Re: [Xen-devel] [PATCH] i386: remove NMI deferral

[Keir Fraser <keir@xxxxxxxxxxxxx>]

On 19/6/07 11:01, "Jan Beulich" <jbeulich@xxxxxxxxxx> wrote:

> +        cmpw %cx,%di;                                   \
> +        mov %ecx,%ds;                                   \
>          mov %fs,UREGS_fs(%esp);                         \
> +        cmovel UREGS_ds(%esp),%edi;                     \
> +        cmpw %cx,%si;                                   \
> +        mov %edi,UREGS_ds(%esp);                        \
> +        cmovel UREGS_es(%esp),%esi;                     \
> +        mov %ecx,%es;                                   \
>          mov %gs,UREGS_gs(%esp);                         \

It looks like this section of SAVE_ALL gets executed only if we interrupted
the guest in protected mode (i.e., not vm86 mode, and not an outer Xen
context). So I'm not sure how the cmove's will ever be a no-op -- how could
we have UREGS_cs&3!=0, yet have HYPERVISOR_DS in %ds or %es? I guess I must
be missing something fundamental about how this patch works...

The cmove's themselves raise another question -- can we really rely on all
32-bit CPUs that Xen is likely to run on to actually have CMOV support? It
seems possible, but I'm not certain.

 -- Keir


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