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_RNDIS_MSG_LEN \
136 (sizeof(rndis_msg) + \
137 RNDIS_VLAN_PPI_SIZE + \
138 RNDIS_TSO_PPI_SIZE + \
140 #define HN_RNDIS_MSG_BOUNDARY PAGE_SIZE
141 #define HN_RNDIS_MSG_ALIGN CACHE_LINE_SIZE
143 #define HN_TX_DATA_BOUNDARY PAGE_SIZE
144 #define HN_TX_DATA_MAXSIZE IP_MAXPACKET
145 #define HN_TX_DATA_SEGSIZE PAGE_SIZE
146 #define HN_TX_DATA_SEGCNT_MAX \
147 (NETVSC_PACKET_MAXPAGE - HV_RF_NUM_TX_RESERVED_PAGE_BUFS)
149 #define HN_DIRECT_TX_SIZE_DEF 128
152 SLIST_ENTRY(hn_txdesc) link;
156 uint32_t flags; /* HN_TXD_FLAG_ */
157 netvsc_packet netvsc_pkt; /* XXX to be removed */
159 bus_dmamap_t data_dmap;
161 bus_addr_t rndis_msg_paddr;
162 rndis_msg *rndis_msg;
163 bus_dmamap_t rndis_msg_dmap;
166 #define HN_TXD_FLAG_ONLIST 0x1
167 #define HN_TXD_FLAG_DMAMAP 0x2
170 * Only enable UDP checksum offloading when it is on 2012R2 or
171 * later. UDP checksum offloading doesn't work on earlier
174 #define HN_CSUM_ASSIST_WIN8 (CSUM_TCP)
175 #define HN_CSUM_ASSIST (CSUM_IP | CSUM_UDP | CSUM_TCP)
177 /* XXX move to netinet/tcp_lro.h */
178 #define HN_LRO_HIWAT_MAX 65535
179 #define HN_LRO_HIWAT_DEF HN_LRO_HIWAT_MAX
180 /* YYY 2*MTU is a bit rough, but should be good enough. */
181 #define HN_LRO_HIWAT_MTULIM(ifp) (2 * (ifp)->if_mtu)
182 #define HN_LRO_HIWAT_ISVALID(sc, hiwat) \
183 ((hiwat) >= HN_LRO_HIWAT_MTULIM((sc)->hn_ifp) || \
184 (hiwat) <= HN_LRO_HIWAT_MAX)
187 * Be aware that this sleepable mutex will exhibit WITNESS errors when
188 * certain TCP and ARP code paths are taken. This appears to be a
189 * well-known condition, as all other drivers checked use a sleeping
190 * mutex to protect their transmit paths.
191 * Also Be aware that mutexes do not play well with semaphores, and there
192 * is a conflicting semaphore in a certain channel code path.
194 #define NV_LOCK_INIT(_sc, _name) \
195 mtx_init(&(_sc)->hn_lock, _name, MTX_NETWORK_LOCK, MTX_DEF)
196 #define NV_LOCK(_sc) mtx_lock(&(_sc)->hn_lock)
197 #define NV_TRYLOCK(_sc) mtx_trylock(&(_sc)->hn_lock)
198 #define NV_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->hn_lock, MA_OWNED)
199 #define NV_UNLOCK(_sc) mtx_unlock(&(_sc)->hn_lock)
200 #define NV_LOCK_DESTROY(_sc) mtx_destroy(&(_sc)->hn_lock)
207 int hv_promisc_mode = 0; /* normal mode by default */
209 /* Trust tcp segements verification on host side. */
210 static int hn_trust_hosttcp = 1;
211 TUNABLE_INT("dev.hn.trust_hosttcp", &hn_trust_hosttcp);
213 /* Trust udp datagrams verification on host side. */
214 static int hn_trust_hostudp = 1;
215 TUNABLE_INT("dev.hn.trust_hostudp", &hn_trust_hostudp);
217 /* Trust ip packets verification on host side. */
218 static int hn_trust_hostip = 1;
219 TUNABLE_INT("dev.hn.trust_hostip", &hn_trust_hostip);
221 #if __FreeBSD_version >= 1100045
222 /* Limit TSO burst size */
223 static int hn_tso_maxlen = 0;
224 TUNABLE_INT("dev.hn.tso_maxlen", &hn_tso_maxlen);
227 /* Limit chimney send size */
228 static int hn_tx_chimney_size = 0;
229 TUNABLE_INT("dev.hn.tx_chimney_size", &hn_tx_chimney_size);
231 /* Limit the size of packet for direct transmission */
232 static int hn_direct_tx_size = HN_DIRECT_TX_SIZE_DEF;
233 TUNABLE_INT("dev.hn.direct_tx_size", &hn_direct_tx_size);
236 * Forward declarations
238 static void hn_stop(hn_softc_t *sc);
239 static void hn_ifinit_locked(hn_softc_t *sc);
240 static void hn_ifinit(void *xsc);
241 static int hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data);
242 static int hn_start_locked(struct ifnet *ifp, int len);
243 static void hn_start(struct ifnet *ifp);
244 static void hn_start_txeof(struct ifnet *ifp);
245 static int hn_ifmedia_upd(struct ifnet *ifp);
246 static void hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr);
248 static int hn_lro_hiwat_sysctl(SYSCTL_HANDLER_ARGS);
250 static int hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS);
251 static int hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS);
252 static int hn_check_iplen(const struct mbuf *, int);
253 static int hn_create_tx_ring(struct hn_softc *sc);
254 static void hn_destroy_tx_ring(struct hn_softc *sc);
255 static void hn_start_taskfunc(void *xsc, int pending);
256 static void hn_txeof_taskfunc(void *xsc, int pending);
257 static int hn_encap(struct hn_softc *, struct hn_txdesc *, struct mbuf **);
260 hn_set_lro_hiwat(struct hn_softc *sc, int hiwat)
262 sc->hn_lro_hiwat = hiwat;
264 sc->hn_lro.lro_hiwat = sc->hn_lro_hiwat;
269 hn_ifmedia_upd(struct ifnet *ifp __unused)
276 hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
278 struct hn_softc *sc = ifp->if_softc;
280 ifmr->ifm_status = IFM_AVALID;
281 ifmr->ifm_active = IFM_ETHER;
283 if (!sc->hn_carrier) {
284 ifmr->ifm_active |= IFM_NONE;
287 ifmr->ifm_status |= IFM_ACTIVE;
288 ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
291 /* {F8615163-DF3E-46c5-913F-F2D2F965ED0E} */
292 static const hv_guid g_net_vsc_device_type = {
293 .data = {0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
294 0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E}
298 * Standard probe entry point.
302 netvsc_probe(device_t dev)
306 p = vmbus_get_type(dev);
307 if (!memcmp(p, &g_net_vsc_device_type.data, sizeof(hv_guid))) {
308 device_set_desc(dev, "Synthetic Network Interface");
310 printf("Netvsc probe... DONE \n");
312 return (BUS_PROBE_DEFAULT);
319 * Standard attach entry point.
321 * Called when the driver is loaded. It allocates needed resources,
322 * and initializes the "hardware" and software.
