Xen devel - May 2006 - RE: RE: [Xen-changelog] Fix MOVS instruction emulation for HVM MMIO.

> -----Original Message-----
> From: xen-devel-bounces@lists.xensource.com 
> [mailto:xen-devel-bounces@lists.xensource.com] On Behalf Of 
> Gerd Hoffmann
> Sent: 17 May 2006 07:43
> To: Petersson, Mats
> Cc: xen-devel@lists.xensource.com; xen-changelog@lists.xensource.com
> Subject: Re: [Xen-devel] RE: [Xen-changelog] Fix MOVS 
> instruction emulation for HVM MMIO.
> 
> Petersson, Mats wrote:
> >> Subject: [Xen-changelog] Fix MOVS instruction emulation 
> for HVM MMIO.
> >>
> >> diff -r aab3cd33d2ba -r 7fdc4a8b782b xen/arch/x86/hvm/platform.c
> >> --- a/xen/arch/x86/hvm/platform.c	Tue May 16 16:34:27 2006 +0100
> >> +++ b/xen/arch/x86/hvm/platform.c	Tue May 16 19:50:23 2006 +0100
> >> @@ -865,7 +865,7 @@ void handle_mmio(unsigned long va, unsig
> >>           * copy ourself. After this copy succeeds, "rep
movs" is
> >> executed
> >>           * again.
> >>           */
> >> -        if ((addr & PAGE_MASK) != ((addr + size - 1) & 
> PAGE_MASK)) {
> >> +        if ((addr & PAGE_MASK) != ((addr + sign * (size - 1))
&
> >> + PAGE_MASK)) {
> > 
> > With the risk of being wrong (again), I''d say this is 
> incorrect: The 
> > MOVS instruction will start reading at ESI, and write at 
> the address 
> > indicated by EDI and write with size bytes, even when it''s
copying
> > backwards. So there should be no multiplication of sign on 
> this line.
> 
> I still think this is correct.  If I understand things 
> correctly the point of the test is to figure whenever the 
> _next_ repz movs interation will access another page (and if 
> so copy just one data word and let the emulator kick in again 
> for the remaining data on the page above/below).
No, the test is for if the first (current) operation is crossing a page
boundary, not if the NEXT one is... That''s why the code in this
condition is using hvm_copy to fetch the data that is being accessed...

But leave it as is right now, I''m still working on getting a piece of
test-code to do MOVS in various ways...

--
Mats
> 
> cheers,
> 
>   Gerd
> 
> --
> Gerd Hoffmann <kraxel@suse.de>
> Erst mal heiraten, ein, zwei Kinder, und wenn alles läuft 
> geh'' ich nach drei Jahren mit der Familie an die Börse.
> http://www.suse.de/~kraxel/julika-dora.jpeg
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
> 
> 

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Hi,
> No, the test is for if the first (current) operation is crossing a
> page boundary, not if the NEXT one is... That''s why the code in
this
> condition is using hvm_copy to fetch the data that is being
> accessed...
Ok, point taken.  But that also means that the access range calculation
isn''t that simple, it''s quite different for the two directions
...

Patch against 9681 (testing repository) attached.

cheers,

  Gerd

-- 
Gerd Hoffmann <kraxel@suse.de>
Erst mal heiraten, ein, zwei Kinder, und wenn alles läuft
geh'' ich nach drei Jahren mit der Familie an die Börse.
http://www.suse.de/~kraxel/julika-dora.jpeg


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel