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)
46 static int ring_size = 128;
47 module_param(ring_size, int, S_IRUGO);
48 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
50 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
51 NETIF_MSG_LINK | NETIF_MSG_IFUP |
52 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
55 static int debug = -1;
56 module_param(debug, int, S_IRUGO);
57 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
59 static void do_set_multicast(struct work_struct *w)
61 struct net_device_context *ndevctx =
62 container_of(w, struct net_device_context, work);
63 struct hv_device *device_obj = ndevctx->device_ctx;
64 struct net_device *ndev = hv_get_drvdata(device_obj);
65 struct netvsc_device *nvdev = ndevctx->nvdev;
66 struct rndis_device *rdev;
71 rdev = nvdev->extension;
75 if (ndev->flags & IFF_PROMISC)
76 rndis_filter_set_packet_filter(rdev,
77 NDIS_PACKET_TYPE_PROMISCUOUS);
79 rndis_filter_set_packet_filter(rdev,
80 NDIS_PACKET_TYPE_BROADCAST |
81 NDIS_PACKET_TYPE_ALL_MULTICAST |
82 NDIS_PACKET_TYPE_DIRECTED);
85 static void netvsc_set_multicast_list(struct net_device *net)
87 struct net_device_context *net_device_ctx = netdev_priv(net);
89 schedule_work(&net_device_ctx->work);
92 static int netvsc_open(struct net_device *net)
94 struct netvsc_device *nvdev = net_device_to_netvsc_device(net);
95 struct rndis_device *rdev;
98 netif_carrier_off(net);
100 /* Open up the device */
101 ret = rndis_filter_open(nvdev);
103 netdev_err(net, "unable to open device (ret %d).\n", ret);
107 netif_tx_wake_all_queues(net);
109 rdev = nvdev->extension;
110 if (!rdev->link_state)
111 netif_carrier_on(net);
116 static int netvsc_close(struct net_device *net)
118 struct net_device_context *net_device_ctx = netdev_priv(net);
119 struct netvsc_device *nvdev = net_device_ctx->nvdev;
121 u32 aread, awrite, i, msec = 10, retry = 0, retry_max = 20;
122 struct vmbus_channel *chn;
124 netif_tx_disable(net);
126 /* Make sure netvsc_set_multicast_list doesn't re-enable filter! */
127 cancel_work_sync(&net_device_ctx->work);
128 ret = rndis_filter_close(nvdev);
130 netdev_err(net, "unable to close device (ret %d).\n", ret);
134 /* Ensure pending bytes in ring are read */
137 for (i = 0; i < nvdev->num_chn; i++) {
138 chn = nvdev->chn_table[i];
142 hv_get_ringbuffer_availbytes(&chn->inbound, &aread,
148 hv_get_ringbuffer_availbytes(&chn->outbound, &aread,
156 if (retry > retry_max || aread == 0)
166 netdev_err(net, "Ring buffer not empty after closing rndis\n");
173 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
176 struct rndis_packet *rndis_pkt;
177 struct rndis_per_packet_info *ppi;
179 rndis_pkt = &msg->msg.pkt;
180 rndis_pkt->data_offset += ppi_size;
182 ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
183 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
185 ppi->size = ppi_size;
186 ppi->type = pkt_type;
187 ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
189 rndis_pkt->per_pkt_info_len += ppi_size;
194 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
195 void *accel_priv, select_queue_fallback_t fallback)
197 struct net_device_context *net_device_ctx = netdev_priv(ndev);
198 struct netvsc_device *nvsc_dev = net_device_ctx->nvdev;
202 if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1)
205 hash = skb_get_hash(skb);
206 q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] %
207 ndev->real_num_tx_queues;
209 if (!nvsc_dev->chn_table[q_idx])
215 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
216 struct hv_page_buffer *pb)
220 /* Deal with compund pages by ignoring unused part
223 page += (offset >> PAGE_SHIFT);
224 offset &= ~PAGE_MASK;
229 bytes = PAGE_SIZE - offset;
232 pb[j].pfn = page_to_pfn(page);
233 pb[j].offset = offset;
239 if (offset == PAGE_SIZE && len) {
249 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
250 struct hv_netvsc_packet *packet,
251 struct hv_page_buffer **page_buf)
253 struct hv_page_buffer *pb = *page_buf;
255 char *data = skb->data;
256 int frags = skb_shinfo(skb)->nr_frags;
259 /* The packet is laid out thus:
260 * 1. hdr: RNDIS header and PPI
262 * 3. skb fragment data
265 slots_used += fill_pg_buf(virt_to_page(hdr),
267 len, &pb[slots_used]);
269 packet->rmsg_size = len;
270 packet->rmsg_pgcnt = slots_used;
272 slots_used += fill_pg_buf(virt_to_page(data),
273 offset_in_page(data),
274 skb_headlen(skb), &pb[slots_used]);
276 for (i = 0; i < frags; i++) {
277 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
279 slots_used += fill_pg_buf(skb_frag_page(frag),
281 skb_frag_size(frag), &pb[slots_used]);
286 static int count_skb_frag_slots(struct sk_buff *skb)
288 int i, frags = skb_shinfo(skb)->nr_frags;
291 for (i = 0; i < frags; i++) {
292 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
293 unsigned long size = skb_frag_size(frag);
294 unsigned long offset = frag->page_offset;
296 /* Skip unused frames from start of page */