325 netvsc_attach(device_t dev)
327 struct hv_device *device_ctx = vmbus_get_devctx(dev);
328 netvsc_device_info device_info;
330 int unit = device_get_unit(dev);
331 struct ifnet *ifp = NULL;
332 struct sysctl_oid_list *child;
333 struct sysctl_ctx_list *ctx;
335 #if __FreeBSD_version >= 1100045
339 sc = device_get_softc(dev);
344 bzero(sc, sizeof(hn_softc_t));
347 sc->hn_lro_hiwat = HN_LRO_HIWAT_DEF;
348 sc->hn_direct_tx_size = hn_direct_tx_size;
349 if (hn_trust_hosttcp)
350 sc->hn_trust_hcsum |= HN_TRUST_HCSUM_TCP;
351 if (hn_trust_hostudp)
352 sc->hn_trust_hcsum |= HN_TRUST_HCSUM_UDP;
354 sc->hn_trust_hcsum |= HN_TRUST_HCSUM_IP;
356 sc->hn_tx_taskq = taskqueue_create_fast("hn_tx", M_WAITOK,
357 taskqueue_thread_enqueue, &sc->hn_tx_taskq);
358 taskqueue_start_threads(&sc->hn_tx_taskq, 1, PI_NET, "%s tx",
359 device_get_nameunit(dev));
360 TASK_INIT(&sc->hn_start_task, 0, hn_start_taskfunc, sc);
361 TASK_INIT(&sc->hn_txeof_task, 0, hn_txeof_taskfunc, sc);
363 error = hn_create_tx_ring(sc);
367 NV_LOCK_INIT(sc, "NetVSCLock");
369 sc->hn_dev_obj = device_ctx;
371 ifp = sc->hn_ifp = sc->arpcom.ac_ifp = if_alloc(IFT_ETHER);
374 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
375 ifp->if_dunit = unit;
376 ifp->if_dname = NETVSC_DEVNAME;
378 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
379 ifp->if_ioctl = hn_ioctl;
380 ifp->if_start = hn_start;
381 ifp->if_init = hn_ifinit;
382 /* needed by hv_rf_on_device_add() code */
383 ifp->if_mtu = ETHERMTU;
384 IFQ_SET_MAXLEN(&ifp->if_snd, 512);
385 ifp->if_snd.ifq_drv_maxlen = 511;
386 IFQ_SET_READY(&ifp->if_snd);
388 ifmedia_init(&sc->hn_media, 0, hn_ifmedia_upd, hn_ifmedia_sts);
389 ifmedia_add(&sc->hn_media, IFM_ETHER | IFM_AUTO, 0, NULL);
390 ifmedia_set(&sc->hn_media, IFM_ETHER | IFM_AUTO);
391 /* XXX ifmedia_set really should do this for us */
392 sc->hn_media.ifm_media = sc->hn_media.ifm_cur->ifm_media;
395 * Tell upper layers that we support full VLAN capability.
397 ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
398 ifp->if_capabilities |=
399 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
402 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
405 if (hv_vmbus_protocal_version >= HV_VMBUS_VERSION_WIN8_1)
406 sc->hn_csum_assist = HN_CSUM_ASSIST;
408 sc->hn_csum_assist = HN_CSUM_ASSIST_WIN8;
409 ifp->if_hwassist = sc->hn_csum_assist | CSUM_TSO;
411 error = hv_rf_on_device_add(device_ctx, &device_info);
415 if (device_info.link_state == 0) {
419 #if defined(INET) || defined(INET6)
420 tcp_lro_init(&sc->hn_lro);
421 /* Driver private LRO settings */
422 sc->hn_lro.ifp = ifp;
424 sc->hn_lro.lro_hiwat = sc->hn_lro_hiwat;
426 #endif /* INET || INET6 */
428 #if __FreeBSD_version >= 1100045
429 tso_maxlen = hn_tso_maxlen;
430 if (tso_maxlen <= 0 || tso_maxlen > IP_MAXPACKET)
431 tso_maxlen = IP_MAXPACKET;
433 ifp->if_hw_tsomaxsegcount = HN_TX_DATA_SEGCNT_MAX;
434 ifp->if_hw_tsomaxsegsize = PAGE_SIZE;
435 ifp->if_hw_tsomax = tso_maxlen -
436 (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
439 ether_ifattach(ifp, device_info.mac_addr);
441 #if __FreeBSD_version >= 1100045
442 if_printf(ifp, "TSO: %u/%u/%u\n", ifp->if_hw_tsomax,
443 ifp->if_hw_tsomaxsegcount, ifp->if_hw_tsomaxsegsize);
446 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
447 sc->hn_tx_chimney_size = sc->hn_tx_chimney_max;
448 if (hn_tx_chimney_size > 0 &&
449 hn_tx_chimney_size < sc->hn_tx_chimney_max)
450 sc->hn_tx_chimney_size = hn_tx_chimney_size;
452 ctx = device_get_sysctl_ctx(dev);
453 child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
455 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "lro_queued",
456 CTLFLAG_RW, &sc->hn_lro.lro_queued, 0, "LRO queued");
457 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "lro_flushed",
458 CTLFLAG_RW, &sc->hn_lro.lro_flushed, 0, "LRO flushed");
459 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "lro_tried",
460 CTLFLAG_RW, &sc->hn_lro_tried, "# of LRO tries");
462 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_hiwat",
463 CTLTYPE_INT | CTLFLAG_RW, sc, 0, hn_lro_hiwat_sysctl,
464 "I", "LRO high watermark");
466 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hosttcp",
467 CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_TCP,
468 hn_trust_hcsum_sysctl, "I",
469 "Trust tcp segement verification on host side, "
470 "when csum info is missing");
471 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostudp",
472 CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_UDP,
473 hn_trust_hcsum_sysctl, "I",
474 "Trust udp datagram verification on host side, "
475 "when csum info is missing");
476 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostip",
477 CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_IP,
478 hn_trust_hcsum_sysctl, "I",
479 "Trust ip packet verification on host side, "
480 "when csum info is missing");
481 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "csum_ip",
482 CTLFLAG_RW, &sc->hn_csum_ip, "RXCSUM IP");
483 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "csum_tcp",
484 CTLFLAG_RW, &sc->hn_csum_tcp, "RXCSUM TCP");
485 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "csum_udp",
486 CTLFLAG_RW, &sc->hn_csum_udp, "RXCSUM UDP");
487 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "csum_trusted",
488 CTLFLAG_RW, &sc->hn_csum_trusted,
489 "# of packets that we trust host's csum verification");
490 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "small_pkts",
491 CTLFLAG_RW, &sc->hn_small_pkts, "# of small packets received");
492 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "no_txdescs",
493 CTLFLAG_RW, &sc->hn_no_txdescs, "# of times short of TX descs");
494 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "send_failed",
495 CTLFLAG_RW, &sc->hn_send_failed, "# of hyper-v sending failure");
496 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "txdma_failed",
