]> CyberLeo.Net >> Repos - FreeBSD/releng/10.2.git/blob - sys/dev/hyperv/netvsc/hv_netvsc_drv_freebsd.c
- Copy stable/10@285827 to releng/10.2 in preparation for 10.2-RC1
[FreeBSD/releng/10.2.git] / sys / dev / hyperv / netvsc / hv_netvsc_drv_freebsd.c
1 /*-
2  * Copyright (c) 2010-2012 Citrix Inc.
3  * Copyright (c) 2009-2012 Microsoft Corp.
4  * Copyright (c) 2012 NetApp Inc.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice unmodified, this list of conditions, and the following
12  *    disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  */
28
29 /*-
30  * Copyright (c) 2004-2006 Kip Macy
31  * All rights reserved.
32  *
33  * Redistribution and use in source and binary forms, with or without
34  * modification, are permitted provided that the following conditions
35  * are met:
36  * 1. Redistributions of source code must retain the above copyright
37  *    notice, this list of conditions and the following disclaimer.
38  * 2. Redistributions in binary form must reproduce the above copyright
39  *    notice, this list of conditions and the following disclaimer in the
40  *    documentation and/or other materials provided with the distribution.
41  *
42  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
43  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
44  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
45  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
46  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
47  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
48  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
49  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
50  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
51  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
52  * SUCH DAMAGE.
53  */
54
55 #include <sys/cdefs.h>
56 __FBSDID("$FreeBSD$");
57
58 #include "opt_inet6.h"
59 #include "opt_inet.h"
60
61 #include <sys/param.h>
62 #include <sys/systm.h>
63 #include <sys/sockio.h>
64 #include <sys/mbuf.h>
65 #include <sys/malloc.h>
66 #include <sys/module.h>
67 #include <sys/kernel.h>
68 #include <sys/socket.h>
69 #include <sys/queue.h>
70 #include <sys/lock.h>
71 #include <sys/sx.h>
72
73 #include <net/if.h>
74 #include <net/if_arp.h>
75 #include <net/ethernet.h>
76 #include <net/if_dl.h>
77 #include <net/if_media.h>
78
79 #include <net/bpf.h>
80
81 #include <net/if_types.h>
82 #include <net/if_vlan_var.h>
83 #include <net/if.h>
84
85 #include <netinet/in_systm.h>
86 #include <netinet/in.h>
87 #include <netinet/ip.h>
88 #include <netinet/if_ether.h>
89 #include <netinet/tcp.h>
90 #include <netinet/udp.h>
91 #include <netinet/ip6.h>
92
93 #include <vm/vm.h>
94 #include <vm/vm_param.h>
95 #include <vm/vm_kern.h>
96 #include <vm/pmap.h>
97
98 #include <machine/bus.h>
99 #include <machine/resource.h>
100 #include <machine/frame.h>
101 #include <machine/vmparam.h>
102
103 #include <sys/bus.h>
104 #include <sys/rman.h>
105 #include <sys/mutex.h>
106 #include <sys/errno.h>
107 #include <sys/types.h>
108 #include <machine/atomic.h>
109
110 #include <machine/intr_machdep.h>
111
112 #include <machine/in_cksum.h>
113
114 #include <dev/hyperv/include/hyperv.h>
115 #include "hv_net_vsc.h"
116 #include "hv_rndis.h"
117 #include "hv_rndis_filter.h"
118
119
120 /* Short for Hyper-V network interface */
121 #define NETVSC_DEVNAME    "hn"
122
123 /*
124  * It looks like offset 0 of buf is reserved to hold the softc pointer.
125  * The sc pointer evidently not needed, and is not presently populated.
126  * The packet offset is where the netvsc_packet starts in the buffer.
127  */
128 #define HV_NV_SC_PTR_OFFSET_IN_BUF         0
129 #define HV_NV_PACKET_OFFSET_IN_BUF         16
130
131
132 /*
133  * Data types
134  */
135
136 struct hv_netvsc_driver_context {
137         uint32_t                drv_inited;
138 };
139
140 /*
141  * Be aware that this sleepable mutex will exhibit WITNESS errors when
142  * certain TCP and ARP code paths are taken.  This appears to be a
143  * well-known condition, as all other drivers checked use a sleeping
144  * mutex to protect their transmit paths.
145  * Also Be aware that mutexes do not play well with semaphores, and there
146  * is a conflicting semaphore in a certain channel code path.
