2 * Copyright (c) 2009, Microsoft Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, see <http://www.gnu.org/licenses/>.
17 * Haiyang Zhang <haiyangz@microsoft.com>
18 * Hank Janssen <hjanssen@microsoft.com>
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
22 #include <linux/init.h>
23 #include <linux/atomic.h>
24 #include <linux/module.h>
25 #include <linux/highmem.h>
26 #include <linux/device.h>
28 #include <linux/delay.h>
29 #include <linux/netdevice.h>
30 #include <linux/inetdevice.h>
31 #include <linux/etherdevice.h>
32 #include <linux/skbuff.h>
33 #include <linux/if_vlan.h>
35 #include <linux/slab.h>
37 #include <net/route.h>
39 #include <net/pkt_sched.h>
41 #include "hyperv_net.h"
43 #define RING_SIZE_MIN 64
44 #define LINKCHANGE_INT (2 * HZ)
45 #define NETVSC_HW_FEATURES (NETIF_F_RXCSUM | \
51 /* Restrict GSO size to account for NVGRE */
52 #define NETVSC_GSO_MAX_SIZE 62768
54 static int ring_size = 128;
55 module_param(ring_size, int, S_IRUGO);
56 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
58 static int max_num_vrss_chns = 8;
60 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
61 NETIF_MSG_LINK | NETIF_MSG_IFUP |
62 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
65 static int debug = -1;
66 module_param(debug, int, S_IRUGO);
67 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
69 static void do_set_multicast(struct work_struct *w)
71 struct net_device_context *ndevctx =
72 container_of(w, struct net_device_context, work);
73 struct hv_device *device_obj = ndevctx->device_ctx;
74 struct net_device *ndev = hv_get_drvdata(device_obj);
75 struct netvsc_device *nvdev = ndevctx->nvdev;
76 struct rndis_device *rdev;
81 rdev = nvdev->extension;
85 if (ndev->flags & IFF_PROMISC)
86 rndis_filter_set_packet_filter(rdev,
87 NDIS_PACKET_TYPE_PROMISCUOUS);
89 rndis_filter_set_packet_filter(rdev,
90 NDIS_PACKET_TYPE_BROADCAST |
91 NDIS_PACKET_TYPE_ALL_MULTICAST |
92 NDIS_PACKET_TYPE_DIRECTED);
95 static void netvsc_set_multicast_list(struct net_device *net)
97 struct net_device_context *net_device_ctx = netdev_priv(net);
99 schedule_work(&net_device_ctx->work);
102 static int netvsc_open(struct net_device *net)
104 struct netvsc_device *nvdev = net_device_to_netvsc_device(net);
105 struct rndis_device *rdev;
108 netif_carrier_off(net);
110 /* Open up the device */
111 ret = rndis_filter_open(nvdev);
113 netdev_err(net, "unable to open device (ret %d).\n", ret);
117 netif_tx_wake_all_queues(net);
119 rdev = nvdev->extension;
120 if (!rdev->link_state)
121 netif_carrier_on(net);
126 static int netvsc_close(struct net_device *net)
128 struct net_device_context *net_device_ctx = netdev_priv(net);
129 struct netvsc_device *nvdev = net_device_ctx->nvdev;
131 u32 aread, awrite, i, msec = 10, retry = 0, retry_max = 20;
132 struct vmbus_channel *chn;
134 netif_tx_disable(net);
136 /* Make sure netvsc_set_multicast_list doesn't re-enable filter! */
137 cancel_work_sync(&net_device_ctx->work);
138 ret = rndis_filter_close(nvdev);
140 netdev_err(net, "unable to close device (ret %d).\n", ret);
144 /* Ensure pending bytes in ring are read */
147 for (i = 0; i < nvdev->num_chn; i++) {
148 chn = nvdev->chn_table[i];
152 hv_get_ringbuffer_availbytes(&chn->inbound, &aread,
158 hv_get_ringbuffer_availbytes(&chn->outbound, &aread,
166 if (retry > retry_max || aread == 0)
176 netdev_err(net, "Ring buffer not empty after closing rndis\n");
183 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
186 struct rndis_packet *rndis_pkt;
187 struct rndis_per_packet_info *ppi;
189 rndis_pkt = &msg->msg.pkt;
190 rndis_pkt->data_offset += ppi_size;
192 ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
193 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
195 ppi->size = ppi_size;
196 ppi->type = pkt_type;
197 ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
199 rndis_pkt->per_pkt_info_len += ppi_size;
204 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
205 void *accel_priv, select_queue_fallback_t fallback)
207 struct net_device_context *net_device_ctx = netdev_priv(ndev);
208 struct netvsc_device *nvsc_dev = net_device_ctx->nvdev;
212 if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1)
215 hash = skb_get_hash(skb);
216 q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] %
217 ndev->real_num_tx_queues;
219 if (!nvsc_dev->chn_table[q_idx])
225 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
226 struct hv_page_buffer *pb)
230 /* Deal with compund pages by ignoring unused part
233 page += (offset >> PAGE_SHIFT);
234 offset &= ~PAGE_MASK;
239 bytes = PAGE_SIZE - offset;
242 pb[j].pfn = page_to_pfn(page);
243 pb[j].offset = offset;
249 if (offset == PAGE_SIZE && len) {
259 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
260 struct hv_netvsc_packet *packet,
261 struct hv_page_buffer **page_buf)
