examples/vmdq: use new VMDQ API
authorHuawei Xie <huawei.xie@intel.com>
Mon, 10 Nov 2014 00:29:38 +0000 (08:29 +0800)
committerThomas Monjalon <thomas.monjalon@6wind.com>
Tue, 11 Nov 2014 22:48:05 +0000 (23:48 +0100)
This patch supports new VMDQ API in vmdq example.
Besides, it allows users to specify num_pools different with
max_nb_pools, thus the polling thread needn't to poll queues
of all pools.

Due to i40e implementation issue, there is no default mac for
VMDQ pool, so app needs to specify mac address for each pool
explicitly.

Signed-off-by: Huawei Xie <huawei.xie@intel.com>
Acked-by: Jing Chen <jing.d.chen@intel.com>
examples/vmdq/main.c

index c51e2fb..5a2305f 100644 (file)
@@ -144,6 +144,13 @@ const uint16_t vlan_tags[] = {
        48, 49, 50, 51, 52, 53, 54, 55,
        56, 57, 58, 59, 60, 61, 62, 63,
 };
+const uint16_t num_vlans = RTE_DIM(vlan_tags);
+static uint16_t num_pf_queues,  num_vmdq_queues;
+static uint16_t vmdq_pool_base, vmdq_queue_base;
+/* pool mac addr template, pool mac addr is like: 52 54 00 12 port# pool# */
+static struct ether_addr pool_addr_template = {
+       .addr_bytes = {0x52, 0x54, 0x00, 0x12, 0x00, 0x00}
+};
 
 /* ethernet addresses of ports */
 static struct ether_addr vmdq_ports_eth_addr[RTE_MAX_ETHPORTS];
@@ -163,22 +170,9 @@ get_eth_conf(struct rte_eth_conf *eth_conf, uint32_t num_pools)
        unsigned i;
 
        conf.nb_queue_pools = (enum rte_eth_nb_pools)num_pools;
+       conf.nb_pool_maps = num_pools;
        conf.enable_default_pool = 0;
        conf.default_pool = 0; /* set explicit value, even if not used */
-       switch (num_pools) {
-       /* For 10G NIC like 82599, 128 is valid for queue number */
-       case MAX_POOL_NUM_10G:
-               num_queues = MAX_QUEUE_NUM_10G;
-               conf.nb_pool_maps = MAX_POOL_MAP_NUM_10G;
-               break;
-       /* For 1G NIC like i350, 82580 and 82576, 8 is valid for queue number */
-       case MAX_POOL_NUM_1G:
-               num_queues = MAX_QUEUE_NUM_1G;
-               conf.nb_pool_maps = MAX_POOL_MAP_NUM_1G;
-               break;
-       default:
-               return -1;
-       }
 
        for (i = 0; i < conf.nb_pool_maps; i++){
                conf.pool_map[i].vlan_id = vlan_tags[ i ];
@@ -191,40 +185,6 @@ get_eth_conf(struct rte_eth_conf *eth_conf, uint32_t num_pools)
        return 0;
 }
 
-/*
- * Validate the pool number accrording to the max pool number gotten form dev_info
- * If the pool number is invalid, give the error message and return -1
- */
-static inline int
-validate_num_pools(uint32_t max_nb_pools)
-{
-       if (num_pools > max_nb_pools) {
-               printf("invalid number of pools\n");
-               return -1;
-       }
-
-       switch (max_nb_pools) {
-       /* For 10G NIC like 82599, 64 is valid for pool number */
-       case MAX_POOL_NUM_10G:
-               if (num_pools != MAX_POOL_NUM_10G) {
-                       printf("invalid number of pools\n");
-                       return -1;
-               }
-               break;
-       /* For 1G NIC like i350, 82580 and 82576, 8 is valid for pool number */
-       case MAX_POOL_NUM_1G:
-               if (num_pools != MAX_POOL_NUM_1G) {
-                       printf("invalid number of pools\n");
-                       return -1;
-               }
-               break;
-       default:
-               return -1;
-       }
-
-       return 0;
-}
-
 /*
  * Initialises a given port using global settings and with the rx buffers
  * coming from the mbuf_pool passed as parameter
@@ -235,26 +195,57 @@ port_init(uint8_t port, struct rte_mempool *mbuf_pool)
        struct rte_eth_dev_info dev_info;
        struct rte_eth_rxconf *rxconf;
        struct rte_eth_conf port_conf;
-       uint16_t rxRings, txRings = (uint16_t)rte_lcore_count();
+       uint16_t rxRings, txRings;
        const uint16_t rxRingSize = RTE_TEST_RX_DESC_DEFAULT, txRingSize = RTE_TEST_TX_DESC_DEFAULT;
        int retval;
        uint16_t q;
+       uint16_t queues_per_pool;
        uint32_t max_nb_pools;
 
