Re: [Qemu-devel] [RFC 2/3] intc/arm_gic: Support PPI injection for more than 256 vpus

[Auger Eric]

Hi,
On 8/29/19 9:58 AM, Auger Eric wrote:
> Hi Zenghui,
> 
> On 8/29/19 4:53 AM, Zenghui Yu wrote:
>> Hi Eric,
>>
>> On 2019/8/28 0:05, Eric Auger wrote:
>>> Host kernels that expose the KVM_CAP_ARM_IRQ_LINE_LAYOUT_2 capability
>>> allow injection of PPIs along with vcpu ids larger than 255. Let's
>>> encode the vpcu id on 12 bits according to the upgraded KVM_IRQ_LINE
>>> ABI when needed.
>>>
>>> Without that patch qemu exits with "kvm_set_irq: Invalid argument"
>>> message.
>>>
>>> Signed-off-by: Eric Auger <address@hidden>
>>> Reported-by: Zenghui Yu <address@hidden>
>>> ---
>>>   hw/intc/arm_gic_kvm.c | 10 +++++++---
>>>   1 file changed, 7 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/hw/intc/arm_gic_kvm.c b/hw/intc/arm_gic_kvm.c
>>> index b56fda144f..889293e97f 100644
>>> --- a/hw/intc/arm_gic_kvm.c
>>> +++ b/hw/intc/arm_gic_kvm.c
>>> @@ -56,6 +56,7 @@ void kvm_arm_gic_set_irq(uint32_t num_irq, int irq,
>>> int level)
>>>        * CPU number and interrupt number.
>>>        */
>>>       int kvm_irq, irqtype, cpu;
>>> +    int cpu_idx1 = 0, cpu_idx2 = 0;
>>>         if (irq < (num_irq - GIC_INTERNAL)) {
>>>           /* External interrupt. The kernel numbers these like the GIC
>>> @@ -63,17 +64,20 @@ void kvm_arm_gic_set_irq(uint32_t num_irq, int
>>> irq, int level)
>>>            * internal ones.
>>>            */
>>>           irqtype = KVM_ARM_IRQ_TYPE_SPI;
>>> -        cpu = 0;
>>>           irq += GIC_INTERNAL;
>>>       } else {
>>>           /* Internal interrupt: decode into (cpu, interrupt id) */
>>>           irqtype = KVM_ARM_IRQ_TYPE_PPI;
>>>           irq -= (num_irq - GIC_INTERNAL);
>>>           cpu = irq / GIC_INTERNAL;
>>> +        cpu_idx2 = cpu / 256;
>>> +        cpu_idx1 = cpu % 256;
>>>           irq %= GIC_INTERNAL;
>>>       }
>>> -    kvm_irq = (irqtype << KVM_ARM_IRQ_TYPE_SHIFT)
>>> -        | (cpu << KVM_ARM_IRQ_VCPU_SHIFT) | irq;
>>> +    kvm_irq = (irqtype << KVM_ARM_IRQ_TYPE_SHIFT) |
>>> +              (cpu_idx1 << KVM_ARM_IRQ_VCPU_SHIFT) |
>>> +              ((cpu_idx2 & KVM_ARM_IRQ_VCPU2_MASK) <<
>>> KVM_ARM_IRQ_VCPU2_SHIFT) |
>>> +              irq;
>>>         kvm_set_irq(kvm_state, kvm_irq, !!level);
>>>   }
>>>
>>
>> For confirmation, should we also adjust the vcpu_index in
>> arm_cpu_kvm_set_irq(), just like above?
> 
> I am not familiar with this path. in arm_cpu_initfn(), there is a
> comment saying "VIRQ and VFIQ are unused with KVM but we add them to
> maintain the same interface as non-KVM CPUs." So I don't know when that
> code gets executed.
> 
> But maybe it would be more cautious to implement your suggestion here as
> well.
> 
> Maybe Peter can provide more info here?

If this is supposed to get used along with kernel_irqchip=off, it seems
this latter is not supported with GICv3 anyway. So max number of vcpus
with GICv2 is 8.

Thanks

Eric
> 
> Thanks
> 
> Eric
> 
> 
>>
>>
>> Thanks,
>> zenghui
>>
>>
>