263 struct hv_page_buffer *pb = *page_buf;
265 char *data = skb->data;
266 int frags = skb_shinfo(skb)->nr_frags;
269 /* The packet is laid out thus:
270 * 1. hdr: RNDIS header and PPI
272 * 3. skb fragment data
275 slots_used += fill_pg_buf(virt_to_page(hdr),
277 len, &pb[slots_used]);
279 packet->rmsg_size = len;
280 packet->rmsg_pgcnt = slots_used;
282 slots_used += fill_pg_buf(virt_to_page(data),
283 offset_in_page(data),
284 skb_headlen(skb), &pb[slots_used]);
286 for (i = 0; i < frags; i++) {
287 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
289 slots_used += fill_pg_buf(skb_frag_page(frag),
291 skb_frag_size(frag), &pb[slots_used]);
296 static int count_skb_frag_slots(struct sk_buff *skb)
298 int i, frags = skb_shinfo(skb)->nr_frags;
301 for (i = 0; i < frags; i++) {
302 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
303 unsigned long size = skb_frag_size(frag);
304 unsigned long offset = frag->page_offset;
306 /* Skip unused frames from start of page */
307 offset &= ~PAGE_MASK;
308 pages += PFN_UP(offset + size);
313 static int netvsc_get_slots(struct sk_buff *skb)
315 char *data = skb->data;
316 unsigned int offset = offset_in_page(data);
317 unsigned int len = skb_headlen(skb);
321 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
322 frag_slots = count_skb_frag_slots(skb);
323 return slots + frag_slots;
326 static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
328 u32 ret_val = TRANSPORT_INFO_NOT_IP;
330 if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
331 (eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
335 *trans_off = skb_transport_offset(skb);
337 if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
338 struct iphdr *iphdr = ip_hdr(skb);
340 if (iphdr->protocol == IPPROTO_TCP)
341 ret_val = TRANSPORT_INFO_IPV4_TCP;
342 else if (iphdr->protocol == IPPROTO_UDP)
343 ret_val = TRANSPORT_INFO_IPV4_UDP;
345 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
346 ret_val = TRANSPORT_INFO_IPV6_TCP;
347 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
348 ret_val = TRANSPORT_INFO_IPV6_UDP;
355 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
357 struct net_device_context *net_device_ctx = netdev_priv(net);
358 struct hv_netvsc_packet *packet = NULL;
360 unsigned int num_data_pgs;
361 struct rndis_message *rndis_msg;
362 struct rndis_packet *rndis_pkt;
364 struct rndis_per_packet_info *ppi;
365 struct ndis_tcp_ip_checksum_info *csum_info;
370 struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
371 struct hv_page_buffer *pb = page_buf;
373 /* We will atmost need two pages to describe the rndis
374 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
375 * of pages in a single packet. If skb is scattered around
376 * more pages we try linearizing it.
379 skb_length = skb->len;
380 num_data_pgs = netvsc_get_slots(skb) + 2;
382 if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
383 ++net_device_ctx->eth_stats.tx_scattered;
385 if (skb_linearize(skb))
388 num_data_pgs = netvsc_get_slots(skb) + 2;
389 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
390 ++net_device_ctx->eth_stats.tx_too_big;
396 * Place the rndis header in the skb head room and
397 * the skb->cb will be used for hv_netvsc_packet
400 ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
404 /* Use the skb control buffer for building up the packet */
405 BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
406 FIELD_SIZEOF(struct sk_buff, cb));
407 packet = (struct hv_netvsc_packet *)skb->cb;
409 packet->q_idx = skb_get_queue_mapping(skb);
411 packet->total_data_buflen = skb->len;
413 rndis_msg = (struct rndis_message *)skb->head;
415 memset(rndis_msg, 0, RNDIS_AND_PPI_SIZE);
417 /* Add the rndis header */
418 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
419 rndis_msg->msg_len = packet->total_data_buflen;
420 rndis_pkt = &rndis_msg->msg.pkt;
421 rndis_pkt->data_offset = sizeof(struct rndis_packet);
422 rndis_pkt->data_len = packet->total_data_buflen;
423 rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
425 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
427 hash = skb_get_hash_raw(skb);
428 if (hash != 0 && net->real_num_tx_queues > 1) {
429 rndis_msg_size += NDIS_HASH_PPI_SIZE;
430 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
432 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
435 if (skb_vlan_tag_present(skb)) {
436 struct ndis_pkt_8021q_info *vlan;
438 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
439 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
441 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
443 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
444 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
448 net_trans_info = get_net_transport_info(skb, &hdr_offset);
451 * Setup the sendside checksum offload only if this is not a
