2 * Copyright (c) 2010-2012 Citrix Inc.
3 * Copyright (c) 2009-2012 Microsoft Corp.
4 * Copyright (c) 2012 NetApp Inc.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice unmodified, this list of conditions, and the following
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.
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.
30 * Copyright (c) 2004-2006 Kip Macy
31 * All rights reserved.
33 * Redistribution and use in source and binary forms, with or without
34 * modification, are permitted provided that the following conditions
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.
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
55 #include <sys/cdefs.h>
56 __FBSDID("$FreeBSD$");
58 #include "opt_inet6.h"
61 #include <sys/param.h>
62 #include <sys/systm.h>
63 #include <sys/sockio.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>
72 #include <sys/sysctl.h>
75 #include <net/if_arp.h>
76 #include <net/ethernet.h>
77 #include <net/if_dl.h>
78 #include <net/if_media.h>
82 #include <net/if_types.h>
83 #include <net/if_vlan_var.h>
86 #include <netinet/in_systm.h>
87 #include <netinet/in.h>
88 #include <netinet/ip.h>
89 #include <netinet/if_ether.h>
90 #include <netinet/tcp.h>
91 #include <netinet/udp.h>
92 #include <netinet/ip6.h>
95 #include <vm/vm_param.h>
96 #include <vm/vm_kern.h>
99 #include <machine/bus.h>
100 #include <machine/resource.h>
101 #include <machine/frame.h>
102 #include <machine/vmparam.h>
105 #include <sys/rman.h>
106 #include <sys/mutex.h>
107 #include <sys/errno.h>
108 #include <sys/types.h>
109 #include <machine/atomic.h>
111 #include <machine/intr_machdep.h>
113 #include <machine/in_cksum.h>
115 #include <dev/hyperv/include/hyperv.h>
116 #include "hv_net_vsc.h"
117 #include "hv_rndis.h"
118 #include "hv_rndis_filter.h"
121 /* Short for Hyper-V network interface */
122 #define NETVSC_DEVNAME "hn"
125 * It looks like offset 0 of buf is reserved to hold the softc pointer.
126 * The sc pointer evidently not needed, and is not presently populated.
127 * The packet offset is where the netvsc_packet starts in the buffer.
129 #define HV_NV_SC_PTR_OFFSET_IN_BUF 0
130 #define HV_NV_PACKET_OFFSET_IN_BUF 16
132 /* YYY should get it from the underlying channel */
133 #define HN_TX_DESC_CNT 512
135 #define HN_LROENT_CNT_DEF 128
137 #define HN_RNDIS_MSG_LEN \
138 (sizeof(rndis_msg) + \
139 RNDIS_VLAN_PPI_SIZE + \
140 RNDIS_TSO_PPI_SIZE + \
142 #define HN_RNDIS_MSG_BOUNDARY PAGE_SIZE
143 #define HN_RNDIS_MSG_ALIGN CACHE_LINE_SIZE
145 #define HN_TX_DATA_BOUNDARY PAGE_SIZE
146 #define HN_TX_DATA_MAXSIZE IP_MAXPACKET
147 #define HN_TX_DATA_SEGSIZE PAGE_SIZE
148 #define HN_TX_DATA_SEGCNT_MAX \
149 (NETVSC_PACKET_MAXPAGE - HV_RF_NUM_TX_RESERVED_PAGE_BUFS)
151 #define HN_DIRECT_TX_SIZE_DEF 128
154 SLIST_ENTRY(hn_txdesc) link;
158 uint32_t flags; /* HN_TXD_FLAG_ */
159 netvsc_packet netvsc_pkt; /* XXX to be removed */
161 bus_dmamap_t data_dmap;
163 bus_addr_t rndis_msg_paddr;
164 rndis_msg *rndis_msg;
165 bus_dmamap_t rndis_msg_dmap;
168 #define HN_TXD_FLAG_ONLIST 0x1
169 #define HN_TXD_FLAG_DMAMAP 0x2
172 * Only enable UDP checksum offloading when it is on 2012R2 or
173 * later. UDP checksum offloading doesn't work on earlier
176 #define HN_CSUM_ASSIST_WIN8 (CSUM_TCP)
177 #define HN_CSUM_ASSIST (CSUM_IP | CSUM_UDP | CSUM_TCP)
179 /* XXX move to netinet/tcp_lro.h */
180 #define HN_LRO_HIWAT_MAX 65535
181 #define HN_LRO_HIWAT_DEF HN_LRO_HIWAT_MAX
182 /* YYY 2*MTU is a bit rough, but should be good enough. */
183 #define HN_LRO_HIWAT_MTULIM(ifp) (2 * (ifp)->if_mtu)
184 #define HN_LRO_HIWAT_ISVALID(sc, hiwat) \
185 ((hiwat) >= HN_LRO_HIWAT_MTULIM((sc)->hn_ifp) || \
186 (hiwat) <= HN_LRO_HIWAT_MAX)
189 * Be aware that this sleepable mutex will exhibit WITNESS errors when
190 * certain TCP and ARP code paths are taken. This appears to be a
191 * well-known condition, as all other drivers checked use a sleeping
192 * mutex to protect their transmit paths.
193 * Also Be aware that mutexes do not play well with semaphores, and there
194 * is a conflicting semaphore in a certain channel code path.
196 #define NV_LOCK_INIT(_sc, _name) \
197 mtx_init(&(_sc)->hn_lock, _name, MTX_NETWORK_LOCK, MTX_DEF)
198 #define NV_LOCK(_sc) mtx_lock(&(_sc)->hn_lock)
199 #define NV_TRYLOCK(_sc) mtx_trylock(&(_sc)->hn_lock)
200 #define NV_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->hn_lock, MA_OWNED)
201 #define NV_UNLOCK(_sc) mtx_unlock(&(_sc)->hn_lock)
202 #define NV_LOCK_DESTROY(_sc) mtx_destroy(&(_sc)->hn_lock)
209 int hv_promisc_mode = 0; /* normal mode by default */
211 /* Trust tcp segements verification on host side. */
212 static int hn_trust_hosttcp = 1;
213 TUNABLE_INT("dev.hn.trust_hosttcp", &hn_trust_hosttcp);
215 /* Trust udp datagrams verification on host side. */
216 static int hn_trust_hostudp = 1;
217 TUNABLE_INT("dev.hn.trust_hostudp", &hn_trust_hostudp);
219 /* Trust ip packets verification on host side. */
220 static int hn_trust_hostip = 1;
221 TUNABLE_INT("dev.hn.trust_hostip", &hn_trust_hostip);
223 #if __FreeBSD_version >= 1100045
224 /* Limit TSO burst size */
225 static int hn_tso_maxlen = 0;
226 TUNABLE_INT("dev.hn.tso_maxlen", &hn_tso_maxlen);
229 /* Limit chimney send size */
230 static int hn_tx_chimney_size = 0;
231 TUNABLE_INT("dev.hn.tx_chimney_size", &hn_tx_chimney_size);
233 /* Limit the size of packet for direct transmission */
234 static int hn_direct_tx_size = HN_DIRECT_TX_SIZE_DEF;
235 TUNABLE_INT("dev.hn.direct_tx_size", &hn_direct_tx_size);
237 #if defined(INET) || defined(INET6)
238 #if __FreeBSD_version >= 1100095
239 static int hn_lro_entry_count = HN_LROENT_CNT_DEF;
240 TUNABLE_INT("dev.hn.lro_entry_count", &hn_lro_entry_count);
245 * Forward declarations
247 static void hn_stop(hn_softc_t *sc);
248 static void hn_ifinit_locked(hn_softc_t *sc);
249 static void hn_ifinit(void *xsc);
250 static int hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data);
251 static int hn_start_locked(struct ifnet *ifp, int len);
252 static void hn_start(struct ifnet *ifp);
253 static void hn_start_txeof(struct ifnet *ifp);
254 static int hn_ifmedia_upd(struct ifnet *ifp);
255 static void hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr);
257 static int hn_lro_hiwat_sysctl(SYSCTL_HANDLER_ARGS);
259 static int hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS);
260 static int hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS);
261 static int hn_check_iplen(const struct mbuf *, int);
262 static int hn_create_tx_ring(struct hn_softc *sc);
263 static void hn_destroy_tx_ring(struct hn_softc *sc);
264 static void hn_start_taskfunc(void *xsc, int pending);
265 static void hn_txeof_taskfunc(void *xsc, int pending);
266 static int hn_encap(struct hn_softc *, struct hn_txdesc *, struct mbuf **);
269 hn_set_lro_hiwat(struct hn_softc *sc, int hiwat)
271 sc->hn_lro_hiwat = hiwat;
273 sc->hn_lro.lro_hiwat = sc->hn_lro_hiwat;
278 hn_ifmedia_upd(struct ifnet *ifp __unused)
285 hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
287 struct hn_softc *sc = ifp->if_softc;
289 ifmr->ifm_status = IFM_AVALID;
290 ifmr->ifm_active = IFM_ETHER;
292 if (!sc->hn_carrier) {
293 ifmr->ifm_active |= IFM_NONE;
296 ifmr->ifm_status |= IFM_ACTIVE;
297 ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
300 /* {F8615163-DF3E-46c5-913F-F2D2F965ED0E} */
301 static const hv_guid g_net_vsc_device_type = {
302 .data = {0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
303 0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E}
307 * Standard probe entry point.
311 netvsc_probe(device_t dev)
315 p = vmbus_get_type(dev);
316 if (!memcmp(p, &g_net_vsc_device_type.data, sizeof(hv_guid))) {
317 device_set_desc(dev, "Synthetic Network Interface");
319 printf("Netvsc probe... DONE \n");
321 return (BUS_PROBE_DEFAULT);
328 * Standard attach entry point.
330 * Called when the driver is loaded. It allocates needed resources,
331 * and initializes the "hardware" and software.
334 netvsc_attach(device_t dev)
336 struct hv_device *device_ctx = vmbus_get_devctx(dev);
337 netvsc_device_info device_info;
339 int unit = device_get_unit(dev);
340 struct ifnet *ifp = NULL;
341 struct sysctl_oid_list *child;
342 struct sysctl_ctx_list *ctx;
344 #if __FreeBSD_version >= 1100045
347 #if defined(INET) || defined(INET6)
348 #if __FreeBSD_version >= 1100095
353 sc = device_get_softc(dev);
358 bzero(sc, sizeof(hn_softc_t));
361 sc->hn_lro_hiwat = HN_LRO_HIWAT_DEF;
362 sc->hn_direct_tx_size = hn_direct_tx_size;
363 if (hn_trust_hosttcp)
364 sc->hn_trust_hcsum |= HN_TRUST_HCSUM_TCP;
365 if (hn_trust_hostudp)
366 sc->hn_trust_hcsum |= HN_TRUST_HCSUM_UDP;
368 sc->hn_trust_hcsum |= HN_TRUST_HCSUM_IP;
370 sc->hn_tx_taskq = taskqueue_create_fast("hn_tx", M_WAITOK,
371 taskqueue_thread_enqueue, &sc->hn_tx_taskq);
372 taskqueue_start_threads(&sc->hn_tx_taskq, 1, PI_NET, "%s tx",
373 device_get_nameunit(dev));
374 TASK_INIT(&sc->hn_start_task, 0, hn_start_taskfunc, sc);
375 TASK_INIT(&sc->hn_txeof_task, 0, hn_txeof_taskfunc, sc);
377 error = hn_create_tx_ring(sc);
381 NV_LOCK_INIT(sc, "NetVSCLock");
383 sc->hn_dev_obj = device_ctx;
385 ifp = sc->hn_ifp = sc->arpcom.ac_ifp = if_alloc(IFT_ETHER);
388 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
389 ifp->if_dunit = unit;
390 ifp->if_dname = NETVSC_DEVNAME;
392 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
393 ifp->if_ioctl = hn_ioctl;
394 ifp->if_start = hn_start;
395 ifp->if_init = hn_ifinit;
396 /* needed by hv_rf_on_device_add() code */
397 ifp->if_mtu = ETHERMTU;
398 IFQ_SET_MAXLEN(&ifp->if_snd, 512);
399 ifp->if_snd.ifq_drv_maxlen = 511;
400 IFQ_SET_READY(&ifp->if_snd);
402 ifmedia_init(&sc->hn_media, 0, hn_ifmedia_upd, hn_ifmedia_sts);
403 ifmedia_add(&sc->hn_media, IFM_ETHER | IFM_AUTO, 0, NULL);
404 ifmedia_set(&sc->hn_media, IFM_ETHER | IFM_AUTO);
405 /* XXX ifmedia_set really should do this for us */
406 sc->hn_media.ifm_media = sc->hn_media.ifm_cur->ifm_media;
409 * Tell upper layers that we support full VLAN capability.
