On Wed, Oct 09, 2019 at 05:08:17PM +1100, David Gibson wrote: > pseries machine type can have one of two different interrupt controllers in > use depending on feature negotiation with the guest. Usually this is > invisible to devices, because they route to a common set of qemu_irqs which > in turn dispatch to the correct back end. > > VFIO passthrough devices, however, wire themselves up directly to the KVM > irqchip for performance, which means they are affected by this change in > interrupt controller. > > Luckily, there's a notifier chain that will tell VFIO devices to update > their mappings - we just need to call it whenever the intc backend might > change. > > In addition, we make sure we set an active intc earlier, because otherwise > vfio can issue a false warning, because it doesn't think a KVM irqchip is > in use (which is essentially for good INTx performance). > > Cc: Alex Williamson <[email protected]> > Cc: Alexey Kardashevskiy <[email protected]> > > Signed-off-by: David Gibson <[email protected]>
For reference, the reason I described this as RFC only in the cover
letter, is that this doesn't work as it stands.
With this patch, when we switch intc we call the intx_routing
notifiers, and vfio_intx_update() in particular. However that exits
early without calling vfio_intx_enable_kvm() - the bit we actually
need - because the test on pci_intx_route_changed() thinks there's
nothing to do. The difficulty is that our use case isn't really the
same as the x86 one this notifier path was designed for: they're
changing routing of INTx to global interrupts. For us the routing to
global interrupts remains the same, but the interrupt controller
handling those global interrupts has changed.
> ---
> hw/ppc/spapr_irq.c | 6 ++++++
> hw/ppc/spapr_pci.c | 9 +++++++++
> include/hw/ppc/spapr.h | 1 +
> 3 files changed, 16 insertions(+)
>
> diff --git a/hw/ppc/spapr_irq.c b/hw/ppc/spapr_irq.c
> index 473fc8780a..7964e4a1b8 100644
> --- a/hw/ppc/spapr_irq.c
> +++ b/hw/ppc/spapr_irq.c
> @@ -409,6 +409,12 @@ static void set_active_intc(SpaprMachineState *spapr,
> }
>
> spapr->active_intc = new_intc;
> +
> + /*
> + * We've changed the interrupt routing at the KVM level, let VFIO
> + * devices know they need to readjust.
> + */
> + spapr_pci_fire_intx_routing_notifiers(spapr);
> }
>
> void spapr_irq_update_active_intc(SpaprMachineState *spapr)
> diff --git a/hw/ppc/spapr_pci.c b/hw/ppc/spapr_pci.c
> index cc0e7829b6..3bcf6325d4 100644
> --- a/hw/ppc/spapr_pci.c
> +++ b/hw/ppc/spapr_pci.c
> @@ -93,6 +93,15 @@ PCIDevice *spapr_pci_find_dev(SpaprMachineState *spapr,
> uint64_t buid,
> return pci_find_device(phb->bus, bus_num, devfn);
> }
>
> +void spapr_pci_fire_intx_routing_notifiers(SpaprMachineState *spapr)
> +{
> + SpaprPhbState *sphb;
> +
> + QLIST_FOREACH(sphb, &spapr->phbs, list) {
> + pci_bus_fire_intx_routing_notifier(PCI_HOST_BRIDGE(sphb)->bus);
> + }
> +}
> +
> static uint32_t rtas_pci_cfgaddr(uint32_t arg)
> {
> /* This handles the encoding of extended config space addresses */
> diff --git a/include/hw/ppc/spapr.h b/include/hw/ppc/spapr.h
> index d3b4dd7de3..66b68fdd5e 100644
> --- a/include/hw/ppc/spapr.h
> +++ b/include/hw/ppc/spapr.h
> @@ -805,6 +805,7 @@ void spapr_clear_pending_events(SpaprMachineState *spapr);
> int spapr_max_server_number(SpaprMachineState *spapr);
> void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
> uint64_t pte0, uint64_t pte1);
> +void spapr_pci_fire_intx_routing_notifiers(SpaprMachineState *spapr);
>
> /* DRC callbacks. */
> void spapr_core_release(DeviceState *dev);
--
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
