From patchwork Tue Sep 10 09:59:37 2019 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Hyong Youb Kim (hyonkim)" X-Patchwork-Id: 59094 X-Patchwork-Delegate: ferruh.yigit@amd.com Return-Path: X-Original-To: patchwork@dpdk.org Delivered-To: patchwork@dpdk.org Received: from [92.243.14.124] (localhost [127.0.0.1]) by dpdk.org (Postfix) with ESMTP id 72B141ED84; Tue, 10 Sep 2019 12:01:07 +0200 (CEST) Received: from alln-iport-8.cisco.com (alln-iport-8.cisco.com [173.37.142.95]) by dpdk.org (Postfix) with ESMTP id 12AC71EBA6 for ; Tue, 10 Sep 2019 12:01:04 +0200 (CEST) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=cisco.com; i=@cisco.com; l=96791; q=dns/txt; s=iport; t=1568109665; x=1569319265; h=from:to:cc:subject:date:message-id:in-reply-to: references:mime-version:content-transfer-encoding; bh=ss3PAW4rq7zSBqXHZfiDAyfd0YkjffV4QGBTKVLlZkM=; b=k9bAH+vEeNpIPUVSQFQ9QnXqHM9AoGdRQxmoWWLVRYTEx/toq9iK/A0d 35R4fbaROmGsAU4eXq0LQM6a4fEF+62EZ0erFJuIx8vcD9GlIROmwbz7c mbExSEgLYYU04DoWO2O1namLnd+9gerVcxotH+BdYMN0Rq2VPPTKeOVB9 g=; X-IronPort-AV: E=Sophos;i="5.64,489,1559520000"; d="scan'208";a="328864727" Received: from rcdn-core-9.cisco.com ([173.37.93.145]) by alln-iport-8.cisco.com with ESMTP/TLS/DHE-RSA-SEED-SHA; 10 Sep 2019 10:01:02 +0000 Received: from cisco.com (savbu-usnic-a.cisco.com [10.193.184.48]) by rcdn-core-9.cisco.com (8.15.2/8.15.2) with ESMTP id x8AA12Xx023494; Tue, 10 Sep 2019 10:01:02 GMT Received: by cisco.com (Postfix, from userid 508933) id 1CEC720F2003; Tue, 10 Sep 2019 03:01:02 -0700 (PDT) From: Hyong Youb Kim To: Ferruh Yigit Cc: dev@dpdk.org, John Daley , Hyong Youb Kim Date: Tue, 10 Sep 2019 02:59:37 -0700 Message-Id: <20190910095937.9731-3-hyonkim@cisco.com> X-Mailer: git-send-email 2.22.0 In-Reply-To: <20190910095937.9731-1-hyonkim@cisco.com> References: <20190910095937.9731-1-hyonkim@cisco.com> MIME-Version: 1.0 X-Outbound-SMTP-Client: 10.193.184.48, savbu-usnic-a.cisco.com X-Outbound-Node: rcdn-core-9.cisco.com Subject: [dpdk-dev] [PATCH 2/2] net/enic: add flow implementation based on Flow Manager API X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" Flow Manager API is the new firmware interface that exposes match action capabilities in 1400 series VIC adapters. It is intended for virtual switch offloads and enables more advanced features than the existing filter API. For example, it supports VXLAN encap and decap actions, and exposes TCAM and exact match tables. Add the new flow implementation using Flow Manager and use it when available. When Flow Manager is not available, the driver will continue to use the old filter-based flow implementation. Signed-off-by: John Daley Signed-off-by: Hyong Youb Kim --- doc/guides/nics/enic.rst | 9 + doc/guides/rel_notes/release_19_11.rst | 1 + drivers/net/enic/Makefile | 1 + drivers/net/enic/base/vnic_dev.c | 79 +- drivers/net/enic/base/vnic_dev.h | 1 + drivers/net/enic/base/vnic_devcmd.h | 12 + drivers/net/enic/base/vnic_flowman.h | 364 ++++ drivers/net/enic/enic.h | 21 +- drivers/net/enic/enic_ethdev.c | 10 +- drivers/net/enic/enic_fm_flow.c | 2459 ++++++++++++++++++++++++ drivers/net/enic/enic_main.c | 8 +- drivers/net/enic/enic_res.c | 9 +- drivers/net/enic/meson.build | 1 + 13 files changed, 2967 insertions(+), 8 deletions(-) create mode 100644 drivers/net/enic/base/vnic_flowman.h create mode 100644 drivers/net/enic/enic_fm_flow.c diff --git a/doc/guides/nics/enic.rst b/doc/guides/nics/enic.rst index 2384cb5b8..b324d4506 100644 --- a/doc/guides/nics/enic.rst +++ b/doc/guides/nics/enic.rst @@ -260,6 +260,15 @@ Generic Flow API is supported. The baseline support is: - Selectors: 'is', 'spec' and 'mask'. 'last' is not supported - In total, up to 64 bytes of mask is allowed across all headers +- **1400 and later series VICs with Flow Manager API enabled** + + - Attributes: ingress, egress + - Items: eth, vlan, ipv4, ipv6, sctp, udp, tcp, vxlan, raw, inner eth, vlan, ipv4, ipv6, sctp, udp, tcp + - Ingress Actions: count, drop, flag, jump, mark, port_id, passthru, queue, rss, vxlan_decap, vxlan_encap, and void + - Egress Actions: count, drop, jump, passthru, vxlan_encap, and void + - Selectors: 'is', 'spec' and 'mask'. 'last' is not supported + - In total, up to 64 bytes of mask is allowed across all headers + The VIC performs packet matching after applying VLAN strip. If VLAN stripping is enabled, EtherType in the ETH item corresponds to the stripped VLAN header's EtherType. Stripping does not affect the VLAN diff --git a/doc/guides/rel_notes/release_19_11.rst b/doc/guides/rel_notes/release_19_11.rst index afe71f8fa..82e38a904 100644 --- a/doc/guides/rel_notes/release_19_11.rst +++ b/doc/guides/rel_notes/release_19_11.rst @@ -59,6 +59,7 @@ New Features * **Updated the enic driver.** * Added support for Geneve with options offload. + * Added flow API implementation based on VIC Flow Manager API. Removed Items diff --git a/drivers/net/enic/Makefile b/drivers/net/enic/Makefile index 04bae35e3..10736a567 100644 --- a/drivers/net/enic/Makefile +++ b/drivers/net/enic/Makefile @@ -32,6 +32,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_ENIC_PMD) += enic_rxtx.c SRCS-$(CONFIG_RTE_LIBRTE_ENIC_PMD) += enic_clsf.c SRCS-$(CONFIG_RTE_LIBRTE_ENIC_PMD) += enic_res.c SRCS-$(CONFIG_RTE_LIBRTE_ENIC_PMD) += enic_flow.c +SRCS-$(CONFIG_RTE_LIBRTE_ENIC_PMD) += enic_fm_flow.c SRCS-$(CONFIG_RTE_LIBRTE_ENIC_PMD) += base/vnic_cq.c SRCS-$(CONFIG_RTE_LIBRTE_ENIC_PMD) += base/vnic_wq.c SRCS-$(CONFIG_RTE_LIBRTE_ENIC_PMD) += base/vnic_dev.c diff --git a/drivers/net/enic/base/vnic_dev.c b/drivers/net/enic/base/vnic_dev.c index 807d4096f..a52f7430b 100644 --- a/drivers/net/enic/base/vnic_dev.c +++ b/drivers/net/enic/base/vnic_dev.c @@ -12,6 +12,7 @@ #include "vnic_devcmd.h" #include "vnic_nic.h" #include "vnic_stats.h" +#include "vnic_flowman.h" enum vnic_proxy_type { @@ -47,6 +48,8 @@ struct vnic_dev { dma_addr_t stats_pa; struct vnic_devcmd_fw_info *fw_info; dma_addr_t fw_info_pa; + struct fm_info *flowman_info; + dma_addr_t flowman_info_pa; enum vnic_proxy_type proxy; u32 proxy_index; u64 args[VNIC_DEVCMD_NARGS]; @@ -500,8 +503,74 @@ int vnic_dev_capable_adv_filters(struct vnic_dev *vdev) return (a1 >= (u32)FILTER_DPDK_1); } -/* Determine the "best" filtering mode VIC is capaible of. Returns one of 3 +int vnic_dev_flowman_cmd(struct vnic_dev *vdev, u64 *args, int nargs) +{ + int wait = 1000; + + return vnic_dev_cmd_args(vdev, CMD_FLOW_MANAGER_OP, args, nargs, wait); +} + +static int vnic_dev_flowman_enable(struct vnic_dev *vdev, u32 *mode, + u8 *filter_actions) +{ + char name[NAME_MAX]; + u64 args[3]; + u64 ops; + static u32 instance; + + /* flowman devcmd available? */ + if (!vnic_dev_capable(vdev, CMD_FLOW_MANAGER_OP)) + return 0; + /* Have the version we are using? */ + args[0] = FM_API_VERSION_QUERY; + if (vnic_dev_flowman_cmd(vdev, args, 1)) + return 0; + if ((args[0] & (1ULL << FM_VERSION)) == 0) + return 0; + /* Select the version */ + args[0] = FM_API_VERSION_SELECT; + args[1] = FM_VERSION; + if (vnic_dev_flowman_cmd(vdev, args, 2)) + return 0; + /* Can we get fm_info? */ + if (!vdev->flowman_info) { + snprintf((char *)name, sizeof(name), "vnic_flowman_info-%u", + instance++); + vdev->flowman_info = vdev->alloc_consistent(vdev->priv, + sizeof(struct fm_info), + &vdev->flowman_info_pa, (u8 *)name); + if (!vdev->flowman_info) + return 0; + } + args[0] = FM_INFO_QUERY; + args[1] = vdev->flowman_info_pa; + args[2] = sizeof(struct fm_info); + if (vnic_dev_flowman_cmd(vdev, args, 3)) + return 0; + /* Have required operations? */ + ops = (1ULL << FMOP_END) | + (1ULL << FMOP_DROP) | + (1ULL << FMOP_RQ_STEER) | + (1ULL << FMOP_EXACT_MATCH) | + (1ULL << FMOP_MARK) | + (1ULL << FMOP_TAG) | + (1ULL << FMOP_EG_HAIRPIN) | + (1ULL << FMOP_ENCAP) | + (1ULL << FMOP_DECAP_NOSTRIP); + if ((vdev->flowman_info->fm_op_mask & ops) != ops) + return 0; + /* Good to use flowman now */ + *mode = FILTER_FLOWMAN; + *filter_actions = FILTER_ACTION_RQ_STEERING_FLAG | + FILTER_ACTION_FILTER_ID_FLAG | + FILTER_ACTION_COUNTER_FLAG | + FILTER_ACTION_DROP_FLAG; + return 1; +} + +/* Determine the "best" filtering mode VIC is capaible of. Returns one of 4 * value or 0 on error: + * FILTER_FLOWMAN- flowman api capable * FILTER_DPDK_1- advanced filters availabile * FILTER_USNIC_IP_FLAG - advanced filters but with the restriction that * the IP layer must explicitly specified. I.e. cannot have a UDP @@ -517,6 +586,10 @@ int vnic_dev_capable_filter_mode(struct vnic_dev *vdev, u32 *mode, int err; u32 max_level = 0; + /* If flowman is available, use it as it is the most capable API */ + if (vnic_dev_flowman_enable(vdev, mode, filter_actions)) + return 0; + err = vnic_dev_advanced_filters_cap(vdev, args, 4); /* determine supported filter actions */ @@ -977,6 +1050,10 @@ void vnic_dev_unregister(struct vnic_dev *vdev) vdev->free_consistent(vdev->priv, sizeof(struct vnic_stats), vdev->stats, vdev->stats_pa); + if (vdev->flowman_info) + vdev->free_consistent(vdev->priv, + sizeof(struct fm_info), + vdev->flowman_info, vdev->flowman_info_pa); if (vdev->fw_info) vdev->free_consistent(vdev->priv, sizeof(struct vnic_devcmd_fw_info), diff --git a/drivers/net/enic/base/vnic_dev.h b/drivers/net/enic/base/vnic_dev.h index 8a1363486..d84e0c773 100644 --- a/drivers/net/enic/base/vnic_dev.h +++ b/drivers/net/enic/base/vnic_dev.h @@ -182,6 +182,7 @@ int vnic_dev_deinit_done(struct vnic_dev *vdev, int *status); int vnic_dev_set_mac_addr(struct vnic_dev *vdev, u8 *mac_addr); int vnic_dev_classifier(struct vnic_dev *vdev, u8 cmd, u16 *entry, struct filter_v2 *data, struct filter_action_v2 *action_v2); +int vnic_dev_flowman_cmd(struct vnic_dev *vdev, u64 *args, int nargs); int vnic_dev_overlay_offload_ctrl(struct vnic_dev *vdev, u8 overlay, u8 config); int vnic_dev_overlay_offload_cfg(struct vnic_dev *vdev, u8 overlay, diff --git a/drivers/net/enic/base/vnic_devcmd.h b/drivers/net/enic/base/vnic_devcmd.h index da60be7b0..d85d40be2 100644 --- a/drivers/net/enic/base/vnic_devcmd.h +++ b/drivers/net/enic/base/vnic_devcmd.h @@ -600,6 +600,16 @@ enum vnic_devcmd_cmd { * a3 = bitmask of supported actions */ CMD_ADD_ADV_FILTER = _CMDC(_CMD_DIR_RW, _CMD_VTYPE_ENET, 77), + + /* + * Perform a Flow Manager Operation (see flowman_api.h) + * in: (u32) a0 = sub-command + * (u64) a1..15 = (sub-command specific) + * + * All arguments that have not been assigned a meaning should be + * initialized to 0 to allow for better driver forward compatibility. + */ + CMD_FLOW_MANAGER_OP = _CMDC(_CMD_DIR_RW, _CMD_VTYPE_ENET, 88), }; /* Modes for exchanging advanced filter capabilities. The modes supported by @@ -865,6 +875,7 @@ struct filter_action { #define FILTER_ACTION_RQ_STEERING_FLAG (1 << 0) #define FILTER_ACTION_FILTER_ID_FLAG (1 << 1) #define FILTER_ACTION_DROP_FLAG (1 << 2) +#define FILTER_ACTION_COUNTER_FLAG (1 << 3) #define FILTER_ACTION_V2_ALL (FILTER_ACTION_RQ_STEERING_FLAG \ | FILTER_ACTION_DROP_FLAG \ | FILTER_ACTION_FILTER_ID_FLAG) @@ -889,6 +900,7 @@ enum filter_type { FILTER_NVGRE_VMQ = 4, FILTER_USNIC_IP = 5, FILTER_DPDK_1 = 6, + FILTER_FLOWMAN = 7, FILTER_MAX }; diff --git a/drivers/net/enic/base/vnic_flowman.h b/drivers/net/enic/base/vnic_flowman.h new file mode 100644 index 000000000..b586e8a0c --- /dev/null +++ b/drivers/net/enic/base/vnic_flowman.h @@ -0,0 +1,364 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright 2018-2019 Cisco Systems, Inc. All rights reserved. + */ +#ifndef _VNIC_FLOWMAN_H_ +#define _VNIC_FLOWMAN_H_ + +/* This file contains Flow Manager (FM) API of the firmware */ + +/* Flow manager sub-ops */ +enum { + FM_EXACT_TABLE_ALLOC, + FM_TCAM_TABLE_ALLOC, + FM_MATCH_TABLE_FREE, + FM_COUNTER_BRK, + FM_COUNTER_QUERY, + FM_COUNTER_CLEAR_ALL, + FM_COUNTER_DMA, + FM_ACTION_ALLOC, + FM_ACTION_FREE, + FM_EXACT_ENTRY_INSTALL, + FM_TCAM_ENTRY_INSTALL, + FM_MATCH_ENTRY_REMOVE, + FM_VNIC_FIND, + FM_API_VERSION_QUERY, + FM_API_VERSION_SELECT, + FM_INFO_QUERY +}; + +/* + * FKM (flow key metadata) flags used to match packet metadata + * (e.g. packet is tcp) + */ +#define FKM_BITS \ + FBIT(FKM_QTAG) \ + FBIT(FKM_CMD) \ + FBIT(FKM_IPV4) \ + FBIT(FKM_IPV6) \ + FBIT(FKM_ROCE) \ + FBIT(FKM_UDP) \ + FBIT(FKM_TCP) \ + FBIT(FKM_TCPORUDP) \ + FBIT(FKM_IPFRAG) \ + FBIT(FKM_NVGRE) \ + FBIT(FKM_VXLAN) \ + FBIT(FKM_GENEVE) \ + FBIT(FKM_NSH) \ + FBIT(FKM_ROCEV2) \ + FBIT(FKM_VLAN_PRES) \ + FBIT(FKM_IPOK) \ + FBIT(FKM_L4OK) \ + FBIT(FKM_ROCEOK) \ + FBIT(FKM_FCSOK) \ + FBIT(FKM_EG_SPAN) \ + FBIT(FKM_IG_SPAN) \ + FBIT(FKM_EG_HAIRPINNED) + +/* + * FKH (flow key header) flags. + * This selects which headers are valid in the struct. + * This is distinct from metadata in that metadata is requesting actual + * selection criteria. If, for example, a TCAM match with metadata "FKM_UDP" + * is feeding into an exact match table, there may be no need for the + * exact match table to also specify FKM_UDP, so FKH_UDP is used to + * specify that the UDP header fields should be used in the match. + */ +#define FKH_BITS \ + FBIT(FKH_ETHER) \ + FBIT(FKH_QTAG) \ + FBIT(FKH_L2RAW) \ + FBIT(FKH_IPV4) \ + FBIT(FKH_IPV6) \ + FBIT(FKH_L3RAW) \ + FBIT(FKH_UDP) \ + FBIT(FKH_TCP) \ + FBIT(FKH_ICMP) \ + FBIT(FKH_VXLAN) \ + FBIT(FKH_L4RAW) + +#define FBIT(X) X##_BIT, +enum { + FKM_BITS + FKM_BIT_COUNT +}; + +enum { + FKH_BITS + FKH_BIT_COUNT +}; +#undef FBIT +#define FBIT(X) X = (1 << X##_BIT), +enum { + FKM_BITS +}; +enum { + FKH_BITS +}; +#undef FBIT + +#define FM_ETH_ALEN 6 +#define FM_LAYER_SIZE 64 + +/* Header match pattern */ +struct fm_header_set { + uint32_t fk_metadata; /* FKM flags */ + uint32_t fk_header_select; /* FKH flags */ + uint16_t fk_vlan; + /* L2: Ethernet Header (valid if FKH_ETHER) */ + union { + struct { + uint8_t fk_dstmac[FM_ETH_ALEN]; + uint8_t fk_srcmac[FM_ETH_ALEN]; + uint16_t fk_ethtype; + } __rte_packed eth; + uint8_t rawdata[FM_LAYER_SIZE]; + } __rte_packed l2; + /* L3: IPv4 or IPv6 (valid if FKH_IPV4,6) */ + union { + /* Valid if FKH_IPV4 */ + struct { + uint8_t fk_ihl_vers; + uint8_t fk_tos; + uint16_t fk_tot_len; + uint16_t fk_id; + uint16_t fk_frag_off; + uint8_t fk_ttl; + uint8_t fk_proto; + uint16_t fk_check; + uint32_t fk_saddr; + uint32_t fk_daddr; + } __rte_packed ip4; + /* Valid if FKH_IPV6 */ + struct { + union { + struct { + uint32_t fk_un1_flow; + uint16_t fk_un1_plen; + uint8_t fk_un1_nxt; + uint8_t fk_un1_hlim; + } unl; + uint8_t fk_un2_vfc; + } ctl; + uint8_t fk_srcip[16]; + uint8_t fk_dstip[16]; + } __rte_packed ip6; + uint8_t rawdata[FM_LAYER_SIZE]; + } __rte_packed l3; + /* L4: UDP, TCP, or ICMP (valid if FKH_UDP,TCP,ICMP) */ + union { + struct { + uint16_t fk_source; + uint16_t fk_dest; + uint16_t fk_len; + uint16_t fk_check; + } __rte_packed udp; + struct { + uint16_t fk_source; + uint16_t fk_dest; + uint32_t fk_seq; + uint32_t fk_ack_seq; + uint16_t fk_flags; + uint16_t fk_window; + uint16_t fk_check; + uint16_t fk_urg_ptr; + } __rte_packed tcp; + struct { + uint8_t fk_code; + uint8_t fk_type; + } __rte_packed icmp; + uint8_t rawdata[FM_LAYER_SIZE]; + } __rte_packed l4; + /* VXLAN (valid if FKH_VXLAN) */ + struct { + uint8_t fkvx_flags; + uint8_t fkvx_res0[3]; + uint8_t fkvx_vni[3]; + uint8_t fkvx_res1; + } __rte_packed vxlan; + /* Payload or unknown inner-most protocol */ + uint8_t fk_l5_data[64]; +} __rte_packed; + +/* + * FK (flow key) template. + * fk_hdrset specifies a set of headers per layer of encapsulation. + * Currently FM supports two header sets: outer (0) and inner(1) + */ +#define FM_HDRSET_MAX 2 + +struct fm_key_template { + struct fm_header_set fk_hdrset[FM_HDRSET_MAX]; + uint32_t fk_flags; + uint16_t fk_packet_tag; + uint16_t fk_packet_size; + uint16_t fk_port_id; + uint32_t fk_wq_id; /* WQ index */ + uint64_t fk_wq_vnic; /* VNIC handle for WQ index */ +} __rte_packed; + +/* Action operation types */ +enum { + /* End the action chain. */ + FMOP_END, + /* Drop packet and end the action chain. */ + FMOP_DROP, + /* Steer packet to an RQ. */ + FMOP_RQ_STEER, + /* + * Jump to an exact match table. + * arg1: exact match table handle + */ + FMOP_EXACT_MATCH, + /* Apply CQ-visible mark on packet. Mark is written to RSS HASH. */ + FMOP_MARK, + /* + * Apply CQ-visible mark on packet. Mark is written to a field in + * extended CQ. RSS HASH is preserved. + */ + FMOP_EXT_MARK, + /* + * Apply internal tag which can be matched in subsequent + * stages or hairpin. + */ + FMOP_TAG, + /* Hairpin packet from EG -> IG */ + FMOP_EG_HAIRPIN, + /* Hairpin packet from IG -> EG */ + FMOP_IG_HAIRPIN, + /* Encap with VXLAN and inner VLAN from metadata. */ + FMOP_ENCAP_IVLAN, + /* Encap, no inner VLAN. */ + FMOP_ENCAP_NOIVLAN, + /* Encap, add inner VLAN if present. */ + FMOP_ENCAP, + /* Set outer VLAN. */ + FMOP_SET_OVLAN, + /* Decap when vlan_strip is off */ + FMOP_DECAP_NOSTRIP, + FMOP_OP_MAX, +}; + +/* + * Action operation. + * Complex actions are achieved by a series of "transform operations" + * We can have complex transform operations like "decap" or "vxlan + * encap" and also simple ops like insert this data, add PACKET_LEN to + * this address, etc. + */ +struct fm_action_op { + uint32_t fa_op; /* FMOP flags */ + + union { + struct { + uint8_t len1_offset; + uint8_t len1_delta; + uint8_t len2_offset; + uint8_t len2_delta; + uint16_t outer_vlan; + uint8_t template_offset; + uint8_t template_len; + } __rte_packed encap; + struct { + uint32_t rq_index; + uint64_t vnic_handle; + } __rte_packed rq_steer; + struct { + uint16_t vlan; + } __rte_packed ovlan; + struct { + uint16_t mark; + } __rte_packed mark; + struct { + uint32_t ext_mark; + } __rte_packed ext_mark; + struct { + uint8_t tag; + } __rte_packed tag; + struct { + uint64_t handle; + } __rte_packed exact; + } __rte_packed; +} __rte_packed; + +#define FM_ACTION_OP_MAX 64 +#define FM_ACTION_DATA_MAX 96 + +/* + * Action is a series of action operations applied to matched + * packet. FMA (flowman action). + */ +struct fm_action { + struct fm_action_op fma_action_ops[FM_ACTION_OP_MAX]; + uint8_t fma_data[FM_ACTION_DATA_MAX]; +} __rte_packed; + +/* Match entry flags. FMEF (flow match entry flag) */ +#define FMEF_COUNTER 0x0001 /* counter index is valid */ + +/* FEM (flow exact match) entry */ +struct fm_exact_match_entry { + struct fm_key_template fem_data; /* Match data. Mask is per table */ + uint32_t fem_flags; /* FMEF_xxx */ + uint64_t fem_action; /* Action handle */ + uint32_t fem_counter; /* Counter index */ +} __rte_packed; + +/* FTM (flow TCAM match) entry */ +struct fm_tcam_match_entry { + struct fm_key_template ftm_mask; /* Key mask */ + struct fm_key_template ftm_data; /* Match data */ + uint32_t ftm_flags; /* FMEF_xxx */ + uint32_t ftm_position; /* Entry position */ + uint64_t ftm_action; /* Action handle */ + uint32_t ftm_counter; /* Counter index */ +} __rte_packed; + +/* Match directions */ +enum { + FM_INGRESS, + FM_EGRESS, + FM_DIR_CNT +}; + +/* Last stage ID, independent of the number of stages in hardware */ +#define FM_STAGE_LAST 0xff + +/* Hash based exact match table. FET (flow exact match table) */ +struct fm_exact_match_table { + uint8_t fet_direction; /* FM_INGRESS or EGRESS*/ + uint8_t fet_stage; + uint8_t pad[2]; + uint32_t fet_max_entries; + uint64_t fet_dflt_action; + struct fm_key_template fet_key; +} __rte_packed; + +/* TCAM based match table. FTT (flow TCAM match table) */ +struct fm_tcam_match_table { + uint8_t ftt_direction; + uint8_t ftt_stage; + uint8_t pad[2]; + uint32_t ftt_max_entries; +} __rte_packed; + +struct fm_counter_counts { + uint64_t fcc_packets; + uint64_t fcc_bytes; +} __rte_packed; + +/* + * Return structure for FM_INFO_QUERY devcmd + */ +#define FM_VERSION 1 /* This header file is for version 1 */ + +struct fm_info { + uint64_t fm_op_mask; /* Bitmask of action supported ops */ + uint64_t fm_current_ts; /* Current VIC timestamp */ + uint64_t fm_clock_freq; /* Timestamp clock frequency */ + uint16_t fm_max_ops; /* Max ops in an action */ + uint8_t fm_stages; /* Number of match-action stages */ + uint8_t pad[5]; + uint32_t fm_counter_count; /* Number of allocated counters */ +} __rte_packed; + +#endif /* _VNIC_FLOWMAN_H_ */ diff --git a/drivers/net/enic/enic.h b/drivers/net/enic/enic.h index 87792deec..7413e9354 100644 --- a/drivers/net/enic/enic.h +++ b/drivers/net/enic/enic.h @@ -8,6 +8,7 @@ #include "vnic_enet.h" #include "vnic_dev.h" +#include "vnic_flowman.h" #include "vnic_wq.h" #include "vnic_rq.h" #include "vnic_cq.h" @@ -88,10 +89,17 @@ struct enic_memzone_entry { LIST_ENTRY(enic_memzone_entry) entries; }; +/* Defined in enic_fm_flow.c */ +struct enic_flowman; +struct enic_fm_flow; + struct rte_flow { LIST_ENTRY(rte_flow) next; - u16 enic_filter_id; + /* Data for filter API based flow (enic_flow.c) */ + uint16_t enic_filter_id; struct filter_v2 enic_filter; + /* Data for flow manager based flow (enic_fm_flow.c) */ + struct enic_fm_flow *fm; }; /* Per-instance private data structure */ @@ -195,6 +203,9 @@ struct enic { /* Multicast MAC addresses added to the NIC */ uint32_t mc_count; struct rte_ether_addr mc_addrs[ENIC_MULTICAST_PERFECT_FILTERS]; + + /* Flow manager API */ + struct enic_flowman *fm; }; /* Compute ethdev's max packet size from MTU */ @@ -276,6 +287,7 @@ enic_ring_incr(uint32_t n_descriptors, uint32_t idx) return idx; } +int dev_is_enic(struct rte_eth_dev *dev); void enic_fdir_stats_get(struct enic *enic, struct rte_eth_fdir_stats *stats); int enic_fdir_add_fltr(struct enic *enic, @@ -321,6 +333,12 @@ void enic_post_wq_index(struct vnic_wq *wq); int enic_probe(struct enic *enic); int enic_clsf_init(struct enic *enic); void enic_clsf_destroy(struct enic *enic); +int enic_fm_init(struct enic *enic); +void enic_fm_destroy(struct enic *enic); +void *enic_alloc_consistent(void *priv, size_t size, dma_addr_t *dma_handle, + u8 *name); +void enic_free_consistent(void *priv, size_t size, void *vaddr, + dma_addr_t dma_handle); uint16_t enic_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts); uint16_t enic_noscatter_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, @@ -340,4 +358,5 @@ bool enic_use_vector_rx_handler(struct enic *enic); void enic_fdir_info(struct enic *enic); void enic_fdir_info_get(struct enic *enic, struct rte_eth_fdir_info *stats); extern const struct rte_flow_ops enic_flow_ops; +extern const struct rte_flow_ops enic_fm_flow_ops; #endif /* _ENIC_H_ */ diff --git a/drivers/net/enic/enic_ethdev.c b/drivers/net/enic/enic_ethdev.c index e1c6da3b3..e8afc2dcd 100644 --- a/drivers/net/enic/enic_ethdev.c +++ b/drivers/net/enic/enic_ethdev.c @@ -143,7 +143,10 @@ enicpmd_dev_filter_ctrl(struct rte_eth_dev *dev, case RTE_ETH_FILTER_GENERIC: if (filter_op != RTE_ETH_FILTER_GET) return -EINVAL; - *(const void **)arg = &enic_flow_ops; + if (enic->flow_filter_mode == FILTER_FLOWMAN) + *(const void **)arg = &enic_fm_flow_ops; + else + *(const void **)arg = &enic_flow_ops; break; case RTE_ETH_FILTER_FDIR: ret = enicpmd_fdir_ctrl_func(dev, filter_op, arg); @@ -1260,6 +1263,11 @@ static struct rte_pci_driver rte_enic_pmd = { .remove = eth_enic_pci_remove, }; +int dev_is_enic(struct rte_eth_dev *dev) +{ + return dev->device->driver == &rte_enic_pmd.driver; +} + RTE_PMD_REGISTER_PCI(net_enic, rte_enic_pmd); RTE_PMD_REGISTER_PCI_TABLE(net_enic, pci_id_enic_map); RTE_PMD_REGISTER_KMOD_DEP(net_enic, "* igb_uio | uio_pci_generic | vfio-pci"); diff --git a/drivers/net/enic/enic_fm_flow.c b/drivers/net/enic/enic_fm_flow.c new file mode 100644 index 000000000..a1598b822 --- /dev/null +++ b/drivers/net/enic/enic_fm_flow.c @@ -0,0 +1,2459 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright 2008-2019 Cisco Systems, Inc. All rights reserved. + */ + +#include +#include +#include +#include +#include +#include +#include +#include + +#include "enic_compat.h" +#include "enic.h" +#include "vnic_dev.h" +#include "vnic_nic.h" + +#define IP_DEFTTL 64 /* from RFC 1340. */ +#define IP_VERSION 0x40 +#define IP_HDRLEN 0x05 /* default IP header length == five 32-bits words. */ +#define IP_VHL_DEF (IP_VERSION | IP_HDRLEN) +#define IP6_VTC_FLOW 0x60000000 + +/* Highest Item type supported by Flowman */ +#define FM_MAX_ITEM_TYPE RTE_FLOW_ITEM_TYPE_VXLAN + +/* Up to 1024 TCAM entries */ +#define FM_MAX_TCAM_TABLE_SIZE 1024 + +/* Up to 4096 entries per exact match table */ +#define FM_MAX_EXACT_TABLE_SIZE 4096 + +/* Number of counters to increase on for each increment */ +#define FM_COUNTERS_EXPAND 100 + +#define FM_INVALID_HANDLE 0 + +/* + * Flow exact match tables (FET) in the VIC and rte_flow groups. + * Use a simple scheme to map groups to tables. + * Group 0 uses the single TCAM tables, one for each direction. + * Group 1, 2, ... uses its own exact match table. + * + * The TCAM tables are allocated upfront during init. + * + * Exact match tables are allocated on demand. 