Re: [PATCH V2 07/12] virtio-pci: harden INTX interrupts

["Michael S. Tsirkin" <mst@redhat.com>]

On Thu, Oct 14, 2021 at 02:32:19PM +0800, Jason Wang wrote:
> On Thu, Oct 14, 2021 at 2:26 PM Michael S. Tsirkin <mst@redhat.com> wrote:
> >
> > On Thu, Oct 14, 2021 at 02:20:17PM +0800, Jason Wang wrote:
> > > On Thu, Oct 14, 2021 at 1:50 PM Michael S. Tsirkin <mst@redhat.com> wrote:
> > > >
> > > > On Thu, Oct 14, 2021 at 10:35:48AM +0800, Jason Wang wrote:
> > > > > On Wed, Oct 13, 2021 at 5:42 PM Michael S. Tsirkin <mst@redhat.com> wrote:
> > > > > >
> > > > > > On Tue, Oct 12, 2021 at 02:52:22PM +0800, Jason Wang wrote:
> > > > > > > This patch tries to make sure the virtio interrupt handler for INTX
> > > > > > > won't be called after a reset and before virtio_device_ready(). We
> > > > > > > can't use IRQF_NO_AUTOEN since we're using shared interrupt
> > > > > > > (IRQF_SHARED). So this patch tracks the INTX enabling status in a new
> > > > > > > intx_soft_enabled variable and toggle it during in
> > > > > > > vp_disable/enable_vectors(). The INTX interrupt handler will check
> > > > > > > intx_soft_enabled before processing the actual interrupt.
> > > > > > >
> > > > > > > Cc: Boqun Feng <boqun.feng@gmail.com>
> > > > > > > Cc: Thomas Gleixner <tglx@linutronix.de>
> > > > > > > Cc: Peter Zijlstra <peterz@infradead.org>
> > > > > > > Cc: Paul E. McKenney <paulmck@kernel.org>
> > > > > > > Signed-off-by: Jason Wang <jasowang@redhat.com>
> > > > > > > ---
> > > > > > >  drivers/virtio/virtio_pci_common.c | 24 ++++++++++++++++++++++--
> > > > > > >  drivers/virtio/virtio_pci_common.h |  1 +
> > > > > > >  2 files changed, 23 insertions(+), 2 deletions(-)
> > > > > > >
> > > > > > > diff --git a/drivers/virtio/virtio_pci_common.c b/drivers/virtio/virtio_pci_common.c
> > > > > > > index 0b9523e6dd39..5ae6a2a4eb77 100644
> > > > > > > --- a/drivers/virtio/virtio_pci_common.c
> > > > > > > +++ b/drivers/virtio/virtio_pci_common.c
> > > > > > > @@ -30,8 +30,16 @@ void vp_disable_vectors(struct virtio_device *vdev)
> > > > > > >       struct virtio_pci_device *vp_dev = to_vp_device(vdev);
> > > > > > >       int i;
> > > > > > >
> > > > > > > -     if (vp_dev->intx_enabled)
> > > > > > > +     if (vp_dev->intx_enabled) {
> > > > > > > +             /*
> > > > > > > +              * The below synchronize() guarantees that any
> > > > > > > +              * interrupt for this line arriving after
> > > > > > > +              * synchronize_irq() has completed is guaranteed to see
> > > > > > > +              * intx_soft_enabled == false.
> > > > > > > +              */
> > > > > > > +             WRITE_ONCE(vp_dev->intx_soft_enabled, false);
> > > > > > >               synchronize_irq(vp_dev->pci_dev->irq);
> > > > > > > +     }
> > > > > > >
> > > > > > >       for (i = 0; i < vp_dev->msix_vectors; ++i)
> > > > > > >               disable_irq(pci_irq_vector(vp_dev->pci_dev, i));
> > > > > > > @@ -43,8 +51,16 @@ void vp_enable_vectors(struct virtio_device *vdev)
> > > > > > >       struct virtio_pci_device *vp_dev = to_vp_device(vdev);
> > > > > > >       int i;
> > > > > > >
> > > > > > > -     if (vp_dev->intx_enabled)
> > > > > > > +     if (vp_dev->intx_enabled) {
> > > > > > > +             disable_irq(vp_dev->pci_dev->irq);
> > > > > > > +             /*
> > > > > > > +              * The above disable_irq() provides TSO ordering and
> > > > > > > +              * as such promotes the below store to store-release.
> > > > > > > +              */
> > > > > > > +             WRITE_ONCE(vp_dev->intx_soft_enabled, true);
> > > > > > > +             enable_irq(vp_dev->pci_dev->irq);
> > > > > > >               return;
> > > > > > > +     }
> > > > > > >
> > > > > > >       for (i = 0; i < vp_dev->msix_vectors; ++i)
> > > > > > >               enable_irq(pci_irq_vector(vp_dev->pci_dev, i));
> > > > > > > @@ -97,6 +113,10 @@ static irqreturn_t vp_interrupt(int irq, void *opaque)
> > > > > > >       struct virtio_pci_device *vp_dev = opaque;
> > > > > > >       u8 isr;
> > > > > > >
> > > > > > > +     /* read intx_soft_enabled before read others */
> > > > > > > +     if (!smp_load_acquire(&vp_dev->intx_soft_enabled))
> > > > > > > +             return IRQ_NONE;
> > > > > > > +
> > > > > > >       /* reading the ISR has the effect of also clearing it so it's very
> > > > > > >        * important to save off the value. */
> > > > > > >       isr = ioread8(vp_dev->isr);
> > > > > >
> > > > > > I don't see why we need this ordering guarantee here.
> > > > > >
> > > > > > synchronize_irq above makes sure no interrupt handler
> > > > > > is in progress.
> > > > >
> > > > > Yes.
> > > > >
> > > > > > the handler itself thus does not need
> > > > > > any specific order, it is ok if intx_soft_enabled is read
> > > > > > after, not before the rest of it.
> > > > >
> > > > > But the interrupt could be raised after synchronize_irq() which may
> > > > > see a false of the intx_soft_enabled.
> > > >
> > > > You mean a "true" value right? false is what we are writing there.
> > >
> > > I meant that we want to not go for stuff like vq->callback after the
> > > synchronize_irq() after setting intx_soft_enabled to false. Otherwise
> > > we may get unexpected results like use after free. Host can craft ISR
> > > in this case.
> > > >
> > > > Are you sure it can happen? I think that synchronize_irq makes the value
> > > > visible on all CPUs running the irq.
> > >
> > > Yes, so the false is visible by vp_interrupt(), we can't do the other
> > > task before we check intx_soft_enabled.
> >
> > But the order does not matter. synchronize_irq will make sure
> > everything is visible.
> 
> Not the thing that happens after synchronize_irq().
> 
> E.g for remove_vq_common():
> 
> static void remove_vq_common(struct virtnet_info *vi)
> {
>         vi->vdev->config->reset(vi->vdev);
> 
>         /* Free unused buffers in both send and recv, if any. */
>         free_unused_bufs(vi);
> 
>         free_receive_bufs(vi);
> 
>         free_receive_page_frags(vi);
> 
>         virtnet_del_vqs(vi);
> }
> 
> The interrupt could be raised by the device after .reset().
> 
> Thanks