454 if ((net_trans_info & (INFO_TCP | INFO_UDP)) && skb_is_gso(skb)) {
455 struct ndis_tcp_lso_info *lso_info;
457 rndis_msg_size += NDIS_LSO_PPI_SIZE;
458 ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
459 TCP_LARGESEND_PKTINFO);
461 lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
464 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
465 if (net_trans_info & (INFO_IPV4 << 16)) {
466 lso_info->lso_v2_transmit.ip_version =
467 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
468 ip_hdr(skb)->tot_len = 0;
469 ip_hdr(skb)->check = 0;
470 tcp_hdr(skb)->check =
471 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
472 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
474 lso_info->lso_v2_transmit.ip_version =
475 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
476 ipv6_hdr(skb)->payload_len = 0;
477 tcp_hdr(skb)->check =
478 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
479 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
481 lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
482 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
483 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
484 if (net_trans_info & INFO_TCP) {
485 rndis_msg_size += NDIS_CSUM_PPI_SIZE;
486 ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
487 TCPIP_CHKSUM_PKTINFO);
489 csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
492 if (net_trans_info & (INFO_IPV4 << 16))
493 csum_info->transmit.is_ipv4 = 1;
495 csum_info->transmit.is_ipv6 = 1;
497 csum_info->transmit.tcp_checksum = 1;
498 csum_info->transmit.tcp_header_offset = hdr_offset;
500 /* UDP checksum (and other) offload is not supported. */
501 if (skb_checksum_help(skb))
506 /* Start filling in the page buffers with the rndis hdr */
507 rndis_msg->msg_len += rndis_msg_size;
508 packet->total_data_buflen = rndis_msg->msg_len;
509 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
512 /* timestamp packet in software */
513 skb_tx_timestamp(skb);
514 ret = netvsc_send(net_device_ctx->device_ctx, packet,
515 rndis_msg, &pb, skb);
516 if (likely(ret == 0)) {
517 struct netvsc_stats *tx_stats = this_cpu_ptr(net_device_ctx->tx_stats);
519 u64_stats_update_begin(&tx_stats->syncp);
521 tx_stats->bytes += skb_length;
522 u64_stats_update_end(&tx_stats->syncp);
526 if (ret == -EAGAIN) {
527 ++net_device_ctx->eth_stats.tx_busy;
528 return NETDEV_TX_BUSY;
532 ++net_device_ctx->eth_stats.tx_no_space;
535 dev_kfree_skb_any(skb);
536 net->stats.tx_dropped++;
541 ++net_device_ctx->eth_stats.tx_no_memory;
546 * netvsc_linkstatus_callback - Link up/down notification
548 void netvsc_linkstatus_callback(struct hv_device *device_obj,
549 struct rndis_message *resp)
551 struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
552 struct net_device *net;
553 struct net_device_context *ndev_ctx;
554 struct netvsc_reconfig *event;
557 net = hv_get_drvdata(device_obj);
562 ndev_ctx = netdev_priv(net);
564 /* Update the physical link speed when changing to another vSwitch */
565 if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
568 speed = *(u32 *)((void *)indicate + indicate->
569 status_buf_offset) / 10000;
570 ndev_ctx->speed = speed;
574 /* Handle these link change statuses below */
575 if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
576 indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
577 indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
580 if (net->reg_state != NETREG_REGISTERED)
583 event = kzalloc(sizeof(*event), GFP_ATOMIC);
586 event->event = indicate->status;
588 spin_lock_irqsave(&ndev_ctx->lock, flags);
589 list_add_tail(&event->list, &ndev_ctx->reconfig_events);
590 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
592 schedule_delayed_work(&ndev_ctx->dwork, 0);
595 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
596 struct hv_netvsc_packet *packet,
597 struct ndis_tcp_ip_checksum_info *csum_info,
598 void *data, u16 vlan_tci)
602 skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
607 * Copy to skb. This copy is needed here since the memory pointed by
608 * hv_netvsc_packet cannot be deallocated
610 memcpy(skb_put(skb, packet->total_data_buflen), data,
611 packet->total_data_buflen);
613 skb->protocol = eth_type_trans(skb, net);
615 /* skb is already created with CHECKSUM_NONE */
616 skb_checksum_none_assert(skb);
619 * In Linux, the IP checksum is always checked.
620 * Do L4 checksum offload if enabled and present.
622 if (csum_info && (net->features & NETIF_F_RXCSUM)) {
623 if (csum_info->receive.tcp_checksum_succeeded ||
624 csum_info->receive.udp_checksum_succeeded)
625 skb->ip_summed = CHECKSUM_UNNECESSARY;
628 if (vlan_tci & VLAN_TAG_PRESENT)
629 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
636 * netvsc_recv_callback - Callback when we receive a packet from the
637 * "wire" on the specified device.
