static void
app_init_kni(struct app_params *app) {
uint32_t i;
if (app->n_pktq_kni == 0)
return;
rte_kni_init(app->n_pktq_kni);
for (i = 0; i < app->n_pktq_kni; i++) {
struct app_pktq_kni_params *p_kni = &app->kni_params[i];
struct app_link_params *p_link;
struct rte_eth_dev_info dev_info;
struct app_mempool_params *mempool_params;
struct rte_mempool *mempool;
struct rte_kni_conf conf;
struct rte_kni_ops ops;
/* LINK */
p_link = app_get_link_for_kni(app, p_kni);
memset(&dev_info, 0, sizeof(dev_info));
rte_eth_dev_info_get(p_link->pmd_id, &dev_info);
/* MEMPOOL */
mempool_params = &app->mempool_params[p_kni->mempool_id];
mempool = app->mempool[p_kni->mempool_id];
/* KNI */
memset(&conf, 0, sizeof(conf));
snprintf(conf.name, RTE_KNI_NAMESIZE, "%s", p_kni->name);
conf.force_bind = p_kni->force_bind;
if (conf.force_bind) {
int lcore_id;
lcore_id = cpu_core_map_get_lcore_id(app->core_map,
p_kni->socket_id,
p_kni->core_id,
p_kni->hyper_th_id);
if (lcore_id < 0)
rte_panic("%s invalid CPU core\n", p_kni->name);
conf.core_id = (uint32_t) lcore_id;
}
conf.group_id = p_link->pmd_id;
conf.mbuf_size = mempool_params->buffer_size;
conf.addr = dev_info.pci_dev->addr;
conf.id = dev_info.pci_dev->id;
memset(&ops, 0, sizeof(ops));
ops.port_id = (uint8_t) p_link->pmd_id;
ops.change_mtu = kni_change_mtu;
ops.config_network_if = kni_config_network_interface;
APP_LOG(app, HIGH, "Initializing %s ...", p_kni->name);
app->kni[i] = rte_kni_alloc(mempool, &conf, &ops);
if (!app->kni[i])
rte_panic("%s init error\n", p_kni->name);
}
}
/* Initialize KNI subsystem */
void
init_kni(void)
{
unsigned int num_of_kni_ports = 0, i;
struct kni_port_params** params = kni_port_params_array;
/* Calculate the maximum number of KNI interfaces that will be used */
for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
if (kni_port_params_array[i]) {
RTE_LOG(INFO, KNI, "number of kni lcore %d\n",
params[i]->nb_lcore_k);
num_of_kni_ports +=
(params[i]->nb_lcore_k ? params[i]->nb_lcore_k : 1);
}
}
RTE_LOG(INFO, KNI, "number of kni %d\n", num_of_kni_ports);
/* Invoke rte KNI init to preallocate the ports */
rte_kni_init(num_of_kni_ports);
RTE_LOG(INFO, KNI, "finished init_kni\n");
}
/* Init KNI RX queue */
struct vr_dpdk_queue *
vr_dpdk_kni_rx_queue_init(unsigned lcore_id, struct vr_interface *vif,
unsigned host_lcore_id)
{
struct vr_dpdk_lcore *lcore = vr_dpdk.lcores[lcore_id];
const unsigned socket_id = rte_lcore_to_socket_id(lcore_id);
uint8_t port_id = 0;
unsigned vif_idx = vif->vif_idx;
struct vr_dpdk_queue *rx_queue = &lcore->lcore_rx_queues[vif_idx];
struct vr_dpdk_queue_params *rx_queue_params
= &lcore->lcore_rx_queue_params[vif_idx];
if (vif->vif_type == VIF_TYPE_HOST) {
port_id = (((struct vr_dpdk_ethdev *)(vif->vif_bridge->vif_os))->
ethdev_port_id);
}
/* init queue */
rx_queue->rxq_ops = dpdk_knidev_reader_ops;
rx_queue->q_queue_h = NULL;
rx_queue->q_vif = vrouter_get_interface(vif->vif_rid, vif_idx);
/* create the queue */
struct dpdk_knidev_reader_params reader_params = {
.kni = vif->vif_os,
};
rx_queue->q_queue_h = rx_queue->rxq_ops.f_create(&reader_params, socket_id);
