> -----Original Message-----
> From: yang_y...@163.com <yang_y...@163.com>
> Sent: Monday, November 16, 2020 9:11 AM
> To: dev@dpdk.org
> Cc: Hu, Jiayu <jiayu...@intel.com>; Ananyev, Konstantin
> <konstantin.anan...@intel.com>; tho...@monjalon.net;
> yangy...@inspur.com; yang_y...@163.com
> Subject: [PATCH v3] gso: add VXLAN UDP GSO support
>
> From: Yi Yang <yangy...@inspur.com>
>
> Many NICs can't offload VXLAN UFO, so it is very important
> to do VXLAN UDP GSO by software to improve VM-to-VM UDP
> performance, especially for the case that VM MTU is just
> 1500 but not 9000.
>
> With this enabled in DPDK, OVS DPDK can leverage it to
> improve VM-to-VM UDP performance, performance gain is very
> huge, over 2 times.
GSO means software segmentation, so we don't say "VXLAN UDP GSO
by software".
I think it's better to make the commit message and log more accurate.
For example:
gso: add VxLAN UDP/IPv4 support
As some NICs do not support segmentation for VxLAN-encapsulated
UDP/IPv4 packets, this patch adds VxLAN UDP/IPv4 GSO support.
With leveraging VxLAN UDP/IPv4 GSO, OVS DPDK can significantly
improve performance for VxLAN UDP/IPv4 traffic.
Thanks,
Jiayu
>
> Signed-off-by: Yi Yang <yangy...@inspur.com>
> ---
> Changelog:
>
> v2 -> v3:
> - Correct gso type check for UDP TSO.
>
> v1 -> v2:
> - Remove condition check for outer udp header because it
> is always true for VXLAN.
> - Remove inner udp header update because it is wrong and
> unnecessary.
>
> ---
> .../generic_segmentation_offload_lib.rst | 14 ++--
> doc/guides/rel_notes/release_20_11.rst | 4 +
> lib/librte_gso/gso_common.h | 5 ++
> lib/librte_gso/gso_tunnel_udp4.c | 97
> ++++++++++++++++++++++
> lib/librte_gso/gso_tunnel_udp4.h | 44 ++++++++++
> lib/librte_gso/meson.build | 2 +-
> lib/librte_gso/rte_gso.c | 8 ++
> 7 files changed, 166 insertions(+), 8 deletions(-)
> create mode 100644 lib/librte_gso/gso_tunnel_udp4.c
> create mode 100644 lib/librte_gso/gso_tunnel_udp4.h
>
> diff --git a/doc/guides/prog_guide/generic_segmentation_offload_lib.rst
> b/doc/guides/prog_guide/generic_segmentation_offload_lib.rst
> index ad91c6e..c1d06e1 100644
> --- a/doc/guides/prog_guide/generic_segmentation_offload_lib.rst
> +++ b/doc/guides/prog_guide/generic_segmentation_offload_lib.rst
> @@ -46,7 +46,7 @@ Limitations
> - TCP
> - UDP
> - VxLAN
> - - GRE
> + - GRE TCP
>
> See `Supported GSO Packet Types`_ for further details.
>
> @@ -157,14 +157,14 @@ does not modify it during segmentation. Therefore,
> after UDP GSO, only the
> first output packet has the original UDP header, and others just have l2
> and l3 headers.
>
> -VxLAN GSO
> -~~~~~~~~~
> +VxLAN IPv4 GSO
> +~~~~~~~~~~~~~~
> VxLAN packets GSO supports segmentation of suitably large VxLAN packets,
> -which contain an outer IPv4 header, inner TCP/IPv4 headers, and optional
> -inner and/or outer VLAN tag(s).
> +which contain an outer IPv4 header, inner TCP/IPv4 or UDP/IPv4 headers,
> and
> +optional inner and/or outer VLAN tag(s).
