Re: [PATCH v6 4/5] intel_iommu: allow Extended Interrupt Mode when using userspace APIC

[Bui Quang Minh <minhquangbui99@gmail.com>]

On 7/17/23 17:47, Joao Martins wrote:
> +Peter, +Jason (intel-iommu maintainer/reviewer)
> 
> On 15/07/2023 16:22, Bui Quang Minh wrote:
>> As userspace APIC now supports x2APIC, intel interrupt remapping
>> hardware can be set to EIM mode when userspace local APIC is used.
>>
>> Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
>> Signed-off-by: Bui Quang Minh <minhquangbui99@gmail.com>
>> ---
>>   hw/i386/intel_iommu.c | 11 -----------
>>   1 file changed, 11 deletions(-)
>>
>> diff --git a/hw/i386/intel_iommu.c b/hw/i386/intel_iommu.c
>> index dcc334060c..5e576f6059 100644
>> --- a/hw/i386/intel_iommu.c
>> +++ b/hw/i386/intel_iommu.c
>> @@ -4043,17 +4043,6 @@ static bool vtd_decide_config(IntelIOMMUState *s, Error **errp)
>>                         && x86_iommu_ir_supported(x86_iommu) ?
>>                                                 ON_OFF_AUTO_ON : ON_OFF_AUTO_OFF;
>>       }
>> -    if (s->intr_eim == ON_OFF_AUTO_ON && !s->buggy_eim) {
>> -        if (!kvm_irqchip_is_split()) {
>> -            error_setg(errp, "eim=on requires accel=kvm,kernel-irqchip=split");
>> -            return false;
>> -        }
>> -        if (!kvm_enable_x2apic()) {
>> -            error_setg(errp, "eim=on requires support on the KVM side"
>> -                             "(X2APIC_API, first shipped in v4.7)");
>> -            return false;
>> -        }
>> -    }
>>   
> Given commit 20ca47429e ('Revert "intel_iommu: Fix irqchip / X2APIC
> configuration checks"'), won't we regress behaviour again  for the accel=kvm
> case by dropping the kvm_enable_x2apic() call here?
> 
> Perhaps if we support userspace APIC with TCG the check just needs to be redone
> to instead avoid always requiring kvm e.g.:
> 
> if (kvm_irqchip_in_kernel()) {
>      error_setg(errp, "eim=on requires accel=kvm,kernel-irqchip=split"
>                 "(X2APIC_API, first shipped in v4.7)");
> }
> 
> if (kvm_irqchip_is_split() && !kvm_enable_x2apic()) {
>      error_setg(errp, "eim=on requires support on the KVM side"
>                 "(X2APIC_API, first shipped in v4.7)");
>      return false;
> }

Thank you for your review. I think the check for kvm_irqchip_in_kernel() 
is not correct, AFAIK, kvm_irqchip_is_split() == true also means 
kvm_irqchip_in_kernel() == true on x86. To check if kernel-irqchip = on, 
we need to do something like in x86_iommu_realize

     bool irq_all_kernel = kvm_irqchip_in_kernel() && 
!kvm_irqchip_is_split();

The original check for !kvm_irqchip_is_split means emulated/userspace 
APIC. It's because to reach that check x86_iommu_ir_supported(...) == 
true and x86_iommu_ir_supported(...) == true is not supported when 
kernel-irqchip = on (there is a check for this in x86_iommu_realize)

So I think we need to change the check to

     if (s->intr_eim == ON_OFF_AUTO_ON && !s->buggy_eim) {
         if (kvm_irqchip_is_split() && !kvm_enable_x2apic()) {
             error_setg(errp, "eim=on requires support on the KVM side"
                              "(X2APIC_API, first shipped in v4.7)");
             return false;
         }
     }

Is it OK?

Thanks,
Quang Minh.