411 ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
412 ifp->if_capabilities |=
413 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
416 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
419 if (hv_vmbus_protocal_version >= HV_VMBUS_VERSION_WIN8_1)
420 sc->hn_csum_assist = HN_CSUM_ASSIST;
422 sc->hn_csum_assist = HN_CSUM_ASSIST_WIN8;
423 ifp->if_hwassist = sc->hn_csum_assist | CSUM_TSO;
425 error = hv_rf_on_device_add(device_ctx, &device_info);
429 if (device_info.link_state == 0) {
433 #if defined(INET) || defined(INET6)
434 #if __FreeBSD_version >= 1100095
435 lroent_cnt = hn_lro_entry_count;
436 if (lroent_cnt < TCP_LRO_ENTRIES)
437 lroent_cnt = TCP_LRO_ENTRIES;
438 tcp_lro_init_args(&sc->hn_lro, ifp, lroent_cnt, 0);
439 device_printf(dev, "LRO: entry count %d\n", lroent_cnt);
441 tcp_lro_init(&sc->hn_lro);
442 /* Driver private LRO settings */
443 sc->hn_lro.ifp = ifp;
446 sc->hn_lro.lro_hiwat = sc->hn_lro_hiwat;
448 #endif /* INET || INET6 */
450 #if __FreeBSD_version >= 1100045
451 tso_maxlen = hn_tso_maxlen;
452 if (tso_maxlen <= 0 || tso_maxlen > IP_MAXPACKET)
453 tso_maxlen = IP_MAXPACKET;
455 ifp->if_hw_tsomaxsegcount = HN_TX_DATA_SEGCNT_MAX;
456 ifp->if_hw_tsomaxsegsize = PAGE_SIZE;
457 ifp->if_hw_tsomax = tso_maxlen -
458 (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
461 ether_ifattach(ifp, device_info.mac_addr);
463 #if __FreeBSD_version >= 1100045
464 if_printf(ifp, "TSO: %u/%u/%u\n", ifp->if_hw_tsomax,
465 ifp->if_hw_tsomaxsegcount, ifp->if_hw_tsomaxsegsize);
468 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
469 sc->hn_tx_chimney_size = sc->hn_tx_chimney_max;
470 if (hn_tx_chimney_size > 0 &&
471 hn_tx_chimney_size < sc->hn_tx_chimney_max)
472 sc->hn_tx_chimney_size = hn_tx_chimney_size;
474 ctx = device_get_sysctl_ctx(dev);
475 child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
477 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "lro_queued",
478 CTLFLAG_RW, &sc->hn_lro.lro_queued, 0, "LRO queued");
479 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "lro_flushed",
480 CTLFLAG_RW, &sc->hn_lro.lro_flushed, 0, "LRO flushed");
481 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "lro_tried",
482 CTLFLAG_RW, &sc->hn_lro_tried, "# of LRO tries");
484 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_hiwat",
485 CTLTYPE_INT | CTLFLAG_RW, sc, 0, hn_lro_hiwat_sysctl,
486 "I", "LRO high watermark");
488 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hosttcp",
489 CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_TCP,
490 hn_trust_hcsum_sysctl, "I",
491 "Trust tcp segement verification on host side, "
492 "when csum info is missing");
493 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostudp",
494 CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_UDP,
495 hn_trust_hcsum_sysctl, "I",
496 "Trust udp datagram verification on host side, "
497 "when csum info is missing");
498 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostip",
499 CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_IP,
500 hn_trust_hcsum_sysctl, "I",
501 "Trust ip packet verification on host side, "
502 "when csum info is missing");
503 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "csum_ip",
504 CTLFLAG_RW, &sc->hn_csum_ip, "RXCSUM IP");
505 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "csum_tcp",
506 CTLFLAG_RW, &sc->hn_csum_tcp, "RXCSUM TCP");
507 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "csum_udp",
508 CTLFLAG_RW, &sc->hn_csum_udp, "RXCSUM UDP");
509 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "csum_trusted",
510 CTLFLAG_RW, &sc->hn_csum_trusted,
511 "# of packets that we trust host's csum verification");
512 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "small_pkts",
513 CTLFLAG_RW, &sc->hn_small_pkts, "# of small packets received");
514 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "no_txdescs",
515 CTLFLAG_RW, &sc->hn_no_txdescs, "# of times short of TX descs");
516 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "send_failed",
517 CTLFLAG_RW, &sc->hn_send_failed, "# of hyper-v sending failure");
518 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "txdma_failed",
519 CTLFLAG_RW, &sc->hn_txdma_failed, "# of TX DMA failure");
520 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_collapsed",
521 CTLFLAG_RW, &sc->hn_tx_collapsed, "# of TX mbuf collapsed");
522 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_chimney",
523 CTLFLAG_RW, &sc->hn_tx_chimney, "# of chimney send");
524 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_cnt",
525 CTLFLAG_RD, &sc->hn_txdesc_cnt, 0, "# of total TX descs");
526 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_avail",
527 CTLFLAG_RD, &sc->hn_txdesc_avail, 0, "# of available TX descs");
528 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_chimney_max",
529 CTLFLAG_RD, &sc->hn_tx_chimney_max, 0,
530 "Chimney send packet size upper boundary");
531 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_size",
532 CTLTYPE_INT | CTLFLAG_RW, sc, 0, hn_tx_chimney_size_sysctl,
533 "I", "Chimney send packet size limit");
534 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "direct_tx_size",
535 CTLFLAG_RW, &sc->hn_direct_tx_size, 0,
536 "Size of the packet for direct transmission");
537 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "sched_tx",
538 CTLFLAG_RW, &sc->hn_sched_tx, 0,
539 "Always schedule transmission "
540 "instead of doing direct transmission");
543 struct sysctl_ctx_list *dc_ctx;
544 struct sysctl_oid_list *dc_child;
548 * Add sysctl nodes for devclass
550 dc = device_get_devclass(dev);
551 dc_ctx = devclass_get_sysctl_ctx(dc);
552 dc_child = SYSCTL_CHILDREN(devclass_get_sysctl_tree(dc));
554 SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "trust_hosttcp",
555 CTLFLAG_RD, &hn_trust_hosttcp, 0,
556 "Trust tcp segement verification on host side, "
557 "when csum info is missing (global setting)");
558 SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "trust_hostudp",
559 CTLFLAG_RD, &hn_trust_hostudp, 0,
560 "Trust udp datagram verification on host side, "
561 "when csum info is missing (global setting)");
562 SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "trust_hostip",
563 CTLFLAG_RD, &hn_trust_hostip, 0,
564 "Trust ip packet verification on host side, "
565 "when csum info is missing (global setting)");
566 SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "tx_chimney_size",
567 CTLFLAG_RD, &hn_tx_chimney_size, 0,
568 "Chimney send packet size limit");
569 #if __FreeBSD_version >= 1100045
570 SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "tso_maxlen",
571 CTLFLAG_RD, &hn_tso_maxlen, 0, "TSO burst limit");
573 SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "direct_tx_size",
574 CTLFLAG_RD, &hn_direct_tx_size, 0,
575 "Size of the packet for direct transmission");
576 #if defined(INET) || defined(INET6)
577 #if __FreeBSD_version >= 1100095
578 SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "lro_entry_count",
579 CTLFLAG_RD, &hn_lro_entry_count, 0, "LRO entry count");
586 hn_destroy_tx_ring(sc);
593 * Standard detach entry point
596 netvsc_detach(device_t dev)
598 struct hn_softc *sc = device_get_softc(dev);
599 struct hv_device *hv_device = vmbus_get_devctx(dev);
602 printf("netvsc_detach\n");
605 * XXXKYS: Need to clean up all our
606 * driver state; this is the driver
611 * XXXKYS: Need to stop outgoing traffic and unregister