297 offset &= ~PAGE_MASK;
298 pages += PFN_UP(offset + size);
303 static int netvsc_get_slots(struct sk_buff *skb)
305 char *data = skb->data;
306 unsigned int offset = offset_in_page(data);
307 unsigned int len = skb_headlen(skb);
311 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
312 frag_slots = count_skb_frag_slots(skb);
313 return slots + frag_slots;
316 static u32 net_checksum_info(struct sk_buff *skb)
318 if (skb->protocol == htons(ETH_P_IP)) {
319 struct iphdr *ip = ip_hdr(skb);
321 if (ip->protocol == IPPROTO_TCP)
322 return TRANSPORT_INFO_IPV4_TCP;
323 else if (ip->protocol == IPPROTO_UDP)
324 return TRANSPORT_INFO_IPV4_UDP;
326 struct ipv6hdr *ip6 = ipv6_hdr(skb);
328 if (ip6->nexthdr == IPPROTO_TCP)
329 return TRANSPORT_INFO_IPV6_TCP;
330 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
331 return TRANSPORT_INFO_IPV6_UDP;
334 return TRANSPORT_INFO_NOT_IP;
337 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
339 struct net_device_context *net_device_ctx = netdev_priv(net);
340 struct hv_netvsc_packet *packet = NULL;
342 unsigned int num_data_pgs;
343 struct rndis_message *rndis_msg;
344 struct rndis_packet *rndis_pkt;
346 struct rndis_per_packet_info *ppi;
349 struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
350 struct hv_page_buffer *pb = page_buf;
352 /* We will atmost need two pages to describe the rndis
353 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
354 * of pages in a single packet. If skb is scattered around
355 * more pages we try linearizing it.
358 skb_length = skb->len;
359 num_data_pgs = netvsc_get_slots(skb) + 2;
361 if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
362 ++net_device_ctx->eth_stats.tx_scattered;
364 if (skb_linearize(skb))
367 num_data_pgs = netvsc_get_slots(skb) + 2;
368 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
369 ++net_device_ctx->eth_stats.tx_too_big;
375 * Place the rndis header in the skb head room and
376 * the skb->cb will be used for hv_netvsc_packet
379 ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
383 /* Use the skb control buffer for building up the packet */
384 BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
385 FIELD_SIZEOF(struct sk_buff, cb));
386 packet = (struct hv_netvsc_packet *)skb->cb;
388 packet->q_idx = skb_get_queue_mapping(skb);
390 packet->total_data_buflen = skb->len;
392 rndis_msg = (struct rndis_message *)skb->head;
394 memset(rndis_msg, 0, RNDIS_AND_PPI_SIZE);
396 /* Add the rndis header */
397 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
398 rndis_msg->msg_len = packet->total_data_buflen;
399 rndis_pkt = &rndis_msg->msg.pkt;
400 rndis_pkt->data_offset = sizeof(struct rndis_packet);
401 rndis_pkt->data_len = packet->total_data_buflen;
402 rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
404 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
406 hash = skb_get_hash_raw(skb);
407 if (hash != 0 && net->real_num_tx_queues > 1) {
408 rndis_msg_size += NDIS_HASH_PPI_SIZE;
409 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
411 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
414 if (skb_vlan_tag_present(skb)) {
415 struct ndis_pkt_8021q_info *vlan;
417 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
418 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
420 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
422 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
423 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
427 if (skb_is_gso(skb)) {
428 struct ndis_tcp_lso_info *lso_info;
430 rndis_msg_size += NDIS_LSO_PPI_SIZE;
431 ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
432 TCP_LARGESEND_PKTINFO);
434 lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
437 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
438 if (skb->protocol == htons(ETH_P_IP)) {
439 lso_info->lso_v2_transmit.ip_version =
440 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
441 ip_hdr(skb)->tot_len = 0;
442 ip_hdr(skb)->check = 0;
443 tcp_hdr(skb)->check =
444 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
445 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
447 lso_info->lso_v2_transmit.ip_version =
448 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
449 ipv6_hdr(skb)->payload_len = 0;
450 tcp_hdr(skb)->check =
451 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
452 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
454 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
455 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
456 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
457 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
458 struct ndis_tcp_ip_checksum_info *csum_info;
460 rndis_msg_size += NDIS_CSUM_PPI_SIZE;
461 ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
462 TCPIP_CHKSUM_PKTINFO);
464 csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
467 csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
