Re: [Xen-devel] Re: [b.g.o.358549] blkfront: Move blkif_interrupt into a tasklet.

[Daniel Stodden <daniel.stodden@xxxxxxxxxx>]

On Thu, 2011-05-05 at 18:57 -0400, Marcin MirosÅaw wrote:
> W dniu 2011-05-05 23:22, Konrad Rzeszutek Wilk pisze:
> > On Wed, May 04, 2011 at 04:08:08PM +0200, Marcin MirosÅaw wrote:
> >> Hello!
> >> I'm hitted by issue described there:
> >> http://lists.xensource.com/archives/html/xen-devel/2010-09/msg00164.html
> >> . I'm using 32bit guest (no SMP) on 64 xen-server. On hardened-kernel
> >> (grsec)-2.6.{35,36} there is silence in dmesg (i don't remember what was
> >> going on on earlier kernels). On hardened-kernel-2.6.{37,38} and on
> >> vanilla-kernel-2.6.{37,38,39} and git-sources-2.6.39_rc5-r7 problem exists.
> > 
> > Do you always get this when you try to mount a filesystem with
> > barriers? And do you see this _only_ with 32-bit guests, but not with 
> > 64-bit ones?
> 
> I'm not admin of Dom0, i've got only one vm (with PV enabled), which is
> 32 bit. I didn't try to start them with 64 bit kernel (and 32 bit
> userland), do you think it's worth to try it?
> I don't force barrier in fstab, on kernel hardened-2.6.36 (which doesn't
> throws warnings), /proc/mounts shows barrier=1, there is in dmesg:
> [...]
> [    0.209834] Initialising Xen virtual ethernet driver.
> [    0.217792] blkfront: xvda: barriers enabled (tag)
> [...]
> 
> On vanilla-2.6.39-rc5-git7 i've got in log:
> [    6.505772] blkfront: xvda: empty write barrier op failed
> [    6.505776] blkfront: xvda: barriers disabled
> 
> (I hope i didn't omit important part from dmesg, i didn't want to paste
> tons of line)

Iirc this stuff ultimately lead us to an irq migration bug. Are you
running irqbalance? If so, can you turn it off and see if that helps?

Daniel



_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel