On Thu, Oct 23, 2014 at 10:39:52PM +0000, Mendez Montejano Pedro wrote: > Hello, > I am Pedro Mendez I am a master student from ETH Zurich, the reason that I am reaching you is because, > right now I am doing some recording of the events happening during launching in a vm, using KVM + qemu, > and for getting the events I am using the trace_pipe of the trace kernel tool utility with KVM, I already have events enable. > > I am redirecting the output of the trace_pipe with “cat" to a file so I can the events from the launching, but > when I see the size of the logs I am getting really incongruent results, sometimes I get a file > of 45MB and when I launch again the vm and recording the events of the launching I end up > with a file of 400K or 1.2MB. > > I don’t thinks this is normal, I wonder if I am not getting the event from the right place of if I should > use another kernel tool for tracing the events, or do you know if this behaviour is expected > to be normal. > > I hope you can tell me more about this or if you have seen this behaviour before. I'm not sure what "the trace_pipe of the trace kernel tool utility with KVM" really means. Sounds like you are using ftrace and the trace events in /sys/kernel/debug/tracing/events/kvm. What are the exact command-lines you are using? If you use "perf record -e kvm:\* -a" it tells you whether any events were dropped. I never had serious problems with dropped events. If that you are hitting dropped events, try one or more of the following: 1. Trace fewer high-frequency events (be selective about events you enable) 2. Reduce the workload so more cycles are available to harvest ring buffers 3. Increase ring buffer size Stefan
Attachment:
pgpCUA0U7dtMC.pgp
Description: PGP signature
