[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [PATCH RFC 2/2] xen/x86: Introduce XEN_SYSCTL_cpuid hypercall

On 27/02/14 12:26, Jan Beulich wrote:
>>>> On 27.02.14 at 13:11, Andrew Cooper <andrew.cooper3@xxxxxxxxxx> wrote:
>> Apart from forcibly messing with a balanced numa setup, what about cpu
>> pools, or toolstack disaggregation where pinning is restricted?
> There no messing with a balanced setup here - everything should
> of course be transient, i.e. get restored to previous values right
> after.

But anything else happening on that vcpu at the same time is transiently
out of balance.

>  And I can't see a reason not to permit the hardware domain
> to temporarily escape its enclave, as it can do worse to the overall
> system anyway.


> The new hypercall is simple enough (yet very x86-centric) that I'm
> not really against it; what I'm against is adding functionality to the
> hypervisor that is already available by other means.
> Jan

As I said before, the cache information is faked up by the cpuid policy,
so might be subject to policy depending on faulting or non-dom0 hardware
domian, or PVH dom0 in the near future. 

I think there is a legitimate case for "I really truely need the real
hardware values, without anything in the policy getting in the way". 
'cpuid' is indeed very x86 specific, but it is not information readily
available by other means.


Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.