Skip to contentSkip to conversation

qemu/kvm vhost process looping (RHEL6)

DISCUSSION Chris Maltby 2 participants 2 messages Feb 20, 2014 – Feb 24, 2014 KVM

lore
Share & tools
lore
#0 lore raw

I'm not sure that this is the right place to raise this issue, so apologies in advance.

I have an RHEL6 system with two qemu/kvm guests, one running RHEL5 (x64) and one running Windows 7 (x64). These variously provide local and remote access net services in a small office environment. I have configured networking using a bridge config and with guest virtio drivers. There is no shortage of CPU or RAM on the host system or in the guests.

The problem is that bursts of traffic (eg rsync) between host and guest or guest/guest result in the associated vhost process entering a CPU loop which can last for up to 30 minutes. During this period the net interface is offline. It recovers spontaneously after some time and for no obvious reason. Using ifconfig down/up sequences has no impact. Nothing is logged by the kernel or qemu-kvm. All other network interfaces (physical and virtual) continue to function normally.

I can reproduce the problem in any virtual setup that uses vhost networking. I have looked extensively in the Redhat forums etc and haven't seen anyone else reporting a similar problem, nor have I found any mention of anything like this in the kvm forums. Obviously, these are not the current versions of the kernel or qemu-kvm, but it may be that the problem still exists.

The installed versions are:

qemu-kvm-0.12.1.2-2.415.el6_5.3.x86_64
qemu-img-0.12.1.2-2.415.el6_5.3.x86_64
libvirt-0.10.2-29.el6_5.3.x86_64
kernel-2.6.32-431.3.1.el6.x86_64

I can work around the problem by configuring qemu to perform the network IO in process and not in a vhost thread, so it's clearly a vhost issue.

I have a very similar system at another site running an older version of the kernel and qemu-kvm which does not exhibit the problem:

qemu-kvm-0.12.1.2-2.295.el6_3.10.x86_64
qemu-img-0.12.1.2-2.295.el6_3.10.x86_64
libvirt-0.9.10-21.el6_3.8.x86_64
kernel-2.6.32-279.19.1.el6.x86_64

#1 lore raw
Quoted 37 lines · Chris Maltby

I'm not sure that this is the right place to raise this issue, so apologies in advance.

I have an RHEL6 system with two qemu/kvm guests, one running RHEL5 (x64) and one running Windows 7 (x64). These variously provide local and remote access net services in a small office environment. I have configured networking using a bridge config and with guest virtio drivers. There is no shortage of CPU or RAM on the host system or in the guests.

The problem is that bursts of traffic (eg rsync) between host and guest or guest/guest result in the associated vhost process entering a CPU loop which can last for up to 30 minutes. During this period the net interface is offline. It recovers spontaneously after some time and for no obvious reason. Using ifconfig down/up sequences has no impact. Nothing is logged by the kernel or qemu-kvm. All other network interfaces (physical and virtual) continue to function normally.

I can reproduce the problem in any virtual setup that uses vhost networking. I have looked extensively in the Redhat forums etc and haven't seen anyone else reporting a similar problem, nor have I found any mention of anything like this in the kvm forums. Obviously, these are not the current versions of the kernel or qemu-kvm, but it may be that the problem still exists.

The installed versions are:

qemu-kvm-0.12.1.2-2.415.el6_5.3.x86_64
qemu-img-0.12.1.2-2.415.el6_5.3.x86_64
libvirt-0.10.2-29.el6_5.3.x86_64
kernel-2.6.32-431.3.1.el6.x86_64

I can work around the problem by configuring qemu to perform the network IO in process and not in a vhost thread, so it's clearly a vhost issue.

Quoted 7 lines

I have a very similar system at another site running an older version of the kernel and qemu-kvm which does not exhibit the problem:

qemu-kvm-0.12.1.2-2.295.el6_3.10.x86_64
qemu-img-0.12.1.2-2.295.el6_3.10.x86_64
libvirt-0.9.10-21.el6_3.8.x86_64
kernel-2.6.32-279.19.1.el6.x86_64

Does the issue reproduce with the latest upstream release or git tree (linux.git + qemu.git)?

If not (or if you cannot test upstream), please report this issue to Red Hat. The upstream mailing lists focus on linux.git/qemu.git.

Stefan