On 12.06.14 05:56, Paul Mackerras wrote:
On Tue, Jun 10, 2014 at 07:23:00PM +0200, Alexander Graf wrote:
When we're using PR KVM we must not allow the CPU to take interrupts
in virtual mode, as the SLB does not contain host kernel mappings
when running inside the guest context.
To make sure we get good performance for non-KVM tasks but still
properly functioning PR KVM, let's just disable AIL whenever a vcpu
is scheduled in.
This patch fixes running PR KVM on POWER8 bare metal for me.
We already handle this for the situation where we're running under a
hypervisor with the calls to pSeries_disable_reloc_on_exc() and
pSeries_enable_reloc_on_exc() in kvmppc_core_init_vm_pr() and
kvmppc_core_destroy_vm_pr() respectively.
The obvious approach to fixing this problem would be to generalize
those calls, perhaps via a ppc_md callback, to work on the powernv
platform too. If you don't want to do that, for instance because
those calls are defined to operate across the whole machine rather
than a single CPU thread, and you prefer to affect just the one thread
we're running on, then I think you need to explain that in the commit
message.
It's what I've done at first, yes. Unfortunately the pSeries call is
system global, while we need to do the AIL switching per cpu on bare metal.
Once you start considering CPU hotplug and how that affects the
secondary bringup paths you start wondering whether it's worth the
hassle :). This way you can have a single guest running which only slows
down (disables AIL) on its own vcpu, while all the others still enjoy
the benefits of AIL.
Alex
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