3 paths that lead allocations. + * + * 1. Add a flow that jumps from group 0 to group N. + * + * If N does not exist, we allocate an exact match table for it, using + * a dummy key. A key is required for the table. + * + * 2. Add a flow that uses group N. + * + * If N does not exist, we allocate an exact match table for it, using + * the flow's key. Subsequent flows to the same group all should have + * the same key. + * + * Without a jump flow to N, N is not reachable in hardware. No packets + * reach N and match. + * + * 3. Add a flow to an empty group N. + * + * N has been created via (1) and the dummy key. We free that table, allocate + * a new table using the new flow's key. Also re-do the existing jump flow to + * point to the new table. + */ +#define FM_TCAM_RTE_GROUP 0 + +struct enic_fm_fet { + TAILQ_ENTRY(enic_fm_fet) list; + uint32_t group; /* rte_flow group ID */ + uint64_t handle; /* Exact match table handle from flowman */ + uint8_t ingress; + uint8_t default_key; + int ref; /* Reference count via get/put */ + struct fm_key_template key; /* Key associated with the table */ +}; + +struct enic_fm_counter { + SLIST_ENTRY(enic_fm_counter) next; + uint32_t handle; +}; + +/* rte_flow.fm */ +struct enic_fm_flow { + bool counter_valid; + uint64_t entry_handle; + uint64_t action_handle; + struct enic_fm_counter *counter; + struct enic_fm_fet *fet; +}; + +struct enic_fm_jump_flow { + TAILQ_ENTRY(enic_fm_jump_flow) list; + struct rte_flow *flow; + uint32_t group; + struct fm_tcam_match_entry match; + struct fm_action action; +}; + +/* + * Flowman uses host memory for commands. This structure is allocated + * in DMA-able memory. + */ +union enic_flowman_cmd_mem { + struct fm_tcam_match_table fm_tcam_match_table; + struct fm_exact_match_table fm_exact_match_table; + struct fm_tcam_match_entry fm_tcam_match_entry; + struct fm_exact_match_entry fm_exact_match_entry; + struct fm_action fm_action; +}; + +struct enic_flowman { + struct enic *enic; + /* Command buffer */ + struct { + union enic_flowman_cmd_mem *va; + dma_addr_t pa; + } cmd; + /* TCAM tables allocated upfront, used for group 0 */ + uint64_t ig_tcam_hndl; + uint64_t eg_tcam_hndl; + /* Counters */ + SLIST_HEAD(enic_free_counters, enic_fm_counter) counters; + void *counter_stack; + uint32_t counters_alloced; + /* Exact match tables for groups != 0, dynamically allocated */ + TAILQ_HEAD(fet_list, enic_fm_fet) fet_list; + /* + * Default exact match tables used for jump actions to + * non-existent groups. + */ + struct enic_fm_fet *default_eg_fet; + struct enic_fm_fet *default_ig_fet; + /* Flows that jump to the default table above */ + TAILQ_HEAD(jump_flow_list, enic_fm_jump_flow) jump_list; + /* + * Scratch data used during each invocation of flow_create + * and flow_validate. + */ + struct enic_fm_fet *fet; + struct fm_tcam_match_entry tcam_entry; + struct fm_action action; + struct fm_action action_tmp; /* enic_fm_reorder_action_op */ + int action_op_count; +}; + +static int enic_fm_tbl_free(struct enic_flowman *fm, uint64_t handle); + +/* + * Common arguments passed to copy_item functions. Use this structure + * so we can easily add new arguments. + * item: Item specification. + * fm_tcam_entry: Flowman TCAM match entry. + * header_level: 0 for outer header, 1 for inner header. + */ +struct copy_item_args { + const struct rte_flow_item *item; + struct fm_tcam_match_entry *fm_tcam_entry; + uint8_t header_level; +}; + +/* functions for copying items into flowman match */ +typedef int (enic_copy_item_fn)(struct copy_item_args *arg); + +/* Info about how to copy items into flowman match */ +struct enic_fm_items { + /* Function for copying and validating an item. */ + enic_copy_item_fn * const copy_item; + /* List of valid previous items. */ + const enum rte_flow_item_type * const prev_items; + /* + * True if it's OK for this item to be the first item. For some NIC + * versions, it's invalid to start the stack above layer 3. + */ + const uint8_t valid_start_item; +}; + +static enic_copy_item_fn enic_fm_copy_item_eth; +static enic_copy_item_fn enic_fm_copy_item_ipv4; +static enic_copy_item_fn enic_fm_copy_item_ipv6; +static enic_copy_item_fn enic_fm_copy_item_raw; +static enic_copy_item_fn enic_fm_copy_item_sctp; +static enic_copy_item_fn enic_fm_copy_item_tcp; +static enic_copy_item_fn enic_fm_copy_item_udp; +static enic_copy_item_fn enic_fm_copy_item_vlan; +static enic_copy_item_fn enic_fm_copy_item_vxlan; + +/* Ingress actions */ +static const enum rte_flow_action_type enic_fm_supported_ig_actions[] = { + RTE_FLOW_ACTION_TYPE_COUNT, + RTE_FLOW_ACTION_TYPE_DROP, + RTE_FLOW_ACTION_TYPE_FLAG, + RTE_FLOW_ACTION_TYPE_JUMP, + RTE_FLOW_ACTION_TYPE_MARK, + RTE_FLOW_ACTION_TYPE_PORT_ID, + RTE_FLOW_ACTION_TYPE_PASSTHRU, + RTE_FLOW_ACTION_TYPE_QUEUE, + RTE_FLOW_ACTION_TYPE_RSS, + RTE_FLOW_ACTION_TYPE_VOID, + RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP, + RTE_FLOW_ACTION_TYPE_VXLAN_DECAP, + RTE_FLOW_ACTION_TYPE_END, /* END must be the last entry */ +}; + +/* Egress actions */ +static const enum rte_flow_action_type enic_fm_supported_eg_actions[] = { + RTE_FLOW_ACTION_TYPE_COUNT, + RTE_FLOW_ACTION_TYPE_DROP, + RTE_FLOW_ACTION_TYPE_JUMP, + RTE_FLOW_ACTION_TYPE_PASSTHRU, + RTE_FLOW_ACTION_TYPE_VOID, + RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP, + RTE_FLOW_ACTION_TYPE_END, +}; + +static const struct enic_fm_items enic_fm_items[] = { + [RTE_FLOW_ITEM_TYPE_RAW] = { + .copy_item = enic_fm_copy_item_raw, + .valid_start_item = 0, + .prev_items = (const enum rte_flow_item_type[]) { + RTE_FLOW_ITEM_TYPE_UDP, + RTE_FLOW_ITEM_TYPE_END, + }, + }, + [RTE_FLOW_ITEM_TYPE_ETH] = { + .copy_item = enic_fm_copy_item_eth, + .valid_start_item = 1, + .prev_items = (const enum rte_flow_item_type[]) { + RTE_FLOW_ITEM_TYPE_END, + }, + }, + [RTE_FLOW_ITEM_TYPE_VLAN] = { + .copy_item = enic_fm_copy_item_vlan, + .valid_start_item = 1, + .prev_items = (const enum rte_flow_item_type[]) { + RTE_FLOW_ITEM_TYPE_ETH, + RTE_FLOW_ITEM_TYPE_END, + }, + }, + [RTE_FLOW_ITEM_TYPE_IPV4] = { + .copy_item = enic_fm_copy_item_ipv4, + .valid_start_item = 1, + .prev_items = (const enum rte_flow_item_type[]) { + RTE_FLOW_ITEM_TYPE_ETH, + RTE_FLOW_ITEM_TYPE_VLAN, + RTE_FLOW_ITEM_TYPE_END, + }, + }, + [RTE_FLOW_ITEM_TYPE_IPV6] = { + .copy_item = enic_fm_copy_item_ipv6, + .valid_start_item = 1, + .prev_items = (const enum rte_flow_item_type[]) { + RTE_FLOW_ITEM_TYPE_ETH, + RTE_FLOW_ITEM_TYPE_VLAN, + RTE_FLOW_ITEM_TYPE_END, + }, + }, + [RTE_FLOW_ITEM_TYPE_UDP] = { + .copy_item = enic_fm_copy_item_udp, + .valid_start_item = 1, + .prev_items = (const enum rte_flow_item_type[]) { + RTE_FLOW_ITEM_TYPE_IPV4, + RTE_FLOW_ITEM_TYPE_IPV6, + RTE_FLOW_ITEM_TYPE_END, + }, + }, + [RTE_FLOW_ITEM_TYPE_TCP] = { + .copy_item = enic_fm_copy_item_tcp, + .valid_start_item = 1, + .prev_items = (const enum rte_flow_item_type[]) { + RTE_FLOW_ITEM_TYPE_IPV4, + RTE_FLOW_ITEM_TYPE_IPV6, + RTE_FLOW_ITEM_TYPE_END, + }, + }, + [RTE_FLOW_ITEM_TYPE_SCTP] = { + .copy_item = enic_fm_copy_item_sctp, + .valid_start_item = 0, + .prev_items = (const enum rte_flow_item_type[]) { + RTE_FLOW_ITEM_TYPE_IPV4, + RTE_FLOW_ITEM_TYPE_IPV6, + RTE_FLOW_ITEM_TYPE_END, + }, + }, + [RTE_FLOW_ITEM_TYPE_VXLAN] = { + .copy_item = enic_fm_copy_item_vxlan, + .valid_start_item = 1, + .prev_items = (const enum rte_flow_item_type[]) { + RTE_FLOW_ITEM_TYPE_UDP, + RTE_FLOW_ITEM_TYPE_END, + }, + }, +}; + +static int +enic_fm_copy_item_eth(struct copy_item_args *arg) +{ + const struct rte_flow_item *item = arg->item; + const struct rte_flow_item_eth *spec = item->spec; + const struct rte_flow_item_eth *mask = item->mask; + const uint8_t lvl = arg->header_level; + struct fm_tcam_match_entry *entry = arg->fm_tcam_entry; + struct fm_header_set *fm_data, *fm_mask; + + ENICPMD_FUNC_TRACE(); + /* Match all if no spec */ + if (!spec) + return 0; + if (!mask) + mask = &rte_flow_item_eth_mask; + fm_data = &entry->ftm_data.fk_hdrset[lvl]; + fm_mask = &entry->ftm_mask.fk_hdrset[lvl]; + fm_data->fk_header_select |= FKH_ETHER; + fm_mask->fk_header_select |= FKH_ETHER; + memcpy(&fm_data->l2.eth, spec, sizeof(*spec)); + memcpy(&fm_mask->l2.eth, mask, sizeof(*mask)); + return 0; +} + +static int +enic_fm_copy_item_vlan(struct copy_item_args *arg) +{ + const struct rte_flow_item *item = arg->item; + const struct rte_flow_item_vlan *spec = item->spec; + const struct rte_flow_item_vlan *mask = item->mask; + const uint8_t lvl = arg->header_level; + struct fm_tcam_match_entry *entry = arg->fm_tcam_entry; + struct fm_header_set *fm_data, *fm_mask; + struct rte_ether_hdr *eth_mask; + struct rte_ether_hdr *eth_val; + uint32_t meta; + + ENICPMD_FUNC_TRACE(); + fm_data = &entry->ftm_data.fk_hdrset[lvl]; + fm_mask = &entry->ftm_mask.fk_hdrset[lvl]; + /* Outer and inner packet vlans need different flags */ + meta = FKM_VLAN_PRES; + if (lvl > 0) + meta = FKM_QTAG; + fm_data->fk_metadata |= meta; + fm_mask->fk_metadata |= meta; + + /* Match all if no spec */ + if (!spec) + return 0; + if (!mask) + mask = &rte_flow_item_vlan_mask; + + eth_mask = (void *)&fm_mask->l2.eth; + eth_val = (void *)&fm_data->l2.eth; + + /* Outer TPID cannot be matched */ + if (eth_mask->ether_type) + return -ENOTSUP; + + /* + * When packet matching, the VIC always compares vlan-stripped + * L2, regardless of vlan stripping settings. So, the inner type + * from vlan becomes the ether type of the eth header. + */ + eth_mask->ether_type = mask->inner_type; + eth_val->ether_type = spec->inner_type; + fm_data->fk_header_select |= FKH_ETHER | FKH_QTAG; + fm_mask->fk_header_select |= FKH_ETHER | FKH_QTAG; + fm_data->fk_vlan = rte_be_to_cpu_16(spec->tci); + fm_mask->fk_vlan = rte_be_to_cpu_16(mask->tci); + return 0; +} + +static int +enic_fm_copy_item_ipv4(struct copy_item_args *arg) +{ + const struct rte_flow_item *item = arg->item; + const struct rte_flow_item_ipv4 *spec = item->spec; + const struct rte_flow_item_ipv4 *mask = item->mask; + const uint8_t lvl = arg->header_level; + struct fm_tcam_match_entry *entry = arg->fm_tcam_entry; + struct fm_header_set *fm_data, *fm_mask; + + ENICPMD_FUNC_TRACE(); + fm_data = &entry->ftm_data.fk_hdrset[lvl]; + fm_mask = &entry->ftm_mask.fk_hdrset[lvl]; + fm_data->fk_metadata |= FKM_IPV4; + fm_mask->fk_metadata |= FKM_IPV4; + + if (!spec) + return 0; + if (!mask) + mask = &rte_flow_item_ipv4_mask; + + fm_data->fk_header_select |= FKH_IPV4; + fm_mask->fk_header_select |= FKH_IPV4; + memcpy(&fm_data->l3.ip4, spec, sizeof(*spec)); + memcpy(&fm_mask->l3.ip4, mask, sizeof(*mask)); + return 0; +} + +static int +enic_fm_copy_item_ipv6(struct copy_item_args *arg) +{ + const struct rte_flow_item *item = arg->item; + const struct rte_flow_item_ipv6 *spec = item->spec; + const struct rte_flow_item_ipv6 *mask = item->mask; + const uint8_t lvl = arg->header_level; + struct fm_tcam_match_entry *entry = arg->fm_tcam_entry; + struct fm_header_set *fm_data, *fm_mask; + + ENICPMD_FUNC_TRACE(); + fm_data = &entry->ftm_data.fk_hdrset[lvl]; + fm_mask = &entry->ftm_mask.fk_hdrset[lvl]; + fm_data->fk_metadata |= FKM_IPV6; + fm_mask->fk_metadata |= FKM_IPV6; + + if (!spec) + return 0; + if (!mask) + mask = &rte_flow_item_ipv6_mask; + + fm_data->fk_header_select |= FKH_IPV6; + fm_mask->fk_header_select |= FKH_IPV6; + memcpy(&fm_data->l3.ip6, spec, sizeof(*spec)); + memcpy(&fm_mask->l3.ip6, mask, sizeof(*mask)); + return 0; +} + +static int +enic_fm_copy_item_udp(struct copy_item_args *arg) +{ + const struct rte_flow_item *item = arg->item; + const struct rte_flow_item_udp *spec = item->spec; + const struct rte_flow_item_udp *mask = item->mask; + const uint8_t lvl = arg->header_level; + struct fm_tcam_match_entry *entry = arg->fm_tcam_entry; + struct fm_header_set *fm_data, *fm_mask; + + ENICPMD_FUNC_TRACE(); + fm_data = &entry->ftm_data.fk_hdrset[lvl]; + fm_mask = &entry->ftm_mask.fk_hdrset[lvl]; + fm_data->fk_metadata |= FKM_UDP; + fm_mask->fk_metadata |= FKM_UDP; + + if (!spec) + return 0; + if (!mask) + mask = &rte_flow_item_udp_mask; + + fm_data->fk_header_select |= FKH_UDP; + fm_mask->fk_header_select |= FKH_UDP; + memcpy(&fm_data->l4.udp, spec, sizeof(*spec)); + memcpy(&fm_mask->l4.udp, mask, sizeof(*mask)); + return 0; +} + +static int +enic_fm_copy_item_tcp(struct copy_item_args *arg) +{ + const struct rte_flow_item *item = arg->item; + const struct rte_flow_item_tcp *spec = item->spec; + const struct rte_flow_item_tcp *mask = item->mask; + const uint8_t lvl = arg->header_level; + struct fm_tcam_match_entry *entry = arg->fm_tcam_entry; + struct fm_header_set *fm_data, *fm_mask; + + ENICPMD_FUNC_TRACE(); + fm_data = &entry->ftm_data.fk_hdrset[lvl]; + fm_mask = &entry->ftm_mask.fk_hdrset[lvl]; + fm_data->fk_metadata |= FKM_TCP; + fm_mask->fk_metadata |= FKM_TCP; + + if (!spec) + return 0; + if (!mask) + mask = &rte_flow_item_tcp_mask; + + fm_data->fk_header_select |= FKH_TCP; + fm_mask->fk_header_select |= FKH_TCP; + memcpy(&fm_data->l4.tcp, spec, sizeof(*spec)); + memcpy(&fm_mask->l4.tcp, mask, sizeof(*mask)); + return 0; +} + +static int +enic_fm_copy_item_sctp(struct copy_item_args *arg) +{ + const struct rte_flow_item *item = arg->item; + const struct rte_flow_item_sctp *spec = item->spec; + const struct rte_flow_item_sctp *mask = item->mask; + const uint8_t lvl = arg->header_level; + struct fm_tcam_match_entry *entry = arg->fm_tcam_entry; + struct fm_header_set *fm_data, *fm_mask; + uint8_t *ip_proto_mask = NULL; + uint8_t *ip_proto = NULL; + uint32_t l3_fkh; + + ENICPMD_FUNC_TRACE(); + fm_data = &entry->ftm_data.fk_hdrset[lvl]; + fm_mask = &entry->ftm_mask.fk_hdrset[lvl]; + /* + * The NIC filter API has no flags for "match sctp", so explicitly + * set the protocol number in the IP pattern. + */ + if (fm_data->fk_metadata & FKM_IPV4) { + struct rte_ipv4_hdr *ip; + ip = (struct rte_ipv4_hdr *)&fm_mask->l3.ip4; + ip_proto_mask = &ip->next_proto_id; + ip = (struct rte_ipv4_hdr *)&fm_data->l3.ip4; + ip_proto = &ip->next_proto_id; + l3_fkh = FKH_IPV4; + } else if (fm_data->fk_metadata & FKM_IPV6) { + struct rte_ipv6_hdr *ip; + ip = (struct rte_ipv6_hdr *)&fm_mask->l3.ip6; + ip_proto_mask = &ip->proto; + ip = (struct rte_ipv6_hdr *)&fm_data->l3.ip6; + ip_proto = &ip->proto; + l3_fkh = FKH_IPV6; + } else { + /* Need IPv4/IPv6 pattern first */ + return -EINVAL; + } + *ip_proto = IPPROTO_SCTP; + *ip_proto_mask = 0xff; + fm_data->fk_header_select |= l3_fkh; + fm_mask->fk_header_select |= l3_fkh; + + if (!spec) + return 0; + if (!mask) + mask = &rte_flow_item_sctp_mask; + + fm_data->fk_header_select |= FKH_L4RAW; + fm_mask->fk_header_select |= FKH_L4RAW; + memcpy(fm_data->l4.rawdata, spec, sizeof(*spec)); + memcpy(fm_mask->l4.rawdata, mask, sizeof(*mask)); + return 0; +} + +static int +enic_fm_copy_item_vxlan(struct copy_item_args *arg) +{ + const struct rte_flow_item *item = arg->item; + const struct rte_flow_item_vxlan *spec = item->spec; + const struct rte_flow_item_vxlan *mask = item->mask; + struct fm_tcam_match_entry *entry = arg->fm_tcam_entry; + struct fm_header_set *fm_data, *fm_mask; + + ENICPMD_FUNC_TRACE(); + /* Only 2 header levels (outer and inner) allowed */ + if (arg->header_level > 0) + return -EINVAL; + + fm_data = &entry->ftm_data.fk_hdrset[0]; + fm_mask = &entry->ftm_mask.fk_hdrset[0]; + fm_data->fk_metadata |= FKM_VXLAN; + fm_mask->fk_metadata |= FKM_VXLAN; + /* items from here on out are inner header items */ + arg->header_level = 1; + + /* Match all if no spec */ + if (!spec) + return 0; + if (!mask) + mask = &rte_flow_item_vxlan_mask; + + fm_data->fk_header_select |= FKH_VXLAN; + fm_mask->fk_header_select |= FKH_VXLAN; + memcpy(&fm_data->vxlan, spec, sizeof(*spec)); + memcpy(&fm_mask->vxlan, mask, sizeof(*mask)); + return 0; +} + +/* + * Currently, raw pattern match is very limited. It is intended for matching + * UDP tunnel header (e.g. vxlan or geneve). + */ +static int +enic_fm_copy_item_raw(struct copy_item_args *arg) +{ + const struct rte_flow_item *item = arg->item; + const struct rte_flow_item_raw *spec = item->spec; + const struct rte_flow_item_raw *mask = item->mask; + const uint8_t lvl = arg->header_level; + struct fm_tcam_match_entry *entry = arg->fm_tcam_entry; + struct fm_header_set *fm_data, *fm_mask; + + ENICPMD_FUNC_TRACE(); + /* Cannot be used for inner packet */ + if (lvl > 0) + return -EINVAL; + /* Need both spec and mask */ + if (!spec || !mask) + return -EINVAL; + /* Only supports relative with offset 0 */ + if (!spec->relative || spec->offset != 0 || spec->search || + spec->limit) + return -EINVAL; + /* Need non-null pattern that fits within the NIC's filter pattern */ + if (spec->length == 0 || + spec->length + sizeof(struct rte_udp_hdr) > FM_LAYER_SIZE || + !spec->pattern || !mask->pattern) + return -EINVAL; + /* + * Mask fields, including length, are often set to zero. Assume that + * means "same as spec" to avoid breaking existing apps. If length + * is not zero, then it should be >= spec length. + * + * No more pattern follows this, so append to the L4 layer instead of + * L5 to work with both recent and older VICs. + */ + if (mask->length != 0 && mask->length < spec->length) + return -EINVAL; + + fm_data = &entry->ftm_data.fk_hdrset[lvl]; + fm_mask = &entry->ftm_mask.fk_hdrset[lvl]; + fm_data->fk_header_select |= FKH_L4RAW; + fm_mask->fk_header_select |= FKH_L4RAW; + fm_data->fk_header_select &= ~FKH_UDP; + fm_mask->fk_header_select &= ~FKH_UDP; + memcpy(fm_data->l4.rawdata + sizeof(struct rte_udp_hdr), + spec->pattern, spec->length); + memcpy(fm_mask->l4.rawdata + sizeof(struct rte_udp_hdr), + mask->pattern, spec->length); + return 0; +} + +static int +enic_fet_alloc(struct enic_flowman *fm, uint8_t ingress, + struct fm_key_template *key, int entries, + struct enic_fm_fet **fet_out) +{ + struct fm_exact_match_table *cmd; + struct fm_header_set *hdr; + struct enic_fm_fet *fet; + u64 args[3]; + int ret; + + ENICPMD_FUNC_TRACE(); + fet = calloc(1, sizeof(struct enic_fm_fet)); + if (fet == NULL) + return -ENOMEM; + cmd = &fm->cmd.va->fm_exact_match_table; + memset(cmd, 0, sizeof(*cmd)); + cmd->fet_direction = ingress ? FM_INGRESS : FM_EGRESS; + cmd->fet_stage = FM_STAGE_LAST; + cmd->fet_max_entries = entries ? entries : FM_MAX_EXACT_TABLE_SIZE; + if (key == NULL) { + hdr = &cmd->fet_key.fk_hdrset[0]; + memset(hdr, 0, sizeof(*hdr)); + hdr->fk_header_select = FKH_IPV4 | FKH_UDP; + hdr->l3.ip4.fk_saddr = 0xFFFFFFFF; + hdr->l3.ip4.fk_daddr = 0xFFFFFFFF; + hdr->l4.udp.fk_source = 0xFFFF; + hdr->l4.udp.fk_dest = 0xFFFF; + fet->default_key = 1; + } else { + memcpy(&cmd->fet_key, key, sizeof(*key)); + memcpy(&fet->key, key, sizeof(*key)); + fet->default_key = 0; + } + cmd->fet_key.fk_packet_tag = 1; + + args[0] = FM_EXACT_TABLE_ALLOC; + args[1] = fm->cmd.pa; + ret = vnic_dev_flowman_cmd(fm->enic->vdev, args, 2); + if (ret) { + ENICPMD_LOG(ERR, "cannot alloc exact match table: rc=%d", ret); + free(fet); + return ret; + } + fet->handle = args[0]; + fet->ingress = ingress; + ENICPMD_LOG(DEBUG, "allocated exact match table: handle=0x%" PRIx64, + fet->handle); + *fet_out = fet; + return 0; +} + +static void +enic_fet_free(struct enic_flowman *fm, struct enic_fm_fet *fet) +{ + ENICPMD_FUNC_TRACE(); + enic_fm_tbl_free(fm, fet->handle); + if (!fet->default_key) + TAILQ_REMOVE(&fm->fet_list, fet, list); + free(fet); +} + +/* + * Get the exact match table for the given combination of + * . Allocate one on the fly as necessary. + */ +static int +enic_fet_get(struct enic_flowman *fm, + uint32_t group, + uint8_t ingress, + struct fm_key_template *key, + struct enic_fm_fet **fet_out, + struct rte_flow_error *error) +{ + struct enic_fm_fet *fet; + + ENICPMD_FUNC_TRACE(); + /* See if we already have this table open */ + TAILQ_FOREACH(fet, &fm->fet_list, list) { + if (fet->group == group && fet->ingress == ingress) + break; + } + if (fet == NULL) { + /* Jumping to a non-existing group? Use the default table */ + if (key == NULL) { + fet = ingress ? fm->default_ig_fet : fm->default_eg_fet; + } else if (enic_fet_alloc(fm, ingress, key, 0, &fet)) { + return rte_flow_error_set(error, EINVAL, + RTE_FLOW_ERROR_TYPE_UNSPECIFIED, + NULL, "enic: cannot get exact match table"); + } + fet->group = group; + /* Default table is never on the open table list */ + if (!fet->default_key) + TAILQ_INSERT_HEAD(&fm->fet_list, fet, list); + } + fet->ref++; + *fet_out = fet; + ENICPMD_LOG(DEBUG, "fet_get: %s %s group=%u ref=%u", + fet->default_key ? "default" : "", + fet->ingress ? "ingress" : "egress", + fet->group, fet->ref); + return 0; +} + +static void +enic_fet_put(struct enic_flowman *fm, struct enic_fm_fet *fet) +{ + ENICPMD_FUNC_TRACE(); + RTE_ASSERT(fet->ref > 0); + fet->ref--; + ENICPMD_LOG(DEBUG, "fet_put: %s %s group=%u ref=%u", + fet->default_key ? "default" : "", + fet->ingress ? "ingress" : "egress", + fet->group, fet->ref); + if (fet->ref == 0) + enic_fet_free(fm, fet); +} + +/* Return 1 if current item is valid on top of the previous one. */ +static int +fm_item_stacking_valid(enum rte_flow_item_type prev_item, + const struct enic_fm_items *item_info, + uint8_t is_first_item) +{ + enum rte_flow_item_type const *allowed_items = item_info->prev_items; + + ENICPMD_FUNC_TRACE(); + for (; *allowed_items != RTE_FLOW_ITEM_TYPE_END; allowed_items++) { + if (prev_item == *allowed_items) + return 1; + } + + /* This is the first item in the stack. Check if that's cool */ + if (is_first_item && item_info->valid_start_item) + return 1; + return 0; +} + +/* + * Build the flow manager match entry structure from the provided pattern. + * The pattern is validated as the items are copied. + */ +static int +enic_fm_copy_entry(struct enic_flowman *fm, + const struct rte_flow_item pattern[], + struct rte_flow_error *error) +{ + const struct