Re: [XenPPC] [pushed] [ppc] FPSCR and XER are really 64 bit quantities

[Hollis Blanchard <hollisb@xxxxxxxxxx>]

On Tue, 2006-04-25 at 16:54 -0400, Jimi Xenidis wrote:
> On Apr 25, 2006, at 4:08 PM, Hollis Blanchard wrote:
> 
> > On Tue, 2006-04-25 at 14:14 -0400, Jimi Xenidis wrote:
> >> changeset:   9950:249f22f51bb31148f896be0c7e85618b20f8c5fd
> >> tag:         tip
> >> user:        jimix@xxxxxxxxxxxxxxxxxxxxx
> >> date:        Tue Apr 25 15:04:52 2006 -0400
> >> files:       xen/arch/ppc/float.S xen/arch/ppc/gdbstub.c xen/arch/ 
> >> ppc/ppc64/exceptions.S xen/include/public/arch-ppc64.h
> >> description:
> >> [ppc] FPSCR and XER are really 64 bit quantities
> >
> > Neither the PEM PDF (page 64) nor the Architecture Book I PDF (ver  
> > 2.02,
> > page 97) agree with you about FPSCR. Please explain?
> 
> look at the original:
>    http://lists.xensource.com/archives/html/xen-ppc-devel/2006-04/ 
> msg00050.html
> 
> it is explained in the rest of the comment:
>    - FPSCR should be save/restored with stfd/lfd since the 32 bit form
>      can mangle the value.

I read that; I still don't get it. If FPSCR is 32 bits, as all available
documentation claims, then couldn't you use stfiwx?

Your comment, on the other hand, claims it is 64 bits, and if that's
true then you wouldn't use stfiwx. However, in that case, I'd like to
know why you think it's 64 bits.

> > The PEM PDF (again page 64) contradicts the Architecture Book I PDF
> > (page 42), which is unfortunate, but I'm ok with that change.
> 
> If you are questioning the XER here, you need to look at the 64bit  
> PEM (pem_64bit_v3.0.2005jul15.pdf)
> on page 50. Snippet below if you don't have it.

Ah, great, I had an older copy.

-- 
Hollis Blanchard
IBM Linux Technology Center


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