639 int netvsc_recv_callback(struct hv_device *device_obj,
640 struct hv_netvsc_packet *packet,
642 struct ndis_tcp_ip_checksum_info *csum_info,
643 struct vmbus_channel *channel,
646 struct net_device *net = hv_get_drvdata(device_obj);
647 struct net_device_context *net_device_ctx = netdev_priv(net);
648 struct net_device *vf_netdev;
650 struct netvsc_stats *rx_stats;
652 if (net->reg_state != NETREG_REGISTERED)
653 return NVSP_STAT_FAIL;
656 * If necessary, inject this packet into the VF interface.
657 * On Hyper-V, multicast and brodcast packets are only delivered
658 * to the synthetic interface (after subjecting these to
659 * policy filters on the host). Deliver these via the VF
660 * interface in the guest.
662 vf_netdev = rcu_dereference(net_device_ctx->vf_netdev);
663 if (vf_netdev && (vf_netdev->flags & IFF_UP))
666 /* Allocate a skb - TODO direct I/O to pages? */
667 skb = netvsc_alloc_recv_skb(net, packet, csum_info, *data, vlan_tci);
668 if (unlikely(!skb)) {
669 ++net->stats.rx_dropped;
670 return NVSP_STAT_FAIL;
673 if (net != vf_netdev)
674 skb_record_rx_queue(skb,
675 channel->offermsg.offer.sub_channel_index);
678 * Even if injecting the packet, record the statistics
679 * on the synthetic device because modifying the VF device
680 * statistics will not work correctly.
682 rx_stats = this_cpu_ptr(net_device_ctx->rx_stats);
683 u64_stats_update_begin(&rx_stats->syncp);
685 rx_stats->bytes += packet->total_data_buflen;
687 if (skb->pkt_type == PACKET_BROADCAST)
688 ++rx_stats->broadcast;
689 else if (skb->pkt_type == PACKET_MULTICAST)
690 ++rx_stats->multicast;
691 u64_stats_update_end(&rx_stats->syncp);
694 * Pass the skb back up. Network stack will deallocate the skb when it
703 static void netvsc_get_drvinfo(struct net_device *net,
704 struct ethtool_drvinfo *info)
706 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
707 strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
710 static void netvsc_get_channels(struct net_device *net,
711 struct ethtool_channels *channel)
713 struct net_device_context *net_device_ctx = netdev_priv(net);
714 struct netvsc_device *nvdev = net_device_ctx->nvdev;
717 channel->max_combined = nvdev->max_chn;
718 channel->combined_count = nvdev->num_chn;
722 static int netvsc_set_channels(struct net_device *net,
723 struct ethtool_channels *channels)
725 struct net_device_context *net_device_ctx = netdev_priv(net);
726 struct hv_device *dev = net_device_ctx->device_ctx;
727 struct netvsc_device *nvdev = net_device_ctx->nvdev;
728 struct netvsc_device_info device_info;
732 bool recovering = false;
734 if (net_device_ctx->start_remove || !nvdev || nvdev->destroy)
737 num_chn = nvdev->num_chn;
738 max_chn = min_t(u32, nvdev->max_chn, num_online_cpus());
740 if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5) {
741 pr_info("vRSS unsupported before NVSP Version 5\n");
745 /* We do not support rx, tx, or other */
747 channels->rx_count ||
748 channels->tx_count ||
749 channels->other_count ||
750 (channels->combined_count < 1))
753 if (channels->combined_count > max_chn) {
754 pr_info("combined channels too high, using %d\n", max_chn);
755 channels->combined_count = max_chn;
758 ret = netvsc_close(net);
763 net_device_ctx->start_remove = true;
764 rndis_filter_device_remove(dev);
766 nvdev->num_chn = channels->combined_count;
768 memset(&device_info, 0, sizeof(device_info));
769 device_info.num_chn = nvdev->num_chn; /* passed to RNDIS */
770 device_info.ring_size = ring_size;
771 device_info.max_num_vrss_chns = max_num_vrss_chns;
773 ret = rndis_filter_device_add(dev, &device_info);
776 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
782 nvdev = net_device_ctx->nvdev;
784 ret = netif_set_real_num_tx_queues(net, nvdev->num_chn);
787 netdev_err(net, "could not set tx queue count (ret %d)\n", ret);
793 ret = netif_set_real_num_rx_queues(net, nvdev->num_chn);
796 netdev_err(net, "could not set rx queue count (ret %d)\n", ret);
804 net_device_ctx->start_remove = false;
805 /* We may have missed link change notifications */
806 schedule_delayed_work(&net_device_ctx->dwork, 0);
811 /* If the above failed, we attempt to recover through the same
812 * process but with the original number of channels.
