RE: [Xen-devel] USB disk corruption on Xen 4.1.0 & Linux 2.6.38.7

[Ian Pratt <Ian.Pratt@xxxxxxxxxxxxx>]

> I hit it every time, though with slight variations on which blocks are 
> affected. I
> get the problem on two out of three of my machines. I don't supply any
> command line options to Xen or the kernel. I do use an initramfs.

If you're seeing the issue I think you are, you'll be vulnerable to it whenever 
any of dom0's memory is allocated above 4GB. 

If you patch xen to always allocate dom0's memory from below 4GB I bet you 
don't see the issue (though obviously you'll be limited to having a dom0 that 
is a bit smaller than 4GB). The real bug is probably in the EHCI driver's use 
of swiotlb.


Ian

> ----- Original Message -----
> From: "Konrad Rzeszutek Wilk" <konrad.wilk@xxxxxxxxxx>
> To: "Anthony Wright" <anthony@xxxxxxxxxxxxxxx>
> Cc: "Ian Pratt" <Ian.Pratt@xxxxxxxxxxxxx>, xen-devel@xxxxxxxxxxxxxxxxxxx
> Sent: Monday, 13 June, 2011 11:06:25 PM
> Subject: Re: [Xen-devel] USB disk corruption on Xen 4.1.0 & Linux 2.6.38.7
> 
> On Fri, Jun 10, 2011 at 12:46:56PM -0400, Konrad Rzeszutek Wilk wrote:
> > On Thu, Jun 09, 2011 at 09:32:08AM -0400, Konrad Rzeszutek Wilk wrote:
> > > On Thu, Jun 09, 2011 at 02:11:27PM +0100, Anthony Wright wrote:
> > > > I've got things running on a number of systems now, and have tried a
> > > > number of combinations. I have changed machines and USB memory
> sticks,
> > > > and the fault seems to be related to the machine rather than the
> memory
> > > > stick. I have 3 machines with >4GB ram & 2 memory sticks, changing
> > > > memory sticks doesn't have any effect, but only two out of the three
> > > > machines I tried has a problem, the third works fine.
> > >
> > > One of them was an MCP61 based (BIOSTAR) and the other is
> > > SB600 (Gigabyte gs_ma69vm_s2) Southbridge. Both are AMD based ...
> and
> > > I vagually remember something about quirks for those boards for the
> AHCI
> > > controller.
> > >
> > > Let me see if I can reproduce this issue on my box next week.
> >
> > I tried on my BIOSTAR Group N61PB-M2S which has 4GB physically
> > and couldn't reproduce it. But let me try to put in some more memory and
> see.
> 
> So, playing around on a 8gB machine with ballooning memory up (I had
> initially set
> only 2GB to dom0) and then using the md5sum I was able to trigger this once.
> 
> You said you are hitting this everytime right? Do you limit the amount of
> memory
> to the initial domain?
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@xxxxxxxxxxxxxxxxxxx
> http://lists.xensource.com/xen-devel

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