>
> -GRE GSO
> -~~~~~~~
> +GRE TCP/IPv4 GSO
> +~~~~~~~~~~~~~~~~
> GRE GSO supports segmentation of suitably large GRE packets, which
> contain
> an outer IPv4 header, inner TCP/IPv4 headers, and an optional VLAN tag.
>
> diff --git a/doc/guides/rel_notes/release_20_11.rst
> b/doc/guides/rel_notes/release_20_11.rst
> index 24cedba..bd13c5f 100644
> --- a/doc/guides/rel_notes/release_20_11.rst
> +++ b/doc/guides/rel_notes/release_20_11.rst
> @@ -422,6 +422,10 @@ New Features
> leverage IOAT DMA channel with vhost asynchronous APIs.
> See the :doc:`../sample_app_ug/vhost` for more details.
>
> +* **Added VxLAN UDP/IPv4 GSO support.**
> +
> + Added inner UDP/IPv4 support for VxLAN IPv4 GSO.
> +
>
> Removed Items
> -------------
> diff --git a/lib/librte_gso/gso_common.h b/lib/librte_gso/gso_common.h
> index a0b8343..4d5f303 100644
> --- a/lib/librte_gso/gso_common.h
> +++ b/lib/librte_gso/gso_common.h
> @@ -26,6 +26,11 @@
> (PKT_TX_TCP_SEG | PKT_TX_IPV4 | PKT_TX_OUTER_IPV4 | \
> PKT_TX_TUNNEL_VXLAN))
>
> +#define IS_IPV4_VXLAN_UDP4(flag) (((flag) & (PKT_TX_UDP_SEG |
> PKT_TX_IPV4 | \
> + PKT_TX_OUTER_IPV4 |
> PKT_TX_TUNNEL_MASK)) == \
> + (PKT_TX_UDP_SEG | PKT_TX_IPV4 | PKT_TX_OUTER_IPV4 | \
> + PKT_TX_TUNNEL_VXLAN))
> +
> #define IS_IPV4_GRE_TCP4(flag) (((flag) & (PKT_TX_TCP_SEG | PKT_TX_IPV4
> | \
> PKT_TX_OUTER_IPV4 |
> PKT_TX_TUNNEL_MASK)) == \
> (PKT_TX_TCP_SEG | PKT_TX_IPV4 | PKT_TX_OUTER_IPV4 | \
> diff --git a/lib/librte_gso/gso_tunnel_udp4.c
> b/lib/librte_gso/gso_tunnel_udp4.c
> new file mode 100644
> index 0000000..1fc7a8d
> --- /dev/null
> +++ b/lib/librte_gso/gso_tunnel_udp4.c
> @@ -0,0 +1,97 @@
> +/* SPDX-License-Identifier: BSD-3-Clause
> + * Copyright(c) 2020 Inspur Corporation
> + */
> +
> +#include "gso_common.h"
> +#include "gso_tunnel_udp4.h"
> +
> +#define IPV4_HDR_MF_BIT (1U << 13)
> +
> +static void
> +update_tunnel_ipv4_udp_headers(struct rte_mbuf *pkt, struct rte_mbuf
> **segs,
> + uint16_t nb_segs)
> +{
> + struct rte_ipv4_hdr *ipv4_hdr;
> + uint16_t outer_id, inner_id, tail_idx, i, length;
> + uint16_t outer_ipv4_offset, inner_ipv4_offset;
> + uint16_t outer_udp_offset;
> + uint16_t frag_offset = 0, is_mf;
> +
> + outer_ipv4_offset = pkt->outer_l2_len;
> + outer_udp_offset = outer_ipv4_offset + pkt->outer_l3_len;
> + inner_ipv4_offset = outer_udp_offset + pkt->l2_len;
> +
> + /* Outer IPv4 header. */
> + ipv4_hdr = (struct rte_ipv4_hdr *)(rte_pktmbuf_mtod(pkt, char *) +
> + outer_ipv4_offset);
> + outer_id = rte_be_to_cpu_16(ipv4_hdr->packet_id);
> +
> + /* Inner IPv4 header. */
> + ipv4_hdr = (struct rte_ipv4_hdr *)(rte_pktmbuf_mtod(pkt, char *) +
> + inner_ipv4_offset);
> + inner_id = rte_be_to_cpu_16(ipv4_hdr->packet_id);
> +
> + tail_idx = nb_segs - 1;
> +
> + for (i = 0; i < nb_segs; i++) {
> + update_ipv4_header(segs[i], outer_ipv4_offset, outer_id);
> + update_udp_header(segs[i], outer_udp_offset);
> + update_ipv4_header(segs[i], inner_ipv4_offset, inner_id);
> + /* For the case inner packet is UDP, we must keep UDP
> + * datagram boundary, it must be handled as IP fragment.
