[PATCH v12 04/14] s390/vfio-ap: Reset migration state on VFIO_DEVICE_RESET and device close

[Anthony Krowiak <akrowiak@linux.ibm.com>]

According to the VFIO migration specifications, userspace must issue a
VFIO_DEVICE_RESET ioctl if a vfio device enters an error state or fails a
state transition to recover the device and return its migration state to
RUNNING. This patch adds a function that is called when a VFIO_DEVICE_RESET
is issued that releases any open migration file buffers and resets the
migration state to RUNNING.

The same function is called when the vfio device is closed. The migration
data is not freed on close because a migration file descriptor can outlive
the vfio device file descriptor, so the state left over from one open must
be reset for the next. To do so under the locks required to detach the
guest, vfio_ap_mdev_unset_kvm() no longer takes the locks itself. Instead,
vfio_ap_mdev_close_device() takes them using get_update_locks_for_kvm()
and calls vfio_ap_mdev_unset_kvm() and vfio_ap_reset_migration_state()
with the locks held. vfio_ap_mdev_unset_kvm() asserts that the locks it
requires are held.

Signed-off-by: Anthony Krowiak <akrowiak@linux.ibm.com>
---
 drivers/s390/crypto/vfio_ap_migration.c | 21 +++++++++++++++++
 drivers/s390/crypto/vfio_ap_ops.c       | 30 ++++++++++++++++++++++---
 drivers/s390/crypto/vfio_ap_private.h   |  1 +
 3 files changed, 49 insertions(+), 3 deletions(-)

diff --git a/drivers/s390/crypto/vfio_ap_migration.c b/drivers/s390/crypto/vfio_ap_migration.c
index e2e9a8e7a833..94772cab1e85 100644
--- a/drivers/s390/crypto/vfio_ap_migration.c
+++ b/drivers/s390/crypto/vfio_ap_migration.c
@@ -236,3 +236,24 @@ void vfio_ap_release_migration_data(struct ap_matrix_mdev *matrix_mdev)
 	kfree(matrix_mdev->mig_data);
 	matrix_mdev->mig_data = NULL;
 }
+
+/**
+ * vfio_ap_reset_migration_state - Reset the vfio-ap migration state
+ *
+ * @matrix_mdev: pointer to the object maintaining the vfio-ap device state
+ *
+ * Called during VFIO_DEVICE_RESET to clean up any active migration
+ * state and reset the device to RUNNING state as required by the VFIO
+ * migration specification. It is also called when the vfio device is closed
+ * so the migration state of one open does not carry over to the next.
+ */
+void vfio_ap_reset_migration_state(struct ap_matrix_mdev *matrix_mdev)
+{
+	lockdep_assert_held(&matrix_dev->mdevs_lock);
+
+	if (!matrix_mdev->mig_data)
+		return;
+
+	vfio_ap_release_mig_files(matrix_mdev);
+	matrix_mdev->mig_data->mig_state = VFIO_DEVICE_STATE_RUNNING;
+}
diff --git a/drivers/s390/crypto/vfio_ap_ops.c b/drivers/s390/crypto/vfio_ap_ops.c
index 77b9328bdcb1..8d70089eb9c9 100644
--- a/drivers/s390/crypto/vfio_ap_ops.c
+++ b/drivers/s390/crypto/vfio_ap_ops.c
@@ -1899,20 +1899,25 @@ static void vfio_ap_mdev_dma_unmap(struct vfio_device *vdev, u64 iova,
  * by @matrix_mdev.
  *
  * @matrix_mdev: a matrix mediated device
+ *
+ * The locks acquired by get_update_locks_for_kvm() for the KVM guest
+ * @matrix_mdev->kvm must be held by the caller.
  */
 static void vfio_ap_mdev_unset_kvm(struct ap_matrix_mdev *matrix_mdev)
 {
 	struct kvm *kvm = matrix_mdev->kvm;
 
+	lockdep_assert_held(&matrix_dev->mdevs_lock);
+
 	if (kvm && kvm->arch.crypto.crycbd) {
-		get_update_locks_for_kvm(kvm);
+		lockdep_assert_held(&kvm->lock);
+		lockdep_assert_held_write(&kvm->arch.crypto.pqap_hook_rwsem);
+
 		kvm->arch.crypto.pqap_hook = NULL;
 
 		kvm_arch_crypto_clear_masks(kvm);
 		vfio_ap_mdev_reset_queues(matrix_mdev);
 		matrix_mdev->kvm = NULL;
-
-		release_update_locks_for_kvm(kvm);
 	}
 }
 
@@ -2098,8 +2103,16 @@ static void vfio_ap_mdev_close_device(struct vfio_device *vdev)
 {
 	struct ap_matrix_mdev *matrix_mdev =
 		container_of(vdev, struct ap_matrix_mdev, vdev);
+	/*
+	 * Save the KVM pointer because vfio_ap_mdev_unset_kvm() clears it and
+	 * the same pointer is needed to release the locks.
+	 */
+	struct kvm *kvm = matrix_mdev->kvm;
 
+	get_update_locks_for_kvm(kvm);
 	vfio_ap_mdev_unset_kvm(matrix_mdev);
+	vfio_ap_reset_migration_state(matrix_mdev);
+	release_update_locks_for_kvm(kvm);
 }
 
 static void vfio_ap_mdev_request(struct vfio_device *vdev, unsigned int count)
@@ -2316,6 +2329,17 @@ static ssize_t vfio_ap_mdev_ioctl(struct vfio_device *vdev,
 		break;
 	case VFIO_DEVICE_RESET:
 		ret = vfio_ap_mdev_reset_queues(matrix_mdev);
+		/*
+		 * Reset the migration state unconditionally, regardless of
+		 * whether the hardware queue reset succeeded. The migration
+		 * state tracks the AP configuration transfer lifecycle and is
+		 * independent of the hardware queue reset outcome. The VFIO
+		 * migration spec designates VFIO_DEVICE_RESET as the mechanism
+		 * to return device_state to RUNNING (e.g. to recover from
+		 * ERROR), so the migration state must be cleaned up even when
+		 * one or more queues fail to reset.
+		 */
+		vfio_ap_reset_migration_state(matrix_mdev);
 		break;
 	case VFIO_DEVICE_GET_IRQ_INFO:
 		ret = vfio_ap_get_irq_info(arg);
diff --git a/drivers/s390/crypto/vfio_ap_private.h b/drivers/s390/crypto/vfio_ap_private.h
index aa8a22585204..c6e577064e7e 100644
--- a/drivers/s390/crypto/vfio_ap_private.h
+++ b/drivers/s390/crypto/vfio_ap_private.h
@@ -174,5 +174,6 @@ void vfio_ap_on_scan_complete(struct ap_config_info *new_config_info,
 
 int vfio_ap_init_migration_data(struct ap_matrix_mdev *matrix_mdev);
 void vfio_ap_release_migration_data(struct ap_matrix_mdev *matrix_mdev);
+void vfio_ap_reset_migration_state(struct ap_matrix_mdev *matrix_mdev);
 
 #endif /* _VFIO_AP_PRIVATE_H_ */
-- 
2.53.0