Re: [PATCH v3 41/59] KVM: arm/arm64: GICv4: Wire mapping/unmapping of VLPIs in VFIO irq bypass

[Auger Eric <eric.auger@redhat.com>]

Hi Marc,

On 30/08/2017 12:42, Marc Zyngier wrote:
> Hi Eric,
> 
> On 30/08/17 11:20, Auger Eric wrote:
>> Hi Marc,
>>
>> On 30/08/2017 11:42, Marc Zyngier wrote:
>>> On 26/08/17 20:48, Christoffer Dall wrote:
>>>> On Mon, Jul 31, 2017 at 06:26:19PM +0100, Marc Zyngier wrote:
>>>>> Let's use the irq bypass mechanism introduced for platform device
>>>>> interrupts to intercept the virtual PCIe endpoint configuration
>>>>> and establish our LPI->VLPI mapping.
>>>>>
>>>>> Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
>>>>> ---
>>>>>  include/kvm/arm_vgic.h      |   8 ++++
>>>>>  virt/kvm/arm/arm.c          |  27 ++++++++----
>>>>>  virt/kvm/arm/vgic/vgic-v4.c | 103 ++++++++++++++++++++++++++++++++++++++++++++
>>>>>  3 files changed, 130 insertions(+), 8 deletions(-)
>>>>>
>>>>> diff --git a/include/kvm/arm_vgic.h b/include/kvm/arm_vgic.h
>>>>> index 359eeffe9857..050f78d4fb42 100644
>>>>> --- a/include/kvm/arm_vgic.h
>>>>> +++ b/include/kvm/arm_vgic.h
>>>>> @@ -367,4 +367,12 @@ int kvm_vgic_set_forwarding(struct kvm *kvm, unsigned int host_irq,
>>>>>  void kvm_vgic_unset_forwarding(struct kvm *kvm, unsigned int host_irq,
>>>>>  			       unsigned int vintid);
>>>>>  
>>>>> +struct kvm_kernel_irq_routing_entry;
>>>>> +
>>>>> +int kvm_vgic_v4_set_forwarding(struct kvm *kvm, int irq,
>>>>> +			       struct kvm_kernel_irq_routing_entry *irq_entry);
>>>>> +
>>>>> +int kvm_vgic_v4_unset_forwarding(struct kvm *kvm, int irq,
>>>>> +				 struct kvm_kernel_irq_routing_entry *irq_entry);
>>>>> +
>>>>>  #endif /* __KVM_ARM_VGIC_H */
>>>>> diff --git a/virt/kvm/arm/arm.c b/virt/kvm/arm/arm.c
>>>>> index ebab6c29e3be..6803ea27c47d 100644
>>>>> --- a/virt/kvm/arm/arm.c
>>>>> +++ b/virt/kvm/arm/arm.c
>>>>> @@ -1457,11 +1457,16 @@ int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
>>>>>  	struct kvm_kernel_irqfd *irqfd =
>>>>>  		container_of(cons, struct kvm_kernel_irqfd, consumer);
>>>>>  
>>>>> -	if (prod->type != IRQ_BYPASS_VFIO_PLATFORM)
>>>>> +	switch (prod->type) {
>>>>> +	case IRQ_BYPASS_VFIO_PLATFORM:
>>>>> +		return kvm_vgic_set_forwarding(irqfd->kvm, prod->irq,
>>>>> +					       irqfd->gsi + VGIC_NR_PRIVATE_IRQS);
>>>>> +	case IRQ_BYPASS_VFIO_PCI_MSI:
>>>>> +		return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
>>>>> +						  &irqfd->irq_entry);
>>>>> +	default:
>>>>>  		return 0;
>>>>> -
>>>>> -	return kvm_vgic_set_forwarding(irqfd->kvm, prod->irq,
>>>>> -				       irqfd->gsi + VGIC_NR_PRIVATE_IRQS);
>>>>> +	}
>>>>>  }
>>>>>  void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
>>>>>  				      struct irq_bypass_producer *prod)
>>>>> @@ -1469,11 +1474,17 @@ void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
>>>>>  	struct kvm_kernel_irqfd *irqfd =
>>>>>  		container_of(cons, struct kvm_kernel_irqfd, consumer);
>>>>>  
>>>>> -	if (prod->type != IRQ_BYPASS_VFIO_PLATFORM)
>>>>> -		return;
>>>>> +	switch (prod->type) {
>>>>> +	case IRQ_BYPASS_VFIO_PLATFORM:
>>>>> +		kvm_vgic_unset_forwarding(irqfd->kvm, prod->irq,
>>>>> +					  irqfd->gsi + VGIC_NR_PRIVATE_IRQS);
>>>>> +		break;
>>>>>  
>>>>> -	kvm_vgic_unset_forwarding(irqfd->kvm, prod->irq,
>>>>> -				  irqfd->gsi + VGIC_NR_PRIVATE_IRQS);
>>>>> +	case IRQ_BYPASS_VFIO_PCI_MSI:
>>>>> +		kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
>>>>> +					     &irqfd->irq_entry);
>>>>> +		break;
>>>>> +	}
