support VXLAN-GPE header fields(flags, rsvd0 and rsvd1) matching
cover support VXLAN-GPE header fields(flags, rsvd0 and rsvd1) matching
Previously, VXLAN-GPE in DPDK only supports VNI and next protocol header fields. This patch series add support for flags and reserved field 0 and 1.
Below is the VXLAN-GPE header defined in the lasted draft.
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |R|R|Ver|I|P|B|O| Reserved |Next Protocol | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | VXLAN Network Identifier (VNI) | Reserved | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Gavin Li (5):
net/mlx5: support VXLAN-GPE flags matching app/testpmd: support VXLAN-GPE flags net/mlx5: support VXLAN-GPE reserved fields matching app/testpmd: support VXLAN-GPE reserved fields net/mlx5/hws: support VXLAN-GPE matching
app/test-pmd/cmdline_flow.c | 30 +++++ doc/guides/nics/mlx5.rst | 5 + doc/guides/rel_notes/release_24_03.rst | 6 + doc/guides/testpmd_app_ug/testpmd_funcs.rst | 3 + drivers/net/mlx5/hws/mlx5dr_definer.c | 117 ++++++++++++++++++++ drivers/net/mlx5/hws/mlx5dr_definer.h | 13 +++ drivers/net/mlx5/mlx5_flow.c | 6 + drivers/net/mlx5/mlx5_flow_dv.c | 32 +++++- drivers/net/mlx5/mlx5_flow_hw.c | 1 + 9 files changed, 207 insertions(+), 6 deletions(-)
Signature · 1 line
2.39.1
PATCH 1/5 net/mlx5: support VXLAN-GPE flags matching
This commit adds support for matching on the flags field of VXLAN-GPE header (the first 8-bits).
--- drivers/net/mlx5/mlx5_flow.c | 1 + 1 file changed, 1 insertion(+)
drivers/net/mlx5/mlx5_flow.c +1 -0
@@ -3292,6 +3292,7 @@ mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,struct rte_flow_item_vxlan_gpe nic_mask = { .vni ="\xff\xff\xff" , .protocol =0xff , + .flags =0xff , };if (!priv->sh->config.l3_vxlan_en)
Signature · 1 line
2.39.1
PATCH 2/5 app/testpmd: support VXLAN-GPE flags
In reply to Gavin LiAdd support for VXLAN-GPE flags field in flow rule pattern to testpmd.
--- app/test-pmd/cmdline_flow.c | 10 ++++++++++ doc/guides/testpmd_app_ug/testpmd_funcs.rst | 1 + 2 files changed, 11 insertions(+)
app/test-pmd/cmdline_flow.c +10 -0
@@ -401,6 +401,7 @@ enum index {ITEM_VXLAN_GPE ,ITEM_VXLAN_GPE_VNI ,ITEM_VXLAN_GPE_PROTO , +ITEM_VXLAN_GPE_FLAGS ,ITEM_ARP_ETH_IPV4 ,ITEM_ARP_ETH_IPV4_SHA , ITEM_ARP_ETH_IPV4_SPA,
@@ -1789,6 +1790,7 @@ static const enum index item_geneve[] = {static const enum index item_vxlan_gpe[] = {ITEM_VXLAN_GPE_VNI ,ITEM_VXLAN_GPE_PROTO , +ITEM_VXLAN_GPE_FLAGS ,ITEM_NEXT ,ZERO , };
@@ -4936,6 +4938,14 @@ static const struct token token_list[] = { .args = ARGS(ARGS_ENTRY_HTON(struct rte_flow_item_vxlan_gpe, protocol)), }, + [ITEM_VXLAN_GPE_FLAGS] = { + .name = "flags", + .help = "VXLAN-GPE flags", + .next =NEXT (item_vxlan_gpe,NEXT_ENTRY (COMMON_UNSIGNED ), + item_param), + .args = ARGS(ARGS_ENTRY_HTON(struct rte_flow_item_vxlan_gpe, + flags)), + }, [ITEM_ARP_ETH_IPV4] = { .name = "arp_eth_ipv4", .help = "match ARP header for Ethernet/IPv4",
doc/guides/testpmd_app_ug/testpmd_funcs.rst +1 -0
@@ -3703,6 +3703,7 @@ This section lists supported pattern items and their attributes, if any. - ``vxlan-gpe``: match VXLAN-GPE header. - ``vni {unsigned}``: VXLAN-GPE identifier. + - ``flags {unsigned}``: VXLAN-GPE flags. - ``arp_eth_ipv4``: match ARP header for Ethernet/IPv4.