469 if (skb->protocol == htons(ETH_P_IP)) {
470 csum_info->transmit.is_ipv4 = 1;
472 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
473 csum_info->transmit.tcp_checksum = 1;
475 csum_info->transmit.udp_checksum = 1;
477 csum_info->transmit.is_ipv6 = 1;
479 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
480 csum_info->transmit.tcp_checksum = 1;
482 csum_info->transmit.udp_checksum = 1;
485 /* Can't do offload of this type of checksum */
486 if (skb_checksum_help(skb))
491 /* Start filling in the page buffers with the rndis hdr */
492 rndis_msg->msg_len += rndis_msg_size;
493 packet->total_data_buflen = rndis_msg->msg_len;
494 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
497 /* timestamp packet in software */
498 skb_tx_timestamp(skb);
499 ret = netvsc_send(net_device_ctx->device_ctx, packet,
500 rndis_msg, &pb, skb);
501 if (likely(ret == 0)) {
502 struct netvsc_stats *tx_stats = this_cpu_ptr(net_device_ctx->tx_stats);
504 u64_stats_update_begin(&tx_stats->syncp);
506 tx_stats->bytes += skb_length;
507 u64_stats_update_end(&tx_stats->syncp);
511 if (ret == -EAGAIN) {
512 ++net_device_ctx->eth_stats.tx_busy;
513 return NETDEV_TX_BUSY;
517 ++net_device_ctx->eth_stats.tx_no_space;
520 dev_kfree_skb_any(skb);
521 net->stats.tx_dropped++;
526 ++net_device_ctx->eth_stats.tx_no_memory;
531 * netvsc_linkstatus_callback - Link up/down notification
533 void netvsc_linkstatus_callback(struct hv_device *device_obj,
534 struct rndis_message *resp)
536 struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
537 struct net_device *net;
538 struct net_device_context *ndev_ctx;
539 struct netvsc_reconfig *event;
542 net = hv_get_drvdata(device_obj);
547 ndev_ctx = netdev_priv(net);
549 /* Update the physical link speed when changing to another vSwitch */
550 if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
553 speed = *(u32 *)((void *)indicate + indicate->
554 status_buf_offset) / 10000;
555 ndev_ctx->speed = speed;
559 /* Handle these link change statuses below */
560 if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
561 indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
562 indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
565 if (net->reg_state != NETREG_REGISTERED)
568 event = kzalloc(sizeof(*event), GFP_ATOMIC);
571 event->event = indicate->status;
573 spin_lock_irqsave(&ndev_ctx->lock, flags);
574 list_add_tail(&event->list, &ndev_ctx->reconfig_events);
575 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
577 schedule_delayed_work(&ndev_ctx->dwork, 0);
580 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
581 struct hv_netvsc_packet *packet,
582 struct ndis_tcp_ip_checksum_info *csum_info,
583 void *data, u16 vlan_tci)
587 skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
592 * Copy to skb. This copy is needed here since the memory pointed by
593 * hv_netvsc_packet cannot be deallocated
595 memcpy(skb_put(skb, packet->total_data_buflen), data,
596 packet->total_data_buflen);
598 skb->protocol = eth_type_trans(skb, net);
600 /* skb is already created with CHECKSUM_NONE */
601 skb_checksum_none_assert(skb);
604 * In Linux, the IP checksum is always checked.
605 * Do L4 checksum offload if enabled and present.
607 if (csum_info && (net->features & NETIF_F_RXCSUM)) {
608 if (csum_info->receive.tcp_checksum_succeeded ||
609 csum_info->receive.udp_checksum_succeeded)
610 skb->ip_summed = CHECKSUM_UNNECESSARY;
613 if (vlan_tci & VLAN_TAG_PRESENT)
614 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
621 * netvsc_recv_callback - Callback when we receive a packet from the
622 * "wire" on the specified device.
624 int netvsc_recv_callback(struct hv_device *device_obj,
625 struct hv_netvsc_packet *packet,
627 struct ndis_tcp_ip_checksum_info *csum_info,
628 struct vmbus_channel *channel,
631 struct net_device *net = hv_get_drvdata(device_obj);
632 struct net_device_context *net_device_ctx = netdev_priv(net);
633 struct net_device *vf_netdev;
635 struct netvsc_stats *rx_stats;
637 if (net->reg_state != NETREG_REGISTERED)
638 return NVSP_STAT_FAIL;
641 * If necessary, inject this packet into the VF interface.
642 * On Hyper-V, multicast and brodcast packets are only delivered
643 * to the synthetic interface (after subjecting these to
644 * policy filters on the host). Deliver these via the VF
645 * interface in the guest.
648 vf_netdev = rcu_dereference(net_device_ctx->vf_netdev);
649 if (vf_netdev && (vf_netdev->flags & IFF_UP))
652 /* Allocate a skb - TODO direct I/O to pages? */
653 skb = netvsc_alloc_recv_skb(net, packet, csum_info, *data, vlan_tci);
654 if (unlikely(!skb)) {
655 ++net->stats.rx_dropped;
657 return NVSP_STAT_FAIL;
660 if (net != vf_netdev)
661 skb_record_rx_queue(skb,
662 channel->offermsg.offer.sub_channel_index);
665 * Even if injecting the packet, record the statistics
666 * on the synthetic device because modifying the VF device
667 * statistics will not work correctly.