814 netdev_err(net, "could not set channels, recovering\n");
816 channels->combined_count = num_chn;
820 static bool netvsc_validate_ethtool_ss_cmd(const struct ethtool_cmd *cmd)
822 struct ethtool_cmd diff1 = *cmd;
823 struct ethtool_cmd diff2 = {};
825 ethtool_cmd_speed_set(&diff1, 0);
827 /* advertising and cmd are usually set */
828 diff1.advertising = 0;
830 /* We set port to PORT_OTHER */
831 diff2.port = PORT_OTHER;
833 return !memcmp(&diff1, &diff2, sizeof(diff1));
836 static void netvsc_init_settings(struct net_device *dev)
838 struct net_device_context *ndc = netdev_priv(dev);
840 ndc->speed = SPEED_UNKNOWN;
841 ndc->duplex = DUPLEX_UNKNOWN;
844 static int netvsc_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
846 struct net_device_context *ndc = netdev_priv(dev);
848 ethtool_cmd_speed_set(cmd, ndc->speed);
849 cmd->duplex = ndc->duplex;
850 cmd->port = PORT_OTHER;
855 static int netvsc_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
857 struct net_device_context *ndc = netdev_priv(dev);
860 speed = ethtool_cmd_speed(cmd);
861 if (!ethtool_validate_speed(speed) ||
862 !ethtool_validate_duplex(cmd->duplex) ||
863 !netvsc_validate_ethtool_ss_cmd(cmd))
867 ndc->duplex = cmd->duplex;
872 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
874 struct net_device_context *ndevctx = netdev_priv(ndev);
875 struct netvsc_device *nvdev = ndevctx->nvdev;
876 struct hv_device *hdev = ndevctx->device_ctx;
877 struct netvsc_device_info device_info;
881 if (ndevctx->start_remove || !nvdev || nvdev->destroy)
884 ret = netvsc_close(ndev);
888 num_chn = nvdev->num_chn;
890 ndevctx->start_remove = true;
891 rndis_filter_device_remove(hdev);
895 memset(&device_info, 0, sizeof(device_info));
896 device_info.ring_size = ring_size;
897 device_info.num_chn = num_chn;
898 device_info.max_num_vrss_chns = max_num_vrss_chns;
899 rndis_filter_device_add(hdev, &device_info);
903 ndevctx->start_remove = false;
905 /* We may have missed link change notifications */
906 schedule_delayed_work(&ndevctx->dwork, 0);
911 static struct rtnl_link_stats64 *netvsc_get_stats64(struct net_device *net,
912 struct rtnl_link_stats64 *t)
914 struct net_device_context *ndev_ctx = netdev_priv(net);
917 for_each_possible_cpu(cpu) {
918 struct netvsc_stats *tx_stats = per_cpu_ptr(ndev_ctx->tx_stats,
920 struct netvsc_stats *rx_stats = per_cpu_ptr(ndev_ctx->rx_stats,
922 u64 tx_packets, tx_bytes, rx_packets, rx_bytes, rx_multicast;
926 start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
927 tx_packets = tx_stats->packets;
928 tx_bytes = tx_stats->bytes;
929 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
932 start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
933 rx_packets = rx_stats->packets;
934 rx_bytes = rx_stats->bytes;
935 rx_multicast = rx_stats->multicast + rx_stats->broadcast;
936 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
938 t->tx_bytes += tx_bytes;
939 t->tx_packets += tx_packets;
940 t->rx_bytes += rx_bytes;
941 t->rx_packets += rx_packets;
942 t->multicast += rx_multicast;
945 t->tx_dropped = net->stats.tx_dropped;
946 t->tx_errors = net->stats.tx_dropped;
948 t->rx_dropped = net->stats.rx_dropped;
949 t->rx_errors = net->stats.rx_errors;
954 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
956 struct sockaddr *addr = p;
957 char save_adr[ETH_ALEN];
958 unsigned char save_aatype;
961 memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
962 save_aatype = ndev->addr_assign_type;
964 err = eth_mac_addr(ndev, p);
968 err = rndis_filter_set_device_mac(ndev, addr->sa_data);
970 /* roll back to saved MAC */
971 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
972 ndev->addr_assign_type = save_aatype;
978 static const struct {
979 char name[ETH_GSTRING_LEN];
982 { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
983 { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
984 { "tx_no_space", offsetof(struct netvsc_ethtool_stats, tx_no_space) },
985 { "tx_too_big", offsetof(struct netvsc_ethtool_stats, tx_too_big) },
986 { "tx_busy", offsetof(struct netvsc_ethtool_stats, tx_busy) },
989 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
991 switch (string_set) {
993 return ARRAY_SIZE(netvsc_stats);
999 static void netvsc_get_ethtool_stats(struct net_device *dev,
1000 struct ethtool_stats *stats, u64 *data)
1002 struct net_device_context *ndc = netdev_priv(dev);
1003 const void *nds = &ndc->eth_stats;
1006 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1007 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1010 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1014 switch (stringset) {
1016 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1017 memcpy(data + i * ETH_GSTRING_LEN,
1018 netvsc_stats[i].name, ETH_GSTRING_LEN);
1023 #ifdef CONFIG_NET_POLL_CONTROLLER
1024 static void netvsc_poll_controller(struct net_device *net)
1026 /* As netvsc_start_xmit() works synchronous we don't have to
1027 * trigger anything here.