Signature · 1 line
2.39.1
Re: [V1 2/5] app/testpmd: support VXLAN-GPE flags
Quoted 4 lines · Gavin Li
Add support for VXLAN-GPE flags field in flow rule pattern to testpmd.
Signed-off-by Gavin Li <[email protected]>
<snip>
PATCH 3/5 net/mlx5: support VXLAN-GPE reserved fields matching
In reply to Gavin LiThis adds matching on the reserved fields of VXLAN-GPE header (the 16-bits before Next Protocol and the last 8-bits).
To support all the header fields, tunnel_header_0_1 should be supported by FW and misc5_cap is set.
If one of the reserved fields is matched on, misc5 is used for matching. Otherwise, keep using misc3
--- doc/guides/nics/mlx5.rst | 5 +++++ drivers/net/mlx5/mlx5_flow.c | 5 +++++ drivers/net/mlx5/mlx5_flow_dv.c | 32 ++++++++++++++++++++++++++------ 3 files changed, 36 insertions(+), 6 deletions(-)
doc/guides/nics/mlx5.rst +5 -0
@@ -311,6 +311,11 @@ Limitations Group zero's behavior may differ which depends on FW. Matching value equals 0 (value & mask) is not supported. +- Matching on VXLAN-GPE header fields: + + - ``rsvd0``/``rsvd1`` matching support depends on FW version when using DV flow + engine (``dv_flow_en`` = 1). + - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP. - MPLSoGRE is not supported in HW steering (``dv_flow_en`` = 2).
drivers/net/mlx5/mlx5_flow.c +5 -0
@@ -3316,6 +3316,11 @@ mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item, "no outer UDP layer found");if (!mask) mask = &rte_flow_item_vxlan_gpe_mask; +if (priv->sh->misc5_cap && priv->sh->tunnel_header_0_1) { + nic_mask.rsvd0[0 ] =0xff ; + nic_mask.rsvd0[1 ] =0xff ; + nic_mask.rsvd1 =0xff ; + } ret = mlx5_flow_item_acceptable (item, (const uint8_t *)mask, (const uint8_t *)&nic_mask,
drivers/net/mlx5/mlx5_flow_dv.c +26 -6
@@ -9813,14 +9813,10 @@ flow_dv_translate_item_vxlan_gpe(void *key, const struct rte_flow_item *item, vxlan_v = vxlan_m; else if (key_type == MLX5_SET_MATCHER_HS_V) vxlan_m = vxlan_v; -for (i =0 ; i < size; ++i) - vni_v[i] = vxlan_m->hdr.vni[i] & vxlan_v->hdr.vni[i];if (vxlan_m->hdr.flags) { flags_m = vxlan_m->hdr.flags; flags_v = vxlan_v->hdr.flags; } -MLX5_SET (fte_match_set_misc3, misc_v, outer_vxlan_gpe_flags, - flags_m & flags_v); m_protocol = vxlan_m->hdr.protocol; v_protocol = vxlan_v->hdr.protocol;if (!m_protocol) {
@@ -9839,8 +9835,32 @@ flow_dv_translate_item_vxlan_gpe(void *key, const struct rte_flow_item *item,if (key_type &MLX5_SET_MATCHER_M ) v_protocol = m_protocol; } -MLX5_SET (fte_match_set_misc3, misc_v, - outer_vxlan_gpe_next_protocol, m_protocol & v_protocol); +/* +* If only match flags/protocol/vni field, keep using misc3 for matching. +* If need to match rsvd0 or rsvd1, using misc5 and do not need using misc3. +*/ +if (!