615 hv_rf_on_device_remove(hv_device, HV_RF_NV_DESTROY_CHANNEL);
617 taskqueue_drain(sc->hn_tx_taskq, &sc->hn_start_task);
618 taskqueue_drain(sc->hn_tx_taskq, &sc->hn_txeof_task);
619 taskqueue_free(sc->hn_tx_taskq);
621 ifmedia_removeall(&sc->hn_media);
622 #if defined(INET) || defined(INET6)
623 tcp_lro_free(&sc->hn_lro);
625 hn_destroy_tx_ring(sc);
631 * Standard shutdown entry point
634 netvsc_shutdown(device_t dev)
640 hn_txdesc_dmamap_load(struct hn_softc *sc, struct hn_txdesc *txd,
641 struct mbuf **m_head, bus_dma_segment_t *segs, int *nsegs)
643 struct mbuf *m = *m_head;
646 error = bus_dmamap_load_mbuf_sg(sc->hn_tx_data_dtag, txd->data_dmap,
647 m, segs, nsegs, BUS_DMA_NOWAIT);
648 if (error == EFBIG) {
651 m_new = m_collapse(m, M_NOWAIT, HN_TX_DATA_SEGCNT_MAX);
656 sc->hn_tx_collapsed++;
658 error = bus_dmamap_load_mbuf_sg(sc->hn_tx_data_dtag,
659 txd->data_dmap, m, segs, nsegs, BUS_DMA_NOWAIT);
662 bus_dmamap_sync(sc->hn_tx_data_dtag, txd->data_dmap,
663 BUS_DMASYNC_PREWRITE);
664 txd->flags |= HN_TXD_FLAG_DMAMAP;
670 hn_txdesc_dmamap_unload(struct hn_softc *sc, struct hn_txdesc *txd)
673 if (txd->flags & HN_TXD_FLAG_DMAMAP) {
674 bus_dmamap_sync(sc->hn_tx_data_dtag,
675 txd->data_dmap, BUS_DMASYNC_POSTWRITE);
676 bus_dmamap_unload(sc->hn_tx_data_dtag,
678 txd->flags &= ~HN_TXD_FLAG_DMAMAP;
683 hn_txdesc_put(struct hn_softc *sc, struct hn_txdesc *txd)
686 KASSERT((txd->flags & HN_TXD_FLAG_ONLIST) == 0,
687 ("put an onlist txd %#x", txd->flags));
689 KASSERT(txd->refs > 0, ("invalid txd refs %d", txd->refs));
690 if (atomic_fetchadd_int(&txd->refs, -1) != 1)
693 hn_txdesc_dmamap_unload(sc, txd);
694 if (txd->m != NULL) {
699 txd->flags |= HN_TXD_FLAG_ONLIST;
701 mtx_lock_spin(&sc->hn_txlist_spin);
702 KASSERT(sc->hn_txdesc_avail >= 0 &&
703 sc->hn_txdesc_avail < sc->hn_txdesc_cnt,
704 ("txdesc_put: invalid txd avail %d", sc->hn_txdesc_avail));
705 sc->hn_txdesc_avail++;
706 SLIST_INSERT_HEAD(&sc->hn_txlist, txd, link);
707 mtx_unlock_spin(&sc->hn_txlist_spin);
712 static __inline struct hn_txdesc *
713 hn_txdesc_get(struct hn_softc *sc)
715 struct hn_txdesc *txd;
717 mtx_lock_spin(&sc->hn_txlist_spin);
718 txd = SLIST_FIRST(&sc->hn_txlist);
720 KASSERT(sc->hn_txdesc_avail > 0,
721 ("txdesc_get: invalid txd avail %d", sc->hn_txdesc_avail));
722 sc->hn_txdesc_avail--;
723 SLIST_REMOVE_HEAD(&sc->hn_txlist, link);
725 mtx_unlock_spin(&sc->hn_txlist_spin);
728 KASSERT(txd->m == NULL && txd->refs == 0 &&
729 (txd->flags & HN_TXD_FLAG_ONLIST), ("invalid txd"));
730 txd->flags &= ~HN_TXD_FLAG_ONLIST;
737 hn_txdesc_hold(struct hn_txdesc *txd)
740 /* 0->1 transition will never work */
741 KASSERT(txd->refs > 0, ("invalid refs %d", txd->refs));
742 atomic_add_int(&txd->refs, 1);
746 * Send completion processing
748 * Note: It looks like offset 0 of buf is reserved to hold the softc
749 * pointer. The sc pointer is not currently needed in this function, and
750 * it is not presently populated by the TX function.
753 netvsc_xmit_completion(void *context)
755 netvsc_packet *packet = context;
756 struct hn_txdesc *txd;
759 txd = (struct hn_txdesc *)(uintptr_t)
760 packet->compl.send.send_completion_tid;
764 hn_txdesc_put(sc, txd);
768 netvsc_channel_rollup(struct hv_device *device_ctx)
770 struct hn_softc *sc = device_get_softc(device_ctx->device);
771 #if defined(INET) || defined(INET6)
772 struct lro_ctrl *lro = &sc->hn_lro;
773 struct lro_entry *queued;
775 while ((queued = SLIST_FIRST(&lro->lro_active)) != NULL) {
776 SLIST_REMOVE_HEAD(&lro->lro_active, next);
777 tcp_lro_flush(lro, queued);
785 hn_start_txeof(sc->hn_ifp);
790 * If this function fails, then both txd and m_head0 will be freed.
793 hn_encap(struct hn_softc *sc, struct hn_txdesc *txd, struct mbuf **m_head0)
795 bus_dma_segment_t segs[HN_TX_DATA_SEGCNT_MAX];
797 struct mbuf *m_head = *m_head0;
798 netvsc_packet *packet;
799 rndis_msg *rndis_mesg;
800 rndis_packet *rndis_pkt;
801 rndis_per_packet_info *rppi;
802 uint32_t rndis_msg_size;
804 packet = &txd->netvsc_pkt;
805 packet->is_data_pkt = TRUE;
806 packet->tot_data_buf_len = m_head->m_pkthdr.len;
809 * extension points to the area reserved for the
810 * rndis_filter_packet, which is placed just after
811 * the netvsc_packet (and rppi struct, if present;
812 * length is updated later).
814 rndis_mesg = txd->rndis_msg;
815 /* XXX not necessary */
816 memset(rndis_mesg, 0, HN_RNDIS_MSG_LEN);
817 rndis_mesg->ndis_msg_type = REMOTE_NDIS_PACKET_MSG;
819 rndis_pkt = &rndis_mesg->msg.packet;
820 rndis_pkt->data_offset = sizeof(rndis_packet);
821 rndis_pkt->data_length = packet->tot_data_buf_len;
822 rndis_pkt->per_pkt_info_offset = sizeof(rndis_packet);
824 rndis_msg_size = RNDIS_MESSAGE_SIZE(rndis_packet);
826 if (m_head->m_flags & M_VLANTAG) {
827 ndis_8021q_info *rppi_vlan_info;
829 rndis_msg_size += RNDIS_VLAN_PPI_SIZE;
830 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_VLAN_PPI_SIZE,
833 rppi_vlan_info = (ndis_8021q_info *)((uint8_t *)rppi +
834 rppi->per_packet_info_offset);
835 rppi_vlan_info->u1.s1.vlan_id =
836 m_head->m_pkthdr.ether_vtag & 0xfff;
839 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
840 rndis_tcp_tso_info *tso_info;
841 struct ether_vlan_header *eh;
845 * XXX need m_pullup and use mtodo
847 eh = mtod(m_head, struct ether_vlan_header*);
848 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN))
849 ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
851 ether_len = ETHER_HDR_LEN;
853 rndis_msg_size += RNDIS_TSO_PPI_SIZE;
854 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_TSO_PPI_SIZE,
855 tcp_large_send_info);
857 tso_info = (rndis_tcp_tso_info *)((uint8_t *)rppi +
858 rppi->per_packet_info_offset);
859 tso_info->lso_v2_xmit.type =
860 RNDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
863 if (m_head->m_pkthdr.csum_flags & CSUM_IP_TSO) {
865 (struct ip *)(m_head->m_data + ether_len);
866 unsigned long iph_len = ip->ip_hl << 2;
868 (struct tcphdr *)((caddr_t)ip + iph_len);
870 tso_info->lso_v2_xmit.ip_version =
871 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
875 th->th_sum = in_pseudo(ip->ip_src.s_addr,
876 ip->ip_dst.s_addr, htons(IPPROTO_TCP));
879 #if defined(INET6) && defined(INET)
884 struct ip6_hdr *ip6 = (struct ip6_hdr *)
885 (m_head->m_data + ether_len);
886 struct tcphdr *th = (struct tcphdr *)(ip6 + 1);
888 tso_info->lso_v2_xmit.ip_version =
889 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
891 th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
894 tso_info->lso_v2_xmit.tcp_header_offset = 0;
895 tso_info->lso_v2_xmit.mss = m_head->m_pkthdr.tso_segsz;
896 } else if (m_head->m_pkthdr.csum_flags & sc->hn_csum_assist) {
897 rndis_tcp_ip_csum_info *csum_info;
899 rndis_msg_size += RNDIS_CSUM_PPI_SIZE;
900 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_CSUM_PPI_SIZE,
902 csum_info = (rndis_tcp_ip_csum_info *)((uint8_t *)rppi +
903 rppi->per_packet_info_offset);
905 csum_info->xmit.is_ipv4 = 1;
906 if (m_head->m_pkthdr.csum_flags & CSUM_IP)
907 csum_info->xmit.ip_header_csum = 1;
909 if (m_head->m_pkthdr.csum_flags & CSUM_TCP) {
910 csum_info->xmit.tcp_csum = 1;
911 csum_info->xmit.tcp_header_offset = 0;
912 } else if (m_head->m_pkthdr.csum_flags & CSUM_UDP) {
913 csum_info->xmit.udp_csum = 1;
917 rndis_mesg->msg_len = packet->tot_data_buf_len + rndis_msg_size;
918 packet->tot_data_buf_len = rndis_mesg->msg_len;
921 * Chimney send, if the packet could fit into one chimney buffer.
