]> git.karo-electronics.de Git - karo-tx-linux.git/blob - drivers/net/hyperv/netvsc_drv.c
Merge branch 'drm-next' of git://people.freedesktop.org/~airlied/linux
[karo-tx-linux.git] / drivers / net / hyperv / netvsc_drv.c
1 /*
2  * Copyright (c) 2009, Microsoft Corporation.
3  *
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.
7  *
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
11  * more details.
12  *
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/>.
15  *
16  * Authors:
17  *   Haiyang Zhang <haiyangz@microsoft.com>
18  *   Hank Janssen  <hjanssen@microsoft.com>
19  */
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21
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>
27 #include <linux/io.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>
34 #include <linux/in.h>
35 #include <linux/slab.h>
36 #include <net/arp.h>
37 #include <net/route.h>
38 #include <net/sock.h>
39 #include <net/pkt_sched.h>
40
41 #include "hyperv_net.h"
42
43
44 #define RING_SIZE_MIN 64
45 #define LINKCHANGE_INT (2 * HZ)
46 #define NETVSC_HW_FEATURES      (NETIF_F_RXCSUM | \
47                                  NETIF_F_SG | \
48                                  NETIF_F_TSO | \
49                                  NETIF_F_TSO6 | \
50                                  NETIF_F_HW_CSUM)
51 static int ring_size = 128;
52 module_param(ring_size, int, S_IRUGO);
53 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
54
55 static int max_num_vrss_chns = 8;
56
57 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
58                                 NETIF_MSG_LINK | NETIF_MSG_IFUP |
59                                 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
60                                 NETIF_MSG_TX_ERR;
61
62 static int debug = -1;
63 module_param(debug, int, S_IRUGO);
64 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
65
66 static void do_set_multicast(struct work_struct *w)
67 {
68         struct net_device_context *ndevctx =
69                 container_of(w, struct net_device_context, work);
70         struct netvsc_device *nvdev;
71         struct rndis_device *rdev;
72
73         nvdev = hv_get_drvdata(ndevctx->device_ctx);
74         if (nvdev == NULL || nvdev->ndev == NULL)
75                 return;
76
77         rdev = nvdev->extension;
78         if (rdev == NULL)
79                 return;
80
81         if (nvdev->ndev->flags & IFF_PROMISC)
82                 rndis_filter_set_packet_filter(rdev,
83                         NDIS_PACKET_TYPE_PROMISCUOUS);
84         else
85                 rndis_filter_set_packet_filter(rdev,
86                         NDIS_PACKET_TYPE_BROADCAST |
87                         NDIS_PACKET_TYPE_ALL_MULTICAST |
88                         NDIS_PACKET_TYPE_DIRECTED);
89 }
90
91 static void netvsc_set_multicast_list(struct net_device *net)
92 {
93         struct net_device_context *net_device_ctx = netdev_priv(net);
94
95         schedule_work(&net_device_ctx->work);
96 }
97
98 static int netvsc_open(struct net_device *net)
99 {
100         struct net_device_context *net_device_ctx = netdev_priv(net);
101         struct hv_device *device_obj = net_device_ctx->device_ctx;
102         struct netvsc_device *nvdev;
103         struct rndis_device *rdev;
104         int ret = 0;
105
106         netif_carrier_off(net);
107
108         /* Open up the device */
109         ret = rndis_filter_open(device_obj);
110         if (ret != 0) {
111                 netdev_err(net, "unable to open device (ret %d).\n", ret);
112                 return ret;
113         }
114
115         netif_tx_wake_all_queues(net);
116
117         nvdev = hv_get_drvdata(device_obj);
118         rdev = nvdev->extension;
119         if (!rdev->link_state)
120                 netif_carrier_on(net);
121
122         return ret;
123 }
124
125 static int netvsc_close(struct net_device *net)
126 {
127         struct net_device_context *net_device_ctx = netdev_priv(net);
128         struct hv_device *device_obj = net_device_ctx->device_ctx;
129         struct netvsc_device *nvdev = hv_get_drvdata(device_obj);
130         int ret;
131         u32 aread, awrite, i, msec = 10, retry = 0, retry_max = 20;
132         struct vmbus_channel *chn;
133
134         netif_tx_disable(net);
135
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(device_obj);
139         if (ret != 0) {
140                 netdev_err(net, "unable to close device (ret %d).\n", ret);
141                 return ret;
142         }
143
144         /* Ensure pending bytes in ring are read */
145         while (true) {
146                 aread = 0;
147                 for (i = 0; i < nvdev->num_chn; i++) {
148                         chn = nvdev->chn_table[i];
149                         if (!chn)
150                                 continue;
151
152                         hv_get_ringbuffer_availbytes(&chn->inbound, &aread,
153                                                      &awrite);
154
155                         if (aread)
156                                 break;
157
158                         hv_get_ringbuffer_availbytes(&chn->outbound, &aread,
159                                                      &awrite);
160
161                         if (aread)
162                                 break;
163                 }
164
165                 retry++;
166                 if (retry > retry_max || aread == 0)
167                         break;
168
169                 msleep(msec);
170
171                 if (msec < 1000)
172                         msec *= 2;
173         }
174
175         if (aread) {
176                 netdev_err(net, "Ring buffer not empty after closing rndis\n");
177                 ret = -ETIMEDOUT;
178         }
179
180         return ret;
181 }
182
183 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
184                                 int pkt_type)
185 {
186         struct rndis_packet *rndis_pkt;
187         struct rndis_per_packet_info *ppi;
188
189         rndis_pkt = &msg->msg.pkt;
190         rndis_pkt->data_offset += ppi_size;
191
192         ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
193                 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
194
195         ppi->size = ppi_size;
196         ppi->type = pkt_type;
197         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
198
199         rndis_pkt->per_pkt_info_len += ppi_size;
200
201         return ppi;
202 }
203
204 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
