From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mga11.intel.com (mga11.intel.com [192.55.52.93]) by dpdk.org (Postfix) with ESMTP id AAD6E6A87 for ; Tue, 30 Sep 2014 08:13:12 +0200 (CEST) Received: from fmsmga002.fm.intel.com ([10.253.24.26]) by fmsmga102.fm.intel.com with ESMTP; 29 Sep 2014 23:19:50 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.04,625,1406617200"; d="scan'208";a="607384523" Received: from irvmail001.ir.intel.com ([163.33.26.43]) by fmsmga002.fm.intel.com with ESMTP; 29 Sep 2014 23:19:46 -0700 Received: from sivswdev01.ir.intel.com (sivswdev01.ir.intel.com [10.237.217.45]) by irvmail001.ir.intel.com (8.14.3/8.13.6/MailSET/Hub) with ESMTP id s8U6Jj02027342; Tue, 30 Sep 2014 07:19:45 +0100 Received: from sivswdev01.ir.intel.com (localhost [127.0.0.1]) by sivswdev01.ir.intel.com with ESMTP id s8U6Jj1x029381; Tue, 30 Sep 2014 07:19:45 +0100 Received: (from pwodkowx@localhost) by sivswdev01.ir.intel.com with id s8U6JjJE029322; Tue, 30 Sep 2014 07:19:45 +0100 From: Pawel Wodkowski To: dev@dpdk.org Date: Tue, 30 Sep 2014 07:19:33 +0100 Message-Id: <1412057973-29160-1-git-send-email-pawelx.wodkowski@intel.com> X-Mailer: git-send-email 1.7.4.1 In-Reply-To: <1411996978-3123-1-git-send-email-pawelx.wodkowski@intel.com> References: <1411996978-3123-1-git-send-email-pawelx.wodkowski@intel.com> Subject: [dpdk-dev] [PATCH v2] bond: Add mode 4 support. X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: patches and discussions about DPDK List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Tue, 30 Sep 2014 06:13:14 -0000 This patch adds support mode 4 of link bonding. It depend on Delcan Doherty patches v3 and rte alarms patch v2 or above. New version handles race issues with setting/cancelin callbacks, fixes promiscus mode setting in mode 4 and some other minor errors in mode 4 implementation. Signed-off-by: Pawel Wodkowski --- lib/librte_ether/rte_ether.h | 1 + lib/librte_pmd_bond/Makefile | 1 + lib/librte_pmd_bond/rte_eth_bond.h | 4 + lib/librte_pmd_bond/rte_eth_bond_8023ad.c | 1070 ++++++++++++++++++++++++++++ lib/librte_pmd_bond/rte_eth_bond_8023ad.h | 405 +++++++++++ lib/librte_pmd_bond/rte_eth_bond_api.c | 82 ++- lib/librte_pmd_bond/rte_eth_bond_args.c | 1 + lib/librte_pmd_bond/rte_eth_bond_pmd.c | 261 ++++++- lib/librte_pmd_bond/rte_eth_bond_private.h | 42 +- 9 files changed, 1821 insertions(+), 46 deletions(-) create mode 100644 lib/librte_pmd_bond/rte_eth_bond_8023ad.c create mode 100644 lib/librte_pmd_bond/rte_eth_bond_8023ad.h diff --git a/lib/librte_ether/rte_ether.h b/lib/librte_ether/rte_ether.h index 2e08f23..1a3711b 100644 --- a/lib/librte_ether/rte_ether.h +++ b/lib/librte_ether/rte_ether.h @@ -293,6 +293,7 @@ struct vlan_hdr { #define ETHER_TYPE_RARP 0x8035 /**< Reverse Arp Protocol. */ #define ETHER_TYPE_VLAN 0x8100 /**< IEEE 802.1Q VLAN tagging. */ #define ETHER_TYPE_1588 0x88F7 /**< IEEE 802.1AS 1588 Precise Time Protocol. */ +#define ETHER_TYPE_SLOW 0x8809 /**< Slow protocols (LACP and Marker). */ #ifdef __cplusplus } diff --git a/lib/librte_pmd_bond/Makefile b/lib/librte_pmd_bond/Makefile index 953d75e..c2312c2 100644 --- a/lib/librte_pmd_bond/Makefile +++ b/lib/librte_pmd_bond/Makefile @@ -44,6 +44,7 @@ CFLAGS += $(WERROR_FLAGS) # SRCS-$(CONFIG_RTE_LIBRTE_PMD_BOND) += rte_eth_bond_api.c SRCS-$(CONFIG_RTE_LIBRTE_PMD_BOND) += rte_eth_bond_pmd.c +SRCS-$(CONFIG_RTE_LIBRTE_PMD_BOND) += rte_eth_bond_8023ad.c SRCS-$(CONFIG_RTE_LIBRTE_PMD_BOND) += rte_eth_bond_args.c # diff --git a/lib/librte_pmd_bond/rte_eth_bond.h b/lib/librte_pmd_bond/rte_eth_bond.h index 6811c7b..b0223c2 100644 --- a/lib/librte_pmd_bond/rte_eth_bond.h +++ b/lib/librte_pmd_bond/rte_eth_bond.h @@ -75,6 +75,10 @@ extern "C" { /**< Broadcast (Mode 3). * In this mode all transmitted packets will be transmitted on all available * active slaves of the bonded. */ +#define BONDING_MODE_8023AD (4) +/**< 802.3AD (Mode 4). + * In this mode transmission and reception of packets is managed by LACP + * protocol specified in 802.3AD documentation. */ /* Balance Mode Transmit Policies */ #define BALANCE_XMIT_POLICY_LAYER2 (0) diff --git a/lib/librte_pmd_bond/rte_eth_bond_8023ad.c b/lib/librte_pmd_bond/rte_eth_bond_8023ad.c new file mode 100644 index 0000000..de416c6 --- /dev/null +++ b/lib/librte_pmd_bond/rte_eth_bond_8023ad.c @@ -0,0 +1,1070 @@ +/*- + * BSD LICENSE + * + * Copyright(c) 2010-2014 Intel Corporation. All rights reserved. + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * * Neither the name of Intel Corporation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include +#include + +#include +#include +#include + +#include "rte_eth_bond_private.h" +#include "rte_eth_bond_8023ad.h" + +#include + +#ifdef RTE_LIBRTE_BOND_DEBUG_8023AD +#define MODE4_DEBUG(fmt, ...) RTE_LOG(DEBUG, PMD, "%6u [Port %u: %s] " fmt, \ + bond_dbg_get_time_diff_ms(), internals->active_slaves[port_num], \ + __FUNCTION__, ##__VA_ARGS__) + +static unsigned +bond_dbg_get_time_diff_ms(void) +{ + static uint64_t start_time = 0; + uint64_t now; + + now = rte_rdtsc(); + if (start_time == 0) + start_time = now; + + return ((now - start_time) * 1000) / rte_get_tsc_hz(); +} + +static void +bond_print_lacp(struct lacpdu *l) +{ + char a_address[18]; + char p_address[18]; + char a_state[256] = { 0 }; + char p_state[256] = { 0 }; + + static const char *state_labels[] = { + "ACT", "TIMEOUT", "AGG", "SYNC", "COL", "DIST", "DEF", "EXP" + }; + + int a_len = 0; + int p_len = 0; + uint8_t i; + uint8_t *addr; + + addr = l->actor.port_params.system.addr_bytes; + snprintf(a_address, sizeof(a_address), "%02X:%02X:%02X:%02X:%02X:%02X", + addr[0], addr[1], addr[2], addr[3], addr[4], addr[5]); + + addr = l->partner.port_params.system.addr_bytes; + snprintf(p_address, sizeof(p_address), "%02X:%02X:%02X:%02X:%02X:%02X", + addr[0], addr[1], addr[2], addr[3], addr[4], addr[5]); + + for (i = 0; i < 8; i++) { + if ((l->actor.state >> i) & 1) { + a_len += snprintf(a_state + a_len, sizeof(a_state) - a_len, "%s ", + state_labels[i]); + } + + if ((l->partner.state >> i) & 1) { + p_len += snprintf(p_state + p_len, sizeof(p_state) - p_len, "%s ", + state_labels[i]); + } + } + + if (a_len && a_state[a_len-1] == ' ') + a_state[a_len-1] = '\0'; + + if (p_len && p_state[p_len-1] == ' ') + p_state[p_len-1] = '\0'; + + RTE_LOG(DEBUG, PMD, "LACP: {\n"\ + " subtype= %02X\n"\ + " ver_num=%02X\n"\ + " actor={ tlv=%02X, len=%02X\n"\ + " pri=%04X, system=%s, key=%04X, p_pri=%04X p_num=%04X\n"\ + " state={ %s }\n"\ + " }\n"\ + " partner={ tlv=%02X, len=%02X\n"\ + " pri=%04X, system=%s, key=%04X, p_pri=%04X p_num=%04X\n"\ + " state={ %s }\n"\ + " }\n"\ + " collector={info=%02X, length=%02X, max_delay=%04X\n, " \ + "type_term=%02X, terminator_length = %02X}\n",\ + l->subtype,\ + l->version_number,\ + l->actor.tlv_type_info,\ + l->actor.info_length,\ + l->actor.port_params.system_priority,\ + a_address,\ + l->actor.port_params.key,\ + l->actor.port_params.port_priority,\ + l->actor.port_params.port_number,\ + a_state,\ + l->partner.tlv_type_info,\ + l->partner.info_length,\ + l->partner.port_params.system_priority,\ + p_address,\ + l->partner.port_params.key,\ + l->partner.port_params.port_priority,\ + l->partner.port_params.port_number,\ + p_state,\ + l->tlv_type_collector_info,\ + l->collector_info_length,\ + l->collector_max_delay,\ + l->tlv_type_terminator,\ + l->terminator_length); + +} +#define BOND_PRINT_LACP(lacpdu) bond_print_lacp(lacpdu) + +#else +#define BOND_PRINT_LACP(lacpdu) do { } while (0) +#define MODE4_DEBUG(fmt, ...) do { } while (0) +#endif + +static const struct ether_addr lacp_mac_addr = { + .addr_bytes = { 0x01, 0x80, 0xC2, 0x00, 0x00, 0x02 } +}; + +static void +timer_cancel(uint64_t *timer) +{ + *timer = 0; +} + +static void +timer_set(uint64_t *timer, uint64_t timeout_ms) +{ + *timer = rte_rdtsc() + timeout_ms * rte_get_tsc_hz() / 1000; +} + +/* Forces given timer to be in expired state. */ +static void +timer_force_expired(uint64_t *timer) +{ + *timer = rte_rdtsc(); +} + +static bool +timer_is_stopped(uint64_t *timer) +{ + return *timer == 0; +} + +static bool +timer_is_expired(uint64_t *timer) +{ + return *timer <= rte_rdtsc(); +} + +/* Timer is in running state if it is not stopped nor expired */ +static bool +timer_is_running(uint64_t *timer) +{ + return !timer_is_stopped(timer) && !timer_is_expired(timer); +} + +static void +record_default(struct port *port) +{ + /* Record default parametes for partner. Partner admin parameters + * are not implemented