ARM Guest VM Type

ARM Guest VM Type

[Wei Huang]

According to enlighten.c in Kernel, ARM DomU is considered as HVM 
domain. However when the DomU is created, its property isn't configured 
to reflect this (XEN_DOMCTL_CDF_xxx) in tools. So is_hvm_domain(d) will 
return FALSE inside hypervisor.

Error?

-Wei

Re: ARM Guest VM Type

[Stefano Stabellini]

On Fri, 25 Apr 2014, Wei Huang wrote:
> According to enlighten.c in Kernel, ARM DomU is considered as HVM domain.
> However when the DomU is created, its property isn't configured to reflect
> this (XEN_DOMCTL_CDF_xxx) in tools. So is_hvm_domain(d) will return FALSE
> inside hypervisor.
> 
> Error?

It is not really an error, more an inconsistency. In reality the old
"PV" and "HVM" labels don't fit ARM guests very well.

>From Linux POV it makes sense to think of ARM guests as HVM guests.
If it makes sense to do the same in Xen, then we might as well set the
flag.

Re: ARM Guest VM Type

[Tamas Lengyel]

[-- Attachment #1.1: Type: text/plain, Size: 1239 bytes --]

What about paging? On x86 it's configurable whether I want the guest to use
shadow or HAP via hap=0/1 but I haven't find anything about that for the
ARM side. Looking at the source I got the feeling that it's shadow paging
if the guest is created with the config shown in
http://wiki.xenproject.org/wiki/Xen_ARM_with_Virtualization_Extensions#DomU_kernel_and_DTS.
Please correct me if I'm wrong! Thanks =)


On Sat, Apr 26, 2014 at 4:17 PM, Stefano Stabellini <
stefano.stabellini@eu.citrix.com> wrote:

> On Fri, 25 Apr 2014, Wei Huang wrote:
> > According to enlighten.c in Kernel, ARM DomU is considered as HVM domain.
> > However when the DomU is created, its property isn't configured to
> reflect
> > this (XEN_DOMCTL_CDF_xxx) in tools. So is_hvm_domain(d) will return FALSE
> > inside hypervisor.
> >
> > Error?
>
> It is not really an error, more an inconsistency. In reality the old
> "PV" and "HVM" labels don't fit ARM guests very well.
>
> From Linux POV it makes sense to think of ARM guests as HVM guests.
> If it makes sense to do the same in Xen, then we might as well set the
> flag.
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel
>

[-- Attachment #1.2: Type: text/html, Size: 1950 bytes --]

[-- Attachment #2: Type: text/plain, Size: 126 bytes --]

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

Re: ARM Guest VM Type

[Julien Grall]

Hello Tamas,

On 27/04/14 11:25, Tamas Lengyel wrote:
> What about paging? On x86 it's configurable whether I want the guest to
> use shadow or HAP via hap=0/1 but I haven't find anything about that for
> the ARM side. Looking at the source I got the feeling that it's shadow
> paging if the guest is created with the config shown in
> http://wiki.xenproject.org/wiki/Xen_ARM_with_Virtualization_Extensions#DomU_kernel_and_DTS.
> Please correct me if I'm wrong! Thanks =)

Xen on ARM takes directly advantage of the hardware virtualization 
feature.  ARM processors provide a stage 2 translation (i.e guest PA -> 
host PA).

We don't support other kind of P2M translation (such as shadow PT) on ARM.

Regards,

-- 
Julien Grall

Re: ARM Guest VM Type

[Ian Campbell]

On Sat, 2014-04-26 at 15:17 +0100, Stefano Stabellini wrote:
> On Fri, 25 Apr 2014, Wei Huang wrote:
> > According to enlighten.c in Kernel, ARM DomU is considered as HVM domain.
> > However when the DomU is created, its property isn't configured to reflect
> > this (XEN_DOMCTL_CDF_xxx) in tools. So is_hvm_domain(d) will return FALSE
> > inside hypervisor.
> > 
> > Error?
> 
> It is not really an error, more an inconsistency. In reality the old
> "PV" and "HVM" labels don't fit ARM guests very well.
> 
> From Linux POV it makes sense to think of ARM guests as HVM guests.
> If it makes sense to do the same in Xen, then we might as well set the
> flag.

The "PV" and "HVM" labels are x86 specific and have no real bearing on
the ARM hypervisor. We've been slowly eliminating the is_{pv,hvm}_foo
calls from the generic code in the hypervisor in favour of more specific
functionality tests (often paging-mode-translate based ones). We should
continue to do this cleanup, setting any kind of PV or HVM based flag
for an ARM guest would be wrong.

IMHO we should be doing the same on Linux (and all guest kernels) as
well, but the leakage of x86-isms into the generic code is more
pervasive there it seems (although I hope that between ARM and x86 PVH
we can eventually clean it up there too)

Ian.