xen: sched: don't call hooks of the wrong scheduler via VCPU2OP
authorDario Faggioli <dario.faggioli@citrix.com>
Fri, 17 Mar 2017 18:19:37 +0000 (19:19 +0100)
committerGeorge Dunlap <george.dunlap@citrix.com>
Mon, 27 Mar 2017 13:28:05 +0000 (14:28 +0100)
commita3653e6a279213ba4e883b2252415dc98633106a
treebac163454a0251b453ceacd9b5f9f28e90c7e788
parentbd819203e05b5204b5d6911294149609cf436387
xen: sched: don't call hooks of the wrong scheduler via VCPU2OP

Within context_saved(), we call the context_saved hook,
and we use VCPU2OP() to determine from what scheduler.
VCPU2OP uses DOM2OP, which uses d->cpupool, which is
NULL when d is the idle domain. And in that case,
DOM2OP just returns ops, the scheduler of cpupool0.

Therefore, if:
- cpupool0's scheduler defines context_saved (like
  Credit2 and RTDS do),
- we are not in cpupool0 (i.e., our scheduler is
  not ops),
- we are context switching from idle,

we call VCPU2OP(idle_vcpu), which means
DOM2OP(idle->cpupool), which is ops.

Therefore, we both:
- check if context_saved is defined in the wrong
  scheduler;
- if yes, call the wrong one.

When using Credit2 at boot, and also Credit2 in
the other cpupool, this is wrong but innocuous,
because it only involves the idle vcpus.

When using Credit2 at boot, and Credit1 in the
other cpupool, this is *totally* wrong, and
it's by chance it does not explode!

When using Credit2 and other schedulers I'm
developping, I hit the following assert (in
sched_credit2.c, on a CPU inside a cpupool that
does not use Credit2):

csched2_context_saved()
{
 ...
 ASSERT(!vcpu_on_runq(svc));
 ...
}

Fix this by dealing explicitly, in VCPU2OP, with
idle vcpus, returning the scheduler of the pCPU
they (always) run on.

Signed-off-by: Dario Faggioli <dario.faggioli@citrix.com>
Reviewed-by: Juergen Gross <jgross@suse.com>
Reviewed-by: George Dunlap <george.dunlap@citrix.com>
xen/common/schedule.c