1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2016-2018 Microsoft Corporation
3 * Copyright(c) 2013-2016 Brocade Communications Systems, Inc.
13 #include <rte_ethdev.h>
14 #include <rte_memcpy.h>
15 #include <rte_string_fns.h>
16 #include <rte_memzone.h>
17 #include <rte_devargs.h>
18 #include <rte_malloc.h>
19 #include <rte_kvargs.h>
20 #include <rte_atomic.h>
21 #include <rte_branch_prediction.h>
22 #include <rte_ether.h>
23 #include <rte_ethdev_driver.h>
24 #include <rte_cycles.h>
25 #include <rte_errno.h>
26 #include <rte_memory.h>
29 #include <rte_bus_vmbus.h>
37 #define HN_TX_OFFLOAD_CAPS (DEV_TX_OFFLOAD_IPV4_CKSUM | \
38 DEV_TX_OFFLOAD_TCP_CKSUM | \
39 DEV_TX_OFFLOAD_UDP_CKSUM | \
40 DEV_TX_OFFLOAD_TCP_TSO | \
41 DEV_TX_OFFLOAD_MULTI_SEGS | \
42 DEV_TX_OFFLOAD_VLAN_INSERT)
44 #define HN_RX_OFFLOAD_CAPS (DEV_RX_OFFLOAD_CHECKSUM | \
45 DEV_RX_OFFLOAD_VLAN_STRIP)
48 int hn_logtype_driver;
50 struct hn_xstats_name_off {
51 char name[RTE_ETH_XSTATS_NAME_SIZE];
55 static const struct hn_xstats_name_off hn_stat_strings[] = {
56 { "good_packets", offsetof(struct hn_stats, packets) },
57 { "good_bytes", offsetof(struct hn_stats, bytes) },
58 { "errors", offsetof(struct hn_stats, errors) },
59 { "ring full", offsetof(struct hn_stats, ring_full) },
60 { "multicast_packets", offsetof(struct hn_stats, multicast) },
61 { "broadcast_packets", offsetof(struct hn_stats, broadcast) },
62 { "undersize_packets", offsetof(struct hn_stats, size_bins[0]) },
63 { "size_64_packets", offsetof(struct hn_stats, size_bins[1]) },
64 { "size_65_127_packets", offsetof(struct hn_stats, size_bins[2]) },
65 { "size_128_255_packets", offsetof(struct hn_stats, size_bins[3]) },
66 { "size_256_511_packets", offsetof(struct hn_stats, size_bins[4]) },
67 { "size_512_1023_packets", offsetof(struct hn_stats, size_bins[5]) },
68 { "size_1024_1518_packets", offsetof(struct hn_stats, size_bins[6]) },
69 { "size_1519_max_packets", offsetof(struct hn_stats, size_bins[7]) },
72 static struct rte_eth_dev *
73 eth_dev_vmbus_allocate(struct rte_vmbus_device *dev, size_t private_data_size)
75 struct rte_eth_dev *eth_dev;
81 name = dev->device.name;
83 if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
84 eth_dev = rte_eth_dev_allocate(name);
86 PMD_DRV_LOG(NOTICE, "can not allocate rte ethdev");
90 if (private_data_size) {
91 eth_dev->data->dev_private =
92 rte_zmalloc_socket(name, private_data_size,
93 RTE_CACHE_LINE_SIZE, dev->device.numa_node);
94 if (!eth_dev->data->dev_private) {
95 PMD_DRV_LOG(NOTICE, "can not allocate driver data");
96 rte_eth_dev_release_port(eth_dev);
101 eth_dev = rte_eth_dev_attach_secondary(name);
103 PMD_DRV_LOG(NOTICE, "can not attach secondary");
108 eth_dev->device = &dev->device;
110 /* interrupt is simulated */
111 dev->intr_handle.type = RTE_INTR_HANDLE_EXT;
112 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
113 eth_dev->intr_handle = &dev->intr_handle;
119 eth_dev_vmbus_release(struct rte_eth_dev *eth_dev)
121 /* free ether device */
122 rte_eth_dev_release_port(eth_dev);
124 if (rte_eal_process_type() == RTE_PROC_PRIMARY)
125 rte_free(eth_dev->data->dev_private);
127 eth_dev->data->dev_private = NULL;
130 * Secondary process will check the name to attach.
131 * Clear this field to avoid attaching a released ports.
