[Bug 857662] Re: Should xenbus_probe_frontend be built-in?
Colin Watson
cjwatson at canonical.com
Fri Sep 23 22:42:43 UTC 2011
** Description changed:
+ Stable update justification:
+
+ Impact: When installing in a Xen domU, the installer fails to find any network interfaces.
+ Development branch: Fixed in hw-detect 1.81ubuntu3 by loading xenbus_probe_frontend if we seem to be running under the Xen hypervisor. (It should not be catastrophic if we load this even under other conditions, but it should be a bit more efficient this way.)
+ Patch: http://bazaar.launchpad.net/~ubuntu-core-dev/hw-detect/natty-proposed/revision/1466
+ TEST CASE: Start up d-i in a Xen domU and run through to the point where it detects network interfaces. Without this patch, it should fail; with this patch, it should work.
+ Regression potential: I think the worst thing that can plausibly happen here is an error message via debconf if the module fails to load. The /sys/hypervisor/type and /sys/bus/xen checks around this code are mainly to defend against this possibility in non-Xen installations.
+
+ Original report follows:
+
xenbus_probe_frontend is currently modular in several flavours, notably
in the i386/generic-pae and amd64/generic flavours which we use to build
d-i Xen netboot images. The purpose of xenbus_probe_frontend appears to
be to connect to the Xen bus and then emit uevents for any devices it
finds there. Once it's loaded, autoloading for such things as xen-
netfront works fine, allowing the guest to find its network interface.
I believe that xen-blkfront works in a similar way.
However, with xenbus_probe_frontend modular, the early stages of the
installer plus the early stages of normal system boot need to load
xenbus_probe_frontend in order to find block and network devices. If
we're going to do this, it seems as though we might as well just build
xenbus_probe_frontend into the kernel. It seems to me that it should be
pretty harmless on non-Xen systems, although of course somebody
knowledgeable should verify that.
Right now, it's built in for the -virtual flavours, but we aren't
currently building installer images from that, and I think I'm happier
building them directly from -generic-pae / -generic anyway.
This is true in both Natty and Oneiric. If you can't do this for
Oneiric and/or are unwilling to do this in a stable update for Natty,
then please let me know so that I can add a workaround to the installer
in time. Thanks!
** Description changed:
Stable update justification:
Impact: When installing in a Xen domU, the installer fails to find any network interfaces.
Development branch: Fixed in hw-detect 1.81ubuntu3 by loading xenbus_probe_frontend if we seem to be running under the Xen hypervisor. (It should not be catastrophic if we load this even under other conditions, but it should be a bit more efficient this way.)
Patch: http://bazaar.launchpad.net/~ubuntu-core-dev/hw-detect/natty-proposed/revision/1466
- TEST CASE: Start up d-i in a Xen domU and run through to the point where it detects network interfaces. Without this patch, it should fail; with this patch, it should work.
+ TEST CASE: Start up d-i in a Xen domU and run through to the point where it detects network interfaces. Without this patch, it should fail; with this patch, it should work. Note that for testing you will need to use a kernel/initrd from natty-proposed (which I'll need to upload separately after this patch is accepted), and in production after this upload is validated you'll need to use a kernel/initrd from natty-updates.
Regression potential: I think the worst thing that can plausibly happen here is an error message via debconf if the module fails to load. The /sys/hypervisor/type and /sys/bus/xen checks around this code are mainly to defend against this possibility in non-Xen installations.
Original report follows:
xenbus_probe_frontend is currently modular in several flavours, notably
in the i386/generic-pae and amd64/generic flavours which we use to build
d-i Xen netboot images. The purpose of xenbus_probe_frontend appears to
be to connect to the Xen bus and then emit uevents for any devices it
finds there. Once it's loaded, autoloading for such things as xen-
netfront works fine, allowing the guest to find its network interface.
I believe that xen-blkfront works in a similar way.