1032 static const struct ethtool_ops ethtool_ops = {
1033 .get_drvinfo = netvsc_get_drvinfo,
1034 .get_link = ethtool_op_get_link,
1035 .get_ethtool_stats = netvsc_get_ethtool_stats,
1036 .get_sset_count = netvsc_get_sset_count,
1037 .get_strings = netvsc_get_strings,
1038 .get_channels = netvsc_get_channels,
1039 .set_channels = netvsc_set_channels,
1040 .get_ts_info = ethtool_op_get_ts_info,
1041 .get_settings = netvsc_get_settings,
1042 .set_settings = netvsc_set_settings,
1045 static const struct net_device_ops device_ops = {
1046 .ndo_open = netvsc_open,
1047 .ndo_stop = netvsc_close,
1048 .ndo_start_xmit = netvsc_start_xmit,
1049 .ndo_set_rx_mode = netvsc_set_multicast_list,
1050 .ndo_change_mtu = netvsc_change_mtu,
1051 .ndo_validate_addr = eth_validate_addr,
1052 .ndo_set_mac_address = netvsc_set_mac_addr,
1053 .ndo_select_queue = netvsc_select_queue,
1054 .ndo_get_stats64 = netvsc_get_stats64,
1055 #ifdef CONFIG_NET_POLL_CONTROLLER
1056 .ndo_poll_controller = netvsc_poll_controller,
1061 * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1062 * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1063 * present send GARP packet to network peers with netif_notify_peers().
1065 static void netvsc_link_change(struct work_struct *w)
1067 struct net_device_context *ndev_ctx =
1068 container_of(w, struct net_device_context, dwork.work);
1069 struct hv_device *device_obj = ndev_ctx->device_ctx;
1070 struct net_device *net = hv_get_drvdata(device_obj);
1071 struct netvsc_device *net_device;
1072 struct rndis_device *rdev;
1073 struct netvsc_reconfig *event = NULL;
1074 bool notify = false, reschedule = false;
1075 unsigned long flags, next_reconfig, delay;
1078 if (ndev_ctx->start_remove)
1081 net_device = ndev_ctx->nvdev;
1082 rdev = net_device->extension;
1084 next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1085 if (time_is_after_jiffies(next_reconfig)) {
1086 /* link_watch only sends one notification with current state
1087 * per second, avoid doing reconfig more frequently. Handle
1090 delay = next_reconfig - jiffies;
1091 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1092 schedule_delayed_work(&ndev_ctx->dwork, delay);
1095 ndev_ctx->last_reconfig = jiffies;
1097 spin_lock_irqsave(&ndev_ctx->lock, flags);
1098 if (!list_empty(&ndev_ctx->reconfig_events)) {
1099 event = list_first_entry(&ndev_ctx->reconfig_events,
1100 struct netvsc_reconfig, list);
1101 list_del(&event->list);
1102 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1104 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1109 switch (event->event) {
1110 /* Only the following events are possible due to the check in
1111 * netvsc_linkstatus_callback()
1113 case RNDIS_STATUS_MEDIA_CONNECT:
1114 if (rdev->link_state) {
1115 rdev->link_state = false;
1116 netif_carrier_on(net);
1117 netif_tx_wake_all_queues(net);
1123 case RNDIS_STATUS_MEDIA_DISCONNECT:
1124 if (!rdev->link_state) {
1125 rdev->link_state = true;
1126 netif_carrier_off(net);
1127 netif_tx_stop_all_queues(net);
1131 case RNDIS_STATUS_NETWORK_CHANGE:
1132 /* Only makes sense if carrier is present */
1133 if (!rdev->link_state) {
1134 rdev->link_state = true;
1135 netif_carrier_off(net);
1136 netif_tx_stop_all_queues(net);
1137 event->event = RNDIS_STATUS_MEDIA_CONNECT;
1138 spin_lock_irqsave(&ndev_ctx->lock, flags);
1139 list_add(&event->list, &ndev_ctx->reconfig_events);
1140 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1149 netdev_notify_peers(net);
1151 /* link_watch only sends one notification with current state per
1152 * second, handle next reconfig event in 2 seconds.