(vxlan_m->hdr.rsvd0[0 ] || vxlan_m->hdr.rsvd0[1 ] || vxlan_m->hdr.rsvd1)) { +for (i =0 ; i < size; ++i) + vni_v[i] = vxlan_m->hdr.vni[i] & vxlan_v->hdr.vni[i]; +MLX5_SET (fte_match_set_misc3, misc_v, outer_vxlan_gpe_flags, + flags_m & flags_v); +MLX5_SET (fte_match_set_misc3, misc_v, + outer_vxlan_gpe_next_protocol, m_protocol & v_protocol); + }else { +uint32_t tunnel_v; +void *misc5_v =MLX5_ADDR_OF (fte_match_param, key, misc_parameters_5); + + tunnel_v = (flags_m & flags_v) <<24 | + (vxlan_v->hdr.rsvd0[0 ] & vxlan_m->hdr.rsvd0[0 ]) <<16 | + (vxlan_v->hdr.rsvd0[1 ] & vxlan_m->hdr.rsvd0[1 ]) <<8 | + (m_protocol & v_protocol); +MLX5_SET (fte_match_set_misc5, misc5_v, tunnel_header_0, tunnel_v); + tunnel_v = (vxlan_v->hdr.vni[0 ] & vxlan_m->hdr.vni[0 ]) <<24 | + (vxlan_v->hdr.vni[1 ] & vxlan_m->hdr.vni[1 ]) <<16 | + (vxlan_v->hdr.vni[2 ] & vxlan_m->hdr.vni[2 ]) <<8 | + (vxlan_v->hdr.rsvd1 & vxlan_m->hdr.rsvd1); +MLX5_SET (fte_match_set_misc5, misc5_v, tunnel_header_1, tunnel_v); + } }/**
Signature · 1 line
2.39.1
PATCH 4/5 app/testpmd: support VXLAN-GPE reserved fields
In reply to Gavin LiAdd support for VXLAN-GPE rsvd0 and rsvd1 fields in flow rule pattern to testpmd
--- app/test-pmd/cmdline_flow.c | 20 ++++++++++++++++++++ doc/guides/testpmd_app_ug/testpmd_funcs.rst | 2 ++ 2 files changed, 22 insertions(+)
app/test-pmd/cmdline_flow.c +20 -0
@@ -402,6 +402,8 @@ enum index {ITEM_VXLAN_GPE_VNI ,ITEM_VXLAN_GPE_PROTO ,ITEM_VXLAN_GPE_FLAGS , +ITEM_VXLAN_GPE_RSVD0 , +ITEM_VXLAN_GPE_RSVD1 ,ITEM_ARP_ETH_IPV4 ,ITEM_ARP_ETH_IPV4_SHA , ITEM_ARP_ETH_IPV4_SPA,
@@ -1791,6 +1793,8 @@ static const enum index item_vxlan_gpe[] = {ITEM_VXLAN_GPE_VNI ,ITEM_VXLAN_GPE_PROTO ,ITEM_VXLAN_GPE_FLAGS , +ITEM_VXLAN_GPE_RSVD0 , +ITEM_VXLAN_GPE_RSVD1 ,ITEM_NEXT , ZERO, };
@@ -4946,6 +4950,22 @@ static const struct token token_list[] = { .args = ARGS(ARGS_ENTRY_HTON(struct rte_flow_item_vxlan_gpe, flags)), }, + [ITEM_VXLAN_GPE_RSVD0] = { + .name = "rsvd0", + .help = "VXLAN-GPE rsvd0", + .next =NEXT (item_vxlan_gpe,NEXT_ENTRY (COMMON_UNSIGNED ), + item_param), + .args = ARGS(ARGS_ENTRY_HTON(struct rte_flow_item_vxlan_gpe, + rsvd0)), + }, + [ITEM_VXLAN_GPE_RSVD1] = { + .name = "rsvd1", + .help = "VXLAN-GPE rsvd1", + .next =NEXT (item_vxlan_gpe,NEXT_ENTRY (COMMON_UNSIGNED ), + item_param), + .args = ARGS(ARGS_ENTRY_HTON(struct rte_flow_item_vxlan_gpe, + rsvd1)), + }, [ITEM_ARP_ETH_IPV4] = { .name = "arp_eth_ipv4", .help = "match ARP header for Ethernet/IPv4",
doc/guides/testpmd_app_ug/testpmd_funcs.rst +2 -0
@@ -3704,6 +3704,8 @@ This section lists supported pattern items and their attributes, if any. - ``vni {unsigned}``: VXLAN-GPE identifier. - ``flags {unsigned}``: VXLAN-GPE flags. + - ``rsvd0 {unsigned}``: VXLAN-GPE reserved field 0. + - ``rsvd1 {unsigned}``: VXLAN-GPE reserved field 1. - ``arp_eth_ipv4``: match ARP header for Ethernet/IPv4.