enic_fm_items *item_info; + enum rte_flow_item_type prev_item; + const struct rte_flow_item *item; + struct copy_item_args args; + uint8_t prev_header_level; + uint8_t is_first_item; + int ret; + + ENICPMD_FUNC_TRACE(); + item = pattern; + is_first_item = 1; + prev_item = RTE_FLOW_ITEM_TYPE_END; + + args.fm_tcam_entry = &fm->tcam_entry; + args.header_level = 0; + prev_header_level = 0; + for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) { + /* + * Get info about how to validate and copy the item. If NULL + * is returned the nic does not support the item. + */ + if (item->type == RTE_FLOW_ITEM_TYPE_VOID) + continue; + + item_info = &enic_fm_items[item->type]; + + if (item->type > FM_MAX_ITEM_TYPE || + item_info->copy_item == NULL) { + return rte_flow_error_set(error, ENOTSUP, + RTE_FLOW_ERROR_TYPE_ITEM, + NULL, "enic: unsupported item"); + } + + /* check to see if item stacking is valid */ + if (!fm_item_stacking_valid(prev_item, item_info, + is_first_item)) + goto stacking_error; + + args.item = item; + ret = item_info->copy_item(&args); + if (ret) + goto item_not_supported; + /* Going from outer to inner? Treat it as a new packet start */ + if (prev_header_level != args.header_level) { + prev_item = RTE_FLOW_ITEM_TYPE_END; + is_first_item = 1; + } else { + prev_item = item->type; + is_first_item = 0; + } + prev_header_level = args.header_level; + } + return 0; + +item_not_supported: + return rte_flow_error_set(error, -ret, RTE_FLOW_ERROR_TYPE_ITEM, + NULL, "enic: unsupported item type"); + +stacking_error: + return rte_flow_error_set(error, EINVAL, RTE_FLOW_ERROR_TYPE_ITEM, + item, "enic: unsupported item stack"); +} + +static void +flow_item_skip_void(const struct rte_flow_item **item) +{ + for ( ; ; (*item)++) + if ((*item)->type != RTE_FLOW_ITEM_TYPE_VOID) + return; +} + +static void +append_template(void **template, uint8_t *off, const void *data, int len) +{ + memcpy(*template, data, len); + *template = (char *)*template + len; + *off = *off + len; +} + +static int +enic_fm_append_action_op(struct enic_flowman *fm, + struct fm_action_op *fm_op, + struct rte_flow_error *error) +{ + int count; + + count = fm->action_op_count; + ENICPMD_LOG(DEBUG, "append action op: idx=%d op=%u", + count, fm_op->fa_op); + if (count == FM_ACTION_OP_MAX) { + return rte_flow_error_set(error, EINVAL, + RTE_FLOW_ERROR_TYPE_ACTION, NULL, + "too many action operations"); + } + fm->action.fma_action_ops[count] = *fm_op; + fm->action_op_count = count + 1; + return 0; +} + +/* Steer operations need to appear before other ops */ +static void +enic_fm_reorder_action_op(struct enic_flowman *fm) +{ + struct fm_action_op *dst, *dst_head, *src, *src_head; + + ENICPMD_FUNC_TRACE(); + /* Move steer ops to the front. */ + src = fm->action.fma_action_ops; + src_head = src; + dst = fm->action_tmp.fma_action_ops; + dst_head = dst; + /* Copy steer ops to tmp */ + while (src->fa_op != FMOP_END) { + if (src->fa_op == FMOP_RQ_STEER) { + ENICPMD_LOG(DEBUG, "move op: %ld -> dst %ld", + (long)(src - src_head), + (long)(dst - dst_head)); + *dst = *src; + dst++; + } + src++; + } + /* Then append non-steer ops */ + src = src_head; + while (src->fa_op != FMOP_END) { + if (src->fa_op != FMOP_RQ_STEER) { + ENICPMD_LOG(DEBUG, "move op: %ld -> dst %ld", + (long)(src - src_head), + (long)(dst - dst_head)); + *dst = *src; + dst++; + } + src++; + } + /* Copy END */ + *dst = *src; + /* Finally replace the original action with the reordered one */ + memcpy(fm->action.fma_action_ops, fm->action_tmp.fma_action_ops, + sizeof(fm->action.fma_action_ops)); +} + +/* VXLAN decap is done via flowman compound action */ +static int +enic_fm_copy_vxlan_decap(struct enic_flowman *fm, + struct fm_tcam_match_entry *fmt, + const struct rte_flow_action *action, + struct rte_flow_error *error) +{ + struct fm_header_set *fm_data; + struct fm_action_op fm_op; + + ENICPMD_FUNC_TRACE(); + fm_data = &fmt->ftm_data.fk_hdrset[0]; + if (!(fm_data->fk_metadata & FKM_VXLAN)) { + return rte_flow_error_set(error, EINVAL, + RTE_FLOW_ERROR_TYPE_ACTION, action, + "vxlan-decap: vxlan must be in pattern"); + } + + memset(&fm_op, 0, sizeof(fm_op)); + fm_op.fa_op = FMOP_DECAP_NOSTRIP; + return enic_fm_append_action_op(fm, &fm_op, error); +} + +/* VXLAN encap is done via flowman compound action */ +static int +enic_fm_copy_vxlan_encap(struct enic_flowman *fm, + const struct rte_flow_item *item, + struct rte_flow_error *error) +{ + struct fm_action_op fm_op; + struct rte_ether_hdr *eth; + uint16_t *ethertype; + void *template; + uint8_t off; + + ENICPMD_FUNC_TRACE(); + memset(&fm_op, 0, sizeof(fm_op)); + fm_op.fa_op = FMOP_ENCAP; + template = fm->action.fma_data; + off = 0; + /* + * Copy flow items to the flowman template starting L2. + * L2 must be ethernet. + */ + flow_item_skip_void(&item); + if (item->type != RTE_FLOW_ITEM_TYPE_ETH) + return rte_flow_error_set(error, EINVAL, + RTE_FLOW_ERROR_TYPE_ITEM, item, + "vxlan-encap: first item should be ethernet"); + eth = (struct rte_ether_hdr *)template; + ethertype = ð->ether_type; + append_template(&template, &off, item->spec, + sizeof(struct rte_flow_item_eth)); + item++; + flow_item_skip_void(&item); + /* Optional VLAN */ + if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) { + const struct rte_flow_item_vlan *spec; + + ENICPMD_LOG(DEBUG, "vxlan-encap: vlan"); + spec = item->spec; + fm_op.encap.outer_vlan = rte_be_to_cpu_16(spec->tci); + item++; + flow_item_skip_void(&item); + } + /* L3 must be IPv4, IPv6 */ + switch (item->type) { + case RTE_FLOW_ITEM_TYPE_IPV4: + { + struct rte_ipv4_hdr *ip4; + + ENICPMD_LOG(DEBUG, "vxlan-encap: ipv4"); + *ethertype = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4); + ip4 = (struct rte_ipv4_hdr *)template; + /* + * Offset of IPv4 length field and its initial value + * (IP + UDP + VXLAN) are specified in the action. The NIC + * will add inner packet length. + */ + fm_op.encap.len1_offset = off + + offsetof(struct rte_ipv4_hdr, total_length); + fm_op.encap.len1_delta = sizeof(struct rte_ipv4_hdr) + + sizeof(struct rte_udp_hdr) + + sizeof(struct rte_vxlan_hdr); + append_template(&template, &off, item->spec, + sizeof(struct rte_ipv4_hdr)); + ip4->version_ihl = IP_VHL_DEF; + if (ip4->time_to_live == 0) + ip4->time_to_live = IP_DEFTTL; + ip4->next_proto_id = IPPROTO_UDP; + break; + } + case RTE_FLOW_ITEM_TYPE_IPV6: + { + struct rte_ipv6_hdr *ip6; + + ENICPMD_LOG(DEBUG, "vxlan-encap: ipv6"); + *ethertype = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6); + ip6 = (struct rte_ipv6_hdr *)template; + fm_op.encap.len1_offset = off + + offsetof(struct rte_ipv6_hdr, payload_len); + fm_op.encap.len1_delta = sizeof(struct rte_udp_hdr) + + sizeof(struct rte_vxlan_hdr); + append_template(&template, &off, item->spec, + sizeof(struct rte_ipv6_hdr)); + ip6->vtc_flow |= rte_cpu_to_be_32(IP6_VTC_FLOW); + if (ip6->hop_limits == 0) + ip6->hop_limits = IP_DEFTTL; + ip6->proto = IPPROTO_UDP; + break; + } + default: + return rte_flow_error_set(error, + EINVAL, RTE_FLOW_ERROR_TYPE_ITEM, item, + "vxlan-encap: L3 must be IPv4/IPv6"); + } + item++; + flow_item_skip_void(&item); + + /* L4 is UDP */ + if (item->type != RTE_FLOW_ITEM_TYPE_UDP) + return rte_flow_error_set(error, EINVAL, + RTE_FLOW_ERROR_TYPE_ITEM, item, + "vxlan-encap: UDP must follow IPv4/IPv6"); + /* UDP length = UDP + VXLAN. NIC will add inner packet length. */ + fm_op.encap.len2_offset = + off + offsetof(struct rte_udp_hdr, dgram_len); + fm_op.encap.len2_delta = + sizeof(struct rte_udp_hdr) + sizeof(struct rte_vxlan_hdr); + append_template(&template, &off, item->spec, + sizeof(struct rte_udp_hdr)); + item++; + flow_item_skip_void(&item); + + /* Finally VXLAN */ + if (item->type != RTE_FLOW_ITEM_TYPE_VXLAN) + return rte_flow_error_set(error, + EINVAL, RTE_FLOW_ERROR_TYPE_ITEM, item, + "vxlan-encap: VXLAN must follow UDP"); + append_template(&template, &off, item->spec, + sizeof(struct rte_flow_item_vxlan)); + + /* + * Fill in the rest of the action structure. + * Indicate that we want to encap with vxlan at packet start. + */ + fm_op.encap.template_offset = 0; + fm_op.encap.template_len = off; + return enic_fm_append_action_op(fm, &fm_op, error); +} + +static int +enic_fm_find_vnic(struct enic *enic, const struct rte_pci_addr *addr, + uint64_t *handle) +{ + uint32_t bdf; + u64 args[2]; + int rc; + + ENICPMD_FUNC_TRACE(); + ENICPMD_LOG(DEBUG, "bdf=%x:%x:%x", addr->bus, addr->devid, + addr->function); + bdf = addr->bus << 8 | addr->devid << 3 | addr->function; + args[0] = FM_VNIC_FIND; + args[1] = bdf; + rc = vnic_dev_flowman_cmd(enic->vdev, args, 2); + if (rc != 0) { + ENICPMD_LOG(ERR, "allocating counters rc=%d", rc); + return rc; + } + *handle = args[0]; + ENICPMD_LOG(DEBUG, "found vnic: handle=0x%" PRIx64, *handle); + return 0; +} + +/* Translate flow actions to flowman TCAM entry actions */ +static int +enic_fm_copy_action(struct enic_flowman *fm, + const struct rte_flow_action actions[], + uint8_t ingress, + struct rte_flow_error *error) +{ + enum { + FATE = 1 << 0, + MARK = 1 << 1, + PASSTHRU = 1 << 2, + COUNT = 1 << 3, + ENCAP = 1 << 4, + }; + struct fm_tcam_match_entry *fmt; + struct fm_action_op fm_op; + struct enic *enic; + uint32_t overlap; + uint64_t vnic_h; + bool first_rq; + int ret; + + ENICPMD_FUNC_TRACE(); + fmt = &fm->tcam_entry; + first_rq = true; + enic = fm->enic; + overlap = 0; + vnic_h = 0; /* 0 = current vNIC */ + for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) { + switch (actions->type) { + case RTE_FLOW_ACTION_TYPE_VOID: + continue; + case RTE_FLOW_ACTION_TYPE_PASSTHRU: { + if (overlap & PASSTHRU) + goto unsupported; + overlap |= PASSTHRU; + break; + } + case RTE_FLOW_ACTION_TYPE_JUMP: { + const struct rte_flow_action_jump *jump = + actions->conf; + struct enic_fm_fet *fet; + + if (overlap & FATE) + goto unsupported; + ret = enic_fet_get(fm, jump->group, ingress, NULL, + &fet, error); + if (ret) + return ret; + overlap |= FATE; + memset(&fm_op, 0, sizeof(fm_op)); + fm_op.fa_op = FMOP_EXACT_MATCH; + fm_op.exact.handle = fet->handle; + fm->fet = fet; + ret = enic_fm_append_action_op(fm, &fm_op, error); + if (ret) + return ret; + break; + } + case RTE_FLOW_ACTION_TYPE_MARK: { + const struct rte_flow_action_mark *mark = + actions->conf; + + if (overlap & MARK) + goto unsupported; + overlap |= MARK; + if (mark->id >= ENIC_MAGIC_FILTER_ID - 1) + return rte_flow_error_set(error, EINVAL, + RTE_FLOW_ERROR_TYPE_ACTION, + NULL, "invalid mark id"); + memset(&fm_op, 0, sizeof(fm_op)); + fm_op.fa_op = FMOP_MARK; + fm_op.mark.mark = mark->id + 1; + ret = enic_fm_append_action_op(fm, &fm_op, error); + if (ret) + return ret; + break; + } + case RTE_FLOW_ACTION_TYPE_FLAG: { + if (overlap & MARK) + goto unsupported; + overlap |= MARK; + /* ENIC_MAGIC_FILTER_ID is reserved for flagging */ + memset(&fm_op, 0, sizeof(fm_op)); + fm_op.fa_op = FMOP_MARK; + fm_op.mark.mark = ENIC_MAGIC_FILTER_ID; + ret = enic_fm_append_action_op(fm, &fm_op, error); + if (ret) + return ret; + break; + } + case RTE_FLOW_ACTION_TYPE_QUEUE: { + const struct rte_flow_action_queue *queue = + actions->conf; + + /* + * If other fate kind is set, fail. Multiple + * queue actions are ok. + */ + if ((overlap & FATE) && first_rq) + goto unsupported; + first_rq = false; + overlap |= FATE; + memset(&fm_op, 0, sizeof(fm_op)); + fm_op.fa_op = FMOP_RQ_STEER; + fm_op.rq_steer.rq_index = + enic_rte_rq_idx_to_sop_idx(queue->index); + fm_op.rq_steer.vnic_handle = vnic_h; + ret = enic_fm_append_action_op(fm, &fm_op, error); + if (ret) + return ret; + ENICPMD_LOG(DEBUG, "create QUEUE action rq: %u", + fm_op.rq_steer.rq_index); + break; + } + case RTE_FLOW_ACTION_TYPE_DROP: { + if (overlap & FATE) + goto unsupported; + overlap |= FATE; + memset(&fm_op, 0, sizeof(fm_op)); + fm_op.fa_op = FMOP_DROP; + ret = enic_fm_append_action_op(fm, &fm_op, error); + if (ret) + return ret; + ENICPMD_LOG(DEBUG, "create DROP action"); + break; + } + case RTE_FLOW_ACTION_TYPE_COUNT: { + if (overlap & COUNT) + goto unsupported; + overlap |= COUNT; + /* Count is associated with entry not action on VIC. */ + fmt->ftm_flags |= FMEF_COUNTER; + break; + } + case RTE_FLOW_ACTION_TYPE_RSS: { + const struct rte_flow_action_rss *rss = actions->conf; + bool allow; + uint16_t i; + + /* + * Hardware does not support general RSS actions, but + * we can still support the dummy one that is used to + * "receive normally". + */ + allow = rss->func == RTE_ETH_HASH_FUNCTION_DEFAULT && + rss->level == 0 && + (rss->types == 0 || + rss->types == enic->rss_hf) && + rss->queue_num == enic->rq_count && + rss->key_len == 0; + /* Identity queue map is ok */ + for (i = 0; i < rss->queue_num; i++) + allow = allow && (i == rss->queue[i]); + if (!allow) + goto unsupported; + if (overlap & FATE) + goto unsupported; + /* Need MARK or FLAG */ + if (!