133 eth_dev->data->name[0] = '\0';
135 eth_dev->device = NULL;
136 eth_dev->intr_handle = NULL;
139 /* handle "latency=X" from devargs */
140 static int hn_set_latency(const char *key, const char *value, void *opaque)
142 struct hn_data *hv = opaque;
147 lat = strtoul(value, &endp, 0);
149 if (*value == '\0' || *endp != '\0') {
150 PMD_DRV_LOG(ERR, "invalid parameter %s=%s", key, value);
154 PMD_DRV_LOG(DEBUG, "set latency %lu usec", lat);
156 hv->latency = lat * 1000; /* usec to nsec */
160 /* Parse device arguments */
161 static int hn_parse_args(const struct rte_eth_dev *dev)
163 struct hn_data *hv = dev->data->dev_private;
164 struct rte_devargs *devargs = dev->device->devargs;
165 static const char * const valid_keys[] = {
169 struct rte_kvargs *kvlist;
174 PMD_INIT_LOG(DEBUG, "device args %s %s",
175 devargs->name, devargs->args);
177 kvlist = rte_kvargs_parse(devargs->args, valid_keys);
179 PMD_DRV_LOG(NOTICE, "invalid parameters");
183 rte_kvargs_process(kvlist, "latency", hn_set_latency, hv);
184 rte_kvargs_free(kvlist);
188 /* Update link status.
189 * Note: the DPDK definition of "wait_to_complete"
190 * means block this call until link is up.
191 * which is not worth supporting.
194 hn_dev_link_update(struct rte_eth_dev *dev,
195 __rte_unused int wait_to_complete)
197 struct hn_data *hv = dev->data->dev_private;
198 struct rte_eth_link link, old;
201 old = dev->data->dev_link;
203 error = hn_rndis_get_linkstatus(hv);
207 hn_rndis_get_linkspeed(hv);
209 link = (struct rte_eth_link) {
210 .link_duplex = ETH_LINK_FULL_DUPLEX,
211 .link_autoneg = ETH_LINK_SPEED_FIXED,
212 .link_speed = hv->link_speed / 10000,
215 if (hv->link_status == NDIS_MEDIA_STATE_CONNECTED)
216 link.link_status = ETH_LINK_UP;
218 link.link_status = ETH_LINK_DOWN;
220 if (old.link_status == link.link_status)
223 PMD_INIT_LOG(DEBUG, "Port %d is %s", dev->data->port_id,
224 (link.link_status == ETH_LINK_UP) ? "up" : "down");
226 return rte_eth_linkstatus_set(dev, &link);
229 static void hn_dev_info_get(struct rte_eth_dev *dev,
230 struct rte_eth_dev_info *dev_info)
232 struct hn_data *hv = dev->data->dev_private;
234 dev_info->speed_capa = ETH_LINK_SPEED_10G;
235 dev_info->min_rx_bufsize = HN_MIN_RX_BUF_SIZE;
236 dev_info->max_rx_pktlen = HN_MAX_XFER_LEN;
237 dev_info->max_mac_addrs = 1;
239 dev_info->hash_key_size = NDIS_HASH_KEYSIZE_TOEPLITZ;
240 dev_info->flow_type_rss_offloads =
241 ETH_RSS_IPV4 | ETH_RSS_IPV6 | ETH_RSS_TCP | ETH_RSS_UDP;
243 dev_info->max_rx_queues = hv->max_queues;
244 dev_info->max_tx_queues = hv->max_queues;
246 hn_rndis_get_offload(hv, dev_info);
250 hn_dev_promiscuous_enable(struct rte_eth_dev *dev)
252 struct hn_data *hv = dev->data->dev_private;
254 hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_PROMISCUOUS);
258 hn_dev_promiscuous_disable(struct rte_eth_dev *dev)
260 struct hn_data *hv = dev->data->dev_private;
263 filter = NDIS_PACKET_TYPE_DIRECTED | NDIS_PACKET_TYPE_BROADCAST;
264 if (dev->data->all_multicast)
265 filter |= NDIS_PACKET_TYPE_ALL_MULTICAST;
266 hn_rndis_set_rxfilter(hv, filter);
270 hn_dev_allmulticast_enable(struct rte_eth_dev *dev)
272 struct hn_data *hv = dev->data->dev_private;
274 hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_DIRECTED |
275 NDIS_PACKET_TYPE_ALL_MULTICAST |
276 NDIS_PACKET_TYPE_BROADCAST);
280 hn_dev_allmulticast_disable(struct rte_eth_dev *dev)
282 struct hn_data *hv = dev->data->dev_private;
284 hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_DIRECTED |
285 NDIS_PACKET_TYPE_BROADCAST);
288 /* Setup shared rx/tx queue data */
289 static int hn_subchan_configure(struct hn_data *hv,
292 struct vmbus_channel *primary = hn_primary_chan(hv);