497 CTLFLAG_RW, &sc->hn_txdma_failed, "# of TX DMA failure");
498 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_collapsed",
499 CTLFLAG_RW, &sc->hn_tx_collapsed, "# of TX mbuf collapsed");
500 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_chimney",
501 CTLFLAG_RW, &sc->hn_tx_chimney, "# of chimney send");
502 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_cnt",
503 CTLFLAG_RD, &sc->hn_txdesc_cnt, 0, "# of total TX descs");
504 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_avail",
505 CTLFLAG_RD, &sc->hn_txdesc_avail, 0, "# of available TX descs");
506 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_chimney_max",
507 CTLFLAG_RD, &sc->hn_tx_chimney_max, 0,
508 "Chimney send packet size upper boundary");
509 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_size",
510 CTLTYPE_INT | CTLFLAG_RW, sc, 0, hn_tx_chimney_size_sysctl,
511 "I", "Chimney send packet size limit");
512 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "direct_tx_size",
513 CTLFLAG_RW, &sc->hn_direct_tx_size, 0,
514 "Size of the packet for direct transmission");
517 struct sysctl_ctx_list *dc_ctx;
518 struct sysctl_oid_list *dc_child;
522 * Add sysctl nodes for devclass
524 dc = device_get_devclass(dev);
525 dc_ctx = devclass_get_sysctl_ctx(dc);
526 dc_child = SYSCTL_CHILDREN(devclass_get_sysctl_tree(dc));
528 SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "trust_hosttcp",
529 CTLFLAG_RD, &hn_trust_hosttcp, 0,
530 "Trust tcp segement verification on host side, "
531 "when csum info is missing (global setting)");
532 SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "trust_hostudp",
533 CTLFLAG_RD, &hn_trust_hostudp, 0,
534 "Trust udp datagram verification on host side, "
535 "when csum info is missing (global setting)");
536 SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "trust_hostip",
537 CTLFLAG_RD, &hn_trust_hostip, 0,
538 "Trust ip packet verification on host side, "
539 "when csum info is missing (global setting)");
540 SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "tx_chimney_size",
541 CTLFLAG_RD, &hn_tx_chimney_size, 0,
542 "Chimney send packet size limit");
543 #if __FreeBSD_version >= 1100045
544 SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "tso_maxlen",
545 CTLFLAG_RD, &hn_tso_maxlen, 0, "TSO burst limit");
547 SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "direct_tx_size",
548 CTLFLAG_RD, &hn_direct_tx_size, 0,
549 "Size of the packet for direct transmission");
554 hn_destroy_tx_ring(sc);
561 * Standard detach entry point
564 netvsc_detach(device_t dev)
566 struct hn_softc *sc = device_get_softc(dev);
567 struct hv_device *hv_device = vmbus_get_devctx(dev);
570 printf("netvsc_detach\n");
573 * XXXKYS: Need to clean up all our
574 * driver state; this is the driver
579 * XXXKYS: Need to stop outgoing traffic and unregister
583 hv_rf_on_device_remove(hv_device, HV_RF_NV_DESTROY_CHANNEL);
585 taskqueue_drain(sc->hn_tx_taskq, &sc->hn_start_task);
586 taskqueue_drain(sc->hn_tx_taskq, &sc->hn_txeof_task);
587 taskqueue_free(sc->hn_tx_taskq);
589 ifmedia_removeall(&sc->hn_media);
590 #if defined(INET) || defined(INET6)
591 tcp_lro_free(&sc->hn_lro);
593 hn_destroy_tx_ring(sc);
599 * Standard shutdown entry point
602 netvsc_shutdown(device_t dev)
608 hn_txdesc_dmamap_load(struct hn_softc *sc, struct hn_txdesc *txd,
609 struct mbuf **m_head, bus_dma_segment_t *segs, int *nsegs)
611 struct mbuf *m = *m_head;
614 error = bus_dmamap_load_mbuf_sg(sc->hn_tx_data_dtag, txd->data_dmap,
615 m, segs, nsegs, BUS_DMA_NOWAIT);
616 if (error == EFBIG) {
619 m_new = m_collapse(m, M_NOWAIT, HN_TX_DATA_SEGCNT_MAX);
624 sc->hn_tx_collapsed++;
626 error = bus_dmamap_load_mbuf_sg(sc->hn_tx_data_dtag,
627 txd->data_dmap, m, segs, nsegs, BUS_DMA_NOWAIT);
630 bus_dmamap_sync(sc->hn_tx_data_dtag, txd->data_dmap,
631 BUS_DMASYNC_PREWRITE);
632 txd->flags |= HN_TXD_FLAG_DMAMAP;
638 hn_txdesc_dmamap_unload(struct hn_softc *sc, struct hn_txdesc *txd)
641 if (txd->flags & HN_TXD_FLAG_DMAMAP) {
642 bus_dmamap_sync(sc->hn_tx_data_dtag,
643 txd->data_dmap, BUS_DMASYNC_POSTWRITE);
644 bus_dmamap_unload(sc->hn_tx_data_dtag,
646 txd->flags &= ~HN_TXD_FLAG_DMAMAP;
651 hn_txdesc_put(struct hn_softc *sc, struct hn_txdesc *txd)
654 KASSERT((txd->flags & HN_TXD_FLAG_ONLIST) == 0,
655 ("put an onlist txd %#x", txd->flags));
657 KASSERT(txd->refs > 0, ("invalid txd refs %d", txd->refs));
658 if (atomic_fetchadd_int(&txd->refs, -1) != 1)
661 hn_txdesc_dmamap_unload(sc, txd);
662 if (txd->m != NULL) {
667 txd->flags |= HN_TXD_FLAG_ONLIST;
669 mtx_lock_spin(&sc->hn_txlist_spin);
670 KASSERT(sc->hn_txdesc_avail >= 0 &&
671 sc->hn_txdesc_avail < sc->hn_txdesc_cnt,
672 ("txdesc_put: invalid txd avail %d", sc->hn_txdesc_avail));
673 sc->hn_txdesc_avail++;
674 SLIST_INSERT_HEAD(&sc->hn_txlist, txd, link);
675 mtx_unlock_spin(&sc->hn_txlist_spin);
680 static __inline struct hn_txdesc *
681 hn_txdesc_get(struct hn_softc *sc)
683 struct hn_txdesc *txd;
685 mtx_lock_spin(&sc->hn_txlist_spin);
686 txd = SLIST_FIRST(&sc->hn_txlist);
688 KASSERT(sc->hn_txdesc_avail > 0,
689 ("txdesc_get: invalid txd avail %d", sc->hn_txdesc_avail));
690 sc->hn_txdesc_avail--;
691 SLIST_REMOVE_HEAD(&sc->hn_txlist, link);
693 mtx_unlock_spin(&sc->hn_txlist_spin);
696 KASSERT(txd->m == NULL && txd->refs == 0 &&
697 (txd->flags & HN_TXD_FLAG_ONLIST), ("invalid txd"));
698 txd->flags &= ~HN_TXD_FLAG_ONLIST;
705 hn_txdesc_hold(struct hn_txdesc *txd)
708 /* 0->1 transition will never work */
709 KASSERT(txd->refs > 0, ("invalid refs %d", txd->refs));
710 atomic_add_int(&txd->refs, 1);
714 * Send completion processing
716 * Note: It looks like offset 0 of buf is reserved to hold the softc
717 * pointer. The sc pointer is not currently needed in this function, and
718 * it is not presently populated by the TX function.
