Bug 1346430
| Summary: | Domain using vcpusched element fails to start on non-RT kernel | ||
|---|---|---|---|
| Product: | Red Hat Enterprise Linux 7 | Reporter: | Luiz Capitulino <lcapitulino> |
| Component: | libvirt | Assignee: | Libvirt Maintainers <libvirt-maint> |
| Status: | CLOSED DUPLICATE | QA Contact: | Virtualization Bugs <virt-bugs> |
| Severity: | unspecified | Docs Contact: | |
| Priority: | unspecified | ||
| Version: | 7.3 | CC: | atheurer, krister, mkletzan, rbalakri |
| Target Milestone: | rc | ||
| Target Release: | --- | ||
| Hardware: | Unspecified | ||
| OS: | Unspecified | ||
| Whiteboard: | |||
| Fixed In Version: | Doc Type: | If docs needed, set a value | |
| Doc Text: | Story Points: | --- | |
| Clone Of: | Environment: | ||
| Last Closed: | 2016-06-15 19:45:14 UTC | Type: | Bug |
| Regression: | --- | Mount Type: | --- |
| Documentation: | --- | CRM: | |
| Verified Versions: | Category: | --- | |
| oVirt Team: | --- | RHEL 7.3 requirements from Atomic Host: | |
| Cloudforms Team: | --- | Target Upstream Version: | |
| Embargoed: | |||
| Bug Depends On: | |||
| Bug Blocks: | 1175463 | ||
Karl Rister debugged this issue some time ago and found the following: """ I believe I have traced it back to a cgroup partitioning issue. Libvirt is placing the guest process in a specific partition which does not have any RT runtime allocated to it, by doing so and then propagating the values through the cgroup hierarchy I am finally able to manually chrt the vcpu pids: root@host ~ $ virsh qemu-monitor-command master --hmp info cpus * CPU #0: pc=0xffffffff8115ac90 thread_id=3784 CPU #1: pc=0xffffffff8104a086 (halted) thread_id=3785 CPU #2: pc=0xffffffff8104a086 (halted) thread_id=3786 root@host ~ $ chrt -f -p 1 3785 chrt: failed to set pid 3785's policy: Operation not permitted root@host ~ $ chrt -f -p 1 3786 chrt: failed to set pid 3786's policy: Operation not permitted root@host ~ $ cat /sys/fs/cgroup/cpu,cpuacct/cpu.rt_runtime_us 950000 root@host ~ $ cat /sys/fs/cgroup/cpu,cpuacct/machine.slice/cpu.rt_runtime_us 0 root@host ~ $ echo 950000 > /sys/fs/cgroup/cpu,cpuacct/machine.slice/cpu.rt_runtime_us root@host ~ $ cat /sys/fs/cgroup/cpu\,cpuacct/machine.slice/machine-qemu\\x2dmaster.scope/cpu.rt_runtime_us 0 root@host ~ $ echo 950000 > /sys/fs/cgroup/cpu\,cpuacct/machine.slice/machine-qemu\\x2dmaster.scope/cpu.rt_runtime_us root@host ~ $ cat /sys/fs/cgroup/cpu\,cpuacct/machine.slice/machine-qemu\\x2dmaster.scope/vcpu1/cpu.rt_runtime_us 0 root@host ~ $ echo 950000 > /sys/fs/cgroup/cpu\,cpuacct/machine.slice/machine-qemu\\x2dmaster.scope/vcpu1/cpu.rt_runtime_us root@host ~ $ chrt -f -p 1 3785 root@host ~ $ cat /sys/fs/cgroup/cpu\,cpuacct/machine.slice/machine-qemu\\x2dmaster.scope/vcpu2/cpu.rt_runtime_us 0 root@host ~ $ echo 950000 > /sys/fs/cgroup/cpu\,cpuacct/machine.slice/machine-qemu\\x2dmaster.scope/vcpu2/cpu.rt_runtime_us bash: echo: write error: Invalid argument I have to split the available runtime... """ This makes sense because it seems that the RT kernel has less cgroups options enabled, so maybe it's missing a kernel config option that enforces this behavior which makes the guest work on the RT kernel. Harald Jensås also got this issue when debugged it down to a tuned profile setting. Closing as a dupe. *** This bug has been marked as a duplicate of bug 1346715 *** |
Description of problem: A domain using this vCPU placement & configuration: <vcpu placement='static'>2</vcpu> <cputune> <vcpupin vcpu='0' cpuset='14'/> <vcpupin vcpu='1' cpuset='15'/> <emulatorpin cpuset='0,2,4,6'/> <vcpusched vcpus='0-1' scheduler='fifo' priority='1'/> </cputune> Won't start on a system using the NFV profile but running the non-RT kernel: [root@virtlab508 ~]# virsh start RHEL73-RT error: Failed to start domain RHEL73-RT error: Cannot set scheduler parameters for pid 161919: Operation not permitted [root@virtlab508 ~]# However, it starts fine if we do one of the following: - Run the RT kernel in the host - Drop the vcpusched line Version-Release number of selected component (if applicable): libvirt-1.3.4-1.el7.x86_64 How reproducible: Steps to Reproduce: 1. Install the tuned profile-nfv in the host and isolate some cores (this step might not be needed) 2. Use the cputune and vcpusched elements to pin and assign RT priority to vCPUs 3. Try to start the guest with "virsh start" 4. Guest won't start