Port-xen archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: Ubuntu 16.04 (xenial) partial success with NetBSD and Xen



Nice investigations :) I have a couple of quick data points which may
be of some interest (below).

On 1 January 2017 at 05:30, Aaron J. Grier <agrier%poofygoof.com@localhost> wrote:
> with HVM, I can successfully boot Ubuntu 16.04 (both install and update
> from 14.04) under NetBSD-7 Dom0 with both Xen 4.2.3 and 4.6.3.
>
> the Linux kernel option xen_emul_unplug=unnecessary (or =never) is what
> gets things nominally working.  this yields a booting system with
> emulated disk and PV network, with hypercalls used for Xen services
> (clocks, timers, etc).  not quite full PVHVM, but not completely HVM
> either.
>
> part of the problem seems to be the "unplug" protocol for emulated
> devices which was implemented eight years ago [1] in QEMU, and doesn't
> seem to be fully supported in the version shipped with NetBSD's Xen
> packages.  (both xentools42 and xentools46 use QEMU 0.10.2.)  it's
> supposed to unplug the emulated devices, leaving only the native Xen
> block devices.
>
> on the PV front, between Xen 4.2.3 and 4.6.3, pygrub stopped working,
> possibly due to libxl's changing the handling of POLLIN and POLLHUP [2].
> I didn't pursue tracking this down further, since getting pvgrub seemed
> like a better solution.  (broken NFS client for NetBSD-7 slows progress,
> too.)

Manuel fixed the signal issue in xen 4.6 on the 20th of December
(xentools46nb4) - after this I found pygrub to work fine under 4.6.

> PV Linux kernels boot (with pvgrub), but the ramdisk can't find the Xen
> block root device.  updates to blkfront drivers somewhere between kernel
> 3.13.0 (ubuntu 14.04) and 4.4.0 (ubuntu 16.04) seem to prevent newer
> kernels from attaching the Xen block devices.  I haven't tried to boot
> anything other than Linux with pvgrub yet.

Interestingly a 4.6 kernel dropped into a Ubuntu 16.04 image seems to
run fine under xen 4.5 and 4.6, so later Linux kernels seem to handle
this case gracefully.


Home | Main Index | Thread Index | Old Index