669 rx_stats = this_cpu_ptr(net_device_ctx->rx_stats);
670 u64_stats_update_begin(&rx_stats->syncp);
672 rx_stats->bytes += packet->total_data_buflen;
674 if (skb->pkt_type == PACKET_BROADCAST)
675 ++rx_stats->broadcast;
676 else if (skb->pkt_type == PACKET_MULTICAST)
677 ++rx_stats->multicast;
678 u64_stats_update_end(&rx_stats->syncp);
681 * Pass the skb back up. Network stack will deallocate the skb when it
691 static void netvsc_get_drvinfo(struct net_device *net,
692 struct ethtool_drvinfo *info)
694 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
695 strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
698 static void netvsc_get_channels(struct net_device *net,
699 struct ethtool_channels *channel)
701 struct net_device_context *net_device_ctx = netdev_priv(net);
702 struct netvsc_device *nvdev = net_device_ctx->nvdev;
705 channel->max_combined = nvdev->max_chn;
706 channel->combined_count = nvdev->num_chn;
710 static int netvsc_set_queues(struct net_device *net, struct hv_device *dev,
713 struct netvsc_device_info device_info;
716 memset(&device_info, 0, sizeof(device_info));
717 device_info.num_chn = num_chn;
718 device_info.ring_size = ring_size;
719 device_info.max_num_vrss_chns = num_chn;
721 ret = rndis_filter_device_add(dev, &device_info);
725 ret = netif_set_real_num_tx_queues(net, num_chn);
729 ret = netif_set_real_num_rx_queues(net, num_chn);
734 static int netvsc_set_channels(struct net_device *net,
735 struct ethtool_channels *channels)
737 struct net_device_context *net_device_ctx = netdev_priv(net);
738 struct hv_device *dev = net_device_ctx->device_ctx;
739 struct netvsc_device *nvdev = net_device_ctx->nvdev;
740 unsigned int count = channels->combined_count;
743 /* We do not support separate count for rx, tx, or other */
745 channels->rx_count || channels->tx_count || channels->other_count)
748 if (count > net->num_tx_queues || count > net->num_rx_queues)
751 if (net_device_ctx->start_remove || !nvdev || nvdev->destroy)
754 if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
757 if (count > nvdev->max_chn)
760 ret = netvsc_close(net);
764 net_device_ctx->start_remove = true;
765 rndis_filter_device_remove(dev);
767 ret = netvsc_set_queues(net, dev, count);
769 nvdev->num_chn = count;
771 netvsc_set_queues(net, dev, nvdev->num_chn);
774 net_device_ctx->start_remove = false;
776 /* We may have missed link change notifications */
777 schedule_delayed_work(&net_device_ctx->dwork, 0);
782 static bool netvsc_validate_ethtool_ss_cmd(const struct ethtool_cmd *cmd)
784 struct ethtool_cmd diff1 = *cmd;
785 struct ethtool_cmd diff2 = {};
787 ethtool_cmd_speed_set(&diff1, 0);
789 /* advertising and cmd are usually set */
790 diff1.advertising = 0;
792 /* We set port to PORT_OTHER */
793 diff2.port = PORT_OTHER;
795 return !memcmp(&diff1, &diff2, sizeof(diff1));
798 static void netvsc_init_settings(struct net_device *dev)
800 struct net_device_context *ndc = netdev_priv(dev);
802 ndc->speed = SPEED_UNKNOWN;
803 ndc->duplex = DUPLEX_UNKNOWN;
806 static int netvsc_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
808 struct net_device_context *ndc = netdev_priv(dev);
810 ethtool_cmd_speed_set(cmd, ndc->speed);
811 cmd->duplex = ndc->duplex;
812 cmd->port = PORT_OTHER;
817 static int netvsc_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
819 struct net_device_context *ndc = netdev_priv(dev);
822 speed = ethtool_cmd_speed(cmd);
823 if (!ethtool_validate_speed(speed) ||
824 !ethtool_validate_duplex(cmd->duplex) ||
825 !netvsc_validate_ethtool_ss_cmd(cmd))
829 ndc->duplex = cmd->duplex;
834 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
836 struct net_device_context *ndevctx = netdev_priv(ndev);
837 struct netvsc_device *nvdev = ndevctx->nvdev;
838 struct hv_device *hdev = ndevctx->device_ctx;
839 struct netvsc_device_info device_info;
842 if (ndevctx->start_remove || !nvdev || nvdev->destroy)