147  */
148 #define NV_LOCK_INIT(_sc, _name) \
149             mtx_init(&(_sc)->hn_lock, _name, MTX_NETWORK_LOCK, MTX_DEF)
150 #define NV_LOCK(_sc)            mtx_lock(&(_sc)->hn_lock)
151 #define NV_LOCK_ASSERT(_sc)     mtx_assert(&(_sc)->hn_lock, MA_OWNED)
152 #define NV_UNLOCK(_sc)          mtx_unlock(&(_sc)->hn_lock)
153 #define NV_LOCK_DESTROY(_sc)    mtx_destroy(&(_sc)->hn_lock)
154
155
156 /*
157  * Globals
158  */
159
160 int hv_promisc_mode = 0;    /* normal mode by default */
161
162 /* The one and only one */
163 static struct hv_netvsc_driver_context g_netvsc_drv;
164
165
166 /*
167  * Forward declarations
168  */
169 static void hn_stop(hn_softc_t *sc);
170 static void hn_ifinit_locked(hn_softc_t *sc);
171 static void hn_ifinit(void *xsc);
172 static int  hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data);
173 static int  hn_start_locked(struct ifnet *ifp);
174 static void hn_start(struct ifnet *ifp);
175
176 /*
177  * NetVsc get message transport protocol type 
178  */
179 static uint32_t get_transport_proto_type(struct mbuf *m_head)
180 {
181         uint32_t ret_val = TRANSPORT_TYPE_NOT_IP;
182         uint16_t ether_type = 0;
183         int ether_len = 0;
184         struct ether_vlan_header *eh;
185 #ifdef INET
186         struct ip *iph;
187 #endif
188 #ifdef INET6
189         struct ip6_hdr *ip6;
190 #endif
191
192         eh = mtod(m_head, struct ether_vlan_header*);
193         if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
194                 ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
195                 ether_type = eh->evl_proto;
196         } else {
197                 ether_len = ETHER_HDR_LEN;
198                 ether_type = eh->evl_encap_proto;
199         }
200
201         switch (ntohs(ether_type)) {
202 #ifdef INET6
203         case ETHERTYPE_IPV6:
204                 ip6 = (struct ip6_hdr *)(m_head->m_data + ether_len);
205
206                 if (IPPROTO_TCP == ip6->ip6_nxt) {
207                         ret_val = TRANSPORT_TYPE_IPV6_TCP;
208                 } else if (IPPROTO_UDP == ip6->ip6_nxt) {
209                         ret_val = TRANSPORT_TYPE_IPV6_UDP;
210                 }
211                 break;
212 #endif
213 #ifdef INET
214         case ETHERTYPE_IP:
215                 iph = (struct ip *)(m_head->m_data + ether_len);
216
217                 if (IPPROTO_TCP == iph->ip_p) {
218                         ret_val = TRANSPORT_TYPE_IPV4_TCP;
219                 } else if (IPPROTO_UDP == iph->ip_p) {
220                         ret_val = TRANSPORT_TYPE_IPV4_UDP;
221                 }
222                 break;
223 #endif
224         default:
225                 ret_val = TRANSPORT_TYPE_NOT_IP;
226                 break;
227         }
228
229         return (ret_val);
230 }
231
232 /*
233  * NetVsc driver initialization
234  * Note:  Filter init is no longer required
235  */
236 static int
237 netvsc_drv_init(void)
238 {
239         return (0);
240 }
241
242 /*
243  * NetVsc global initialization entry point
244  */
245 static void
246 netvsc_init(void)
247 {
248         if (bootverbose)
249                 printf("Netvsc initializing... ");
250
251         /*
252          * XXXKYS: cleanup initialization
253          */
254         if (!cold && !g_netvsc_drv.drv_inited) {
255                 g_netvsc_drv.drv_inited = 1;
256                 netvsc_drv_init();
257                 if (bootverbose)
258                         printf("done!\n");
259         } else if (bootverbose)
260                 printf("Already initialized!\n");
261 }
262
263 /* {F8615163-DF3E-46c5-913F-F2D2F965ED0E} */
264 static const hv_guid g_net_vsc_device_type = {
265         .data = {0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
266                 0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E}
267 };
268
269 /*
270  * Standard probe entry point.
271  *
272  */
273 static int
274 netvsc_probe(device_t dev)
275 {
276         const char *p;
277
278         p = vmbus_get_type(dev);
279         if (!memcmp(p, &g_net_vsc_device_type.data, sizeof(hv_guid))) {
280                 device_set_desc(dev, "Synthetic Network Interface");
281                 if (bootverbose)
282                         printf("Netvsc probe... DONE \n");
283
284                 return (BUS_PROBE_DEFAULT);
285         }
286
287         return (ENXIO);
288 }
289
290 /*
291  * Standard attach entry point.
292  *
293  * Called when the driver is loaded.  It allocates needed resources,
294  * and initializes the "hardware" and software.
295  */
296 static int
297 netvsc_attach(device_t dev)
298 {
299         struct hv_device *device_ctx = vmbus_get_devctx(dev);
300         netvsc_device_info device_info;
301         hn_softc_t *sc;
302         int unit = device_get_unit(dev);
303         struct ifnet *ifp;
304         int ret;
305
306         netvsc_init();
307
308         sc = device_get_softc(dev);
309         if (sc == NULL) {
310                 return (ENOMEM);
311         }
312
313         bzero(sc, sizeof(hn_softc_t));
314         sc->hn_unit = unit;
315         sc->hn_dev = dev;
316
317         NV_LOCK_INIT(sc, "NetVSCLock");
318
319         sc->hn_dev_obj = device_ctx;
320
321         ifp = sc->hn_ifp = sc->arpcom.ac_ifp = if_alloc(IFT_ETHER);
322         ifp->if_softc = sc;
323
324         if_initname(ifp, device_get_name(dev), device_get_unit(dev));
325         ifp->if_dunit = unit;
326         ifp->if_dname = NETVSC_DEVNAME;
327
328         ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
329         ifp->if_ioctl = hn_ioctl;
330         ifp->if_start = hn_start;
331         ifp->if_init = hn_ifinit;
332         /* needed by hv_rf_on_device_add() code */
333         ifp->if_mtu = ETHERMTU;
334         IFQ_SET_MAXLEN(&ifp->if_snd, 512);
335         ifp->if_snd.ifq_drv_maxlen = 511;
336         IFQ_SET_READY(&ifp->if_snd);
337
338         /*
339          * Tell upper layers that we support full VLAN capability.
