Bug 1746282
| Summary: | qemu-kvm: backport cpuidle-haltpoll support [rhel-7.7.z] | ||||||
|---|---|---|---|---|---|---|---|
| Product: | Red Hat Enterprise Linux 7 | Reporter: | RAD team bot copy to z-stream <autobot-eus-copy> | ||||
| Component: | qemu-kvm-rhev | Assignee: | Marcelo Tosatti <mtosatti> | ||||
| Status: | CLOSED ERRATA | QA Contact: | jingzhao <jinzhao> | ||||
| Severity: | unspecified | Docs Contact: | |||||
| Priority: | high | ||||||
| Version: | 7.8 | CC: | chayang, hhuang, jinzhao, juzhang, mrezanin, mtessun, mtosatti, virt-maint | ||||
| Target Milestone: | rc | Keywords: | ZStream | ||||
| Target Release: | --- | ||||||
| Hardware: | Unspecified | ||||||
| OS: | Unspecified | ||||||
| Whiteboard: | |||||||
| Fixed In Version: | qemu-kvm-rhev-2.12.0-33.el7_7.3 | Doc Type: | If docs needed, set a value | ||||
| Doc Text: | Story Points: | --- | |||||
| Clone Of: | 1734502 | Environment: | |||||
| Last Closed: | 2019-10-22 15:24:05 UTC | Type: | --- | ||||
| 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: | 1734502, 1759283 | ||||||
| Bug Blocks: | |||||||
| Attachments: |
|
||||||
|
Description
RAD team bot copy to z-stream
2019-08-28 07:26:03 UTC
Created attachment 1618874 [details]
dmesg log of guest
Since the problem described in this bug report should be resolved in a recent advisory, it has been closed with a resolution of ERRATA. For information on the advisory, and where to find the updated files, follow the link below. If the solution does not work for you, open a new bug report. https://access.redhat.com/errata/RHSA-2019:3179 Rationale for cpuidle haltpoll driver and usage instructions:
That for baremetal systems, when a CPU waits for a certain event,
such as a waitqueue notification, the kernel implements this as
a "mwait" instruction based wait (waiter). This avoids the CPU that wakes
the waiter from sending an IPI (and waiter from processing that IPI).
That can have a significant performance impact (up to 10%) if the
application is multithreaded, the threads communicate with each
other, and sleep for short intervals of time.
The mwait instruction can't be emulated on KVM easily: so instead
the haltpoll driver busy spins for a specified amount of time
when a guest vCPU halts.
To enable the driver in a guest, the user has to:
1) Enable the kvm-poll-control CPU flag for the emulated vCPU. For
example, if using qemu directly the relevant command line section
would be (libvirt configuration should be mentioned):
-cpu SandyBridge,tsc-deadline=on,pmu=off,vmx=off,kvm-poll-control=on
2) Load the cpuidle-haltpoll driver in the guest persistently ("PERSISTENT MODULE LOADING" section of RHEL documentation).
3) Tune the guest_halt_poll_ns module parameter: should be slightly larger
than the average delay the application in question performs. Suggest to benchmark
the app while increasing/decreasing guest_halt_poll_ns.
|