so set them to arbitrary default (last known) and + * mark actor that parner is in defaulted state. */ + port->partner_state = STATE_LACP_ACTIVE; + ACTOR_STATE_SET(port, DEFAULTED); +} + +/** Function handles rx state machine. + * + * This function implements Receive State Machine from point 5.4.12 in + * 802.1AX documentation. It should be called periodically. + * + * @param lacpdu LACPDU received. + * @param port Port on which LACPDU was received. + */ +static void +rx_machine(struct bond_dev_private *internals, uint8_t port_num, + struct lacpdu *lacp) +{ + struct port *port = &internals->mode4.port_list[port_num]; + + if (SM_FLAG(port, BEGIN)) { + /* Initialize stuff */ + MODE4_DEBUG("-> INITIALIZE\n"); + SM_FLAG_CLR(port, MOVED); + port->selected = UNSELECTED; + + record_default(port); + + ACTOR_STATE_CLR(port, EXPIRED); + timer_cancel(&port->current_while_timer); + + /* DISABLED: On initialization partner is out of sync */ + PARTNER_STATE_CLR(port, SYNCHRONIZATION); + + /* LACP DISABLED stuff if LACP not enabled on this port */ + if (!SM_FLAG(port, LACP_ENABLED)) + PARTNER_STATE_CLR(port, AGGREGATION); + } + + if (!SM_FLAG(port, LACP_ENABLED)) { + /* Update parameters only if state changed */ + if (!timer_is_stopped(&port->current_while_timer)) { + port->selected = UNSELECTED; + record_default(port); + PARTNER_STATE_CLR(port, AGGREGATION); + ACTOR_STATE_CLR(port, EXPIRED); + timer_cancel(&port->current_while_timer); + } + return; + } + + if (lacp) { + MODE4_DEBUG("LACP -> CURRENT\n"); + BOND_PRINT_LACP(lacp); + /* Update selected flag. If partner parameters are defaulted assume they + * are match. If not defaulted compare LACP actor with ports parner + * params. */ + if (!(port->actor_state & STATE_DEFAULTED) && + (((port->partner_state ^ lacp->actor.state) & STATE_AGGREGATION) || + memcmp(&port->partner, &lacp->actor.port_params, + sizeof(port->partner)) != 0)) { + MODE4_DEBUG("selected <- UNSELECTED\n"); + port->selected = UNSELECTED; + } + + /* Record this PDU actor params as partner params */ + memcpy(&port->partner, &lacp->actor.port_params, + sizeof(struct port_params)); + port->partner_state = lacp->actor.state; + + /* Partner parameters are not defaulted any more */ + ACTOR_STATE_CLR(port, DEFAULTED); + + /* Update NTT if partners information are outdated */ + uint8_t state_mask = STATE_LACP_ACTIVE | STATE_LACP_SHORT_TIMEOUT | + STATE_SYNCHRONIZATION | STATE_AGGREGATION; + + if (((port->actor_state ^ lacp->partner.state) & state_mask) || + memcmp(&port->actor, &lacp->partner.port_params, + sizeof(struct port_params)) != 0) { + port->sm_flags |= SM_FLAGS_NTT; + } + + /* If LACP partner params match this port actor params */ + if (memcmp(&port->actor, &lacp->partner.port_params, + sizeof(port->actor)) == 0 && + (port->partner_state & STATE_AGGREGATION) == (port->actor_state + & STATE_AGGREGATION)) + PARTNER_STATE_SET(port, SYNCHRONIZATION); + else if (!(port->partner_state & STATE_AGGREGATION) && + (port->actor_state & STATE_AGGREGATION)) + PARTNER_STATE_SET(port, SYNCHRONIZATION); + else + PARTNER_STATE_CLR(port, SYNCHRONIZATION); + + if (ACTOR_STATE(port, LACP_SHORT_TIMEOUT)) + timer_set(&port->current_while_timer, BOND_8023AD_SHORT_TIMEOUT_MS); + else + timer_set(&port->current_while_timer, BOND_8023AD_LONG_TIMEOUT_MS); + + ACTOR_STATE_CLR(port, EXPIRED); + return; /* No state change */ + } + + /* If CURRENT state timer is not running (stopped or expired) + * transit to EXPIRED state from DISABLED or CURRENT */ + if (!timer_is_running(&port->current_while_timer)) { + ACTOR_STATE_SET(port, EXPIRED); + PARTNER_STATE_CLR(port, SYNCHRONIZATION); + PARTNER_STATE_SET(port, LACP_SHORT_TIMEOUT); + timer_set(&port->current_while_timer, BOND_8023AD_SHORT_TIMEOUT_MS); + } +} + +/** + * Function handles periodic tx state machine. + * + * Function implements Periodic Transmission state machine from point 5.4.13 + * in 802.1AX documentation. It should be called periodically. + * + * @param port Port to handle state machine. + */ +static void +periodic_machine(struct bond_dev_private *internals, uint8_t port_num) +{ + struct port *port = &internals->mode4.port_list[port_num]; + /* Calculate if either site is LACP enabled */ + uint32_t timeout; + uint8_t active = ACTOR_STATE(port, LACP_ACTIVE) || + PARTNER_STATE(port, LACP_ACTIVE); + + uint8_t is_partner_fast, was_partner_fast; + /* No periodic is on BEGIN, LACP DISABLE or when both sides are pasive */ + if (SM_FLAG(port, BEGIN) || !SM_FLAG(port, LACP_ENABLED) || + active == 0) { + timer_cancel(&port->periodic_timer); + timer_force_expired(&port->tx_machine_timer); + SM_FLAG_CLR(port, PARTNER_SHORT_TIMEOUT); + + MODE4_DEBUG("-> NO_PERIODIC ( %s%s%s)\n", + SM_FLAG(port, BEGIN) ? "begind " : "", + SM_FLAG(port, LACP_ENABLED) ? "" : "LACP disabled ", + active ? "LACP active " : "LACP pasive "); + return; + } + + is_partner_fast = PARTNER_STATE(port, LACP_SHORT_TIMEOUT); + was_partner_fast = SM_FLAG(port, PARTNER_SHORT_TIMEOUT); + + /* If periodic timer is not started, transit from NO PERIODIC to FAST/SLOW. + * Other case: check if timer expire or partners settings changed. */ + if (!timer_is_stopped(&port->periodic_timer)) { + if (timer_is_expired(&port->periodic_timer)) { + SM_FLAG_SET(port, NTT); + } else if (is_partner_fast != was_partner_fast) { + /* Partners timeout was slow and now it is fast -> send LACP. + * In other case (was fast and now it is slow) just switch + * timeout to slow without forcing send of LACP (because standard + * say so)*/ + if (!is_partner_fast) + SM_FLAG_SET(port, NTT); + } else + return; /* Nothing changed */ + } + + /* Handle state transition to FAST/SLOW LACP timeout */ + if (is_partner_fast) { + timeout = BOND_8023AD_FAST_PERIODIC_MS; + SM_FLAG_SET(port, PARTNER_SHORT_TIMEOUT); + } else { + timeout = BOND_8023AD_SLOW_PERIODIC_MS; + SM_FLAG_CLR(port, PARTNER_SHORT_TIMEOUT); + } + + timer_set(&port->periodic_timer, timeout); +} + +/** + * Function handles mux state machine. + * + * Function implements Mux Machine from point 5.4.15 in 802.1AX documentation. + * It should be called periodically. + * + * @param port Port to handle state machine. + */ +static int +mux_machine(struct bond_dev_private *internals, uint8_t port_num) +{ + bool ntt = false; + struct port *port = &internals->mode4.port_list[port_num]; + + /* Save current state for later use */ + const uint8_t state_mask = STATE_SYNCHRONIZATION | STATE_DISTRIBUTING | + STATE_COLLECTING; + + /* Enter DETACHED state on BEGIN condition or from any other state if + * port was unselected */ + if (SM_FLAG(port, BEGIN) || + port->selected == UNSELECTED || (port->selected == STANDBY && + (port->actor_state & state_mask) != 0)) { + /* detach mux from aggregator not used */ + port->actor_state &= ~state_mask; + /* Set ntt to true if BEGIN condition or transition from any other state + * which is indicated that wait_while_timer was started */ + if (SM_FLAG(port, BEGIN) || + !timer_is_stopped(&port->wait_while_timer)) { + SM_FLAG_SET(port, NTT); + MODE4_DEBUG("-> DETACHED\n"); + } + timer_cancel(&port->wait_while_timer); + } + + if (timer_is_stopped(&port->wait_while_timer)) { + if (port->selected == SELECTED || port->selected == STANDBY) { + timer_set(&port->wait_while_timer, + BOND_8023AD_AGGREGATE_WAIT_TIMEOUT_MS); + + MODE4_DEBUG("DETACHED -> WAITING\n"); + } + /* Waiting state entered */ + return 0; + } + + /* Transit next state if port is ready */ + if (!timer_is_expired(&port->wait_while_timer)) + return 0; + + if ((ACTOR_STATE(port, DISTRIBUTING) || ACTOR_STATE(port, COLLECTING)) && + !PARTNER_STATE(port, SYNCHRONIZATION)) { + /* If in COLLECTING or DISTRIBUTING state and partner becomes out of + * sync transit to ATACHED state. */ + ACTOR_STATE_CLR(port, DISTRIBUTING); + ACTOR_STATE_CLR(port, COLLECTING); + /* Clear actor sync to activate transit ATACHED in condition bellow */ + ACTOR_STATE_CLR(port, SYNCHRONIZATION); + MODE4_DEBUG("Out of sync -> ATTACHED\n"); + } else if (!ACTOR_STATE(port, SYNCHRONIZATION)) { + /* attach mux to aggregator */ + RTE_VERIFY((port->actor_state & (STATE_COLLECTING | + STATE_DISTRIBUTING)) == 0); + ACTOR_STATE_SET(port, SYNCHRONIZATION); + ntt = true; + MODE4_DEBUG("ATTACHED Entered\n"); + } else if (!ACTOR_STATE(port, COLLECTING)) { + /* Start collecting if in sync */ + if (PARTNER_STATE(port, SYNCHRONIZATION)) { + MODE4_DEBUG("ATTACHED -> COLLECTING\n"); + ACTOR_STATE_SET(port, COLLECTING); + } + } else if (ACTOR_STATE(port, COLLECTING)) { + /* Check