Re: [Xen-devel] [bug] 'VT-d 1G super page' feature is blocked

[Tim Deegan <tim@xxxxxxx>]

At 17:25 +0100 on 10 Aug (1312997134), Tim Deegan wrote:
> This turns out not to work.  Allen, how about the attached patch
> instead?

So this breaks on dom0 boot, but that's a symptom of a deeper illness. 
By chance, the old code (before 23247) only actually enabled EPT-sharing
when the first HVM domain tried to use it.  23247 pulled that enabling
up to boot time, and exposed the bug that PV guests aren't properly
supported.  AFAICS the code just tries to share the P2M table without
making sure there is one there to share. :(

Tim.


> diff -r 1f08b380d438 xen/drivers/passthrough/vtd/iommu.c
> --- a/xen/drivers/passthrough/vtd/iommu.c     Wed Aug 10 14:43:34 2011 +0100
> +++ b/xen/drivers/passthrough/vtd/iommu.c     Wed Aug 10 17:24:19 2011 +0100
> @@ -1731,15 +1731,15 @@ void iommu_pte_flush(struct domain *d, u
>  
>  static int vtd_ept_page_compatible(struct iommu *iommu)
>  {
> -    u64 cap = iommu->cap;
> +    u64 ept_cap, vtd_cap = iommu->cap;
>  
> -    if ( ept_has_2mb(cpu_has_vmx_ept_2mb) != cap_sps_2mb(cap) )
> +    /* EPT is not initialised yet, so we must check the capability in
> +     * the MSR explicitly rather than use cpu_has_vmx_ept_*() */
> +    if ( rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP, ept_cap) != 0 ) 
>          return 0;
>  
> -    if ( ept_has_1gb(cpu_has_vmx_ept_1gb) != cap_sps_1gb(cap) )
> -        return 0;
> -
> -    return 1;
> +    return ( ept_has_2mb(ept_cap) == cap_sps_2mb(vtd_cap) 
> +             && ept_has_1gb(ept_cap) == cap_sps_1gb(vtd_cap) );
>  }
>  
>  /*

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


-- 
Tim Deegan <tim@xxxxxxx>
Principal Software Engineer, Xen Platform Team
Citrix Systems UK Ltd.  (Company #02937203, SL9 0BG)

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