>>>>>  }
>>>>>  
>>>>>  void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
>>>>> diff --git a/virt/kvm/arm/vgic/vgic-v4.c b/virt/kvm/arm/vgic/vgic-v4.c
>>>>> index 207e1fda0dcd..338c86c5159f 100644
>>>>> --- a/virt/kvm/arm/vgic/vgic-v4.c
>>>>> +++ b/virt/kvm/arm/vgic/vgic-v4.c
>>>>> @@ -72,3 +72,106 @@ void vgic_v4_teardown(struct kvm *kvm)
>>>>>  	its_vm->nr_vpes = 0;
>>>>>  	its_vm->vpes = NULL;
>>>>>  }
>>>>> +
>>>>> +static struct vgic_its *vgic_get_its(struct kvm *kvm,
>>>>> +				     struct kvm_kernel_irq_routing_entry *irq_entry)
>>>>> +{
>>>>> +	struct kvm_msi msi  = (struct kvm_msi) {
>>>>> +		.address_lo	= irq_entry->msi.address_lo,
>>>>> +		.address_hi	= irq_entry->msi.address_hi,
>>>>> +		.data		= irq_entry->msi.data,
>>>>> +		.flags		= irq_entry->msi.flags,
>>>>> +		.devid		= irq_entry->msi.devid,
>>>>> +	};
>>>>> +
>>>>> +	/*
>>>>> +	 * Get a reference on the LPI. If NULL, this is not a valid
>>>>> +	 * translation for any of our vITSs.
>>>>> +	 */
>>>>> +	return vgic_msi_to_its(kvm, &msi);
>>>>> +}
>>>>> +
>>>>> +int kvm_vgic_v4_set_forwarding(struct kvm *kvm, int virq,
>>>>> +			       struct kvm_kernel_irq_routing_entry *irq_entry)
>>>>> +{
>>>>> +	struct vgic_its *its;
>>>>> +	struct vgic_irq *irq;
>>>>> +	struct its_vlpi_map map;
>>>>> +	int ret;
>>>>> +
>>>>> +	if (!vgic_is_v4_capable(kvm))
>>>>> +		return 0;
>>>>> +
>>>>> +	/*
>>>>> +	 * Get the ITS, and escape early on error (not a valid
>>>>> +	 * doorbell for any of our vITSs).
>>>>> +	 */
>>>>> +	its = vgic_get_its(kvm, irq_entry);
>>>>> +	if (IS_ERR(its))
>>>>> +		return 0;
>>>>> +
>>>>> +	mutex_lock(&its->its_lock);
>>>>> +
>>>>> +	/* Perform then actual DevID/EventID -> LPI translation. */
>>>>> +	ret = vgic_its_resolve_lpi(kvm, its, irq_entry->msi.devid,
>>>>> +				   irq_entry->msi.data, &irq);
>>>>> +	if (ret)
>>>>> +		goto out;
>>>>> +
>>>>> +	/*
>>>>> +	 * Emit the mapping request. If it fails, the ITS probably
>>>>> +	 * isn't v4 compatible, so let's silently bail out. Holding
>>>>> +	 * the ITS lock should ensure that nothing can modify the
>>>>> +	 * target vcpu.
>>>>> +	 */
>>>>> +	map = (struct its_vlpi_map) {
>>>>> +		.vm		= &kvm->arch.vgic.its_vm,
>>>>> +		.vintid		= irq->intid,
>>>>> +		.db_enabled	= true,
>>>>> +		.vpe_idx	= irq->target_vcpu->vcpu_id,
>>>>> +	};
>>>>> +
>>>>> +	if (its_map_vlpi(virq, &map))
>>>>> +		goto out;
>>>>
>>>> This seems to be able to return things like -ENOMEM, whould we really
>>>> not report this back to the caller in any way?
>>>
>>>
>>> That's a good question.
>>>
>>> If we return -ENOMEM, we'll probably end-up returning an error to
>>> userspace (as a result of the VFIO ioctl), which will in turn probably
>>> terminate the guest (I'm guessing, I haven't actually looked at what
>>> userspace does).
>>>
>>> If we don't return the error, then we have a chance to keep the guest
>>> running by sticking to software injection.
>> I have not read the whole stuff yet but userspace is not aware of this
>> negotiation. Everything happens under the hood in kernel, see
>> virt/lib/irqbypass.c __connect(): if add_producer() fails
>> prod->del_consumer() is called and we should return to the not optimized
>> injection.
> 
> Ah, fair enough. I guess del_consumer() does nothing on PCI?

Correct. For vfio-platform, it is switching back to automasked handler
but if you haven't implemented anything specific on vfio side in this
series, it is not even implemented.

Thanks

Eric
> 
> Thanks,
> 
> 	M.
>