340          */
341         ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
342         ifp->if_capabilities |=
343             IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO;
344         ifp->if_capenable |=
345             IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO;
346         ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_TSO;
347
348         ret = hv_rf_on_device_add(device_ctx, &device_info);
349         if (ret != 0) {
350                 if_free(ifp);
351
352                 return (ret);
353         }
354         if (device_info.link_state == 0) {
355                 sc->hn_carrier = 1;
356         }
357
358         ether_ifattach(ifp, device_info.mac_addr);
359
360         return (0);
361 }
362
363 /*
364  * Standard detach entry point
365  */
366 static int
367 netvsc_detach(device_t dev)
368 {
369         struct hv_device *hv_device = vmbus_get_devctx(dev); 
370
371         if (bootverbose)
372                 printf("netvsc_detach\n");
373
374         /*
375          * XXXKYS:  Need to clean up all our
376          * driver state; this is the driver
377          * unloading.
378          */
379
380         /*
381          * XXXKYS:  Need to stop outgoing traffic and unregister
382          * the netdevice.
383          */
384
385         hv_rf_on_device_remove(hv_device, HV_RF_NV_DESTROY_CHANNEL);
386
387         return (0);
388 }
389
390 /*
391  * Standard shutdown entry point
392  */
393 static int
394 netvsc_shutdown(device_t dev)
395 {
396         return (0);
397 }
398
399 /*
400  * Send completion processing
401  *
402  * Note:  It looks like offset 0 of buf is reserved to hold the softc
403  * pointer.  The sc pointer is not currently needed in this function, and
404  * it is not presently populated by the TX function.
405  */
406 void
407 netvsc_xmit_completion(void *context)
408 {
409         netvsc_packet *packet = (netvsc_packet *)context;
410         struct mbuf *mb;
411         uint8_t *buf;
412
413         mb = (struct mbuf *)(uintptr_t)packet->compl.send.send_completion_tid;
414         buf = ((uint8_t *)packet) - HV_NV_PACKET_OFFSET_IN_BUF;
415
416         free(buf, M_NETVSC);
417
418         if (mb != NULL) {
419                 m_freem(mb);
420         }
421 }
422
423 /*
424  * Start a transmit of one or more packets
425  */
426 static int
427 hn_start_locked(struct ifnet *ifp)
428 {
429         hn_softc_t *sc = ifp->if_softc;
430         struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
431         netvsc_dev *net_dev = sc->net_dev;
432         device_t dev = device_ctx->device;
433         uint8_t *buf;
434         netvsc_packet *packet;
435         struct mbuf *m_head, *m;
436         struct mbuf *mc_head = NULL;
437         struct ether_vlan_header *eh;
438         rndis_msg *rndis_mesg;
439         rndis_packet *rndis_pkt;
440         rndis_per_packet_info *rppi;
441         ndis_8021q_info *rppi_vlan_info;
442         rndis_tcp_ip_csum_info *csum_info;
443         rndis_tcp_tso_info *tso_info;   
444         int ether_len;
445         int i;
446         int num_frags;
447         int len;
448         int retries = 0;
449         int ret = 0;    
450         uint32_t rndis_msg_size = 0;
451         uint32_t trans_proto_type;
452         uint32_t send_buf_section_idx =
453             NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX;
454
455         while (!IFQ_DRV_IS_EMPTY(&sc->hn_ifp->if_snd)) {
456                 IFQ_DRV_DEQUEUE(&sc->hn_ifp->if_snd, m_head);
457                 if (m_head == NULL) {
458                         break;
459                 }
460
461                 len = 0;
462                 num_frags = 0;
463
464                 /* Walk the mbuf list computing total length and num frags */
465                 for (m = m_head; m != NULL; m = m->m_next) {
466                         if (m->m_len != 0) {
467                                 num_frags++;
468                                 len += m->m_len;
469                         }
470                 }
471
472                 /*
473                  * Reserve the number of pages requested.  Currently,
474                  * one page is reserved for the message in the RNDIS
475                  * filter packet
476                  */
477                 num_frags += HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
478
479                 /* If exceeds # page_buffers in netvsc_packet */
480                 if (num_frags > NETVSC_PACKET_MAXPAGE) {
481                         device_printf(dev, "exceed max page buffers,%d,%d\n",
482                             num_frags, NETVSC_PACKET_MAXPAGE);
483                         m_freem(m_head);
484                         if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
485                         return (EINVAL);
486                 }
487
488                 /*
489                  * Allocate a buffer with space for a netvsc packet plus a
490                  * number of reserved areas.  First comes a (currently 16
491                  * bytes, currently unused) reserved data area.  Second is
492                  * the netvsc_packet. Third is an area reserved for an 
493                  * rndis_filter_packet struct. Fourth (optional) is a 
494                  * rndis_per_packet_info struct.
495                  * Changed malloc to M_NOWAIT to avoid sleep under spin lock.
496                  * No longer reserving extra space for page buffers, as they
497                  * are already part of the netvsc_packet.