721 netvsc_xmit_completion(void *context)
723 netvsc_packet *packet = context;
724 struct hn_txdesc *txd;
727 txd = (struct hn_txdesc *)(uintptr_t)
728 packet->compl.send.send_completion_tid;
732 hn_txdesc_put(sc, txd);
736 netvsc_channel_rollup(struct hv_device *device_ctx)
738 struct hn_softc *sc = device_get_softc(device_ctx->device);
744 hn_start_txeof(sc->hn_ifp);
749 * This this function fails, then both txd and m_head0 will be freed
752 hn_encap(struct hn_softc *sc, struct hn_txdesc *txd, struct mbuf **m_head0)
754 bus_dma_segment_t segs[HN_TX_DATA_SEGCNT_MAX];
756 struct mbuf *m_head = *m_head0;
757 netvsc_packet *packet;
758 rndis_msg *rndis_mesg;
759 rndis_packet *rndis_pkt;
760 rndis_per_packet_info *rppi;
761 uint32_t rndis_msg_size;
763 packet = &txd->netvsc_pkt;
764 packet->is_data_pkt = TRUE;
765 packet->tot_data_buf_len = m_head->m_pkthdr.len;
768 * extension points to the area reserved for the
769 * rndis_filter_packet, which is placed just after
770 * the netvsc_packet (and rppi struct, if present;
771 * length is updated later).
773 rndis_mesg = txd->rndis_msg;
774 /* XXX not necessary */
775 memset(rndis_mesg, 0, HN_RNDIS_MSG_LEN);
776 rndis_mesg->ndis_msg_type = REMOTE_NDIS_PACKET_MSG;
778 rndis_pkt = &rndis_mesg->msg.packet;
779 rndis_pkt->data_offset = sizeof(rndis_packet);
780 rndis_pkt->data_length = packet->tot_data_buf_len;
781 rndis_pkt->per_pkt_info_offset = sizeof(rndis_packet);
783 rndis_msg_size = RNDIS_MESSAGE_SIZE(rndis_packet);
785 if (m_head->m_flags & M_VLANTAG) {
786 ndis_8021q_info *rppi_vlan_info;
788 rndis_msg_size += RNDIS_VLAN_PPI_SIZE;
789 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_VLAN_PPI_SIZE,
792 rppi_vlan_info = (ndis_8021q_info *)((uint8_t *)rppi +
793 rppi->per_packet_info_offset);
794 rppi_vlan_info->u1.s1.vlan_id =
795 m_head->m_pkthdr.ether_vtag & 0xfff;
798 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
799 rndis_tcp_tso_info *tso_info;
800 struct ether_vlan_header *eh;
804 * XXX need m_pullup and use mtodo
806 eh = mtod(m_head, struct ether_vlan_header*);
807 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN))
808 ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
810 ether_len = ETHER_HDR_LEN;
812 rndis_msg_size += RNDIS_TSO_PPI_SIZE;
813 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_TSO_PPI_SIZE,
814 tcp_large_send_info);
816 tso_info = (rndis_tcp_tso_info *)((uint8_t *)rppi +
817 rppi->per_packet_info_offset);
818 tso_info->lso_v2_xmit.type =
819 RNDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
822 if (m_head->m_pkthdr.csum_flags & CSUM_IP_TSO) {
824 (struct ip *)(m_head->m_data + ether_len);
825 unsigned long iph_len = ip->ip_hl << 2;
827 (struct tcphdr *)((caddr_t)ip + iph_len);
829 tso_info->lso_v2_xmit.ip_version =
830 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
834 th->th_sum = in_pseudo(ip->ip_src.s_addr,
835 ip->ip_dst.s_addr, htons(IPPROTO_TCP));
838 #if defined(INET6) && defined(INET)
843 struct ip6_hdr *ip6 = (struct ip6_hdr *)
844 (m_head->m_data + ether_len);
845 struct tcphdr *th = (struct tcphdr *)(ip6 + 1);
847 tso_info->lso_v2_xmit.ip_version =
848 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
850 th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
853 tso_info->lso_v2_xmit.tcp_header_offset = 0;
854 tso_info->lso_v2_xmit.mss = m_head->m_pkthdr.tso_segsz;
855 } else if (m_head->m_pkthdr.csum_flags & sc->hn_csum_assist) {
856 rndis_tcp_ip_csum_info *csum_info;
858 rndis_msg_size += RNDIS_CSUM_PPI_SIZE;
859 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_CSUM_PPI_SIZE,
861 csum_info = (rndis_tcp_ip_csum_info *)((uint8_t *)rppi +
862 rppi->per_packet_info_offset);
864 csum_info->xmit.is_ipv4 = 1;
865 if (m_head->m_pkthdr.csum_flags & CSUM_IP)
866 csum_info->xmit.ip_header_csum = 1;
868 if (m_head->m_pkthdr.csum_flags & CSUM_TCP) {
869 csum_info->xmit.tcp_csum = 1;
870 csum_info->xmit.tcp_header_offset = 0;
871 } else if (m_head->m_pkthdr.csum_flags & CSUM_UDP) {
872 csum_info->xmit.udp_csum = 1;
876 rndis_mesg->msg_len = packet->tot_data_buf_len + rndis_msg_size;
877 packet->tot_data_buf_len = rndis_mesg->msg_len;
880 * Chimney send, if the packet could fit into one chimney buffer.
882 if (packet->tot_data_buf_len < sc->hn_tx_chimney_size) {
883 netvsc_dev *net_dev = sc->net_dev;
884 uint32_t send_buf_section_idx;
886 send_buf_section_idx =
887 hv_nv_get_next_send_section(net_dev);
888 if (send_buf_section_idx !=
889 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX) {
890 uint8_t *dest = ((uint8_t *)net_dev->send_buf +
891 (send_buf_section_idx *
892 net_dev->send_section_size));
894 memcpy(dest, rndis_mesg, rndis_msg_size);
895 dest += rndis_msg_size;
896 m_copydata(m_head, 0, m_head->m_pkthdr.len, dest);
898 packet->send_buf_section_idx = send_buf_section_idx;
899 packet->send_buf_section_size =
900 packet->tot_data_buf_len;
901 packet->page_buf_count = 0;
907 error = hn_txdesc_dmamap_load(sc, txd, &m_head, segs, &nsegs);
912 * This mbuf is not linked w/ the txd yet, so free it now.
917 freed = hn_txdesc_put(sc, txd);
919 ("fail to free txd upon txdma error"));
921 sc->hn_txdma_failed++;
922 if_inc_counter(sc->hn_ifp, IFCOUNTER_OERRORS, 1);
927 packet->page_buf_count = nsegs + HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
929 /* send packet with page buffer */
930 packet->page_buffers[0].pfn = atop(txd->rndis_msg_paddr);
931 packet->page_buffers[0].offset = txd->rndis_msg_paddr & PAGE_MASK;
932 packet->page_buffers[0].length = rndis_msg_size;
935 * Fill the page buffers with mbuf info starting at index
936 * HV_RF_NUM_TX_RESERVED_PAGE_BUFS.
