RHEL 9.3.0 / CentOS clone +++ This bug was initially created as a clone of Bug #2214884 +++ Description of problem: Under ocp4.13/ocp-v 4.13.1-188 env, the real time worker node is installed with RHCOS based on RHEL9.2, it's found "# Max Latencies: 00090 00096 00141" when run 1 hour cyclictest, while a maximum of 50us is acceptable. From Marcelo, it's caused by "query-stats QMP command interrupts vcpus.", "[PATCH] kvm: reuse per-vcpu stats fd to avoid vcpu interruption" was submitted. Version-Release number of selected component (if applicable): Host kernel: Scratch kernel 5.14.0-284.13.1.rt14.298kvmrtv1.el9_2.x86_64 Guest kernel: Scratch kernel 5.14.0-284.13.1.rt14.298kvmrtv1.el9_2.x86_64 Qemu: qemu-kvm-core-7.2.0-14.el9_2.x86_64 How reproducible: 100% Steps to Reproduce: 1. 2. 3. Actual results: Expected results: Additional info:
Adding zstream and ZTR=9.2.0, see bug 2214884 for details, especially https://bugzilla.redhat.com/show_bug.cgi?id=2214884#c2
*** Bug 2214884 has been marked as a duplicate of this bug. ***
Test steps to verify bug: 1) Run qemu-kvm in parallel with rt-trace-bpf, with isolated vcpu. 2) Run query-stats command on QMP (https://patchew.org/Libvirt/20220624081449.70085-1-natto@weirdnatto.in/20220624081449.70085-3-natto@weirdnatto.in/), should see IPIs to pcpu where isolated vcpu runs. 3) with fixed qemu-kvm, should be able to run query-stats QMP command and not see the IPIs.
QE bot(pre verify): Set 'Verified:Tested,SanityOnly' as gating/tier1 test pass.