> + *
> + * Set IP fragment offset for inner IP header.
> + */
> + ipv4_hdr = (struct rte_ipv4_hdr *)
> + (rte_pktmbuf_mtod(segs[i], char *) +
> + inner_ipv4_offset);
> + is_mf = i < tail_idx ? IPV4_HDR_MF_BIT : 0;
> + ipv4_hdr->fragment_offset =
> + rte_cpu_to_be_16(frag_offset | is_mf);
> + length = segs[i]->pkt_len - inner_ipv4_offset - pkt->l3_len;
> + frag_offset += (length >> 3);
> + outer_id++;
> + }
> +}
> +
> +int
> +gso_tunnel_udp4_segment(struct rte_mbuf *pkt,
> + uint16_t gso_size,
> + struct rte_mempool *direct_pool,
> + struct rte_mempool *indirect_pool,
> + struct rte_mbuf **pkts_out,
> + uint16_t nb_pkts_out)
> +{
> + struct rte_ipv4_hdr *inner_ipv4_hdr;
> + uint16_t pyld_unit_size, hdr_offset, frag_off;
> + int ret;
> +
> + hdr_offset = pkt->outer_l2_len + pkt->outer_l3_len + pkt->l2_len;
> + inner_ipv4_hdr = (struct rte_ipv4_hdr *)(rte_pktmbuf_mtod(pkt,
> char *) +
> + hdr_offset);
> + /*
> + * Don't process the packet whose MF bit or offset in the inner
> + * IPv4 header are non-zero.
> + */
> + frag_off = rte_be_to_cpu_16(inner_ipv4_hdr->fragment_offset);
> + if (unlikely(IS_FRAGMENTED(frag_off)))
> + return 0;
> +
> + hdr_offset += pkt->l3_len;
> + /* Don't process the packet without data */
> + if ((hdr_offset + pkt->l4_len) >= pkt->pkt_len)
> + return 0;
> +
> + /* pyld_unit_size must be a multiple of 8 because frag_off
> + * uses 8 bytes as unit.
> + */
> + pyld_unit_size = (gso_size - hdr_offset) & ~7U;
> +
> + /* Segment the payload */
> + ret = gso_do_segment(pkt, hdr_offset, pyld_unit_size, direct_pool,
> + indirect_pool, pkts_out, nb_pkts_out);
> + if (ret > 1)
> + update_tunnel_ipv4_udp_headers(pkt, pkts_out, ret);
> +
> + return ret;
> +}
> diff --git a/lib/librte_gso/gso_tunnel_udp4.h
> b/lib/librte_gso/gso_tunnel_udp4.h
> new file mode 100644
> index 0000000..c49b43f
> --- /dev/null
> +++ b/lib/librte_gso/gso_tunnel_udp4.h
> @@ -0,0 +1,44 @@
> +/* SPDX-License-Identifier: BSD-3-Clause
> + * Copyright(c) 2020 Inspur Corporation
> + */
> +
> +#ifndef _GSO_TUNNEL_UDP4_H_
> +#define _GSO_TUNNEL_UDP4_H_
> +
> +#include <stdint.h>
> +#include <rte_mbuf.h>
> +
> +/**
> + * Segment a tunneling packet with inner UDP/IPv4 headers. This function
> + * does not check if the input packet has correct checksums, and does not
> + * update checksums for output GSO segments. Furthermore, it does not
> + * process IP fragment packets.
