Re: Panic in 9.0_STABLE with ZFS module on DOMU PV with recent build

[Jaromír Doleček <jaromir.dolecek%gmail.com@localhost>]

OK, it seems what happens is that xbd_diskstart() routine gets a
request for starting I/O bigger than it can handle. The NetBSD 9.0
xbd(4) has a limit of 44 KB maximum I/O, and XEN3_DOMU has specific
override for MAXPHYS to 32 KB for this reason.

It seems the ZFS module does not take this into account, and is
compiled with default 64 KB MAXPHYS.

On -current xbd(4) was changed to remove this limitation, and works
with default 64KB MAXPHYS. One option is then to just upgrade the
kernel to -current. You can leave userland on 9.0. If that is not
possible for you, you can build your own custom ZFS module with an
override of MAXPHYS=32768, using still netbsd-9 sources.

Jaromir

Le mer. 9 sept. 2020 à 08:17, Staffan Thomén <duck%shangtai.net@localhost> a écrit :
>
> Brad Spencer wrote:
> > I have been running a PV with NetBSD 9.0_STABLE and ZFS filesets for
> > while with pretty good success.  I went to update the system with a
> > build as of 20200820 and the ZFS module panics with the included panic
> > as soon as the ZFS module loads (and maybe starts to do something, don't
> > know about that part).  The ZFS module appears to work from 20200509
> > with the new kernel.  Anyone else have this??  The Xen version is older,
> > a DOM0 from October 2018, 8.99.25 and Xen 4.8.
>
> I saw this very crash yesterday when trying to replicate a hang that I found
> with zvols on the dom0 (testing on the dom0 is a bit annoying...).
>
> It seems to happen when zfs imports the pool (tries the scan the xbd:s?)
> because I could load the module and create a pool on a fresh domU, but when I
> rebooted the system it would crash. I am running Xen 4.11.3nb2 on 9.0_STABLE,
> sources updated on September 7th or 8th.
>
> When I booted the system with zfs=NO in rc.conf but zfs in /etc/modules.conf
> the system booted properly, but instantly crashed with the following backtrace
> when I ran zpool status:
>
> test# zpool status
> [  28.5500728] panic: kernel diagnostic assertion "seg <=
> BLKIF_MAX_SEGMENTS_PER_REQUEST" failed: file
> "/usr/build/src/sys/arch/xen/xen/xbd_xenbus.c", line 1032
> [  28.5500728] cpu0: Begin traceback...
> [  28.5500728] vpanic() at netbsd:vpanic+0x143
> [  28.5500728] kern_assert() at netbsd:kern_assert+0x48
> [  28.5500728] xbd_diskstart() at netbsd:xbd_diskstart+0x3c7
> [  28.5500728] dk_start() at netbsd:dk_start+0x104