if partner is in COLLECTING state. If so this port can + * distribute frames to it */ + if (!ACTOR_STATE(port, DISTRIBUTING)) { + if (PARTNER_STATE(port, COLLECTING)) { + /* Enable DISTRIBUTING if partner is collecting */ + ACTOR_STATE_SET(port, DISTRIBUTING); + ntt = true; + MODE4_DEBUG("COLLECTING -> DISTRIBUTING\n"); + } + } else { + if (!PARTNER_STATE(port, COLLECTING)) { + /* Disable DISTRIBUTING (enter COLLECTING state) if partner + * is not collecting */ + ACTOR_STATE_CLR(port, DISTRIBUTING); + ntt = true; + MODE4_DEBUG("DISTRIBUTING -> COLLECTING\n"); + } + } + } + + if (ntt != false) + SM_FLAG_SET(port, NTT); + + return ntt; +} + +/** + * Function handles transmit state machine. + * + * Function implements Transmit Machine from point 5.4.16 in 802.1AX + * documentation. + * + * @param port + */ +static void +tx_machine(struct rte_eth_dev *bond_dev, uint8_t port_num) +{ + struct bond_dev_private *internals = bond_dev->data->dev_private; + struct port *port = &internals->mode4.port_list[port_num]; + struct mode8023ad_data *data = &internals->mode4; + + struct slow_protocol_msg *msg = NULL; + struct lacpdu_header *hdr; + struct lacpdu *lacpdu; + + /* If periodic timer is not running periodic machine is in NO PERIODIC and + * acording to 802.3ax standard tx machine should not transmit any frames + * and set ntt to false. */ + if (timer_is_stopped(&port->periodic_timer)) + SM_FLAG_CLR(port, NTT); + + if (!SM_FLAG(port, NTT) || !timer_is_expired(&port->tx_machine_timer)) + return; + + /* If all conditions are met construct packet to send */ + if (rte_ring_dequeue(data->free_ring, (void **)&msg) == -ENOBUFS) { + MODE4_DEBUG("tx_machine: no free_lacpdu_ring\n"); + return; + } + + msg->pkt = rte_pktmbuf_alloc(data->mbuf_pool); + if (msg->pkt == NULL) { + rte_ring_enqueue(data->free_ring, msg); + RTE_LOG(ERR, PMD, "Failed to allocate LACP packet from pool\n"); + return; + } + + msg->port_id = internals->active_slaves[port_num]; + hdr = rte_pktmbuf_mtod(msg->pkt, struct lacpdu_header *); + + msg->pkt->data_len = sizeof(*hdr); + msg->pkt->pkt_len = sizeof(*hdr); + /* Source and destination MAC */ + ether_addr_copy(&lacp_mac_addr, &hdr->eth_hdr.d_addr); + ether_addr_copy(&port->actor.system, &hdr->eth_hdr.s_addr); + hdr->eth_hdr.ether_type = rte_cpu_to_be_16(ETHER_TYPE_SLOW); + + lacpdu = &hdr->lacpdu; + memset(lacpdu, 0, sizeof(*lacpdu)); + + /* Initialize LACP part */ + lacpdu->subtype = SUBTYPE_LACP; + lacpdu->version_number = 1; + + /* ACTOR */ + lacpdu->actor.tlv_type_info = TLV_TYPE_ACTOR_INFORMATION; + lacpdu->actor.info_length = sizeof(struct lacpdu_actor_partner_params); + memcpy(&hdr->lacpdu.actor.port_params, &port->actor, + sizeof(port->actor)); + lacpdu->actor.state = port->actor_state; + + /* PARTNER */ + lacpdu->partner.tlv_type_info = TLV_TYPE_PARTNER_INFORMATION; + lacpdu->partner.info_length = sizeof(struct lacpdu_actor_partner_params); + memcpy(&lacpdu->partner.port_params, &port->partner, + sizeof(struct port_params)); + lacpdu->partner.state = port->partner_state; + + /* Other fields */ + lacpdu->tlv_type_collector_info = TLV_TYPE_COLLECTOR_INFORMATION; + lacpdu->collector_info_length = 0x10; + lacpdu->collector_max_delay = 0; + + lacpdu->tlv_type_terminator = TLV_TYPE_TERMINATOR_INFORMATION; + lacpdu->terminator_length = 0; + + if (rte_ring_enqueue(data->tx_ring, msg) == -ENOBUFS) { + /* If TX ring full, drop packet and free message. Retransmission + * will happen in next function call. */ + rte_pktmbuf_free(msg->pkt); + rte_ring_enqueue(data->free_ring, msg); + + RTE_LOG(ERR, PMD, "Failed to enqueue LACP packet into tx ring.\n" + "Receive and transmit functions must be invoked on bonded interface" + " at least 10 times per second or LACP will not work correctly\n"); + return; + } + + MODE4_DEBUG("sending LACP frame\n"); + BOND_PRINT_LACP(lacpdu); + + SM_FLAG_CLR(port, NTT); + /* Add 10% random backoff time to better distribute slow packets + * between tx bursts. */ + timer_set(&port->tx_machine_timer, BOND_8023AD_TX_PERIOD_MS + + rand() % ((BOND_8023AD_TX_PERIOD_MS * 10) / 100)); +} + +/** + * Function assigns port to aggregator. + * + * @param bond_dev_private Pointer to bond_dev_private structure. + * @param port_pos Port to assign. + */ +static void +selection_logic(struct bond_dev_private *internals, uint8_t port_num) +{ + struct mode8023ad_data *data = &internals->mode4; + struct port *agg, *port, *port_list; + uint8_t ports_count; + uint8_t i; + + ports_count = internals->slave_count; + port_list = data->port_list; + port = &port_list[port_num]; + + /* Skip port if it is selected */ + if (port->selected == SELECTED) + return; + + /* Search for aggregator suitable for this port */ + for (i = 0; i < ports_count; ++i) { + agg = &port_list[i]; + /* Skip ports that are not aggreagators */ + if (agg->agregator_idx != i && i == port_num) + continue; + + /* Actors system ID is not checked since all slave device have the same + * ID (MAC address). */ + if ((agg->actor.key == port->actor.key && + agg->partner.system_priority == port->partner.system_priority && + is_same_ether_addr(&agg->partner.system, &port->partner.system) == 1 + && (agg->partner.key == port->partner.key)) && + is_zero_ether_addr(&port->partner.system) != 1 && + (agg->actor.key & + rte_cpu_to_be_16(BOND_LINK_FULL_DUPLEX_KEY)) != 0) { + + port->agregator_idx = i; + break; + } + } + + /* By default, port uses it self as agregator */ + if (i == ports_count) + port->agregator_idx = port_num; + + port->selected = SELECTED; + + MODE4_DEBUG("-> SELECTED: ID=%3u pos=%3u\n" + "\t%s ID=%3u pos=%3u\n", + internals->active_slaves[port_num], port_num, + port->agregator_idx == port_num ? + "agregator not found, using default" : "agregator found", + port->agregator_idx, + internals->active_slaves[port->agregator_idx]); +} + +/** + * Helper function which updates current port + */ +static void +update_mux_slaves(struct bond_dev_private *internals) +{ + struct mode8023ad_data *data = &internals->mode4; + struct port *port; + uint8_t current[RTE_MAX_ETHPORTS]; + uint8_t count = 0; + uint8_t i; + + for (i = 0; i < internals->slave_count; i++) { + port = &data->port_list[i]; + if (ACTOR_STATE(port, DISTRIBUTING)) + current[count++] = i; + } + + memcpy(data->distibuting_slaves_offsets, current, + sizeof(current[0]) * count); + data->distibuting_slaves_count = count; +} + +/* Function maps DPDK speed to bonding speed stored in key field */ +static uint16_t +link_speed_key(uint16_t speed) { + uint16_t key_speed; + + switch (speed) { + case ETH_LINK_SPEED_AUTONEG: + key_speed = 0x00; + break; + case ETH_LINK_SPEED_10: + key_speed = BOND_LINK_SPEED_KEY_10M; + break; + case ETH_LINK_SPEED_100: + key_speed = BOND_LINK_SPEED_KEY_100M; + break; + case ETH_LINK_SPEED_1000: + key_speed = BOND_LINK_SPEED_KEY_1000M; + break; + case ETH_LINK_SPEED_10G: + key_speed = BOND_LINK_SPEED_KEY_10G; + break; + case ETH_LINK_SPEED_20G: + key_speed = BOND_LINK_SPEED_KEY_20G; + break; + case ETH_LINK_SPEED_40G: + key_speed = BOND_LINK_SPEED_KEY_40G; + break; + default: + /* Unknown speed*/ + key_speed = 0xFFFF; + } + + return key_speed; +} + +static void +bond_mode_8023ad_periodic_cb(void *arg) +{ + struct rte_eth_dev *bond_dev = arg; + struct bond_dev_private *internals = bond_dev->data->dev_private; + struct mode8023ad_data *data = &internals->mode4; + + struct port *port; + struct slow_protocol_frame *slow_hdr; + struct rte_eth_link link_info; + struct ether_addr slave_addr; + + struct slow_protocol_msg *msgs[BOND_MODE_8023AX_RX_RING_SIZE]; + uint16_t port_num, j, nb_msgs; + /* if not 0 collecting/distibuting array need update */ + uint16_t slaves_changed = 0; + bool machines_invoked; + + /* Update link status on each port */ + for (port_num = 0; port_num < internals->active_slave_count; port_num++) { + uint16_t key; + + rte_eth_link_get(internals->active_slaves[port_num], &link_info); + rte_eth_macaddr_get(internals->active_slaves[port_num], &slave_addr); + + if (link_info.link_status != 0) { + key = link_speed_key(link_info.link_speed) << 1; + if (link_info.link_duplex == ETH_LINK_FULL_DUPLEX) + key |= BOND_LINK_FULL_DUPLEX_KEY; + } else + key = 0; + + port = &data->port_list[port_num]; + key = rte_cpu_to_be_16(key); + + if (key != port->actor.key) { + port->actor.key = key; + SM_FLAG_SET(port, NTT); + } + + if (!is_same_ether_addr(&port->actor.system, &slave_addr)) { + SM_FLAG_SET(port, NTT); + ether_addr_copy(&slave_addr, &port->actor.system); + } + } + + nb_msgs = (uint16_t)rte_ring_dequeue_burst(data->rx_ring, (void **) msgs, + BOND_MODE_8023AX_RX_RING_SIZE); + + for (port_num = 0; port_num < internals->active_slave_count; port_num++) { + port = &data->port_list[port_num]; + if ((port->actor.key & + rte_cpu_to_be_16(BOND_LINK_FULL_DUPLEX_KEY)) == 0) { + + SM_FLAG_SET(port, BEGIN); + + /* LACP is disabled on half duples or link is down */ + if (SM_FLAG(port, LACP_ENABLED)) { + /* If port was enabled set it to BEGIN state */ + SM_FLAG_CLR(port, LACP_ENABLED); + ACTOR_STATE_CLR(port, DISTRIBUTING); + ACTOR_STATE_CLR(port, COLLECTING); + slaves_changed++; + } + + MODE4_DEBUG("Port %u is not LACP capable!