923 if (packet->tot_data_buf_len < sc->hn_tx_chimney_size) {
924 netvsc_dev *net_dev = sc->net_dev;
925 uint32_t send_buf_section_idx;
927 send_buf_section_idx =
928 hv_nv_get_next_send_section(net_dev);
929 if (send_buf_section_idx !=
930 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX) {
931 uint8_t *dest = ((uint8_t *)net_dev->send_buf +
932 (send_buf_section_idx *
933 net_dev->send_section_size));
935 memcpy(dest, rndis_mesg, rndis_msg_size);
936 dest += rndis_msg_size;
937 m_copydata(m_head, 0, m_head->m_pkthdr.len, dest);
939 packet->send_buf_section_idx = send_buf_section_idx;
940 packet->send_buf_section_size =
941 packet->tot_data_buf_len;
942 packet->page_buf_count = 0;
948 error = hn_txdesc_dmamap_load(sc, txd, &m_head, segs, &nsegs);
953 * This mbuf is not linked w/ the txd yet, so free it now.
958 freed = hn_txdesc_put(sc, txd);
960 ("fail to free txd upon txdma error"));
962 sc->hn_txdma_failed++;
963 if_inc_counter(sc->hn_ifp, IFCOUNTER_OERRORS, 1);
968 packet->page_buf_count = nsegs + HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
970 /* send packet with page buffer */
971 packet->page_buffers[0].pfn = atop(txd->rndis_msg_paddr);
972 packet->page_buffers[0].offset = txd->rndis_msg_paddr & PAGE_MASK;
973 packet->page_buffers[0].length = rndis_msg_size;
976 * Fill the page buffers with mbuf info starting at index
977 * HV_RF_NUM_TX_RESERVED_PAGE_BUFS.
979 for (i = 0; i < nsegs; ++i) {
980 hv_vmbus_page_buffer *pb = &packet->page_buffers[
981 i + HV_RF_NUM_TX_RESERVED_PAGE_BUFS];
983 pb->pfn = atop(segs[i].ds_addr);
984 pb->offset = segs[i].ds_addr & PAGE_MASK;
985 pb->length = segs[i].ds_len;
988 packet->send_buf_section_idx =
989 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX;
990 packet->send_buf_section_size = 0;
994 /* Set the completion routine */
995 packet->compl.send.on_send_completion = netvsc_xmit_completion;
996 packet->compl.send.send_completion_context = packet;
997 packet->compl.send.send_completion_tid = (uint64_t)(uintptr_t)txd;
1003 * Start a transmit of one or more packets
1006 hn_start_locked(struct ifnet *ifp, int len)
1008 struct hn_softc *sc = ifp->if_softc;
1009 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1011 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1015 while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
1016 int error, send_failed = 0;
1017 struct hn_txdesc *txd;
1018 struct mbuf *m_head;
1020 IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
1024 if (len > 0 && m_head->m_pkthdr.len > len) {
1026 * This sending could be time consuming; let callers
1027 * dispatch this packet sending (and sending of any
1028 * following up packets) to tx taskqueue.
1030 IF_PREPEND(&ifp->if_snd, m_head);
1034 txd = hn_txdesc_get(sc);
1036 sc->hn_no_txdescs++;
1037 IF_PREPEND(&ifp->if_snd, m_head);
1038 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1042 error = hn_encap(sc, txd, &m_head);
1044 /* Both txd and m_head are freed */
1049 * Make sure that txd is not freed before ETHER_BPF_MTAP.
1051 hn_txdesc_hold(txd);
1052 error = hv_nv_on_send(device_ctx, &txd->netvsc_pkt);
1054 ETHER_BPF_MTAP(ifp, m_head);
1055 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1057 hn_txdesc_put(sc, txd);
1059 if (__predict_false(error)) {
1063 * This should "really rarely" happen.
1065 * XXX Too many RX to be acked or too many sideband
1066 * commands to run? Ask netvsc_channel_rollup()
1067 * to kick start later.
1071 sc->hn_send_failed++;
1074 * Try sending again after set hn_txeof;
1075 * in case that we missed the last
1076 * netvsc_channel_rollup().
1080 if_printf(ifp, "send failed\n");
1083 * This mbuf will be prepended, don't free it
1084 * in hn_txdesc_put(); only unload it from the
1085 * DMA map in hn_txdesc_put(), if it was loaded.
1088 freed = hn_txdesc_put(sc, txd);
1090 ("fail to free txd upon send error"));
1092 sc->hn_send_failed++;
1093 IF_PREPEND(&ifp->if_snd, m_head);
1094 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1102 * Link up/down notification
1105 netvsc_linkstatus_callback(struct hv_device *device_obj, uint32_t status)
1107 hn_softc_t *sc = device_get_softc(device_obj->device);
1121 * Append the specified data to the indicated mbuf chain,
1122 * Extend the mbuf chain if the new data does not fit in
1125 * This is a minor rewrite of m_append() from sys/kern/uipc_mbuf.c.
1126 * There should be an equivalent in the kernel mbuf code,
1127 * but there does not appear to be one yet.
1129 * Differs from m_append() in that additional mbufs are
1130 * allocated with cluster size MJUMPAGESIZE, and filled
1133 * Return 1 if able to complete the job; otherwise 0.
1136 hv_m_append(struct mbuf *m0, int len, c_caddr_t cp)
1139 int remainder, space;
1141 for (m = m0; m->m_next != NULL; m = m->m_next)
1144 space = M_TRAILINGSPACE(m);
1147 * Copy into available space.
1149 if (space > remainder)
1151 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1156 while (remainder > 0) {
1158 * Allocate a new mbuf; could check space
1159 * and allocate a cluster instead.
1161 n = m_getjcl(M_DONTWAIT, m->m_type, 0, MJUMPAGESIZE);
1164 n->m_len = min(MJUMPAGESIZE, remainder);
1165 bcopy(cp, mtod(n, caddr_t), n->m_len);
1167 remainder -= n->m_len;
1171 if (m0->m_flags & M_PKTHDR)
1172 m0->m_pkthdr.len += len - remainder;
1174 return (remainder == 0);
1179 * Called when we receive a data packet from the "wire" on the
1182 * Note: This is no longer used as a callback
1185 netvsc_recv(struct hv_device *device_ctx, netvsc_packet *packet,
1186 rndis_tcp_ip_csum_info *csum_info)
1188 hn_softc_t *sc = (hn_softc_t *)device_get_softc(device_ctx->device);
1191 device_t dev = device_ctx->device;
1192 int size, do_lro = 0, do_csum = 1;
1195 return (0); /* TODO: KYS how can this be! */
1200 ifp = sc->arpcom.ac_ifp;
1202 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1207 * Bail out if packet contains more data than configured MTU.