205                         void *accel_priv, select_queue_fallback_t fallback)
206 {
207         struct net_device_context *net_device_ctx = netdev_priv(ndev);
208         struct hv_device *hdev =  net_device_ctx->device_ctx;
209         struct netvsc_device *nvsc_dev = hv_get_drvdata(hdev);
210         u32 hash;
211         u16 q_idx = 0;
212
213         if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1)
214                 return 0;
215
216         hash = skb_get_hash(skb);
217         q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] %
218                 ndev->real_num_tx_queues;
219
220         if (!nvsc_dev->chn_table[q_idx])
221                 q_idx = 0;
222
223         return q_idx;
224 }
225
226 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
227                         struct hv_page_buffer *pb)
228 {
229         int j = 0;
230
231         /* Deal with compund pages by ignoring unused part
232          * of the page.
233          */
234         page += (offset >> PAGE_SHIFT);
235         offset &= ~PAGE_MASK;
236
237         while (len > 0) {
238                 unsigned long bytes;
239
240                 bytes = PAGE_SIZE - offset;
241                 if (bytes > len)
242                         bytes = len;
243                 pb[j].pfn = page_to_pfn(page);
244                 pb[j].offset = offset;
245                 pb[j].len = bytes;
246
247                 offset += bytes;
248                 len -= bytes;
249
250                 if (offset == PAGE_SIZE && len) {
251                         page++;
252                         offset = 0;
253                         j++;
254                 }
255         }
256
257         return j + 1;
258 }
259
260 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
261                            struct hv_netvsc_packet *packet,
262                            struct hv_page_buffer **page_buf)
263 {
264         struct hv_page_buffer *pb = *page_buf;
265         u32 slots_used = 0;
266         char *data = skb->data;
267         int frags = skb_shinfo(skb)->nr_frags;
268         int i;
269
270         /* The packet is laid out thus:
271          * 1. hdr: RNDIS header and PPI
272          * 2. skb linear data
273          * 3. skb fragment data
274          */
275         if (hdr != NULL)
276                 slots_used += fill_pg_buf(virt_to_page(hdr),
277                                         offset_in_page(hdr),
278                                         len, &pb[slots_used]);
279
280         packet->rmsg_size = len;
281         packet->rmsg_pgcnt = slots_used;
282
283         slots_used += fill_pg_buf(virt_to_page(data),
284                                 offset_in_page(data),
285                                 skb_headlen(skb), &pb[slots_used]);
286
287         for (i = 0; i < frags; i++) {
288                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
289
290                 slots_used += fill_pg_buf(skb_frag_page(frag),
291                                         frag->page_offset,
292                                         skb_frag_size(frag), &pb[slots_used]);
293         }
294         return slots_used;
295 }
296
297 static int count_skb_frag_slots(struct sk_buff *skb)
298 {
299         int i, frags = skb_shinfo(skb)->nr_frags;
300         int pages = 0;
301
302         for (i = 0; i < frags; i++) {
303                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
304                 unsigned long size = skb_frag_size(frag);
305                 unsigned long offset = frag->page_offset;
306
307                 /* Skip unused frames from start of page */
308                 offset &= ~PAGE_MASK;
309                 pages += PFN_UP(offset + size);
310         }
311         return pages;
312 }
313
314 static int netvsc_get_slots(struct sk_buff *skb)
315 {
316         char *data = skb->data;
317         unsigned int offset = offset_in_page(data);
318         unsigned int len = skb_headlen(skb);
319         int slots;
320         int frag_slots;
321
322         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
323         frag_slots = count_skb_frag_slots(skb);
324         return slots + frag_slots;
325 }
326
327 static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
328 {
329         u32 ret_val = TRANSPORT_INFO_NOT_IP;
330
331         if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
332                 (eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
333                 goto not_ip;
334         }
335
336         *trans_off = skb_transport_offset(skb);
337
338         if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
339                 struct iphdr *iphdr = ip_hdr(skb);
340
341                 if (iphdr->protocol == IPPROTO_TCP)
342                         ret_val = TRANSPORT_INFO_IPV4_TCP;
343                 else if (iphdr->protocol == IPPROTO_UDP)
344                         ret_val = TRANSPORT_INFO_IPV4_UDP;
345         } else {
346                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
347                         ret_val = TRANSPORT_INFO_IPV6_TCP;
348                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
349                         ret_val = TRANSPORT_INFO_IPV6_UDP;
350         }
351
352 not_ip:
353         return ret_val;
354 }
355
356 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
357 {
358         struct net_device_context *net_device_ctx = netdev_priv(net);
359         struct hv_netvsc_packet *packet = NULL;
360         int ret;
361         unsigned int num_data_pgs;
362         struct rndis_message *rndis_msg;
363         struct rndis_packet *rndis_pkt;
364         u32 rndis_msg_size;
365         bool isvlan;
366         bool linear = false;
367         struct rndis_per_packet_info *ppi;
368         struct ndis_tcp_ip_checksum_info *csum_info;
369         struct ndis_tcp_lso_info *lso_info;
370         int  hdr_offset;
371         u32 net_trans_info;
372         u32 hash;
373         u32 skb_length;
374         struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
375         struct hv_page_buffer *pb = page_buf;
376         struct netvsc_stats *tx_stats = this_cpu_ptr(net_device_ctx->tx_stats);
377
378         /* We will atmost need two pages to describe the rndis
379          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
380          * of pages in a single packet. If skb is scattered around
381          * more pages we try linearizing it.
