[PATCH v4 1/2] virtio_blk: Add control virtqueue support

[Linlin Zhang <linlin.zhang@oss.qualcomm.com>]

Add support for the optional virtio-blk control virtqueue.

If control queue feature bit is negotiated, this allows the driver
to manage control-queue requests independently from the data path
and to safely handle outstanding requests during device removal
and suspend.

No control command is submitted by this change. The control virtqueue
will be used by a subsequent inline encryption implementation.

Signed-off-by: Linlin Zhang <linlin.zhang@oss.qualcomm.com>
---
 drivers/block/virtio_blk.c      | 244 +++++++++++++++++++++++++++++++-
 include/uapi/linux/virtio_blk.h |   1 +
 2 files changed, 241 insertions(+), 4 deletions(-)

diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
index 32bf3ba07a9d..6e9729134efa 100644
--- a/drivers/block/virtio_blk.c
+++ b/drivers/block/virtio_blk.c
@@ -6,6 +6,7 @@
 #include <linux/hdreg.h>
 #include <linux/module.h>
 #include <linux/mutex.h>
+#include <linux/completion.h>
 #include <linux/interrupt.h>
 #include <linux/virtio.h>
 #include <linux/virtio_blk.h>
@@ -52,6 +53,15 @@ struct virtio_blk_vq {
 	char name[VQ_NAME_LEN];
 } ____cacheline_aligned_in_smp;
 
+struct virtio_blk_ctrl_vq {
+	struct virtqueue *vq;
+	struct mutex mutex;
+	spinlock_t lock;
+	unsigned int inflight;
+	bool dead;
+	struct completion drained;
+};
+
 struct virtio_blk {
 	/*
 	 * This mutex must be held by anything that may run after
@@ -83,6 +93,9 @@ struct virtio_blk {
 
 	/* For zoned device */
 	unsigned int zone_sectors;
+
+	/* Control virtqueue state. */
+	struct virtio_blk_ctrl_vq ctrl_vq;
 };
 
 struct virtblk_req {
@@ -110,6 +123,27 @@ struct virtblk_req {
 	struct scatterlist sg[];
 };
 
+/*
+ * Software-only completion state for a control-queue request.
+ */
+struct virtblk_ctrl_completion {
+	struct completion done;
+	/*
+	 * Set when virtblk_ctrl_vq_request()'s waiter timed out and moved on
+	 * without freeing this request. Whichever of virtblk_ctrlq_callback()
+	 * or virtblk_ctrl_vq_drain() later retrieves the buffer must free
+	 * this struct and the request instead of calling complete() on it.
+	 */
+	bool abandoned;
+};
+
+struct virtblk_ctrl_request {
+	__virtio32 type;
+	u8 status;
+
+	struct virtblk_ctrl_completion *compl;
+};
+
 static inline blk_status_t virtblk_result(u8 status)
 {
 	switch (status) {
@@ -863,11 +897,181 @@ static int virtblk_getgeo(struct gendisk *disk, struct hd_geometry *geo)
 	return ret;
 }
 
