Re: PV hybrid: vmlaunch failure help...

[Mukesh Rathor <mukesh.rathor@ORACLE.COM>]

On Fri, 04 Mar 2011 07:42:07 +0000
Keir Fraser <keir.xen@gmail.com> wrote:

> On 04/03/2011 03:21, "Mukesh Rathor" <mukesh.rathor@oracle.com> wrote:
> 
> > Hi all,
> I was going to look into this feature myself, it's good someone else
> has picked it up so I don't have to. :-)
> 
> For this particular vmentry failure, Vol.3B Section 23.3.1 lists all
> the guest-state checks carried out on vmentry. There are quite a few
> of them, but you simply need to check your state dump against every
> check until you find a mismatch. I believe someone had a function
> that implemented all/most of these checks in a diagnostic helper
> function -- I can't find the patch just now unfortunately.
> 
>  -- Keir

Yey, success finally! I am able to enter the vmx to the first
instruction after seeing hundreds for "guest state invalid" for
sometime.  It appears that attrs must be set for all descriptors. That
seems in conflict with somewhere in manual that null selectors cause
descriptors to be ignored.  I wish the vendors would provide flow chart
of exact microcode execution for 'vmlaunch'. It is an assembler
instruction after all, and the programmer deserves to know everything
it's doing and the order it's doing in :)... 

Anyways, I hope to make good progress on it now, and if I can get my
hacked up version at least stable enough, will run some benchmarks. I'm
on vacation next month, but I hope to have something to report to you guys
middle/end of May.

thanks,
Mukesh