1209 if (packet->tot_data_buf_len > (ifp->if_mtu + ETHER_HDR_LEN)) {
1211 } else if (packet->tot_data_buf_len <= MHLEN) {
1212 m_new = m_gethdr(M_NOWAIT, MT_DATA);
1215 memcpy(mtod(m_new, void *), packet->data,
1216 packet->tot_data_buf_len);
1217 m_new->m_pkthdr.len = m_new->m_len = packet->tot_data_buf_len;
1218 sc->hn_small_pkts++;
1221 * Get an mbuf with a cluster. For packets 2K or less,
1222 * get a standard 2K cluster. For anything larger, get a
1223 * 4K cluster. Any buffers larger than 4K can cause problems
1224 * if looped around to the Hyper-V TX channel, so avoid them.
1227 if (packet->tot_data_buf_len > MCLBYTES) {
1229 size = MJUMPAGESIZE;
1232 m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, size);
1233 if (m_new == NULL) {
1234 device_printf(dev, "alloc mbuf failed.\n");
1238 hv_m_append(m_new, packet->tot_data_buf_len, packet->data);
1240 m_new->m_pkthdr.rcvif = ifp;
1242 if (__predict_false((ifp->if_capenable & IFCAP_RXCSUM) == 0))
1245 /* receive side checksum offload */
1246 if (csum_info != NULL) {
1247 /* IP csum offload */
1248 if (csum_info->receive.ip_csum_succeeded && do_csum) {
1249 m_new->m_pkthdr.csum_flags |=
1250 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1254 /* TCP/UDP csum offload */
1255 if ((csum_info->receive.tcp_csum_succeeded ||
1256 csum_info->receive.udp_csum_succeeded) && do_csum) {
1257 m_new->m_pkthdr.csum_flags |=
1258 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1259 m_new->m_pkthdr.csum_data = 0xffff;
1260 if (csum_info->receive.tcp_csum_succeeded)
1266 if (csum_info->receive.ip_csum_succeeded &&
1267 csum_info->receive.tcp_csum_succeeded)
1270 const struct ether_header *eh;
1275 if (m_new->m_len < hoff)
1277 eh = mtod(m_new, struct ether_header *);
1278 etype = ntohs(eh->ether_type);
1279 if (etype == ETHERTYPE_VLAN) {
1280 const struct ether_vlan_header *evl;
1282 hoff = sizeof(*evl);
1283 if (m_new->m_len < hoff)
1285 evl = mtod(m_new, struct ether_vlan_header *);
1286 etype = ntohs(evl->evl_proto);
1289 if (etype == ETHERTYPE_IP) {
1292 pr = hn_check_iplen(m_new, hoff);
1293 if (pr == IPPROTO_TCP) {
1295 (sc->hn_trust_hcsum & HN_TRUST_HCSUM_TCP)) {
1296 sc->hn_csum_trusted++;
1297 m_new->m_pkthdr.csum_flags |=
1298 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1299 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1300 m_new->m_pkthdr.csum_data = 0xffff;
1302 /* Rely on SW csum verification though... */
1304 } else if (pr == IPPROTO_UDP) {
1306 (sc->hn_trust_hcsum & HN_TRUST_HCSUM_UDP)) {
1307 sc->hn_csum_trusted++;
1308 m_new->m_pkthdr.csum_flags |=
1309 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1310 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1311 m_new->m_pkthdr.csum_data = 0xffff;
1313 } else if (pr != IPPROTO_DONE && do_csum &&
1314 (sc->hn_trust_hcsum & HN_TRUST_HCSUM_IP)) {
1315 sc->hn_csum_trusted++;
1316 m_new->m_pkthdr.csum_flags |=
1317 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1322 if ((packet->vlan_tci != 0) &&
1323 (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0) {
1324 m_new->m_pkthdr.ether_vtag = packet->vlan_tci;
1325 m_new->m_flags |= M_VLANTAG;
1329 * Note: Moved RX completion back to hv_nv_on_receive() so all
1330 * messages (not just data messages) will trigger a response.
1335 if ((ifp->if_capenable & IFCAP_LRO) && do_lro) {
1336 #if defined(INET) || defined(INET6)
1337 struct lro_ctrl *lro = &sc->hn_lro;
1341 if (tcp_lro_rx(lro, m_new, 0) == 0) {
1349 /* We're not holding the lock here, so don't release it */
1350 (*ifp->if_input)(ifp, m_new);
1356 netvsc_recv_rollup(struct hv_device *device_ctx __unused)
1361 * Rules for using sc->temp_unusable:
1362 * 1. sc->temp_unusable can only be read or written while holding NV_LOCK()
1363 * 2. code reading sc->temp_unusable under NV_LOCK(), and finding
1364 * sc->temp_unusable set, must release NV_LOCK() and exit
1365 * 3. to retain exclusive control of the interface,
1366 * sc->temp_unusable must be set by code before releasing NV_LOCK()
1367 * 4. only code setting sc->temp_unusable can clear sc->temp_unusable
1368 * 5. code setting sc->temp_unusable must eventually clear sc->temp_unusable
1372 * Standard ioctl entry point. Called when the user wants to configure
1376 hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1378 hn_softc_t *sc = ifp->if_softc;
1379 struct ifreq *ifr = (struct ifreq *)data;
1381 struct ifaddr *ifa = (struct ifaddr *)data;
1383 netvsc_device_info device_info;
1384 struct hv_device *hn_dev;
1385 int mask, error = 0;
1386 int retry_cnt = 500;
1392 if (ifa->ifa_addr->sa_family == AF_INET) {
1393 ifp->if_flags |= IFF_UP;
1394 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1396 arp_ifinit(ifp, ifa);
1399 error = ether_ioctl(ifp, cmd, data);
1402 hn_dev = vmbus_get_devctx(sc->hn_dev);
1404 /* Check MTU value change */
1405 if (ifp->if_mtu == ifr->ifr_mtu)
1408 if (ifr->ifr_mtu > NETVSC_MAX_CONFIGURABLE_MTU) {
1413 /* Obtain and record requested MTU */
1414 ifp->if_mtu = ifr->ifr_mtu;
1416 * Make sure that LRO high watermark is still valid,
1417 * after MTU change (the 2*MTU limit).
1419 if (!HN_LRO_HIWAT_ISVALID(sc, sc->hn_lro_hiwat))
1420 hn_set_lro_hiwat(sc, HN_LRO_HIWAT_MTULIM(ifp));
1424 if (!sc->temp_unusable) {
1425 sc->temp_unusable = TRUE;
1429 if (retry_cnt > 0) {
1433 } while (retry_cnt > 0);
1435 if (retry_cnt == 0) {
1440 /* We must remove and add back the device to cause the new
1441 * MTU to take effect. This includes tearing down, but not
1442 * deleting the channel, then bringing it back up.