> + *
> + * @param pkt
> + * The packet mbuf to segment.
> + * @param gso_size
> + * The max length of a GSO segment, measured in bytes.
> + * @param direct_pool
> + * MBUF pool used for allocating direct buffers for output segments.
> + * @param indirect_pool
> + * MBUF pool used for allocating indirect buffers for output segments.
> + * @param pkts_out
> + * Pointer array used to store the MBUF addresses of output GSO
> + * segments, when it succeeds. If the memory space in pkts_out is
> + * insufficient, it fails and returns -EINVAL.
> + * @param nb_pkts_out
> + * The max number of items that 'pkts_out' can keep.
> + *
> + * @return
> + * - The number of GSO segments filled in pkts_out on success.
> + * - Return 0 if it needn't GSO.
> + * - Return -ENOMEM if run out of memory in MBUF pools.
> + * - Return -EINVAL for invalid parameters.
> + */
> +int gso_tunnel_udp4_segment(struct rte_mbuf *pkt,
> + uint16_t gso_size,
> + struct rte_mempool *direct_pool,
> + struct rte_mempool *indirect_pool,
> + struct rte_mbuf **pkts_out,
> + uint16_t nb_pkts_out);
> +#endif
> diff --git a/lib/librte_gso/meson.build b/lib/librte_gso/meson.build
> index ad8dd85..05904f2 100644
> --- a/lib/librte_gso/meson.build
> +++ b/lib/librte_gso/meson.build
> @@ -2,6 +2,6 @@
> # Copyright(c) 2017 Intel Corporation
>
> sources = files('gso_common.c', 'gso_tcp4.c', 'gso_udp4.c',
> - 'gso_tunnel_tcp4.c', 'rte_gso.c')
> + 'gso_tunnel_tcp4.c', 'gso_tunnel_udp4.c', 'rte_gso.c')
> headers = files('rte_gso.h')
> deps += ['ethdev']
> diff --git a/lib/librte_gso/rte_gso.c b/lib/librte_gso/rte_gso.c
> index 896350e..0d02ec3 100644
> --- a/lib/librte_gso/rte_gso.c
> +++ b/lib/librte_gso/rte_gso.c
> @@ -11,6 +11,7 @@
> #include "gso_common.h"
> #include "gso_tcp4.h"
> #include "gso_tunnel_tcp4.h"
> +#include "gso_tunnel_udp4.h"
> #include "gso_udp4.h"
>
> #define ILLEGAL_UDP_GSO_CTX(ctx) \
> @@ -60,6 +61,13 @@
> ret = gso_tunnel_tcp4_segment(pkt, gso_size, ipid_delta,
> direct_pool, indirect_pool,
> pkts_out, nb_pkts_out);
> + } else if (IS_IPV4_VXLAN_UDP4(pkt->ol_flags) &&
> + (gso_ctx->gso_types &
> DEV_TX_OFFLOAD_VXLAN_TNL_TSO) &&
> + (gso_ctx->gso_types & DEV_TX_OFFLOAD_UDP_TSO))
> {
> + pkt->ol_flags &= (~PKT_TX_UDP_SEG);
> + ret = gso_tunnel_udp4_segment(pkt, gso_size,
> + direct_pool, indirect_pool,
> + pkts_out, nb_pkts_out);
> } else if (IS_IPV4_TCP(pkt->ol_flags) &&
> (gso_ctx->gso_types & DEV_TX_OFFLOAD_TCP_TSO))
> {
> pkt->ol_flags &= (~PKT_TX_TCP_SEG);
> --
> 1.8.3.1
