2009-03-03 Feature freeze (6 days) 2009-03-10 Beta Freeze (13 days) 2009-04-14 Final freeze (48 days)
An early report this week because I'm about to disappear for three weeks. Just before the feature/beta freeze ... nice timing, I know! :-)
F11 Features ============
Dale Bewley provides a timely reminder that all but one of the F11 virt features have not yet been approved by FESCo:
http://www.redhat.com/archives/fedora-virt/2009-February/msg00088.html
FWN ===
This weeks Fedora Weekly News is out with another great virtualization section:
https://fedoraproject.org/wiki/FWN/Issue164#Virtualization
Tablet Pointer ==============
Dan Berrange did some experimentation with making Fedora be able to sanely handle a tablet pointer without extra configuration and filed these two bugs:
https://bugzilla.redhat.com/487025 Add a HAL FDI file for the QEMU tablet device
https://bugzilla.redhat.com/487028 Default to using the table in python-virtinst for F11 guests:
The idea here is that if we use the tablet pointer with F11 guest by default it will fix the screwiness with mouse pointers in guests by using absolute pointer positioning rather than relative positioning.
High Resolution in QEMU =======================
In order to use a higher resolution when running Fedora on qemu/kvm, we're planning on using the PCI subvendor ID to allow the Xorg cirrus driver to detect that it's a QEMU emulated card and use 1024x768.
Dan Berrange filed two bugs to track this:
https://bugzilla.redhat.com/487118 Need to add subsystem vendor ID for kvm/qemu cirrus card
https://bugzilla.redhat.com/251264 xorg should use 1024x768 for kvm/qemu cirrus card
Device Assignment =================
Several libvirt patch sets were posted to add device dettach, re-attach and reset support:
http://www.redhat.com/archives/libvir-list/2009-February/msg00393.html http://www.redhat.com/archives/libvir-list/2009-February/msg00438.html http://www.redhat.com/archives/libvir-list/2009-February/msg00441.html
Also, Chris Wright posted the patch for "remove_id" in sysfs:
http://marc.info/?l=kvm&m=123545476703098 https://bugzilla.redhat.com/487103
QEMU ====
Rumour has it that there may be an upstream release Real Soon Now.
The package reviews for the three new BIOS packages are still in progress:
https://bugzilla.redhat.com/485417 https://bugzilla.redhat.com/485418 https://bugzilla.redhat.com/485420
Installer Debugging ===================
A bug was filed about headless installs:
https://bugzilla.redhat.com/487136 KVM Having problem In fedora 10 With no-gui
This prompted a new tip to be added to:
https://fedoraproject.org/wiki/Reporting_virtualization_bugs#virt-install
In order to gain access to a serial console during the install, you can use -x "console=ttyS0". Using a serial console combined with a VNC install can be very useful for debugging e.g. <code>--nographics -x "console=ttyS0 vnc".
Perhaps it might make sense for Fedora to use any available serial console if there is no graphics card? Probably not, though, since you have no idea whether anything is connected to the other end of the serial line - on a bare-metal headless server, such a policy would have us pointlessly driving unused serial lines.
kvm-84 ======
The recently released kvm-84 has hit rawhide. Avi's release announcement is here:
http://marc.info/?l=kvm&m=123465878918472
Tom London reports that qemu-kvm segaults for him when using the SDL graphics frontend. It appears the issue may be related to SDL's use of SSE/MMX instructions for bit blitting.
https://bugzilla.redhat.com/487018 Segfault with kvm-84-1.fc11.x86_64.rpm (SDL related)
libvirt-0.6.0 =============
libvirt-0.6.0-3.fc10 was pushed to f10-updates-testing this week.
Another issue with libvirt-0.6.0 was reported:
https://bugzilla.redhat.com/487013 virsh setvcpus hangs with libvirt-0.6.0
This turns out to be a recursive locking bug introduced by the major threading changes in the release.