845 ret = netvsc_close(ndev);
849 ndevctx->start_remove = true;
850 rndis_filter_device_remove(hdev);
854 memset(&device_info, 0, sizeof(device_info));
855 device_info.ring_size = ring_size;
856 device_info.num_chn = nvdev->num_chn;
857 device_info.max_num_vrss_chns = nvdev->num_chn;
858 rndis_filter_device_add(hdev, &device_info);
862 ndevctx->start_remove = false;
864 /* We may have missed link change notifications */
865 schedule_delayed_work(&ndevctx->dwork, 0);
870 static void netvsc_get_stats64(struct net_device *net,
871 struct rtnl_link_stats64 *t)
873 struct net_device_context *ndev_ctx = netdev_priv(net);
876 for_each_possible_cpu(cpu) {
877 struct netvsc_stats *tx_stats = per_cpu_ptr(ndev_ctx->tx_stats,
879 struct netvsc_stats *rx_stats = per_cpu_ptr(ndev_ctx->rx_stats,
881 u64 tx_packets, tx_bytes, rx_packets, rx_bytes, rx_multicast;
885 start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
886 tx_packets = tx_stats->packets;
887 tx_bytes = tx_stats->bytes;
888 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
891 start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
892 rx_packets = rx_stats->packets;
893 rx_bytes = rx_stats->bytes;
894 rx_multicast = rx_stats->multicast + rx_stats->broadcast;
895 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
897 t->tx_bytes += tx_bytes;
898 t->tx_packets += tx_packets;
899 t->rx_bytes += rx_bytes;
900 t->rx_packets += rx_packets;
901 t->multicast += rx_multicast;
904 t->tx_dropped = net->stats.tx_dropped;
905 t->tx_errors = net->stats.tx_dropped;
907 t->rx_dropped = net->stats.rx_dropped;
908 t->rx_errors = net->stats.rx_errors;
911 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
913 struct sockaddr *addr = p;
914 char save_adr[ETH_ALEN];
915 unsigned char save_aatype;
918 memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
919 save_aatype = ndev->addr_assign_type;
921 err = eth_mac_addr(ndev, p);
925 err = rndis_filter_set_device_mac(ndev, addr->sa_data);
927 /* roll back to saved MAC */
928 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
929 ndev->addr_assign_type = save_aatype;
935 static const struct {
936 char name[ETH_GSTRING_LEN];
939 { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
940 { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
941 { "tx_no_space", offsetof(struct netvsc_ethtool_stats, tx_no_space) },
942 { "tx_too_big", offsetof(struct netvsc_ethtool_stats, tx_too_big) },
943 { "tx_busy", offsetof(struct netvsc_ethtool_stats, tx_busy) },
946 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
948 switch (string_set) {
950 return ARRAY_SIZE(netvsc_stats);
956 static void netvsc_get_ethtool_stats(struct net_device *dev,
957 struct ethtool_stats *stats, u64 *data)
959 struct net_device_context *ndc = netdev_priv(dev);
960 const void *nds = &ndc->eth_stats;
963 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
964 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
967 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
973 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
974 memcpy(data + i * ETH_GSTRING_LEN,
975 netvsc_stats[i].name, ETH_GSTRING_LEN);
981 netvsc_get_rss_hash_opts(struct netvsc_device *nvdev,
982 struct ethtool_rxnfc *info)
984 info->data = RXH_IP_SRC | RXH_IP_DST;
986 switch (info->flow_type) {
989 info->data |= RXH_L4_B_0_1 | RXH_L4_B_2_3;
1005 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1008 struct net_device_context *ndc = netdev_priv(dev);
1009 struct netvsc_device *nvdev = ndc->nvdev;
1011 switch (info->cmd) {
1012 case ETHTOOL_GRXRINGS:
1013 info->data = nvdev->num_chn;
1017 return netvsc_get_rss_hash_opts(nvdev, info);
1022 #ifdef CONFIG_NET_POLL_CONTROLLER
1023 static void netvsc_poll_controller(struct net_device *net)
1025 /* As netvsc_start_xmit() works synchronous we don't have to
1026 * trigger anything here.