938 for (i = 0; i < nsegs; ++i) {
939 hv_vmbus_page_buffer *pb = &packet->page_buffers[
940 i + HV_RF_NUM_TX_RESERVED_PAGE_BUFS];
942 pb->pfn = atop(segs[i].ds_addr);
943 pb->offset = segs[i].ds_addr & PAGE_MASK;
944 pb->length = segs[i].ds_len;
947 packet->send_buf_section_idx =
948 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX;
949 packet->send_buf_section_size = 0;
953 /* Set the completion routine */
954 packet->compl.send.on_send_completion = netvsc_xmit_completion;
955 packet->compl.send.send_completion_context = packet;
956 packet->compl.send.send_completion_tid = (uint64_t)(uintptr_t)txd;
962 * Start a transmit of one or more packets
965 hn_start_locked(struct ifnet *ifp, int len)
967 struct hn_softc *sc = ifp->if_softc;
968 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
970 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
974 while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
975 int error, send_failed = 0;
976 struct hn_txdesc *txd;
979 IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
983 if (len > 0 && m_head->m_pkthdr.len > len) {
985 * This sending could be time consuming; let callers
986 * dispatch this packet sending (and sending of any
987 * following up packets) to tx taskqueue.
989 IF_PREPEND(&ifp->if_snd, m_head);
993 txd = hn_txdesc_get(sc);
996 IF_PREPEND(&ifp->if_snd, m_head);
997 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1001 error = hn_encap(sc, txd, &m_head);
1003 /* Both txd and m_head are freed */
1008 * Make sure that txd is not freed before ETHER_BPF_MTAP.
1010 hn_txdesc_hold(txd);
1011 error = hv_nv_on_send(device_ctx, &txd->netvsc_pkt);
1013 ETHER_BPF_MTAP(ifp, m_head);
1014 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1016 hn_txdesc_put(sc, txd);
1018 if (__predict_false(error)) {
1022 * This should "really rarely" happen.
1024 * XXX Too many RX to be acked or too many sideband
1025 * commands to run? Ask netvsc_channel_rollup()
1026 * to kick start later.
1030 sc->hn_send_failed++;
1033 * Try sending again after set hn_txeof;
1034 * in case that we missed the last
1035 * netvsc_channel_rollup().
1039 if_printf(ifp, "send failed\n");
1042 * This mbuf will be prepended, don't free it
1043 * in hn_txdesc_put(); only unload it from the
1044 * DMA map in hn_txdesc_put(), if it was loaded.
1047 freed = hn_txdesc_put(sc, txd);
1049 ("fail to free txd upon send error"));
1051 sc->hn_send_failed++;
1052 IF_PREPEND(&ifp->if_snd, m_head);
1053 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1061 * Link up/down notification
1064 netvsc_linkstatus_callback(struct hv_device *device_obj, uint32_t status)
1066 hn_softc_t *sc = device_get_softc(device_obj->device);
1080 * Append the specified data to the indicated mbuf chain,
1081 * Extend the mbuf chain if the new data does not fit in
1084 * This is a minor rewrite of m_append() from sys/kern/uipc_mbuf.c.
1085 * There should be an equivalent in the kernel mbuf code,
1086 * but there does not appear to be one yet.
1088 * Differs from m_append() in that additional mbufs are
1089 * allocated with cluster size MJUMPAGESIZE, and filled
1092 * Return 1 if able to complete the job; otherwise 0.
1095 hv_m_append(struct mbuf *m0, int len, c_caddr_t cp)
1098 int remainder, space;
1100 for (m = m0; m->m_next != NULL; m = m->m_next)
1103 space = M_TRAILINGSPACE(m);
1106 * Copy into available space.
1108 if (space > remainder)
1110 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1115 while (remainder > 0) {
1117 * Allocate a new mbuf; could check space
1118 * and allocate a cluster instead.
1120 n = m_getjcl(M_DONTWAIT, m->m_type, 0, MJUMPAGESIZE);
1123 n->m_len = min(MJUMPAGESIZE, remainder);
1124 bcopy(cp, mtod(n, caddr_t), n->m_len);
1126 remainder -= n->m_len;
1130 if (m0->m_flags & M_PKTHDR)
1131 m0->m_pkthdr.len += len - remainder;
1133 return (remainder == 0);
1138 * Called when we receive a data packet from the "wire" on the
1141 * Note: This is no longer used as a callback
1144 netvsc_recv(struct hv_device *device_ctx, netvsc_packet *packet,
1145 rndis_tcp_ip_csum_info *csum_info)
1147 hn_softc_t *sc = (hn_softc_t *)device_get_softc(device_ctx->device);
1150 device_t dev = device_ctx->device;
1151 int size, do_lro = 0, do_csum = 1;
1154 return (0); /* TODO: KYS how can this be! */
1159 ifp = sc->arpcom.ac_ifp;
1161 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1166 * Bail out if packet contains more data than configured MTU.
1168 if (packet->tot_data_buf_len > (ifp->if_mtu + ETHER_HDR_LEN)) {
1170 } else if (packet->tot_data_buf_len <= MHLEN) {
1171 m_new = m_gethdr(M_NOWAIT, MT_DATA);
1174 memcpy(mtod(m_new, void *), packet->data,
1175 packet->tot_data_buf_len);
1176 m_new->m_pkthdr.len = m_new->m_len = packet->tot_data_buf_len;
1177 sc->hn_small_pkts++;
1180 * Get an mbuf with a cluster. For packets 2K or less,
1181 * get a standard 2K cluster. For anything larger, get a
1182 * 4K cluster. Any buffers larger than 4K can cause problems
1183 * if looped around to the Hyper-V TX channel, so avoid them.