294 unsigned int retry = 0;
297 "open %u subchannels", subchan);
299 /* Send create sub channels command */
300 err = hn_nvs_alloc_subchans(hv, &subchan);
304 while (subchan > 0) {
305 struct vmbus_channel *new_sc;
308 err = rte_vmbus_subchan_open(primary, &new_sc);
309 if (err == -ENOENT && ++retry < 1000) {
310 /* This can happen if not ready yet */
317 "open subchannel failed: %d", err);
321 rte_vmbus_set_latency(hv->vmbus, new_sc, hv->latency);
324 chn_index = rte_vmbus_sub_channel_index(new_sc);
325 if (chn_index == 0 || chn_index > hv->max_queues) {
327 "Invalid subchannel offermsg channel %u",
332 PMD_DRV_LOG(DEBUG, "new sub channel %u", chn_index);
333 hv->channels[chn_index] = new_sc;
340 static int hn_dev_configure(struct rte_eth_dev *dev)
342 const struct rte_eth_conf *dev_conf = &dev->data->dev_conf;
343 const struct rte_eth_rxmode *rxmode = &dev_conf->rxmode;
344 const struct rte_eth_txmode *txmode = &dev_conf->txmode;
346 const struct rte_eth_rss_conf *rss_conf =
347 &dev_conf->rx_adv_conf.rss_conf;
348 struct hn_data *hv = dev->data->dev_private;
349 uint64_t unsupported;
352 PMD_INIT_FUNC_TRACE();
354 unsupported = txmode->offloads & ~HN_TX_OFFLOAD_CAPS;
357 "unsupported TX offload: %#" PRIx64,
362 unsupported = rxmode->offloads & ~HN_RX_OFFLOAD_CAPS;
365 "unsupported RX offload: %#" PRIx64,
370 err = hn_rndis_conf_offload(hv, txmode->offloads,
374 "offload configure failed");
378 hv->num_queues = RTE_MAX(dev->data->nb_rx_queues,
379 dev->data->nb_tx_queues);
380 subchan = hv->num_queues - 1;
382 err = hn_subchan_configure(hv, subchan);
385 "subchannel configuration failed");
389 err = hn_rndis_conf_rss(hv, rss_conf);
392 "rss configuration failed");
400 static int hn_dev_stats_get(struct rte_eth_dev *dev,
401 struct rte_eth_stats *stats)
405 for (i = 0; i < dev->data->nb_tx_queues; i++) {
406 const struct hn_tx_queue *txq = dev->data->tx_queues[i];
411 stats->opackets += txq->stats.packets;
412 stats->obytes += txq->stats.bytes;
413 stats->oerrors += txq->stats.errors;
415 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
416 stats->q_opackets[i] = txq->stats.packets;
417 stats->q_obytes[i] = txq->stats.bytes;
421 for (i = 0; i < dev->data->nb_rx_queues; i++) {
422 const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
427 stats->ipackets += rxq->stats.packets;
428 stats->ibytes += rxq->stats.bytes;
429 stats->ierrors += rxq->stats.errors;
430 stats->imissed += rxq->stats.ring_full;
432 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
433 stats->q_ipackets[i] = rxq->stats.packets;
434 stats->q_ibytes[i] = rxq->stats.bytes;
438 stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
443 hn_dev_stats_reset(struct rte_eth_dev *dev)
447 PMD_INIT_FUNC_TRACE();
449 for (i = 0; i < dev->data->nb_tx_queues; i++) {
450 struct hn_tx_queue *txq = dev->data->tx_queues[i];
454 memset(&txq->stats, 0, sizeof(struct hn_stats));
457 for (i = 0; i < dev->data->nb_rx_queues; i++) {
458 struct hn_rx_queue *rxq = dev->data->rx_queues[i];
463 memset(&rxq->stats, 0, sizeof(struct hn_stats));
468 hn_dev_xstats_get_names(struct rte_eth_dev *dev,
469 struct rte_eth_xstat_name *xstats_names,
470 __rte_unused unsigned int limit)
472 unsigned int i, t, count = 0;
474 PMD_INIT_FUNC_TRACE();
477 return dev->data->nb_tx_queues * RTE_DIM(hn_stat_strings)
478 + dev->data->nb_rx_queues * RTE_DIM(hn_stat_strings);
480 /* Note: limit checked in rte_eth_xstats_names() */
481 for (i = 0; i < dev->data->nb_tx_queues; i++) {
482 const struct hn_tx_queue *txq = dev->data->tx_queues[i];
487 for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
488 snprintf(xstats_names[count++].name,
489 RTE_ETH_XSTATS_NAME_SIZE,
