From patchwork Thu Nov 27 16:33:40 2014 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-Patchwork-Submitter: Daniel Mrzyglod X-Patchwork-Id: 1662 Return-Path: X-Original-To: patchwork@dpdk.org Delivered-To: patchwork@dpdk.org Received: from [92.243.14.124] (localhost [IPv6:::1]) by dpdk.org (Postfix) with ESMTP id 50C50803B; Thu, 27 Nov 2014 17:33:53 +0100 (CET) Received: from mga03.intel.com (mga03.intel.com [134.134.136.65]) by dpdk.org (Postfix) with ESMTP id B1FDC803A for ; Thu, 27 Nov 2014 17:33:50 +0100 (CET) Received: from orsmga001.jf.intel.com ([10.7.209.18]) by orsmga103.jf.intel.com with ESMTP; 27 Nov 2014 08:30:43 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.07,470,1413270000"; d="scan'208";a="615089159" Received: from irvmail001.ir.intel.com ([163.33.26.43]) by orsmga001.jf.intel.com with ESMTP; 27 Nov 2014 08:33:49 -0800 Received: from sivswdev02.ir.intel.com (sivswdev02.ir.intel.com [10.237.217.46]) by irvmail001.ir.intel.com (8.14.3/8.13.6/MailSET/Hub) with ESMTP id sARGXmJi013332; Thu, 27 Nov 2014 16:33:48 GMT Received: from sivswdev02.ir.intel.com (localhost [127.0.0.1]) by sivswdev02.ir.intel.com with ESMTP id sARGXm1d022898; Thu, 27 Nov 2014 16:33:48 GMT Received: (from dtmrzglx@localhost) by sivswdev02.ir.intel.com with id sARGXme4022894; Thu, 27 Nov 2014 16:33:48 GMT From: Daniel Mrzyglod To: dev@dpdk.org Date: Thu, 27 Nov 2014 16:33:40 +0000 Message-Id: <1417106021-22842-2-git-send-email-danielx.t.mrzyglod@intel.com> X-Mailer: git-send-email 1.7.4.1 In-Reply-To: <1417106021-22842-1-git-send-email-danielx.t.mrzyglod@intel.com> References: <1417106021-22842-1-git-send-email-danielx.t.mrzyglod@intel.com> MIME-Version: 1.0 Subject: [dpdk-dev] =?utf-8?q?=5BPATCH_v4_1/2=5D_This_patch_add_support_of?= =?utf-8?q?_mode_5_to_link_bonding_pmd?= 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: , Errors-To: dev-bounces@dpdk.org Sender: "dev" Add support for mode 5 (Transmit load balancing) into pmd driver This patch add support for Adaptive transmit load balancing (mode 5) to the librte_pmd_bond library. This mode provides an adaptive transmit load balancing. It dynamically changes the transmitting slave, according to the computed load. Further details are described here: https://www.kernel.org/doc/Documentation/networking/bonding.txt In implementation callback is used for sorting slave order - providing statistics for burst function about slave bandwith usage and sort interfaces due to usage. Difference in this implementation vs Linux implementation: - We Are trying send all pkts – If one interface hasn’t send packets we are trying to send rest of packets by other slaves sorted previously by callback function. Some implementation details: - Every 100ms is taken obytes statistics from every slave. - Every 10 ms the slaves in table are sorted and updated by callback - bandwidth and successfully transmitted bytes from previous iteration which happens every 100 ms - There is callback function which updates this statistics for transparency and for rather intensive computation involved in this mode. Signed-off-by: Daniel Mrzyglod --- lib/librte_pmd_bond/rte_eth_bond.h | 6 ++ lib/librte_pmd_bond/rte_eth_bond_args.c | 1 + lib/librte_pmd_bond/rte_eth_bond_pmd.c | 160 ++++++++++++++++++++++++++++- lib/librte_pmd_bond/rte_eth_bond_private.h | 2 +- 4 files changed, 166 insertions(+), 3 deletions(-) diff --git a/lib/librte_pmd_bond/rte_eth_bond.h b/lib/librte_pmd_bond/rte_eth_bond.h index 085500b..3831f56 100644 --- a/lib/librte_pmd_bond/rte_eth_bond.h +++ b/lib/librte_pmd_bond/rte_eth_bond.h @@ -77,6 +77,12 @@ extern "C" { * In this mode all transmitted packets will be transmitted on all available * active slaves of the bonded. */ #endif +#define BONDING_MODE_ADAPTIVE_TRANSMIT_LOAD_BALANCING (5) +/**< Adaptive TLB (Mode 5) + * This mode provides an adaptive transmit load balancing. It dynamically + * changes the transmitting slave, according to