Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Jason Wang <jasowang@redhat.com>
To:     jasowang@redhat.com, mst@redhat.com
Cc:     elic@nvidia.com, hdanton@sina.com,
        virtualization@lists.linux-foundation.org,
        linux-kernel@vger.kernel.org
Subject: [PATCH RESEND V2 1/3] vdpa: mlx5: prevent cvq work from hogging CPU
Date:   Tue, 29 Mar 2022 12:21:07 +0800
Message-Id: <20220329042109.4029-1-jasowang@redhat.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

A userspace triggerable infinite loop could happen in
mlx5_cvq_kick_handler() if userspace keeps sending a huge amount of
cvq requests.

Fixing this by introducing a quota and re-queue the work if we're out
of the budget (currently the implicit budget is one) . While at it,
using a per device work struct to avoid on demand memory allocation
for cvq.

Fixes: 5262912ef3cfc ("vdpa/mlx5: Add support for control VQ and MAC setting")
Signed-off-by: Jason Wang <jasowang@redhat.com>
---
Changes since V1:
- Using 1 as the budget
---
 drivers/vdpa/mlx5/net/mlx5_vnet.c | 21 +++++++++------------
 1 file changed, 9 insertions(+), 12 deletions(-)

diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c b/drivers/vdpa/mlx5/net/mlx5_vnet.c
index d0f91078600e..b2afd2b6fbca 100644
--- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
+++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
@@ -163,6 +163,7 @@ struct mlx5_vdpa_net {
 	u32 cur_num_vqs;
 	struct notifier_block nb;
 	struct vdpa_callback config_cb;
+	struct mlx5_vdpa_wq_ent cvq_ent;
 };
 
 static void free_resources(struct mlx5_vdpa_net *ndev);
@@ -1616,10 +1617,10 @@ static void mlx5_cvq_kick_handler(struct work_struct *work)
 	ndev = to_mlx5_vdpa_ndev(mvdev);
 	cvq = &mvdev->cvq;
 	if (!(ndev->mvdev.actual_features & BIT_ULL(VIRTIO_NET_F_CTRL_VQ)))
-		goto out;
+		return;
 
 	if (!cvq->ready)
-		goto out;
+		return;
 
 	while (true) {
 		err = vringh_getdesc_iotlb(&cvq->vring, &cvq->riov, &cvq->wiov, &cvq->head,
@@ -1653,9 +1654,10 @@ static void mlx5_cvq_kick_handler(struct work_struct *work)
 
 		if (vringh_need_notify_iotlb(&cvq->vring))
 			vringh_notify(&cvq->vring);
+
+		queue_work(mvdev->wq, &wqent->work);
+		break;
 	}
-out:
-	kfree(wqent);
 }
 
 static void mlx5_vdpa_kick_vq(struct vdpa_device *vdev, u16 idx)
@@ -1663,7 +1665,6 @@ static void mlx5_vdpa_kick_vq(struct vdpa_device *vdev, u16 idx)
 	struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
 	struct mlx5_vdpa_net *ndev = to_mlx5_vdpa_ndev(mvdev);
 	struct mlx5_vdpa_virtqueue *mvq;
-	struct mlx5_vdpa_wq_ent *wqent;
 
 	if (!is_index_valid(mvdev, idx))
 		return;
@@ -1672,13 +1673,7 @@ static void mlx5_vdpa_kick_vq(struct vdpa_device *vdev, u16 idx)
 		if (!mvdev->cvq.ready)
 			return;
 
-		wqent = kzalloc(sizeof(*wqent), GFP_ATOMIC);
-		if (!wqent)
-			return;
-
-		wqent->mvdev = mvdev;
-		INIT_WORK(&wqent->work, mlx5_cvq_kick_handler);
-		queue_work(mvdev->wq, &wqent->work);
+		queue_work(mvdev->wq, &ndev->cvq_ent.work);
 		return;
 	}
 
@@ -2668,6 +2663,8 @@ static int mlx5_vdpa_dev_add(struct vdpa_mgmt_dev *v_mdev, const char *name,
 	if (err)
 		goto err_mr;
 
+	ndev->cvq_ent.mvdev = mvdev;
+	INIT_WORK(&ndev->cvq_ent.work, mlx5_cvq_kick_handler);
 	mvdev->wq = create_singlethread_workqueue("mlx5_vdpa_wq");
 	if (!mvdev->wq) {
 		err = -ENOMEM;
-- 
2.18.1