Re: [Xen-devel] [xen-4.2-testing test] 19031: regressions - FAIL

[Konrad Rzeszutek Wilk <konrad.wilk@xxxxxxxxxx>]

On Wed, Sep 04, 2013 at 04:11:15PM +0100, Ian Jackson wrote:
> Konrad Rzeszutek Wilk writes ("Re: [Xen-devel] [xen-4.2-testing test] 19031: 
> regressions - FAIL"):
> > It is interesting that it fails at the same problem the linux-real had
> > been failing. That is when doing rsync.
> > 
> > 
> > 5120 bytes (5.1 kB) copied, 0.0194138 s, 264 kB/s
> > 2013-09-04 04:42:13 Z execution took 19 seconds [**>30/2**]: ssh -o 
> > StrictHostKeyChecking=no -o BatchMode=yes -o ConnectTimeout=100 -o 
> > ServerAliveInterval=100 -o PasswordAuthentication=no -o 
> > ChallengeResponseAuthentication=no -o 
> > UserKnownHostsFile=tmp/t.known_hosts_19031.test-amd64-i386-xend-qemut-winxpsp3
> >  root@xxxxxxxxxxxxx dd if=/dev/zero 
> > of=/dev/woodlouse/win.guest.osstest-disk count=10
> > 2013-09-04 04:42:13 Z executing rsync ... -p 
> > /home/xc_osstest/images/winxpsp3.iso 
> > root@xxxxxxxxxxxxx:/root/19031.test-amd64-i386-xend-qemut-winxpsp3.winxpsp3.iso
> 
> This is obviously badly broken.  It's copying a tiny known_hosts file,
> and it took 19 seconds.
> 
> > Ian, is it possible to a bit more verbose - perhaps that will shed some 
> > light
> > at this awful speed.
> 
> I could add a -v but I'm not sure it will be much more illuminating.
> 
> > There does not seem to be anything about the 'tg3' blowing up?
> > Ian, would it also be able to get the interrupt count _Before_ and
> > _after_ the timeout? Perhaps the interrupts aren't going to the
> > device?
> 
> Surely that wouldn't produce these symptoms...

Does this happen _only_ with 32-bit dom0? The theory about /proc/interrupts
is that the tg3 is not using the DMA API properly and the device driver
has submitted commands. The hardware sits there and occassionaly polls
instead of going over the commands quickly. That would explain the
slowness. But that won't explain why it works with 64-bit dom0s unless
there is some truncation of physical addresses.

Perhaps you could build with CONFIG_DEBUG_DMA_API=? Perhaps that will shed
some light on it?


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel