vtsc in domain builder

vtsc in domain builder

[Dan Magenheimer]

I'm working on passing vtsc=[0|1] (rdtsc emulation) from
the vm.cfg file into xen.

I was using the superpages vm.cfg implementation as a guide,
but I see the domain builder interface for superpages changed
in c/s 19963 to pass the value via a flag in a pad byte in
struct xc_dom_image.  Was this change due to some backward
compatibility reason?  This looks like a nicer interface,
but is there another pad byte I can/should use?  If not,
what's the best way then to pass the vtsc value from the
tools to Xen?

Also, will I be able to use the same interface for both
HVM and PV, or should I use the HVM param interface for
HVM domains (using the HPET parameter as a model)?

Thanks,
Dan

RE: vtsc in domain builder

[Dan Magenheimer]

> Was this change due to some backward compatibility reason?

I found the compatibility issue describe here:

http://xen.markmail.org/message/usjxnjyp3csm3uct?q=superpages+date:200907

So it seems even more important that I somehow claim
a field (or at least a bit) in struct xc_dom_image.
Any help to get this right would be appreciated
as getting it wrong clearly has some non-obvious
consequences.

Thanks,
Dan

> -----Original Message-----
> From: Dan Magenheimer 
> Sent: Tuesday, September 15, 2009 3:58 PM
> To: Xen-Devel (E-mail)
> Subject: vtsc in domain builder
> 
> 
> I'm working on passing vtsc=[0|1] (rdtsc emulation) from
> the vm.cfg file into xen.
> 
> I was using the superpages vm.cfg implementation as a guide,
> but I see the domain builder interface for superpages changed
> in c/s 19963 to pass the value via a flag in a pad byte in
> struct xc_dom_image.  Was this change due to some backward
> compatibility reason?  This looks like a nicer interface,
> but is there another pad byte I can/should use?  If not,
> what's the best way then to pass the vtsc value from the
> tools to Xen?
> 
> Also, will I be able to use the same interface for both
> HVM and PV, or should I use the HVM param interface for
> HVM domains (using the HPET parameter as a model)?
> 
> Thanks,
> Dan
>

Re: vtsc in domain builder

[Keir Fraser]

On 16/09/2009 00:32, "Dan Magenheimer" <dan.magenheimer@oracle.com> wrote:

>> Was this change due to some backward compatibility reason?
> 
> I found the compatibility issue describe here:
> 
> http://xen.markmail.org/message/usjxnjyp3csm3uct?q=superpages+date:200907
> 
> So it seems even more important that I somehow claim
> a field (or at least a bit) in struct xc_dom_image.
> Any help to get this right would be appreciated
> as getting it wrong clearly has some non-obvious
> consequences.

Well, the main issue was the original patch broke API compatibility. Fixing
ABI compatibility in the fixup patch was just gravy.

I suggest you split the superpages int8_t into a bitfield, e.g:
uint8_t superpages:1, vtsc:1;
Putting superpages first should maintain compatibility as far as possible.
The bitfield has to be switched to be unsigned - we got bit by that before.

 -- Keir