However, with xenbus_probe_frontend modular, the early stages of the
installer plus the early stages of normal system boot need to load
xenbus_probe_frontend in order to find block and network devices. If
we're going to do this, it seems as though we might as well just build
xenbus_probe_frontend into the kernel. It seems to me that it should be
pretty harmless on non-Xen systems, although of course somebody
knowledgeable should verify that.
Right now, it's built in for the -virtual flavours, but we aren't
currently building installer images from that, and I think I'm happier
building them directly from -generic-pae / -generic anyway.
This is true in both Natty and Oneiric. If you can't do this for
Oneiric and/or are unwilling to do this in a stable update for Natty,
then please let me know so that I can add a workaround to the installer
in time. Thanks!
--
You received this bug notification because you are a member of Ubuntu
Foundations Bugs, which is subscribed to debian-installer in Ubuntu.
https://bugs.launchpad.net/bugs/857662
Title:
Should xenbus_probe_frontend be built-in?
Status in “debian-installer” package in Ubuntu:
Triaged
Status in “hw-detect” package in Ubuntu:
Triaged
Status in “linux” package in Ubuntu:
Invalid
Status in “debian-installer” source package in Natty:
Won't Fix
Status in “hw-detect” source package in Natty:
Triaged
Status in “linux” source package in Natty:
Won't Fix
Status in “debian-installer” source package in Oneiric:
Won't Fix
Status in “hw-detect” source package in Oneiric:
Triaged
Status in “linux” source package in Oneiric:
Won't Fix
Status in “debian-installer” source package in p-series:
Triaged
Status in “hw-detect” source package in p-series:
Invalid
Status in “linux” source package in p-series:
Invalid
Bug description:
Stable update justification:
Impact: When installing in a Xen domU, the installer fails to find any network interfaces.
Development branch: Fixed in hw-detect 1.81ubuntu3 by loading xenbus_probe_frontend if we seem to be running under the Xen hypervisor. (It should not be catastrophic if we load this even under other conditions, but it should be a bit more efficient this way.)
Patch: http://bazaar.launchpad.net/~ubuntu-core-dev/hw-detect/natty-proposed/revision/1466
TEST CASE: Start up d-i in a Xen domU and run through to the point where it detects network interfaces. Without this patch, it should fail; with this patch, it should work. Note that for testing you will need to use a kernel/initrd from natty-proposed (which I'll need to upload separately after this patch is accepted), and in production after this upload is validated you'll need to use a kernel/initrd from natty-updates.
Regression potential: I think the worst thing that can plausibly happen here is an error message via debconf if the module fails to load. The /sys/hypervisor/type and /sys/bus/xen checks around this code are mainly to defend against this possibility in non-Xen installations.
Original report follows:
xenbus_probe_frontend is currently modular in several flavours,
notably in the i386/generic-pae and amd64/generic flavours which we
use to build d-i Xen netboot images. The purpose of
xenbus_probe_frontend appears to be to connect to the Xen bus and then
emit uevents for any devices it finds there. Once it's loaded,
autoloading for such things as xen-netfront works fine, allowing the
guest to find its network interface. I believe that xen-blkfront
works in a similar way.
However, with xenbus_probe_frontend modular, the early stages of the
installer plus the early stages of normal system boot need to load
xenbus_probe_frontend in order to find block and network devices. If
we're going to do this, it seems as though we might as well just build
xenbus_probe_frontend into the kernel. It seems to me that it should
be pretty harmless on non-Xen systems, although of course somebody
knowledgeable should verify that.
Right now, it's built in for the -virtual flavours, but we aren't
currently building installer images from that, and I think I'm happier
building them directly from -generic-pae / -generic anyway.
This is true in both Natty and Oneiric. If you can't do this for
Oneiric and/or are unwilling to do this in a stable update for Natty,
then please let me know so that I can add a workaround to the
installer in time. Thanks!
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/debian-installer/+bug/857662/+subscriptions
More information about the foundations-bugs
mailing list