if (rx_queue->q_queue_h == NULL) {
RTE_LOG(ERR, VROUTER, " error creating KNI device %s RX queue"
" at eth device %" PRIu8 "\n", vif->vif_name, port_id);
return NULL;
}
/* store queue params */
rx_queue_params->qp_release_op = &dpdk_kni_rx_queue_release;
return rx_queue;
}
/* Release KNI TX queue */
static void
dpdk_kni_tx_queue_release(unsigned lcore_id, struct vr_interface *vif)
{
struct vr_dpdk_lcore *lcore = vr_dpdk.lcores[lcore_id];
struct vr_dpdk_queue *tx_queue = &lcore->lcore_tx_queues[vif->vif_idx];
struct vr_dpdk_queue_params *tx_queue_params
= &lcore->lcore_tx_queue_params[vif->vif_idx];
tx_queue->txq_ops.f_tx = NULL;
rte_wmb();
/* flush and free the queue */
if (tx_queue->txq_ops.f_free(tx_queue->q_queue_h)) {
RTE_LOG(ERR, VROUTER, " error freeing lcore %u KNI device TX queue\n",
lcore_id);
}
/* reset the queue */
vrouter_put_interface(tx_queue->q_vif);
memset(tx_queue, 0, sizeof(*tx_queue));
memset(tx_queue_params, 0, sizeof(*tx_queue_params));
}
/* Init KNI TX queue */
struct vr_dpdk_queue *
vr_dpdk_kni_tx_queue_init(unsigned lcore_id, struct vr_interface *vif,
unsigned host_lcore_id)
{
struct vr_dpdk_lcore *lcore = vr_dpdk.lcores[lcore_id];
const unsigned socket_id = rte_lcore_to_socket_id(lcore_id);
uint8_t port_id = 0;
unsigned vif_idx = vif->vif_idx;
struct vr_dpdk_queue *tx_queue = &lcore->lcore_tx_queues[vif_idx];
struct vr_dpdk_queue_params *tx_queue_params
= &lcore->lcore_tx_queue_params[vif_idx];
struct vr_dpdk_ethdev *ethdev;
if (vif->vif_type == VIF_TYPE_HOST) {
ethdev = vif->vif_bridge->vif_os;
if (ethdev == NULL) {
RTE_LOG(ERR, VROUTER, " error creating KNI device %s TX queue:"
" bridge vif %u ethdev is not initialized\n",
vif->vif_name, vif->vif_bridge->vif_idx);
return NULL;
}
port_id = ethdev->ethdev_port_id;
}
/* init queue */
tx_queue->txq_ops = dpdk_knidev_writer_ops;
tx_queue->q_queue_h = NULL;
tx_queue->q_vif = vrouter_get_interface(vif->vif_rid, vif_idx);
/* create the queue */
struct dpdk_knidev_writer_params writer_params = {
.kni = vif->vif_os,
.tx_burst_sz = VR_DPDK_TX_BURST_SZ,
};
tx_queue->q_queue_h = tx_queue->txq_ops.f_create(&writer_params, socket_id);
if (tx_queue->q_queue_h == NULL) {
RTE_LOG(ERR, VROUTER, " error creating KNI device %s TX queue"
" at eth device %" PRIu8 "\n", vif->vif_name, port_id);
return NULL;
}
/* store queue params */
tx_queue_params->qp_release_op = &dpdk_kni_tx_queue_release;
return tx_queue;
}
/* Change KNI MTU size callback */
static int
dpdk_knidev_change_mtu(uint8_t port_id, unsigned new_mtu)
{
struct vrouter *router = vrouter_get(0);
struct vr_interface *vif;
int i, ret;
uint8_t ethdev_port_id, slave_port_id;
struct vr_dpdk_ethdev *ethdev = NULL;
RTE_LOG(INFO, VROUTER, "Changing eth device %" PRIu8 " MTU to %u\n",
port_id, new_mtu);
if (port_id >= rte_eth_dev_count()) {
RTE_LOG(ERR, VROUTER, "Error changing eth device %"PRIu8" MTU: invalid eth device\n", port_id);
return -EINVAL;
}
/*
* TODO: DPDK bond PMD does not implement mtu_set op, so we need to
* set the MTU manually for all the slaves.