        /* The max pool number from dev_info will be used to validate the pool number specified in cmd line */
        rte_eth_dev_info_get (port, &dev_info);
        max_nb_pools = (uint32_t)dev_info.max_vmdq_pools;
-       retval = validate_num_pools(max_nb_pools);
-       if (retval < 0)
-               return retval;
-
-       retval = get_eth_conf(&port_conf, num_pools);
+       /*
+        * We allow to process part of VMDQ pools specified by num_pools in
+        * command line.
+        */
+       if (num_pools > max_nb_pools) {
+               printf("num_pools %d >max_nb_pools %d\n",
+                       num_pools, max_nb_pools);
+               return -1;
+       }
+       retval = get_eth_conf(&port_conf, max_nb_pools);
        if (retval < 0)
                return retval;
 
+       /*
+        * NIC queues are divided into pf queues and vmdq queues.
+        */
+       /* There is assumption here all ports have the same configuration! */
+       num_pf_queues = dev_info.max_rx_queues - dev_info.vmdq_queue_num;
+       queues_per_pool = dev_info.vmdq_queue_num / dev_info.max_vmdq_pools;
+       num_vmdq_queues = num_pools * queues_per_pool;
+       num_queues = num_pf_queues + num_vmdq_queues;
+       vmdq_queue_base = dev_info.vmdq_queue_base;
+       vmdq_pool_base  = dev_info.vmdq_pool_base;
+
+       printf("pf queue num: %u, configured vmdq pool num: %u,"
+               " each vmdq pool has %u queues\n",
+               num_pf_queues, num_pools, queues_per_pool);
+       printf("vmdq queue base: %d pool base %d\n",
+               vmdq_queue_base, vmdq_pool_base);
        if (port >= rte_eth_dev_count()) return -1;
 
-       rxRings = (uint16_t)num_queues,
+       /*
+        * Though in this example, we only receive packets from the first queue
+        * of each pool and send packets through first rte_lcore_count() tx
+        * queues of vmdq queues, all queues including pf queues are setup.
+        * This is because VMDQ queues doesn't always start from zero, and the
+        * PMD layer doesn't support selectively initialising part of rx/tx
+        * queues.
+        */
+       rxRings = (uint16_t)dev_info.max_rx_queues;
+       txRings = (uint16_t)dev_info.max_tx_queues;
        retval = rte_eth_dev_configure(port, rxRings, txRings, &port_conf);
        if (retval != 0)
                return retval;
@@ -267,21 +258,27 @@ port_init(uint8_t port, struct rte_mempool *mbuf_pool)
                                        rte_eth_dev_socket_id(port),
                                        rxconf,
                                        mbuf_pool);
-               if (retval < 0)
+               if (retval < 0) {
+                       printf("initialise rx queue %d failed\n", q);
                        return retval;
+               }
        }
 
        for (q = 0; q < txRings; q ++) {
                retval = rte_eth_tx_queue_setup(port, q, txRingSize,
                                        rte_eth_dev_socket_id(port),
                                        NULL);
-               if (retval < 0)
+               if (retval < 0) {
+                       printf("initialise tx queue %d failed\n", q);
                        return retval;
+               }
        }
 
        retval  = rte_eth_dev_start(port);
-       if (retval < 0)
+       if (retval < 0) {
+               printf("port %d start failed\n", port);
                return retval;
+       }
 
        rte_eth_macaddr_get(port, &vmdq_ports_eth_addr[port]);
        printf("Port %u MAC: %02"PRIx8" %02"PRIx8" %02"PRIx8
@@ -294,6 +291,29 @@ port_init(uint8_t port, struct rte_mempool *mbuf_pool)
                        vmdq_ports_eth_addr[port].addr_bytes[4],
                        vmdq_ports_eth_addr[port].addr_bytes[5]);
 