1031 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1033 return NETVSC_HASH_KEYLEN;
1036 static u32 netvsc_rss_indir_size(struct net_device *dev)
1041 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1044 struct net_device_context *ndc = netdev_priv(dev);
1045 struct netvsc_device *ndev = ndc->nvdev;
1046 struct rndis_device *rndis_dev = ndev->extension;
1050 *hfunc = ETH_RSS_HASH_TOP; /* Toeplitz */
1053 for (i = 0; i < ITAB_NUM; i++)
1054 indir[i] = rndis_dev->ind_table[i];
1058 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1063 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1064 const u8 *key, const u8 hfunc)
1066 struct net_device_context *ndc = netdev_priv(dev);
1067 struct netvsc_device *ndev = ndc->nvdev;
1068 struct rndis_device *rndis_dev = ndev->extension;
1071 if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1075 for (i = 0; i < ITAB_NUM; i++)
1076 if (indir[i] >= dev->num_rx_queues)
1079 for (i = 0; i < ITAB_NUM; i++)
1080 rndis_dev->ind_table[i] = indir[i];
1087 key = rndis_dev->rss_key;
1090 return rndis_filter_set_rss_param(rndis_dev, key, ndev->num_chn);
1093 static const struct ethtool_ops ethtool_ops = {
1094 .get_drvinfo = netvsc_get_drvinfo,
1095 .get_link = ethtool_op_get_link,
1096 .get_ethtool_stats = netvsc_get_ethtool_stats,
1097 .get_sset_count = netvsc_get_sset_count,
1098 .get_strings = netvsc_get_strings,
1099 .get_channels = netvsc_get_channels,
1100 .set_channels = netvsc_set_channels,
1101 .get_ts_info = ethtool_op_get_ts_info,
1102 .get_settings = netvsc_get_settings,
1103 .set_settings = netvsc_set_settings,
1104 .get_rxnfc = netvsc_get_rxnfc,
1105 .get_rxfh_key_size = netvsc_get_rxfh_key_size,
1106 .get_rxfh_indir_size = netvsc_rss_indir_size,
1107 .get_rxfh = netvsc_get_rxfh,
1108 .set_rxfh = netvsc_set_rxfh,
1111 static const struct net_device_ops device_ops = {
1112 .ndo_open = netvsc_open,
1113 .ndo_stop = netvsc_close,
1114 .ndo_start_xmit = netvsc_start_xmit,
1115 .ndo_set_rx_mode = netvsc_set_multicast_list,
1116 .ndo_change_mtu = netvsc_change_mtu,
1117 .ndo_validate_addr = eth_validate_addr,
1118 .ndo_set_mac_address = netvsc_set_mac_addr,
1119 .ndo_select_queue = netvsc_select_queue,
1120 .ndo_get_stats64 = netvsc_get_stats64,
1121 #ifdef CONFIG_NET_POLL_CONTROLLER
1122 .ndo_poll_controller = netvsc_poll_controller,
1127 * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1128 * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1129 * present send GARP packet to network peers with netif_notify_peers().
1131 static void netvsc_link_change(struct work_struct *w)
1133 struct net_device_context *ndev_ctx =
1134 container_of(w, struct net_device_context, dwork.work);
1135 struct hv_device *device_obj = ndev_ctx->device_ctx;
1136 struct net_device *net = hv_get_drvdata(device_obj);
1137 struct netvsc_device *net_device;
1138 struct rndis_device *rdev;
1139 struct netvsc_reconfig *event = NULL;
1140 bool notify = false, reschedule = false;
1141 unsigned long flags, next_reconfig, delay;
1144 if (ndev_ctx->start_remove)
1147 net_device = ndev_ctx->nvdev;
1148 rdev = net_device->extension;
1150 next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1151 if (time_is_after_jiffies(next_reconfig)) {
1152 /* link_watch only sends one notification with current state
1153 * per second, avoid doing reconfig more frequently. Handle
1156 delay = next_reconfig - jiffies;
1157 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1158 schedule_delayed_work(&ndev_ctx->dwork, delay);
1161 ndev_ctx->last_reconfig = jiffies;
1163 spin_lock_irqsave(&ndev_ctx->lock, flags);
1164 if (!list_empty(&ndev_ctx->reconfig_events)) {
1165 event = list_first_entry(&ndev_ctx->reconfig_events,
1166 struct netvsc_reconfig, list);
1167 list_del(&event->list);
1168 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1170 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1175 switch (event->event) {
1176 /* Only the following events are possible due to the check in
1177 * netvsc_linkstatus_callback()
1179 case RNDIS_STATUS_MEDIA_CONNECT:
1180 if (rdev->link_state) {
1181 rdev->link_state = false;
1182 netif_carrier_on(net);
1183 netif_tx_wake_all_queues(net);
1189 case RNDIS_STATUS_MEDIA_DISCONNECT:
1190 if (!rdev->link_state) {
1191 rdev->link_state = true;
1192 netif_carrier_off(net);
1193 netif_tx_stop_all_queues(net);
1197 case RNDIS_STATUS_NETWORK_CHANGE:
1198 /* Only makes sense if carrier is present */
1199 if (!rdev->link_state) {
1200 rdev->link_state = true;
1201 netif_carrier_off(net);
1202 netif_tx_stop_all_queues(net);
1203 event->event = RNDIS_STATUS_MEDIA_CONNECT;
1204 spin_lock_irqsave(&ndev_ctx->lock, flags);
1205 list_add(&event->list, &ndev_ctx->reconfig_events);
1206 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1215 netdev_notify_peers(net);
1217 /* link_watch only sends one notification with current state per
1218 * second, handle next reconfig event in 2 seconds.
