On Mon, Mar 17, 2025 at 12:43:53PM -0700, Sean Christopherson wrote:
> On Mon, Mar 17, 2025, Yosry Ahmed wrote:
> > On Fri, Mar 14, 2025 at 07:56:15PM -0700, Sean Christopherson wrote:
> > > Add a module param to allow disabling device posted interrupts without
> > > having to sacrifice all of APICv/AVIC, and to also effectively enumerate
> > > to userspace whether or not KVM may be utilizing device posted IRQs.
> > > Disabling device posted interrupts is very desirable for testing, and can
> > > even be desirable for production environments, e.g. if the host kernel
> > > wants to interpose on device interrupts.
> > >
> > > Signed-off-by: Sean Christopherson <seanjc@xxxxxxxxxx>
> > > ---
> > > arch/x86/include/asm/kvm_host.h | 1 +
> > > arch/x86/kvm/svm/avic.c | 3 +--
> > > arch/x86/kvm/vmx/posted_intr.c | 7 +++----
> > > arch/x86/kvm/x86.c | 9 ++++++++-
> > > 4 files changed, 13 insertions(+), 7 deletions(-)
> > >
> > > diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
> > > index d881e7d276b1..bf11c5ee50cb 100644
> > > --- a/arch/x86/include/asm/kvm_host.h
> > > +++ b/arch/x86/include/asm/kvm_host.h
> > > @@ -1922,6 +1922,7 @@ struct kvm_arch_async_pf {
> > > extern u32 __read_mostly kvm_nr_uret_msrs;
> > > extern bool __read_mostly allow_smaller_maxphyaddr;
> > > extern bool __read_mostly enable_apicv;
> > > +extern bool __read_mostly enable_device_posted_irqs;
> > > extern struct kvm_x86_ops kvm_x86_ops;
> > >
> > > #define kvm_x86_call(func) static_call(kvm_x86_##func)
> > > diff --git a/arch/x86/kvm/svm/avic.c b/arch/x86/kvm/svm/avic.c
> > > index 65fd245a9953..e0f519565393 100644
> > > --- a/arch/x86/kvm/svm/avic.c
> > > +++ b/arch/x86/kvm/svm/avic.c
> > > @@ -898,8 +898,7 @@ int avic_pi_update_irte(struct kvm *kvm, unsigned int host_irq,
> > > struct kvm_irq_routing_table *irq_rt;
> > > int idx, ret = 0;
> > >
> > > - if (!kvm_arch_has_assigned_device(kvm) ||
> > > - !irq_remapping_cap(IRQ_POSTING_CAP))
> > > + if (!kvm_arch_has_assigned_device(kvm) || !enable_device_posted_irqs)
> >
> > This function will now also be skipped if enable_apicv is false. Is this
> > always the case here for some reason? Sorry if I missed something
> > obvious.
>
> Working as intended, though I failed to document it. Hrm, but I wasn't expecting
> this to be a functional change. Oh, I know what happened. I had originally
> tacked this on to a big series to clean up the IRTE stuff (spoiler alert), and in
> that series common code checked kvm_arch_has_irq_bypass() (which incorporates
> enable_apicv) before calling pi_update_irte().
>
> I'll prepend a patch or three to do minimal cleanup before introducing the new
> module param.
>
> > > @@ -9772,6 +9776,9 @@ int kvm_x86_vendor_init(struct kvm_x86_init_ops *ops)
> > > if (r != 0)
> > > goto out_mmu_exit;
> > >
> > > + enable_device_posted_irqs = enable_device_posted_irqs && enable_apicv &&
> > > + irq_remapping_cap(IRQ_POSTING_CAP);
> >
> > Maybe this is clearer:
> >
> > enable_device_posted_irqs &= enable_avivc && irq_remapping_cap(IRQ_POSTING_CAP);
>
> I don't have a strong opinion. I went with the "self check" approach purely
> because SVM does so for a few params, e.b.
>
> nrips = nrips && boot_cpu_has(X86_FEATURE_NRIPS);
>
> Anyone else care either way? If not, I'll go with Yosry's suggestion.
I can understand a consistency argument, so I am fine either way too.
The main reason I suggested this is that it took me a second to realize
this is the same thing on both sides of the assignment.