(overlap & MARK)) + goto unsupported; + overlap |= FATE; + break; + } + case RTE_FLOW_ACTION_TYPE_PORT_ID: { + const struct rte_flow_action_port_id *port; + struct rte_pci_device *pdev; + struct rte_eth_dev *dev; + + port = actions->conf; + if (port->original) { + vnic_h = 0; /* This port */ + break; + } + ENICPMD_LOG(DEBUG, "port id %u", port->id); + if (!rte_eth_dev_is_valid_port(port->id)) { + return rte_flow_error_set(error, EINVAL, + RTE_FLOW_ERROR_TYPE_ACTION, + NULL, "invalid port_id"); + } + dev = &rte_eth_devices[port->id]; + if (!dev_is_enic(dev)) { + return rte_flow_error_set(error, EINVAL, + RTE_FLOW_ERROR_TYPE_ACTION, + NULL, "port_id is not enic"); + } + pdev = RTE_ETH_DEV_TO_PCI(dev); + if (enic_fm_find_vnic(enic, &pdev->addr, &vnic_h)) { + return rte_flow_error_set(error, EINVAL, + RTE_FLOW_ERROR_TYPE_ACTION, + NULL, "port_id is not vnic"); + } + break; + } + case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP: { + ret = enic_fm_copy_vxlan_decap(fm, fmt, actions, + error); + if (ret != 0) + return ret; + break; + } + case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP: { + const struct rte_flow_action_vxlan_encap *encap; + + encap = actions->conf; + if (overlap & ENCAP) + goto unsupported; + overlap |= ENCAP; + ret = enic_fm_copy_vxlan_encap(fm, encap->definition, + error); + if (ret != 0) + return ret; + break; + } + default: + goto unsupported; + } + } + + if (!(overlap & (FATE | PASSTHRU | COUNT))) + goto unsupported; + memset(&fm_op, 0, sizeof(fm_op)); + fm_op.fa_op = FMOP_END; + ret = enic_fm_append_action_op(fm, &fm_op, error); + if (ret) + return ret; + enic_fm_reorder_action_op(fm); + return 0; + +unsupported: + return rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION, + NULL, "enic: unsupported action"); +} + +/** Check if the action is supported */ +static int +enic_fm_match_action(const struct rte_flow_action *action, + const enum rte_flow_action_type *supported_actions) +{ + for (; *supported_actions != RTE_FLOW_ACTION_TYPE_END; + supported_actions++) { + if (action->type == *supported_actions) + return 1; + } + return 0; +} + +/* Debug function to dump internal NIC action structure. */ +static void +enic_fm_dump_tcam_actions(const struct fm_action *fm_action) +{ + /* Manually keep in sync with FMOP commands */ + const char *fmop_str[FMOP_OP_MAX] = { + [FMOP_END] = "end", + [FMOP_DROP] = "drop", + [FMOP_RQ_STEER] = "steer", + [FMOP_EXACT_MATCH] = "exmatch", + [FMOP_MARK] = "mark", + [FMOP_EXT_MARK] = "ext_mark", + [FMOP_TAG] = "tag", + [FMOP_EG_HAIRPIN] = "eg_hairpin", + [FMOP_IG_HAIRPIN] = "ig_hairpin", + [FMOP_ENCAP_IVLAN] = "encap_ivlan", + [FMOP_ENCAP_NOIVLAN] = "encap_noivlan", + [FMOP_ENCAP] = "encap", + [FMOP_SET_OVLAN] = "set_ovlan", + [FMOP_DECAP_NOSTRIP] = "decap_nostrip", + }; + const struct fm_action_op *op = &fm_action->fma_action_ops[0]; + char buf[128], *bp = buf; + const char *op_str; + int i, n, buf_len; + + buf[0] = '\0'; + buf_len = sizeof(buf); + for (i = 0; i < FM_ACTION_OP_MAX; i++) { + if (op->fa_op == FMOP_END) + break; + if (op->fa_op >= FMOP_OP_MAX) + op_str = "unknown"; + else + op_str = fmop_str[op->fa_op]; + n = snprintf(bp, buf_len, "%s,", op_str); + if (n > 0 && n < buf_len) { + bp += n; + buf_len -= n; + } + op++; + } + /* Remove trailing comma */ + if (buf[0]) + *(bp - 1) = '\0'; + ENICPMD_LOG(DEBUG, " Acions: %s", buf); +} + +static int +bits_to_str(uint32_t bits, const char *strings[], int max, + char *buf, int buf_len) +{ + int i, n = 0, len = 0; + + for (i = 0; i < max; i++) { + if (bits & (1 << i)) { + n = snprintf(buf, buf_len, "%s,", strings[i]); + if (n > 0 && n < buf_len) { + buf += n; + buf_len -= n; + len += n; + } + } + } + /* Remove trailing comma */ + if (len) { + *(buf - 1) = '\0'; + len--; + } + return len; +} + +/* Debug function to dump internal NIC filter structure. */ +static void +__enic_fm_dump_tcam_match(const struct fm_header_set *fk_hdrset, char *buf, + int buf_len) +{ + /* Manually keep in sync with FKM_BITS */ + const char *fm_fkm_str[FKM_BIT_COUNT] = { + [FKM_QTAG_BIT] = "qtag", + [FKM_CMD_BIT] = "cmd", + [FKM_IPV4_BIT] = "ip4", + [FKM_IPV6_BIT] = "ip6", + [FKM_ROCE_BIT] = "roce", + [FKM_UDP_BIT] = "udp", + [FKM_TCP_BIT] = "tcp", + [FKM_TCPORUDP_BIT] = "tcpportudp", + [FKM_IPFRAG_BIT] = "ipfrag", + [FKM_NVGRE_BIT] = "nvgre", + [FKM_VXLAN_BIT] = "vxlan", + [FKM_GENEVE_BIT] = "geneve", + [FKM_NSH_BIT] = "nsh", + [FKM_ROCEV2_BIT] = "rocev2", + [FKM_VLAN_PRES_BIT] = "vlan_pres", + [FKM_IPOK_BIT] = "ipok", + [FKM_L4OK_BIT] = "l4ok", + [FKM_ROCEOK_BIT] = "roceok", + [FKM_FCSOK_BIT] = "fcsok", + [FKM_EG_SPAN_BIT] = "eg_span", + [FKM_IG_SPAN_BIT] = "ig_span", + [FKM_EG_HAIRPINNED_BIT] = "eg_hairpinned", + }; + /* Manually keep in sync with FKH_BITS */ + const char *fm_fkh_str[FKH_BIT_COUNT] = { + [FKH_ETHER_BIT] = "eth", + [FKH_QTAG_BIT] = "qtag", + [FKH_L2RAW_BIT] = "l2raw", + [FKH_IPV4_BIT] = "ip4", + [FKH_IPV6_BIT] = "ip6", + [FKH_L3RAW_BIT] = "l3raw", + [FKH_UDP_BIT] = "udp", + [FKH_TCP_BIT] = "tcp", + [FKH_ICMP_BIT] = "icmp", + [FKH_VXLAN_BIT] = "vxlan", + [FKH_L4RAW_BIT] = "l4raw", + }; + uint32_t fkh_bits = fk_hdrset->fk_header_select; + uint32_t fkm_bits = fk_hdrset->fk_metadata; + int n; + + if (!fkm_bits && !fkh_bits) + return; + n = snprintf(buf, buf_len, "metadata("); + if (n > 0 && n < buf_len) { + buf += n; + buf_len -= n; + } + n = bits_to_str(fkm_bits, fm_fkm_str, FKM_BIT_COUNT, buf, buf_len); + if (n > 0 && n < buf_len) { + buf += n; + buf_len -= n; + } + n = snprintf(buf, buf_len, ") valid hdr fields("); + if (n > 0 && n < buf_len) { + buf += n; + buf_len -= n; + } + n = bits_to_str(fkh_bits, fm_fkh_str, FKH_BIT_COUNT, buf, buf_len); + if (n > 0 && n < buf_len) { + buf += n; + buf_len -= n; + } + snprintf(buf, buf_len, ")"); +} + +static void +enic_fm_dump_tcam_match(const struct fm_tcam_match_entry *match, + uint8_t ingress) +{ + char buf[256]; + + memset(buf, 0, sizeof(buf)); + __enic_fm_dump_tcam_match(&match->ftm_mask.fk_hdrset[0], + buf, sizeof(buf)); + ENICPMD_LOG(DEBUG, " TCAM %s Outer: %s %scounter", + (ingress) ? "IG" : "EG", buf, + (match->ftm_flags & FMEF_COUNTER) ? "" : "no "); + memset(buf, 0, sizeof(buf)); + __enic_fm_dump_tcam_match(&match->ftm_mask.fk_hdrset[1], + buf, sizeof(buf)); + if (buf[0]) + ENICPMD_LOG(DEBUG, " Inner: %s", buf); +} + +/* Debug function to dump internal NIC flow structures. */ +static void +enic_fm_dump_tcam_entry(const struct fm_tcam_match_entry *fm_match, + const struct fm_action *fm_action, + uint8_t ingress) +{ + if (rte_log_get_level(enic_pmd_logtype) < (int)RTE_LOG_DEBUG) + return; + enic_fm_dump_tcam_match(fm_match, ingress); + enic_fm_dump_tcam_actions(fm_action); +} + +static int +enic_fm_flow_parse(struct enic_flowman *fm, + const struct rte_flow_attr *attrs, + const struct rte_flow_item pattern[], + const struct rte_flow_action actions[], + struct rte_flow_error *error) +{ + const struct rte_flow_action *action; + unsigned int ret; + static const enum rte_flow_action_type *sa; + + ENICPMD_FUNC_TRACE(); + ret = 0; + if (!pattern) { + rte_flow_error_set(error, EINVAL, RTE_FLOW_ERROR_TYPE_ITEM_NUM, + NULL, "no pattern specified"); + return -rte_errno; + } + + if (!actions) { + rte_flow_error_set(error, EINVAL, + RTE_FLOW_ERROR_TYPE_ACTION_NUM, + NULL, "no action specified"); + return -rte_errno; + } + + if (attrs) { + if (attrs->priority) { + rte_flow_error_set(error, ENOTSUP, + RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY, + NULL, + "priorities are not supported"); + return -rte_errno; + } else if (attrs->transfer) { + rte_flow_error_set(error, ENOTSUP, + RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER, + NULL, + "transfer is not supported"); + return -rte_errno; + } else if (attrs->ingress && attrs->egress) { + rte_flow_error_set(error, ENOTSUP, + RTE_FLOW_ERROR_TYPE_ATTR_INGRESS, + NULL, + "bidirectional rules not supported"); + return -rte_errno; + } + + } else { + rte_flow_error_set(error, EINVAL, + RTE_FLOW_ERROR_TYPE_ATTR, + NULL, "no attribute specified"); + return -rte_errno; + } + + /* Verify Actions. */ + sa = (attrs->ingress) ? enic_fm_supported_ig_actions : + enic_fm_supported_eg_actions; + for (action = &actions[0]; action->type != RTE_FLOW_ACTION_TYPE_END; + action++) { + if (action->type == RTE_FLOW_ACTION_TYPE_VOID) + continue; + else if (!enic_fm_match_action(action, sa)) + break; + } + if (action->type != RTE_FLOW_ACTION_TYPE_END) { + rte_flow_error_set(error, EPERM, RTE_FLOW_ERROR_TYPE_ACTION, + action, "invalid action"); + return -rte_errno; + } + ret = enic_fm_copy_entry(fm, pattern, error); + if (ret) + return ret; + ret = enic_fm_copy_action(fm, actions, attrs->ingress, error); + return ret; +} + +static void +enic_fm_counter_free(struct enic_flowman *fm, struct enic_fm_flow *fm_flow) +{ + if (!fm_flow->counter_valid) + return; + SLIST_INSERT_HEAD(&fm->counters, fm_flow->counter, next); + fm_flow->counter_valid = false; +} + +static int +enic_fm_more_counters(struct enic_flowman *fm) +{ + struct enic_fm_counter *new_stack; + struct enic_fm_counter *ctrs; + struct enic *enic; + int i, rc; + u64 args[2]; + + ENICPMD_FUNC_TRACE(); + enic = fm->enic; + new_stack = rte_realloc(fm->counter_stack, (fm->counters_alloced + + FM_COUNTERS_EXPAND) * + sizeof(struct enic_fm_counter), 0); + if (new_stack == NULL) { + ENICPMD_LOG(ERR, "cannot alloc counter memory"); + return -ENOMEM; + } + fm->counter_stack = new_stack; + + args[0] = FM_COUNTER_BRK; + args[1] = fm->counters_alloced + FM_COUNTERS_EXPAND; + rc = vnic_dev_flowman_cmd(enic->vdev, args, 2); + if (rc != 0) { + ENICPMD_LOG(ERR, "cannot alloc counters rc=%d", rc); + return rc; + } + ctrs = (struct enic_fm_counter *)fm->counter_stack + + fm->counters_alloced; + for (i = 0; i < FM_COUNTERS_EXPAND; i++, ctrs++) { + ctrs->handle = fm->counters_alloced + i; + SLIST_INSERT_HEAD(&fm->counters, ctrs, next); + } + fm->counters_alloced += FM_COUNTERS_EXPAND; + ENICPMD_LOG(DEBUG, "%u counters allocated, total: %u", + FM_COUNTERS_EXPAND, fm->counters_alloced); + return 0; +} + +static int +enic_fm_counter_zero(struct enic_flowman *fm, struct enic_fm_counter *c) +{ + struct enic *enic; + u64 args[3]; + int ret; + + ENICPMD_FUNC_TRACE(); + enic = fm->enic; + args[0] = FM_COUNTER_QUERY; + args[1] = c->handle; + args[2] = 1; /* clear */ + ret = vnic_dev_flowman_cmd(enic->vdev, args, 3); + if (ret) { + ENICPMD_LOG(ERR, "counter init: rc=%d