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

Re: [Xen-devel] I cannot get any message from domU by console / pv_ops domU kernel crashes with xen_create_contiguous_region failed



xen-devel-bounces@xxxxxxxxxxxxxxxxxxx wrote on 12/22/2009 08:04:45 AM:

> xen-devel-bounces@xxxxxxxxxxxxxxxxxxx wrote on 12/22/2009 04:20:15 AM:
> 
> > On Tue, 2009-12-22 at 08:42 +0000, Pasi Kärkkäinen wrote:
> > > >    (early) [    0.000000] PCI-DMA: Using Xen software bounce
> > > buffering for IO
> > > >    (Xen-SWIOTLB)
> > > >    (early) [    0.000000] Kernel panic - not syncing:
> > > >    <3>xen_create_contiguous_region failed
> > > > 
> > > 
> > > Did you try cgroup_disable=memory kernel option? 
> > 
> > I think this is more likely to relate to the failure to initialise
> > swiotlb due to the failure of xen_create_contiguous_region, leading to 
a
> > panic. IIRC the guest won't be privileged enough to make this call
> > unless it has PCI passthrough enabled.
> > 
> > In a pre-pvops kernel the swiotlb would be inactive in a domU unless
> > specifically requested on the command line but in Jeremy's xen/master
> > branch it looks like it is unconditionally enabled for dom0 and domU 
as
> > a consequence of:
> >         commit 4bc6b1a9dd5d7447af8c3d27c1449f73f5f764ec
> >         Author: root <root@xxxxxxxxxxxxxxxxxxxxx>
> >         Date:   Thu Nov 5 16:33:10 2009 -0500
> > 
> >             Enable Xen-SWIOTLB if running in [non-]privileged and 
> > disable the Xen-IOMMU if an IOMMU is detected.
> > 
> > On a related note I don't think the xen-swiotlb code should panic() if
> > it cannot allocate the contiguous region (at least in domU). Chances 
are
> > if it hasn't got privilege to make the fixups it isn't going to have 
any
> > devices.
> 
> I ran into the same problem yesterday, traced it back to that exact 
> commit.  Reverting it makes my domUs work, but feels like the wrong 
thing 
> to do.  How do I make sure xen_create_contiguous_region is privileged 
> enough?  I've got the PCI frontend and backend  built in to my kernel 
> (used the x86 fedora .config from the pv_ops wiki).

When you said to enable PCI passthrough, did you mean by setting 
CONFIG_XEN_PCIDEV_BACKEND_PASS as apposed to the newer 
CONFIG_XEN_PCIDEV_BACKEND_VPCI?  Is Passthrough going to replace VPCI as 
the preferred PCI Backend Mode for Xen 4.0?

-Mike


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


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.