[5/6] net/vhost: perform SW checksum in Rx path
Checks
Commit Message
Virtio specification supports host checksum offloading
for L4, which is enabled with VIRTIO_NET_F_CSUM feature
negotiation. However, the Vhost PMD does not advertise
Rx checksum offload capabilities, so we can end-up with
the VIRTIO_NET_F_CSUM feature being negociated, implying
the Vhost library returns packets with checksum being
offloaded while the application did not request for it.
Advertising these offload capabilities at the ethdev level
is not enough, because we could still end-up with the
application not enabling these offloads while the guest
still negotiate them.
This patch advertizes the Rx checksum offload capabilities,
and introduces a compatibility layer to cover the case
VIRTIO_NET_F_CSUM has been negotiated but the application
does not configure the Rx checksum offloads. This function
performis the L4 Rx checksum in SW for UDP and TCP. Note
that it is not needed to calculate the pseudo-header
checksum, because the Virtio specification requires that
the driver do it.
This patch does not advertize SCTP checksum offloading
capability for now, but it could be handled later if the
need arises.
Reported-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
---
doc/guides/nics/features/vhost.ini | 1 +
drivers/net/vhost/rte_eth_vhost.c | 83 ++++++++++++++++++++++++++++++
2 files changed, 84 insertions(+)
@@ -7,6 +7,7 @@
Link status = Y
Free Tx mbuf on demand = Y
Queue status event = Y
+L4 checksum offload = P
Basic stats = Y
Extended stats = Y
x86-32 = Y
@@ -12,6 +12,7 @@
#include <ethdev_vdev.h>
#include <rte_malloc.h>
#include <rte_memcpy.h>
+#include <rte_net.h>
#include <rte_bus_vdev.h>
#include <rte_kvargs.h>
#include <rte_vhost.h>
@@ -107,10 +108,12 @@ struct pmd_internal {
char *iface_name;
uint64_t flags;
uint64_t disable_flags;
+ uint64_t features;
uint16_t max_queues;
int vid;
rte_atomic32_t started;
bool vlan_strip;
+ bool rx_sw_csum;
};
struct internal_list {
@@ -362,6 +365,70 @@ vhost_update_single_packet_xstats(struct vhost_queue *vq, struct rte_mbuf *buf)
vhost_count_xcast_packets(vq, buf);
}
+static void
+vhost_dev_csum_configure(struct rte_eth_dev *eth_dev)
+{
+ struct pmd_internal *internal = eth_dev->data->dev_private;
+ const struct rte_eth_rxmode *rxmode = ð_dev->data->dev_conf.rxmode;
+
+ internal->rx_sw_csum = false;
+
+ /* SW checksum is not compatible with legacy mode */
+ if (!(internal->flags & RTE_VHOST_USER_NET_COMPLIANT_OL_FLAGS))
+ return;
+
+ if (internal->features & (1ULL << VIRTIO_NET_F_CSUM)) {
+ if (!(rxmode->offloads &
+ (RTE_ETH_RX_OFFLOAD_UDP_CKSUM | RTE_ETH_RX_OFFLOAD_TCP_CKSUM))) {
+ VHOST_LOG(NOTICE, "Rx csum will be done in SW, may impact performance.");
+ internal->rx_sw_csum = true;
+ }
+ }
+}
+
+static void
+vhost_dev_rx_sw_csum(struct rte_mbuf *mbuf)
+{
+ struct rte_net_hdr_lens hdr_lens;
+ uint32_t ptype, hdr_len;
+ uint16_t csum = 0, csum_offset;
+
+ /* Return early if the L4 checksum was not offloaded */
+ if ((mbuf->ol_flags & RTE_MBUF_F_RX_L4_CKSUM_MASK) != RTE_MBUF_F_RX_L4_CKSUM_NONE)
+ return;
+
+ ptype = rte_net_get_ptype(mbuf, &hdr_lens, RTE_PTYPE_ALL_MASK);
+
+ hdr_len = hdr_lens.l2_len + hdr_lens.l3_len;
+
+ switch (ptype & RTE_PTYPE_L4_MASK) {
+ case RTE_PTYPE_L4_TCP:
+ csum_offset = offsetof(struct rte_tcp_hdr, cksum) + hdr_len;
+ break;
+ case RTE_PTYPE_L4_UDP:
+ csum_offset = offsetof(struct rte_udp_hdr, dgram_cksum) + hdr_len;
+ break;
+ default:
+ /* Unsupported packet type */
+ return;
+ }
+
+ /* The pseudo-header checksum is already performed, as per Virtio spec */
+ if (rte_raw_cksum_mbuf(mbuf, hdr_len, rte_pktmbuf_pkt_len(mbuf) - hdr_len, &csum) < 0)
+ return;
+
+ csum = ~csum;
+ /* See RFC768 */
+ if (unlikely((ptype & RTE_PTYPE_L4_UDP) && csum == 0))
+ csum = 0xffff;
+
+ if (rte_pktmbuf_data_len(mbuf) >= csum_offset + 1)
+ *rte_pktmbuf_mtod_offset(mbuf, uint16_t *, csum_offset) = csum;
+
+ mbuf->ol_flags &= ~RTE_MBUF_F_RX_L4_CKSUM_MASK;
+ mbuf->ol_flags |= RTE_MBUF_F_RX_L4_CKSUM_GOOD;
+}
+
static uint16_t
eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
{
@@ -402,6 +469,9 @@ eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
if (r->internal->vlan_strip)
rte_vlan_strip(bufs[i]);
+ if (r->internal->rx_sw_csum)
+ vhost_dev_rx_sw_csum(bufs[i]);
+
r->stats.bytes += bufs[i]->pkt_len;
r->stats.xstats[VHOST_BYTE] += bufs[i]->pkt_len;
@@ -805,6 +875,11 @@ new_device(int vid)
eth_dev->data->numa_node = newnode;
#endif
+ if (rte_vhost_get_negotiated_features(vid, &internal->features)) {
+ VHOST_LOG(ERR, "Failed to get device features\n");
+ return -1;
+ }
+
internal->vid = vid;
if (rte_atomic32_read(&internal->started) == 1) {
queue_setup(eth_dev, internal);
@@ -827,6 +902,8 @@ new_device(int vid)
eth_dev->data->dev_link.link_status = RTE_ETH_LINK_UP;
+ vhost_dev_csum_configure(eth_dev);
+
rte_atomic32_set(&internal->dev_attached, 1);
update_queuing_status(eth_dev);
@@ -1131,6 +1208,8 @@ eth_dev_configure(struct rte_eth_dev *dev)
internal->vlan_strip = !!(rxmode->offloads & RTE_ETH_RX_OFFLOAD_VLAN_STRIP);
+ vhost_dev_csum_configure(dev);
+
return 0;
}
@@ -1281,6 +1360,10 @@ eth_dev_info(struct rte_eth_dev *dev,
dev_info->tx_offload_capa = RTE_ETH_TX_OFFLOAD_MULTI_SEGS |
RTE_ETH_TX_OFFLOAD_VLAN_INSERT;
dev_info->rx_offload_capa = RTE_ETH_RX_OFFLOAD_VLAN_STRIP;
+ if (internal->flags & RTE_VHOST_USER_NET_COMPLIANT_OL_FLAGS) {
+ dev_info->rx_offload_capa |= RTE_ETH_RX_OFFLOAD_UDP_CKSUM |
+ RTE_ETH_RX_OFFLOAD_TCP_CKSUM;
+ }
return 0;
}