s390/vfio-ap: use work struct to verify queue reset
authorTony Krowiak <akrowiak@linux.ibm.com>
Tue, 15 Aug 2023 18:43:28 +0000 (14:43 -0400)
committerHeiko Carstens <hca@linux.ibm.com>
Fri, 18 Aug 2023 13:09:29 +0000 (15:09 +0200)
Instead of waiting to verify that a queue is reset in the
vfio_ap_mdev_reset_queue function, let's use a wait queue to check the
the state of the reset. This way, when resetting all of the queues assigned
to a matrix mdev, we don't have to wait for each queue to be reset before
initiating a reset on the next queue to be reset.

Signed-off-by: Tony Krowiak <akrowiak@linux.ibm.com>
Reviewed-by: Jason J. Herne <jjherne@linux.ibm.com>
Suggested-by: Halil Pasic <pasic@linux.ibm.com>
Acked-by: Janosch Frank <frankja@linux.ibm.com>
Tested-by: Viktor Mihajlovski <mihajlov@linux.ibm.com>
Link: https://lore.kernel.org/r/20230815184333.6554-8-akrowiak@linux.ibm.com
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
drivers/s390/crypto/vfio_ap_ops.c
drivers/s390/crypto/vfio_ap_private.h

index 43224f7..3a59f1c 100644 (file)
@@ -35,7 +35,7 @@
 static int vfio_ap_mdev_reset_queues(struct ap_queue_table *qtable);
 static struct vfio_ap_queue *vfio_ap_find_queue(int apqn);
 static const struct vfio_device_ops vfio_ap_matrix_dev_ops;
-static int vfio_ap_mdev_reset_queue(struct vfio_ap_queue *q);
+static void vfio_ap_mdev_reset_queue(struct vfio_ap_queue *q);
 
 /**
  * get_update_locks_for_kvm: Acquire the locks required to dynamically update a
@@ -1623,11 +1623,13 @@ static int apq_status_check(int apqn, struct ap_queue_status *status)
 
 #define WAIT_MSG "Waited %dms for reset of queue %02x.%04x (%u, %u, %u)"
 
-static int apq_reset_check(struct vfio_ap_queue *q)
+static void apq_reset_check(struct work_struct *reset_work)
 {
        int ret = -EBUSY, elapsed = 0;
        struct ap_queue_status status;
+       struct vfio_ap_queue *q;
 
+       q = container_of(reset_work, struct vfio_ap_queue, reset_work);
        memcpy(&status, &q->reset_status, sizeof(status));
        while (true) {
                msleep(AP_RESET_INTERVAL);
@@ -1635,7 +1637,7 @@ static int apq_reset_check(struct vfio_ap_queue *q)
                status = ap_tapq(q->apqn, NULL);
                ret = apq_status_check(q->apqn, &status);
                if (ret == -EIO)
-                       return ret;
+                       return;
                if (ret == -EBUSY) {
                        pr_notice_ratelimited(WAIT_MSG, elapsed,
                                              AP_QID_CARD(q->apqn),
@@ -1663,34 +1665,32 @@ static int apq_reset_check(struct vfio_ap_queue *q)
                        break;
                }
        }
-       return ret;
 }
 
-static int vfio_ap_mdev_reset_queue(struct vfio_ap_queue *q)
+static void vfio_ap_mdev_reset_queue(struct vfio_ap_queue *q)
 {
        struct ap_queue_status status;
-       int ret = 0;
 
        if (!q)
-               return 0;
+               return;
        status = ap_zapq(q->apqn, 0);
        memcpy(&q->reset_status, &status, sizeof(status));
        switch (status.response_code) {
        case AP_RESPONSE_NORMAL:
        case AP_RESPONSE_RESET_IN_PROGRESS:
        case AP_RESPONSE_BUSY:
-               /* Let's verify whether the ZAPQ completed successfully */
-               ret = apq_reset_check(q);
+               /*
+                * Let's verify whether the ZAPQ completed successfully on a work queue.
+                */
+               queue_work(system_long_wq, &q->reset_work);
                break;
        case AP_RESPONSE_DECONFIGURED:
                /*
                 * When an AP adapter is deconfigured, the associated
                 * queues are reset, so let's set the status response code to 0
-                * so the queue may be passed through (i.e., not filtered) and
-                * return a value indicating the reset completed successfully.
+                * so the queue may be passed through (i.e., not filtered).
                 */
                q->reset_status.response_code = 0;
-               ret = 0;
                vfio_ap_free_aqic_resources(q);
                break;
        default:
@@ -1698,29 +1698,25 @@ static int vfio_ap_mdev_reset_queue(struct vfio_ap_queue *q)
                     "PQAP/ZAPQ for %02x.%04x failed with invalid rc=%u\n",
                     AP_QID_CARD(q->apqn), AP_QID_QUEUE(q->apqn),
                     status.response_code);
-               return -EIO;
        }
-
-       return ret;
 }
 
 static int vfio_ap_mdev_reset_queues(struct ap_queue_table *qtable)
 {
-       int ret, loop_cursor, rc = 0;
+       int ret = 0, loop_cursor;
        struct vfio_ap_queue *q;
 
+       hash_for_each(qtable->queues, loop_cursor, q, mdev_qnode)
+               vfio_ap_mdev_reset_queue(q);
+
        hash_for_each(qtable->queues, loop_cursor, q, mdev_qnode) {
-               ret = vfio_ap_mdev_reset_queue(q);
-               /*
-                * Regardless whether a queue turns out to be busy, or
-                * is not operational, we need to continue resetting
-                * the remaining queues.
-                */
-               if (ret)
-                       rc = ret;
+               flush_work(&q->reset_work);
+
+               if (q->reset_status.response_code)
+                       ret = -EIO;
        }
 
-       return rc;
+       return ret;
 }
 
 static int vfio_ap_mdev_open_device(struct vfio_device *vdev)
@@ -2045,6 +2041,7 @@ int vfio_ap_mdev_probe_queue(struct ap_device *apdev)
        q->apqn = to_ap_queue(&apdev->device)->qid;
        q->saved_isc = VFIO_AP_ISC_INVALID;
        memset(&q->reset_status, 0, sizeof(q->reset_status));
+       INIT_WORK(&q->reset_work, apq_reset_check);
        matrix_mdev = get_update_locks_by_apqn(q->apqn);
 
        if (matrix_mdev) {
@@ -2094,6 +2091,7 @@ void vfio_ap_mdev_remove_queue(struct ap_device *apdev)
        }
 
        vfio_ap_mdev_reset_queue(q);
+       flush_work(&q->reset_work);
        dev_set_drvdata(&apdev->device, NULL);
        kfree(q);
        release_update_locks_for_mdev(matrix_mdev);
index d6eb352..88aff8b 100644 (file)
@@ -134,6 +134,7 @@ struct ap_matrix_mdev {
  * @saved_isc: the guest ISC registered with the GIB interface
  * @mdev_qnode: allows the vfio_ap_queue struct to be added to a hashtable
  * @reset_status: the status from the last reset of the queue
+ * @reset_work: work to wait for queue reset to complete
  */
 struct vfio_ap_queue {
        struct ap_matrix_mdev *matrix_mdev;
@@ -143,6 +144,7 @@ struct vfio_ap_queue {
        unsigned char saved_isc;
        struct hlist_node mdev_qnode;
        struct ap_queue_status reset_status;
+       struct work_struct reset_work;
 };
 
 int vfio_ap_mdev_register(void);