XENMAPSPACE_vlapic vs XENMAPSPACE_vlapic_compat

XENMAPSPACE_vlapic vs XENMAPSPACE_vlapic_compat

[James Harper]

GPLPV tries to map vlapic for TPR acceleration. Under what circumstances/versions are XENMAPSPACE_vlapic and XENMAPSPACE_vlapic_compat supported or not supported?

Thanks

James

Re: XENMAPSPACE_vlapic vs XENMAPSPACE_vlapic_compat

[Andrew Cooper]

On 01/10/2013 23:39, James Harper wrote:
> GPLPV tries to map vlapic for TPR acceleration. Under what circumstances/versions are XENMAPSPACE_vlapic and XENMAPSPACE_vlapic_compat supported or not supported?
>
> Thanks
>
> James

Short answer: not at all.
Long answer: will continue to work on XenServer for the forseeable
future for compatibility with our older windows drivers.

Both of them were gross XenServer hacks to solve the WinXP TPR
performance problem on hardware without vTPR, and never upstreamed.

XENMAPSPACE_vlapic_compat (defined as 3) became a binary incompatibility
with Xen 4.2 with the introduction of XENMAPSPACE_gmfn_range

XENMAPSPACE_vlapic (defined as 0x80000000) is luckily a long way away
from being a binary incompatibility.


With the lapic fastpath in commit
5d43891bf4002b754cd90d83e91d9190e8c8b9d0, there is less of a need for
the vlapic mapping anyway.  Furthermore, allowing a guest a RW mapping
of its own hypervisor vlapic page is a hard sell for the security
concious, although I did spend quite a long time investigating the
possible vulnerabilities and came to the conclusion that the worst a
guest could do was monkey with its own interrupt injection.

As these are just performance tweaks for WinXP (which is almost out of
extended support) on older hardware only, there are no plans to upstream
the patches. 

~Andrew

Re: XENMAPSPACE_vlapic vs XENMAPSPACE_vlapic_compat

[James Harper]

> 
> On 01/10/2013 23:39, James Harper wrote:
> > GPLPV tries to map vlapic for TPR acceleration. Under what
> circumstances/versions are XENMAPSPACE_vlapic and
> XENMAPSPACE_vlapic_compat supported or not supported?
> >
> > Thanks
> >
> > James
> 
> Short answer: not at all.
> Long answer: will continue to work on XenServer for the forseeable
> future for compatibility with our older windows drivers.
> 
> Both of them were gross XenServer hacks to solve the WinXP TPR
> performance problem on hardware without vTPR, and never upstreamed.
> 
> XENMAPSPACE_vlapic_compat (defined as 3) became a binary
> incompatibility
> with Xen 4.2 with the introduction of XENMAPSPACE_gmfn_range
> 

When I try and map it, it fails with -EINVAL, and then crashes the DomU shortly after (xen 4.3.1). I'll remove that code from GPLPV.

I wonder why I thought it was supported?

James