[dpdk-dev] [PATCH v2 6/7] net/virtio_user: enable offloading

[Jianfeng Tan]

When used with vhost kernel backend, we can offload at both directions.
  - From vhost kernel to virtio_user, the offload is enabled so that
    DPDK app can trust the flow is checksum-correct; and if DPDK app
    sends it through another port, the checksum needs to be
    recalculated or offloaded. It also applies to TSO.
  - From virtio_user to vhost_kernel, the offload is enabled so that
    kernel can trust the flow is L4-checksum-correct, no need to verify
    it; if kernel will consume it, DPDK app should make sure the
    l3-checksum is correctly set.

Signed-off-by: Jianfeng Tan <jianfeng.tan at intel.com>
---
 drivers/net/virtio/virtio_user/vhost_kernel.c | 61 ++++++++++++++++++++++++++-
 1 file changed, 59 insertions(+), 2 deletions(-)

diff --git a/drivers/net/virtio/virtio_user/vhost_kernel.c b/drivers/net/virtio/virtio_user/vhost_kernel.c
index 8984c5c..fb3c454 100644
--- a/drivers/net/virtio/virtio_user/vhost_kernel.c
+++ b/drivers/net/virtio/virtio_user/vhost_kernel.c
@@ -91,6 +91,13 @@ struct vhost_memory_kernel {
 #define IFF_ATTACH_QUEUE 0x0200
 #define IFF_DETACH_QUEUE 0x0400
 
+/* Features for GSO (TUNSETOFFLOAD). */
+#define TUN_F_CSUM	0x01	/* You can hand me unchecksummed packets. */
+#define TUN_F_TSO4	0x02	/* I can handle TSO for IPv4 packets */
+#define TUN_F_TSO6	0x04	/* I can handle TSO for IPv6 packets */
+#define TUN_F_TSO_ECN	0x08	/* I can handle TSO with ECN bits. */
+#define TUN_F_UFO	0x10	/* I can handle UFO packets */
+
 /* Constants */
 #define TUN_DEF_SNDBUF	(1ull << 20)
 #define PATH_NET_TUN	"/dev/net/tun"
@@ -173,6 +180,28 @@ prepare_vhost_memory_kernel(void)
 	return vm;
 }
 
+/* with below features, vhost kernel does not need to do the checksum and TSO,
+ * these info will be passed to virtio_user through virtio net header.
+ */
+static const uint64_t guest_offloads_mask =
+	(1ULL << VIRTIO_NET_F_GUEST_CSUM) |
+	(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
+	(1ULL << VIRTIO_NET_F_GUEST_TSO6) |
+	(1ULL << VIRTIO_NET_F_GUEST_ECN)  |
+	(1ULL << VIRTIO_NET_F_GUEST_UFO);
+
+/* with below features, when flows from virtio_user to vhost kernel
+ * (1) if flows goes up through the kernel networking stack, it does not need
+ * to verify checksum, which can save CPU cycles;
+ * (2) if flows goes through a Linux bridge and outside from an interface
+ * (kernel driver), checksum and TSO will be done by GSO in kernel or even
+ * offloaded into real physical device.
+ */
+static const uint64_t host_offloads_mask =
+	(1ULL << VIRTIO_NET_F_CSUM) |
+	(1ULL << VIRTIO_NET_F_HOST_TSO4) |
+	(1ULL << VIRTIO_NET_F_HOST_TSO6);
+
 static int
 vhost_kernel_ioctl(struct virtio_user_dev *dev,
 		   enum vhost_user_request req,
@@ -191,10 +220,15 @@ vhost_kernel_ioctl(struct virtio_user_dev *dev,
 		arg = (void *)vm;
 	}
 
-	/* Does not work when VIRTIO_F_IOMMU_PLATFORM now, why? */
-	if (req_kernel == VHOST_SET_FEATURES)
+	if (req_kernel == VHOST_SET_FEATURES) {
+		/* Does not work when VIRTIO_F_IOMMU_PLATFORM now, why? */
 		*(uint64_t *)arg &= ~(1ULL << VIRTIO_F_IOMMU_PLATFORM);
 
+		/* VHOST kernel does not know about below flags */
+		*(uint64_t *)arg &= ~guest_offloads_mask;
+		*(uint64_t *)arg &= ~host_offloads_mask;
+	}
+
 	for (i = 0; i < VHOST_KERNEL_MAX_QUEUES; ++i) {
 		if (dev->vhostfds[i] < 0)
 			continue;
@@ -204,6 +238,15 @@ vhost_kernel_ioctl(struct virtio_user_dev *dev,
 			break;
 	}
 
+	if (!ret && req_kernel == VHOST_GET_FEATURES) {
+		/* with tap as the backend, all these features are supported
+		 * but not claimed by vhost-net, so we add them back when
+		 * reporting to upper layer.
+		 */
+		*((uint64_t *)arg) |= guest_offloads_mask;
+		*((uint64_t *)arg) |= host_offloads_mask;
+	}
+
 	if (vm)
 		free(vm);
 
@@ -271,6 +314,12 @@ vhost_kernel_enable_queue_pair(struct virtio_user_dev *dev,
 	int hdr_size;
 	int vhostfd;
 	int tapfd;
+	unsigned int offload =
+			TUN_F_CSUM |
+			TUN_F_TSO4 |
+			TUN_F_TSO6 |
+			TUN_F_TSO_ECN |
+			TUN_F_UFO;
 
 	vhostfd = dev->vhostfds[pair_idx];
 
@@ -345,6 +394,14 @@ vhost_kernel_enable_queue_pair(struct virtio_user_dev *dev,
 		goto error;
 	}
 
+	/* TODO: before set the offload capabilities, we'd better (1) check
+	 * negotiated features to see if necessary to offload; (2) query tap
+	 * to see if it supports the offload capabilities.
+	 */
+	if (ioctl(tapfd, TUNSETOFFLOAD, offload) != 0)
+		PMD_DRV_LOG(ERR, "TUNSETOFFLOAD ioctl() failed: %s",
+			   strerror(errno));
+
 	if (vhost_kernel_set_backend(vhostfd, tapfd) < 0)
 		goto error;
 
-- 
2.7.4