Xen devel - Apr 2006 - RE: Hypercall number assignment convension (was Re: Re:[PATCH]: kexec: framework and i386)

>From: Keir Fraser [mailto:Keir.Fraser@cl.cam.ac.uk]
>Sent: 2006年4月26日 15:38
>
>On 26 Apr 2006, at 03:34, Tian, Kevin wrote:
>
>> I prefer to the first one. However not the current
>> __HYPERVISOR_arch_specific_0, *_1, *_2, ..., how about just call
>> it as __HYPERVISOR_arch_specific_ops which contains another
>> namespace defined by different architecture seperately?
>
>Sometimes you might want a fast hypercall that doesn''t have two
levels
>of demultiplexing, or where the register/stack parameters are carefully
>crafted and would not fit with an ioctl()-style hypercall (see
x86''s
>IRET hypercall).
See.
>
>How about reserving 8 or 16 arch-specific hypercalls up front?
>#define __HYPERVISOR_arch_0  32
>  ...
>#define __HYPERVISOR_arch_7  39
>
>We could give a bit of breathing space to the non-arch range by
>starting __HYPERVISOR_arch_* at. e.g., 40 or 48?
>
>  -- Keir
Then we may need to fill that breathing space with do_ni_hypercall 
to ensure no leakage from NR_hypercall check. If that''s the case, 
how about define the __HYPERVISOR_arch_* at end of 256 spaces, 
and fill all unused entries with do_ni_hypercall. By that way, the check
to illegal hypercall (<256) is a bit slower, however it shouldn''t
matter
for that rare cases.

Thanks,
Kevin

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

On 26 Apr 2006, at 08:54, Tian, Kevin wrote:
> Then we may need to fill that breathing space with do_ni_hypercall
> to ensure no leakage from NR_hypercall check. If that''s the case,
> how about define the __HYPERVISOR_arch_* at end of 256 spaces,
> and fill all unused entries with do_ni_hypercall. By that way, the 
> check
> to illegal hypercall (<256) is a bit slower, however it
shouldn''t
> matter
> for that rare cases.
Yes, it would need filling with ni_hypercall: we already do that on x86 
anyway (since hypercall table is rounded up to a power of two).

I don''t want to put the hypercalls that far up: with one hypercall page
x86 will currently have a problem implementing more than 128 
hypercalls. I also don''t want to put them right at the end of the 
hypercall space because that would make it harder/uglier to add extra 
arch hypercalls later on.

I think 48-55 would be reasonable.

  -- Keir


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