382          */
383
384 check_size:
385         skb_length = skb->len;
386         num_data_pgs = netvsc_get_slots(skb) + 2;
387         if (num_data_pgs > MAX_PAGE_BUFFER_COUNT && linear) {
388                 net_alert_ratelimited("packet too big: %u pages (%u bytes)\n",
389                                       num_data_pgs, skb->len);
390                 ret = -EFAULT;
391                 goto drop;
392         } else if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
393                 if (skb_linearize(skb)) {
394                         net_alert_ratelimited("failed to linearize skb\n");
395                         ret = -ENOMEM;
396                         goto drop;
397                 }
398                 linear = true;
399                 goto check_size;
400         }
401
402         /*
403          * Place the rndis header in the skb head room and
404          * the skb->cb will be used for hv_netvsc_packet
405          * structure.
406          */
407         ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
408         if (ret) {
409                 netdev_err(net, "unable to alloc hv_netvsc_packet\n");
410                 ret = -ENOMEM;
411                 goto drop;
412         }
413         /* Use the skb control buffer for building up the packet */
414         BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
415                         FIELD_SIZEOF(struct sk_buff, cb));
416         packet = (struct hv_netvsc_packet *)skb->cb;
417
418
419         packet->q_idx = skb_get_queue_mapping(skb);
420
421         packet->total_data_buflen = skb->len;
422
423         rndis_msg = (struct rndis_message *)skb->head;
424
425         memset(rndis_msg, 0, RNDIS_AND_PPI_SIZE);
426
427         isvlan = skb->vlan_tci & VLAN_TAG_PRESENT;
428
429         /* Add the rndis header */
430         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
431         rndis_msg->msg_len = packet->total_data_buflen;
432         rndis_pkt = &rndis_msg->msg.pkt;
433         rndis_pkt->data_offset = sizeof(struct rndis_packet);
434         rndis_pkt->data_len = packet->total_data_buflen;
435         rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
436
437         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
438
439         hash = skb_get_hash_raw(skb);
440         if (hash != 0 && net->real_num_tx_queues > 1) {
441                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
442                 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
443                                     NBL_HASH_VALUE);
444                 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
445         }
446
447         if (isvlan) {
448                 struct ndis_pkt_8021q_info *vlan;
449
450                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
451                 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
452                                         IEEE_8021Q_INFO);
453                 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
454                                                 ppi->ppi_offset);
455                 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
456                 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
457                                 VLAN_PRIO_SHIFT;
458         }
459
460         net_trans_info = get_net_transport_info(skb, &hdr_offset);
461         if (net_trans_info == TRANSPORT_INFO_NOT_IP)
462                 goto do_send;
463
464         /*
465          * Setup the sendside checksum offload only if this is not a
466          * GSO packet.
467          */
468         if (skb_is_gso(skb))
469                 goto do_lso;
470
471         if ((skb->ip_summed == CHECKSUM_NONE) ||
472             (skb->ip_summed == CHECKSUM_UNNECESSARY))
473                 goto do_send;
474
475         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
476         ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
477                             TCPIP_CHKSUM_PKTINFO);
478
479         csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
480                         ppi->ppi_offset);
481
482         if (net_trans_info & (INFO_IPV4 << 16))
483                 csum_info->transmit.is_ipv4 = 1;
484         else
485                 csum_info->transmit.is_ipv6 = 1;
486
487         if (net_trans_info & INFO_TCP) {
488                 csum_info->transmit.tcp_checksum = 1;
489                 csum_info->transmit.tcp_header_offset = hdr_offset;
490         } else if (net_trans_info & INFO_UDP) {
491                 /* UDP checksum offload is not supported on ws2008r2.
492                  * Furthermore, on ws2012 and ws2012r2, there are some
493                  * issues with udp checksum offload from Linux guests.
494                  * (these are host issues).
495                  * For now compute the checksum here.