498                  */
499                 buf = malloc(HV_NV_PACKET_OFFSET_IN_BUF +
500                         sizeof(netvsc_packet) + 
501                         sizeof(rndis_msg) +
502                         RNDIS_VLAN_PPI_SIZE +
503                         RNDIS_TSO_PPI_SIZE +
504                         RNDIS_CSUM_PPI_SIZE,
505                         M_NETVSC, M_ZERO | M_NOWAIT);
506                 if (buf == NULL) {
507                         device_printf(dev, "hn:malloc packet failed\n");
508                         m_freem(m_head);
509                         if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
510                         return (ENOMEM);
511                 }
512
513                 packet = (netvsc_packet *)(buf + HV_NV_PACKET_OFFSET_IN_BUF);
514                 *(vm_offset_t *)buf = HV_NV_SC_PTR_OFFSET_IN_BUF;
515
516                 packet->is_data_pkt = TRUE;
517
518                 /* Set up the rndis header */
519                 packet->page_buf_count = num_frags;
520
521                 /* Initialize it from the mbuf */
522                 packet->tot_data_buf_len = len;
523
524                 /*
525                  * extension points to the area reserved for the
526                  * rndis_filter_packet, which is placed just after
527                  * the netvsc_packet (and rppi struct, if present;
528                  * length is updated later).
529                  */
530                 packet->rndis_mesg = packet + 1;
531                 rndis_mesg = (rndis_msg *)packet->rndis_mesg;
532                 rndis_mesg->ndis_msg_type = REMOTE_NDIS_PACKET_MSG;
533
534                 rndis_pkt = &rndis_mesg->msg.packet;
535                 rndis_pkt->data_offset = sizeof(rndis_packet);
536                 rndis_pkt->data_length = packet->tot_data_buf_len;
537                 rndis_pkt->per_pkt_info_offset = sizeof(rndis_packet);
538
539                 rndis_msg_size = RNDIS_MESSAGE_SIZE(rndis_packet);
540
541                 /*
542                  * If the Hyper-V infrastructure needs to embed a VLAN tag,
543                  * initialize netvsc_packet and rppi struct values as needed.
544                  */
545                 if (m_head->m_flags & M_VLANTAG) {
546                         /*
547                          * set up some additional fields so the Hyper-V infrastructure will stuff the VLAN tag
548                          * into the frame.
549                          */
550                         packet->vlan_tci = m_head->m_pkthdr.ether_vtag;
551
552                         rndis_msg_size += RNDIS_VLAN_PPI_SIZE;
553
554                         rppi = hv_set_rppi_data(rndis_mesg, RNDIS_VLAN_PPI_SIZE,
555                             ieee_8021q_info);
556                 
557                         /* VLAN info immediately follows rppi struct */
558                         rppi_vlan_info = (ndis_8021q_info *)((char*)rppi + 
559                             rppi->per_packet_info_offset);
560                         /* FreeBSD does not support CFI or priority */
561                         rppi_vlan_info->u1.s1.vlan_id =
562                             packet->vlan_tci & 0xfff;
563                 }
564
565                 if (0 == m_head->m_pkthdr.csum_flags) {
566                         goto pre_send;
567                 }
568
569                 eh = mtod(m_head, struct ether_vlan_header*);
570                 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
571                         ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
572                 } else {
573                         ether_len = ETHER_HDR_LEN;
574                 }
575
576                 trans_proto_type = get_transport_proto_type(m_head);
577                 if (TRANSPORT_TYPE_NOT_IP == trans_proto_type) {
578                         goto pre_send;
579                 }
580
581                 /*
582                  * TSO packet needless to setup the send side checksum
583                  * offload.
584                  */
585                 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
586                         goto do_tso;
587                 }
588
589                 /* setup checksum offload */
590                 rndis_msg_size += RNDIS_CSUM_PPI_SIZE;
591                 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_CSUM_PPI_SIZE,
592                     tcpip_chksum_info);
593                 csum_info = (rndis_tcp_ip_csum_info *)((char*)rppi +
594                     rppi->per_packet_info_offset);
595
596                 if (trans_proto_type & (TYPE_IPV4 << 16)) {
597                         csum_info->xmit.is_ipv4 = 1;
598                 } else {
599                         csum_info->xmit.is_ipv6 = 1;
600                 }
601
602                 if (trans_proto_type & TYPE_TCP) {
603                         csum_info->xmit.tcp_csum = 1;
604                         csum_info->xmit.tcp_header_offset = 0;
605                 } else if (trans_proto_type & TYPE_UDP) {
606                         csum_info->xmit.udp_csum = 1;
607                 }
608
609                 goto pre_send;
610
611 do_tso:
612                 /* setup TCP segmentation offload */
613                 rndis_msg_size += RNDIS_TSO_PPI_SIZE;
614                 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_TSO_PPI_SIZE,
615                     tcp_large_send_info);
616                 
617                 tso_info = (rndis_tcp_tso_info *)((char *)rppi +
618                     rppi->per_packet_info_offset);
619                 tso_info->lso_v2_xmit.type =
620                     RNDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
621                 
622 #ifdef INET
623                 if (trans_proto_type & (TYPE_IPV4 << 16)) {
624                         struct ip *ip =
625                             (struct ip *)(m_head->m_data + ether_len);
626                         unsigned long iph_len = ip->ip_hl << 2;
627                         struct tcphdr *th =
628                             (struct tcphdr *)((caddr_t)ip + iph_len);
629                 
630                         tso_info->lso_v2_xmit.ip_version =
631                             RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
632                         ip->ip_len = 0;
633                         ip->ip_sum = 0;
634                 
635                         th->th_sum = in_pseudo(ip->ip_src.s_addr,
636                             ip->ip_dst.s_addr,
637                             htons(IPPROTO_TCP));
638                 }
639 #endif
640 #if defined(INET6) && defined(INET)
641                 else
642 #endif
643 #ifdef INET6
644                 {
645                         struct ip6_hdr *ip6 =
646                             (struct ip6_hdr *)(m_head->m_data + ether_len);
647                         struct tcphdr *th = (struct tcphdr *)(ip6 + 1);
648
649                         tso_info->lso_v2_xmit.ip_version =
