vhost: add non-blocking API for posting interrupt

Vhost-user library locks all VQ's access lock when processing
vring based messages, such as SET_VRING_KICK and SET_VRING_CALL,
and the data processing thread may already be started, e.g: SPDK
vhost-blk and vhost-scsi will start the data processing thread
when one vring is ready, then deadlock may happen when SPDK is
posting interrupts to VM.  Here, we add a new API which allows
caller to try again later for this case.

Bugzilla ID: 1015
Fixes: c573699 ("vhost: fix missing virtqueue lock protection")
Cc: stable@dpdk.org

Signed-off-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>

doc/guides/prog_guide/vhost_lib.rst

@@ -297,6 +297,12 @@ The following is an overview of some key Vhost API functions:
   Clear in-flight packets which are submitted to async channel in vhost async data
   path. Completed packets are returned to applications through ``pkts``.
 
+* ``rte_vhost_vring_call_nonblock(int vid, uint16_t vring_idx)``
+
+  Notify the guest that used descriptors have been added to the vring. This function
+  will return -EAGAIN when vq's access lock is held by other thread, user should try
+  again later.
+
 * ``rte_vhost_vring_stats_get_names(int vid, uint16_t queue_id, struct rte_vhost_stat_name *names, unsigned int size)``
 
   This function returns the names of the queue statistics. It requires

doc/guides/rel_notes/release_22_11.rst

@@ -188,6 +188,12 @@ New Features
   Added support to unconfigure DMA vChannel that is no longer used
   by the vhost library.
 
+* **Added non-blocking notify API to vhost library.**
+
+  Added ``rte_vhost_vring_call_nonblock`` API to notify the guest that
+  used descriptors have been added to the vring in non-blocking way.
+  User should check the return value of this API and try again if needed.
+
 * **Added support for MACsec in rte_security.**
 
   Added MACsec transform for rte_security session and added new API

lib/vhost/rte_vhost.h

@@ -909,6 +909,21 @@ rte_vhost_clr_inflight_desc_packed(int vid, uint16_t vring_idx,
  */
 int rte_vhost_vring_call(int vid, uint16_t vring_idx);
 
+/**
+ * Notify the guest that used descriptors have been added to the vring.  This
+ * function acts as a memory barrier.  This function will return -EAGAIN when
+ * vq's access lock is held by other thread, user should try again later.
+ *
+ * @param vid
+ *  vhost device ID
+ * @param vring_idx
+ *  vring index
+ * @return
+ *  0 on success, -1 on failure, -EAGAIN for another retry
+ */
+__rte_experimental
+int rte_vhost_vring_call_nonblock(int vid, uint16_t vring_idx);
+
 /**
  * Get vhost RX queue avail count.
  *

lib/vhost/version.map

@@ -97,6 +97,7 @@ EXPERIMENTAL {
 
 	# added in 22.11
 	rte_vhost_async_dma_unconfigure;
+	rte_vhost_vring_call_nonblock;
 };
 
 INTERNAL {

lib/vhost/vhost.c

@@ -1318,6 +1318,36 @@ rte_vhost_vring_call(int vid, uint16_t vring_idx)
 	return 0;
 }
 
+int
+rte_vhost_vring_call_nonblock(int vid, uint16_t vring_idx)
+{
+	struct virtio_net *dev;
+	struct vhost_virtqueue *vq;
+
+	dev = get_device(vid);
+	if (!dev)
+		return -1;
+
+	if (vring_idx >= VHOST_MAX_VRING)
+		return -1;
+
+	vq = dev->virtqueue[vring_idx];
+	if (!vq)
+		return -1;
+
+	if (!rte_spinlock_trylock(&vq->access_lock))
+		return -EAGAIN;
+
+	if (vq_is_packed(dev))
+		vhost_vring_call_packed(dev, vq);
+	else
+		vhost_vring_call_split(dev, vq);
+
+	rte_spinlock_unlock(&vq->access_lock);
+
+	return 0;
+}
+
 uint16_t
 rte_vhost_avail_entries(int vid, uint16_t queue_id)
 {