Re: Introducing paravirt_ops for x86_64

Glauber de Oliveira Costa wrote:

On 8/8/07, Nakajima, Jun [off-list ref] wrote:
  

quoted

Glauber de Oliveira Costa wrote:
    

quoted

Hi folks,

After some time away from it, and a big rebase as a consequence, here
      

is
    

quoted

the updated version of paravirt_ops for x86_64, heading to inclusion.

Your criticism is of course, very welcome.

Have fun
      

Do you assume that the kernel ougtht to use 2MB pages for its mappings
(e.g. initilal text/data,  direct mapping of physical memory) under your
paravirt_ops?  As far as I look at the patches, I don't find one.
    

I don't think how it could be relevant here. lguest kernel does use
2MB pages, and it goes smootly. For 2MB pages, we will update the page
tables in the very same way, and in the very places we did before.
Just that the operations can now be overwritten.

So, unless I'm very wrong,  it only makes sense to talk about not
supporting large pages in the guest level. But it is not a
paravirt_ops problem.
  

At the moment Xen can't support guests with 2M pages.  In 32-bit this
isn't a huge problem, since the kernel doesn't assume it can map itself
with 2M pages.  But I think the 64-bit kernel assumes 2M pages are
always available for mapping the kernel.  I don't know how pervasive
this assumption is, but it would be nice to parameterize it in pv-ops.


    J