Re: [Qemu-arm] [Qemu-devel] [PATCH] ARM: KVM: Enable in-kernel timers with user space gic

[Andrew Jones]

On Tue, Jun 27, 2017 at 03:00:10PM +0200, Alexander Graf wrote:
> > > +        /* XXX PMU IRQ is missing */
> > 
> > So I was wondering why you chose not to wire up the PMU in order to finish
> 
> Because the PMU IRQ is only available from the machine file, so it'll need
> some more plumbing to get the qemu_irq object available from kvm.c.

OK, I just wrote a patch based on this patch for the PMU and will post it
in a second.

> 
> > this off, since I knew KVM has commit 3dbbdf78636e "KVM: arm/arm64: Report
> > PMU overflow interrupts to userspace irqchip", which I presume should
> > allow the PMU to work too. However, attempting to start a guest with
> > -machine virt,accel=kvm,kernel-irqchip=off -cpu host,pmu=on doesn't provide
> > it a PMU. AFAICT, that's only because KVM still has
> > 
> >          /*
> >           * We currently require an in-kernel VGIC to use the PMU emulation,
> >           * because we do not support forwarding PMU overflow interrupts to
> >           * userspace yet.
> >           */
> >          if (!irqchip_in_kernel(vcpu->kvm) || !vgic_initialized(vcpu->kvm))
> >                  return -ENODEV;
> > 
> > in kvm_arm_pmu_v3_init(). Anyone know why that wasn't removed with
> > 3dbbdf78636e?
> 
> I guess it should get removed when the PMU sync works?

OK, I'll post the KVM patch at the same time.


I noticed another thing while testing with a virtio-scsi disk; I get
this on output when booting with userspace irqchip

 virtio-scsi: Failed to set guest notifiers (-11), ensure -enable-kvm is set
 qemu-system-aarch64: virtio_bus_start_ioeventfd: failed. Fallback to userspace 
(slower).

Just a warning, but I guess it should be removed, at least when the
user has explicitly asked for a userspace irqchip with kernel-irqchip=no.

Thanks,
drew