+       /*
+        * Set mac for each pool.
+        * There is no default mac for the pools in i40.
+        * Removes this after i40e fixes this issue.
+        */
+       for (q = 0; q < num_pools; q++) {
+               struct ether_addr mac;
+               mac = pool_addr_template;
+               mac.addr_bytes[4] = port;
+               mac.addr_bytes[5] = q;
+               printf("Port %u vmdq pool %u set mac %02x:%02x:%02x:%02x:%02x:%02x\n",
+                       port, q,
+                       mac.addr_bytes[0], mac.addr_bytes[1],
+                       mac.addr_bytes[2], mac.addr_bytes[3],
+                       mac.addr_bytes[4], mac.addr_bytes[5]);
+               retval = rte_eth_dev_mac_addr_add(port, &mac,
+                               q + vmdq_pool_base);
+               if (retval) {
+                       printf("mac addr add failed at pool %d\n", q);
+                       return retval;
+               }
+       }
+
        return 0;
 }
 
@@ -309,6 +329,11 @@ vmdq_parse_num_pools(const char *q_arg)
        if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
                return -1;
 
+       if (num_pools > num_vlans) {
+               printf("num_pools %d > num_vlans %d\n", num_pools, num_vlans);
+               return -1;
+       }
+
        num_pools = n;
 
        return 0;
@@ -437,7 +462,7 @@ lcore_main(__attribute__((__unused__)) void* dummy)
        uint16_t core_id = 0;
        uint16_t startQueue, endQueue;
        uint16_t q, i, p;
-       const uint16_t remainder = (uint16_t)(num_queues % num_cores);
+       const uint16_t remainder = (uint16_t)(num_vmdq_queues % num_cores);
 
        for (i = 0; i < num_cores; i ++)
                if (lcore_ids[i] == lcore_id) {
@@ -447,17 +472,27 @@ lcore_main(__attribute__((__unused__)) void* dummy)
 
        if (remainder != 0) {
                if (core_id < remainder) {
-                       startQueue = (uint16_t)(core_id * (num_queues/num_cores + 1));
-                       endQueue = (uint16_t)(startQueue + (num_queues/num_cores) + 1);
+                       startQueue = (uint16_t)(core_id *
+                                       (num_vmdq_queues / num_cores + 1));
+                       endQueue = (uint16_t)(startQueue +
+                                       (num_vmdq_queues / num_cores) + 1);
                } else {
-                       startQueue = (uint16_t)(core_id * (num_queues/num_cores) + remainder);
-                       endQueue = (uint16_t)(startQueue + (num_queues/num_cores));
+                       startQueue = (uint16_t)(core_id *
+                                       (num_vmdq_queues / num_cores) +
+                                       remainder);
+                       endQueue = (uint16_t)(startQueue +
+                                       (num_vmdq_queues / num_cores));
                }
        } else {
-               startQueue = (uint16_t)(core_id * (num_queues/num_cores));
-               endQueue = (uint16_t)(startQueue + (num_queues/num_cores));
+               startQueue = (uint16_t)(core_id *
+                               (num_vmdq_queues / num_cores));
+               endQueue = (uint16_t)(startQueue +
+                               (num_vmdq_queues / num_cores));
        }
 
+       /* vmdq queue idx doesn't always start from zero.*/
+       startQueue += vmdq_queue_base;
+       endQueue   += vmdq_queue_base;
        printf("core %u(lcore %u) reading queues %i-%i\n", (unsigned)core_id,
                (unsigned)lcore_id, startQueue, endQueue - 1);
 
@@ -490,7 +525,9 @@ lcore_main(__attribute__((__unused__)) void* dummy)
                                        update_mac_address(buf[i], dport);
 
                                const uint16_t txCount = rte_eth_tx_burst(dport,
-                                       core_id, buf, rxCount);
+                                       vmdq_queue_base + core_id,
+                                       buf,
+                                       rxCount);
 
                                if (txCount != rxCount) {
                                        for (i = txCount; i < rxCount; i++)