490 "tx_q%u_%s", i, hn_stat_strings[t].name);
493 for (i = 0; i < dev->data->nb_rx_queues; i++) {
494 const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
499 for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
500 snprintf(xstats_names[count++].name,
501 RTE_ETH_XSTATS_NAME_SIZE,
503 hn_stat_strings[t].name);
510 hn_dev_xstats_get(struct rte_eth_dev *dev,
511 struct rte_eth_xstat *xstats,
514 unsigned int i, t, count = 0;
516 const unsigned int nstats =
517 dev->data->nb_tx_queues * RTE_DIM(hn_stat_strings)
518 + dev->data->nb_rx_queues * RTE_DIM(hn_stat_strings);
521 PMD_INIT_FUNC_TRACE();
526 for (i = 0; i < dev->data->nb_tx_queues; i++) {
527 const struct hn_tx_queue *txq = dev->data->tx_queues[i];
532 stats = (const char *)&txq->stats;
533 for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
534 xstats[count++].value = *(const uint64_t *)
535 (stats + hn_stat_strings[t].offset);
538 for (i = 0; i < dev->data->nb_rx_queues; i++) {
539 const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
544 stats = (const char *)&rxq->stats;
545 for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
546 xstats[count++].value = *(const uint64_t *)
547 (stats + hn_stat_strings[t].offset);
554 hn_dev_start(struct rte_eth_dev *dev)
556 struct hn_data *hv = dev->data->dev_private;
558 PMD_INIT_FUNC_TRACE();
560 return hn_rndis_set_rxfilter(hv,
561 NDIS_PACKET_TYPE_BROADCAST |
562 NDIS_PACKET_TYPE_ALL_MULTICAST |
563 NDIS_PACKET_TYPE_DIRECTED);
567 hn_dev_stop(struct rte_eth_dev *dev)
569 struct hn_data *hv = dev->data->dev_private;
571 PMD_INIT_FUNC_TRACE();
573 hn_rndis_set_rxfilter(hv, 0);
577 hn_dev_close(struct rte_eth_dev *dev __rte_unused)
579 PMD_INIT_LOG(DEBUG, "close");
582 static const struct eth_dev_ops hn_eth_dev_ops = {
583 .dev_configure = hn_dev_configure,
584 .dev_start = hn_dev_start,
585 .dev_stop = hn_dev_stop,
586 .dev_close = hn_dev_close,
587 .dev_infos_get = hn_dev_info_get,
588 .txq_info_get = hn_dev_tx_queue_info,
589 .rxq_info_get = hn_dev_rx_queue_info,
590 .promiscuous_enable = hn_dev_promiscuous_enable,
591 .promiscuous_disable = hn_dev_promiscuous_disable,
592 .allmulticast_enable = hn_dev_allmulticast_enable,
593 .allmulticast_disable = hn_dev_allmulticast_disable,
594 .tx_queue_setup = hn_dev_tx_queue_setup,
595 .tx_queue_release = hn_dev_tx_queue_release,
596 .tx_done_cleanup = hn_dev_tx_done_cleanup,
597 .rx_queue_setup = hn_dev_rx_queue_setup,
598 .rx_queue_release = hn_dev_rx_queue_release,
599 .link_update = hn_dev_link_update,
600 .stats_get = hn_dev_stats_get,
601 .xstats_get = hn_dev_xstats_get,
602 .xstats_get_names = hn_dev_xstats_get_names,
603 .stats_reset = hn_dev_stats_reset,
604 .xstats_reset = hn_dev_stats_reset,
608 * Setup connection between PMD and kernel.
611 hn_attach(struct hn_data *hv, unsigned int mtu)
616 error = hn_nvs_attach(hv, mtu);
621 error = hn_rndis_attach(hv);
627 * Under certain conditions on certain versions of Hyper-V,
628 * the RNDIS rxfilter is _not_ zero on the hypervisor side
629 * after the successful RNDIS initialization.
631 hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_NONE);
640 hn_detach(struct hn_data *hv)
647 eth_hn_dev_init(struct rte_eth_dev *eth_dev)
649 struct hn_data *hv = eth_dev->data->dev_private;
650 struct rte_device *device = eth_dev->device;
651 struct rte_vmbus_device *vmbus;
652 unsigned int rxr_cnt;
655 PMD_INIT_FUNC_TRACE();
657 vmbus = container_of(device, struct rte_vmbus_device, device);
658 eth_dev->dev_ops = &hn_eth_dev_ops;
659 eth_dev->tx_pkt_burst = &hn_xmit_pkts;
660 eth_dev->rx_pkt_burst = &hn_recv_pkts;
663 * for secondary processes, we don't initialize any further as primary
664 * has already done this work.
