/* // Copyright (c) 2010-2017 Intel Corporation // // Licensed under the Apache License, Version 2.0 (the "License"); // you may not use this file except in compliance with the License. // You may obtain a copy of the License at // // http://www.apache.org/licenses/LICENSE-2.0 // // Unless required by applicable law or agreed to in writing, software // distributed under the License is distributed on an "AS IS" BASIS, // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. // See the License for the specific language governing permissions and // limitations under the License. */ #include #include #include #include #include #include #include "prox_malloc.h" #include "log.h" #include "quit.h" #include "stats_port.h" #include "prox_port_cfg.h" #include "rw_reg.h" #include "prox_compat.h" #if defined(PROX_STATS) && defined(PROX_HW_DIRECT_STATS) /* Directly access hardware counters instead of going through DPDK. This allows getting * specific counters that DPDK does not report or aggregates with other ones. */ /* Definitions for IXGBE (taken from PMD) */ #define PROX_IXGBE_MPC(_i) (0x03FA0 + ((_i) * 4)) /* 8 of these 3FA0-3FBC*/ #define PROX_IXGBE_QBRC_L(_i) (0x01034 + ((_i) * 0x40)) /* 16 of these */ #define PROX_IXGBE_QBRC_H(_i) (0x01038 + ((_i) * 0x40)) /* 16 of these */ #define PROX_IXGBE_QPRC(_i) (0x01030 + ((_i) * 0x40)) /* 16 of these */ #define PROX_IXGBE_GPTC 0x04080 #define PROX_IXGBE_TPR 0x040D0 #define PROX_IXGBE_TORL 0x040C0 #define PROX_IXGBE_TORH 0x040C4 #define PROX_IXGBE_GOTCL 0x04090 #define PROX_IXGBE_GOTCH 0x04094 #define IXGBE_QUEUE_STAT_COUNTERS 16 static void ixgbe_read_stats(uint8_t port_id, struct port_stats_sample* stats, struct port_stats_sample *prev, int last_stat) { uint64_t before, after; unsigned i; struct rte_eth_dev* dev = &rte_eth_devices[port_id]; /* WARNING: Assumes hardware address is first field of structure! This may change! */ struct _dev_hw* hw = (struct _dev_hw *)(dev->data->dev_private); stats->no_mbufs = dev->data->rx_mbuf_alloc_failed; /* Since we only read deltas from the NIC, we have to add to previous values * even though we actually substract again later to find out the rates! */ stats->ierrors = prev->ierrors; stats->imissed = prev->imissed; stats->rx_bytes = prev->rx_bytes; stats->rx_tot = prev->rx_tot; stats->tx_bytes = prev->tx_bytes; stats->tx_tot = prev->tx_tot; /* WARNING: In this implementation, we count as imiised only the "no descriptor" * missed packets cases and not the actual receive errors. */ before = rte_rdtsc(); for (i = 0; i < 8; i++) { stats->imissed += PROX_READ_REG(hw, PROX_IXGBE_MPC(i)); } /* RX stats */ #if 0 /* This version is equivalent to what ixgbe PMD does. It only accounts for packets * actually received on the host. */ for (i = 0; i < IXGBE_QUEUE_STAT_COUNTERS; i++) { /* ipackets: */ stats->rx_tot += PROX_READ_REG(hw, PROX_IXGBE_QPRC(i)); /* ibytes: */ stats->rx_bytes += PROX_READ_REG(hw, PROX_IXGBE_QBRC_L(i)); stats->rx_bytes += ((uint64_t)PROX_READ_REG(hw, PROX_IXGBE_QBRC_H(i)) << 32); } #else /* This version reports the packets received by the NIC, regardless of whether they * reached the host or not, etc. (no need to add ierrors or imissedto this packet count) */ stats->rx_tot += PROX_READ_REG(hw, PROX_IXGBE_TPR); stats->rx_bytes += PROX_READ_REG(hw, PROX_IXGBE_TORL); stats->rx_bytes += ((uint64_t)PROX_READ_REG(hw, PROX_IXGBE_TORH) << 32); #endif /* TX stats */ /* opackets: */ stats->tx_tot += PROX_READ_REG(hw, PROX_IXGBE_GPTC); /* obytes: */ stats->tx_bytes += PROX_READ_REG(hw, PROX_IXGBE_GOTCL); stats->tx_bytes += ((uint64_t)PROX_READ_REG(hw, PROX_IXGBE_GOTCH) << 32); after = rte_rdtsc(); stats->tsc = (before >> 1) + (after >> 1); } #endif extern int last_stat; static struct port_stats port_stats[PROX_MAX_PORTS]; static uint8_t nb_interface; static uint8_t n_ports; static int num_xstats[PROX_MAX_PORTS] = {0}; static int num_ixgbe_xstats = 0; #if RTE_VERSION >= RTE_VERSION_NUM(2,1,0,1) #define XSTATS_SUPPORT 1 #else #define XSTATS_SUPPORT 0 #endif #if XSTATS_SUPPORT #if RTE_VERSION >= RTE_VERSION_NUM(16,7,0,0) static struct rte_eth_xstat *eth_xstats[PROX_MAX_PORTS] = {0}; static struct rte_eth_xstat_name *eth_xstat_names[PROX_MAX_PORTS] = {0}; #else static struct rte_eth_xstats *eth_xstats[PROX_MAX_PORTS] = {0}; static struct rte_eth_xstats *eth_xstat_names[PROX_MAX_PORTS] = {0}; #endif static int xstat_tpr_offset[PROX_MAX_PORTS] ={0}, xstat_tor_offset[PROX_MAX_PORTS] = {0}; static int tx_pkt_size_offset[PROX_MAX_PORTS][PKT_SIZE_COUNT]; #endif #if RTE_VERSION >= RTE_VERSION_NUM(16,7,0,0) static int find_xstats_str(struct rte_eth_xstat_name *xstats, int n, const char *name) #else static int find_xstats_str(struct rte_eth_xstats *xstats, int n, const char *name) #endif { for (int i = 0; i < n; i++) { if (strcmp(xstats[i].name, name) == 0) return i; } return -1; } void stats_port_init(void) { int potential_ixgbe_warn = 0; for (int i = 0; i < PROX_MAX_PORTS; i++) { xstat_tpr_offset[i] = -1; xstat_tor_offset[i] = -1; for (int j = 0; j < PKT_SIZE_COUNT; j++) { tx_pkt_size_offset[i][j] = -1; } } #if XSTATS_SUPPORT nb_interface = prox_last_port_active() + 1; n_ports = prox_nb_active_ports(); for (uint8_t port_id = 0; port_id < nb_interface; ++port_id) { if (prox_port_cfg[port_id].active) { #if RTE_VERSION >= RTE_VERSION_NUM(16,7,0,0) num_xstats[port_id] = rte_eth_xstats_get_names(port_id, NULL, 0); eth_xstat_names[port_id] = prox_zmalloc(num_xstats[port_id] * sizeof(struct rte_eth_xstat_name), prox_port_cfg[port_id].socket); PROX_PANIC(eth_xstat_names[port_id] == NULL, "Error allocating memory for xstats"); num_xstats[port_id] = rte_eth_xstats_get_names(port_id, eth_xstat_names[port_id], num_xstats[port_id]); eth_xstats[port_id] = prox_zmalloc(num_xstats[port_id] * sizeof(struct rte_eth_xstat), prox_port_cfg[port_id].socket); PROX_PANIC(eth_xstats[port_id] == NULL, "Error allocating memory for xstats"); #else num_xstats[port_id] = rte_eth_xstats_get(port_id, NULL, 0); eth_xstats[port_id] = prox_zmalloc(num_xstats[port_id] * sizeof(struct rte_eth_xstats), prox_port_cfg[port_id].socket); PROX_PANIC(eth_xstats[port_id] == NULL, "Error allocating memory for xstats"); eth_xstat_names[port_id] = eth_xstats[port_id]; num_xstats[port_id] = rte_eth_xstats_get(port_id, eth_xstats[port_id], num_xstats[port_id]); #endif if (!strcmp(prox_port_cfg[port_id].short_name, "ixgbe")) { potential_ixgbe_warn = 1; xstat_tor_offset[port_id] = find_xstats_str(eth_xstat_names[port_id], num_xstats[port_id], "rx_total_bytes"); xstat_tpr_offset[port_id] = find_xstats_str(eth_xstat_names[port_id], num_xstats[port_id], "rx_total_packets"); } tx_pkt_size_offset[port_id][PKT_SIZE_64] = find_xstats_str(eth_xstat_names[port_id], num_xstats[port_id], "tx_size_64_packets"); tx_pkt_size_offset[port_id][PKT_SIZE_65] = find_xstats_str(eth_xstat_names[port_id], num_xstats[port_id], "tx_size_65_to_127_packets"); tx_pkt_size_offset[port_id][PKT_SIZE_128] = find_xstats_str(eth_xstat_names[port_id], num_xstats[port_id], "tx_size_128_to_255_packets"); tx_pkt_size_offset[port_id][PKT_SIZE_256] = find_xstats_str(eth_xstat_names[port_id], num_xstats[port_id], "tx_size_256_to_511_packets"); tx_pkt_size_offset[port_id][PKT_SIZE_512] = find_xstats_str(eth_xstat_names[port_id], num_xstats[port_id], "tx_size_512_to_1023_packets"); if (0 == strcmp(prox_port_cfg[port_id].short_name, "ixgbe")) { tx_pkt_size_offset[port_id][PKT_SIZE_1024] = find_xstats_str(eth_xstat_names[port_id], num_xstats[port_id], "tx_size_1024_to_max_packets"); } else { tx_pkt_size_offset[port_id][PKT_SIZE_1024] = find_xstats_str(eth_xstat_names[port_id], num_xstats[port_id], "tx_size_1024_to_1522_packets"); tx_pkt_size_offset[port_id][PKT_SIZE_1522] = find_xstats_str(eth_xstat_names[port_id], num_xstats[port_id], "tx_size_1523_to_max_packets"); } plog_info("offset = %d, %d, %d, %d, %d, %d %d\n", tx_pkt_size_offset[port_id][PKT_SIZE_64], tx_pkt_size_offset[port_id][PKT_SIZE_65], tx_pkt_size_offset[port_id][PKT_SIZE_128], tx_pkt_size_offset[port_id][PKT_SIZE_256], tx_pkt_size_offset[port_id][PKT_SIZE_512], tx_pkt_size_offset[port_id][PKT_SIZE_1024], tx_pkt_size_offset[port_id][PKT_SIZE_1522]); #if RTE_VERSION >= RTE_VERSION_NUM(16,7,0,0) prox_free(eth_xstat_names[port_id]); #endif if (num_xstats[port_id] == 0 || eth_xstats[port_id] == NULL) { plog_warn("Failed to initialize xstat for port %d, running without xstats\n", port_id); num_xstats[port_id] = 0; } } } for (uint8_t port_id = 0; port_id < nb_interface; ++port_id) { if ((xstat_tor_offset[port_id] != -1) && (xstat_tpr_offset[port_id] != -1)) { num_ixgbe_xstats = 2; // ixgbe PMD supports tor and tpr xstats break; } } if ((num_ixgbe_xstats == 0) && (potential_ixgbe_warn)) plog_warn("Failed to initialize ixgbe xstat, running without ixgbe xstats\n"); #endif } static void nic_read_stats(uint8_t port_id) { unsigned is_ixgbe = (0 == strcmp(prox_port_cfg[port_id].short_name, "ixgbe")); struct port_stats_sample *stats = &port_stats[port_id].sample[last_stat]; #if defined(PROX_STATS) && defined(PROX_HW_DIRECT_STATS) if (is_ixgbe) { struct port_stats_sample *prev = &port_stats[port_id].sample[!last_stat]; ixgbe_read_stats(port_id, stats, prev, last_stat); return; } #endif uint64_t before, after; struct rte_eth_stats eth_stat; before = rte_rdtsc(); rte_eth_stats_get(port_id, ð_stat); after = rte_rdtsc(); stats->tsc = (before >> 1) + (after >> 1); stats->no_mbufs = eth_stat.rx_nombuf; stats->ierrors = eth_stat.ierrors; stats->imissed = eth_stat.imissed; stats->oerrors = eth_stat.oerrors; stats->rx_bytes = eth_stat.ibytes; /* The goal would be to get the total number of bytes received by the NIC (including overhead). Without the patch (i.e. num_ixgbe_xstats == 0) we can't do this directly with DPDK 2.1 API. So, we report the number of bytes (including overhead) received by the host. */ #if XSTATS_SUPPORT if (num_xstats[port_id]) { rte_eth_xstats_get(port_id, eth_xstats[port_id], num_xstats[port_id]); for (size_t i = 0; i < sizeof(tx_pkt_size_offset[0])/sizeof(tx_pkt_size_offset[0][0]); ++i) { if (tx_pkt_size_offset[port_id][i] != -1) stats->tx_pkt_size[i] = (eth_xstats[port_id][tx_pkt_size_offset[port_id][i]]).value; else stats->tx_pkt_size[i] = -1; } } else { for (size_t i = 0; i < sizeof(tx_pkt_size_offset[0])/sizeof(tx_pkt_size_offset[0][0]); ++i) { stats->tx_pkt_size[i] = -1; } } #endif if (is_ixgbe) { #if XSTATS_SUPPORT if (num_ixgbe_xstats) { stats->rx_tot = eth_xstats[port_id][xstat_tpr_offset[port_id]].value; stats->rx_bytes = eth_xstats[port_id][xstat_tor_offset[port_id]].value; } else #endif { stats->rx_tot = eth_stat.ipackets + eth_stat.ierrors + eth_stat.imissed; /* On ixgbe, the rx_bytes counts bytes received by Host without overhead. The rx_tot counts the number of packets received by the NIC. If we only add 20 * rx_tot to rx_bytes, the result will also take into account 20 * "number of packets dropped by the nic". Note that in case CRC is stripped on ixgbe, the CRC bytes are not counted. */ if (prox_port_cfg[port_id].requested_rx_offload & DEV_RX_OFFLOAD_CRC_STRIP) stats->rx_bytes = eth_stat.ibytes + (24 * eth_stat.ipackets - 20 * (eth_stat.ierrors + eth_stat.imissed)); else stats->rx_bytes = eth_stat.ibytes + (20 * eth_stat.ipackets - 20 * (eth_stat.ierrors + eth_stat.imissed)); } } else if (strcmp(prox_port_cfg[port_id].short_name, "i40e_vf") == 0) { // For I40E VF, imissed already part of received packets stats->rx_tot = eth_stat.ipackets; } else { stats->rx_tot = eth_stat.ipackets + eth_stat.imissed; } stats->tx_tot = eth_stat.opackets; stats->tx_bytes = eth_stat.obytes; } void stats_port_reset(void) { for (uint8_t port_id = 0; port_id < nb_interface; ++port_id) { if (prox_port_cfg[port_id].active) { rte_eth_stats_reset(port_id); memset(&port_stats[port_id], 0, sizeof(struct port_stats)); } } } void stats_port_update(void) { for (uint8_t port_id = 0; port_id < nb_interface; ++port_id) { if (prox_port_cfg[port_id].active) { nic_read_stats(port_id); } } } uint64_t stats_port_get_ierrors(void) { uint64_t ret = 0; for (uint8_t port_id = 0; port_id < nb_interface; ++port_id) { if (prox_port_cfg[port_id].active) ret += port_stats[port_id].sample[last_stat].ierrors; } return ret; } uint64_t stats_port_get_imissed(void) { uint64_t ret = 0; for (uint8_t port_id = 0; port_id < nb_interface; ++port_id) { if (prox_port_cfg[port_id].active) ret += port_stats[port_id].sample[last_stat].imissed; } return ret; } uint64_t stats_port_get_rx_packets(void) { uint64_t ret = 0; for (uint8_t port_id = 0; port_id < nb_interface; ++port_id) { if (prox_port_cfg[port_id].active) ret += port_stats[port_id].sample[last_stat].rx_tot; } return ret; } uint64_t stats_port_get_tx_packets(void) { uint64_t ret = 0; for (uint8_t port_id = 0; port_id < nb_interface; ++port_id) { if (prox_port_cfg[port_id].active) ret += port_stats[port_id].sample[last_stat].tx_tot; } return ret; } int stats_get_n_ports(void) { return n_ports; } struct port_stats_sample *stats_get_port_stats_sample(uint32_t port_id, int l) { return &port_stats[port_id].sample[l == last_stat]; } int stats_port(uint8_t port_id, struct get_port_stats *gps) { if (!prox_port_cfg[port_id].active) return -1; struct port_stats_sample *last = &port_stats[port_id].sample[last_stat]; struct port_stats_sample *prev = &port_stats[port_id].sample[!last_stat]; gps->no_mbufs_diff = last->no_mbufs - prev->no_mbufs; gps->ierrors_diff = last->ierrors - prev->ierrors; gps->imissed_diff = last->imissed - prev->imissed; gps->rx_bytes_diff = last->rx_bytes - prev->rx_bytes; gps->tx_bytes_diff = last->tx_bytes - prev->tx_bytes; gps->rx_pkts_diff = last->rx_tot - prev->rx_tot; if (unlikely(prev->rx_tot > last->rx_tot)) gps->rx_pkts_diff = 0; gps->tx_pkts_diff = last->tx_tot - prev->tx_tot; if (unlikely(prev->tx_tot > last->tx_tot)) gps->rx_pkts_diff = 0; gps->rx_tot = last->rx_tot; gps->tx_tot = last->tx_tot; gps->no_mbufs_tot = last->no_mbufs; gps->ierrors_tot = last->ierrors; gps->imissed_tot = last->imissed; gps->last_tsc = last->tsc; gps->prev_tsc = prev->tsc; return 0; }