+#define VIRTBLK_CTRL_VQ_TIMEOUT (10 * HZ)
+
+/* Prevent new submissions and wait for in-flight requests to complete. */
+static void virtblk_ctrl_vq_quiesce(struct virtio_blk *vblk)
+{
+	unsigned long flags;
+	bool need_wait;
+
+	if (!vblk->ctrl_vq.vq)
+		return;
+
+	init_completion(&vblk->ctrl_vq.drained);
+
+	spin_lock_irqsave(&vblk->ctrl_vq.lock, flags);
+	vblk->ctrl_vq.dead = true;
+	need_wait = vblk->ctrl_vq.inflight != 0;
+	spin_unlock_irqrestore(&vblk->ctrl_vq.lock, flags);
+
+	if (need_wait &&
+	    !wait_for_completion_timeout(&vblk->ctrl_vq.drained,
+					  VIRTBLK_CTRL_VQ_TIMEOUT))
+		dev_warn(&vblk->vdev->dev,
+			 "timed out waiting for control queue requests to complete\n");
+}
+
+/* Fail requests left in the control queue after reset. */
+static void virtblk_ctrl_vq_drain(struct virtio_blk *vblk)
+{
+	struct virtblk_ctrl_request *creq;
+	unsigned long flags;
+
+	if (!vblk->ctrl_vq.vq)
+		return;
+
+	spin_lock_irqsave(&vblk->ctrl_vq.lock, flags);
+	while ((creq = virtqueue_detach_unused_buf(vblk->ctrl_vq.vq)) != NULL) {
+		bool abandoned = creq->compl->abandoned;
+
+		if (!WARN_ON_ONCE(!vblk->ctrl_vq.inflight))
+			vblk->ctrl_vq.inflight--;
+
+		if (abandoned) {
+			kfree(creq->compl);
+			kfree(creq);
+		} else {
+			creq->status = VIRTIO_BLK_S_IOERR;
+			complete(&creq->compl->done);
+		}
+	}
+	spin_unlock_irqrestore(&vblk->ctrl_vq.lock, flags);
+}
+
+static void virtblk_ctrlq_callback(struct virtqueue *vq)
+{
+	struct virtio_blk *vblk = vq->vdev->priv;
+	struct virtblk_ctrl_request *creq;
+	unsigned long flags;
+	unsigned int len;
+
+	spin_lock_irqsave(&vblk->ctrl_vq.lock, flags);
+	do {
+		virtqueue_disable_cb(vq);
+		while ((creq = virtqueue_get_buf(vq, &len)) != NULL) {
+			bool drained = false;
+			bool abandoned = creq->compl->abandoned;
+
+			/*
+			 * Skip the inflight decrement/drained check when
+			 * inflight was already 0 (a bug, hence WARN_ON_ONCE)
+			 * — but always fall through to resolve @creq below
+			 * regardless.  Never leave a synchronous caller
+			 * blocked because the accounting state was already
+			 * inconsistent.
+			 */
+			if (!WARN_ON_ONCE(!vblk->ctrl_vq.inflight) &&
+			    --vblk->ctrl_vq.inflight == 0 && vblk->ctrl_vq.dead)
+				drained = true;
+
+			/*
+			 * Decide and act on @abandoned without dropping the lock
+			 * to avoid the memory leakage of @creq and its completion
+			 * in the race condition that virtblk_ctrl_vq_request()
+			 * wakes up from the timeout, and the callback resumes at
+			 * the same time.
+			 */
+			if (abandoned) {
+				kfree(creq->compl);
+				kfree(creq);
+			} else {
+				complete(&creq->compl->done);
+			}
+			if (drained)
+				complete(&vblk->ctrl_vq.drained);
+		}
+	} while (!virtqueue_enable_cb(vq));
+	spin_unlock_irqrestore(&vblk->ctrl_vq.lock, flags);
+}
+
+/* Submit a control-queue request and wait for completion. */
+static int virtblk_ctrl_vq_request(struct virtio_blk *vblk,
+				    struct virtblk_ctrl_request *creq,
+				    struct scatterlist *sgs[],
+				    unsigned int out_sgs, unsigned int in_sgs)
+{
+	struct virtblk_ctrl_completion *comp;
+	unsigned long flags;
+	int err;
+
+	/*
+	 * GFP_NOIO: this may be reached on the bio-submission path
+	 * (memory reclaim writing back dirty pages to this same device),
+	 * so GFP_KERNEL could self-deadlock.
+	 */
+	comp = kmalloc_obj(*comp, GFP_NOIO);
+	if (!comp)
+		return -ENOMEM;
+	init_completion(&comp->done);
+	comp->abandoned = false;
+
+	mutex_lock(&vblk->ctrl_vq.mutex);
+	creq->compl = comp;
+
+	spin_lock_irqsave(&vblk->ctrl_vq.lock, flags);
+	if (vblk->ctrl_vq.dead) {
+		spin_unlock_irqrestore(&vblk->ctrl_vq.lock, flags);
+		mutex_unlock(&vblk->ctrl_vq.mutex);
+		kfree(comp);
+		creq->compl = NULL;
+		return -ENODEV;
+	}
+	err = virtqueue_add_sgs(vblk->ctrl_vq.vq, sgs, out_sgs, in_sgs, creq, GFP_ATOMIC);
+	if (!err) {
+		vblk->ctrl_vq.inflight++;
+		virtqueue_kick(vblk->ctrl_vq.vq);
+	}
+	spin_unlock_irqrestore(&vblk->ctrl_vq.lock, flags);
+	if (err) {
+		mutex_unlock(&vblk->ctrl_vq.mutex);
+		kfree(comp);
+		creq->compl = NULL;
+		return err;
+	}
+
+	if (wait_for_completion_timeout(&comp->done, VIRTBLK_CTRL_VQ_TIMEOUT)) {
+		mutex_unlock(&vblk->ctrl_vq.mutex);
+		kfree(comp);
+		creq->compl = NULL;
+		return 0;
+	}
+
+	/*
+	 * The host hasn't responded within the timeout. @creq is still
+	 * owned by the device, so don't touch its DMA-target fields or
+	 * free it here. Mark it abandoned and hand ownership of both @creq
+	 * and @comp to whichever of virtblk_ctrlq_callback() or
+	 * virtblk_ctrl_vq_drain() retrieves the buffer later; unlock the
+	 * mutex so subsequent requests aren't serialized behind an
+	 * unresponsive host.
+	 */
+	spin_lock_irqsave(&vblk->ctrl_vq.lock, flags);
+	comp->abandoned = true;
+	spin_unlock_irqrestore(&vblk->ctrl_vq.lock, flags);
+	mutex_unlock(&vblk->ctrl_vq.mutex);
+
+	dev_warn(&vblk->vdev->dev,
+		 "control queue request timed out, abandoning\n");
+	return -ETIMEDOUT;
+}
+
 static void virtblk_free_disk(struct gendisk *disk)
 {
 	struct virtio_blk *vblk = disk->private_data;
 