handle=0x%x", + ret, c->handle); + return ret; + } + return 0; +} + +static int +enic_fm_counter_alloc(struct enic_flowman *fm, struct rte_flow_error *error, + struct enic_fm_counter **ctr) +{ + struct enic_fm_counter *c; + int ret; + + ENICPMD_FUNC_TRACE(); + *ctr = NULL; + if (SLIST_EMPTY(&fm->counters)) { + ret = enic_fm_more_counters(fm); + if (ret) + return rte_flow_error_set(error, -ret, + RTE_FLOW_ERROR_TYPE_UNSPECIFIED, + NULL, "enic: out of counters"); + } + c = SLIST_FIRST(&fm->counters); + SLIST_REMOVE_HEAD(&fm->counters, next); + *ctr = c; + return 0; +} + +static int +enic_fm_action_free(struct enic_flowman *fm, uint64_t handle) +{ + u64 args[2]; + int rc; + + ENICPMD_FUNC_TRACE(); + args[0] = FM_ACTION_FREE; + args[1] = handle; + rc = vnic_dev_flowman_cmd(fm->enic->vdev, args, 2); + if (rc) + ENICPMD_LOG(ERR, "cannot free action: rc=%d handle=0x%" PRIx64, + rc, handle); + return rc; +} + +static int +enic_fm_entry_free(struct enic_flowman *fm, uint64_t handle) +{ + u64 args[2]; + int rc; + + ENICPMD_FUNC_TRACE(); + args[0] = FM_MATCH_ENTRY_REMOVE; + args[1] = handle; + rc = vnic_dev_flowman_cmd(fm->enic->vdev, args, 2); + if (rc) + ENICPMD_LOG(ERR, "cannot free match entry: rc=%d" + " handle=0x%" PRIx64, rc, handle); + return rc; +} + +static struct enic_fm_jump_flow * +find_jump_flow(struct enic_flowman *fm, uint32_t group) +{ + struct enic_fm_jump_flow *j; + + ENICPMD_FUNC_TRACE(); + TAILQ_FOREACH(j, &fm->jump_list, list) { + if (j->group == group) + return j; + } + return NULL; +} + +static void +remove_jump_flow(struct enic_flowman *fm, struct rte_flow *flow) +{ + struct enic_fm_jump_flow *j; + + ENICPMD_FUNC_TRACE(); + TAILQ_FOREACH(j, &fm->jump_list, list) { + if (j->flow == flow) { + TAILQ_REMOVE(&fm->jump_list, j, list); + free(j); + return; + } + } +} + +static int +save_jump_flow(struct enic_flowman *fm, + struct rte_flow *flow, + uint32_t group, + struct fm_tcam_match_entry *match, + struct fm_action *action) +{ + struct enic_fm_jump_flow *j; + + ENICPMD_FUNC_TRACE(); + j = calloc(1, sizeof(struct enic_fm_jump_flow)); + if (j == NULL) + return -ENOMEM; + j->flow = flow; + j->group = group; + j->match = *match; + j->action = *action; + TAILQ_INSERT_HEAD(&fm->jump_list, j, list); + ENICPMD_LOG(DEBUG, "saved jump flow: flow=%p group=%u", flow, group); + return 0; +} + +static void +__enic_fm_flow_free(struct enic_flowman *fm, struct enic_fm_flow *fm_flow) +{ + if (fm_flow->entry_handle != FM_INVALID_HANDLE) { + enic_fm_entry_free(fm, fm_flow->entry_handle); + fm_flow->entry_handle = FM_INVALID_HANDLE; + } + if (fm_flow->action_handle != FM_INVALID_HANDLE) { + enic_fm_action_free(fm, fm_flow->action_handle); + fm_flow->action_handle = FM_INVALID_HANDLE; + } + enic_fm_counter_free(fm, fm_flow); + if (fm_flow->fet) { + enic_fet_put(fm, fm_flow->fet); + fm_flow->fet = NULL; + } +} + +static void +enic_fm_flow_free(struct enic_flowman *fm, struct rte_flow *flow) +{ + if (flow->fm->fet && flow->fm->fet->default_key) + remove_jump_flow(fm, flow); + __enic_fm_flow_free(fm, flow->fm); + free(flow->fm); + free(flow); +} + +static int +enic_fm_add_tcam_entry(struct enic_flowman *fm, + struct fm_tcam_match_entry *match_in, + uint64_t *entry_handle, + uint8_t ingress, + struct rte_flow_error *error) +{ + struct fm_tcam_match_entry *ftm; + u64 args[3]; + int ret; + + ENICPMD_FUNC_TRACE(); + /* Copy entry to the command buffer */ + ftm = &fm->cmd.va->fm_tcam_match_entry; + memcpy(ftm, match_in, sizeof(*ftm)); + /* Add TCAM entry */ + args[0] = FM_TCAM_ENTRY_INSTALL; + args[1] = ingress ? fm->ig_tcam_hndl : fm->eg_tcam_hndl; + args[2] = fm->cmd.pa; + ret = vnic_dev_flowman_cmd(fm->enic->vdev, args, 3); + if (ret != 0) { + ENICPMD_LOG(ERR, "cannot add %s TCAM entry: rc=%d", + ingress ? "ingress" : "egress", ret); + rte_flow_error_set(error, ret, RTE_FLOW_ERROR_TYPE_UNSPECIFIED, + NULL, "enic: devcmd(tcam-entry-install)"); + return ret; + } + ENICPMD_LOG(DEBUG, "installed %s TCAM entry: handle=0x%" PRIx64, + ingress ? "ingress" : "egress", (uint64_t)args[0]); + *entry_handle = args[0]; + return 0; +} + +static int +enic_fm_add_exact_entry(struct enic_flowman *fm, + struct fm_tcam_match_entry *match_in, + uint64_t *entry_handle, + struct enic_fm_fet *fet, + struct rte_flow_error *error) +{ + struct fm_exact_match_entry *fem; + u64 args[3]; + int ret; + + ENICPMD_FUNC_TRACE(); + /* The new entry must have the table's key */ + if (memcmp(fet->key.fk_hdrset, match_in->ftm_mask.fk_hdrset, + sizeof(struct fm_header_set) * FM_HDRSET_MAX)) { + return rte_flow_error_set(error, EINVAL, + RTE_FLOW_ERROR_TYPE_ITEM, NULL, + "enic: key does not match group's key"); + } + + /* Copy entry to the command buffer */ + fem = &fm->cmd.va->fm_exact_match_entry; + /* + * Translate TCAM entry to exact entry. As is only need to drop + * position and mask. The mask is part of the exact match table. + * Position (aka priority) is not supported in the exact match table. + */ + fem->fem_data = match_in->ftm_data; + fem->fem_flags = match_in->ftm_flags; + fem->fem_action = match_in->ftm_action; + fem->fem_counter = match_in->ftm_counter; + + /* Add exact entry */ + args[0] = FM_EXACT_ENTRY_INSTALL; + args[1] = fet->handle; + args[2] = fm->cmd.pa; + ret = vnic_dev_flowman_cmd(fm->enic->vdev, args, 3); + if (ret != 0) { + ENICPMD_LOG(ERR, "cannot add %s exact entry: group=%u", + fet->ingress ? "ingress" : "egress", fet->group); + rte_flow_error_set(error, ret, RTE_FLOW_ERROR_TYPE_UNSPECIFIED, + NULL, "enic: devcmd(exact-entry-install)"); + return ret; + } + ENICPMD_LOG(DEBUG, "installed %s exact entry: group=%u" + " handle=0x%" PRIx64, + fet->ingress ? "ingress" : "egress", fet->group, + (uint64_t)args[0]); + *entry_handle = args[0]; + return 0; +} + +/* Push match-action to the NIC. */ +static int +__enic_fm_flow_add_entry(struct enic_flowman *fm, + struct enic_fm_flow *fm_flow, + struct fm_tcam_match_entry *match_in, + struct fm_action *action_in, + uint32_t group, + uint8_t ingress, + struct rte_flow_error *error) +{ + struct enic_fm_counter *ctr; + struct fm_action *fma; + uint64_t action_h; + uint64_t entry_h; + u64 args[3]; + int ret; + + ENICPMD_FUNC_TRACE(); + /* Allocate action. */ + fma = &fm->cmd.va->fm_action; + memcpy(fma, action_in, sizeof(*fma)); + args[0] = FM_ACTION_ALLOC; + args[1] = fm->cmd.pa; + ret = vnic_dev_flowman_cmd(fm->enic->vdev, args, 2); + if (ret != 0) { + ENICPMD_LOG(ERR, "allocating TCAM table action rc=%d", ret); + rte_flow_error_set(error, ret, RTE_FLOW_ERROR_TYPE_UNSPECIFIED, + NULL, "enic: devcmd(action-alloc)"); + return ret; + } + action_h = args[0]; + fm_flow->action_handle = action_h; + match_in->ftm_action = action_h; + ENICPMD_LOG(DEBUG, "action allocated: handle=0x%" PRIx64, action_h); + + /* Allocate counter if requested. */ + if (match_in->ftm_flags & FMEF_COUNTER) { + ret = enic_fm_counter_alloc(fm, error, &ctr); + if (ret) /* error has been filled in */ + return ret; + fm_flow->counter_valid = true; + fm_flow->counter = ctr; + match_in->ftm_counter = ctr->handle; + } + + /* + * Get the group's table (either TCAM or exact match table) and + * add entry to it. If we use the exact match table, the handler + * will translate the TCAM entry (match_in) to the appropriate + * exact match entry and use that instead. + */ + entry_h = FM_INVALID_HANDLE; + if (group == FM_TCAM_RTE_GROUP) { + ret = enic_fm_add_tcam_entry(fm, match_in, &entry_h, ingress, + error); + if (ret) + return ret; + /* Jump action might have a ref to fet */ + fm_flow->fet = fm->fet; + fm->fet = NULL; + } else { + struct enic_fm_fet *fet = NULL; + + ret = enic_fet_get(fm, group, ingress, + &match_in->ftm_mask, &fet, error); + if (ret) + return ret; + fm_flow->fet = fet; + ret = enic_fm_add_exact_entry(fm, match_in, &entry_h, fet, + error); + if (ret) + return ret; + } + /* Clear counter after adding entry, as it requires in-use counter */ + if (fm_flow->counter_valid) { + ret = enic_fm_counter_zero(fm, fm_flow->counter); + if (ret) + return ret; + } + fm_flow->entry_handle = entry_h; + return 0; +} + +/* Push match-action to the NIC. */ +static struct rte_flow * +enic_fm_flow_add_entry(struct enic_flowman *fm, + struct fm_tcam_match_entry *match_in, + struct fm_action *action_in, + const struct rte_flow_attr *attrs, + struct rte_flow_error *error) +{ + struct enic_fm_flow *fm_flow; + struct rte_flow *flow; + + ENICPMD_FUNC_TRACE(); + enic_fm_dump_tcam_entry(match_in, action_in, attrs->ingress); + flow = calloc(1, sizeof(*flow)); + fm_flow = calloc(1, sizeof(*fm_flow)); + if (flow == NULL || fm_flow == NULL) { + rte_flow_error_set(error, ENOMEM, RTE_FLOW_ERROR_TYPE_HANDLE, + NULL, "enic: cannot allocate rte_flow"); + free(flow); + free(fm_flow); + return NULL; + } + flow->fm = fm_flow; + fm_flow->action_handle = FM_INVALID_HANDLE; + fm_flow->entry_handle = FM_INVALID_HANDLE; + if (__enic_fm_flow_add_entry(fm, fm_flow, match_in, action_in, + attrs->group, attrs->ingress, error)) { + enic_fm_flow_free(fm, flow); + return NULL; + } + return flow; +} + +static void +convert_jump_flows(struct enic_flowman *fm, struct enic_fm_fet *fet, + struct rte_flow_error *error) +{ + struct enic_fm_flow *fm_flow; + struct enic_fm_jump_flow *j; + struct fm_action *fma; + uint32_t group; + + ENICPMD_FUNC_TRACE(); + /* + * Find the saved flows that should jump to the new table (fet). + * Then delete the old TCAM entry that jumps to the default table, + * and add a new one that jumps to the new table. + */ + group = fet->group; + j = find_jump_flow(fm, group); + while (j) { + ENICPMD_LOG(DEBUG, "convert jump flow: flow=%p group=%u", + j->flow, group); + /* Delete old entry */ + fm_flow = j->flow->fm; + __enic_fm_flow_free(fm, fm_flow); + + /* Add new entry */ + fma = &j->action; + fma->fma_action_ops[0].exact.handle = fet->handle; + if (__enic_fm_flow_add_entry(fm, fm_flow, &j->match, fma, + FM_TCAM_RTE_GROUP, fet->ingress, error)) { + /* Cannot roll back changes at the moment */ + ENICPMD_LOG(ERR, "cannot convert jump flow: flow=%p", + j->flow); + } else { + fm_flow->fet = fet; + fet->ref++; + ENICPMD_LOG(DEBUG, "convert ok: group=%u ref=%u", + fet->group, fet->ref); + } + + TAILQ_REMOVE(&fm->jump_list, j, list); + free(j); + j = find_jump_flow(fm, group); + } +} + +static void +enic_fm_open_scratch(struct enic_flowman *fm) +{ + fm->action_op_count = 0; + fm->fet = NULL; + memset(&fm->tcam_entry, 0, sizeof(fm->tcam_entry)); + memset(&fm->action, 0, sizeof(fm->action)); +} + +static void +enic_fm_close_scratch(struct enic_flowman *fm) +{ + if (fm->fet) { + enic_fet_put(fm, fm->fet); + fm->fet = NULL; + } + fm->action_op_count = 0; +} + +static int +enic_fm_flow_validate(struct rte_eth_dev *dev, + const struct rte_flow_attr *attrs, + const struct rte_flow_item pattern[], + const struct rte_flow_action actions[], + struct rte_flow_error *error) +{ + struct fm_tcam_match_entry *fm_tcam_entry; + struct fm_action *fm_action; + struct enic_flowman *fm; + int ret; + + ENICPMD_FUNC_TRACE(); + fm = pmd_priv(dev)->fm; + if (fm == NULL) + return -ENOTSUP; + enic_fm_open_scratch(fm); + ret = enic_fm_flow_parse(fm, attrs, pattern, actions, error); + if (!ret) { + fm_tcam_entry = &fm->tcam_entry; + fm_action = &fm->action; + enic_fm_dump_tcam_entry(fm_tcam_entry, fm_action, + attrs->ingress); + } + enic_fm_close_scratch(fm); + return ret; +} + +static int +enic_fm_flow_query_count(struct rte_eth_dev *dev, + struct rte_flow *flow, void *data, + struct rte_flow_error *error) +{ + struct rte_flow_query_count *query; + struct enic_fm_flow *fm_flow; + struct enic *enic; + u64 args[3]; + int rc; + + ENICPMD_FUNC_TRACE(); + enic = pmd_priv(dev); + query = data; + fm_flow = flow->fm; + if (!fm_flow->counter_valid) + return rte_flow_error_set(error, ENOTSUP, + RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, + "enic: flow does not have counter"); + + args[0] = FM_COUNTER_QUERY; + args[1] = fm_flow->counter->handle; + args[2] = query->reset; + rc = vnic_dev_flowman_cmd(enic->vdev, args, 3); + if (rc) { + ENICPMD_LOG(ERR, "cannot query counter: rc=%d handle=0x%x", + rc, fm_flow->counter->handle); + return rc; + } + query->hits_set = 1; + query->hits = args[0]; + query->bytes_set = 1; + query->bytes = args[1]; + return 0; +} + +static int +enic_fm_flow_query(struct rte_eth_dev *dev, + struct rte_flow *flow, + const struct rte_flow_action *actions, + void *data, + struct rte_flow_error *error) +{ + int ret = 0; + + ENICPMD_FUNC_TRACE(); + for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) { + switch (actions->type) { + case RTE_FLOW_ACTION_TYPE_VOID: + break; + case RTE_FLOW_ACTION_TYPE_COUNT: + ret = enic_fm_flow_query_count(dev, flow, data, error); + break; + default: + return rte_flow_error_set(error, ENOTSUP, + RTE_FLOW_ERROR_TYPE_ACTION, + actions, + "action not supported"); + } + if (ret < 0) + return ret; + } + return 0; +} + +static struct rte_flow * +enic_fm_flow_create(struct rte_eth_dev *dev, + const struct rte_flow_attr *attrs, + const struct rte_flow_item pattern[], + const struct rte_flow_action actions[], + struct rte_flow_error *error) +{ + struct fm_tcam_match_entry *fm_tcam_entry; + struct fm_action *fm_action; + struct enic_flowman *fm; + struct enic_fm_fet *fet; + struct rte_flow *flow; + struct enic *enic; + int ret; + + ENICPMD_FUNC_TRACE(); + enic = pmd_priv(dev); + fm = enic->fm; + if (fm == NULL) { + rte_flow_error_set(error, ENOTSUP, + RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, + "flowman is not initialized"); + return NULL; + } + enic_fm_open_scratch(fm); + flow = NULL; + ret = enic_fm_flow_parse(fm, attrs, pattern, actions, error); + if (ret < 0) + goto error_with_scratch; + fm_tcam_entry = &fm->tcam_entry; + fm_action = &fm->action; + flow = enic_fm_flow_add_entry(fm, fm_tcam_entry, fm_action, + attrs, error); + if (flow) { + LIST_INSERT_HEAD(&enic->flows, flow, next); + fet = flow->fm->fet; + if (fet && fet->default_key) { + /* + * Jump to non-existent group? Save the relevant info + * so we can convert this flow when that group + * materializes. + */ + save_jump_flow(fm, flow, fet->group, + fm_tcam_entry, fm_action); + } else if (fet && fet->ref == 1) { + /* + * A new table is created. Convert the saved flows + * that should jump to this group. + */ + convert_jump_flows(fm, fet, error); + } + } + +error_with_scratch: + enic_fm_close_scratch(fm); + return flow; +} + +static int +enic_fm_flow_destroy(struct rte_eth_dev *dev, struct rte_flow *flow, + __rte_unused struct rte_flow_error *error) +{ + struct enic *enic = pmd_priv(dev); + + ENICPMD_FUNC_TRACE(); + if (enic->fm == NULL) + return 0; + LIST_REMOVE(flow, next); + enic_fm_flow_free(enic->fm, flow); + return 0; +} + +static int +enic_fm_flow_flush(struct rte_eth_dev *dev, + __rte_unused struct rte_flow_error *error) +{ + struct enic_fm_flow *fm_flow; + struct enic_flowman *fm; + struct rte_flow *flow; + struct enic *enic = pmd_priv(dev); + + ENICPMD_FUNC_TRACE(); + if (enic->fm == NULL) + return 0; + fm = enic->fm; + while (!LIST_EMPTY(&enic->flows)) { + flow = LIST_FIRST(&enic->flows); + fm_flow = flow->fm; + LIST_REMOVE(flow, next); + /* + * If tables are null, then vNIC is closing, and the firmware + * has already cleaned up flowman state. So do not try to free + * resources, as it only causes errors. + */ + if (fm->ig_tcam_hndl == FM_INVALID_HANDLE) { + fm_flow->entry_handle = FM_INVALID_HANDLE; + fm_flow->action_handle = FM_INVALID_HANDLE; + fm_flow->fet = NULL; + } + enic_fm_flow_free(fm, flow); + } + return 0; +} + +static int +enic_fm_tbl_free(struct enic_flowman *fm, uint64_t handle) +{ + u64 args[2]; + int rc; + + args[0] = FM_MATCH_TABLE_FREE; + args[1] = handle; + rc = vnic_dev_flowman_cmd(fm->enic->vdev, args, 2); + if (rc) + ENICPMD_LOG(ERR, "cannot free table: rc=%d handle=0x%" PRIx64, + rc, handle); + return rc; +} + +static int +enic_fm_tcam_tbl_alloc(struct enic_flowman *fm, uint32_t direction, + uint32_t max_entries, uint64_t *handle) +{ + struct fm_tcam_match_table *tcam_tbl; + struct enic *enic; + u64 args[2]; + int rc; + + ENICPMD_FUNC_TRACE(); + enic = fm->enic; + tcam_tbl = &fm->cmd.va->fm_tcam_match_table; + tcam_tbl->ftt_direction = direction; + tcam_tbl->ftt_stage = FM_STAGE_LAST; + tcam_tbl->ftt_max_entries = max_entries; + args[0] = FM_TCAM_TABLE_ALLOC; + args[1] = fm->cmd.pa; + rc = vnic_dev_flowman_cmd(enic->vdev, args, 2); + if (rc) { + ENICPMD_LOG(ERR, "cannot alloc %s TCAM table: rc=%d", + (direction == FM_INGRESS) ? "IG" : "EG", rc); + return rc; + } + *handle = args[0]; + ENICPMD_LOG(DEBUG, "%s TCAM table allocated, handle=0x%" PRIx64, + (direction == FM_INGRESS) ? "IG" : "EG", *handle); + return 0; +} + +static int +enic_fm_init_counters(struct enic_flowman *fm) +{ + ENICPMD_FUNC_TRACE(); + SLIST_INIT(&fm->counters); + return enic_fm_more_counters(fm); +} + +static void +enic_fm_free_all_counters(struct enic_flowman *fm) +{ + struct enic *enic; + u64 args[2]; + int rc; + + enic = fm->enic; + args[0] = FM_COUNTER_BRK; + args[1] = 0; + rc = vnic_dev_flowman_cmd(enic->vdev, args, 2); + if (rc != 0) + ENICPMD_LOG(ERR, "cannot free counters: rc=%d", rc); + rte_free(fm->counter_stack); +} + +static int +enic_fm_alloc_tcam_tables(struct enic_flowman *fm) +{ + int rc; + + ENICPMD_FUNC_TRACE(); + rc = enic_fm_tcam_tbl_alloc(fm, FM_INGRESS, FM_MAX_TCAM_TABLE_SIZE, + &fm->ig_tcam_hndl); + if (rc) + return rc; + rc = enic_fm_tcam_tbl_alloc(fm, FM_EGRESS, FM_MAX_TCAM_TABLE_SIZE, + &fm->eg_tcam_hndl); + return rc; +} + +static void +enic_fm_free_tcam_tables(struct enic_flowman *fm) +{ + ENICPMD_FUNC_TRACE(); + if (fm->ig_tcam_hndl) { + ENICPMD_LOG(DEBUG, "free IG TCAM table handle=0x%" PRIx64, + fm->ig_tcam_hndl); + enic_fm_tbl_free(fm, fm->ig_tcam_hndl); + fm->ig_tcam_hndl = FM_INVALID_HANDLE; + } + if (fm->eg_tcam_hndl) { + ENICPMD_LOG(DEBUG, "free EG TCAM table handle=0x%" PRIx64, + fm->eg_tcam_hndl); + enic_fm_tbl_free(fm, fm->eg_tcam_hndl); + fm->eg_tcam_hndl = FM_INVALID_HANDLE; + } +} + +int +enic_fm_init(struct enic *enic) +{ + struct enic_flowman *fm; + u8 name[NAME_MAX]; + int rc; + + if (enic->flow_filter_mode != FILTER_FLOWMAN) + return 0; + ENICPMD_FUNC_TRACE(); + fm = calloc(1, sizeof(*fm)); + if (fm == NULL) { + ENICPMD_LOG(ERR, "cannot alloc flowman struct"); + return -ENOMEM; + } + fm->enic = enic; + TAILQ_INIT(&fm->fet_list); + TAILQ_INIT(&fm->jump_list); + /* Allocate host memory for flowman commands */ + snprintf((char *)name, NAME_MAX, "fm-cmd-%s", enic->bdf_name); + fm->cmd.va = enic_alloc_consistent(enic, + sizeof(union enic_flowman_cmd_mem), &fm->cmd.pa, name); + if (!fm->cmd.va) { + ENICPMD_LOG(ERR, "cannot allocate flowman command memory"); + rc = -ENOMEM; + goto error_fm; + } + /* Allocate TCAM tables upfront as they are the main tables */ + rc = enic_fm_alloc_tcam_tables(fm); + if (rc) { + ENICPMD_LOG(ERR, "cannot alloc TCAM tables"); + goto error_cmd; + } + /* Then a number of counters */ + rc = enic_fm_init_counters(fm); + if (rc) { + ENICPMD_LOG(ERR, "cannot alloc counters"); + goto error_tables; + } + /* + * One default exact match table for each direction. We hold onto + * it until close. + */ + rc = enic_fet_alloc(fm, 1, NULL, 128, &fm->default_ig_fet); + if (rc) { + ENICPMD_LOG(ERR, "cannot alloc default IG exact match table"); + goto error_counters; + } + fm->default_ig_fet->ref = 1; + rc = enic_fet_alloc(fm, 0, NULL, 128, &fm->default_eg_fet); + if (rc) { + ENICPMD_LOG(ERR, "cannot alloc default EG exact match table"); + goto error_ig_fet; + } + fm->default_eg_fet->ref = 1; + enic->fm = fm; + return 0; + +error_ig_fet: + enic_fet_free(fm, fm->default_ig_fet); +error_counters: + enic_fm_free_all_counters(fm); +error_tables: + enic_fm_free_tcam_tables(fm); +error_cmd: + enic_free_consistent(enic, sizeof(union enic_flowman_cmd_mem), + fm->cmd.va, fm->cmd.pa); +error_fm: + free(fm); + return rc; +} + +void +enic_fm_destroy(struct enic *enic) +{ + struct enic_flowman *fm; + struct enic_fm_fet *fet; + + if (enic->fm == NULL) + return; + ENICPMD_FUNC_TRACE(); + fm = enic->fm; + enic_fet_free(fm, fm->default_eg_fet); + enic_fet_free(fm, fm->default_ig_fet); + /* Free all exact match tables still open */ + while (!TAILQ_EMPTY(&fm->fet_list)) { + fet = TAILQ_FIRST(&fm->fet_list); + enic_fet_free(fm, fet); + } + enic_fm_free_tcam_tables(fm); + enic_fm_free_all_counters(fm); + enic_free_consistent(enic, sizeof(union enic_flowman_cmd_mem), + fm->cmd.va, fm->cmd.pa); + fm->cmd.va = NULL; + free(fm); + enic->fm = NULL; +} + +const struct rte_flow_ops enic_fm_flow_ops = { + .validate = enic_fm_flow_validate, + .create = enic_fm_flow_create, + .destroy = enic_fm_flow_destroy, + .flush = enic_fm_flow_flush, + .query = enic_fm_flow_query, +}; diff --git a/drivers/net/enic/enic_main.c b/drivers/net/enic/enic_main.c index fa993c657..40bdd0963 100644 --- a/drivers/net/enic/enic_main.c +++ b/drivers/net/enic/enic_main.c @@ -350,7 +350,7 @@ enic_initial_post_rx(struct enic *enic, struct vnic_rq *rq) rq->need_initial_post = false; } -static void * +void * enic_alloc_consistent(void *priv, size_t size, dma_addr_t *dma_handle, u8 *name) { @@ -390,7 +390,7 @@ enic_alloc_consistent(void *priv, size_t size, return vaddr; } -static void +void enic_free_consistent(void *priv, __rte_unused size_t size, void *vaddr, @@ -587,6 +587,9 @@ int enic_enable(struct enic *enic) dev_warning(enic, "Init of hash table for clsf failed."\ "Flow director feature will not work\n"); + if (enic_fm_init(enic)) + dev_warning(enic, "Init of flowman failed.\n"); + for (index = 0; index < enic->rq_count; index++) { err = enic_alloc_rx_queue_mbufs(enic, &enic->rq[enic_rte_rq_idx_to_sop_idx(index)]); @@ -1043,6 +1046,7 @@ int enic_disable(struct enic *enic) vnic_dev_disable(enic->vdev); enic_clsf_destroy(enic); + enic_fm_destroy(enic); if (!enic_is_sriov_vf(enic)) vnic_dev_del_addr(enic->vdev, enic->mac_addr); diff --git a/drivers/net/enic/enic_res.c b/drivers/net/enic/enic_res.c index 742999cd0..f403a0b66 100644 --- a/drivers/net/enic/enic_res.c +++ b/drivers/net/enic/enic_res.c @@ -85,17 +85,20 @@ int enic_get_vnic_config(struct enic *enic) vnic_dev_capable_udp_rss_weak(enic->vdev, &enic->nic_cfg_chk, &enic->udp_rss_weak); - dev_info(enic, "Flow api filter mode: %s Actions: %s%s%s\n", + dev_info(enic, "Flow api filter mode: %s Actions: %s%s%s%s\n", + ((enic->flow_filter_mode == FILTER_FLOWMAN) ? "FLOWMAN" : ((enic->flow_filter_mode == FILTER_DPDK_1) ? "DPDK" : ((enic->flow_filter_mode == FILTER_USNIC_IP) ? "USNIC" : ((enic->flow_filter_mode == FILTER_IPV4_5TUPLE) ? "5TUPLE" : - "NONE"))), + "NONE")))), ((enic->filter_actions & FILTER_ACTION_RQ_STEERING_FLAG) ? "steer " : ""), ((enic->filter_actions & FILTER_ACTION_FILTER_ID_FLAG) ? "tag " : ""), ((enic->filter_actions & FILTER_ACTION_DROP_FLAG) ? - "drop " : "")); + "drop " : ""), + ((enic->filter_actions & FILTER_ACTION_COUNTER_FLAG) ? + "count " : "")); c->wq_desc_count = min_t(u32, ENIC_MAX_WQ_DESCS, diff --git a/drivers/net/enic/meson.build b/drivers/net/enic/meson.build index 6716cdaf3..1bd7cc7e1 100644 --- a/drivers/net/enic/meson.build +++ b/drivers/net/enic/meson.build @@ -10,6 +10,7 @@ sources = files( 'enic_clsf.c', 'enic_ethdev.c', 'enic_flow.c', + 'enic_fm_flow.c', 'enic_main.c', 'enic_res.c', 'enic_rxtx.c',