Red Hat Bugzilla – Bug 547410
guest cpu stuck for 10s when guest is loaded
Last modified: 2013-01-09 17:07:01 EST
Created attachment 378255 [details]
call trace in guest
Description of problem:
1. found calltrace in guest(attached is the log), I met the calltrace twice. one is when scp a large file(3.4G) from a remote machine to the guest. another one is when running iozone test inside guest.
2. nothing interesting in host was found.
3. the image files are on remote nfs server.
qemu-kvm CLI :/usr/libexec/qemu-kvm -m 2048 -smp 2 -drive file=/data/images/t78/images/ChenZhou/5.4-64.virtio.virtio.qcow2.20,if=virtio,boot=on,cache=off -drive file=/data/images/t78/images/test2.qcow2.64k,if=virtio,cache=off -drive file=/data/images/t195/images/test1.qcow2.4k,if=virtio,cache=off -net nic -net tap -vnc :2 -monitor stdio -usbdevice tablet -uuid c3105705-6c51-40b5-80eb-a0ffa387575a
iozone CLI :./iozone -aZR -n 16g -g 22g -y 32k -q 64k -M -b /root/result.xls -f /mnt/1/testfile
5. guest is RHEL5.4 X86_64
Version-Release number of selected component (if applicable):
Red Hat Enterprise Virtualization Hypervisor release 5.4-2.1 (2)
reproduce twice so far
Steps to Reproduce:
I'm not sure this is a bug, it might be due too stall by synchronous first time allocation done by qcow2.
Can you first fully populate the guest image by writing to the entire disk and afterwards erase these dummy files and return the test again?
(In reply to comment #1)
> I'm not sure this is a bug, it might be due too stall by synchronous first time
> allocation done by qcow2.
> Can you first fully populate the guest image by writing to the entire disk and
> afterwards erase these dummy files and return the test again?
Yes. test 3 times,only reproduce when writing a fresh disk.