Re: [Xen-devel] [PATCH v4 16/26] x86/cpu: Context switch cpuid masks and faulting state in context_switch()

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 28/03/16 20:27, Konrad Rzeszutek Wilk wrote:
> On Wed, Mar 23, 2016 at 04:36:19PM +0000, Andrew Cooper wrote:
>> A single ctxt_switch_levelling() function pointer is provided
>> (defaulting to an empty nop), which is overridden in the appropriate
>> $VENDOR_init_levelling().
>>
>> set_cpuid_faulting() is made private and included within
>> intel_ctxt_switch_levelling().
>>
>> One functional change is that the faulting configuration is no longer special
>> cased for dom0.  There was never any need to, and it will cause dom0 to
> There was. See 1d6ffea6
>     ACPI: add _PDC input override mechanism
>
> And in Linux see xen_check_mwait().

This logic is fundamentally broken.  It is, and has *alway* been wrong
to try and equate dom0's view of cpuid with the hosts view of cpuid...

>
>> observe the same information through native and enlightened cpuid.
> Which will be a regression when it comes to ACPI C-states

... it can't possibly work for an HVM-based dom0, where there is no
distinction between a native and enlightened cpuid.

I will clearly have to re-break this in a similar way to the MTRR bits,
whitelisted for the hardware domain only, but Linux will have to be
changed to get working deep cstates for PVH/HVMLite.

~Andrew

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