Re: [RFC 3/5] vfio/pci: Respond to KVM irqchip change notifier

[David Gibson]

On Thu, Oct 17, 2019 at 02:28:45PM -0600, Alex Williamson wrote:
> On Thu, 17 Oct 2019 16:42:16 +1100
> David Gibson <address@hidden> wrote:
> 
> > VFIO PCI devices already respond to the pci intx routing notifier, in order
> > to update kernel irqchip mappings when routing is updated.  However this
> > won't handle the case where the irqchip itself is replaced by a different
> > model while retaining the same routing.  This case can happen on
> > the pseries machine type due to PAPR feature negotiation.
> > 
> > To handle that case, add a handler for the irqchip change notifier, which
> > does much the same thing as the routing notifier, but is unconditional,
> > rather than being a no-op when the routing hasn't changed.
> > 
> > Cc: Alex Williamson <address@hidden>
> > Cc: Alexey Kardashevskiy <address@hidden>
> > 
> > Signed-off-by: David Gibson <address@hidden>
> > ---
> >  hw/vfio/pci.c | 12 ++++++++++++
> >  hw/vfio/pci.h |  2 ++
> >  2 files changed, 14 insertions(+)
> > 
> > diff --git a/hw/vfio/pci.c b/hw/vfio/pci.c
> > index 529ad13908..6aa806baff 100644
> > --- a/hw/vfio/pci.c
> > +++ b/hw/vfio/pci.c
> > @@ -255,6 +255,14 @@ static void vfio_intx_routing_notifier(PCIDevice *pdev)
> >      }
> >  }
> >  
> > +static void vfio_irqchip_change(Notifier *notify, void *data)
> > +{
> > +    VFIOPCIDevice *vdev = container_of(notify, VFIOPCIDevice,
> > +                                       irqchip_change_notifier);
> > +
> > +    vfio_intx_update(vdev, &vdev->intx.route);
> > +}
> > +
> >  static int vfio_intx_enable(VFIOPCIDevice *vdev, Error **errp)
> >  {
> >      uint8_t pin = vfio_pci_read_config(&vdev->pdev, PCI_INTERRUPT_PIN, 1);
> > @@ -2960,6 +2968,8 @@ static void vfio_realize(PCIDevice *pdev, Error 
> > **errp)
> >                                                    vfio_intx_mmap_enable, 
> > vdev);
> >          pci_device_set_intx_routing_notifier(&vdev->pdev,
> >                                               vfio_intx_routing_notifier);
> > +        vdev->irqchip_change_notifier.notify = vfio_irqchip_change;
> > +        kvm_irqchip_add_change_notifier(&vdev->irqchip_change_notifier);
> >          ret = vfio_intx_enable(vdev, errp);
> >          if (ret) {
> >              goto out_teardown;
> > @@ -3009,6 +3019,7 @@ static void vfio_realize(PCIDevice *pdev, Error 
> > **errp)
> >  
> >  out_teardown:
> >      pci_device_set_intx_routing_notifier(&vdev->pdev, NULL);
> > +    kvm_irqchip_remove_change_notifier(&vdev->irqchip_change_notifier);
> 
> We can get here without having added the irqchip notifier and I don't
> think QLIST_REMOVE is very friendly with gratuitous calls.

Good point, adjusted.


> >      vfio_teardown_msi(vdev);
> >      vfio_bars_exit(vdev);
> >  error:
> > @@ -3042,6 +3053,7 @@ static void vfio_exitfn(PCIDevice *pdev)
> >      vfio_unregister_req_notifier(vdev);
> >      vfio_unregister_err_notifier(vdev);
> >      pci_device_set_intx_routing_notifier(&vdev->pdev, NULL);
> > +    kvm_irqchip_remove_change_notifier(&vdev->irqchip_change_notifier);
> 
> Same here.  Otherwise the approach looks sane to me.  Thanks,
> 
> Alex
> 
> >      vfio_disable_interrupts(vdev);
> >      if (vdev->intx.mmap_timer) {
> >          timer_free(vdev->intx.mmap_timer);
> > diff --git a/hw/vfio/pci.h b/hw/vfio/pci.h
> > index 834a90d646..11324f28ce 100644
> > --- a/hw/vfio/pci.h
> > +++ b/hw/vfio/pci.h
> > @@ -168,6 +168,8 @@ typedef struct VFIOPCIDevice {
> >      bool no_vfio_ioeventfd;
> >      bool enable_ramfb;
> >      VFIODisplay *dpy;
> > +
> > +    Notifier irqchip_change_notifier;
> >  } VFIOPCIDevice;
> >  
> >  uint32_t vfio_pci_read_config(PCIDevice *pdev, uint32_t addr, int len);
> 

-- 
David Gibson                    | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au  | minimalist, thank you.  NOT _the_ _other_
                                | _way_ _around_!
http://www.ozlabs.org/~dgibson

signature.asc
Description: PGP signature