Xen devel - May 2010 - bad interaction between boosted idle vcpus and csched''s tickling?

Keir,

isn''t there a problem after the change to handle tasklets in idle vcpus
in that {default,acpi_processor}_idle() will put the vcpu to sleep with
boosted priority still in place, hence preventing it from being tickled
and/or needlessly tickling other CPUs?

Jan


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

On 07/05/2010 13:36, "Jan Beulich" <JBeulich@novell.com> wrote:
> Keir,
> 
> isn''t there a problem after the change to handle tasklets in idle
vcpus
> in that {default,acpi_processor}_idle() will put the vcpu to sleep with
> boosted priority still in place, hence preventing it from being tickled
> and/or needlessly tickling other CPUs?
See the comment above tasklet_list definition in tasklet.c: we always run
through the scheduler when tasklets are added/removed. That is explicitly to
get the idle thread''s priority set correctly.

I think this could be handled better though, and also in the current
approach I actually still probably don''t have enough raise_softirq()
calls.
:-) I will take another look and see if I can make things more obvious and
robust.

 Thanks,
 Keir
> Jan
> 


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

On 07/05/2010 13:36, "Jan Beulich" <JBeulich@novell.com> wrote:
> Keir,
> 
> isn''t there a problem after the change to handle tasklets in idle
vcpus
> in that {default,acpi_processor}_idle() will put the vcpu to sleep with
> boosted priority still in place, hence preventing it from being tickled
> and/or needlessly tickling other CPUs?
Should be resolved by xen-unstable:21344.

 -- Keir



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