[PATCH 15/17] host: support delayed vq creation

[Mike Christie <michael.christie@oracle.com>]

This allows vq creation to be done when it's first accessed by
userspace. vhost-scsi doesn't know how many queues the user requested
until they are first setup, and we don't want to allocate resources
like the iovecs for 128 vqs when we are only using 1 or 2 most of the
time. In the next pathces, vhost-scsi will also switch to preallocating
cmds per vq instead of per lio session and we don't want to allocate
them for 127 extra vqs if they are not in use.

With this patch when a driver calls vhost_dev_init they pass in the
number of vqs that they know they need and the max they can support.
This patch has all the drivers pass in the same value for both the
initial number of vqs and the max. The next patch will convert scsi.
The other drivers like net/vsock have their vqs hard coded in the
kernel or setup/discovered via other methods like with vdpa.

Signed-off-by: Mike Christie <michael.christie@oracle.com>
---
 drivers/vhost/scsi.c | 48 +++++++++++++++++++++++++-----------------------
 1 file changed, 25 insertions(+), 23 deletions(-)

diff --git a/drivers/vhost/scsi.c b/drivers/vhost/scsi.c
index b348e9c..5d6dc15 100644
--- a/drivers/vhost/scsi.c
+++ b/drivers/vhost/scsi.c
@@ -178,6 +178,7 @@ enum {
 
 struct vhost_scsi_virtqueue {
 	struct vhost_virtqueue vq;
+	struct vhost_scsi *vs;
 	/*
 	 * Reference counting for inflight reqs, used for flush operation. At
 	 * each time, one reference tracks new commands submitted, while we
@@ -192,6 +193,9 @@ struct vhost_scsi_virtqueue {
 	struct vhost_scsi_cmd *scsi_cmds;
 	struct sbitmap scsi_tags;
 	int max_cmds;
+
+	struct vhost_work completion_work;
+	struct llist_head completion_list;
 };
 
 struct vhost_scsi {
@@ -202,9 +206,6 @@ struct vhost_scsi {
 	struct vhost_dev dev;
 	struct vhost_scsi_virtqueue vqs[VHOST_SCSI_MAX_VQ];
 
-	struct vhost_work vs_completion_work; /* cmd completion work item */
-	struct llist_head vs_completion_list; /* cmd completion queue */
-
 	struct vhost_work vs_event_work; /* evt injection work item */
 	struct llist_head vs_event_list; /* evt injection queue */
 
@@ -380,10 +381,11 @@ static void vhost_scsi_release_cmd(struct se_cmd *se_cmd)
 	} else {
 		struct vhost_scsi_cmd *cmd = container_of(se_cmd,
 					struct vhost_scsi_cmd, tvc_se_cmd);
-		struct vhost_scsi *vs = cmd->tvc_vhost;
+		struct vhost_scsi_virtqueue *svq =  container_of(cmd->tvc_vq,
+					struct vhost_scsi_virtqueue, vq);
 
-		llist_add(&cmd->tvc_completion_list, &vs->vs_completion_list);
-		vhost_work_queue(&vs->dev, &vs->vs_completion_work);
+		llist_add(&cmd->tvc_completion_list, &svq->completion_list);
+		vhost_vq_work_queue(&svq->vq, &svq->completion_work);
 	}
 }
 
@@ -545,18 +547,17 @@ static void vhost_scsi_evt_work(struct vhost_work *work)
  */
 static void vhost_scsi_complete_cmd_work(struct vhost_work *work)
 {
-	struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
-					vs_completion_work);
-	DECLARE_BITMAP(signal, VHOST_SCSI_MAX_VQ);
+	struct vhost_scsi_virtqueue *svq = container_of(work,
+				struct vhost_scsi_virtqueue, completion_work);
 	struct virtio_scsi_cmd_resp v_rsp;
 	struct vhost_scsi_cmd *cmd, *t;
 	struct llist_node *llnode;
 	struct se_cmd *se_cmd;
 	struct iov_iter iov_iter;
-	int ret, vq;
+	bool signal = false;
+	int ret;
 
-	bitmap_zero(signal, VHOST_SCSI_MAX_VQ);
-	llnode = llist_del_all(&vs->vs_completion_list);
+	llnode = llist_del_all(&svq->completion_list);
 	llist_for_each_entry_safe(cmd, t, llnode, tvc_completion_list) {
 		se_cmd = &cmd->tvc_se_cmd;
 
@@ -576,21 +577,16 @@ static void vhost_scsi_complete_cmd_work(struct vhost_work *work)
 			      cmd->tvc_in_iovs, sizeof(v_rsp));
 		ret = copy_to_iter(&v_rsp, sizeof(v_rsp), &iov_iter);
 		if (likely(ret = sizeof(v_rsp))) {
-			struct vhost_scsi_virtqueue *q;
+			signal = true;
 			vhost_add_used(cmd->tvc_vq, cmd->tvc_vq_desc, 0);
-			q = container_of(cmd->tvc_vq, struct vhost_scsi_virtqueue, vq);
-			vq = q - vs->vqs;
-			__set_bit(vq, signal);
 		} else
 			pr_err("Faulted on virtio_scsi_cmd_resp\n");
 
 		vhost_scsi_release_cmd_res(se_cmd);
 	}
 
-	vq = -1;
-	while ((vq = find_next_bit(signal, VHOST_SCSI_MAX_VQ, vq + 1))
-		< VHOST_SCSI_MAX_VQ)
-		vhost_signal(&vs->dev, &vs->vqs[vq].vq);
+	if (signal)
+		vhost_signal(&svq->vs->dev, &svq->vq);
 }
 
 static struct vhost_scsi_cmd *
@@ -1805,6 +1801,7 @@ static int vhost_scsi_set_features(struct vhost_scsi *vs, u64 features)
 
 static int vhost_scsi_open(struct inode *inode, struct file *f)
 {
+	struct vhost_scsi_virtqueue *svq;
 	struct vhost_scsi *vs;
 	struct vhost_virtqueue **vqs;
 	int r = -ENOMEM, i;
@@ -1820,7 +1817,6 @@ static int vhost_scsi_open(struct inode *inode, struct file *f)
 	if (!vqs)
 		goto err_vqs;
 
-	vhost_work_init(&vs->vs_completion_work, vhost_scsi_complete_cmd_work);
 	vhost_work_init(&vs->vs_event_work, vhost_scsi_evt_work);
 
 	vs->vs_events_nr = 0;
@@ -1831,8 +1827,14 @@ static int vhost_scsi_open(struct inode *inode, struct file *f)
 	vs->vqs[VHOST_SCSI_VQ_CTL].vq.handle_kick = vhost_scsi_ctl_handle_kick;
 	vs->vqs[VHOST_SCSI_VQ_EVT].vq.handle_kick = vhost_scsi_evt_handle_kick;
 	for (i = VHOST_SCSI_VQ_IO; i < VHOST_SCSI_MAX_VQ; i++) {
-		vqs[i] = &vs->vqs[i].vq;
-		vs->vqs[i].vq.handle_kick = vhost_scsi_handle_kick;
+		svq = &vs->vqs[i];
+
+		vqs[i] = &svq->vq;
+		svq->vs = vs;
+		init_llist_head(&svq->completion_list);
+		vhost_work_init(&svq->completion_work,
+				vhost_scsi_complete_cmd_work);
+		svq->vq.handle_kick = vhost_scsi_handle_kick;
 	}
 
 	/*
-- 
1.8.3.1