Re: [Qemu-devel] [PATCH] RFCv3 kvm irqfd: support msimessage to irq translation in PHB

[Alexey Kardashevskiy <aik@ozlabs.ru>]

On 06/30/2013 12:28 AM, Anthony Liguori wrote:
> On Sat, Jun 29, 2013 at 8:45 AM, Alexey Kardashevskiy <aik@ozlabs.ru> wrote:
>> On PPC64 systems MSI Messages are translated to system IRQ in a PCI
>> host bridge. This is already supported for emulated MSI/MSIX but
>> not for irqfd where the current QEMU allocates IRQ numbers from
>> irqchip and maps MSIMessages to those IRQ in the host kernel.
>>
>> The patch extends irqfd support in order to avoid unnecessary
>> mapping and reuse the one which already exists in a PCI host bridge.
>>
>> Specifically, a map_msi callback is added to PCIBus and pci_bus_map_msi()
>> to PCI API. The latter returns -1 if a specific PHB does not provide
>> with any trsnslation so the existing code will work.
> 
> I think there's a bit of confusion here.  The kernel needs a "virq"
> number to create an eventfd.  virq is just a KVM concept, it doesn't
> correspond to anything useful in hardware.
> 
> On pseries, there is a 1-1 mapping between XICS IRQs and VIRQs and MSI
> can be trivially mapped to a virq.
> 
> On x86, we need to call a special kernel function which essentially
> creates an apic message->virq mapping such that we can deliver the
> irqfd.
> 
> So what this should look like is:
> 
> 1) A PCI bus function to do the MSI -> virq mapping
> 2) On x86 (and e500), this is implemented by calling kvm_irqchip_add_msi_route()
> 3) On pseries, this just returns msi->data
> 
> Perhaps (2) can just be the default PCI bus implementation to simplify things.
> 
>> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
>>
>> ---
>>
>> Looks like we agreed that in general PHB is the right place for this,
>> not KVM, so I am trying again.
>>
>> Probably something should be done to kvm_irqchip_update_msi_route()
>> as well but I do not really understand what exactly. Any suggestions?


Ah. Everybody ignored, I'll try asking again :)

kvm_irqchip_update_msi_route() - where should it go? What is it for?
virtio-pci and pci device assignment use it but vfio does not - is it a bug
of vfio? Thanks.



-- 
Alexey