650                             RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
651                         ip6->ip6_plen = 0;
652                         th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
653                 }
654 #endif
655                 tso_info->lso_v2_xmit.tcp_header_offset = 0;
656                 tso_info->lso_v2_xmit.mss = m_head->m_pkthdr.tso_segsz;
657
658 pre_send:
659                 rndis_mesg->msg_len = packet->tot_data_buf_len + rndis_msg_size;
660                 packet->tot_data_buf_len = rndis_mesg->msg_len;
661
662                 /* send packet with send buffer */
663                 if (packet->tot_data_buf_len < net_dev->send_section_size) {
664                         send_buf_section_idx =
665                             hv_nv_get_next_send_section(net_dev);
666                         if (send_buf_section_idx !=
667                             NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX) {
668                                 char *dest = ((char *)net_dev->send_buf +
669                                     send_buf_section_idx *
670                                     net_dev->send_section_size);
671
672                                 memcpy(dest, rndis_mesg, rndis_msg_size);
673                                 dest += rndis_msg_size;
674                                 for (m = m_head; m != NULL; m = m->m_next) {
675                                         if (m->m_len) {
676                                                 memcpy(dest,
677                                                     (void *)mtod(m, vm_offset_t),
678                                                     m->m_len);
679                                                 dest += m->m_len;
680                                         }
681                                 }
682
683                                 packet->send_buf_section_idx =
684                                     send_buf_section_idx;
685                                 packet->send_buf_section_size =
686                                     packet->tot_data_buf_len;
687                                 packet->page_buf_count = 0;
688                                 goto do_send;
689                         }
690                 }
691
692                 /* send packet with page buffer */
693                 packet->page_buffers[0].pfn =
694                     atop(hv_get_phys_addr(rndis_mesg));
695                 packet->page_buffers[0].offset =
696                     (unsigned long)rndis_mesg & PAGE_MASK;
697                 packet->page_buffers[0].length = rndis_msg_size;
698
699                 /*
700                  * Fill the page buffers with mbuf info starting at index
701                  * HV_RF_NUM_TX_RESERVED_PAGE_BUFS.
702                  */
703                 i = HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
704                 for (m = m_head; m != NULL; m = m->m_next) {
705                         if (m->m_len) {
706                                 vm_offset_t paddr =
707                                     vtophys(mtod(m, vm_offset_t));
708                                 packet->page_buffers[i].pfn =
709                                     paddr >> PAGE_SHIFT;
710                                 packet->page_buffers[i].offset =
711                                     paddr & (PAGE_SIZE - 1);
712                                 packet->page_buffers[i].length = m->m_len;
713                                 i++;
714                         }
715                 }
716
717                 packet->send_buf_section_idx = 
718                     NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX;
719                 packet->send_buf_section_size = 0;
720
721 do_send:
722
723                 /*
724                  * If bpf, copy the mbuf chain.  This is less expensive than
725                  * it appears; the mbuf clusters are not copied, only their
726                  * reference counts are incremented.
727                  * Needed to avoid a race condition where the completion
728                  * callback is invoked, freeing the mbuf chain, before the
729                  * bpf_mtap code has a chance to run.
730                  */
731                 if (ifp->if_bpf) {
732                         mc_head = m_copypacket(m_head, M_DONTWAIT);
733                 }
734 retry_send:
735                 /* Set the completion routine */
736                 packet->compl.send.on_send_completion = netvsc_xmit_completion;
737                 packet->compl.send.send_completion_context = packet;
738                 packet->compl.send.send_completion_tid = (uint64_t)(uintptr_t)m_head;
739
740                 /* Removed critical_enter(), does not appear necessary */
741                 ret = hv_nv_on_send(device_ctx, packet);
742                 if (ret == 0) {
743                         ifp->if_opackets++;
744                         /* if bpf && mc_head, call bpf_mtap code */
745                         if (mc_head) {
746                                 ETHER_BPF_MTAP(ifp, mc_head);
747                         }
748                 } else {
749                         retries++;
750                         if (retries < 4) {
751                                 goto retry_send;
752                         }
753
754                         IF_PREPEND(&ifp->if_snd, m_head);
755                         ifp->if_drv_flags |= IFF_DRV_OACTIVE;
756
757                         /*
758                          * Null the mbuf pointer so the completion function
759                          * does not free the mbuf chain.  We just pushed the
760                          * mbuf chain back on the if_snd queue.
761                          */
762                         packet->compl.send.send_completion_tid = 0;
763
764                         /*
765                          * Release the resources since we will not get any
766                          * send completion
767                          */
768                         netvsc_xmit_completion(packet);
769                         if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
770                 }
771
772                 /* if bpf && mc_head, free the mbuf chain copy */
773                 if (mc_head) {
774                         m_freem(mc_head);
775                 }
776         }
777
778         return (ret);
779 }
780
781 /*
782  * Link up/down notification
783  */
784 void
785 netvsc_linkstatus_callback(struct hv_device *device_obj, uint32_t status)
786 {
787         hn_softc_t *sc = device_get_softc(device_obj->device);
788
789         if (sc == NULL) {
790                 return;
791         }
792
793         if (status == 1) {
794                 sc->hn_carrier = 1;
795         } else {
796                 sc->hn_carrier = 0;
797         }
798 }
799
800 /*
801  * Append the specified data to the indicated mbuf chain,
802  * Extend the mbuf chain if the new data does not fit in
803  * existing space.
