Re: GPU passthrough performance regression in >4GB vms due to XSA-60 changes

["Jan Beulich" <JBeulich@suse.com>]

>>> On 19.05.14 at 13:09, <tomasz.wroblewski@gmail.com> wrote:

> On 05/19/2014 01:01 PM, Jan Beulich wrote:
>>>>> On 19.05.14 at 12:42, <tomasz.wroblewski@gmail.com> wrote:
>>> Actually seems to be like the plausible explanation for the performance
>>> issues we see could be that
>>>
>>> - some region of guest space has been relocated by hvmloader out to
>>> 64bit memory to enlarge pci mmio hole (which stays in 32bit space)
>>> - BUT the caching on that relocated region is UC since at the time of
>>> the relocation MTRR was disabled and that caused the EPT entry to get UC
>>> type.
>> But this would be easily seen from 'D' debug key output - did you
>> check this is what is the case?
> Yes I've checked that, though I used some homebrew EPT dumping code 
> instead. Mentioned it a bit earlier, in the original post as well as 
> http://lists.xen.org/archives/html/xen-devel/2014-05/msg01965.html. I 
> just incorrectly thought that it's the BAR being relocated which 
> introduced some confusion..

Ah, okay. So then the question is how to address this. Which means
it now becomes really important to be certain that this is already
fixed on -unstable, so we can evaluate whether we want to backport
some non-trivial changes, or find a workaround just for the stable
trees (which, as said before, I can't easily see how it would look like).

Jan