1186 if (packet->tot_data_buf_len > MCLBYTES) {
1188 size = MJUMPAGESIZE;
1191 m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, size);
1192 if (m_new == NULL) {
1193 device_printf(dev, "alloc mbuf failed.\n");
1197 hv_m_append(m_new, packet->tot_data_buf_len, packet->data);
1199 m_new->m_pkthdr.rcvif = ifp;
1201 if (__predict_false((ifp->if_capenable & IFCAP_RXCSUM) == 0))
1204 /* receive side checksum offload */
1205 if (csum_info != NULL) {
1206 /* IP csum offload */
1207 if (csum_info->receive.ip_csum_succeeded && do_csum) {
1208 m_new->m_pkthdr.csum_flags |=
1209 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1213 /* TCP/UDP csum offload */
1214 if ((csum_info->receive.tcp_csum_succeeded ||
1215 csum_info->receive.udp_csum_succeeded) && do_csum) {
1216 m_new->m_pkthdr.csum_flags |=
1217 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1218 m_new->m_pkthdr.csum_data = 0xffff;
1219 if (csum_info->receive.tcp_csum_succeeded)
1225 if (csum_info->receive.ip_csum_succeeded &&
1226 csum_info->receive.tcp_csum_succeeded)
1229 const struct ether_header *eh;
1234 if (m_new->m_len < hoff)
1236 eh = mtod(m_new, struct ether_header *);
1237 etype = ntohs(eh->ether_type);
1238 if (etype == ETHERTYPE_VLAN) {
1239 const struct ether_vlan_header *evl;
1241 hoff = sizeof(*evl);
1242 if (m_new->m_len < hoff)
1244 evl = mtod(m_new, struct ether_vlan_header *);
1245 etype = ntohs(evl->evl_proto);
1248 if (etype == ETHERTYPE_IP) {
1251 pr = hn_check_iplen(m_new, hoff);
1252 if (pr == IPPROTO_TCP) {
1254 (sc->hn_trust_hcsum & HN_TRUST_HCSUM_TCP)) {
1255 sc->hn_csum_trusted++;
1256 m_new->m_pkthdr.csum_flags |=
1257 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1258 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1259 m_new->m_pkthdr.csum_data = 0xffff;
1261 /* Rely on SW csum verification though... */
1263 } else if (pr == IPPROTO_UDP) {
1265 (sc->hn_trust_hcsum & HN_TRUST_HCSUM_UDP)) {
1266 sc->hn_csum_trusted++;
1267 m_new->m_pkthdr.csum_flags |=
1268 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1269 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1270 m_new->m_pkthdr.csum_data = 0xffff;
1272 } else if (pr != IPPROTO_DONE && do_csum &&
1273 (sc->hn_trust_hcsum & HN_TRUST_HCSUM_IP)) {
1274 sc->hn_csum_trusted++;
1275 m_new->m_pkthdr.csum_flags |=
1276 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1281 if ((packet->vlan_tci != 0) &&
1282 (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0) {
1283 m_new->m_pkthdr.ether_vtag = packet->vlan_tci;
1284 m_new->m_flags |= M_VLANTAG;
1288 * Note: Moved RX completion back to hv_nv_on_receive() so all
1289 * messages (not just data messages) will trigger a response.
1294 if ((ifp->if_capenable & IFCAP_LRO) && do_lro) {
1295 #if defined(INET) || defined(INET6)
1296 struct lro_ctrl *lro = &sc->hn_lro;
1300 if (tcp_lro_rx(lro, m_new, 0) == 0) {
1308 /* We're not holding the lock here, so don't release it */
1309 (*ifp->if_input)(ifp, m_new);
1315 netvsc_recv_rollup(struct hv_device *device_ctx)
1317 #if defined(INET) || defined(INET6)
1318 hn_softc_t *sc = device_get_softc(device_ctx->device);
1319 struct lro_ctrl *lro = &sc->hn_lro;
1320 struct lro_entry *queued;
1322 while ((queued = SLIST_FIRST(&lro->lro_active)) != NULL) {
1323 SLIST_REMOVE_HEAD(&lro->lro_active, next);
1324 tcp_lro_flush(lro, queued);
1330 * Rules for using sc->temp_unusable:
1331 * 1. sc->temp_unusable can only be read or written while holding NV_LOCK()
1332 * 2. code reading sc->temp_unusable under NV_LOCK(), and finding
1333 * sc->temp_unusable set, must release NV_LOCK() and exit
1334 * 3. to retain exclusive control of the interface,
1335 * sc->temp_unusable must be set by code before releasing NV_LOCK()
1336 * 4. only code setting sc->temp_unusable can clear sc->temp_unusable
1337 * 5. code setting sc->temp_unusable must eventually clear sc->temp_unusable
1341 * Standard ioctl entry point. Called when the user wants to configure
1345 hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1347 hn_softc_t *sc = ifp->if_softc;
1348 struct ifreq *ifr = (struct ifreq *)data;
1350 struct ifaddr *ifa = (struct ifaddr *)data;
1352 netvsc_device_info device_info;
1353 struct hv_device *hn_dev;
1354 int mask, error = 0;
1355 int retry_cnt = 500;
1361 if (ifa->ifa_addr->sa_family == AF_INET) {
1362 ifp->if_flags |= IFF_UP;
1363 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1365 arp_ifinit(ifp, ifa);
1368 error = ether_ioctl(ifp, cmd, data);
1371 hn_dev = vmbus_get_devctx(sc->hn_dev);
1373 /* Check MTU value change */
1374 if (ifp->if_mtu == ifr->ifr_mtu)
1377 if (ifr->ifr_mtu > NETVSC_MAX_CONFIGURABLE_MTU) {
1382 /* Obtain and record requested MTU */
1383 ifp->if_mtu = ifr->ifr_mtu;
1385 * Make sure that LRO high watermark is still valid,
1386 * after MTU change (the 2*MTU limit).
1388 if (!HN_LRO_HIWAT_ISVALID(sc, sc->hn_lro_hiwat))
1389 hn_set_lro_hiwat(sc, HN_LRO_HIWAT_MTULIM(ifp));
1393 if (!sc->temp_unusable) {
1394 sc->temp_unusable = TRUE;
1398 if (retry_cnt > 0) {
1402 } while (retry_cnt > 0);
1404 if (retry_cnt == 0) {
1409 /* We must remove and add back the device to cause the new
1410 * MTU to take effect. This includes tearing down, but not
1411 * deleting the channel, then bringing it back up.
