Salvatore Mazzarino <apocalips...@gmail.com> writes:
> I'm running QEMU v4.2.0 inside a Docker container. The Docker container runs > on a VMware VM running Flatcar Linux OS with kernel version 5.4.92. It > happens that the QEMU process starts to show up an high usage of CPU. The > CPUs that the Qemu process uses go into saturation and the guest OS running > by QEMU becomes totally unresponsive. Note that this issue does not occur > when QEMU runs on a bare metal machine. It only happens when we have nested > virtualization. > > I would like to debug the QEMU process to get where it get stuck. Then I > compiled a version with tracing enabled. Following the tracing doc I was able > to get traces of QEMU. > > Problem here is that I'm not sure exactly what function to trace. If I > trace all the file is huge. In 5 seconds of tracing in fact I get over > 2GB. I would need some guidance to narrow down the issue finding the > right QEMU trace to get some useful info. Which function would you > recommend to trace? Tracing is best used when tracking down very specific performance issues - as you have found it can very quickly saturate logs. Does perf top point you towards any areas where QEMU is particularly busy? Have you compared the perf report of a in-container vs out of container QEMU? -- Alex Bennée
