[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [PATCH for-5.0] xive/kvm: Trigger interrupts from userspace
From: |
Greg Kurz |
Subject: |
Re: [PATCH for-5.0] xive/kvm: Trigger interrupts from userspace |
Date: |
Tue, 19 Nov 2019 09:53:31 +0100 |
On Tue, 19 Nov 2019 09:15:52 +0100
Cédric Le Goater <address@hidden> wrote:
> On 19/11/2019 01:47, David Gibson wrote:
> > On Mon, Nov 18, 2019 at 04:37:16PM +0100, Cédric Le Goater wrote:
> >> On 18/11/2019 16:12, Greg Kurz wrote:
> >>> When using the XIVE KVM device, the trigger page is directly accessible
> >>> in QEMU. Unlike with XICS, no need to ask KVM to fire the interrupt. A
> >>> simple store on the trigger page does the job.
> >>>
> >>> Just call xive_esb_trigger().
> >>
> >> Yes but the KVM XIVE device does a few other checks.
> >>
> >> It checks that the interrupt was correctly initialized at the KVM device
> >> level. We should be fine in QEMU which has similar checks.
> >>
> >> It caches the LSI assertion level. We should be fine also because it is
> >> useless in KVM when using the XIVE native exploitation mode.
> >>
> >> It checks it is not a passthru interrupt. Any idea on how to check this
> >> condition under QEMU ?
> >>
> >>> This may improve performance of emulated devices that go through
> >>> qemu_set_irq(), eg. virtio devices created with ioeventfd=off or
> >>> configured by the guest to use LSI interrupts, which aren't really
> >>> recommended setups.
> >>
> >> LGTM.
> >
> > Ok, between the comments above and this, I'm not sure if this is ready
> > to merge or not.
>
> I think it is.
>
> With this change, we are loosing a check on passthrough interrupts but
> I am not sure how critical this is given that QEMU can anyhow bypass
> KVM and trigger the interrupt using a store on the ESB page.
>
True. Thinking a bit more about this: nothing prevents such a store to
be the result of a bug somewhere else in QEMU, eg. some dangling pointer
with the same value, in a much easier way than doing the KVM ioctl. Is
it a concern we should take into account ?
> >> Any figures to share ?
>
> I am torturing Greg to have numbers :) but he resisted well.
>
Maybe a _liquid_ bribe or two can be convincing enough :-)
> >> C.
> >>
> >>> Signed-off-by: Greg Kurz <address@hidden>
>
> Let's move on.
>
> Reviewed-by: Cédric Le Goater <address@hidden>
>
> C.
>
> >>> ---
> >>> hw/intc/spapr_xive_kvm.c | 16 ++--------------
> >>> 1 file changed, 2 insertions(+), 14 deletions(-)
> >>>
> >>> diff --git a/hw/intc/spapr_xive_kvm.c b/hw/intc/spapr_xive_kvm.c
> >>> index 08012ac7cd76..69e73552f1ef 100644
> >>> --- a/hw/intc/spapr_xive_kvm.c
> >>> +++ b/hw/intc/spapr_xive_kvm.c
> >>> @@ -354,32 +354,20 @@ static void kvmppc_xive_source_get_state(XiveSource
> >>> *xsrc)
> >>> void kvmppc_xive_source_set_irq(void *opaque, int srcno, int val)
> >>> {
> >>> XiveSource *xsrc = opaque;
> >>> - SpaprXive *xive = SPAPR_XIVE(xsrc->xive);
> >>> - struct kvm_irq_level args;
> >>> - int rc;
> >>> -
> >>> - /* The KVM XIVE device should be in use */
> >>> - assert(xive->fd != -1);
> >>>
> >>> - args.irq = srcno;
> >>> if (!xive_source_irq_is_lsi(xsrc, srcno)) {
> >>> if (!val) {
> >>> return;
> >>> }
> >>> - args.level = KVM_INTERRUPT_SET;
> >>> } else {
> >>> if (val) {
> >>> xsrc->status[srcno] |= XIVE_STATUS_ASSERTED;
> >>> - args.level = KVM_INTERRUPT_SET_LEVEL;
> >>> } else {
> >>> xsrc->status[srcno] &= ~XIVE_STATUS_ASSERTED;
> >>> - args.level = KVM_INTERRUPT_UNSET;
> >>> }
> >>> }
> >>> - rc = kvm_vm_ioctl(kvm_state, KVM_IRQ_LINE, &args);
> >>> - if (rc < 0) {
> >>> - error_report("XIVE: kvm_irq_line() failed : %s",
> >>> strerror(errno));
> >>> - }
> >>> +
> >>> + xive_esb_trigger(xsrc, srcno);
> >>> }
> >>
> >>
> >
>