[PATCH v2 1/2] virtio: support unlocked queue poll

[PATCH v2 1/2] virtio: support unlocked queue poll

[Michael S. Tsirkin]

This adds a way to check ring empty state after enable_cb outside any
locks. Will be used by virtio_net.

Note: there's room for more optimization: caller is likely to have a
memory barrier already, which means we might be able to get rid of a
barrier here.  Deferring this optimization until we do some
benchmarking.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
---
 drivers/virtio/virtio_ring.c | 56 ++++++++++++++++++++++++++++++++++----------
 include/linux/virtio.h       |  4 ++++
 2 files changed, 48 insertions(+), 12 deletions(-)

diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
index 5217baf..37d58f8 100644
--- a/drivers/virtio/virtio_ring.c
+++ b/drivers/virtio/virtio_ring.c
@@ -607,19 +607,21 @@ void virtqueue_disable_cb(struct virtqueue *_vq)
 EXPORT_SYMBOL_GPL(virtqueue_disable_cb);
 
 /**
- * virtqueue_enable_cb - restart callbacks after disable_cb.
+ * virtqueue_enable_cb_prepare - restart callbacks after disable_cb
  * @vq: the struct virtqueue we're talking about.
  *
- * This re-enables callbacks; it returns "false" if there are pending
- * buffers in the queue, to detect a possible race between the driver
- * checking for more work, and enabling callbacks.
+ * This re-enables callbacks; it returns current queue state
+ * in an opaque unsigned value. This value should be later tested by
+ * virtqueue_poll, to detect a possible race between the driver checking for
+ * more work, and enabling callbacks.
  *
  * Caller must ensure we don't call this with other virtqueue
  * operations at the same time (except where noted).
  */
-bool virtqueue_enable_cb(struct virtqueue *_vq)
+unsigned virtqueue_enable_cb_prepare(struct virtqueue *_vq)
 {
 	struct vring_virtqueue *vq = to_vvq(_vq);
+	u16 last_used_idx;
 
 	START_USE(vq);
 
@@ -629,15 +631,45 @@ bool virtqueue_enable_cb(struct virtqueue *_vq)
 	 * either clear the flags bit or point the event index at the next
 	 * entry. Always do both to keep code simple. */
 	vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
-	vring_used_event(&vq->vring) = vq->last_used_idx;
+	vring_used_event(&vq->vring) = last_used_idx = vq->last_used_idx;
+	END_USE(vq);
+	return last_used_idx;
+}
+EXPORT_SYMBOL_GPL(virtqueue_enable_cb_prepare);
+
+/**
+ * virtqueue_poll - query pending used buffers
+ * @vq: the struct virtqueue we're talking about.
+ * @last_used_idx: virtqueue state (from call to virtqueue_enable_cb_prepare).
+ *
+ * Returns "true" if there are pending used buffers in the queue.
+ *
+ * This does not need to be serialized.
+ */
+bool virtqueue_poll(struct virtqueue *_vq, unsigned last_used_idx)
+{
+	struct vring_virtqueue *vq = to_vvq(_vq);
+
 	virtio_mb(vq->weak_barriers);
-	if (unlikely(more_used(vq))) {
-		END_USE(vq);
-		return false;
-	}
+	return (u16)last_used_idx != vq->vring.used->idx;
+}
+EXPORT_SYMBOL_GPL(virtqueue_poll);
 
-	END_USE(vq);
-	return true;
+/**
+ * virtqueue_enable_cb - restart callbacks after disable_cb.
+ * @vq: the struct virtqueue we're talking about.
+ *
+ * This re-enables callbacks; it returns "false" if there are pending
+ * buffers in the queue, to detect a possible race between the driver
+ * checking for more work, and enabling callbacks.
+ *
+ * Caller must ensure we don't call this with other virtqueue
+ * operations at the same time (except where noted).
+ */
+bool virtqueue_enable_cb(struct virtqueue *_vq)
+{
+	unsigned last_used_idx = virtqueue_enable_cb_prepare(_vq);
+	return !virtqueue_poll(_vq, last_used_idx);
 }
 EXPORT_SYMBOL_GPL(virtqueue_enable_cb);
 
diff --git a/include/linux/virtio.h b/include/linux/virtio.h
index 9ff8645..72398ee 100644
--- a/include/linux/virtio.h
+++ b/include/linux/virtio.h
@@ -70,6 +70,10 @@ void virtqueue_disable_cb(struct virtqueue *vq);
 
 bool virtqueue_enable_cb(struct virtqueue *vq);
 
+unsigned virtqueue_enable_cb_prepare(struct virtqueue *vq);
+
+bool virtqueue_poll(struct virtqueue *vq, unsigned);
+
 bool virtqueue_enable_cb_delayed(struct virtqueue *vq);
 
 void *virtqueue_detach_unused_buf(struct virtqueue *vq);
-- 
MST

[PATCH v2 2/2] virtio_net: fix race in RX VQ processing

[Michael S. Tsirkin]

virtio net called virtqueue_enable_cq on RX path after napi_complete, so
with NAPI_STATE_SCHED clear - outside the implicit napi lock.
This violates the requirement to synchronize virtqueue_enable_cq wrt
virtqueue_add_buf.  In particular, used event can move backwards,
causing us to lose interrupts.
In a debug build, this can trigger panic within START_USE.

Jason Wang reports that he can trigger the races artificially,
by adding udelay() in virtqueue_enable_cb() after virtio_mb().

However, we must call napi_complete to clear NAPI_STATE_SCHED before
polling the virtqueue for used buffers, otherwise napi_schedule_prep in
a callback will fail, causing us to lose RX events.

To fix, call virtqueue_enable_cb_prepare with NAPI_STATE_SCHED
set (under napi lock), later call virtqueue_poll with
NAPI_STATE_SCHED clear (outside the lock).

Reported-by: Jason Wang <jasowang@redhat.com>
Tested-by: Jason Wang <jasowang@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
---
 drivers/net/virtio_net.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 5305bd1..27f79dd 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -605,7 +605,7 @@ static int virtnet_poll(struct napi_struct *napi, int budget)
 		container_of(napi, struct receive_queue, napi);
 	struct virtnet_info *vi = rq->vq->vdev->priv;
 	void *buf;
-	unsigned int len, received = 0;
+	unsigned int r, len, received = 0;
 
 again:
 	while (received < budget &&
@@ -622,8 +622,9 @@ again:
 
 	/* Out of packets? */
 	if (received < budget) {
+		r = virtqueue_enable_cb_prepare(rq->vq);
 		napi_complete(napi);
-		if (unlikely(!virtqueue_enable_cb(rq->vq)) &&
+		if (unlikely(virtqueue_poll(rq->vq, r)) &&
 		    napi_schedule_prep(napi)) {
 			virtqueue_disable_cb(rq->vq);
 			__napi_schedule(napi);
-- 
MST