1155 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1163 static void netvsc_free_netdev(struct net_device *netdev)
1165 struct net_device_context *net_device_ctx = netdev_priv(netdev);
1167 free_percpu(net_device_ctx->tx_stats);
1168 free_percpu(net_device_ctx->rx_stats);
1169 free_netdev(netdev);
1172 static struct net_device *get_netvsc_bymac(const u8 *mac)
1174 struct net_device *dev;
1178 for_each_netdev(&init_net, dev) {
1179 if (dev->netdev_ops != &device_ops)
1180 continue; /* not a netvsc device */
1182 if (ether_addr_equal(mac, dev->perm_addr))
1189 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1191 struct net_device *dev;
1195 for_each_netdev(&init_net, dev) {
1196 struct net_device_context *net_device_ctx;
1198 if (dev->netdev_ops != &device_ops)
1199 continue; /* not a netvsc device */
1201 net_device_ctx = netdev_priv(dev);
1202 if (net_device_ctx->nvdev == NULL)
1203 continue; /* device is removed */
1205 if (rtnl_dereference(net_device_ctx->vf_netdev) == vf_netdev)
1206 return dev; /* a match */
1212 static int netvsc_register_vf(struct net_device *vf_netdev)
1214 struct net_device *ndev;
1215 struct net_device_context *net_device_ctx;
1216 struct netvsc_device *netvsc_dev;
1218 if (vf_netdev->addr_len != ETH_ALEN)
1222 * We will use the MAC address to locate the synthetic interface to
1223 * associate with the VF interface. If we don't find a matching
1224 * synthetic interface, move on.
1226 ndev = get_netvsc_bymac(vf_netdev->perm_addr);
1230 net_device_ctx = netdev_priv(ndev);
1231 netvsc_dev = net_device_ctx->nvdev;
1232 if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
1235 netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
1237 * Take a reference on the module.
1239 try_module_get(THIS_MODULE);
1241 dev_hold(vf_netdev);
1242 rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
1246 static int netvsc_vf_up(struct net_device *vf_netdev)
1248 struct net_device *ndev;
1249 struct netvsc_device *netvsc_dev;
1250 struct net_device_context *net_device_ctx;
1252 ndev = get_netvsc_byref(vf_netdev);
1256 net_device_ctx = netdev_priv(ndev);
1257 netvsc_dev = net_device_ctx->nvdev;
1259 netdev_info(ndev, "VF up: %s\n", vf_netdev->name);
1262 * Open the device before switching data path.
1264 rndis_filter_open(netvsc_dev);
1267 * notify the host to switch the data path.
1269 netvsc_switch_datapath(ndev, true);
1270 netdev_info(ndev, "Data path switched to VF: %s\n", vf_netdev->name);
1272 netif_carrier_off(ndev);
1274 /* Now notify peers through VF device. */
1275 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, vf_netdev);
1280 static int netvsc_vf_down(struct net_device *vf_netdev)
1282 struct net_device *ndev;
1283 struct netvsc_device *netvsc_dev;
1284 struct net_device_context *net_device_ctx;
1286 ndev = get_netvsc_byref(vf_netdev);
1290 net_device_ctx = netdev_priv(ndev);
1291 netvsc_dev = net_device_ctx->nvdev;
1293 netdev_info(ndev, "VF down: %s\n", vf_netdev->name);
1294 netvsc_switch_datapath(ndev, false);
1295 netdev_info(ndev, "Data path switched from VF: %s\n", vf_netdev->name);
1296 rndis_filter_close(netvsc_dev);
1297 netif_carrier_on(ndev);
1299 /* Now notify peers through netvsc device. */
1300 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, ndev);
1305 static int netvsc_unregister_vf(struct net_device *vf_netdev)
1307 struct net_device *ndev;
1308 struct netvsc_device *netvsc_dev;
1309 struct net_device_context *net_device_ctx;
1311 ndev = get_netvsc_byref(vf_netdev);
1315 net_device_ctx = netdev_priv(ndev);
1316 netvsc_dev = net_device_ctx->nvdev;
1318 netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
1320 RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
1322 module_put(THIS_MODULE);
1326 static int netvsc_probe(struct hv_device *dev,
1327 const struct hv_vmbus_device_id *dev_id)
1329 struct net_device *net = NULL;
1330 struct net_device_context *net_device_ctx;
1331 struct netvsc_device_info device_info;
1332 struct netvsc_device *nvdev;
1335 net = alloc_etherdev_mq(sizeof(struct net_device_context),
1340 netif_carrier_off(net);
1342 netvsc_init_settings(net);
1344 net_device_ctx = netdev_priv(net);
1345 net_device_ctx->device_ctx = dev;
1346 net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1347 if (netif_msg_probe(net_device_ctx))
1348 netdev_dbg(net, "netvsc msg_enable: %d\n",
1349 net_device_ctx->msg_enable);
1351 net_device_ctx->tx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1352 if (!net_device_ctx->tx_stats) {
1356 net_device_ctx->rx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1357 if (!net_device_ctx->rx_stats) {