496                  */
497                 struct udphdr *uh;
498                 u16 udp_len;
499
500                 ret = skb_cow_head(skb, 0);
501                 if (ret)
502                         goto drop;
503
504                 uh = udp_hdr(skb);
505                 udp_len = ntohs(uh->len);
506                 uh->check = 0;
507                 uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
508                                               ip_hdr(skb)->daddr,
509                                               udp_len, IPPROTO_UDP,
510                                               csum_partial(uh, udp_len, 0));
511                 if (uh->check == 0)
512                         uh->check = CSUM_MANGLED_0;
513
514                 csum_info->transmit.udp_checksum = 0;
515         }
516         goto do_send;
517
518 do_lso:
519         rndis_msg_size += NDIS_LSO_PPI_SIZE;
520         ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
521                             TCP_LARGESEND_PKTINFO);
522
523         lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
524                         ppi->ppi_offset);
525
526         lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
527         if (net_trans_info & (INFO_IPV4 << 16)) {
528                 lso_info->lso_v2_transmit.ip_version =
529                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
530                 ip_hdr(skb)->tot_len = 0;
531                 ip_hdr(skb)->check = 0;
532                 tcp_hdr(skb)->check =
533                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
534                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
535         } else {
536                 lso_info->lso_v2_transmit.ip_version =
537                         NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
538                 ipv6_hdr(skb)->payload_len = 0;
539                 tcp_hdr(skb)->check =
540                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
541                                 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
542         }
543         lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
544         lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
545
546 do_send:
547         /* Start filling in the page buffers with the rndis hdr */
548         rndis_msg->msg_len += rndis_msg_size;
549         packet->total_data_buflen = rndis_msg->msg_len;
550         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
551                                                skb, packet, &pb);
552
553         /* timestamp packet in software */
554         skb_tx_timestamp(skb);
555         ret = netvsc_send(net_device_ctx->device_ctx, packet,
556                           rndis_msg, &pb, skb);
557
558 drop:
559         if (ret == 0) {
560                 u64_stats_update_begin(&tx_stats->syncp);
561                 tx_stats->packets++;
562                 tx_stats->bytes += skb_length;
563                 u64_stats_update_end(&tx_stats->syncp);
564         } else {
565                 if (ret != -EAGAIN) {
566                         dev_kfree_skb_any(skb);
567                         net->stats.tx_dropped++;
568                 }
569         }
570
571         return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
572 }
573
574 /*
575  * netvsc_linkstatus_callback - Link up/down notification
576  */
577 void netvsc_linkstatus_callback(struct hv_device *device_obj,
578                                 struct rndis_message *resp)
579 {
580         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
581         struct net_device *net;
582         struct net_device_context *ndev_ctx;
583         struct netvsc_device *net_device;
584         struct netvsc_reconfig *event;
585         unsigned long flags;
586
587         /* Handle link change statuses only */
588         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
589             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
590             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
591                 return;
592
593         net_device = hv_get_drvdata(device_obj);
594         net = net_device->ndev;
595
596         if (!net || net->reg_state != NETREG_REGISTERED)
597                 return;
598
599         ndev_ctx = netdev_priv(net);
600
601         event = kzalloc(sizeof(*event), GFP_ATOMIC);
602         if (!event)
603                 return;
604         event->event = indicate->status;
605
606         spin_lock_irqsave(&ndev_ctx->lock, flags);
607         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
608         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
609
610         schedule_delayed_work(&ndev_ctx->dwork, 0);
611 }
612
613 /*
614  * netvsc_recv_callback -  Callback when we receive a packet from the
615  * "wire" on the specified device.
616  */
617 int netvsc_recv_callback(struct hv_device *device_obj,
618                                 struct hv_netvsc_packet *packet,
619                                 void **data,
620                                 struct ndis_tcp_ip_checksum_info *csum_info,
621                                 struct vmbus_channel *channel,
622                                 u16 vlan_tci)
623 {
624         struct net_device *net;
625         struct net_device_context *net_device_ctx;
626         struct sk_buff *skb;
627         struct netvsc_stats *rx_stats;
628
629         net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev;
630         if (!net || net->reg_state != NETREG_REGISTERED) {
631                 return NVSP_STAT_FAIL;
632         }
633         net_device_ctx = netdev_priv(net);
634         rx_stats = this_cpu_ptr(net_device_ctx->rx_stats);
635
636         /* Allocate a skb - TODO direct I/O to pages? */
637         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
638         if (unlikely(!skb)) {
639                 ++net->stats.rx_dropped;
640                 return NVSP_STAT_FAIL;
641         }
642
643         /*
644          * Copy to skb. This copy is needed here since the memory pointed by
645          * hv_netvsc_packet cannot be deallocated
646          */
647         memcpy(skb_put(skb, packet->total_data_buflen), *data,
648                 packet->total_data_buflen);
649
650         skb->protocol = eth_type_trans(skb, net);
651         if (csum_info) {
652                 /* We only look at the IP checksum here.
653                  * Should we be dropping the packet if checksum
654                  * failed? How do we deal with other checksums - TCP/UDP?