1444 error = hv_rf_on_device_remove(hn_dev, HV_RF_NV_RETAIN_CHANNEL);
1447 sc->temp_unusable = FALSE;
1451 error = hv_rf_on_device_add(hn_dev, &device_info);
1454 sc->temp_unusable = FALSE;
1459 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
1460 if (sc->hn_tx_chimney_size > sc->hn_tx_chimney_max)
1461 sc->hn_tx_chimney_size = sc->hn_tx_chimney_max;
1462 hn_ifinit_locked(sc);
1465 sc->temp_unusable = FALSE;
1471 if (!sc->temp_unusable) {
1472 sc->temp_unusable = TRUE;
1476 if (retry_cnt > 0) {
1480 } while (retry_cnt > 0);
1482 if (retry_cnt == 0) {
1487 if (ifp->if_flags & IFF_UP) {
1489 * If only the state of the PROMISC flag changed,
1490 * then just use the 'set promisc mode' command
1491 * instead of reinitializing the entire NIC. Doing
1492 * a full re-init means reloading the firmware and
1493 * waiting for it to start up, which may take a
1497 /* Fixme: Promiscuous mode? */
1498 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1499 ifp->if_flags & IFF_PROMISC &&
1500 !(sc->hn_if_flags & IFF_PROMISC)) {
1501 /* do something here for Hyper-V */
1502 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1503 !(ifp->if_flags & IFF_PROMISC) &&
1504 sc->hn_if_flags & IFF_PROMISC) {
1505 /* do something here for Hyper-V */
1508 hn_ifinit_locked(sc);
1510 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1515 sc->temp_unusable = FALSE;
1517 sc->hn_if_flags = ifp->if_flags;
1523 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1524 if (mask & IFCAP_TXCSUM) {
1525 ifp->if_capenable ^= IFCAP_TXCSUM;
1526 if (ifp->if_capenable & IFCAP_TXCSUM)
1527 ifp->if_hwassist |= sc->hn_csum_assist;
1529 ifp->if_hwassist &= ~sc->hn_csum_assist;
1532 if (mask & IFCAP_RXCSUM)
1533 ifp->if_capenable ^= IFCAP_RXCSUM;
1535 if (mask & IFCAP_LRO)
1536 ifp->if_capenable ^= IFCAP_LRO;
1538 if (mask & IFCAP_TSO4) {
1539 ifp->if_capenable ^= IFCAP_TSO4;
1540 if (ifp->if_capenable & IFCAP_TSO4)
1541 ifp->if_hwassist |= CSUM_IP_TSO;
1543 ifp->if_hwassist &= ~CSUM_IP_TSO;
1546 if (mask & IFCAP_TSO6) {
1547 ifp->if_capenable ^= IFCAP_TSO6;
1548 if (ifp->if_capenable & IFCAP_TSO6)
1549 ifp->if_hwassist |= CSUM_IP6_TSO;
1551 ifp->if_hwassist &= ~CSUM_IP6_TSO;
1560 /* Fixme: Multicast mode? */
1561 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1563 netvsc_setmulti(sc);
1572 error = ifmedia_ioctl(ifp, ifr, &sc->hn_media, cmd);
1575 error = ether_ioctl(ifp, cmd, data);
1586 hn_stop(hn_softc_t *sc)
1590 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1595 printf(" Closing Device ...\n");
1597 atomic_clear_int(&ifp->if_drv_flags,
1598 (IFF_DRV_RUNNING | IFF_DRV_OACTIVE));
1599 if_link_state_change(ifp, LINK_STATE_DOWN);
1600 sc->hn_initdone = 0;
1602 ret = hv_rf_on_close(device_ctx);
1606 * FreeBSD transmit entry point
1609 hn_start(struct ifnet *ifp)
1611 struct hn_softc *sc = ifp->if_softc;
1613 if (sc->hn_sched_tx)
1616 if (NV_TRYLOCK(sc)) {
1619 sched = hn_start_locked(ifp, sc->hn_direct_tx_size);
1625 taskqueue_enqueue_fast(sc->hn_tx_taskq, &sc->hn_start_task);
1629 hn_start_txeof(struct ifnet *ifp)
1631 struct hn_softc *sc = ifp->if_softc;
1633 if (sc->hn_sched_tx)
1636 if (NV_TRYLOCK(sc)) {
1639 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1640 sched = hn_start_locked(ifp, sc->hn_direct_tx_size);
1643 taskqueue_enqueue_fast(sc->hn_tx_taskq,
1644 &sc->hn_start_task);
1649 * Release the OACTIVE earlier, with the hope, that
1650 * others could catch up. The task will clear the
1651 * flag again with the NV_LOCK to avoid possible
1654 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1655 taskqueue_enqueue_fast(sc->hn_tx_taskq, &sc->hn_txeof_task);
1663 hn_ifinit_locked(hn_softc_t *sc)
1666 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1671 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1675 hv_promisc_mode = 1;
1677 ret = hv_rf_on_open(device_ctx);
1681 sc->hn_initdone = 1;
1683 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1684 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_RUNNING);
1685 if_link_state_change(ifp, LINK_STATE_UP);
1692 hn_ifinit(void *xsc)
1694 hn_softc_t *sc = xsc;
1697 if (sc->temp_unusable) {
1701 sc->temp_unusable = TRUE;
1704 hn_ifinit_locked(sc);
1707 sc->temp_unusable = FALSE;
1716 hn_watchdog(struct ifnet *ifp)
1721 printf("hn%d: watchdog timeout -- resetting\n", sc->hn_unit);
1722 hn_ifinit(sc); /*???*/
1729 hn_lro_hiwat_sysctl(SYSCTL_HANDLER_ARGS)
1731 struct hn_softc *sc = arg1;
1734 hiwat = sc->hn_lro_hiwat;
1735 error = sysctl_handle_int(oidp, &hiwat, 0, req);
1736 if (error || req->newptr == NULL)
1739 if (!HN_LRO_HIWAT_ISVALID(sc, hiwat))
1742 if (sc->hn_lro_hiwat != hiwat)
1743 hn_set_lro_hiwat(sc, hiwat);
1746 #endif /* HN_LRO_HIWAT */
1749 hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS)
1751 struct hn_softc *sc = arg1;
1756 if (sc->hn_trust_hcsum & hcsum)
1759 error = sysctl_handle_int(oidp, &on, 0, req);
1760 if (error || req->newptr == NULL)
1765 sc->hn_trust_hcsum |= hcsum;
1767 sc->hn_trust_hcsum &= ~hcsum;
1773 hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS)
1775 struct hn_softc *sc = arg1;
1776 int chimney_size, error;
1778 chimney_size = sc->hn_tx_chimney_size;
1779 error = sysctl_handle_int(oidp, &chimney_size, 0, req);
1780 if (error || req->newptr == NULL)
1783 if (chimney_size > sc->hn_tx_chimney_max || chimney_size <= 0)
1786 if (sc->hn_tx_chimney_size != chimney_size)
1787 sc->hn_tx_chimney_size = chimney_size;
1792 hn_check_iplen(const struct mbuf *m, int hoff)
1794 const struct ip *ip;
1795 int len, iphlen, iplen;
1796 const struct tcphdr *th;
1797 int thoff; /* TCP data offset */
1799 len = hoff + sizeof(struct ip);
1801 /* The packet must be at least the size of an IP header. */
1802 if (m->m_pkthdr.len < len)
1803 return IPPROTO_DONE;
1805 /* The fixed IP header must reside completely in the first mbuf. */
1807 return IPPROTO_DONE;
1809 ip = mtodo(m, hoff);
1811 /* Bound check the packet's stated IP header length. */
1812 iphlen = ip->ip_hl << 2;
1813 if (iphlen < sizeof(struct ip)) /* minimum header length */
1814 return IPPROTO_DONE;
1816 /* The full IP header must reside completely in the one mbuf. */
1817 if (m->m_len < hoff + iphlen)
1818 return IPPROTO_DONE;
1820 iplen = ntohs(ip->ip_len);
1823 * Check that the amount of data in the buffers is as
1824 * at least much as the IP header would have us expect.
1826 if (m->m_pkthdr.len < hoff + iplen)
1827 return IPPROTO_DONE;
1830 * Ignore IP fragments.
1832 if (ntohs(ip->ip_off) & (IP_OFFMASK | IP_MF))
1833 return IPPROTO_DONE;
1836 * The TCP/IP or UDP/IP header must be entirely contained within
1837 * the first fragment of a packet.
