be applied to either tip tree or kvm tree, but patch 2 can only be applied to kvm tree. Fortunately, patch 2 is just doing the "improvement" work, and it can be picked up independently. Usage: - kvm stat run a command and gather performance counter statistics, it is the alias of perf stat - trace kvm events: perf kvm stat record, or, if other tracepoints are interesting as well, we can append the events like this: perf kvm stat record -e timer:* If many guests are running, we can track the specified guest by using -p or --pid - show the result: perf kvm stat report The output example is following: # pgrep qemu-kvm 27841 27888 27936 total 3 guests are running on the host Then, track the guest whose pid is 27936: # ./perf kvm stat record -p 27936 ^C[ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.177 MB perf.data.guest (~7742 samples) ] See the vmexit events: # ./perf kvm stat report --event=vmexit Analyze events for all VCPUs: VM-EXIT Samples Samples% Time% Avg time APIC_ACCESS 237 80.34% 0.01% 20.86us ( +- 8.03% ) HLT 45 15.25% 99.98% 790874.75us ( +- 16.93% ) EXTERNAL_INTERRUPT 11 3.73% 0.00% 47.70us ( +- 17.45% ) EXCEPTION_NMI 1 0.34% 0.00% 5.11us ( +- -nan% ) CR_ACCESS 1 0.34% 0.00% 6.33us ( +- -nan% ) Total Samples:295, Total events handled time:35594844.34us. See the mmio events: # ./perf kvm stat report --event=mmio Analyze events for all VCPUs: MMIO Access Samples Samples% Time% Avg time 0xfee00380:W 151 82.07% 76.19% 7.72us ( +- 11.96% ) 0xfee00300:W 11 5.98% 18.86% 26.23us ( +- 27.33% ) 0xfee00300:R 11 5.98% 2.16% 3.00us ( +- 9.17% ) 0xfee00310:W 11 5.98% 2.79% 3.88us ( +- 9.15% ) Total Samples:184, Total events handled time:1529.42us. See the ioport event: # ./perf kvm stat report --event=ioport Analyze events for all VCPUs: IO Port Access Samples Samples% Time% Avg time 0x5658:PIN 3357 55.22% 92.13% 162.31us ( +- 0.87% ) 0xc090:POUT 1221 20.09% 2.10% 10.18us ( +- 4.97% ) 0x60:PIN 748 12.30% 3.18% 25.17us ( +- 5.01% ) 0x64:PIN 748 12.30% 2.57% 20.35us ( +- 11.81% ) 0xc050:POUT 5 0.08% 0.01% 8.79us ( +- 12.88% ) Total Samples:6079, Total events handled time:591405.43us. And, --vcpu is used to track the specified vcpu and --key is used to sort the result: # ./perf kvm stat report --event=ioport --vcpu=0 --key=time Analyze events for VCPU 0: IO Port Access Samples Samples% Time% Avg time 0x5658:PIN 145 94.77% 99.67% 133.00us ( +- 2.96% ) 0xc090:POUT 8 5.23% 0.33% 7.99us ( +- 16.85% ) Total Samples:153, Total events handled time:19348.87us. Changelog: - merge "perf kvm-events" into perf "perf kvm stat" as Ingo's suggestion - track kvm events for the specified guest - rename kvm_mmio_done to kvm_io_done - fix compiling-error on i386 Dong Hao (3): KVM: x86: export svm/vmx exit code and vector code to userspace KVM: x86: tracemmio begin and complete KVM: perf kvm events analysis tool arch/x86/include/asm/kvm_host.h | 36 +- arch/x86/include/asm/svm.h | 205 +++++--- arch/x86/include/asm/vmx.h | 126 ++++-- arch/x86/kvm/trace.h | 89 ---- arch/x86/kvm/x86.c | 32 +- include/trace/events/kvm.h | 37 ++ tools/perf/Documentation/perf-kvm.txt | 30 ++- tools/perf/MANIFEST | 3 + tools/perf/builtin-kvm.c | 858 ++++++++++++++++++++++++++++++++- tools/perf/util/header.c | 55 ++- tools/perf/util/header.h | 1 + tools/perf/util/thread.h | 2 + 12 files changed, 1234 insertions(+), 240 deletions(-) -- 1.7.2.5 -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
