On Thu, Aug 19, 2021 at 1:04 AM Marc Zyngier [off-list ref] wrote:

On Thu, 19 Aug 2021 08:41:19 +0100,
Oliver Upton [off-list ref] wrote:

quoted

On Wed, Aug 18, 2021 at 2:45 PM Ricardo Koller [off-list ref] wrote:

quoted

On Wed, Aug 18, 2021 at 02:34:03PM -0700, Oliver Upton wrote:

quoted

Hi Ricardo,

On Wed, Aug 18, 2021 at 2:32 PM Ricardo Koller [off-list ref] wrote:

quoted

vgic_get_irq(intid) is used all over the vgic code in order to get a
reference to a struct irq. It warns whenever intid is not a valid number
(like when it's a reserved IRQ number). The issue is that this warning
can be triggered from userspace (e.g., KVM_IRQ_LINE for intid 1020).

Drop the WARN call from vgic_get_irq.

Signed-off-by: Ricardo Koller <redacted>
---
 arch/arm64/kvm/vgic/vgic.c | 1 -
 1 file changed, 1 deletion(-)

diff --git a/arch/arm64/kvm/vgic/vgic.c b/arch/arm64/kvm/vgic/vgic.c
index 111bff47e471..81cec508d413 100644
--- a/arch/arm64/kvm/vgic/vgic.c
+++ b/arch/arm64/kvm/vgic/vgic.c

@@ -106,7 +106,6 @@ struct vgic_irq *vgic_get_irq(struct kvm *kvm, struct kvm_vcpu *vcpu,
        if (intid >= VGIC_MIN_LPI)
                return vgic_get_lpi(kvm, intid);

-       WARN(1, "Looking up struct vgic_irq for reserved INTID");

Could we maybe downgrade the message to WARN_ONCE() (to get a stack)
or pr_warn_ratelimited()? I agree it is problematic that userspace can
cause this WARN to fire, but it'd be helpful for debugging too.

Was thinking about that, until I found this in bug.h:

        /*
         * WARN(), WARN_ON(), WARN_ON_ONCE, and so on can be used to report
         * significant kernel issues that need prompt attention if they should ever
         * appear at runtime.
         *
         * Do not use these macros when checking for invalid external inputs
         * (e.g. invalid system call arguments, or invalid data coming from
         * network/devices),

Just in case, KVM_IRQ_LINE returns -EINVAL for an invalid intid (like
1020). I think it's more appropriate for the vmm to log it. What do you
think?

vgic_get_irq() is called in a bunch of other places though, right?
IOW, intid doesn't necessarily come from userspace. In fact, I believe
KVM_IRQ_LINE is the only place we pass a value from userspace to
vgic_get_irq() (don't quote me on that).

Perhaps instead the fix could be to explicitly check that the intid
from userspace is valid and exit early rather than count on
vgic_get_irq() to do the right thing.

vgic_get_irq() is designed to do the right thing. Returning NULL is
the way it reports an error, and this NULL value is already checked at
when directly provided either by the VMM or the guest. If we missed
any of those, that would be what needs addressing.  Obtaining a NULL
pointer is a good way to catch those.

In general, the kernel log is not how we report userspace errors (we
have been there before...). It is slow, noisy, unclear and ultimately
leaks information.

Absolutely. My comments were aimed at calls to vgic_get_irq() where
intid is coming from the kernel, not userspace. That being said,
probably no good reason to buy a full fat WARN() in a function such as
this one.  I'm done waffling on this one liner now :)

Reviewed-by: Oliver Upton <redacted>

If you really want something, then a pr_debug is a
potential tool as it can be dynamically enabled with the right
configuration.

Thanks,

        M.

--
Without deviation from the norm, progress is not possible.