Signature · 1 line
2.39.1
Re: [V1 4/5] app/testpmd: support VXLAN-GPE reserved fields
Quoted 5 lines · Gavin Li
Add support for VXLAN-GPE rsvd0 and rsvd1 fields in flow rule pattern to testpmd
Signed-off-by Gavin Li <[email protected]>
Quoted 1 line
---
<snip>
PATCH 5/5 net/mlx5/hws: support VXLAN-GPE matching
In reply to Gavin LiAdd support for matching VXLAN-GPE tunnel header.
--- doc/guides/rel_notes/release_24_03.rst | 6 ++ drivers/net/mlx5/hws/mlx5dr_definer.c | 117 +++++++++++++++++++++++++ drivers/net/mlx5/hws/mlx5dr_definer.h | 13 +++ drivers/net/mlx5/mlx5_flow_hw.c | 1 + 4 files changed, 137 insertions(+)
doc/guides/rel_notes/release_24_03.rst +6 -0
@@ -59,6 +59,12 @@ New Features * Added support for accumulating from src field to dst field. + * Added support for VXLAN-GPE flags/rsvd0/rsvd fields matching in DV flow + engine (``dv_flow_en`` = 1). + + * Added support for VXLAN-GPE matching in HW Steering flow engine + (``dv_flow_en`` = 2). + Removed Items -------------
drivers/net/mlx5/hws/mlx5dr_definer.c +117 -0
147 lines not on this page · raw message
drivers/net/mlx5/hws/mlx5dr_definer.h +13 -0
@@ -91,6 +91,11 @@ enum mlx5dr_definer_fname {MLX5DR_DEFINER_FNAME_VPORT_REG_C_0 ,MLX5DR_DEFINER_FNAME_VXLAN_FLAGS ,MLX5DR_DEFINER_FNAME_VXLAN_VNI , +MLX5DR_DEFINER_FNAME_VXLAN_GPE_FLAGS , +MLX5DR_DEFINER_FNAME_VXLAN_GPE_RSVD0 , +MLX5DR_DEFINER_FNAME_VXLAN_GPE_PROTO , +MLX5DR_DEFINER_FNAME_VXLAN_GPE_VNI , +MLX5DR_DEFINER_FNAME_VXLAN_GPE_RSVD1 ,MLX5DR_DEFINER_FNAME_SOURCE_QP ,MLX5DR_DEFINER_FNAME_REG_0 , MLX5DR_DEFINER_FNAME_REG_1,
@@ -593,6 +598,14 @@ struct mlx5_ifc_header_vxlan_bits {u8 reserved2[0x8 ]; }; +struct mlx5_ifc_header_vxlan_gpe_bits { +u8 flags[0x8 ]; +u8 rsvd0[0x10 ]; +u8 protocol[0x8 ]; +u8 vni[0x18 ]; +u8 rsvd1[0x8 ]; +}; +struct mlx5_ifc_header_gre_bits {union {u8 c_rsvd0_ver[0x10 ];
drivers/net/mlx5/mlx5_flow_hw.c +1 -0
@@ -6827,6 +6827,7 @@ flow_hw_pattern_validate(struct rte_eth_dev *dev,case RTE_FLOW_ITEM_TYPE_GTP :case RTE_FLOW_ITEM_TYPE_GTP_PSC :case RTE_FLOW_ITEM_TYPE_VXLAN : +case RTE_FLOW_ITEM_TYPE_VXLAN_GPE :case RTE_FLOW_ITEM_TYPE_MPLS :case MLX5_RTE_FLOW_ITEM_TYPE_SQ :case RTE_FLOW_ITEM_TYPE_GRE :
Signature · 1 line
2.39.1
Hi,
Quoted 44 lines
-----Original Message-----
From: Gavin Li <[email protected]>
Sent: Friday, January 12, 2024 10:02 AM
To: [email protected]; NBU-Contact-Thomas Monjalon (EXTERNAL) <[email protected]>; Ori Kam <[email protected]>; [email protected]; [email protected]; Dariusz Sosnowski <[email protected]>; Slava Ovsiienko <[email protected]>; Suanming Mou <[email protected]>; Matan Azrad <[email protected]>
Cc: Jiawei(Jonny) Wang <[email protected]>; Raslan Darawsheh <[email protected]>
Subject: [V1 0/5] support VXLAN-GPE headerfields(flags, rsvd0 and rsvd1)matchingPreviously, VXLAN-GPE in DPDK only supports VNI and next protocol header fields. This patch series add support for flags and reserved field 0 and 1.