804  *
805  * This is a minor rewrite of m_append() from sys/kern/uipc_mbuf.c.
806  * There should be an equivalent in the kernel mbuf code,
807  * but there does not appear to be one yet.
808  *
809  * Differs from m_append() in that additional mbufs are
810  * allocated with cluster size MJUMPAGESIZE, and filled
811  * accordingly.
812  *
813  * Return 1 if able to complete the job; otherwise 0.
814  */
815 static int
816 hv_m_append(struct mbuf *m0, int len, c_caddr_t cp)
817 {
818         struct mbuf *m, *n;
819         int remainder, space;
820
821         for (m = m0; m->m_next != NULL; m = m->m_next)
822                 ;
823         remainder = len;
824         space = M_TRAILINGSPACE(m);
825         if (space > 0) {
826                 /*
827                  * Copy into available space.
828                  */
829                 if (space > remainder)
830                         space = remainder;
831                 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
832                 m->m_len += space;
833                 cp += space;
834                 remainder -= space;
835         }
836         while (remainder > 0) {
837                 /*
838                  * Allocate a new mbuf; could check space
839                  * and allocate a cluster instead.
840                  */
841                 n = m_getjcl(M_DONTWAIT, m->m_type, 0, MJUMPAGESIZE);
842                 if (n == NULL)
843                         break;
844                 n->m_len = min(MJUMPAGESIZE, remainder);
845                 bcopy(cp, mtod(n, caddr_t), n->m_len);
846                 cp += n->m_len;
847                 remainder -= n->m_len;
848                 m->m_next = n;
849                 m = n;
850         }
851         if (m0->m_flags & M_PKTHDR)
852                 m0->m_pkthdr.len += len - remainder;
853
854         return (remainder == 0);
855 }
856
857
858 /*
859  * Called when we receive a data packet from the "wire" on the
860  * specified device
861  *
862  * Note:  This is no longer used as a callback
863  */
864 int
865 netvsc_recv(struct hv_device *device_ctx, netvsc_packet *packet,
866     rndis_tcp_ip_csum_info *csum_info)
867 {
868         hn_softc_t *sc = (hn_softc_t *)device_get_softc(device_ctx->device);
869         struct mbuf *m_new;
870         struct ifnet *ifp;
871         device_t dev = device_ctx->device;
872         int size;
873
874         if (sc == NULL) {
875                 return (0); /* TODO: KYS how can this be! */
876         }
877
878         ifp = sc->hn_ifp;
879         
880         ifp = sc->arpcom.ac_ifp;
881
882         if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
883                 return (0);
884         }
885
886         /*
887          * Bail out if packet contains more data than configured MTU.
888          */
889         if (packet->tot_data_buf_len > (ifp->if_mtu + ETHER_HDR_LEN)) {
890                 return (0);
891         }
892
893         /*
894          * Get an mbuf with a cluster.  For packets 2K or less,
895          * get a standard 2K cluster.  For anything larger, get a
896          * 4K cluster.  Any buffers larger than 4K can cause problems
897          * if looped around to the Hyper-V TX channel, so avoid them.
898          */
899         size = MCLBYTES;
900
901         if (packet->tot_data_buf_len > MCLBYTES) {
902                 /* 4096 */
903                 size = MJUMPAGESIZE;
904         }
905
906         m_new = m_getjcl(M_DONTWAIT, MT_DATA, M_PKTHDR, size);
907
908         if (m_new == NULL) {
909                 device_printf(dev, "alloc mbuf failed.\n");
910                 return (0);
911         }
912
913         hv_m_append(m_new, packet->tot_data_buf_len,
914                         packet->data);
915
916         m_new->m_pkthdr.rcvif = ifp;
917
918         /* receive side checksum offload */
919         m_new->m_pkthdr.csum_flags = 0;
920         if (NULL != csum_info) {
921                 /* IP csum offload */
922                 if (csum_info->receive.ip_csum_succeeded) {
923                         m_new->m_pkthdr.csum_flags |=
924                             (CSUM_IP_CHECKED | CSUM_IP_VALID);
925                 }
926
927                 /* TCP csum offload */
928                 if (csum_info->receive.tcp_csum_succeeded) {
929                         m_new->m_pkthdr.csum_flags |=
930                             (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
931                         m_new->m_pkthdr.csum_data = 0xffff;
932                 }
933         }
934
935         if ((packet->vlan_tci != 0) &&
936             (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0) {
937                 m_new->m_pkthdr.ether_vtag = packet->vlan_tci;
938                 m_new->m_flags |= M_VLANTAG;
939         }
940
941         /*
942          * Note:  Moved RX completion back to hv_nv_on_receive() so all
943          * messages (not just data messages) will trigger a response.
944          */
945
946         ifp->if_ipackets++;
947
948         /* We're not holding the lock here, so don't release it */
949         (*ifp->if_input)(ifp, m_new);
950
951         return (0);
952 }
953
954 /*
955  * Rules for using sc->temp_unusable:
956  * 1.  sc->temp_unusable can only be read or written while holding NV_LOCK()
957  * 2.  code reading sc->temp_unusable under NV_LOCK(), and finding 
958  *     sc->temp_unusable set, must release NV_LOCK() and exit
959  * 3.  to retain exclusive control of the interface,
960  *     sc->temp_unusable must be set by code before releasing NV_LOCK()
961  * 4.  only code setting sc->temp_unusable can clear sc->temp_unusable
962  * 5.  code setting sc->temp_unusable must eventually clear sc->temp_unusable
963  */
964
965 /*
966  * Standard ioctl entry point.  Called when the user wants to configure
967  * the interface.