\n", + internals->active_slaves[port_num]); + /* Skip this port processing */ + continue; + } + + SM_FLAG_SET(port, LACP_ENABLED); + machines_invoked = false; + /* Find LACP packet */ + for (j = 0; j < nb_msgs; j++) { + if (msgs[j] == NULL || msgs[j]->port_id != + internals->active_slaves[port_num]) + continue; + + slow_hdr = rte_pktmbuf_mtod(msgs[j]->pkt, + struct slow_protocol_frame *); + + if (slow_hdr->slow_protocol.subtype == SLOW_SUBTYPE_LACP) { + /* This is LACP frame so pass it to rx_machine */ + struct lacpdu *lacp = (struct lacpdu *)&slow_hdr->slow_protocol; + /* Invoke state machines on every active slave port */ + rx_machine(internals, port_num, lacp); + periodic_machine(internals, port_num); + slaves_changed += mux_machine(internals, port_num); + tx_machine(bond_dev, port_num); + selection_logic(internals, port_num); + + machines_invoked = true; + } else if (slow_hdr->slow_protocol.subtype == SLOW_SUBTYPE_MARKER) { + struct marker *marker; + + marker = (struct marker *) &slow_hdr->slow_protocol; + if (marker->tlv_type_marker == MARKER_TLV_TYPE_MARKER_INFO) { + /* Reuse received packet to send frame to Marker Responder + */ + marker->tlv_type_marker = MARKER_TLV_TYPE_MARKER_RESP; + + /* Update source MAC, destination MAC is multicast so we + * don't update it */ + mac_address_get(bond_dev, &slow_hdr->eth_hdr.s_addr); + + if (rte_ring_enqueue(data->tx_ring, msgs[j]) == -ENOBUFS) { + RTE_LOG(ERR, PMD, + "Failed to enqueue packet into tx ring"); + rte_pktmbuf_free(msgs[j]->pkt); + rte_ring_enqueue(data->free_ring, msgs[j]); + } + + msgs[j] = NULL; + } + } + } + + if (machines_invoked == false) { + rx_machine(internals, port_num, NULL); + periodic_machine(internals, port_num); + slaves_changed += mux_machine(internals, port_num); + tx_machine(bond_dev, port_num); + selection_logic(internals, port_num); + machines_invoked = true; + } + + SM_FLAG_CLR(port, BEGIN); + } + + /* Update mux if something changed */ + if (slaves_changed > 0) { + update_mux_slaves(internals); + MODE4_DEBUG("mux count %u [%2u%s%2u%s%2u%s%2u%s%s]\n", + data->distibuting_slaves_count, + data->distibuting_slaves_offsets[0], + data->distibuting_slaves_count > 0 ? " " : "\b\b", + data->distibuting_slaves_offsets[1], + data->distibuting_slaves_count > 1 ? " " : "\b\b", + data->distibuting_slaves_offsets[2], + data->distibuting_slaves_count > 2 ? " " : "\b\b", + data->distibuting_slaves_offsets[3], + data->distibuting_slaves_count > 3 ? " " : "\b\b", + data->distibuting_slaves_count > 4 ? "..." : ""); + } + + /* Free packets that was not reused */ + for (port_num = 0; port_num < nb_msgs; port_num++) { + if (msgs[port_num] != NULL) { + rte_pktmbuf_free(msgs[port_num]->pkt); + rte_ring_enqueue(data->free_ring, msgs[port_num]); + } + } + + rte_eal_alarm_set(BOND_MODE_8023AX_UPDATE_TIMEOUT_MS * 1000, + bond_mode_8023ad_periodic_cb, arg); +} + +static void +bond_mode_8023ad_activate_slave(struct rte_eth_dev *bond_dev, uint8_t slave_idx) +{ + struct bond_dev_private *internals = bond_dev->data->dev_private; + struct mode8023ad_data *data = &internals->mode4; + + struct port *port = &data->port_list[internals->active_slave_count]; + struct port_params initial = { + .system = { { 0 } }, + .system_priority = rte_cpu_to_be_16(0xFFFF), + .key = rte_cpu_to_be_16(BOND_LINK_FULL_DUPLEX_KEY), + .port_priority = rte_cpu_to_be_16(0x00FF), + .port_number = 0, + }; + + uint8_t slave_id = internals->active_slaves[slave_idx]; + + memcpy(&port->actor, &initial, sizeof(struct port_params)); + port->actor.port_number = slave_id_to_port_number(slave_id); + + memcpy(&port->partner, &initial, sizeof(struct port_params)); + + /* default states */ + port->actor_state = STATE_AGGREGATION | STATE_LACP_ACTIVE | STATE_DEFAULTED; + port->partner_state = STATE_LACP_ACTIVE; + port->sm_flags = SM_FLAGS_BEGIN; + + /* use this port as agregator */ + port->agregator_idx = slave_idx; + + rte_eth_promiscuous_enable(slave_id); +} + +void +bond_mode_8023ad_slave_append(struct rte_eth_dev *bond_dev) +{ + struct bond_dev_private *internals = bond_dev->data->dev_private; + + bond_mode_8023ad_activate_slave(bond_dev, internals->active_slave_count); +} + +int +bond_mode_8023ad_deactivate_slave(struct rte_eth_dev *bond_dev, + uint8_t slave_pos) +{ + struct bond_dev_private *internals = bond_dev->data->dev_private; + struct mode8023ad_data *data = &internals->mode4; + struct port *port; + uint8_t i; + + bond_mode_8023ad_stop(bond_dev); + + /* Exclude slave from transmit policy. If this slave is an aggregator + * make all aggregated slaves unselected to force sellection logic + * to select suitable aggregator for this port */ + for (i = 0; i < internals->active_slave_count; i++) { + port = &data->port_list[slave_pos]; + if (port->agregator_idx == slave_pos) { + port->selected = UNSELECTED; + port->actor_state &= ~(STATE_SYNCHRONIZATION | STATE_DISTRIBUTING | + STATE_COLLECTING); + + /* Use default aggregator */ + port->agregator_idx = i; + } + } + + port = &data->port_list[slave_pos]; + + update_mux_slaves(internals); + + /* Remove slave port config */ + if (slave_pos + 1 < internals->active_slave_count) { + memmove(&data->port_list[slave_pos], + &data->port_list[slave_pos + 1], + sizeof(data->port_list[0]) * (internals->active_slave_count - + slave_pos - 1)); + } + + if (bond_dev->data->dev_started) + return bond_mode_8023ad_start(bond_dev); + + return 0; +} + +int +bond_mode_8023ad_init(struct rte_eth_dev *bond_dev) +{ + struct bond_dev_private *internals = bond_dev->data->dev_private; + struct mode8023ad_data *data = &internals->mode4; + char mem_name[RTE_ETH_NAME_MAX_LEN]; + int socket_id = bond_dev->pci_dev->numa_node; + uint8_t i; + + if (data->mbuf_pool == NULL) { + const uint16_t element_size = sizeof(struct slow_protocol_frame) + + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM; + + snprintf(mem_name, sizeof(mem_name), "%s_POOL", bond_dev->data->name); + data->mbuf_pool = rte_mempool_create(mem_name, + /* FIXME: How big memory pool should be? If driver will not + * free packets quick enough there will be ENOMEM in tx_machine. + * For now give 512 packets per slave. Hope it will be enough. */ + (BOND_MODE_8023AX_TX_RING_SIZE + 1) * 512 * RTE_MAX_ETHPORTS, + element_size, + RTE_MEMPOOL_CACHE_MAX_SIZE >= 32 ? 32 : RTE_MEMPOOL_CACHE_MAX_SIZE, + sizeof(struct rte_pktmbuf_pool_private), rte_pktmbuf_pool_init, + NULL, rte_pktmbuf_init, NULL, socket_id, 0); + + /* Any memory allocation failure in initalization is critical because + * resources can't be free, so reinitialization is impossible. */ + if (data->mbuf_pool == NULL) { + RTE_LOG(ERR, PMD, "%s: Failed to initialize LACP rx ring\n", + bond_dev->data->name); + + rte_panic("Failed to alocate memory pool ('%s')\n" + "for bond device '%s'\n", mem_name, bond_dev->data->name); + } + + /* Setup ring for free messages that can be used in RX/TX burst */ + snprintf(mem_name, sizeof(mem_name), "%s_free", bond_dev->data->name); + + uint16_t free_cnt = BOND_MODE_8023AX_RX_RING_SIZE + + BOND_MODE_8023AX_TX_RING_SIZE; + + data->free_ring = rte_ring_create(mem_name, free_cnt, socket_id, 0); + + if (data->free_ring == NULL) { + rte_panic("%s: Failed to create slow messages free ring\n", + bond_dev->data->name); + } + + for (i = 0; i < free_cnt; i++) { + struct slow_protocol_msg *msg; + + snprintf(mem_name, sizeof(mem_name), "%s_slow_msg_%u", + bond_dev->data->name, i); + + msg = (struct slow_protocol_msg *) rte_malloc_socket(mem_name, + sizeof(struct slow_protocol_msg), 0, socket_id); + + if (msg == NULL) { + rte_panic("%s: Failed to allocate slow message\n", + bond_dev->data->name); + } + + rte_ring_enqueue(data->free_ring, msg); + } + + /* Setup rings for usage in rx/tx bursts and machines state + * call back */ + snprintf(mem_name, sizeof(mem_name), "%s_rx", bond_dev->data->name); + data->rx_ring = rte_ring_create(mem_name, + BOND_MODE_8023AX_RX_RING_SIZE, socket_id, 0); + + if (data->rx_ring == NULL) { + rte_panic("%s: Failed to create slow messages rx ring\n", + bond_dev->data->name); + } + + snprintf(mem_name, sizeof(mem_name), "%s_tx", bond_dev->data->name); + data->tx_ring = rte_ring_create(mem_name, BOND_MODE_8023AX_TX_RING_SIZE, + socket_id, RING_F_SP_ENQ); + + if (data->tx_ring == NULL) { + rte_panic("%s: Failed to create slow messages tx ring\n", + bond_dev->data->name); + } + } + + data->distibuting_slaves_count = 0; + + for (i = 0; i < internals->active_slave_count; i++) + bond_mode_8023ad_activate_slave(bond_dev, i); + + return 0; +} + +int +bond_mode_8023ad_start(struct rte_eth_dev *bond_dev) +{ + return rte_eal_alarm_set(BOND_MODE_8023AX_UPDATE_TIMEOUT_MS * 1000, + &bond_mode_8023ad_periodic_cb, bond_dev); +} + +int +bond_mode_8023ad_stop(struct rte_eth_dev *bond_dev) +{ + if (rte_eal_alarm_cancel(&bond_mode_8023ad_periodic_cb, bond_dev)) + return 0; + + return -ENOENT; +} + +void +bond_mode_8023ad_handle_slow_pkt(struct bond_dev_private *internals, + uint8_t slave_pos, struct rte_mbuf *slot_pkt) +{ + struct mode8023ad_data *data; + struct slow_protocol_msg *msg = NULL; + int retval; + + data = &internals->mode4; + + if (unlikely(rte_ring_dequeue(data->free_ring, (void **)&msg) == + -ENOBUFS)) { + rte_pktmbuf_free(slot_pkt); + return; + } + + msg->pkt = slot_pkt; + msg->port_id = internals->active_slaves[slave_pos]; + + retval = rte_ring_enqueue(data->rx_ring, msg); + if (unlikely(retval == -ENOBUFS)) { + /* If RX fing full free lacpdu message and drop packet */ + rte_ring_enqueue(data->free_ring, msg); + rte_pktmbuf_free(slot_pkt); + } +} diff --git a/lib/librte_pmd_bond/rte_eth_bond_8023ad.h b/lib/librte_pmd_bond/rte_eth_bond_8023ad.h new file mode 100644 index 0000000..df250bb --- /dev/null +++ b/lib/librte_pmd_bond/rte_eth_bond_8023ad.h @@ -0,0 +1,405 @@ +/*- + * BSD LICENSE + * + * Copyright(c) 2010-2014 Intel Corporation. All rights reserved. + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * * Neither the name of Intel Corporation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef RTE_ETH_BOND_8023AD_H_ +#define RTE_ETH_BOND_8023AD_H_ + +#include + +#include +#include +#include + +typedef int bool; + +#define true 1 +#define false 0 + +/** + * Timeouts deffinitions (5.4.4 in 802.1AX documentation). + */ +#define BOND_8023AD_FAST_PERIODIC_MS 1000 +#define BOND_8023AD_SLOW_PERIODIC_MS 30000 +#define BOND_8023AD_SHORT_TIMEOUT_MS 3000 +#define BOND_8023AD_LONG_TIMEOUT_MS 90000 +#define BOND_8023AD_CHURN_DETECTION_TIMEOUT_MS 60000 +#define BOND_8023AD_AGGREGATE_WAIT_TIMEOUT_MS 2000 +#define BOND_8023AD_TX_PERIOD_MS 333 +/** + * Actor/partner states + */ +#define STATE_LACP_ACTIVE 0x01 +#define STATE_LACP_SHORT_TIMEOUT 0x02 +#define STATE_AGGREGATION 0x04 +#define STATE_SYNCHRONIZATION 0x08 +#define STATE_COLLECTING 0x10 +#define STATE_DISTRIBUTING 0x20 +/** Partners parameters are defaulted */ +#define STATE_DEFAULTED 0x40 +#define STATE_EXPIRED 0x80 + +/** + * State machine flags + */ +#define SM_FLAGS_BEGIN 0x0001 +#define SM_FLAGS_LACP_ENABLED 0x0002 +#define SM_FLAGS_ACTOR_CHURN 0x0004 +#define SM_FLAGS_PARTNER_CHURN 0x0008 +#define SM_FLAGS_MOVED 0x0100 +#define SM_FLAGS_PARTNER_SHORT_TIMEOUT 0x0200 +#define SM_FLAGS_NTT 0x0400 + +#define BOND_MODE_8023AX_UPDATE_TIMEOUT_MS 100 +#define BOND_MODE_8023AX_RX_RING_SIZE (2 * RTE_MAX_ETHPORTS) +#define BOND_MODE_8023AX_TX_RING_SIZE (2 * RTE_MAX_ETHPORTS) + +#define BOND_LINK_FULL_DUPLEX_KEY 0x01 +#define BOND_LINK_SPEED_KEY_10M 0x02 +#define BOND_LINK_SPEED_KEY_100M 0x04 +#define BOND_LINK_SPEED_KEY_1000M 0x08 +#define BOND_LINK_SPEED_KEY_10G 0x10 +#define BOND_LINK_SPEED_KEY_20G 0x11 +#define BOND_LINK_SPEED_KEY_40G 0x12 + +#define SUBTYPE_LACP 0x01 + +#define TLV_TYPE_ACTOR_INFORMATION 0x01 +#define TLV_TYPE_PARTNER_INFORMATION 0x02 +#define TLV_TYPE_COLLECTOR_INFORMATION 0x03 +#define TLV_TYPE_TERMINATOR_INFORMATION 0x00 + +#define CHECK_FLAGS(_variable, _flags) ((_variable) & (_flags)) +#define SET_FLAGS(_variable, _flags) ((_variable) |= (_flags)) +#define CLEAR_FLAGS(_variable, _flags) ((_variable) &= ~(_flags)) + +#define SM_FLAG(port, flag) (!!CHECK_FLAGS((port)->sm_flags, SM_FLAGS_ ## flag)) +#define SM_FLAG_SET(port, flag) SET_FLAGS((port)->sm_flags, SM_FLAGS_ ## flag) +#define SM_FLAG_CLR(port, flag) CLEAR_FLAGS((port)->sm_flags, SM_FLAGS_ ## flag) + +#define ACTOR_STATE(port, flag) (!!CHECK_FLAGS((port)->actor_state, STATE_ ## flag)) +#define ACTOR_STATE_SET(port, flag) SET_FLAGS((port)->actor_state, STATE_ ## flag) +#define ACTOR_STATE_CLR(port, flag) CLEAR_FLAGS((port)->actor_state, STATE_ ## flag) + +#define PARTNER_STATE(port, flag) (!!CHECK_FLAGS((port)->partner_state, STATE_ ## flag)) +#define PARTNER_STATE_SET(port, flag) SET_FLAGS((port)->partner_state, STATE_ ## flag) +#define PARTNER_STATE_CLR(port, flag) CLEAR_FLAGS((port)->partner_state, STATE_ ## flag) + +/** Slow protocol LACP frame subtype */ +#define SLOW_SUBTYPE_LACP 0x01 + +/** Slow procotol marker frame subtype */ +#define SLOW_SUBTYPE_MARKER 0x02 + +/** Marker type info request */ +#define MARKER_TLV_TYPE_MARKER_INFO 0x01 + +/** Marker type info response */ +#define MARKER_TLV_TYPE_MARKER_RESP 0x02 + +/** Generic slow protocol structure */ +struct slow_protocol { + uint8_t subtype; + uint8_t reserved_119[119]; +} __attribute__((__packed__)); + +/** Generic slow protocol frame type structure */ +struct slow_protocol_frame { + struct ether_hdr eth_hdr; + struct slow_protocol slow_protocol; +} __attribute__((__packed__)); + +struct port_params { + uint16_t system_priority; + /**< System priority (unused in current implementation) */ + struct ether_addr system; + /**< System ID - Slave MAC address, same as bonding MAC address */ + uint16_t key; + /**< Speed information (implementation dependednt) and duplex. */ + uint16_t port_priority; + /**< Priority of this (unused in current implementation) */ + uint16_t port_number; + /**< Port number. It corresponds to slave port id. */ +} __attribute__((__packed__)); + +struct lacpdu_actor_partner_params { + uint8_t tlv_type_info; + uint8_t info_length; + struct port_params port_params; + uint8_t state; + uint8_t reserved_3[3]; +} __attribute__((__packed__)); + +/** LACPDU structure (5.4.2 in 802.1AX documentation). */ +struct lacpdu { + uint8_t subtype; + uint8_t version_number; + + struct lacpdu_actor_partner_params actor; + struct lacpdu_actor_partner_params partner; + + uint8_t tlv_type_collector_info; + uint8_t collector_info_length; + uint16_t collector_max_delay; + uint8_t reserved_12[12]; + + uint8_t tlv_type_terminator; + uint8_t terminator_length; + uint8_t reserved_50[50]; +} __attribute__((__packed__)); + +/** LACPDU frame: Contains ethernet header and LACPDU. */ +struct lacpdu_header { + struct ether_hdr eth_hdr; + struct lacpdu lacpdu; +} __attribute__((__packed__)); + +struct marker { + uint8_t subtype; + uint8_t version_number; + + uint8_t tlv_type_marker; + uint8_t info_length; + uint16_t requester_port; + struct ether_addr requester_system; + uint32_t requester_transaction_id; + uint8_t reserved_2[2]; + + uint8_t tlv_type_terminator; + uint8_t terminator_length; + uint8_t reserved_90[90]; +} __attribute__((__packed__)); + +struct marker_header { + struct ether_hdr eth_hdr; + struct marker marker; +} __attribute__((__packed__)); + +/** Variables associated with the system (5.4.5 in 802.1AX documentation). */ +struct system { + struct ether_addr actor_system; + /**< The MAC address component of the System Identifier of the System */ + uint16_t actor_system_priority; + /**< The System Priority of the System */ +}; + +enum selection { + UNSELECTED, + STANDBY, + SELECTED +}; + +/** Variables associated with each port (5.4.7 in 802.1AX documentation). */ +struct port { + /** + * The operational values of the Actor's state parameters. Bitmask + * of port states. + */ + uint8_t actor_state; + + /** The operational Actor's port parameters */ + struct port_params actor; + + /** + * The operational value of the Actor's view of the current values of + * the Partner's state parameters. The Actor sets this variable either + * to the value received from the Partner in an LACPDU, or to the value + * of Partner_Admin_Port_State. Bitmask of port states. + */ + uint8_t partner_state; + + /** The operational Partner's port parameters */ + struct port_params partner; + + /* Additional port parameters not listed in documentation */ + /** State machine flags */ + uint16_t sm_flags; + enum selection selected; + + uint64_t current_while_timer; + uint64_t periodic_timer; + uint64_t wait_while_timer; + uint64_t tx_machine_timer; + /* Agregator parameters */ + /** + * Index in mode8023ad_data::port_list[] of Aggregator + * the port is currently attached to. + */ + uint16_t agregator_idx; +}; + + +/** + * Struct used to comunicate with 8023ad logic. + */ +struct slow_protocol_msg { + struct rte_mbuf *pkt; + uint8_t port_id; +}; + +/** Data specific to mode 802.1AX */ +struct mode8023ad_data { + /** Memory pool used to allocated rings */ + struct rte_mempool *mbuf_pool; + + /** Ring containing free slow_protocol_msg objects. Used to avoid + * alocating/freeing memory in RX/TX bursts */ + struct rte_ring *free_ring; + + /** Ring of struct slow_protocol_msg from RX burst function */ + struct rte_ring *rx_ring; + + /** Ring of struct slow_protocol_msg to RX burst function */ + struct rte_ring *tx_ring; + + /** list of all enslaved ports in mode 802.1AX */ + struct port port_list[RTE_MAX_ETHPORTS]; + + /** List of offsets in active slaves array used to tansmit packets. */ + uint8_t distibuting_slaves_offsets[RTE_MAX_ETHPORTS]; + uint8_t distibuting_slaves_count; +}; + +/* Forward declaration */ +struct bond_dev_private; + +/** + * Configures 802.1AX mode and all active slaves on bonded interface. + * + * @param dev Bonded interface + * @return + * 0 on success, negative value otherwise. + */ +int +bond_mode_8023ad_init(struct rte_eth_dev *dev); + +/** + * Deconfigures 802.1AX mode of the bonded interface and slaves. + * + * @param dev Bonded interface + * @return + * 0 on success, negative value otherwise. + */ +int bond_mode_8023ad_disable(struct rte_eth_dev *dev); + +/** + * Starts 802.3AX state machines management logic. + * @param dev Bonded interface + * @return + * 0 if machines was started, 1 if machines was already running, + * negative value otherwise. + */ +int +bond_mode_8023ad_start(struct rte_eth_dev *dev); + +/** + * Stops 802.3AX state machines management logic. + * @param dev Bonded interface + * @return + * 0 if this call stopped state machines, -ENOENT if alarm was not set. + */ +int +bond_mode_8023ad_stop(struct rte_eth_dev *dev); + +/** + * Passes given slow packet to state machines management logic. + * @param internals Bonded device private data. + * @param slave_pos Possition in active slaves array on which this packet was received. + * @param slot_pkt Slow packet + */ +void +bond_mode_8023ad_handle_slow_pkt(struct bond_dev_private *internals, + uint8_t slave_pos, struct rte_mbuf *slot_pkt); + +/** + * Appends and initializes slave active_slaves[slave_num] to use with + * 802.1AX mode. + * + * @pre active_slaves[active_slave_count] must contain valid slave id. + * @post active_slave_count must be incremented. + * + * @param dev Bonded interface. + * + * @return + * 0 on success, negative value otherwise. + */ +void +bond_mode_8023ad_slave_append(struct rte_eth_dev *dev); + +/** + * Denitializes and removes given slave from 802.1AX mode. + * + * @pre active_slaves[slave_num] must contain valid slave id corresponding to + * slave initialized in 802.1AX mode. + * @post active_slaves[slave_num] must be removed. + * + * @param dev Bonded interface. + * @param slave_num Position of slave in active_slaves array + * + * @return + * 0 on success, negative value otherwise. + * + */ +int +bond_mode_8023ad_deactivate_slave(struct rte_eth_dev *dev, uint8_t slave_pos); + +/** + * Converts port_number from network byte order to port id. + * + * @param port_number The 8023ad port number to convert. + * @return corresponding slave id + */ +static inline uint8_t +port_number_to_slave_id(uint16_t port_number) +{ + uint16_t port_id = rte_be_to_cpu_16(port_number); + /* Standard requires that port number must be grater than 0. + * Substract 1 to get corresponding slave id */ + return port_id - 1; +} + +/** + * Converts port id to mode 8023ad port number. + * + * @param slave_id Id of slave to convert. + * @return corresponding Port number in network byte order. + */ +static inline uint16_t +slave_id_to_port_number(uint8_t slave_id) +{ + /* Standard requires that port ID must be grater than 0. + * Add 1 do get corresponding port_number */ + uint16_t port_number = (uint16_t)slave_id + 1; + return rte_cpu_to_be_16(port_number); +} + +#endif /* RTE_ETH_BOND_8023AD_H_ */ diff --git a/lib/librte_pmd_bond/rte_eth_bond_api.c b/lib/librte_pmd_bond/rte_eth_bond_api.c index c690ceb..c547164 100644 --- a/lib/librte_pmd_bond/rte_eth_bond_api.c +++ b/lib/librte_pmd_bond/rte_eth_bond_api.c @@ -31,6 +31,8 @@ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ +#include + #include #include #include @@ -104,6 +106,44 @@ valid_slave_port_id(uint8_t port_id) return 0; } +void +activate_slave(struct rte_eth_dev *eth_dev, uint8_t port_id) +{ + struct bond_dev_private *internals = eth_dev->data->dev_private; + uint8_t active_count = internals->active_slave_count; + + internals->active_slaves[active_count] = port_id; + + if (internals->mode == BONDING_MODE_8023AD) + bond_mode_8023ad_slave_append(eth_dev); + + internals->active_slave_count = active_count + 1; +} + +void +deactivate_slave(struct rte_eth_dev *eth_dev, + uint8_t slave_pos) +{ + struct bond_dev_private *internals = eth_dev->data->dev_private; + uint8_t active_count = internals->active_slave_count; + + if (internals->mode == BONDING_MODE_8023AD) + bond_mode_8023ad_deactivate_slave(eth_dev, slave_pos); + + active_count--; + + /* If slave was not at the end of the list + * shift active slaves up active array list */ + if (slave_pos < active_count) { + memmove(internals->active_slaves + slave_pos, + internals->active_slaves + slave_pos + 1, + (active_count - slave_pos) * + sizeof(internals->active_slaves[0])); + } + + internals->active_slave_count = active_count; +} + uint8_t number_of_sockets(void) { @@ -216,12 +256,8 @@ rte_eth_bond_create(const char *name, uint8_t mode, uint8_t socket_id) eth_dev->dev_ops = &default_dev_ops; eth_dev->pci_dev = pci_dev; - if (bond_ethdev_mode_set(eth_dev, mode)) { - RTE_BOND_LOG(ERR, "Failed to set bonded device %d mode too %d", - eth_dev->data->port_id, mode); - goto err; - } - + internals->port_id = eth_dev->data->port_id; + internals->mode = BONDING_MODE_INVALID; internals->current_primary_port = 0; internals->balance_xmit_policy = BALANCE_XMIT_POLICY_LAYER2; internals->user_defined_mac = 0; @@ -241,6 +277,12 @@ rte_eth_bond_create(const char *name, uint8_t mode, uint8_t socket_id) memset(internals->active_slaves, 0, sizeof(internals->active_slaves)); memset(internals->slaves, 0, sizeof(internals->slaves)); + if (bond_ethdev_mode_set(eth_dev, mode)) { + RTE_BOND_LOG(ERR, "Failed to set bonded device %d mode too %d", + eth_dev->data->port_id, mode); + goto err; + } + return eth_dev->data->port_id; err: @@ -348,14 +390,12 @@ __eth_bond_slave_add_lock_free(uint8_t bonded_port_id, uint8_t slave_port_id) rte_eth_link_get_nowait(slave_port_id, &link_props); if (link_props.link_status == 1) - internals->active_slaves[internals->active_slave_count++] = - slave_port_id; + activate_slave(bonded_eth_dev, slave_port_id); } return 0; } - int rte_eth_bond_slave_add(uint8_t bonded_port_id, uint8_t slave_port_id) { @@ -380,31 +420,26 @@ rte_eth_bond_slave_add(uint8_t bonded_port_id, uint8_t slave_port_id) return retval; } - static int __eth_bond_slave_remove_lock_free(uint8_t bonded_port_id, uint8_t slave_port_id) { + struct rte_eth_dev *bonded_eth_dev; struct bond_dev_private *internals; - int i, slave_idx = -1; + int i, slave_idx; if (valid_slave_port_id(slave_port_id) != 0) return -1; - internals = rte_eth_devices[bonded_port_id].data->dev_private; + bonded_eth_dev = &rte_eth_devices[bonded_port_id]; + internals = bonded_eth_dev->data->dev_private; /* first remove from active slave list */ - for (i = 0; i < internals->active_slave_count; i++) { - if (internals->active_slaves[i] == slave_port_id) - slave_idx = i; + slave_idx = find_slave_by_id(internals->active_slaves, internals->active_slave_count, + slave_port_id); - /* shift active slaves up active array list */ - if (slave_idx >= 0 && i < (internals->active_slave_count - 1)) - internals->active_slaves[i] = internals->active_slaves[i+1]; - } - - if (slave_idx >= 0) - internals->active_slave_count--; + if (slave_idx < internals->active_slave_count) + deactivate_slave(bonded_eth_dev, slave_idx); slave_idx = -1; /* now find in slave