*/
/* Bond vif uses first slave port ID. */
if (router->vr_eth_if) {
ethdev = (struct vr_dpdk_ethdev *)router->vr_eth_if->vif_os;
if (ethdev && ethdev->ethdev_nb_slaves > 0) {
for (i = 0; i < ethdev->ethdev_nb_slaves; i++) {
if (port_id == ethdev->ethdev_slaves[i])
break;
}
/* Clear ethdev if no port match. */
if (i >= ethdev->ethdev_nb_slaves)
ethdev = NULL;
}
}
if (ethdev && ethdev->ethdev_nb_slaves > 0) {
for (i = 0; i < ethdev->ethdev_nb_slaves; i++) {
slave_port_id = ethdev->ethdev_slaves[i];
RTE_LOG(INFO, VROUTER, " changing bond member eth device %" PRIu8
" MTU to %u\n", slave_port_id, new_mtu);
ret = rte_eth_dev_set_mtu(slave_port_id, new_mtu);
if (ret < 0) {
RTE_LOG(ERR, VROUTER, " error changing bond member eth device %" PRIu8
" MTU: %s (%d)\n", slave_port_id, rte_strerror(-ret), -ret);
return ret;
}
}
} else {
ret = rte_eth_dev_set_mtu(port_id, new_mtu);
if (ret < 0) {
RTE_LOG(ERR, VROUTER, "Error changing eth device %" PRIu8
" MTU: %s (%d)\n", port_id, rte_strerror(-ret), -ret);
}
return ret;
}
/* On success, inform vrouter about new MTU */
for (i = 0; i < router->vr_max_interfaces; i++) {
vif = __vrouter_get_interface(router, i);
if (vif && (vif->vif_type == VIF_TYPE_PHYSICAL)) {
ethdev_port_id = (((struct vr_dpdk_ethdev *)(vif->vif_os))->
ethdev_port_id);
if (ethdev_port_id == port_id) {
/* Ethernet header size */
new_mtu += sizeof(struct vr_eth);
if (vr_dpdk.vlan_tag != VLAN_ID_INVALID) {
/* 802.1q header size */
new_mtu += sizeof(uint32_t);
}
vif->vif_mtu = new_mtu;
if (vif->vif_bridge)
vif->vif_bridge->vif_mtu = new_mtu;
}
}
}
return 0;
}
/* Configure KNI state callback */
static int
dpdk_knidev_config_network_if(uint8_t port_id, uint8_t if_up)
{
int ret = 0;
RTE_LOG(INFO, VROUTER, "Configuring eth device %" PRIu8 " %s\n",
port_id, if_up ? "UP" : "DOWN");
if (port_id >= rte_eth_dev_count() || port_id >= RTE_MAX_ETHPORTS) {
RTE_LOG(ERR, VROUTER, "Invalid eth device %" PRIu8 "\n", port_id);
return -EINVAL;
}
if (if_up)
ret = rte_eth_dev_start(port_id);
else
rte_eth_dev_stop(port_id);
if (ret < 0) {
RTE_LOG(ERR, VROUTER, "Configuring eth device %" PRIu8 " UP"
"failed (%d)\n", port_id, ret);
}
return ret;
}
/* Init KNI */
int
vr_dpdk_knidev_init(uint8_t port_id, struct vr_interface *vif)
{
int i;
struct rte_eth_dev_info dev_info;
struct rte_kni_conf kni_conf;
struct rte_kni_ops kni_ops;
struct rte_kni *kni;
struct rte_config *rte_conf = rte_eal_get_configuration();
if (!vr_dpdk.kni_inited) {
/*
* If the host does not support KNIs (i.e. RedHat), we'll get
* a panic here.
*/
rte_kni_init(VR_DPDK_MAX_KNI_INTERFACES);
vr_dpdk.kni_inited = true;
}
/* get eth device info */
memset(&dev_info, 0, sizeof(dev_info));
rte_eth_dev_info_get(port_id, &dev_info);
/* create KNI configuration */
memset(&kni_conf, 0, sizeof(kni_conf));
strncpy(kni_conf.name, (char *)vif->vif_name, sizeof(kni_conf.name) - 1);
kni_conf.addr = dev_info.pci_dev->addr;
kni_conf.id = dev_info.pci_dev->id;
kni_conf.group_id = port_id;
kni_conf.mbuf_size = VR_DPDK_MAX_PACKET_SZ;
/*
* Due to DPDK commit 41a6ebd, now to prevent packet reordering in KNI
* we have to bind KNI kernel thread to a first online unused CPU.
*/
for (i = 0; i < RTE_MAX_LCORE; i++) {
if (lcore_config[i].detected
&& rte_conf->lcore_role[VR_DPDK_FWD_LCORE_ID + i] == ROLE_OFF) {
kni_conf.force_bind = 1;
kni_conf.core_id = i;
RTE_LOG(INFO, VROUTER, " bind KNI kernel thread to CPU %d\n", i);
break;
}
}
/* KNI options
*
* Changing state of the KNI interface can change state of the physical
* interface. This is useful for the vhost, but not for the VLAN
* forwarding interface.