655                  */
656                 if (csum_info->receive.ip_checksum_succeeded)
657                         skb->ip_summed = CHECKSUM_UNNECESSARY;
658                 else
659                         skb->ip_summed = CHECKSUM_NONE;
660         }
661
662         if (vlan_tci & VLAN_TAG_PRESENT)
663                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
664                                        vlan_tci);
665
666         skb_record_rx_queue(skb, channel->
667                             offermsg.offer.sub_channel_index);
668
669         u64_stats_update_begin(&rx_stats->syncp);
670         rx_stats->packets++;
671         rx_stats->bytes += packet->total_data_buflen;
672         u64_stats_update_end(&rx_stats->syncp);
673
674         /*
675          * Pass the skb back up. Network stack will deallocate the skb when it
676          * is done.
677          * TODO - use NAPI?
678          */
679         netif_rx(skb);
680
681         return 0;
682 }
683
684 static void netvsc_get_drvinfo(struct net_device *net,
685                                struct ethtool_drvinfo *info)
686 {
687         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
688         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
689 }
690
691 static void netvsc_get_channels(struct net_device *net,
692                                 struct ethtool_channels *channel)
693 {
694         struct net_device_context *net_device_ctx = netdev_priv(net);
695         struct hv_device *dev = net_device_ctx->device_ctx;
696         struct netvsc_device *nvdev = hv_get_drvdata(dev);
697
698         if (nvdev) {
699                 channel->max_combined   = nvdev->max_chn;
700                 channel->combined_count = nvdev->num_chn;
701         }
702 }
703
704 static int netvsc_set_channels(struct net_device *net,
705                                struct ethtool_channels *channels)
706 {
707         struct net_device_context *net_device_ctx = netdev_priv(net);
708         struct hv_device *dev = net_device_ctx->device_ctx;
709         struct netvsc_device *nvdev = hv_get_drvdata(dev);
710         struct netvsc_device_info device_info;
711         u32 num_chn;
712         u32 max_chn;
713         int ret = 0;
714         bool recovering = false;
715
716         if (!nvdev || nvdev->destroy)
717                 return -ENODEV;
718
719         num_chn = nvdev->num_chn;
720         max_chn = min_t(u32, nvdev->max_chn, num_online_cpus());
721
722         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5) {
723                 pr_info("vRSS unsupported before NVSP Version 5\n");
724                 return -EINVAL;
725         }
726
727         /* We do not support rx, tx, or other */
728         if (!channels ||
729             channels->rx_count ||
730             channels->tx_count ||
731             channels->other_count ||
732             (channels->combined_count < 1))
733                 return -EINVAL;
734
735         if (channels->combined_count > max_chn) {
736                 pr_info("combined channels too high, using %d\n", max_chn);
737                 channels->combined_count = max_chn;
738         }
739
740         ret = netvsc_close(net);
741         if (ret)
742                 goto out;
743
744  do_set:
745         nvdev->start_remove = true;
746         rndis_filter_device_remove(dev);
747
748         nvdev->num_chn = channels->combined_count;
749
750         net_device_ctx->device_ctx = dev;
751         hv_set_drvdata(dev, net);
752
753         memset(&device_info, 0, sizeof(device_info));
754         device_info.num_chn = nvdev->num_chn; /* passed to RNDIS */
755         device_info.ring_size = ring_size;
756         device_info.max_num_vrss_chns = max_num_vrss_chns;
757
758         ret = rndis_filter_device_add(dev, &device_info);
759         if (ret) {
760                 if (recovering) {
761                         netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
762                         return ret;
763                 }
764                 goto recover;
765         }
766
767         nvdev = hv_get_drvdata(dev);
768
769         ret = netif_set_real_num_tx_queues(net, nvdev->num_chn);
770         if (ret) {
771                 if (recovering) {
772                         netdev_err(net, "could not set tx queue count (ret %d)\n", ret);
773                         return ret;
774                 }
775                 goto recover;
776         }
777
778         ret = netif_set_real_num_rx_queues(net, nvdev->num_chn);
779         if (ret) {
780                 if (recovering) {
781                         netdev_err(net, "could not set rx queue count (ret %d)\n", ret);
782                         return ret;
783                 }
784                 goto recover;
785         }
786
787  out:
788         netvsc_open(net);
789
790         return ret;
791
792  recover:
793         /* If the above failed, we attempt to recover through the same
794          * process but with the original number of channels.