1221 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1229 static void netvsc_free_netdev(struct net_device *netdev)
1231 struct net_device_context *net_device_ctx = netdev_priv(netdev);
1233 free_percpu(net_device_ctx->tx_stats);
1234 free_percpu(net_device_ctx->rx_stats);
1235 free_netdev(netdev);
1238 static struct net_device *get_netvsc_bymac(const u8 *mac)
1240 struct net_device *dev;
1244 for_each_netdev(&init_net, dev) {
1245 if (dev->netdev_ops != &device_ops)
1246 continue; /* not a netvsc device */
1248 if (ether_addr_equal(mac, dev->perm_addr))
1255 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1257 struct net_device *dev;
1261 for_each_netdev(&init_net, dev) {
1262 struct net_device_context *net_device_ctx;
1264 if (dev->netdev_ops != &device_ops)
1265 continue; /* not a netvsc device */
1267 net_device_ctx = netdev_priv(dev);
1268 if (net_device_ctx->nvdev == NULL)
1269 continue; /* device is removed */
1271 if (rtnl_dereference(net_device_ctx->vf_netdev) == vf_netdev)
1272 return dev; /* a match */
1278 static int netvsc_register_vf(struct net_device *vf_netdev)
1280 struct net_device *ndev;
1281 struct net_device_context *net_device_ctx;
1282 struct netvsc_device *netvsc_dev;
1284 if (vf_netdev->addr_len != ETH_ALEN)
1288 * We will use the MAC address to locate the synthetic interface to
1289 * associate with the VF interface. If we don't find a matching
1290 * synthetic interface, move on.
1292 ndev = get_netvsc_bymac(vf_netdev->perm_addr);
1296 net_device_ctx = netdev_priv(ndev);
1297 netvsc_dev = net_device_ctx->nvdev;
1298 if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
1301 netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
1303 * Take a reference on the module.
1305 try_module_get(THIS_MODULE);
1307 dev_hold(vf_netdev);
1308 rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
1312 static int netvsc_vf_up(struct net_device *vf_netdev)
1314 struct net_device *ndev;
1315 struct netvsc_device *netvsc_dev;
1316 struct net_device_context *net_device_ctx;
1318 ndev = get_netvsc_byref(vf_netdev);
1322 net_device_ctx = netdev_priv(ndev);
1323 netvsc_dev = net_device_ctx->nvdev;
1325 netdev_info(ndev, "VF up: %s\n", vf_netdev->name);
1328 * Open the device before switching data path.
1330 rndis_filter_open(netvsc_dev);
1333 * notify the host to switch the data path.
1335 netvsc_switch_datapath(ndev, true);
1336 netdev_info(ndev, "Data path switched to VF: %s\n", vf_netdev->name);
1338 netif_carrier_off(ndev);
1340 /* Now notify peers through VF device. */
1341 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, vf_netdev);
1346 static int netvsc_vf_down(struct net_device *vf_netdev)
1348 struct net_device *ndev;
1349 struct netvsc_device *netvsc_dev;
1350 struct net_device_context *net_device_ctx;
1352 ndev = get_netvsc_byref(vf_netdev);
1356 net_device_ctx = netdev_priv(ndev);
1357 netvsc_dev = net_device_ctx->nvdev;
1359 netdev_info(ndev, "VF down: %s\n", vf_netdev->name);
1360 netvsc_switch_datapath(ndev, false);
1361 netdev_info(ndev, "Data path switched from VF: %s\n", vf_netdev->name);
1362 rndis_filter_close(netvsc_dev);
1363 netif_carrier_on(ndev);
1365 /* Now notify peers through netvsc device. */
1366 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, ndev);
1371 static int netvsc_unregister_vf(struct net_device *vf_netdev)
1373 struct net_device *ndev;
1374 struct netvsc_device *netvsc_dev;
1375 struct net_device_context *net_device_ctx;
1377 ndev = get_netvsc_byref(vf_netdev);
1381 net_device_ctx = netdev_priv(ndev);
1382 netvsc_dev = net_device_ctx->nvdev;
1384 netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
1386 RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
1388 module_put(THIS_MODULE);
1392 static int netvsc_probe(struct hv_device *dev,
1393 const struct hv_vmbus_device_id *dev_id)
1395 struct net_device *net = NULL;
1396 struct net_device_context *net_device_ctx;
1397 struct netvsc_device_info device_info;
1398 struct netvsc_device *nvdev;
1401 net = alloc_etherdev_mq(sizeof(struct net_device_context),
1406 netif_carrier_off(net);
1408 netvsc_init_settings(net);
1410 net_device_ctx = netdev_priv(net);
1411 net_device_ctx->device_ctx = dev;
1412 net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1413 if (netif_msg_probe(net_device_ctx))
1414 netdev_dbg(net, "netvsc msg_enable: %d\n",
1415 net_device_ctx->msg_enable);
1417 net_device_ctx->tx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1418 if (!net_device_ctx->tx_stats) {
1422 net_device_ctx->rx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1423 if (!net_device_ctx->rx_stats) {