*/
if (vif->vif_type == VIF_TYPE_VLAN) {
memset(&kni_ops, 0, sizeof(kni_ops));
} else {
kni_ops.port_id = port_id;
kni_ops.change_mtu = dpdk_knidev_change_mtu;
kni_ops.config_network_if = dpdk_knidev_config_network_if;
}
/* allocate KNI device */
kni = rte_kni_alloc(vr_dpdk.rss_mempool, &kni_conf, &kni_ops);
if (kni == NULL) {
RTE_LOG(ERR, VROUTER, " error allocation KNI device %s"
" at eth device %" PRIu8 "\n", vif->vif_name, port_id);
return -ENOMEM;
}
/* store pointer to KNI for further use */
vif->vif_os = kni;
/* add interface to the table of KNIs */
for (i = 0; i < VR_DPDK_MAX_KNI_INTERFACES; i++) {
if (vr_dpdk.knis[i] == NULL) {
vr_dpdk.knis[i] = vif->vif_os;
break;
}
}
return 0;
}
/*
* Initialize PIOT Module
*/
int
rw_piot_init(int argc, char **argv, void *instance_ptr, f_rw_piot_log_t log_fn)
{
int rc, ret;
int i, no_huge = 0, memory_req = 0;
struct rte_config *cfg = rte_eal_get_configuration();
/*
* TBD - thread safety
*/
if (piot_initialized) {
int num_arg;
for (num_arg = 0; num_arg < argc; num_arg ++) {
if (strcmp(argv[num_arg], "--") == 0) {
break;
}
}
return num_arg;
}
piot_initialized = 1;
memset(&rw_piot_global_config, 0, sizeof(rw_piot_global_config));
ASSERT(instance_ptr);
rw_piot_global_config.instance_ptr = instance_ptr;
rw_piot_global_config.log_fn = log_fn;
memset(&(rw_piot_lcore[0]), 0, sizeof(rw_piot_lcore));
for (i=0; i<argc; i++) {
if (strcmp("--no-huge", argv[i]) == 0) {
no_huge = 1;
RW_PIOT_LOG(RTE_LOG_INFO, "PIOT: Huge pages disabled by --no-huge cmdarg\n");
}
if (strcmp("-m", argv[i]) == 0) {
if ((i+1) <argc) {
memory_req = atoi(argv[i+1]);
RW_PIOT_LOG(RTE_LOG_INFO, "PIOT: -m cmdarg setting requested memory to %d mb\n", memory_req);
}
}
if (strcmp("--", argv[i]) == 0) {
break;
}
}
/*
* setup system environment for DPDK
*/
rc = dpdk_system_setup(no_huge, memory_req);
if (rc < 0) {
return rc;
}
rte_set_application_log_hook(rw_piot_log_handler);
/*
* Init DPDK EAL without doing thread related inits
*/
ret = rte_eal_init_no_thread(argc, argv);
if (ret < 0) {
return ret;
}
if (geteuid() == 0){
rte_kni_init(RWPIOT_MAX_KNI_DEVICES);
}
/*
* Assign master lcore-id. Should be passed in the init - TBD
*/
cfg->master_lcore = 0; /* This wil be fixed with RW.Sched integration - TBD */
/* set the lcore ID in per-lcore memory area */
RTE_PER_LCORE(_lcore_id) = cfg->master_lcore;
/* set CPU affinity for master thread ??? TBD*/
// if (eal_thread_set_affinity() < 0)
// rte_panic("cannot set affinity\n");
rte_timer_subsystem_init();
return ret; /* number of args consumed by rte_eal_init_no_thread */
}
int kni_dev_init(struct rte_mempool *mbuf_pool)
{
struct conf_sect *s = conf_get_sect("interface");
struct conf_opt *opt;
struct rte_kni_conf conf;
struct rte_kni_ops ops;
struct rte_kni *kni;
struct net_device *dev;
struct knidev *knidev;
pthread_t tid;
struct ifconfig_arg arg;
int i = 0, x = rte_eth_dev_count();
for (opt = s->opt; opt; opt = opt->next) {
const char *busid = conf_get_subopt(opt, "busid");
if (!strcmp(busid, "kni"))
kni_cnt++;
}
if (!kni_cnt)
return 0;
rte_kni_init(kni_cnt);
dev_list = rte_malloc(NULL, kni_cnt * sizeof(void *), 0);
memset(&conf, 0, sizeof(conf));
memset(&ops, 0, sizeof(ops));
ops.change_mtu = kni_change_mtu;
ops.config_network_if = kni_config_network_if;
for (opt = s->opt; opt; opt = opt->next) {
const char *busid = conf_get_subopt(opt, "busid");
if (strcmp(busid, "kni"))
continue;
strcpy(conf.name, opt->name);
conf.group_id = i;
conf.mbuf_size = ETHER_MAX_LEN + 8;
ops.port_id = i;
kni = rte_kni_alloc(mbuf_pool, &conf, &ops);
if (!kni) {
fprintf(stderr, "failed to create %s\n", opt->name);
return -1;
}
dev = netdev_alloc(opt->name, sizeof(*knidev), NULL);
dev->xmit = knidev_xmit;
knidev = netdev_priv(dev);
knidev->port = i;
knidev->xport = i + x;
knidev->dev = dev;
knidev->kni = kni;
dev_list[i] = knidev;
arg.dev = knidev;
arg.opt = opt;
arg.err = -1;
pthread_create(&tid, NULL, kni_ifconfig, &arg);
while (arg.err == -1)
rte_kni_handle_request(kni);
pthread_join(tid, NULL);
if (arg.err != 0)
return -1;
}
return 0;
}
