RE: [Xen-devel] [PATCH] non-flat protected mode HVM support

["Petersson, Mats" <Mats.Petersson@xxxxxxx>]

> -----Original Message-----
> From: xen-devel-bounces@xxxxxxxxxxxxxxxxxxx 
> [mailto:xen-devel-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of 
> Jan Beulich
> Sent: 29 November 2006 15:19
> To: Trolle Selander
> Cc: xen-devel@xxxxxxxxxxxxxxxxxxx
> Subject: Re: [Xen-devel] [PATCH] non-flat protected mode HVM support
> 
> >>> "Trolle Selander" <trolle.selander@xxxxxxxxx> 29.11.06 15:45 >>>
> >The assumption of zero base/limit still exists in insn_fetch in
> >x86_emulate.c. When an EXCEPTION_PF occurs and cs.base is non-zero,
> >x86_emulate_memop fails because it does not take cs.base 
> into account when
> >reading the instruction to be decoded/emulated from memory. 
> I had an email
> >discussion with Mats early last week about this and he made some
> >suggestions, but I didn't have a chance to really look at it until
> >yesterday, and by then you had done a lot of work on the zero segment
> >base/limit assumptions anyway. Looking at it now, I'm not 
> wholly sure what
> >would be the best way to pass this info to 
> x86_emulate_memop. Maybe extend
> >the struct x86_emulate_ctxt to include cs base (and 
> limits/attibutes)? Or
> >would that cause problems elsewhere?
> 
> Indeed, I didn't look outside the hvm directories. But it's 
> not just the instruction
> fetching there, there are other segment based operations in 
> there, too. Hence
> I think adding a callback to obtain the base address would be cleaner;
> hvm_get_segment_base() could be used in the hvm case, and a 
> simple function
> should be put in place for pv guest handling (for those, 
> assuming a zero base
> address seems reasonably safe at this point, since the 
> behavior of them is
> known, and if ever page table updates get done through 
> segments with non-
> zero base addresses this would need to be revisited.

There should be a per-domain function pointer to get the segment-data -
we can either allow this function pointer to be NULL (for non HVM-case),
in which case the base is assumed to be zero. Or implement a function
that just returns zero for non HVM domains. I'm thinking something like
this (written as a function for ease of explaining, rather than
efficiency - a macro may be better...)

unsigned long get_seg_base(int segno)
{
        if (current->get_segment_base)
                return current->get_segment_base;
        else
                return 0;
}

The if-statement should be leaner than calling a function that returns
zero... 

--
Mats


> 
> I'll try to take care of this unless you beat me.
> 
> Jan
> 
> 
> _______________________________________________
> 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