968  */
969 static int
970 hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
971 {
972         hn_softc_t *sc = ifp->if_softc;
973         struct ifreq *ifr = (struct ifreq *)data;
974 #ifdef INET
975         struct ifaddr *ifa = (struct ifaddr *)data;
976 #endif
977         netvsc_device_info device_info;
978         struct hv_device *hn_dev;
979         int mask, error = 0;
980         int retry_cnt = 500;
981         
982         switch(cmd) {
983
984         case SIOCSIFADDR:
985 #ifdef INET
986                 if (ifa->ifa_addr->sa_family == AF_INET) {
987                         ifp->if_flags |= IFF_UP;
988                         if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
989                                 hn_ifinit(sc);
990                         arp_ifinit(ifp, ifa);
991                 } else
992 #endif
993                 error = ether_ioctl(ifp, cmd, data);
994                 break;
995         case SIOCSIFMTU:
996                 hn_dev = vmbus_get_devctx(sc->hn_dev);
997
998                 /* Check MTU value change */
999                 if (ifp->if_mtu == ifr->ifr_mtu)
1000                         break;
1001
1002                 if (ifr->ifr_mtu > NETVSC_MAX_CONFIGURABLE_MTU) {
1003                         error = EINVAL;
1004                         break;
1005                 }
1006
1007                 /* Obtain and record requested MTU */
1008                 ifp->if_mtu = ifr->ifr_mtu;
1009                 
1010                 do {
1011                         NV_LOCK(sc);
1012                         if (!sc->temp_unusable) {
1013                                 sc->temp_unusable = TRUE;
1014                                 retry_cnt = -1;
1015                         }
1016                         NV_UNLOCK(sc);
1017                         if (retry_cnt > 0) {
1018                                 retry_cnt--;
1019                                 DELAY(5 * 1000);
1020                         }
1021                 } while (retry_cnt > 0);
1022
1023                 if (retry_cnt == 0) {
1024                         error = EINVAL;
1025                         break;
1026                 }
1027
1028                 /* We must remove and add back the device to cause the new
1029                  * MTU to take effect.  This includes tearing down, but not
1030                  * deleting the channel, then bringing it back up.
1031                  */
1032                 error = hv_rf_on_device_remove(hn_dev, HV_RF_NV_RETAIN_CHANNEL);
1033                 if (error) {
1034                         NV_LOCK(sc);
1035                         sc->temp_unusable = FALSE;
1036                         NV_UNLOCK(sc);
1037                         break;
1038                 }
1039                 error = hv_rf_on_device_add(hn_dev, &device_info);
1040                 if (error) {
1041                         NV_LOCK(sc);
1042                         sc->temp_unusable = FALSE;
1043                         NV_UNLOCK(sc);
1044                         break;
1045                 }
1046
1047                 hn_ifinit_locked(sc);
1048
1049                 NV_LOCK(sc);
1050                 sc->temp_unusable = FALSE;
1051                 NV_UNLOCK(sc);
1052                 break;
1053         case SIOCSIFFLAGS:
1054                 do {
1055                        NV_LOCK(sc);
1056                        if (!sc->temp_unusable) {
1057                                sc->temp_unusable = TRUE;
1058                                retry_cnt = -1;
1059                        }
1060                        NV_UNLOCK(sc);
1061                        if (retry_cnt > 0) {
1062                                 retry_cnt--;
1063                                 DELAY(5 * 1000);
1064                        }
1065                 } while (retry_cnt > 0);
1066
1067                 if (retry_cnt == 0) {
1068                        error = EINVAL;
1069                        break;
1070                 }
1071
1072                 if (ifp->if_flags & IFF_UP) {
1073                         /*
1074                          * If only the state of the PROMISC flag changed,
1075                          * then just use the 'set promisc mode' command
1076                          * instead of reinitializing the entire NIC. Doing
1077                          * a full re-init means reloading the firmware and
1078                          * waiting for it to start up, which may take a
1079                          * second or two.