Below is the VXLAN-GPE header defined in the lasted draft.
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |R|R|Ver|I|P|B|O| Reserved |Next Protocol | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | VXLAN Network Identifier (VNI) | Reserved | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+Gavin Li (5):
net/mlx5: support VXLAN-GPE flags matching app/testpmd: support VXLAN-GPE flags net/mlx5: support VXLAN-GPE reserved fields matching app/testpmd: support VXLAN-GPE reserved fieldsnet/mlx5/hws: support VXLAN-GPE matchingapp/test-pmd/cmdline_flow.c | 30 +++++ doc/guides/nics/mlx5.rst | 5 + doc/guides/rel_notes/release_24_03.rst | 6 + doc/guides/testpmd_app_ug/testpmd_funcs.rst | 3 + drivers/net/mlx5/hws/mlx5dr_definer.c | 117 ++++++++++++++++++++ drivers/net/mlx5/hws/mlx5dr_definer.h | 13 +++ drivers/net/mlx5/mlx5_flow.c | 6 + drivers/net/mlx5/mlx5_flow_dv.c | 32 +++++- drivers/net/mlx5/mlx5_flow_hw.c | 1 + 9 files changed, 207 insertions(+), 6 deletions(-)
Quoted 2 lines
--
2.39.1
Series applied to next-net-mlx,
Kindest regards
Raslan Darawsheh
Quoted 4 lines · On Fri, 12 Jan 2024 10:02:05 +0200 Gavin Li <[email protected]> wrote:
Previously, VXLAN-GPE in DPDK only supports VNI and next protocol header fields. This patch series add support for flags and reserved field 0 and 1.
Quoted 8 lines
Below is the VXLAN-GPE header defined in the lasted draft.
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |R|R|Ver|I|P|B|O| Reserved |Next Protocol | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | VXLAN Network Identifier (VNI) | Reserved | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Would recommend against implementing anything in a draft RFC. Things can change. Learned the hard way when doing VXLAN driver for Linux. The hardcoded port value in the Linux VXLAN driver is wrong because it matched the draft RFC (got changed in final version). Because of strict compatibility requirements the Linux driver could not be changed to the correct value.
19/02/2024 20:50, Stephen Hemminger:
Quoted 16 lines
Quoted 4 lines · On Fri, 12 Jan 2024 10:02:05 +0200 Gavin Li <[email protected]> wrote:
Previously, VXLAN-GPE in DPDK only supports VNI and next protocol header fields. This patch series add support for flags and reserved field 0 and 1.
Quoted 8 lines
Below is the VXLAN-GPE header defined in the lasted draft.
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |R|R|Ver|I|P|B|O| Reserved |Next Protocol | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | VXLAN Network Identifier (VNI) | Reserved | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Quoted 5 lines
Would recommend against implementing anything in a draft RFC. Things can change. Learned the hard way when doing VXLAN driver for Linux. The hardcoded port value in the Linux VXLAN driver is wrong because it matched the draft RFC (got changed in final version). Because of strict compatibility requirements the Linux driver could not be changed to the correct value.
The problem is that standardization may be slow. Would it be acceptable without any compatibility guarantee?
Quoted 17 lines · On Mon, 19 Feb 2024 23:48:31 +0100 Thomas Monjalon <[email protected]> wrote:
19/02/2024 20:50, Stephen Hemminger:
Quoted 16 lines
Quoted 12 lines · On Fri, 12 Jan 2024 10:02:05 +0200 Gavin Li <[email protected]> wrote:
Previously, VXLAN-GPE in DPDK only supports VNI and next protocol header fields. This patch series add support for flags and reserved field 0 and 1.
Below is the VXLAN-GPE header defined in the lasted draft.
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |R|R|Ver|I|P|B|O| Reserved |Next Protocol | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | VXLAN Network Identifier (VNI) | Reserved | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Quoted 8 lines
Quoted 5 lines
Would recommend against implementing anything in a draft RFC. Things can change. Learned the hard way when doing VXLAN driver for Linux. The hardcoded port value in the Linux VXLAN driver is wrong because it matched the draft RFC (got changed in final version). Because of strict compatibility requirements the Linux driver could not be changed to the correct value.
The problem is that standardization may be slow. Would it be acceptable without any compatibility guarantee?
Never mark it stable until RFC is done.