1358 free_percpu(net_device_ctx->tx_stats);
1363 hv_set_drvdata(dev, net);
1365 net_device_ctx->start_remove = false;
1367 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1368 INIT_WORK(&net_device_ctx->work, do_set_multicast);
1370 spin_lock_init(&net_device_ctx->lock);
1371 INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
1373 net->netdev_ops = &device_ops;
1375 net->hw_features = NETVSC_HW_FEATURES;
1376 net->features = NETVSC_HW_FEATURES | NETIF_F_HW_VLAN_CTAG_TX;
1378 net->ethtool_ops = ðtool_ops;
1379 SET_NETDEV_DEV(net, &dev->device);
1381 /* We always need headroom for rndis header */
1382 net->needed_headroom = RNDIS_AND_PPI_SIZE;
1384 /* Notify the netvsc driver of the new device */
1385 memset(&device_info, 0, sizeof(device_info));
1386 device_info.ring_size = ring_size;
1387 device_info.max_num_vrss_chns = max_num_vrss_chns;
1388 ret = rndis_filter_device_add(dev, &device_info);
1390 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1391 netvsc_free_netdev(net);
1392 hv_set_drvdata(dev, NULL);
1395 memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1397 nvdev = net_device_ctx->nvdev;
1398 netif_set_real_num_tx_queues(net, nvdev->num_chn);
1399 netif_set_real_num_rx_queues(net, nvdev->num_chn);
1400 netif_set_gso_max_size(net, NETVSC_GSO_MAX_SIZE);
1402 /* MTU range: 68 - 1500 or 65521 */
1403 net->min_mtu = NETVSC_MTU_MIN;
1404 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
1405 net->max_mtu = NETVSC_MTU - ETH_HLEN;
1407 net->max_mtu = ETH_DATA_LEN;
1409 ret = register_netdev(net);
1411 pr_err("Unable to register netdev.\n");
1412 rndis_filter_device_remove(dev);
1413 netvsc_free_netdev(net);
1419 static int netvsc_remove(struct hv_device *dev)
1421 struct net_device *net;
1422 struct net_device_context *ndev_ctx;
1423 struct netvsc_device *net_device;
1425 net = hv_get_drvdata(dev);
1428 dev_err(&dev->device, "No net device to remove\n");
1432 ndev_ctx = netdev_priv(net);
1433 net_device = ndev_ctx->nvdev;
1435 /* Avoid racing with netvsc_change_mtu()/netvsc_set_channels()
1436 * removing the device.
1439 ndev_ctx->start_remove = true;
1442 cancel_delayed_work_sync(&ndev_ctx->dwork);
1443 cancel_work_sync(&ndev_ctx->work);
1445 /* Stop outbound asap */
1446 netif_tx_disable(net);
1448 unregister_netdev(net);
1451 * Call to the vsc driver to let it know that the device is being
1454 rndis_filter_device_remove(dev);
1456 hv_set_drvdata(dev, NULL);
1458 netvsc_free_netdev(net);
1462 static const struct hv_vmbus_device_id id_table[] = {
1468 MODULE_DEVICE_TABLE(vmbus, id_table);
1470 /* The one and only one */
1471 static struct hv_driver netvsc_drv = {
1472 .name = KBUILD_MODNAME,
1473 .id_table = id_table,
1474 .probe = netvsc_probe,
1475 .remove = netvsc_remove,
1479 * On Hyper-V, every VF interface is matched with a corresponding
1480 * synthetic interface. The synthetic interface is presented first
1481 * to the guest. When the corresponding VF instance is registered,
1482 * we will take care of switching the data path.
1484 static int netvsc_netdev_event(struct notifier_block *this,
1485 unsigned long event, void *ptr)
1487 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
1489 /* Skip our own events */
1490 if (event_dev->netdev_ops == &device_ops)
1493 /* Avoid non-Ethernet type devices */
1494 if (event_dev->type != ARPHRD_ETHER)
1497 /* Avoid Vlan dev with same MAC registering as VF */
1498 if (event_dev->priv_flags & IFF_802_1Q_VLAN)
1501 /* Avoid Bonding master dev with same MAC registering as VF */
1502 if ((event_dev->priv_flags & IFF_BONDING) &&
1503 (event_dev->flags & IFF_MASTER))
1507 case NETDEV_REGISTER:
1508 return netvsc_register_vf(event_dev);
1509 case NETDEV_UNREGISTER:
1510 return netvsc_unregister_vf(event_dev);
1512 return netvsc_vf_up(event_dev);
1514 return netvsc_vf_down(event_dev);
1520 static struct notifier_block netvsc_netdev_notifier = {
1521 .notifier_call = netvsc_netdev_event,
1524 static void __exit netvsc_drv_exit(void)
1526 unregister_netdevice_notifier(&netvsc_netdev_notifier);
1527 vmbus_driver_unregister(&netvsc_drv);
1530 static int __init netvsc_drv_init(void)
1534 if (ring_size < RING_SIZE_MIN) {
1535 ring_size = RING_SIZE_MIN;
1536 pr_info("Increased ring_size to %d (min allowed)\n",
1539 ret = vmbus_driver_register(&netvsc_drv);
1544 register_netdevice_notifier(&netvsc_netdev_notifier);
1548 MODULE_LICENSE("GPL");
1549 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1551 module_init(netvsc_drv_init);
1552 module_exit(netvsc_drv_exit);