Fedora 12 Schedule
The Fedora 12 release is looming large. At this stage, all efforts
should be focused on polishing off everything for the final release.
2009-09-17 Virt Test Day (6 days)
2009-09-29 Final Development Freeze (18 days)
2009-10-28 Release Candidate Compose (47 days)
Fedora Test Day
Next Thursday, September 17th, we will be holding be holding Fedora
Test Day in order to test all the latest Fedora 12 virt goodies:
* PXE booting
* Live Migration
* The qcow2 image format
* Huge page backed memory
* NIC hotplug
* Stable guest ABI
Preparations for the day are still in their early stages, so if you
want to help out, please dive right in.
Most importantly, though, please do make an effort to turn up on the
day, grab some test cases and report bugs!
KSM Tuning Daemon
On the fedora-virt list, Dan Kenigsberg and Dan Berrange discussed
creating a KSM tuning service:
An unattended host running a variable number of qemu-kvm's needs to
tune ksm automatically, since when memory is tight, it's better to
spend more cpu on merging pages. In more relaxed cases, it's just a
waste of time.
The attached service tries to do just that.
It monitors how much memory is used by qemu-kvm processes, and
starts ksm when a threshold is passed. Ksm usually manages to free
up some memory.
This relates to the new KSM feature in Fedora 12:
Dale Bewley continues to churn out excellent virt updates for Fedora
Dale also posted a sneak preview of the extensive virtualization
section for the Fedora 12 release notes:
Huge amounts of activity is going on the fedora-xen list around
testing out the latest Xen pv_ops Dom0 kernels. It's all too much to
summarize here, so if you're interested, take a look a the list
DOOM-O-METER: 214 open bugs now, 217 three weeks ago
The Fedora 12 blocker list is:
and the Fedora 12 target list is:
All these queries can be found on:
== misc ==
Directory permissions on volume group directory too restrictive
The change to fix volume group directory permissions is still
blocking on a change over to udev. Time is getting tight for
== kernel ==
virtio_net page allocation failure
A report of a 2.6.29 virtio_net oops which Rusty recently fixed
rotational mode is much faster for virtio-blk disks, but uses
non-rotational mode by default
Christoph is having trouble reproducing this issue.
Unable to boot using qemu-kvm and gPXE from virt-preview
This issue should be fixed in 126.96.36.199 which hopefully will
appear in F-11 updates soon.
== qemu ==
qemu-kvm exits when initializing MSI-X fails on older kernels
virtio's MSI-X support causes qemu to exit if the kernel does not
support it. This is now fixed in rawhide by backporting a fix
QEMU does not exit if it can't attach to the host USB device
Dan Berrange notes that qemu doesn't give the user any useful
error it fails to pass through a host USB device to the guest.
QEMU threads should share the same I/O context
Implement support for CLONE_IO
Uli doesn't think CLONE_IO should be a thread attribute and has
closed the glibc bug as WONTFIX. The suggestion now is to make
this a process-wide attribute configurable with prctl().
qemu slows down dramatically using the SDL graphics backend and
ac97 sound driver
It looks like there may be something specific to SDL and ac97
causing qemu to slow down hugely in Fedora 11.
== libvirt ==
libvirt is using untrusted 'info vcpus' PID data for already
running VM after libvirtd restart
Dan Berrange points out that if a qemu process is compromised, it
can cause libvirtd to kill another process.
libvirt/netcf loads modprobe.conf and others - AVC messages
(preventing libvirtd (virtd_t) "getattr" modules_conf_t)
David Lutterkort thought it may be possible to get bonding to work
without modifying modprobe.conf, but that turns out not to be the
case so we need the selinux-policy to allow it.
libvirt should check for -soundhw support; qemu-system-sparc does
not have soundhw support
It turns out that the sparc target does not have -soundhw
support. We need libvirt to detect that.
== misc ==
Guest clock is running aprox. 3 seconds before host clock.
Now that the kernel is configured with CONFIG_RTC_HCTOSYS,
initscripts runs hwclock --systz rather than --hctosys, which in
turn means that the hwclock isn't incorrectly introducing an
offset between the guest and host clocks.
== kernel ==
KSM breaks encryption 157 > kernel > 139 - KSM support now
Andrea tracked down and fixed this KSM blocker and Justin pushed
the fix to rawhide.
== qemu ==
kvm forgets to close /proc/meminfo when hugepages are active
F-11 qemu-kvm is leaking a /proc/meminfo when using huge
pages. Fixed by backporting a change to use statfs rather than
qemu locks up at shutdown with sdl audio driver
It appears several users are using qemu with QEMU_AUDIO_DRV=pa in
order to get working sound. F-11 and F-12 are now switched over to
used this backend by default.
qemu segfault when VNC client disconnects
qemu VNC :: xterm inside VM shows garbled text
Fixes for both of these Fedora 11 bugs have now been backported
and will appear in updates shortly.
== libvirt ==
selinux policy update needed :: libvirt has moved the location of
the qemu monitor socket
libvirt has moved where the monitor socket is created, so the
selinux policy needed updating. This is fixed in rawhide now.
unable to start guest: qemu: could not open serial device 'pty'
This turned out to be caused by incorrect permissions on
/dev/ptmx. We're still not sure how that came about.
libvirtd should chown dirs when qemu configured to run as
Dan Berrange fixed this upstream for libvirt-0.7.1.
RFE: libvirt should support KVM huge page backed memory
Dan Berrange polished off John Cooper's patch and pushed it in
time for libvirt-0.7.1.
libvirt always places IDE disks before virtio/scsi disks
A fix for this has been committed upstream and will be pulled into
== xen ==
2.6.31-rc1 xen domU crashes early during boot
This Xen DomU pv_ops blocker should now be fixed in rawhide thanks
to a patch from Jeremy Fitzhardinge.
kernel backtrace: possible recursive locking detected on Xen domU
A kernel lockdep warning seen under F-12 Xen DomU.