795          */
796         netdev_err(net, "could not set channels, recovering\n");
797         recovering = true;
798         channels->combined_count = num_chn;
799         goto do_set;
800 }
801
802 static bool netvsc_validate_ethtool_ss_cmd(const struct ethtool_cmd *cmd)
803 {
804         struct ethtool_cmd diff1 = *cmd;
805         struct ethtool_cmd diff2 = {};
806
807         ethtool_cmd_speed_set(&diff1, 0);
808         diff1.duplex = 0;
809         /* advertising and cmd are usually set */
810         diff1.advertising = 0;
811         diff1.cmd = 0;
812         /* We set port to PORT_OTHER */
813         diff2.port = PORT_OTHER;
814
815         return !memcmp(&diff1, &diff2, sizeof(diff1));
816 }
817
818 static void netvsc_init_settings(struct net_device *dev)
819 {
820         struct net_device_context *ndc = netdev_priv(dev);
821
822         ndc->speed = SPEED_UNKNOWN;
823         ndc->duplex = DUPLEX_UNKNOWN;
824 }
825
826 static int netvsc_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
827 {
828         struct net_device_context *ndc = netdev_priv(dev);
829
830         ethtool_cmd_speed_set(cmd, ndc->speed);
831         cmd->duplex = ndc->duplex;
832         cmd->port = PORT_OTHER;
833
834         return 0;
835 }
836
837 static int netvsc_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
838 {
839         struct net_device_context *ndc = netdev_priv(dev);
840         u32 speed;
841
842         speed = ethtool_cmd_speed(cmd);
843         if (!ethtool_validate_speed(speed) ||
844             !ethtool_validate_duplex(cmd->duplex) ||
845             !netvsc_validate_ethtool_ss_cmd(cmd))
846                 return -EINVAL;
847
848         ndc->speed = speed;
849         ndc->duplex = cmd->duplex;
850
851         return 0;
852 }
853
854 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
855 {
856         struct net_device_context *ndevctx = netdev_priv(ndev);
857         struct hv_device *hdev =  ndevctx->device_ctx;
858         struct netvsc_device *nvdev = hv_get_drvdata(hdev);
859         struct netvsc_device_info device_info;
860         int limit = ETH_DATA_LEN;
861         int ret = 0;
862
863         if (nvdev == NULL || nvdev->destroy)
864                 return -ENODEV;
865
866         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
867                 limit = NETVSC_MTU - ETH_HLEN;
868
869         if (mtu < NETVSC_MTU_MIN || mtu > limit)
870                 return -EINVAL;
871
872         ret = netvsc_close(ndev);
873         if (ret)
874                 goto out;
875
876         nvdev->start_remove = true;
877         rndis_filter_device_remove(hdev);
878
879         ndev->mtu = mtu;
880
881         ndevctx->device_ctx = hdev;
882         hv_set_drvdata(hdev, ndev);
883
884         memset(&device_info, 0, sizeof(device_info));
885         device_info.ring_size = ring_size;
886         device_info.num_chn = nvdev->num_chn;
887         device_info.max_num_vrss_chns = max_num_vrss_chns;
888         rndis_filter_device_add(hdev, &device_info);
889
890 out:
891         netvsc_open(ndev);
892
893         return ret;
894 }
895
896 static struct rtnl_link_stats64 *netvsc_get_stats64(struct net_device *net,
897                                                     struct rtnl_link_stats64 *t)
898 {
899         struct net_device_context *ndev_ctx = netdev_priv(net);
900         int cpu;
901
902         for_each_possible_cpu(cpu) {
903                 struct netvsc_stats *tx_stats = per_cpu_ptr(ndev_ctx->tx_stats,
904                                                             cpu);
905                 struct netvsc_stats *rx_stats = per_cpu_ptr(ndev_ctx->rx_stats,
906                                                             cpu);
907                 u64 tx_packets, tx_bytes, rx_packets, rx_bytes;
908                 unsigned int start;
909
910                 do {
911                         start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
912                         tx_packets = tx_stats->packets;
913                         tx_bytes = tx_stats->bytes;
914                 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
915
916                 do {
917                         start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
918                         rx_packets = rx_stats->packets;
919                         rx_bytes = rx_stats->bytes;
920                 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
921
922                 t->tx_bytes     += tx_bytes;
923                 t->tx_packets   += tx_packets;
924                 t->rx_bytes     += rx_bytes;
925                 t->rx_packets   += rx_packets;
926         }
927
928         t->tx_dropped   = net->stats.tx_dropped;
929         t->tx_errors    = net->stats.tx_dropped;
930
931         t->rx_dropped   = net->stats.rx_dropped;
932         t->rx_errors    = net->stats.rx_errors;
933
934         return t;
935 }
936
937 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
938 {
939         struct net_device_context *ndevctx = netdev_priv(ndev);
940         struct hv_device *hdev =  ndevctx->device_ctx;
941         struct sockaddr *addr = p;
942         char save_adr[ETH_ALEN];
943         unsigned char save_aatype;
944         int err;
945
946         memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
947         save_aatype = ndev->addr_assign_type;
948
949         err = eth_mac_addr(ndev, p);
950         if (err != 0)
951                 return err;
952
953         err = rndis_filter_set_device_mac(hdev, addr->sa_data);
954         if (err != 0) {
955                 /* roll back to saved MAC */
956                 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
957                 ndev->addr_assign_type = save_aatype;
958         }
959
960         return err;
961 }
962
963 #ifdef CONFIG_NET_POLL_CONTROLLER
964 static void netvsc_poll_controller(struct net_device *net)
965 {
966         /* As netvsc_start_xmit() works synchronous we don't have to
967          * trigger anything here.