That's why your patches set intx_soft_enabled to false within reset.
Then you sync so all other CPUs see the false value.
Then it's ok to proceed with reset.
What does the interrupt handler *do* with the value
does not matter as long as it sees that it is false.

OTOH if you are really worried about spectre type speculative attacks,
that is a different matter, and would force us to stick expensive
barriers around hardware accessible buffers just like we have in
copy_XXX_user. I am not sure this is in scope for TDX, and
certainly out of scope for regular driver ardening.
If yes worth hiding that behind a kernel option.


> >
> > > >
> > > > > In this case we still need the
> > > > > make sure intx_soft_enbled to be read first instead of allowing other
> > > > > operations to be done first, otherwise the intx_soft_enabled is
> > > > > meaningless.
> > > > >
> > > > > Thanks
> > > >
> > > > If intx_soft_enbled were not visible after synchronize_irq then
> > > > it does not matter in which order we read it wrt other values,
> > > > it still wouldn't work right.
> > >
> > > Yes.
> > >
> > > Thanks
> >
> >
> > We are agreed then? No need for a barrier here, READ_ONCE is enough?
> >
> > > >
> > > > > >
> > > > > > Just READ_ONCE should be enough, and we can drop the comment.
> > > > > >
> > > > > >
> > > > > > > diff --git a/drivers/virtio/virtio_pci_common.h b/drivers/virtio/virtio_pci_common.h
> > > > > > > index a235ce9ff6a5..3c06e0f92ee4 100644
> > > > > > > --- a/drivers/virtio/virtio_pci_common.h
> > > > > > > +++ b/drivers/virtio/virtio_pci_common.h
> > > > > > > @@ -64,6 +64,7 @@ struct virtio_pci_device {
> > > > > > >       /* MSI-X support */
> > > > > > >       int msix_enabled;
> > > > > > >       int intx_enabled;
> > > > > > > +     bool intx_soft_enabled;
> > > > > > >       cpumask_var_t *msix_affinity_masks;
> > > > > > >       /* Name strings for interrupts. This size should be enough,
> > > > > > >        * and I'm too lazy to allocate each name separately. */
> > > > > > > --
> > > > > > > 2.25.1
> > > > > >
> > > >
> >