1080                          */
1081 #ifdef notyet
1082                         /* Fixme:  Promiscuous mode? */
1083                         if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1084                             ifp->if_flags & IFF_PROMISC &&
1085                             !(sc->hn_if_flags & IFF_PROMISC)) {
1086                                 /* do something here for Hyper-V */
1087                         } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1088                             !(ifp->if_flags & IFF_PROMISC) &&
1089                             sc->hn_if_flags & IFF_PROMISC) {
1090                                 /* do something here for Hyper-V */
1091                         } else
1092 #endif
1093                                 hn_ifinit_locked(sc);
1094                 } else {
1095                         if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1096                                 hn_stop(sc);
1097                         }
1098                 }
1099                 NV_LOCK(sc);
1100                 sc->temp_unusable = FALSE;
1101                 NV_UNLOCK(sc);
1102                 sc->hn_if_flags = ifp->if_flags;
1103                 error = 0;
1104                 break;
1105         case SIOCSIFCAP:
1106                 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1107                 if (mask & IFCAP_TXCSUM) {
1108                         if (IFCAP_TXCSUM & ifp->if_capenable) {
1109                                 ifp->if_capenable &= ~IFCAP_TXCSUM;
1110                                 ifp->if_hwassist &= ~(CSUM_TCP | CSUM_UDP);
1111                         } else {
1112                                 ifp->if_capenable |= IFCAP_TXCSUM;
1113                                 ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP);
1114                         }
1115                 }
1116
1117                 if (mask & IFCAP_RXCSUM) {
1118                         if (IFCAP_RXCSUM & ifp->if_capenable) {
1119                                 ifp->if_capenable &= ~IFCAP_RXCSUM;
1120                         } else {
1121                                 ifp->if_capenable |= IFCAP_RXCSUM;
1122                         }
1123                 }
1124
1125                 if (mask & IFCAP_TSO4) {
1126                         ifp->if_capenable ^= IFCAP_TSO4;
1127                         ifp->if_hwassist ^= CSUM_IP_TSO;
1128                 }
1129
1130                 if (mask & IFCAP_TSO6) {
1131                         ifp->if_capenable ^= IFCAP_TSO6;
1132                         ifp->if_hwassist ^= CSUM_IP6_TSO;
1133                 }
1134
1135                 error = 0;
1136                 break;
1137         case SIOCADDMULTI:
1138         case SIOCDELMULTI:
1139 #ifdef notyet
1140                 /* Fixme:  Multicast mode? */
1141                 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1142                         NV_LOCK(sc);
1143                         netvsc_setmulti(sc);
1144                         NV_UNLOCK(sc);
1145                         error = 0;
1146                 }
1147 #endif
1148                 /* FALLTHROUGH */
1149         case SIOCSIFMEDIA:
1150         case SIOCGIFMEDIA:
1151                 error = EINVAL;
1152                 break;
1153         default:
1154                 error = ether_ioctl(ifp, cmd, data);
1155                 break;
1156         }
1157
1158         return (error);
1159 }
1160
1161 /*
1162  *
1163  */
1164 static void
1165 hn_stop(hn_softc_t *sc)
1166 {
1167         struct ifnet *ifp;
1168         int ret;
1169         struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1170
1171         ifp = sc->hn_ifp;
1172
1173         if (bootverbose)
1174                 printf(" Closing Device ...\n");
1175
1176         ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
1177         if_link_state_change(ifp, LINK_STATE_DOWN);
1178         sc->hn_initdone = 0;
1179
1180         ret = hv_rf_on_close(device_ctx);
1181 }
1182
1183 /*
1184  * FreeBSD transmit entry point
1185  */
1186 static void
1187 hn_start(struct ifnet *ifp)
1188 {
1189         hn_softc_t *sc;
1190
1191         sc = ifp->if_softc;
1192         NV_LOCK(sc);
1193         if (sc->temp_unusable) {
1194                 NV_UNLOCK(sc);
1195                 return;
1196         }
1197         hn_start_locked(ifp);
1198         NV_UNLOCK(sc);
1199 }
1200
1201 /*
1202  *
1203  */
1204 static void
1205 hn_ifinit_locked(hn_softc_t *sc)
1206 {
1207         struct ifnet *ifp;
1208         struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1209         int ret;
1210
1211         ifp = sc->hn_ifp;
1212
1213         if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1214                 return;
1215         }
1216
1217         hv_promisc_mode = 1;
1218
1219         ret = hv_rf_on_open(device_ctx);
1220         if (ret != 0) {
1221                 return;
1222         } else {
1223                 sc->hn_initdone = 1;
1224         }
1225         ifp->if_drv_flags |= IFF_DRV_RUNNING;
1226         ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1227         if_link_state_change(ifp, LINK_STATE_UP);
1228 }
1229
1230 /*
1231  *
1232  */
1233 static void
1234 hn_ifinit(void *xsc)
1235 {
1236         hn_softc_t *sc = xsc;
1237
1238         NV_LOCK(sc);
1239         if (sc->temp_unusable) {
1240                 NV_UNLOCK(sc);
1241                 return;
1242         }
1243         sc->temp_unusable = TRUE;
1244         NV_UNLOCK(sc);
1245
1246         hn_ifinit_locked(sc);
1247
1248         NV_LOCK(sc);
1249         sc->temp_unusable = FALSE;
1250         NV_UNLOCK(sc);
1251 }
1252
1253 #ifdef LATER
1254 /*
1255  *
1256  */
1257 static void
1258 hn_watchdog(struct ifnet *ifp)
1259 {
1260         hn_softc_t *sc;
1261         sc = ifp->if_softc;
1262
1263         printf("hn%d: watchdog timeout -- resetting\n", sc->hn_unit);
1264         hn_ifinit(sc);    /*???*/
1265         ifp->if_oerrors++;
1266 }
1267 #endif
1268
1269 static device_method_t netvsc_methods[] = {
1270         /* Device interface */
1271         DEVMETHOD(device_probe,         netvsc_probe),
1272         DEVMETHOD(device_attach,        netvsc_attach),
1273         DEVMETHOD(device_detach,        netvsc_detach),
1274         DEVMETHOD(device_shutdown,      netvsc_shutdown),
1275
1276         { 0, 0 }
1277 };
1278
1279 static driver_t netvsc_driver = {
1280         NETVSC_DEVNAME,
1281         netvsc_methods,
1282         sizeof(hn_softc_t)
1283 };
1284
1285 static devclass_t netvsc_devclass;
1286
1287 DRIVER_MODULE(hn, vmbus, netvsc_driver, netvsc_devclass, 0, 0);
1288 MODULE_VERSION(hn, 1);
1289 MODULE_DEPEND(hn, vmbus, 1, 1, 1);
1290 SYSINIT(netvsc_initx, SI_SUB_KTHREAD_IDLE, SI_ORDER_MIDDLE + 1, netvsc_init,
1291      NULL);
1292