Busted Installs ===============
Jeff Garzik reports that he sees x86_64 F10 installs hangs at "installing bootloader" under KVM:
https://bugzilla.redhat.com/474116
This problem went away when he used 1Gb RAM instead of 512Mb. There is already another similar, but not identical bug, detailing an issue with x86_64 installs on 512Mb:
https://bugzilla.redhat.com/480826
Fabian Deutsch also reports some F11 alpha installer issues under KVM. The first looks like a generic anaconda bug that also existed in F10:
https://bugzilla.redhat.com/476433
The second seems to be an anaconda hang during a LiveCD install:
https://bugzilla.redhat.com/484721
Neither of Fabian's issues look related to known pvmmu related crash in F11 alpha:
https://bugzilla.redhat.com/480822
However, Mark Bidewell reported an issue with F11 Alpha installs to kvm@vger and this did turn out to likely be pvmmu related:
http://marc.info/?l=kvm&m=123548745416389
Also, James Laska previously reported a kernel oops during install that may well be yet another pvmmu related issue:
https://bugzilla.redhat.com/483648
Marcelo's pvmmu fix first made it in 2.6.26-rc5.git, so hopefully the world has been a better place since then.
Rawhide Rebuild Failures ========================
Matt Domsch rebuilt all of rawhide in mock and found that etherboot, ruby-libvirt, gnome-applet-vm and xen failed to build:
http://www.redhat.com/archives/fedora-virt/2009-February/msg00079.html http://www.redhat.com/archives/fedora-virt/2009-February/msg00080.html http://www.redhat.com/archives/fedora-virt/2009-February/msg00081.html http://www.redhat.com/archives/fedora-virt/2009-February/msg00082.html
Xen dom0 ========
Michael Young completed the first dom0 build in Koji:
http://www.redhat.com/archives/fedora-virt/2009-February/msg00085.html
These are really for development and debugging purposes only, as I am still having problems getting them to boot, but others have reported more success at getting kernels based on this git repository working, so you might be lucky.
Michael found an interesting issue where recently built F11 RPMs wouldn't install on F10 because F11 RPMs use SHA-256 checksums. F10 updates now contains a version of RPM which can handle these checksums.
https://bugzilla.redhat.com/487140 Packages built with rpm-4.6.0-8.fc11 won't install on Fedora 10 x86_64
Xen DomU ========
Some discussion on fedora-kernel-list about the need to fix anaconda to install kernel-PAE on any machine which supports PAE:
http://www.redhat.com/archives/fedora-kernel-list/2009-February/msg00079.htm...
Michael Young also reported a lockdep warning during DomU boot which Jeremy has fixed in his Dom0 tree:
https://bugzilla.redhat.com/487324 kernel/lockdep.c warning in xen guest boot dmesg
Bugs ====
DOOM-O-METER: 188 open bugs last week, 192 this week. Baa!
https://bugzilla.redhat.com/487190 virt-manager has hardcoded Requires: PolicyKit-gnome
Does virt-manager really require PolicyKit-gnome? The KDE spin doesn't ship it.
Looks like PolicyKit was fixed in 0.7 to allow us to only require PolicyKit itself.
https://bugzilla.redhat.com/467687 Virtual network won't work until restart
It appears that ip_forward was being set to zero because the NetworkManager and libvirtd services were incorrectly ordered.
The "reporting virt bugs" wiki page has some details on how to diagnose the problem in future:
https://fedoraproject.org/wiki/Reporting_virtualization_bugs#Networking
On Wed, Feb 25, 2009 at 09:15:26PM +0000, Mark McLoughlin wrote:
Installer Debugging
A bug was filed about headless installs:
https://bugzilla.redhat.com/487136 KVM Having problem In fedora 10 With no-gui
This prompted a new tip to be added to:
https://fedoraproject.org/wiki/Reporting_virtualization_bugs#virt-install
In order to gain access to a serial console during the install, you can use -x "console=ttyS0". Using a serial console combined with a VNC install can be very useful for debugging e.g. <code>--nographics -x "console=ttyS0 vnc".
Perhaps it might make sense for Fedora to use any available serial console if there is no graphics card? Probably not, though, since you have no idea whether anything is connected to the other end of the serial line - on a bare-metal headless server, such a policy would have us pointlessly driving unused serial lines.
Latest KVM has a new 'virtio console' PCI device present. We just need to hook this upto libirt's <console> XML, to get an always available PV console for all KVM guests without messing around with serial ports. Just need to check Fedora guests have a config /etc/event/hvc0 so the agetty is started on it out of the box.
Daniel