Re: [PATCH] Dont' round-robin the callback interrupt

[Juergen Gross <juergen.gross@ts.fujitsu.com>]

On 07/12/2010 07:41 PM, Keir Fraser wrote:
> On 12/07/2010 18:17, "Keir Fraser"<keir.fraser@eu.citrix.com>  wrote:
>
>>>    However, that's not the motivation for this patch. In the windows code, we
>>> only bind event channels to vcpu 0 since we cannot get callback interrupts on
>>> multiple vcpus simultaneously, since the interrupt is level sensitive. Thus
>>> round-robining is wasteful in terms of kicking certain data structures
>>> between
>>> caches (assuming a reasonably constant vcpu ->  pcpu mapping).
>>
>> Surely that argument can be made for any interrupt that is set up to
>> round-robin among multiple CPUs? Obviously in the PV drivers case the
>> event-channel IRQ is probably the only significant source of round-robin
>> interrupts. But I don't see that it's special in any other way.
>
> Further, the correct semantics for LowestPrio delivery was implemented by
> Juergen Gross at Fujitsu for a reason. Cc'ing him. I suspect he will say
> that relaxing the delivery semantics will cause something he cares about to
> break.

Thanks for CC'ing me, Keir.
Selecting different CPUs gives at least our BS2000 system a performance win of
a few percent. As Keir already said, that's the reason I implemented the LPP
delivery of interrupts.
If you really need a different interrupt delivery behaviour I would at least
recommend a per-domain parameter for violating the correct semantics using the
LPP delivery as default.


Juergen

-- 
Juergen Gross                 Principal Developer Operating Systems
TSP ES&S SWE OS6                       Telephone: +49 (0) 89 3222 2967
Fujitsu Technology Solutions              e-mail: juergen.gross@ts.fujitsu.com
Domagkstr. 28                           Internet: ts.fujitsu.com
D-80807 Muenchen                 Company details: ts.fujitsu.com/imprint.html