1413 error = hv_rf_on_device_remove(hn_dev, HV_RF_NV_RETAIN_CHANNEL);
1416 sc->temp_unusable = FALSE;
1420 error = hv_rf_on_device_add(hn_dev, &device_info);
1423 sc->temp_unusable = FALSE;
1428 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
1429 if (sc->hn_tx_chimney_size > sc->hn_tx_chimney_max)
1430 sc->hn_tx_chimney_size = sc->hn_tx_chimney_max;
1431 hn_ifinit_locked(sc);
1434 sc->temp_unusable = FALSE;
1440 if (!sc->temp_unusable) {
1441 sc->temp_unusable = TRUE;
1445 if (retry_cnt > 0) {
1449 } while (retry_cnt > 0);
1451 if (retry_cnt == 0) {
1456 if (ifp->if_flags & IFF_UP) {
1458 * If only the state of the PROMISC flag changed,
1459 * then just use the 'set promisc mode' command
1460 * instead of reinitializing the entire NIC. Doing
1461 * a full re-init means reloading the firmware and
1462 * waiting for it to start up, which may take a
1466 /* Fixme: Promiscuous mode? */
1467 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1468 ifp->if_flags & IFF_PROMISC &&
1469 !(sc->hn_if_flags & IFF_PROMISC)) {
1470 /* do something here for Hyper-V */
1471 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1472 !(ifp->if_flags & IFF_PROMISC) &&
1473 sc->hn_if_flags & IFF_PROMISC) {
1474 /* do something here for Hyper-V */
1477 hn_ifinit_locked(sc);
1479 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1484 sc->temp_unusable = FALSE;
1486 sc->hn_if_flags = ifp->if_flags;
1492 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1493 if (mask & IFCAP_TXCSUM) {
1494 ifp->if_capenable ^= IFCAP_TXCSUM;
1495 if (ifp->if_capenable & IFCAP_TXCSUM)
1496 ifp->if_hwassist |= sc->hn_csum_assist;
1498 ifp->if_hwassist &= ~sc->hn_csum_assist;
1501 if (mask & IFCAP_RXCSUM)
1502 ifp->if_capenable ^= IFCAP_RXCSUM;
1504 if (mask & IFCAP_LRO)
1505 ifp->if_capenable ^= IFCAP_LRO;
1507 if (mask & IFCAP_TSO4) {
1508 ifp->if_capenable ^= IFCAP_TSO4;
1509 if (ifp->if_capenable & IFCAP_TSO4)
1510 ifp->if_hwassist |= CSUM_IP_TSO;
1512 ifp->if_hwassist &= ~CSUM_IP_TSO;
1515 if (mask & IFCAP_TSO6) {
1516 ifp->if_capenable ^= IFCAP_TSO6;
1517 if (ifp->if_capenable & IFCAP_TSO6)
1518 ifp->if_hwassist |= CSUM_IP6_TSO;
1520 ifp->if_hwassist &= ~CSUM_IP6_TSO;
1529 /* Fixme: Multicast mode? */
1530 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1532 netvsc_setmulti(sc);
1541 error = ifmedia_ioctl(ifp, ifr, &sc->hn_media, cmd);
1544 error = ether_ioctl(ifp, cmd, data);
1555 hn_stop(hn_softc_t *sc)
1559 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1564 printf(" Closing Device ...\n");
1566 atomic_clear_int(&ifp->if_drv_flags,
1567 (IFF_DRV_RUNNING | IFF_DRV_OACTIVE));
1568 if_link_state_change(ifp, LINK_STATE_DOWN);
1569 sc->hn_initdone = 0;
1571 ret = hv_rf_on_close(device_ctx);
1575 * FreeBSD transmit entry point
1578 hn_start(struct ifnet *ifp)
1583 if (NV_TRYLOCK(sc)) {
1586 sched = hn_start_locked(ifp, sc->hn_direct_tx_size);
1591 taskqueue_enqueue_fast(sc->hn_tx_taskq, &sc->hn_start_task);
1595 hn_start_txeof(struct ifnet *ifp)
1600 if (NV_TRYLOCK(sc)) {
1603 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1604 sched = hn_start_locked(ifp, sc->hn_direct_tx_size);
1607 taskqueue_enqueue_fast(sc->hn_tx_taskq,
1608 &sc->hn_start_task);
1612 * Release the OACTIVE earlier, with the hope, that
1613 * others could catch up. The task will clear the
1614 * flag again with the NV_LOCK to avoid possible
1617 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1618 taskqueue_enqueue_fast(sc->hn_tx_taskq, &sc->hn_txeof_task);
1626 hn_ifinit_locked(hn_softc_t *sc)
1629 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1634 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1638 hv_promisc_mode = 1;
1640 ret = hv_rf_on_open(device_ctx);
1644 sc->hn_initdone = 1;
1646 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1647 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_RUNNING);
1648 if_link_state_change(ifp, LINK_STATE_UP);
1655 hn_ifinit(void *xsc)
1657 hn_softc_t *sc = xsc;
1660 if (sc->temp_unusable) {
1664 sc->temp_unusable = TRUE;
1667 hn_ifinit_locked(sc);
1670 sc->temp_unusable = FALSE;
1679 hn_watchdog(struct ifnet *ifp)
1684 printf("hn%d: watchdog timeout -- resetting\n", sc->hn_unit);
1685 hn_ifinit(sc); /*???*/
1692 hn_lro_hiwat_sysctl(SYSCTL_HANDLER_ARGS)
1694 struct hn_softc *sc = arg1;
1697 hiwat = sc->hn_lro_hiwat;
1698 error = sysctl_handle_int(oidp, &hiwat, 0, req);
1699 if (error || req->newptr == NULL)
1702 if (!HN_LRO_HIWAT_ISVALID(sc, hiwat))
1705 if (sc->hn_lro_hiwat != hiwat)
1706 hn_set_lro_hiwat(sc, hiwat);
1709 #endif /* HN_LRO_HIWAT */
1712 hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS)
1714 struct hn_softc *sc = arg1;
1719 if (sc->hn_trust_hcsum & hcsum)
1722 error = sysctl_handle_int(oidp, &on, 0, req);
1723 if (error || req->newptr == NULL)
1728 sc->hn_trust_hcsum |= hcsum;
1730 sc->hn_trust_hcsum &= ~hcsum;
1736 hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS)
1738 struct hn_softc *sc = arg1;
1739 int chimney_size, error;
1741 chimney_size = sc->hn_tx_chimney_size;
1742 error = sysctl_handle_int(oidp, &chimney_size, 0, req);
1743 if (error || req->newptr == NULL)
1746 if (chimney_size > sc->hn_tx_chimney_max || chimney_size <= 0)
1749 if (sc->hn_tx_chimney_size != chimney_size)
1750 sc->hn_tx_chimney_size = chimney_size;
1755 hn_check_iplen(const struct mbuf *m, int hoff)
1757 const struct ip *ip;
1758 int len, iphlen, iplen;
1759 const struct tcphdr *th;
1760 int thoff; /* TCP data offset */
1762 len = hoff + sizeof(struct ip);
1764 /* The packet must be at least the size of an IP header. */
1765 if (m->m_pkthdr.len < len)
1766 return IPPROTO_DONE;
1768 /* The fixed IP header must reside completely in the first mbuf. */
1770 return IPPROTO_DONE;
1772 ip = mtodo(m, hoff);
1774 /* Bound check the packet's stated IP header length. */
1775 iphlen = ip->ip_hl << 2;
1776 if (iphlen < sizeof(struct ip)) /* minimum header length */
1777 return IPPROTO_DONE;
1779 /* The full IP header must reside completely in the one mbuf. */
1780 if (m->m_len < hoff + iphlen)
1781 return IPPROTO_DONE;
1783 iplen = ntohs(ip->ip_len);
1786 * Check that the amount of data in the buffers is as
1787 * at least much as the IP header would have us expect.
1789 if (m->m_pkthdr.len < hoff + iplen)
1790 return IPPROTO_DONE;
1793 * Ignore IP fragments.
1795 if (ntohs(ip->ip_off) & (IP_OFFMASK | IP_MF))
1796 return IPPROTO_DONE;
1799 * The TCP/IP or UDP/IP header must be entirely contained within
1800 * the first fragment of a packet.
