Skip to content
Open
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
198 changes: 167 additions & 31 deletions virtio-queue/src/queue/verification.rs
Original file line number Diff line number Diff line change
Expand Up @@ -459,6 +459,17 @@ impl kani::Arbitrary for ProofContext {
}
}

/// Helper function to determine if `point` is in the wrapping range `[start, stop)`.
fn pred_in_wrapping_range(start: Wrapping<u16>, stop: Wrapping<u16>, point: Wrapping<u16>) -> bool {
// Check if point is in the range [start, stop) in a wrapping manner.
if start <= stop {
start <= point && point < stop
} else {
// If start > stop, we wrap around the maximum value of u16.
point >= start || point < stop
}
}

/// # Specification (VirtIO 1.3, Section 2.7.7.2: "Device Requirements: Used Buffer Notification Suppression")
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The commit 4a8b048 description is not really clear to me.

It seems now this function don't test only the suppression, so should we update this documentation?

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think the proof is testing both w/wo suppression. I think the comment should only mention about the notification.

///
/// Section 2.7.7.2 deals with device-to-driver notification suppression. It
Expand All @@ -468,46 +479,35 @@ impl kani::Arbitrary for ProofContext {
/// specific number of descriptors has been processed. This is done by the
/// driver defining a "used_event" index, which tells the device "please do not
/// notify me until used.ring[used_event] has been written to by you".
///
/// This proof checks:
// - If event_idx is not enabled, a notification is always needed.
// - If event_idx is enabled, a notification is needed if the used_event index
// is in the wrapping range [next_used - num_added, next_used).
#[kani::proof]
// There are no loops anywhere, but kani really enjoys getting stuck in
// std::ptr::drop_in_place. This is a compiler intrinsic that has a "dummy"
// implementation in stdlib that just recursively calls itself. Kani will
// generally unwind this recursion infinitely.
#[kani::unwind(0)]
fn verify_device_notification_suppression() {
let ProofContext {
mut queue,
memory: mem,
} = kani::any();

let num_added_old = queue.num_added.0;
let needs_notification = queue.needs_notification(&mem);
fn verify_device_to_driver_notification() {
let ProofContext { mut queue, memory } = kani::any();

// event_idx_enabled equivalent to VIRTIO_F_EVENT_IDX negotiated
// If the event_idx_enabled is false, we always need a notification
if !queue.event_idx_enabled {
// The specification here says
// After the device writes a descriptor index into the used ring:
// – If flags is 1, the device SHOULD NOT send a notification.
// – If flags is 0, the device MUST send a notification
// flags is the first field in the avail_ring_address, which we completely ignore. We
// always send a notification, and as there only is a SHOULD NOT, that is okay
assert!(needs_notification.unwrap());
assert!(queue.needs_notification(&memory).unwrap());
} else {
// next_used - 1 is where the previous descriptor was placed
if Wrapping(queue.used_event(&mem, Ordering::Relaxed).unwrap())
== std::num::Wrapping(queue.next_used - Wrapping(1))
&& num_added_old > 0
{
// If the idx field in the used ring (which determined where that descriptor index
// was placed) was equal to used_event, the device MUST send a
// notification.
assert!(needs_notification.unwrap());

kani::cover!();
}

// The other case is handled by a "SHOULD NOT send a notification" in the spec.
// So we do not care
// If the event_idx_enabled is true, we only need a notification if
// the used_event is in the range of [next_used - num_added, next_used)
let used_event = queue.used_event(&memory, Ordering::Relaxed).unwrap();
let current_used = queue.next_used;
let old = current_used - queue.num_added;
let should_notify = pred_in_wrapping_range(old, current_used, used_event);
assert_eq!(
queue.needs_notification(&memory).unwrap(), /* actual */
should_notify, /* expected */
);
assert_eq!(queue.num_added.0, 0);
}
}

Expand Down Expand Up @@ -565,3 +565,139 @@ fn verify_add_used() {
kani::cover!();
}
}

/// This proof checks that after setting the `next_used` field of the queue
/// using `set_next_used(x)`, reading back the value of `next_used` returns the
/// same value `x`.
#[kani::proof]
#[kani::unwind(0)]
fn verify_used_ring_avail_event() {
let ProofContext {
mut queue,
memory: _,
} = kani::any();
let x = kani::any();
queue.set_next_used(x);
assert_eq!(x, queue.next_used.0);
}

/// # Specification (VirtIO 1.3, Section 2.7.6.1: "Driver Requirements: The Virtqueue Available Ring")
///
/// This proof checks that:
/// - If there are pending entries in the avail ring (avail_idx != next_avail),
/// `enable_notification` returns true.
/// - If there are no pending entries (avail_idx == next_avail), it returns false.
Copy link
Contributor

@MatiasVara MatiasVara Sep 24, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think VirtIO 1.3, Section 2.7.6.1 states that avail_idx is always >= than next_avail. Although, it is correct that enable_notification() checks only if they are different.

