vfio/mlx5: Run the SAVE state command in an async mode

Use the PF asynchronous command mode for the SAVE state command.

This enables returning earlier to user space upon issuing successfully
the command and improve latency by let things run in parallel.

Link: https://lore.kernel.org/r/20220510090206.90374-5-yishaih@nvidia.com
Signed-off-by: Yishai Hadas <yishaih@nvidia.com>
Reviewed-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>

1 files changed, 37 insertions, 3 deletions

diff --git a/drivers/vfio/pci/mlx5/main.c b/drivers/vfio/pci/mlx5/main.c
index 2db6b816f003..df8b572977da 100644
--- a/drivers/vfio/pci/mlx5/main.c
+++ b/drivers/vfio/pci/mlx5/main.c
@@ -134,12 +134,22 @@ static ssize_t mlx5vf_save_read(struct file *filp, char __user *buf, size_t len,
 		return -ESPIPE;
 	pos = &filp->f_pos;
 
+	if (!(filp->f_flags & O_NONBLOCK)) {
+		if (wait_event_interruptible(migf->poll_wait,
+			     READ_ONCE(migf->total_length) || migf->is_err))
+			return -ERESTARTSYS;
+	}
+
 	mutex_lock(&migf->lock);
+	if ((filp->f_flags & O_NONBLOCK) && !READ_ONCE(migf->total_length)) {
+		done = -EAGAIN;
+		goto out_unlock;
+	}
 	if (*pos > migf->total_length) {
 		done = -EINVAL;
 		goto out_unlock;
 	}
-	if (migf->disabled) {
+	if (migf->disabled || migf->is_err) {
 		done = -ENODEV;
 		goto out_unlock;
 	}
@@ -179,9 +189,28 @@ out_unlock:
 	return done;
 }
 
+static __poll_t mlx5vf_save_poll(struct file *filp,
+				 struct poll_table_struct *wait)
+{
+	struct mlx5_vf_migration_file *migf = filp->private_data;
+	__poll_t pollflags = 0;
+
+	poll_wait(filp, &migf->poll_wait, wait);
+
+	mutex_lock(&migf->lock);
+	if (migf->disabled || migf->is_err)
+		pollflags = EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
+	else if (READ_ONCE(migf->total_length))
+		pollflags = EPOLLIN | EPOLLRDNORM;
+	mutex_unlock(&migf->lock);
+
+	return pollflags;
+}
+
 static const struct file_operations mlx5vf_save_fops = {
 	.owner = THIS_MODULE,
 	.read = mlx5vf_save_read,
+	.poll = mlx5vf_save_poll,
 	.release = mlx5vf_release_file,
 	.llseek = no_llseek,
 };
@@ -207,7 +236,9 @@ mlx5vf_pci_save_device_data(struct mlx5vf_pci_core_device *mvdev)
 
 	stream_open(migf->filp->f_inode, migf->filp);
 	mutex_init(&migf->lock);
-
+	init_waitqueue_head(&migf->poll_wait);
+	mlx5_cmd_init_async_ctx(mvdev->mdev, &migf->async_ctx);
+	INIT_WORK(&migf->async_data.work, mlx5vf_mig_file_cleanup_cb);
 	ret = mlx5vf_cmd_query_vhca_migration_state(mvdev,
 						    &migf->total_length);
 	if (ret)
@@ -218,6 +249,7 @@ mlx5vf_pci_save_device_data(struct mlx5vf_pci_core_device *mvdev)
 	if (ret)
 		goto out_free;
 
+	migf->mvdev = mvdev;
 	ret = mlx5vf_cmd_save_vhca_state(mvdev, migf);
 	if (ret)
 		goto out_free;
@@ -323,7 +355,7 @@ mlx5vf_pci_resume_device_data(struct mlx5vf_pci_core_device *mvdev)
 	return migf;
 }
 
-static void mlx5vf_disable_fds(struct mlx5vf_pci_core_device *mvdev)
+void mlx5vf_disable_fds(struct mlx5vf_pci_core_device *mvdev)
 {
 	if (mvdev->resuming_migf) {
 		mlx5vf_disable_fd(mvdev->resuming_migf);
@@ -331,6 +363,8 @@ static void mlx5vf_disable_fds(struct mlx5vf_pci_core_device *mvdev)
 		mvdev->resuming_migf = NULL;
 	}
 	if (mvdev->saving_migf) {
+		mlx5_cmd_cleanup_async_ctx(&mvdev->saving_migf->async_ctx);
+		cancel_work_sync(&mvdev->saving_migf->async_data.work);
 		mlx5vf_disable_fd(mvdev->saving_migf);
 		fput(mvdev->saving_migf->filp);
 		mvdev->saving_migf = NULL;