list */ @@ -538,6 +573,7 @@ rte_eth_bond_primary_get(uint8_t bonded_port_id) return internals->current_primary_port; } + int rte_eth_bond_slaves_get(uint8_t bonded_port_id, uint8_t slaves[], uint8_t len) { @@ -673,7 +709,6 @@ rte_eth_bond_xmit_policy_get(uint8_t bonded_port_id) return internals->balance_xmit_policy; } - int rte_eth_bond_link_monitoring_set(uint8_t bonded_port_id, uint32_t internal_ms) { @@ -729,7 +764,6 @@ rte_eth_bond_link_down_prop_delay_get(uint8_t bonded_port_id) return internals->link_down_delay_ms; } - int rte_eth_bond_link_up_prop_delay_set(uint8_t bonded_port_id, uint32_t delay_ms) diff --git a/lib/librte_pmd_bond/rte_eth_bond_args.c b/lib/librte_pmd_bond/rte_eth_bond_args.c index bbbc69b..a0be0e6 100644 --- a/lib/librte_pmd_bond/rte_eth_bond_args.c +++ b/lib/librte_pmd_bond/rte_eth_bond_args.c @@ -171,6 +171,7 @@ bond_ethdev_parse_slave_mode_kvarg(const char *key __rte_unused, case BONDING_MODE_ACTIVE_BACKUP: case BONDING_MODE_BALANCE: case BONDING_MODE_BROADCAST: + case BONDING_MODE_8023AD: return 0; default: RTE_BOND_LOG(ERR, "Invalid slave mode value (%s) specified", value); diff --git a/lib/librte_pmd_bond/rte_eth_bond_pmd.c b/lib/librte_pmd_bond/rte_eth_bond_pmd.c index 6d0fb1b..13630d9 100644 --- a/lib/librte_pmd_bond/rte_eth_bond_pmd.c +++ b/lib/librte_pmd_bond/rte_eth_bond_pmd.c @@ -44,6 +44,7 @@ #include "rte_eth_bond.h" #include "rte_eth_bond_private.h" +#include "rte_eth_bond_8023ad.h" static uint16_t bond_ethdev_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts) @@ -168,6 +169,56 @@ bond_ethdev_tx_burst_active_backup(void *queue, bufs, nb_pkts); } +static uint16_t +bond_ethdev_rx_burst_8023ad(void *queue, struct rte_mbuf **bufs, + uint16_t nb_pkts) +{ + /* Cast to structure, containing bonded device's port id and queue id */ + struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue; + struct bond_dev_private *internals = bd_rx_q->dev_private; + struct mode8023ad_data *mode4 = &internals->mode4; + struct ether_addr bond_mac; + + struct ether_hdr *hdr; + struct rte_mbuf *pkts[nb_pkts + 1]; /* one packet more for slow packet */ + + uint16_t num_rx_slave = 0; /* Number of packet received on current slave */ + uint16_t num_rx_total = 0; /* Total number of received packets */ + + uint8_t i, j; + + rte_eth_macaddr_get(internals->port_id, &bond_mac); + + for (i = 0; i < internals->active_slave_count && num_rx_total < nb_pkts; i++) { + /* Read packets from this slave */ + num_rx_slave = rte_eth_rx_burst(internals->active_slaves[i], + bd_rx_q->queue_id, pkts, nb_pkts + 1 - num_rx_total); + + /* Separate slow protocol packets from other packets */ + for (j = 0; j < num_rx_slave; j++) { + hdr = rte_pktmbuf_mtod(pkts[j], struct ether_hdr *); + + uint16_t ether_type = rte_be_to_cpu_16(hdr->ether_type); + if (unlikely(ether_type == ETHER_TYPE_SLOW)) { + bond_mode_8023ad_handle_slow_pkt(internals, i, pkts[j]); + continue; + } + + /* Check if we can receive this packet. Also filter packets if + * bonding interface is not in promiscuous mode (slaves are always + * in promiscuous mode). */ + if (likely(ACTOR_STATE(&mode4->port_list[i], COLLECTING)) && + likely(internals->promiscuous_en || + is_same_ether_addr(&bond_mac, &hdr->d_addr))) { + bufs[num_rx_total++] = pkts[j]; + } else + rte_pktmbuf_free(pkts[j]); + } + } + + return num_rx_total; +} + static inline uint16_t ether_hash(struct ether_hdr *eth_hdr) { @@ -350,6 +401,126 @@ bond_ethdev_tx_burst_balance(void *queue, struct rte_mbuf **bufs, } static uint16_t +bond_ethdev_tx_burst_8023ad(void *queue, struct rte_mbuf **bufs, + uint16_t nb_pkts) +{ + struct bond_dev_private *internals; + struct mode8023ad_data *mode4; + struct bond_tx_queue *bd_tx_q; + + uint8_t num_of_slaves; + uint8_t slaves[RTE_MAX_ETHPORTS]; + /* possitions in slaves, not ID */ + uint8_t distributing_offsets[RTE_MAX_ETHPORTS]; + uint8_t distributing_slaves_count; + + uint16_t num_tx_slave, num_tx_total = 0, tx_fail_total = 0; + uint16_t i, op_slave_idx; + + /* Slow packets from 802.3AX state machines. */ + struct slow_protocol_msg *slow_msg; + + /* Allocate one additional packet in case 8023AD mode. + * First element if not NULL is slow packet. */ + struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][nb_pkts + 1]; + /* Total amount of packets in slave_bufs */ + uint16_t slave_nb_pkts[RTE_MAX_ETHPORTS] = { 0 }; + /* Array of slow packets placed in each slave */ + uint8_t slave_slow_packets[RTE_MAX_ETHPORTS] = { 0 }; + + bd_tx_q = (struct bond_tx_queue *)queue; + internals = bd_tx_q->dev_private; + mode4 = &internals->mode4; + + /* Copy slave list to protect against slave up/down changes during tx + * bursting */ + num_of_slaves = internals->active_slave_count; + if (num_of_slaves < 1) + return num_tx_total; + + memcpy(slaves, internals->active_slaves, sizeof(slaves[0]) * num_of_slaves); + + distributing_slaves_count = mode4->distibuting_slaves_count; + memcpy(distributing_offsets, mode4->distibuting_slaves_offsets, + sizeof(slaves[0]) * distributing_slaves_count); + + for (i = 0; i < num_of_slaves; i++) + slave_bufs[i][0] = NULL; + + /* It is likely that tx ring will be empty. If it is not empty, it is + * likely that there will be only one frame. */ + while (unlikely(!rte_ring_empty(mode4->tx_ring)) && + rte_ring_dequeue(mode4->tx_ring, (void **)&slow_msg) != -ENOENT) { + i = find_slave_by_id(slaves, num_of_slaves, slow_msg->port_id); + + /* Assign slow packet to slave or drop it if slave is not in active list + * (ex: link down). */ + if (likely(i < num_of_slaves)) { + /* If there is more than one slow packet to the same slave, send + * only latest, and drop previouse - tx burst was no called quick + * enough. */ + if (slave_bufs[i][0] != NULL) + rte_pktmbuf_free(slave_bufs[i][0]); + + slave_bufs[i][0] = slow_msg->pkt; + slave_nb_pkts[i] = 1; + slave_slow_packets[i] = 1; + } else + rte_pktmbuf_free(slow_msg->pkt); + + rte_ring_enqueue(mode4->free_ring, slow_msg); + } + + if (likely(distributing_slaves_count > 0)) { + /* Populate slaves mbuf with the packets which are to be sent on it */ + for (i = 0; i < nb_pkts; i++) { + /* Select output slave using hash based on xmit policy */ + op_slave_idx = xmit_slave_hash(bufs[i], distributing_slaves_count, + internals->balance_xmit_policy); + + /* Populate slave mbuf arrays with mbufs for that slave. Use only + * slaves that are currently distributing. */ + uint8_t slave_offset = distributing_offsets[op_slave_idx]; + uint16_t pkt_pos = slave_nb_pkts[slave_offset]; + slave_nb_pkts[slave_offset]++; + + slave_bufs[slave_offset][pkt_pos] = bufs[i]; + } + } + + /* Send packet burst on each slave device */ + for (i = 0; i < num_of_slaves; i++) { + if (slave_nb_pkts[i] > 0) { + num_tx_slave = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id, + slave_bufs[i], slave_nb_pkts[i]); + + /* if tx burst fails move packets to end of bufs */ + if (unlikely(num_tx_slave < slave_nb_pkts[i])) { + uint16_t slave_tx_fail_count = slave_nb_pkts[i] - num_tx_slave; + + /* Free slow packet if it exists and not send. */ + if (slave_slow_packets[i] != 0 && num_tx_slave == 0) { + rte_pktmbuf_free(slave_bufs[i][0]); + slave_tx_fail_count--; + } + + tx_fail_total += slave_tx_fail_count; + memcpy(bufs[nb_pkts - tx_fail_total], + slave_bufs[i][num_tx_slave], + slave_tx_fail_count); + } + + if (num_tx_slave > 0) + num_tx_slave -= slave_slow_packets[i]; + + num_tx_total += num_tx_slave; + } + } + + return num_tx_total; +} + +static uint16_t bond_ethdev_tx_burst_broadcast(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts) { @@ -448,6 +619,27 @@ link_properties_valid(struct rte_eth_link *bonded_dev_link, } int +mac_address_get(struct rte_eth_dev *eth_dev, struct ether_addr *dst_mac_addr) +{ + struct ether_addr *mac_addr; + + mac_addr = eth_dev->data->mac_addrs; + + if (eth_dev == NULL) { + RTE_LOG(ERR, PMD, "%s: NULL pointer eth_dev specified\n", __func__); + return -1; + } + + if (dst_mac_addr == NULL) { + RTE_LOG(ERR, PMD, "%s: NULL pointer MAC specified\n", __func__); + return -1; + } + + ether_addr_copy(mac_addr, dst_mac_addr); + return 0; +} + +int mac_address_set(struct rte_eth_dev *eth_dev, struct ether_addr *new_mac_addr) { struct ether_addr *mac_addr; @@ -455,7 +647,7 @@ mac_address_set(struct rte_eth_dev *eth_dev, struct ether_addr *new_mac_addr) mac_addr = eth_dev->data->mac_addrs; if (eth_dev == NULL) { - RTE_BOND_LOG(ERR, "NULL pointer eth_dev specified"); + RTE_BOND_LOG(ERR, "NULL pointer eth_dev specified"); return -1; } @@ -494,6 +686,8 @@ mac_address_slaves_update(struct rte_eth_dev *bonded_eth_dev) } } break; + case BONDING_MODE_8023AD: + break; case BONDING_MODE_ACTIVE_BACKUP: default: for (i = 0; i < internals->slave_count; i++) { @@ -544,6 +738,13 @@ bond_ethdev_mode_set(struct rte_eth_dev *eth_dev, int mode) eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_broadcast; eth_dev->rx_pkt_burst = bond_ethdev_rx_burst; break; + case BONDING_MODE_8023AD: + if (bond_mode_8023ad_init(eth_dev) != 0) + return -1; + + eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_8023ad; + eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_8023ad; + break; default: return -1; } @@ -751,6 +952,8 @@ bond_ethdev_start(struct rte_eth_dev *eth_dev) if (internals->user_defined_primary_port) bond_ethdev_primary_set(internals, internals->primary_port); + if (internals->mode == BONDING_MODE_8023AD) + bond_mode_8023ad_start(eth_dev); if (internals->link_status_polling_enabled) rte_eal_alarm_set(internals->link_status_polling_interval_ms * 1000, @@ -765,6 +968,25 @@ bond_ethdev_stop(struct rte_eth_dev *eth_dev) { struct bond_dev_private *internals = eth_dev->data->dev_private; + if (internals->mode == BONDING_MODE_8023AD) { + struct mode8023ad_data *data = &internals->mode4; + struct slow_protocol_msg *msg; + + bond_mode_8023ad_stop(eth_dev); + data->distibuting_slaves_count = 0; + + /* Discard all messages to/from mode 4 state machines */ + while (rte_ring_dequeue(data->rx_ring, (void **)&msg) != -ENOENT) { + rte_pktmbuf_free(msg->pkt); + rte_ring_enqueue(data->free_ring, msg); + } + + while (rte_ring_dequeue(data->tx_ring, (void **)&msg) != -ENOENT) { + rte_pktmbuf_free(msg->pkt); + rte_ring_enqueue(data->free_ring, msg); + } + } + internals->active_slave_count = 0; internals->link_status_polling_enabled = 0; @@ -832,7 +1054,7 @@ bond_ethdev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id, 0, dev->pci_dev->numa_node); if (bd_tx_q == NULL) - return -1; + return -1; bd_tx_q->queue_id = tx_queue_id; bd_tx_q->dev_private = dev->data->dev_private; @@ -863,7 +1085,6 @@ bond_ethdev_tx_queue_release(void *queue) rte_free(queue); } - static void bond_ethdev_slave_link_status_change_monitor(void *cb_arg) { @@ -884,7 +1105,7 @@ bond_ethdev_slave_link_status_change_monitor(void *cb_arg) /* If device is currently being configured then don't check slaves link * status, wait until next period */ - if (rte_spinlock_trylock(&internals->lock)){ + if (rte_spinlock_trylock(&internals->lock)) { for (i = 0; i < internals->slave_count; i++) { if (internals->slaves[i].link_status_polling_enabled) { slave_ethdev = &rte_eth_devices[internals->slaves[i].port_id]; @@ -1002,11 +1223,13 @@ bond_ethdev_promiscuous_enable(struct rte_eth_dev *eth_dev) for (i = 0; i < internals->slave_count; i++) rte_eth_promiscuous_enable(internals->slaves[i].port_id); break; + /* In mode4 promiscus mode is managed when slave is added/removed */ + case BONDING_MODE_8023AD: + break; /* Promiscuous mode is propagated only to primary slave */ case BONDING_MODE_ACTIVE_BACKUP: default: rte_eth_promiscuous_enable(internals->current_primary_port); - } } @@ -1017,7 +1240,7 @@ bond_ethdev_promiscuous_disable(struct rte_eth_dev *dev) int i; internals->promiscuous_en = 0; - + switch (internals->mode) { /* Promiscuous mode is propagated to all slaves */ case BONDING_MODE_ROUND_ROBIN: @@ -1026,6 +1249,9 @@ bond_ethdev_promiscuous_disable(struct rte_eth_dev *dev) for (i = 0; i < internals->slave_count; i++) rte_eth_promiscuous_disable(internals->slaves[i].port_id); break; + /* In mode4 promiscus mode is set managed when slave is added/removed */ + case BONDING_MODE_8023AD: + break; /* Promiscuous mode is propagated only to primary slave */ case BONDING_MODE_ACTIVE_BACKUP: default: @@ -1051,7 +1277,8 @@ bond_ethdev_lsc_event_callback(uint8_t port_id, enum rte_eth_event_type type, struct bond_dev_private *internals; struct rte_eth_link link; - int i, valid_slave = 0, active_pos = -1; + int i, valid_slave = 0; + uint8_t active_pos; uint8_t lsc_flag = 0; if (type != RTE_ETH_EVENT_INTR_LSC || param == NULL) @@ -1081,16 +1308,12 @@ bond_ethdev_lsc_event_callback(uint8_t port_id, enum rte_eth_event_type type, return; /* Search for port in active port list */ - for (i = 0; i < internals->active_slave_count; i++) { - if (port_id == internals->active_slaves[i]) { - active_pos = i; - break; - } - } + active_pos = find_slave_by_id(internals->active_slaves, + internals->active_slave_count, port_id); rte_eth_link_get_nowait(port_id, &link); if (link.link_status) { - if (active_pos >= 0) + if (active_pos < internals->active_slave_count) return; /* if no active slave ports then set this port to be primary port */ @@ -1104,21 +1327,19 @@ bond_ethdev_lsc_event_callback(uint8_t port_id, enum rte_eth_event_type type, link_properties_set(bonded_eth_dev, &(slave_eth_dev->data->dev_link)); } - internals->active_slaves[internals->active_slave_count++] = port_id; + + activate_slave(bonded_eth_dev, port_id); /* If user has defined the primary port then default to using it */ if (internals->user_defined_primary_port && internals->primary_port == port_id) bond_ethdev_primary_set(internals, port_id); } else { - if (active_pos < 0) + if (active_pos == internals->active_slave_count) return; /* Remove from active slave list */ - for (i = active_pos; i < (internals->active_slave_count - 1); i++) - internals->active_slaves[i] = internals->active_slaves[i+1]; - - internals->active_slave_count--; + deactivate_slave(bonded_eth_dev, active_pos); /* No active slaves, change link status to down and reset other * link properties */ diff --git a/lib/librte_pmd_bond/rte_eth_bond_private.h b/lib/librte_pmd_bond/rte_eth_bond_private.h index 6db5144..77f7bb0 100644 --- a/lib/librte_pmd_bond/rte_eth_bond_private.h +++ b/lib/librte_pmd_bond/rte_eth_bond_private.h @@ -42,6 +42,7 @@ extern "C" { #include #include "rte_eth_bond.h" +#include "rte_eth_bond_8023ad.h" #define PMD_BOND_SLAVE_PORT_KVARG ("slave") #define PMD_BOND_PRIMARY_SLAVE_KVARG ("primary") @@ -60,6 +61,8 @@ extern "C" { #define RTE_BOND_LOG(lvl, msg, ...) \ RTE_LOG(lvl, PMD, "%s(%d) - " msg "\n", __func__, __LINE__, ##__VA_ARGS__); +#define BONDING_MODE_INVALID 0xFF + extern const char *pmd_bond_init_valid_arguments[]; extern const char *driver_name; @@ -89,7 +92,13 @@ struct bond_tx_queue { /**< Copy of TX configuration structure for queue */ }; - +/** Persisted Slave Configuration Structure */ +struct slave_conf { + uint8_t port_id; + /**< Port Id of slave eth_dev */ + struct ether_addr mac_addr; + /**< Slave eth_dev original MAC address */ +}; /** Bonded slave devices structure */ struct bond_ethdev_slave_ports { uint8_t slaves[RTE_MAX_ETHPORTS]; /**< Slave port id array */ @@ -124,7 +133,7 @@ struct bond_dev_private { uint8_t user_defined_mac; /**< Flag for whether MAC address is user defined or not */ uint8_t promiscuous_en; - /**< Enabled/disable promiscuous mode on slave devices */ + /**< Enabled/disable promiscuous mode on bonding device */ uint8_t link_props_set; /**< flag to denote if the link properties are set */ @@ -143,6 +152,9 @@ struct bond_dev_private { uint8_t slave_count; /**< Number of bonded slaves */ struct bond_slave_details slaves[RTE_MAX_ETHPORTS]; /**< Arary of bonded slaves details */ + + struct mode8023ad_data mode4; + /**< Mode 4 private data */ }; extern struct eth_dev_ops default_dev_ops; @@ -150,6 +162,21 @@ extern struct eth_dev_ops default_dev_ops; int valid_bonded_ethdev(struct rte_eth_dev *eth_dev); +/* Search given slave array to find possition of given id. + * Return slave pos or slaves_count if not found. */ +static inline uint8_t +find_slave_by_id(uint8_t *slaves, uint8_t slaves_count, + uint8_t slave_id ) { + + uint8_t pos; + for (pos = 0; pos < slaves_count; pos++) { + if (slave_id == slaves[pos]) + break; + } + + return pos; +} + int valid_port_id(uint8_t port_id); @@ -160,6 +187,14 @@ int valid_slave_port_id(uint8_t port_id); void +deactivate_slave(struct rte_eth_dev *eth_dev, + uint8_t slave_pos ); + +void +activate_slave(struct rte_eth_dev *eth_dev, + uint8_t port_id ); + +void link_properties_set(struct rte_eth_dev *bonded_eth_dev, struct rte_eth_link *slave_dev_link); void @@ -173,6 +208,9 @@ int mac_address_set(struct rte_eth_dev *eth_dev, struct ether_addr *new_mac_addr); int +mac_address_get(struct rte_eth_dev *eth_dev, struct ether_addr *dst_mac_addr); + +int mac_address_slaves_update(struct rte_eth_dev *bonded_eth_dev); uint8_t -- 1.7.9.5