968          */
969 }
970 #endif
971
972 static const struct ethtool_ops ethtool_ops = {
973         .get_drvinfo    = netvsc_get_drvinfo,
974         .get_link       = ethtool_op_get_link,
975         .get_channels   = netvsc_get_channels,
976         .set_channels   = netvsc_set_channels,
977         .get_ts_info    = ethtool_op_get_ts_info,
978         .get_settings   = netvsc_get_settings,
979         .set_settings   = netvsc_set_settings,
980 };
981
982 static const struct net_device_ops device_ops = {
983         .ndo_open =                     netvsc_open,
984         .ndo_stop =                     netvsc_close,
985         .ndo_start_xmit =               netvsc_start_xmit,
986         .ndo_set_rx_mode =              netvsc_set_multicast_list,
987         .ndo_change_mtu =               netvsc_change_mtu,
988         .ndo_validate_addr =            eth_validate_addr,
989         .ndo_set_mac_address =          netvsc_set_mac_addr,
990         .ndo_select_queue =             netvsc_select_queue,
991         .ndo_get_stats64 =              netvsc_get_stats64,
992 #ifdef CONFIG_NET_POLL_CONTROLLER
993         .ndo_poll_controller =          netvsc_poll_controller,
994 #endif
995 };
996
997 /*
998  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
999  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1000  * present send GARP packet to network peers with netif_notify_peers().
1001  */
1002 static void netvsc_link_change(struct work_struct *w)
1003 {
1004         struct net_device_context *ndev_ctx;
1005         struct net_device *net;
1006         struct netvsc_device *net_device;
1007         struct rndis_device *rdev;
1008         struct netvsc_reconfig *event = NULL;
1009         bool notify = false, reschedule = false;
1010         unsigned long flags, next_reconfig, delay;
1011
1012         ndev_ctx = container_of(w, struct net_device_context, dwork.work);
1013         net_device = hv_get_drvdata(ndev_ctx->device_ctx);
1014         rdev = net_device->extension;
1015         net = net_device->ndev;
1016
1017         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1018         if (time_is_after_jiffies(next_reconfig)) {
1019                 /* link_watch only sends one notification with current state
1020                  * per second, avoid doing reconfig more frequently. Handle
1021                  * wrap around.
1022                  */
1023                 delay = next_reconfig - jiffies;
1024                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1025                 schedule_delayed_work(&ndev_ctx->dwork, delay);
1026                 return;
1027         }
1028         ndev_ctx->last_reconfig = jiffies;
1029
1030         spin_lock_irqsave(&ndev_ctx->lock, flags);
1031         if (!list_empty(&ndev_ctx->reconfig_events)) {
1032                 event = list_first_entry(&ndev_ctx->reconfig_events,
1033                                          struct netvsc_reconfig, list);
1034                 list_del(&event->list);
1035                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1036         }
1037         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1038
1039         if (!event)
1040                 return;
1041
1042         rtnl_lock();
1043
1044         switch (event->event) {
1045                 /* Only the following events are possible due to the check in
1046                  * netvsc_linkstatus_callback()
1047                  */
1048         case RNDIS_STATUS_MEDIA_CONNECT:
1049                 if (rdev->link_state) {
1050                         rdev->link_state = false;
1051                         netif_carrier_on(net);
1052                         netif_tx_wake_all_queues(net);
1053                 } else {
1054                         notify = true;
1055                 }
1056                 kfree(event);
1057                 break;
1058         case RNDIS_STATUS_MEDIA_DISCONNECT:
1059                 if (!rdev->link_state) {
1060                         rdev->link_state = true;
1061                         netif_carrier_off(net);
1062                         netif_tx_stop_all_queues(net);
1063                 }
1064                 kfree(event);
1065                 break;
1066         case RNDIS_STATUS_NETWORK_CHANGE:
1067                 /* Only makes sense if carrier is present */
1068                 if (!rdev->link_state) {
1069                         rdev->link_state = true;
1070                         netif_carrier_off(net);
1071                         netif_tx_stop_all_queues(net);
1072                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
1073                         spin_lock_irqsave(&ndev_ctx->lock, flags);
1074                         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
1075                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1076                         reschedule = true;
1077                 }
1078                 break;
1079         }
1080
1081         rtnl_unlock();
1082
1083         if (notify)
1084                 netdev_notify_peers(net);
1085
1086         /* link_watch only sends one notification with current state per
1087          * second, handle next reconfig event in 2 seconds.