the computed load. Statistics + * are collected in 100ms intervals and scheduled every 10ms */ + /* Balance Mode Transmit Policies */ #define BALANCE_XMIT_POLICY_LAYER2 (0) /**< Layer 2 (Ethernet MAC) */ diff --git a/lib/librte_pmd_bond/rte_eth_bond_args.c b/lib/librte_pmd_bond/rte_eth_bond_args.c index d8ce681..2675cf6 100644 --- a/lib/librte_pmd_bond/rte_eth_bond_args.c +++ b/lib/librte_pmd_bond/rte_eth_bond_args.c @@ -173,6 +173,7 @@ bond_ethdev_parse_slave_mode_kvarg(const char *key __rte_unused, #ifdef RTE_MBUF_REFCNT case BONDING_MODE_BROADCAST: #endif + case BONDING_MODE_ADAPTIVE_TRANSMIT_LOAD_BALANCING: 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 cf2fbab..7a5dae6 100644 --- a/lib/librte_pmd_bond/rte_eth_bond_pmd.c +++ b/lib/librte_pmd_bond/rte_eth_bond_pmd.c @@ -30,7 +30,7 @@ * (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 @@ -41,10 +41,15 @@ #include #include #include +#include #include "rte_eth_bond.h" #include "rte_eth_bond_private.h" +#define REORDER_PERIOD_MS 10 +/* Table for statistics in mode 5 TLB */ +static uint64_t tlb_last_obytets[RTE_MAX_ETHPORTS]; + static uint16_t bond_ethdev_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts) { @@ -288,6 +293,144 @@ xmit_slave_hash(const struct rte_mbuf *buf, uint8_t slave_count, uint8_t policy) return hash % slave_count; } +struct bwg_slave { + uint64_t bwg_left_int; + uint64_t bwg_left_remainder; + uint8_t slave; +}; + +static int +bandwidth_cmp(const void *a, const void *b) +{ + const struct bwg_slave *bwg_a = a; + const struct bwg_slave *bwg_b = b; + int64_t diff = (int64_t)bwg_b->bwg_left_int - (int64_t)bwg_a->bwg_left_int; + int64_t diff2 = (int64_t)bwg_b->bwg_left_remainder - + (int64_t)bwg_a->bwg_left_remainder; + if (diff > 0) + return 1; + else if (diff < 0) + return -1; + else if (diff2 > 0) + return 1; + else if (diff2 < 0) + return -1; + else + return 0; +} + +static void +bandwidth_left(int port_id, uint64_t load, uint8_t update_idx, + struct bwg_slave *bwg_slave) +{ + struct rte_eth_link link_status; + + rte_eth_link_get(port_id, &link_status); + uint64_t link_bwg = link_status.link_speed * 1000000ULL / 8; + if (link_bwg == 0) + return; + link_bwg = (link_bwg * (update_idx+1) * REORDER_PERIOD_MS); + bwg_slave->bwg_left_int = (link_bwg - 1000*load) / link_bwg; + bwg_slave->bwg_left_remainder = (link_bwg - 1000*load) % link_bwg; +} + +static void +bond_ethdev_update_tlb_slave_cb(void *arg) +{ + struct bond_dev_private *internals = arg; + struct rte_eth_stats slave_stats; + struct bwg_slave bwg_array[RTE_MAX_ETHPORTS]; + uint8_t slave_count; + uint64_t tx_bytes; + + uint8_t update_stats = 0; + uint8_t i, slave_id; + + internals->slave_update_idx++; + + + if (internals->slave_update_idx >= REORDER_PERIOD_MS) + update_stats = 1; + + for (i = 0; i < internals->active_slave_count; i++) { + slave_id = internals->active_slaves[i]; + rte_eth_stats_get(slave_id, &slave_stats); + tx_bytes = slave_stats.obytes - tlb_last_obytets[slave_id]; + bandwidth_left(slave_id, tx_bytes, + internals->slave_update_idx, &bwg_array[i]); + bwg_array[i].slave = slave_id; + + if (update_stats) + tlb_last_obytets[slave_id] = slave_stats.obytes; + } + + if (update_stats == 1) + internals->slave_update_idx = 0; + + slave_count = i; + qsort(bwg_array, slave_count, sizeof(bwg_array[0]), bandwidth_cmp); + for (i = 0; i < slave_count; i++) + internals->active_slaves[i] = bwg_array[i].slave; + + rte_eal_alarm_set(REORDER_PERIOD_MS * 1000, bond_ethdev_update_tlb_slave_cb, + (struct bond_dev_private *)internals); +} + +static uint16_t +bond_ethdev_tx_burst_tlb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts) +{ + struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue; + struct bond_dev_private *internals = bd_tx_q->dev_private; + + struct rte_eth_dev *primary_port = + &rte_eth_devices[internals->primary_port]; + uint16_t num_tx_total = 0; + uint8_t i, j; + + uint8_t num_of_slaves = internals->active_slave_count; + uint8_t slaves[RTE_MAX_ETHPORTS]; + + struct ether_hdr *ether_hdr; + struct ether_addr primary_slave_addr; + struct ether_addr active_slave_addr; + + if (num_of_slaves < 1) + return num_tx_total; + + memcpy(slaves, internals->active_slaves, + sizeof(internals->active_slaves[0]) * num_of_slaves); + + + ether_addr_copy(primary_port->data->mac_addrs, &primary_slave_addr); + + if (nb_pkts > 3) { + for (i = 0; i < 3; i++) + rte_prefetch0(rte_pktmbuf_mtod(bufs[i], void*)); + } + + for (i = 0; i < num_of_slaves; i++) { + ether_addr_copy(&internals->slaves[slaves[i]].persisted_mac_addr, + &active_slave_addr); + + for (j = num_tx_total; j < nb_pkts; j++) { + if (j + 3 < nb_pkts) + rte_prefetch0(rte_pktmbuf_mtod(bufs[j+3], void*)); + + ether_hdr = rte_pktmbuf_mtod(bufs[j], struct ether_hdr *); + if (is_same_ether_addr(ðer_hdr->s_addr, &primary_slave_addr)) + ether_addr_copy(&active_slave_addr, ðer_hdr->s_addr); + } + + num_tx_total += rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id, + bufs + num_tx_total, nb_pkts - num_tx_total); + + if (num_tx_total == nb_pkts) + break; + } + + return num_tx_total; +} + static uint16_t bond_ethdev_tx_burst_balance(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts) @@ -500,6 +643,7 @@ mac_address_slaves_update(struct rte_eth_dev *bonded_eth_dev) } break; case BONDING_MODE_ACTIVE_BACKUP: + case BONDING_MODE_ADAPTIVE_TRANSMIT_LOAD_BALANCING: default: for (i = 0; i < internals->slave_count; i++) { if (internals->slaves[i].port_id == @@ -551,6 +695,10 @@ bond_ethdev_mode_set(struct rte_eth_dev *eth_dev, int mode) eth_dev->rx_pkt_burst = bond_ethdev_rx_burst; break; #endif + case BONDING_MODE_ADAPTIVE_TRANSMIT_LOAD_BALANCING: + eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_tlb; + eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_active_backup; + break; default: return -1; } @@ -676,7 +824,7 @@ slave_add(struct bond_dev_private *internals, } slave_details->link_status_wait_to_complete = 0; - + /* clean tlb_last_obytes when adding port for bonding device */ memcpy(&(slave_details->persisted_mac_addr), slave_eth_dev->data->mac_addrs, sizeof(struct ether_addr)); } @@ -762,6 +910,9 @@ 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_ADAPTIVE_TRANSMIT_LOAD_BALANCING) + bond_ethdev_update_tlb_slave_cb(internals); + return 0; } @@ -770,6 +921,9 @@ bond_ethdev_stop(struct rte_eth_dev *eth_dev) { struct bond_dev_private *internals = eth_dev->data->dev_private; + if (internals->mode == BONDING_MODE_ADAPTIVE_TRANSMIT_LOAD_BALANCING) { + rte_eal_alarm_cancel(bond_ethdev_update_tlb_slave_cb, internals); + } internals->active_slave_count = 0; internals->link_status_polling_enabled = 0; @@ -1016,6 +1170,7 @@ bond_ethdev_promiscuous_enable(struct rte_eth_dev *eth_dev) break; /* Promiscuous mode is propagated only to primary slave */ case BONDING_MODE_ACTIVE_BACKUP: + case BONDING_MODE_ADAPTIVE_TRANSMIT_LOAD_BALANCING: default: rte_eth_promiscuous_enable(internals->current_primary_port); @@ -1042,6 +1197,7 @@ bond_ethdev_promiscuous_disable(struct rte_eth_dev *dev) break; /* Promiscuous mode is propagated only to primary slave */ case BONDING_MODE_ACTIVE_BACKUP: + case BONDING_MODE_ADAPTIVE_TRANSMIT_LOAD_BALANCING: default: rte_eth_promiscuous_disable(internals->current_primary_port); } diff --git a/lib/librte_pmd_bond/rte_eth_bond_private.h b/lib/librte_pmd_bond/rte_eth_bond_private.h index 6254c84..2a4e129 100644 --- a/lib/librte_pmd_bond/rte_eth_bond_private.h +++ b/lib/librte_pmd_bond/rte_eth_bond_private.h @@ -102,7 +102,6 @@ struct bond_slave_details { uint8_t link_status_poll_enabled; uint8_t link_status_wait_to_complete; uint8_t last_link_status; - /**< Port Id of slave eth_dev */ struct ether_addr persisted_mac_addr; }; @@ -145,6 +144,7 @@ struct bond_dev_private { /**< Arary of bonded slaves details */ struct rte_kvargs *kvlist; + uint8_t slave_update_idx; }; extern struct eth_dev_ops default_dev_ops;