 	ida_free(&vd_index_ida, vblk->index);
+	mutex_destroy(&vblk->ctrl_vq.mutex);
 	mutex_destroy(&vblk->vdev_mutex);
 	kfree(vblk);
 }
@@ -965,6 +1169,8 @@ static int init_vq(struct virtio_blk *vblk)
 	struct virtqueue **vqs;
 	unsigned short num_vqs;
 	unsigned short num_poll_vqs;
+	unsigned short total_vqs;
+	bool has_ctrl_vq;
 	struct virtio_device *vdev = vblk->vdev;
 	struct irq_affinity desc = { 0, };
 
@@ -993,12 +1199,19 @@ static int init_vq(struct virtio_blk *vblk)
 				vblk->io_queues[HCTX_TYPE_READ],
 				vblk->io_queues[HCTX_TYPE_POLL]);
 
+	/*
+	 * The control vq is appended after the data vqs whenever
+	 * F_CTRL_VQ is negotiated.
+	 */
+	has_ctrl_vq = virtio_has_feature(vdev, VIRTIO_BLK_F_CTRL_VQ);
+	total_vqs = num_vqs + (has_ctrl_vq ? 1 : 0);
+
 	vblk->vqs = kmalloc_objs(*vblk->vqs, num_vqs);
 	if (!vblk->vqs)
 		return -ENOMEM;
 
-	vqs_info = kzalloc_objs(*vqs_info, num_vqs);
-	vqs = kmalloc_objs(*vqs, num_vqs);
+	vqs_info = kzalloc_objs(*vqs_info, total_vqs);
+	vqs = kmalloc_objs(*vqs, total_vqs);
 	if (!vqs_info || !vqs) {
 		err = -ENOMEM;
 		goto out;
@@ -1015,8 +1228,13 @@ static int init_vq(struct virtio_blk *vblk)
 		vqs_info[i].name = vblk->vqs[i].name;
 	}
 
+	if (has_ctrl_vq) {
+		vqs_info[num_vqs].callback = virtblk_ctrlq_callback;
+		vqs_info[num_vqs].name = "control";
+	}
+
 	/* Discover virtqueues and write information to configuration.  */
-	err = virtio_find_vqs(vdev, num_vqs, vqs, vqs_info, &desc);
+	err = virtio_find_vqs(vdev, total_vqs, vqs, vqs_info, &desc);
 	if (err)
 		goto out;
 