666 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
669 /* Since Hyper-V only supports one MAC address, just use local data */
670 eth_dev->data->mac_addrs = &hv->mac_addr;
673 hv->rxbuf_res = &vmbus->resource[HV_RECV_BUF_MAP];
674 hv->chim_res = &vmbus->resource[HV_SEND_BUF_MAP];
675 hv->port_id = eth_dev->data->port_id;
676 hv->latency = HN_CHAN_LATENCY_NS;
678 err = hn_parse_args(eth_dev);
682 /* Initialize primary channel input for control operations */
683 err = rte_vmbus_chan_open(vmbus, &hv->channels[0]);
687 rte_vmbus_set_latency(hv->vmbus, hv->channels[0], hv->latency);
689 hv->primary = hn_rx_queue_alloc(hv, 0,
690 eth_dev->device->numa_node);
695 err = hn_attach(hv, ETHER_MTU);
699 err = hn_tx_pool_init(eth_dev);
703 err = hn_rndis_get_eaddr(hv, hv->mac_addr.addr_bytes);
707 max_chan = rte_vmbus_max_channels(vmbus);
708 PMD_INIT_LOG(DEBUG, "VMBus max channels %d", max_chan);
712 if (hn_rndis_query_rsscaps(hv, &rxr_cnt) != 0)
715 hv->max_queues = RTE_MIN(rxr_cnt, (unsigned int)max_chan);
720 PMD_INIT_LOG(NOTICE, "device init failed");
727 eth_hn_dev_uninit(struct rte_eth_dev *eth_dev)
729 struct hn_data *hv = eth_dev->data->dev_private;
731 PMD_INIT_FUNC_TRACE();
733 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
736 hn_dev_stop(eth_dev);
737 hn_dev_close(eth_dev);
739 eth_dev->dev_ops = NULL;
740 eth_dev->tx_pkt_burst = NULL;
741 eth_dev->rx_pkt_burst = NULL;
744 rte_vmbus_chan_close(hv->primary->chan);
745 rte_free(hv->primary);
747 eth_dev->data->mac_addrs = NULL;
752 static int eth_hn_probe(struct rte_vmbus_driver *drv __rte_unused,
753 struct rte_vmbus_device *dev)
755 struct rte_eth_dev *eth_dev;
758 PMD_INIT_FUNC_TRACE();
760 eth_dev = eth_dev_vmbus_allocate(dev, sizeof(struct hn_data));
764 ret = eth_hn_dev_init(eth_dev);
766 eth_dev_vmbus_release(eth_dev);
768 rte_eth_dev_probing_finish(eth_dev);
773 static int eth_hn_remove(struct rte_vmbus_device *dev)
775 struct rte_eth_dev *eth_dev;
778 PMD_INIT_FUNC_TRACE();
780 eth_dev = rte_eth_dev_allocated(dev->device.name);
784 ret = eth_hn_dev_uninit(eth_dev);
788 eth_dev_vmbus_release(eth_dev);
792 /* Network device GUID */
793 static const rte_uuid_t hn_net_ids[] = {
794 /* f8615163-df3e-46c5-913f-f2d2f965ed0e */
795 RTE_UUID_INIT(0xf8615163, 0xdf3e, 0x46c5, 0x913f, 0xf2d2f965ed0eULL),
799 static struct rte_vmbus_driver rte_netvsc_pmd = {
800 .id_table = hn_net_ids,
801 .probe = eth_hn_probe,
802 .remove = eth_hn_remove,
805 RTE_PMD_REGISTER_VMBUS(net_netvsc, rte_netvsc_pmd);
806 RTE_PMD_REGISTER_KMOD_DEP(net_netvsc, "* uio_hv_generic");
808 RTE_INIT(hn_init_log);
812 hn_logtype_init = rte_log_register("pmd.net.netvsc.init");
813 if (hn_logtype_init >= 0)
814 rte_log_set_level(hn_logtype_init, RTE_LOG_NOTICE);
815 hn_logtype_driver = rte_log_register("pmd.net.netvsc.driver");
816 if (hn_logtype_driver >= 0)
817 rte_log_set_level(hn_logtype_driver, RTE_LOG_NOTICE);