|
|
|
|
|
|
|
|
|
|
xen-devel
Re: [Xen-devel] [Patch 2/4] Refining Xsave/Xrestore support
I have done a quick test on my box, setting XCR0 is around 130 tsc
cycles while reading XCR0 is around 80 tsc cycles.
Shan Haitao
2010/10/28 Jan Beulich <JBeulich@xxxxxxxxxx>:
>>>> On 28.10.10 at 09:52, Haitao Shan <maillists.shan@xxxxxxxxx> wrote:
>> Then I would prefer to write XCR0 unconditionally. Otherwise, I can
>> only refer to the approach for handling CR4 switches: reading CR4
>> first and checking whether there is a need to write actually.
>> But I don't think <a read to XCR0 plus a data comparison> can save any
>> compared with one unconditional write to XCR0.
>> Are you OK with this?
>
> Depends on the performance expectations of xsetbv and xgetbv
> (and its comparison to moves from/to control registers). At least
> there's no word in the documentation that xsetbv would be
> serializing. I would hope Intel could at least provide approximate
> numbers...
>
> Jan
>
>
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel
|
|
|
|
|