1424 free_percpu(net_device_ctx->tx_stats);
1429 hv_set_drvdata(dev, net);
1431 net_device_ctx->start_remove = false;
1433 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1434 INIT_WORK(&net_device_ctx->work, do_set_multicast);
1436 spin_lock_init(&net_device_ctx->lock);
1437 INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
1439 net->netdev_ops = &device_ops;
1440 net->ethtool_ops = ðtool_ops;
1441 SET_NETDEV_DEV(net, &dev->device);
1443 /* We always need headroom for rndis header */
1444 net->needed_headroom = RNDIS_AND_PPI_SIZE;
1446 /* Notify the netvsc driver of the new device */
1447 memset(&device_info, 0, sizeof(device_info));
1448 device_info.ring_size = ring_size;
1449 device_info.max_num_vrss_chns = min_t(u32, VRSS_CHANNEL_DEFAULT,
1451 ret = rndis_filter_device_add(dev, &device_info);
1453 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1454 netvsc_free_netdev(net);
1455 hv_set_drvdata(dev, NULL);
1458 memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1460 /* hw_features computed in rndis_filter_device_add */
1461 net->features = net->hw_features |
1462 NETIF_F_HIGHDMA | NETIF_F_SG |
1463 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
1464 net->vlan_features = net->features;
1466 nvdev = net_device_ctx->nvdev;
1467 netif_set_real_num_tx_queues(net, nvdev->num_chn);
1468 netif_set_real_num_rx_queues(net, nvdev->num_chn);
1470 /* MTU range: 68 - 1500 or 65521 */
1471 net->min_mtu = NETVSC_MTU_MIN;
1472 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
1473 net->max_mtu = NETVSC_MTU - ETH_HLEN;
1475 net->max_mtu = ETH_DATA_LEN;
1477 ret = register_netdev(net);
1479 pr_err("Unable to register netdev.\n");
1480 rndis_filter_device_remove(dev);
1481 netvsc_free_netdev(net);
1487 static int netvsc_remove(struct hv_device *dev)
1489 struct net_device *net;
1490 struct net_device_context *ndev_ctx;
1491 struct netvsc_device *net_device;
1493 net = hv_get_drvdata(dev);
1496 dev_err(&dev->device, "No net device to remove\n");
1500 ndev_ctx = netdev_priv(net);
1501 net_device = ndev_ctx->nvdev;
1503 /* Avoid racing with netvsc_change_mtu()/netvsc_set_channels()
1504 * removing the device.
1507 ndev_ctx->start_remove = true;
1510 cancel_delayed_work_sync(&ndev_ctx->dwork);
1511 cancel_work_sync(&ndev_ctx->work);
1513 /* Stop outbound asap */
1514 netif_tx_disable(net);
1516 unregister_netdev(net);
1519 * Call to the vsc driver to let it know that the device is being
1522 rndis_filter_device_remove(dev);
1524 hv_set_drvdata(dev, NULL);
1526 netvsc_free_netdev(net);
1530 static const struct hv_vmbus_device_id id_table[] = {
1536 MODULE_DEVICE_TABLE(vmbus, id_table);
1538 /* The one and only one */
1539 static struct hv_driver netvsc_drv = {
1540 .name = KBUILD_MODNAME,
1541 .id_table = id_table,
1542 .probe = netvsc_probe,
1543 .remove = netvsc_remove,
1547 * On Hyper-V, every VF interface is matched with a corresponding
1548 * synthetic interface. The synthetic interface is presented first
1549 * to the guest. When the corresponding VF instance is registered,
1550 * we will take care of switching the data path.
1552 static int netvsc_netdev_event(struct notifier_block *this,
1553 unsigned long event, void *ptr)
1555 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
1557 /* Skip our own events */
1558 if (event_dev->netdev_ops == &device_ops)
1561 /* Avoid non-Ethernet type devices */
1562 if (event_dev->type != ARPHRD_ETHER)
1565 /* Avoid Vlan dev with same MAC registering as VF */
1566 if (event_dev->priv_flags & IFF_802_1Q_VLAN)
1569 /* Avoid Bonding master dev with same MAC registering as VF */
1570 if ((event_dev->priv_flags & IFF_BONDING) &&
1571 (event_dev->flags & IFF_MASTER))
1575 case NETDEV_REGISTER:
1576 return netvsc_register_vf(event_dev);
1577 case NETDEV_UNREGISTER:
1578 return netvsc_unregister_vf(event_dev);
1580 return netvsc_vf_up(event_dev);
1582 return netvsc_vf_down(event_dev);
1588 static struct notifier_block netvsc_netdev_notifier = {
1589 .notifier_call = netvsc_netdev_event,
1592 static void __exit netvsc_drv_exit(void)
1594 unregister_netdevice_notifier(&netvsc_netdev_notifier);
1595 vmbus_driver_unregister(&netvsc_drv);
1598 static int __init netvsc_drv_init(void)
1602 if (ring_size < RING_SIZE_MIN) {
1603 ring_size = RING_SIZE_MIN;
1604 pr_info("Increased ring_size to %d (min allowed)\n",
1607 ret = vmbus_driver_register(&netvsc_drv);
1612 register_netdevice_notifier(&netvsc_netdev_notifier);
1616 MODULE_LICENSE("GPL");
1617 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1619 module_init(netvsc_drv_init);
1620 module_exit(netvsc_drv_exit);