1804 if (iplen < iphlen + sizeof(struct tcphdr))
1805 return IPPROTO_DONE;
1806 if (m->m_len < hoff + iphlen + sizeof(struct tcphdr))
1807 return IPPROTO_DONE;
1808 th = (const struct tcphdr *)((const uint8_t *)ip + iphlen);
1809 thoff = th->th_off << 2;
1810 if (thoff < sizeof(struct tcphdr) || thoff + iphlen > iplen)
1811 return IPPROTO_DONE;
1812 if (m->m_len < hoff + iphlen + thoff)
1813 return IPPROTO_DONE;
1816 if (iplen < iphlen + sizeof(struct udphdr))
1817 return IPPROTO_DONE;
1818 if (m->m_len < hoff + iphlen + sizeof(struct udphdr))
1819 return IPPROTO_DONE;
1823 return IPPROTO_DONE;
1830 hn_dma_map_paddr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
1832 bus_addr_t *paddr = arg;
1837 KASSERT(nseg == 1, ("too many segments %d!", nseg));
1838 *paddr = segs->ds_addr;
1842 hn_create_tx_ring(struct hn_softc *sc)
1844 bus_dma_tag_t parent_dtag;
1847 sc->hn_txdesc_cnt = HN_TX_DESC_CNT;
1848 sc->hn_txdesc = malloc(sizeof(struct hn_txdesc) * sc->hn_txdesc_cnt,
1849 M_NETVSC, M_WAITOK | M_ZERO);
1850 SLIST_INIT(&sc->hn_txlist);
1851 mtx_init(&sc->hn_txlist_spin, "hn txlist", NULL, MTX_SPIN);
1853 parent_dtag = bus_get_dma_tag(sc->hn_dev);
1855 /* DMA tag for RNDIS messages. */
1856 error = bus_dma_tag_create(parent_dtag, /* parent */
1857 HN_RNDIS_MSG_ALIGN, /* alignment */
1858 HN_RNDIS_MSG_BOUNDARY, /* boundary */
1859 BUS_SPACE_MAXADDR, /* lowaddr */
1860 BUS_SPACE_MAXADDR, /* highaddr */
1861 NULL, NULL, /* filter, filterarg */
1862 HN_RNDIS_MSG_LEN, /* maxsize */
1864 HN_RNDIS_MSG_LEN, /* maxsegsize */
1866 NULL, /* lockfunc */
1867 NULL, /* lockfuncarg */
1868 &sc->hn_tx_rndis_dtag);
1870 device_printf(sc->hn_dev, "failed to create rndis dmatag\n");
1874 /* DMA tag for data. */
1875 error = bus_dma_tag_create(parent_dtag, /* parent */
1877 HN_TX_DATA_BOUNDARY, /* boundary */
1878 BUS_SPACE_MAXADDR, /* lowaddr */
1879 BUS_SPACE_MAXADDR, /* highaddr */
1880 NULL, NULL, /* filter, filterarg */
1881 HN_TX_DATA_MAXSIZE, /* maxsize */
1882 HN_TX_DATA_SEGCNT_MAX, /* nsegments */
1883 HN_TX_DATA_SEGSIZE, /* maxsegsize */
1885 NULL, /* lockfunc */
1886 NULL, /* lockfuncarg */
1887 &sc->hn_tx_data_dtag);
1889 device_printf(sc->hn_dev, "failed to create data dmatag\n");
1893 for (i = 0; i < sc->hn_txdesc_cnt; ++i) {
1894 struct hn_txdesc *txd = &sc->hn_txdesc[i];
1899 * Allocate and load RNDIS messages.
1901 error = bus_dmamem_alloc(sc->hn_tx_rndis_dtag,
1902 (void **)&txd->rndis_msg,
1903 BUS_DMA_WAITOK | BUS_DMA_COHERENT,
1904 &txd->rndis_msg_dmap);
1906 device_printf(sc->hn_dev,
1907 "failed to allocate rndis_msg, %d\n", i);
1911 error = bus_dmamap_load(sc->hn_tx_rndis_dtag,
1912 txd->rndis_msg_dmap,
1913 txd->rndis_msg, HN_RNDIS_MSG_LEN,
1914 hn_dma_map_paddr, &txd->rndis_msg_paddr,
1917 device_printf(sc->hn_dev,
1918 "failed to load rndis_msg, %d\n", i);
1919 bus_dmamem_free(sc->hn_tx_rndis_dtag,
1920 txd->rndis_msg, txd->rndis_msg_dmap);
1924 /* DMA map for TX data. */
1925 error = bus_dmamap_create(sc->hn_tx_data_dtag, 0,
1928 device_printf(sc->hn_dev,
1929 "failed to allocate tx data dmamap\n");
1930 bus_dmamap_unload(sc->hn_tx_rndis_dtag,
1931 txd->rndis_msg_dmap);
1932 bus_dmamem_free(sc->hn_tx_rndis_dtag,
1933 txd->rndis_msg, txd->rndis_msg_dmap);
1937 /* All set, put it to list */
1938 txd->flags |= HN_TXD_FLAG_ONLIST;
1939 SLIST_INSERT_HEAD(&sc->hn_txlist, txd, link);
1941 sc->hn_txdesc_avail = sc->hn_txdesc_cnt;
1947 hn_destroy_tx_ring(struct hn_softc *sc)
1949 struct hn_txdesc *txd;
1951 while ((txd = SLIST_FIRST(&sc->hn_txlist)) != NULL) {
1952 KASSERT(txd->m == NULL, ("still has mbuf installed"));
1953 KASSERT((txd->flags & HN_TXD_FLAG_DMAMAP) == 0,
1954 ("still dma mapped"));
1955 SLIST_REMOVE_HEAD(&sc->hn_txlist, link);
1957 bus_dmamap_unload(sc->hn_tx_rndis_dtag,
1958 txd->rndis_msg_dmap);
1959 bus_dmamem_free(sc->hn_tx_rndis_dtag,
1960 txd->rndis_msg, txd->rndis_msg_dmap);
1962 bus_dmamap_destroy(sc->hn_tx_data_dtag, txd->data_dmap);
1965 if (sc->hn_tx_data_dtag != NULL)
1966 bus_dma_tag_destroy(sc->hn_tx_data_dtag);
1967 if (sc->hn_tx_rndis_dtag != NULL)
1968 bus_dma_tag_destroy(sc->hn_tx_rndis_dtag);
1969 free(sc->hn_txdesc, M_NETVSC);
1970 mtx_destroy(&sc->hn_txlist_spin);
1974 hn_start_taskfunc(void *xsc, int pending __unused)
1976 struct hn_softc *sc = xsc;
1979 hn_start_locked(sc->hn_ifp, 0);
1984 hn_txeof_taskfunc(void *xsc, int pending __unused)
1986 struct hn_softc *sc = xsc;
1987 struct ifnet *ifp = sc->hn_ifp;
1990 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1991 hn_start_locked(ifp, 0);
1995 static device_method_t netvsc_methods[] = {
1996 /* Device interface */
1997 DEVMETHOD(device_probe, netvsc_probe),
1998 DEVMETHOD(device_attach, netvsc_attach),
1999 DEVMETHOD(device_detach, netvsc_detach),
2000 DEVMETHOD(device_shutdown, netvsc_shutdown),
2005 static driver_t netvsc_driver = {
2011 static devclass_t netvsc_devclass;
2013 DRIVER_MODULE(hn, vmbus, netvsc_driver, netvsc_devclass, 0, 0);
2014 MODULE_VERSION(hn, 1);
2015 MODULE_DEPEND(hn, vmbus, 1, 1, 1);