@@ -1025,6 +1243,9 @@ static int init_vq(struct virtio_blk *vblk)
 		vblk->vqs[i].vq = vqs[i];
 	}
 	vblk->num_vqs = num_vqs;
+	vblk->ctrl_vq.vq = has_ctrl_vq ? vqs[num_vqs] : NULL;
+	vblk->ctrl_vq.dead = false;
+	vblk->ctrl_vq.inflight = 0;
 
 out:
 	kfree(vqs);
@@ -1464,14 +1685,18 @@ static int virtblk_probe(struct virtio_device *vdev)
 	}
 
 	mutex_init(&vblk->vdev_mutex);
+	mutex_init(&vblk->ctrl_vq.mutex);
+	spin_lock_init(&vblk->ctrl_vq.lock);
 
 	vblk->vdev = vdev;
 
 	INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
 
 	err = init_vq(vblk);
-	if (err)
+	if (err) {
+		dev_err(&vdev->dev, "init virt queue failed: err = %d\n", err);
 		goto out_free_vblk;
+	}
 
 	/* Default queue sizing is to fill the ring. */
 	if (!virtblk_queue_depth) {
@@ -1553,6 +1778,7 @@ static int virtblk_probe(struct virtio_device *vdev)
 out_free_vq:
 	vdev->config->del_vqs(vdev);
 	kfree(vblk->vqs);
+	vblk->ctrl_vq.vq = NULL;
 out_free_vblk:
 	kfree(vblk);
 out_free_index:
@@ -1571,16 +1797,21 @@ static void virtblk_remove(struct virtio_device *vdev)
 	del_gendisk(vblk->disk);
 	blk_mq_free_tag_set(&vblk->tag_set);
 
+	virtblk_ctrl_vq_quiesce(vblk);
+
 	mutex_lock(&vblk->vdev_mutex);
 
 	/* Stop all the virtqueues. */
 	virtio_reset_device(vdev);
+	virtblk_ctrl_vq_drain(vblk);
 
 	/* Virtqueues are stopped, nothing can use vblk->vdev anymore. */
 	vblk->vdev = NULL;
 
 	vdev->config->del_vqs(vdev);
 	kfree(vblk->vqs);
+	vblk->vqs = NULL;
+	vblk->ctrl_vq.vq = NULL;
 
 	mutex_unlock(&vblk->vdev_mutex);
 
@@ -1598,8 +1829,11 @@ static int virtblk_freeze_priv(struct virtio_device *vdev)
 	blk_mq_quiesce_queue_nowait(q);
 	blk_mq_unfreeze_queue(q, memflags);
 
+	virtblk_ctrl_vq_quiesce(vblk);
+
 	/* Ensure we don't receive any more interrupts */
 	virtio_reset_device(vdev);
+	virtblk_ctrl_vq_drain(vblk);
 
 	/* Make sure no work handler is accessing the device. */
 	flush_work(&vblk->config_work);
@@ -1612,6 +1846,7 @@ static int virtblk_freeze_priv(struct virtio_device *vdev)
 	 * pointers safely.
 	 */
 	vblk->vqs = NULL;
+	vblk->ctrl_vq.vq = NULL;
 
 	return 0;
 }
@@ -1672,6 +1907,7 @@ static unsigned int features[] = {
 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
 	VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
 	VIRTIO_BLK_F_SECURE_ERASE, VIRTIO_BLK_F_ZONED,
+	VIRTIO_BLK_F_CTRL_VQ,
 };
 
 static struct virtio_driver virtio_blk = {
diff --git a/include/uapi/linux/virtio_blk.h b/include/uapi/linux/virtio_blk.h
index 3744e4da1b2a..1daddd259249 100644
--- a/include/uapi/linux/virtio_blk.h
+++ b/include/uapi/linux/virtio_blk.h
@@ -42,6 +42,7 @@
 #define VIRTIO_BLK_F_WRITE_ZEROES	14	/* WRITE ZEROES is supported */
 #define VIRTIO_BLK_F_SECURE_ERASE	16 /* Secure Erase is supported */
 #define VIRTIO_BLK_F_ZONED		17	/* Zoned block device */
+#define VIRTIO_BLK_F_CTRL_VQ			20	/* Control queue */
 
 /* Legacy feature bits */
 #ifndef VIRTIO_BLK_NO_LEGACY
-- 
2.34.1