On Wed, 2012-08-22 at 03:41 +0300, Michael S. Tsirkin wrote:
> On Tue, Aug 21, 2012 at 03:14:54PM -0600, Alex Williamson wrote:
> > On Tue, 2012-08-21 at 23:41 +0300, Michael S. Tsirkin wrote:
> > > On Tue, Aug 21, 2012 at 02:06:19PM -0600, Alex Williamson wrote:
> > > > On Tue, 2012-08-21 at 22:58 +0300, Michael S. Tsirkin wrote:
> > > > > On Tue, Aug 21, 2012 at 01:29:06PM -0600, Alex Williamson wrote:
> > > > > > KVM_IRQFD currently uses the reserved KVM_USERSPACE_IRQ_SOURCE_ID
> > > > > > which is also shared with userspace injection methods like
> > > > > > KVM_IRQ_LINE. This can cause a conflict if an irqfd triggers on
> > > > > > a GSI asserted through KVM_IRQ_LINE.
> > > > >
> > > > > What kind of conflict do you envision? Pls note level interrupts are
> > > > > unsupported ATM.
> > > >
> > > > If KVM_IRQ_LINE asserts a level interrupt and KVM_IRQFD triggers on the
> > > > same GSI then the pin is no longer asserted as userspace thinks it is.
> > > > Do we just chalk this up to userspace error?
> > >
> > > Yes: using a level GSI with current irqfd is a userspace error
> > > because you can lose interrupts anyway.
> > >
> > > Are edge GSIs affected?
> >
> > I wouldn't think so.
>
> No? If userspace does
>
> . set line to 1
> . trigger irqfd
> . set line to 1
> . trigger irqfd
> . set line to 1
> . trigger irqfd
> . set line to 1
>
> it gets 4 interrupts now
>
> With your patch it will get 1, right?
>
> > > > > > Move irqfd to it's own reserved IRQ source ID. Add a capability for
> > > > > > userspace to test for this fix.
> > > > > >
> > > > > > Signed-off-by: Alex Williamson <alex.williamson@xxxxxxxxxx>
> > > > > > ---
> > > > > >
> > > > > > arch/x86/kvm/x86.c | 3 +++
> > > > > > include/linux/kvm.h | 1 +
> > > > > > include/linux/kvm_host.h | 1 +
> > > > > > virt/kvm/eventfd.c | 6 +++---
> > > > > > 4 files changed, 8 insertions(+), 3 deletions(-)
> > > > > >
> > > > > > diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> > > > > > index 42bce48..cd98673 100644
> > > > > > --- a/arch/x86/kvm/x86.c
> > > > > > +++ b/arch/x86/kvm/x86.c
> > > > > > @@ -2174,6 +2174,7 @@ int kvm_dev_ioctl_check_extension(long ext)
> > > > > > case KVM_CAP_GET_TSC_KHZ:
> > > > > > case KVM_CAP_PCI_2_3:
> > > > > > case KVM_CAP_KVMCLOCK_CTRL:
> > > > > > + case KVM_CAP_IRQFD_IRQ_SOURCE_ID:
> > > > > > r = 1;
> > > > > > break;
> > > > > > case KVM_CAP_COALESCED_MMIO:
> > > > > > @@ -6258,6 +6259,8 @@ int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
> > > > > >
> > > > > > /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
> > > > > > set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
> > > > > > + /* Reserve bit 1 of irq_sources_bitmap for irqfd irq source */
> > > > > > + set_bit(KVM_IRQFD_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
> > > > > >
> > > > > > raw_spin_lock_init(&kvm->arch.tsc_write_lock);
> > > > > >
> > > > > > diff --git a/include/linux/kvm.h b/include/linux/kvm.h
> > > > > > index 2ce09aa..ae66b9c 100644
> > > > > > --- a/include/linux/kvm.h
> > > > > > +++ b/include/linux/kvm.h
> > > > > > @@ -618,6 +618,7 @@ struct kvm_ppc_smmu_info {
> > > > > > #define KVM_CAP_PPC_GET_SMMU_INFO 78
> > > > > > #define KVM_CAP_S390_COW 79
> > > > > > #define KVM_CAP_PPC_ALLOC_HTAB 80
> > > > > > +#define KVM_CAP_IRQFD_IRQ_SOURCE_ID 81
> > > > > >
> > > > > > #ifdef KVM_CAP_IRQ_ROUTING
> > > > > >
> > > > > > diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
> > > > > > index b70b48b..b763230 100644
> > > > > > --- a/include/linux/kvm_host.h
> > > > > > +++ b/include/linux/kvm_host.h
> > > > > > @@ -71,6 +71,7 @@
> > > > > > #define KVM_REQ_PMI 17
> > > > > >
> > > > > > #define KVM_USERSPACE_IRQ_SOURCE_ID 0
> > > > > > +#define KVM_IRQFD_IRQ_SOURCE_ID 1
> > > > > >
> > > > > > struct kvm;
> > > > > > struct kvm_vcpu;
> > > > >
> > > > > Above looks fine but I'm not sure why is the below needed.
> > > > > This changes irqfd behaviour for edge GSIs slightly
> > > > > in a userspace-visible way. Maybe make it a separate patch
> > > > > so it can be considered on merits?
> > > >
> > > > Hmm, the above does nothing without the below.
> > >
> > > Yes. But you can use the above with the new irqfds you are adding.
> >
> > Nope, racy.
> >
> > > > I thought I was just
> > > > implementing your idea that IRQFDs should all share a single IRQ source
> > > > ID...
> > >
> > > Sorry I only meant for level irqfds. You are changing edge here.
> >
> > Ok, I misunderstood then.
> >
> > > > why is that no longer a good idea? Thanks,
> > > >
> > > > Alex
> > >
> > > Maybe it is a good idea. I am just asking for the motivation.
> >
> > I assumed you were pointing out the level vs edge interaction. If we
> > call that a userspace bug, I can just drop this. Thanks,
> >
> > Alex
>
> level is userspace bug I think :)
Dropped. Thanks,
Alex
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
