Re: HVMLOADER test fails on AMD if CR3 write intercepts are enabled (XEN 4.4 stable)

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

On Sun, Jul 27, 2014 at 07:03:35PM -0600, Abhiroop Dabral wrote:
> Hi,

Hey,

Please use the MAINTAINERs file in the future to make sure that
the maintainers of the AMD SVM code can see this email. I've
done that for you.

> 
> The HVMLOADER test fails on AMD SVM with Nested Paging enabled **Only**
> when CR3 control registers write intercepts are enabled.
> (By default the CR3 write intercepts are disabled on HAP domains)

How did you enable the cr3 interception? Was there an configuration
flag or was this something you modified in Xen?

> 
> The first test "rep_io" fails with the following output:
> 
> Bad value at 0x00500000: saw 98765432 expected 987654ff
> 
> Bad value at 0x00500ffc: saw 00000000 expected ff000000
> 
> Bad value at 0x00600000: saw 000000ff expected 00000000
> 
> Bad value at 0x00600ffc: saw ff000000 expected 00000000
> 
> The second test execution results in the triple fault.
> 
> The paging is setup as writing to va 0x600000 a value of 98765432 works
> fine. (Since the test can see the same value at physical address 0x500000).
> 
> But after that the insb mov to address are not writing the memory correctly.
> 
> The only change made in the code is to enable CR3 intercepts on a similar
> lines as done in VMX counterpart.
> 
> Seen in:
> 
> Xen 4.4 stable, AMD SVM with HAP
> 
> Can any one suggest what can be the issue behind this.
> 
> thanks.

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