1841 if (iplen < iphlen + sizeof(struct tcphdr))
1842 return IPPROTO_DONE;
1843 if (m->m_len < hoff + iphlen + sizeof(struct tcphdr))
1844 return IPPROTO_DONE;
1845 th = (const struct tcphdr *)((const uint8_t *)ip + iphlen);
1846 thoff = th->th_off << 2;
1847 if (thoff < sizeof(struct tcphdr) || thoff + iphlen > iplen)
1848 return IPPROTO_DONE;
1849 if (m->m_len < hoff + iphlen + thoff)
1850 return IPPROTO_DONE;
1853 if (iplen < iphlen + sizeof(struct udphdr))
1854 return IPPROTO_DONE;
1855 if (m->m_len < hoff + iphlen + sizeof(struct udphdr))
1856 return IPPROTO_DONE;
1860 return IPPROTO_DONE;
1867 hn_dma_map_paddr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
1869 bus_addr_t *paddr = arg;
1874 KASSERT(nseg == 1, ("too many segments %d!", nseg));
1875 *paddr = segs->ds_addr;
1879 hn_create_tx_ring(struct hn_softc *sc)
1881 bus_dma_tag_t parent_dtag;
1884 sc->hn_txdesc_cnt = HN_TX_DESC_CNT;
1885 sc->hn_txdesc = malloc(sizeof(struct hn_txdesc) * sc->hn_txdesc_cnt,
1886 M_NETVSC, M_WAITOK | M_ZERO);
1887 SLIST_INIT(&sc->hn_txlist);
1888 mtx_init(&sc->hn_txlist_spin, "hn txlist", NULL, MTX_SPIN);
1890 parent_dtag = bus_get_dma_tag(sc->hn_dev);
1892 /* DMA tag for RNDIS messages. */
1893 error = bus_dma_tag_create(parent_dtag, /* parent */
1894 HN_RNDIS_MSG_ALIGN, /* alignment */
1895 HN_RNDIS_MSG_BOUNDARY, /* boundary */
1896 BUS_SPACE_MAXADDR, /* lowaddr */
1897 BUS_SPACE_MAXADDR, /* highaddr */
1898 NULL, NULL, /* filter, filterarg */
1899 HN_RNDIS_MSG_LEN, /* maxsize */
1901 HN_RNDIS_MSG_LEN, /* maxsegsize */
1903 NULL, /* lockfunc */
1904 NULL, /* lockfuncarg */
1905 &sc->hn_tx_rndis_dtag);
1907 device_printf(sc->hn_dev, "failed to create rndis dmatag\n");
1911 /* DMA tag for data. */
1912 error = bus_dma_tag_create(parent_dtag, /* parent */
1914 HN_TX_DATA_BOUNDARY, /* boundary */
1915 BUS_SPACE_MAXADDR, /* lowaddr */
1916 BUS_SPACE_MAXADDR, /* highaddr */
1917 NULL, NULL, /* filter, filterarg */
1918 HN_TX_DATA_MAXSIZE, /* maxsize */
1919 HN_TX_DATA_SEGCNT_MAX, /* nsegments */
1920 HN_TX_DATA_SEGSIZE, /* maxsegsize */
1922 NULL, /* lockfunc */
1923 NULL, /* lockfuncarg */
1924 &sc->hn_tx_data_dtag);
1926 device_printf(sc->hn_dev, "failed to create data dmatag\n");
1930 for (i = 0; i < sc->hn_txdesc_cnt; ++i) {
1931 struct hn_txdesc *txd = &sc->hn_txdesc[i];
1936 * Allocate and load RNDIS messages.
1938 error = bus_dmamem_alloc(sc->hn_tx_rndis_dtag,
1939 (void **)&txd->rndis_msg,
1940 BUS_DMA_WAITOK | BUS_DMA_COHERENT,
1941 &txd->rndis_msg_dmap);
1943 device_printf(sc->hn_dev,
1944 "failed to allocate rndis_msg, %d\n", i);
1948 error = bus_dmamap_load(sc->hn_tx_rndis_dtag,
1949 txd->rndis_msg_dmap,
1950 txd->rndis_msg, HN_RNDIS_MSG_LEN,
1951 hn_dma_map_paddr, &txd->rndis_msg_paddr,
1954 device_printf(sc->hn_dev,
1955 "failed to load rndis_msg, %d\n", i);
1956 bus_dmamem_free(sc->hn_tx_rndis_dtag,
1957 txd->rndis_msg, txd->rndis_msg_dmap);
1961 /* DMA map for TX data. */
1962 error = bus_dmamap_create(sc->hn_tx_data_dtag, 0,
1965 device_printf(sc->hn_dev,
1966 "failed to allocate tx data dmamap\n");
1967 bus_dmamap_unload(sc->hn_tx_rndis_dtag,
1968 txd->rndis_msg_dmap);
1969 bus_dmamem_free(sc->hn_tx_rndis_dtag,
1970 txd->rndis_msg, txd->rndis_msg_dmap);
1974 /* All set, put it to list */
1975 txd->flags |= HN_TXD_FLAG_ONLIST;
1976 SLIST_INSERT_HEAD(&sc->hn_txlist, txd, link);
1978 sc->hn_txdesc_avail = sc->hn_txdesc_cnt;
1984 hn_destroy_tx_ring(struct hn_softc *sc)
1986 struct hn_txdesc *txd;
1988 while ((txd = SLIST_FIRST(&sc->hn_txlist)) != NULL) {
1989 KASSERT(txd->m == NULL, ("still has mbuf installed"));
1990 KASSERT((txd->flags & HN_TXD_FLAG_DMAMAP) == 0,
1991 ("still dma mapped"));
1992 SLIST_REMOVE_HEAD(&sc->hn_txlist, link);
1994 bus_dmamap_unload(sc->hn_tx_rndis_dtag,
1995 txd->rndis_msg_dmap);
1996 bus_dmamem_free(sc->hn_tx_rndis_dtag,
1997 txd->rndis_msg, txd->rndis_msg_dmap);
1999 bus_dmamap_destroy(sc->hn_tx_data_dtag, txd->data_dmap);
2002 if (sc->hn_tx_data_dtag != NULL)
2003 bus_dma_tag_destroy(sc->hn_tx_data_dtag);
2004 if (sc->hn_tx_rndis_dtag != NULL)
2005 bus_dma_tag_destroy(sc->hn_tx_rndis_dtag);
2006 free(sc->hn_txdesc, M_NETVSC);
2007 mtx_destroy(&sc->hn_txlist_spin);
2011 hn_start_taskfunc(void *xsc, int pending __unused)
2013 struct hn_softc *sc = xsc;
2016 hn_start_locked(sc->hn_ifp, 0);
2021 hn_txeof_taskfunc(void *xsc, int pending __unused)
2023 struct hn_softc *sc = xsc;
2024 struct ifnet *ifp = sc->hn_ifp;
2027 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
2028 hn_start_locked(ifp, 0);
2032 static device_method_t netvsc_methods[] = {
2033 /* Device interface */
2034 DEVMETHOD(device_probe, netvsc_probe),
2035 DEVMETHOD(device_attach, netvsc_attach),
2036 DEVMETHOD(device_detach, netvsc_detach),
2037 DEVMETHOD(device_shutdown, netvsc_shutdown),
2042 static driver_t netvsc_driver = {
2048 static devclass_t netvsc_devclass;
2050 DRIVER_MODULE(hn, vmbus, netvsc_driver, netvsc_devclass, 0, 0);
2051 MODULE_VERSION(hn, 1);
2052 MODULE_DEPEND(hn, vmbus, 1, 1, 1);