1088          */
1089         if (reschedule)
1090                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1091 }
1092
1093 static void netvsc_free_netdev(struct net_device *netdev)
1094 {
1095         struct net_device_context *net_device_ctx = netdev_priv(netdev);
1096
1097         free_percpu(net_device_ctx->tx_stats);
1098         free_percpu(net_device_ctx->rx_stats);
1099         free_netdev(netdev);
1100 }
1101
1102 static int netvsc_probe(struct hv_device *dev,
1103                         const struct hv_vmbus_device_id *dev_id)
1104 {
1105         struct net_device *net = NULL;
1106         struct net_device_context *net_device_ctx;
1107         struct netvsc_device_info device_info;
1108         struct netvsc_device *nvdev;
1109         int ret;
1110
1111         net = alloc_etherdev_mq(sizeof(struct net_device_context),
1112                                 num_online_cpus());
1113         if (!net)
1114                 return -ENOMEM;
1115
1116         netif_carrier_off(net);
1117
1118         net_device_ctx = netdev_priv(net);
1119         net_device_ctx->device_ctx = dev;
1120         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1121         if (netif_msg_probe(net_device_ctx))
1122                 netdev_dbg(net, "netvsc msg_enable: %d\n",
1123                            net_device_ctx->msg_enable);
1124
1125         net_device_ctx->tx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1126         if (!net_device_ctx->tx_stats) {
1127                 free_netdev(net);
1128                 return -ENOMEM;
1129         }
1130         net_device_ctx->rx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1131         if (!net_device_ctx->rx_stats) {
1132                 free_percpu(net_device_ctx->tx_stats);
1133                 free_netdev(net);
1134                 return -ENOMEM;
1135         }
1136
1137         hv_set_drvdata(dev, net);
1138         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1139         INIT_WORK(&net_device_ctx->work, do_set_multicast);
1140
1141         spin_lock_init(&net_device_ctx->lock);
1142         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
1143
1144         net->netdev_ops = &device_ops;
1145
1146         net->hw_features = NETVSC_HW_FEATURES;
1147         net->features = NETVSC_HW_FEATURES | NETIF_F_HW_VLAN_CTAG_TX;
1148
1149         net->ethtool_ops = &ethtool_ops;
1150         SET_NETDEV_DEV(net, &dev->device);
1151
1152         /* We always need headroom for rndis header */
1153         net->needed_headroom = RNDIS_AND_PPI_SIZE;
1154
1155         /* Notify the netvsc driver of the new device */
1156         memset(&device_info, 0, sizeof(device_info));
1157         device_info.ring_size = ring_size;
1158         device_info.max_num_vrss_chns = max_num_vrss_chns;
1159         ret = rndis_filter_device_add(dev, &device_info);
1160         if (ret != 0) {
1161                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1162                 netvsc_free_netdev(net);
1163                 hv_set_drvdata(dev, NULL);
1164                 return ret;
1165         }
1166         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1167
1168         nvdev = hv_get_drvdata(dev);
1169         netif_set_real_num_tx_queues(net, nvdev->num_chn);
1170         netif_set_real_num_rx_queues(net, nvdev->num_chn);
1171
1172         netvsc_init_settings(net);
1173
1174         ret = register_netdev(net);
1175         if (ret != 0) {
1176                 pr_err("Unable to register netdev.\n");
1177                 rndis_filter_device_remove(dev);
1178                 netvsc_free_netdev(net);
1179         }
1180
1181         return ret;
1182 }
1183
1184 static int netvsc_remove(struct hv_device *dev)
1185 {
1186         struct net_device *net;
1187         struct net_device_context *ndev_ctx;
1188         struct netvsc_device *net_device;
1189
1190         net_device = hv_get_drvdata(dev);
1191         net = net_device->ndev;
1192
1193         if (net == NULL) {
1194                 dev_err(&dev->device, "No net device to remove\n");
1195                 return 0;
1196         }
1197
1198         net_device->start_remove = true;
1199
1200         ndev_ctx = netdev_priv(net);
1201         cancel_delayed_work_sync(&ndev_ctx->dwork);
1202         cancel_work_sync(&ndev_ctx->work);
1203
1204         /* Stop outbound asap */
1205         netif_tx_disable(net);
1206
1207         unregister_netdev(net);
1208
1209         /*
1210          * Call to the vsc driver to let it know that the device is being
1211          * removed
1212          */
1213         rndis_filter_device_remove(dev);
1214
1215         netvsc_free_netdev(net);
1216         return 0;
1217 }
1218
1219 static const struct hv_vmbus_device_id id_table[] = {
1220         /* Network guid */
1221         { HV_NIC_GUID, },
1222         { },
1223 };
1224
1225 MODULE_DEVICE_TABLE(vmbus, id_table);
1226
1227 /* The one and only one */
1228 static struct  hv_driver netvsc_drv = {
1229         .name = KBUILD_MODNAME,
1230         .id_table = id_table,
1231         .probe = netvsc_probe,
1232         .remove = netvsc_remove,
1233 };
1234
1235 static void __exit netvsc_drv_exit(void)
1236 {
1237         vmbus_driver_unregister(&netvsc_drv);
1238 }
1239
1240 static int __init netvsc_drv_init(void)
1241 {
1242         if (ring_size < RING_SIZE_MIN) {
1243                 ring_size = RING_SIZE_MIN;
1244                 pr_info("Increased ring_size to %d (min allowed)\n",
1245                         ring_size);
1246         }
1247         return vmbus_driver_register(&netvsc_drv);
1248 }
1249
1250 MODULE_LICENSE("GPL");
1251 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1252
1253 module_init(netvsc_drv_init);
1254 module_exit(netvsc_drv_exit);