/// This matches the monotonicity property of the avail ring in VirtIO 1.3 Section 2.7.6.1.
#[kani::proof]
#[kani::unwind(0)]
fn verify_enable_driver_to_device_notification() {
let ProofContext { mut queue, memory } = kani::any();

// The enable_notification method sets notification to true and returns
// - true, if there are pending entries in the `idx` field of the
// avail ring
// - false, if there are no pending entries in the `idx` field of the
// avail ring The check for pending entries is done by comparing the
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
// avail ring The check for pending entries is done by comparing the
// avail ring. The check for pending entries is done by comparing the

// current `avail_idx` with the `next_avail` field of the queue. If they
// are different, there are pending entries, otherwise there are no
// pending entries. The equality check is a consequence of the
// monotonicity property of `idx` (2.7.6.1) that it cannot be decreased
// by the driver.
if queue.enable_notification(&memory).unwrap() {
assert_ne!(
queue.avail_idx(&memory, Ordering::Relaxed).unwrap(),
queue.next_avail
);
} else {
assert_eq!(
queue.avail_idx(&memory, Ordering::Relaxed).unwrap(),
queue.next_avail
);
}
}

// Helper method that reads `val` from the `avail_event` field of the used ring, using
// the provided ordering. Takes used_ring address and queue size directly.
fn get_avail_event<M: GuestMemory>(
used_ring_addr: GuestAddress,
queue_size: u16,
mem: &M,
order: Ordering,
) -> Result<u16, Error> {
// This can not overflow an u64 since it is working with relatively small numbers compar
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
// This can not overflow an u64 since it is working with relatively small numbers compar
// This can not overflow an u64 since it is working with relatively small numbers compared

// to u64::MAX.
let avail_event_offset =
VIRTQ_USED_RING_HEADER_SIZE + VIRTQ_USED_ELEMENT_SIZE * u64::from(queue_size);
let addr = used_ring_addr
.checked_add(avail_event_offset)
.ok_or(Error::AddressOverflow)?;

mem.load::<u16>(addr, order).map_err(Error::GuestMemory)
}

// Get the value of the `flags` field of the used ring, applying the specified ordering.
fn get_used_flags<M: GuestMemory>(queue: &Queue, mem: &M, order: Ordering) -> Result<u16, Error> {
mem.load::<u16>(queue.used_ring, order)
.map(u16::from_le)
.map_err(Error::GuestMemory)
}
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
}
}

/// # Specification (VirtIO 1.3, Section 2.7.7: "Used Buffer Notification Suppression")
///
/// This proof checks:
/// - If event_idx is not enabled, `set_notification(false)` sets used.flags to 1 (NO_NOTIFY).
/// - If event_idx is enabled, the call is a no-op.
#[kani::proof]
#[kani::unwind(0)]
fn verify_set_notification_true() {
let ProofContext { mut queue, memory } = kani::any();
if queue
.set_notification(&memory, true /* enable notification */)
.is_ok()
{
if queue.event_idx_enabled {
// Since VIRTIO_F_EVENT_IDX is negotiated, we make sure that set_notification
// has updated the used.avail_event field with the tail position of
// the avail ring.
let used_ring_addr = queue.used_ring;
let queue_size = queue.size();
kani::cover!();
assert_eq!(
get_avail_event(used_ring_addr, queue_size, &memory, Ordering::Relaxed).unwrap(),
queue.next_avail.0
);
} else {
// If VIRTIO_F_EVENT_IDX is not negotiated, we make sure that the
// used.flags field is set to 0, meaning that the driver should not
// send notifications to the device.
kani::cover!();
assert_eq!(
get_used_flags(&queue, &memory, Ordering::Relaxed).unwrap(),
0
);
}
}
}

/// # Specification (VirtIO 1.3, Section 2.7.7: "Used Buffer Notification Suppression")
///
/// This proof checks:
/// - If event_idx is not enabled, `set_notification(false)` sets used.flags to 1 (NO_NOTIFY).
/// - If event_idx is enabled, the call is a no-op.
#[kani::proof]
#[kani::unwind(0)]
fn verify_set_notification_false() {
let ProofContext { mut queue, memory } = kani::any();
let result = queue.set_notification(&memory, false /* disable notification */);
Copy link
Contributor

@MatiasVara MatiasVara Sep 24, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can't we use kany::any to cover the potential values of the second parameter of set_notification()? This would prevent having two proofs.

if !queue.event_idx_enabled {
// Check for Sec 2.7.10
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sec, should be Spec right?

Also the test doc talks about Specification (VirtIO 1.3, Section 2.7.7: "Used Buffer Notification Suppression") what 2.7.10 is? Is related to virtio 1.3?

Please be more verbose with that references.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think here you could add that Section 2.7.10 of the Virtio spec explains how the device can indicates to the driver that notifications are not required when adding buffers into the avail right.

assert_eq!(
get_used_flags(&queue, &memory, Ordering::Relaxed).unwrap(),
1
);
// don't check Ok() result since that is a property of the
// underlying mem system and out of scope. E.g., it is stubbed for
// this proof and we always expect it to succeed.
} else {
assert!(result.is_ok());
}
}