Re: [Xen-devel] [PATCH v2 30/48] xen/sched: introduce unit_runnable_state()

[Juergen Gross <jgross@xxxxxxxx>]

On 12.09.19 12:24, Dario Faggioli wrote:
> On Fri, 2019-08-09 at 16:58 +0200, Juergen Gross wrote:
> > Today the vcpu runstate of a new scheduled vcpu is always set to
> > "running" even if at that time vcpu_runnable() is already returning
> > false due to a race (e.g. with pausing the vcpu).
> >
> > With core scheduling this can no longer work as not all vcpus of a
> > schedule unit have to be "running" when being scheduled. So the
> > vcpu's
> > new runstate has to be selected at the same time as the runnability
> > of
> > the related schedule unit is probed.
> So, you're saying that this patch is meant at fixing a race. A race
> that, if I understood correctly, it's not there/it's not problematic
> right now, but will be when we have more than 1 vcpu in a unit and we
> enable core scheduling. Is this the case?
Correct.

> If yes:
> - this very patch, at this point in the series, is basically
> introducing no functional (or at least behavioral) change, is this
> right too?
Yes.

> - can you provide some more detail about the race. When/how it occurs
> and how the changes done in credit and rt fix it?
The problem is that with more than one vcpu in a unit you can no longer
tell the new run_state from the mere fact that the unit has been
selected to run next. So testing e.g. vcpu_runnable() after selecting
a unit for running could result in all vcpus not being runnable any
longer (e.g. when a domain was paused via domain_pause_nosync()
after selecting the next unit, but before setting any vcpu's run_state).

In order to avoid that problem the new run_state of the vcpus of a unit
needs to be determined when doing the decision which unit to select.


Juergen

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel