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 #define HN_LRO_LENLIM_DEF (25 * ETHERMTU)
180 /* YYY 2*MTU is a bit rough, but should be good enough. */
181 #define HN_LRO_LENLIM_MIN(ifp) (2 * (ifp)->if_mtu)
183 #define HN_LRO_ACKCNT_DEF 1
186 * Be aware that this sleepable mutex will exhibit WITNESS errors when
187 * certain TCP and ARP code paths are taken. This appears to be a
188 * well-known condition, as all other drivers checked use a sleeping
189 * mutex to protect their transmit paths.
190 * Also Be aware that mutexes do not play well with semaphores, and there
191 * is a conflicting semaphore in a certain channel code path.
193 #define NV_LOCK_INIT(_sc, _name) \
194 mtx_init(&(_sc)->hn_lock, _name, MTX_NETWORK_LOCK, MTX_DEF)
195 #define NV_LOCK(_sc) mtx_lock(&(_sc)->hn_lock)
196 #define NV_TRYLOCK(_sc) mtx_trylock(&(_sc)->hn_lock)
197 #define NV_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->hn_lock, MA_OWNED)
198 #define NV_UNLOCK(_sc) mtx_unlock(&(_sc)->hn_lock)
199 #define NV_LOCK_DESTROY(_sc) mtx_destroy(&(_sc)->hn_lock)
206 int hv_promisc_mode = 0; /* normal mode by default */
208 /* Trust tcp segements verification on host side. */
209 static int hn_trust_hosttcp = 1;
210 TUNABLE_INT("dev.hn.trust_hosttcp", &hn_trust_hosttcp);
212 /* Trust udp datagrams verification on host side. */
213 static int hn_trust_hostudp = 1;
214 TUNABLE_INT("dev.hn.trust_hostudp", &hn_trust_hostudp);
216 /* Trust ip packets verification on host side. */
217 static int hn_trust_hostip = 1;
218 TUNABLE_INT("dev.hn.trust_hostip", &hn_trust_hostip);
220 #if __FreeBSD_version >= 1100045
221 /* Limit TSO burst size */
222 static int hn_tso_maxlen = 0;
223 TUNABLE_INT("dev.hn.tso_maxlen", &hn_tso_maxlen);
226 /* Limit chimney send size */
227 static int hn_tx_chimney_size = 0;
228 TUNABLE_INT("dev.hn.tx_chimney_size", &hn_tx_chimney_size);
230 /* Limit the size of packet for direct transmission */
231 static int hn_direct_tx_size = HN_DIRECT_TX_SIZE_DEF;
232 TUNABLE_INT("dev.hn.direct_tx_size", &hn_direct_tx_size);
234 #if defined(INET) || defined(INET6)
235 #if __FreeBSD_version >= 1100095
236 static int hn_lro_entry_count = HN_LROENT_CNT_DEF;
237 TUNABLE_INT("dev.hn.lro_entry_count", &hn_lro_entry_count);
241 static int hn_share_tx_taskq = 0;
242 TUNABLE_INT("hw.hn.share_tx_taskq", &hn_share_tx_taskq);
244 static struct taskqueue *hn_tx_taskq;
247 * Forward declarations
249 static void hn_stop(hn_softc_t *sc);
250 static void hn_ifinit_locked(hn_softc_t *sc);
251 static void hn_ifinit(void *xsc);
252 static int hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data);
253 static int hn_start_locked(struct ifnet *ifp, int len);
254 static void hn_start(struct ifnet *ifp);
255 static void hn_start_txeof(struct ifnet *ifp);
256 static int hn_ifmedia_upd(struct ifnet *ifp);
257 static void hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr);
258 #if __FreeBSD_version >= 1100099
259 static int hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS);
260 static int hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS);
262 static int hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS);
263 static int hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS);
264 static int hn_check_iplen(const struct mbuf *, int);
265 static int hn_create_tx_ring(struct hn_softc *sc);
266 static void hn_destroy_tx_ring(struct hn_softc *sc);
267 static void hn_start_taskfunc(void *xsc, int pending);
268 static void hn_txeof_taskfunc(void *xsc, int pending);
269 static int hn_encap(struct hn_softc *, struct hn_txdesc *, struct mbuf **);
272 hn_ifmedia_upd(struct ifnet *ifp __unused)
279 hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
281 struct hn_softc *sc = ifp->if_softc;
283 ifmr->ifm_status = IFM_AVALID;
284 ifmr->ifm_active = IFM_ETHER;
286 if (!sc->hn_carrier) {
287 ifmr->ifm_active |= IFM_NONE;
290 ifmr->ifm_status |= IFM_ACTIVE;
291 ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
294 /* {F8615163-DF3E-46c5-913F-F2D2F965ED0E} */
295 static const hv_guid g_net_vsc_device_type = {
296 .data = {0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
297 0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E}
301 * Standard probe entry point.
305 netvsc_probe(device_t dev)
309 p = vmbus_get_type(dev);
310 if (!memcmp(p, &g_net_vsc_device_type.data, sizeof(hv_guid))) {
311 device_set_desc(dev, "Synthetic Network Interface");
313 printf("Netvsc probe... DONE \n");
315 return (BUS_PROBE_DEFAULT);
322 * Standard attach entry point.
324 * Called when the driver is loaded. It allocates needed resources,
325 * and initializes the "hardware" and software.
328 netvsc_attach(device_t dev)
330 struct hv_device *device_ctx = vmbus_get_devctx(dev);
331 netvsc_device_info device_info;
333 int unit = device_get_unit(dev);
334 struct ifnet *ifp = NULL;
335 struct sysctl_oid_list *child;
336 struct sysctl_ctx_list *ctx;
338 #if __FreeBSD_version >= 1100045
341 #if defined(INET) || defined(INET6)
342 #if __FreeBSD_version >= 1100095
347 sc = device_get_softc(dev);
352 bzero(sc, sizeof(hn_softc_t));
355 sc->hn_direct_tx_size = hn_direct_tx_size;
356 if (hn_trust_hosttcp)
357 sc->hn_trust_hcsum |= HN_TRUST_HCSUM_TCP;
358 if (hn_trust_hostudp)
359 sc->hn_trust_hcsum |= HN_TRUST_HCSUM_UDP;
361 sc->hn_trust_hcsum |= HN_TRUST_HCSUM_IP;
363 if (hn_tx_taskq == NULL) {
364 sc->hn_tx_taskq = taskqueue_create_fast("hn_tx", M_WAITOK,
365 taskqueue_thread_enqueue, &sc->hn_tx_taskq);
366 taskqueue_start_threads(&sc->hn_tx_taskq, 1, PI_NET, "%s tx",
367 device_get_nameunit(dev));
369 sc->hn_tx_taskq = hn_tx_taskq;
371 TASK_INIT(&sc->hn_start_task, 0, hn_start_taskfunc, sc);
372 TASK_INIT(&sc->hn_txeof_task, 0, hn_txeof_taskfunc, sc);
374 error = hn_create_tx_ring(sc);
378 NV_LOCK_INIT(sc, "NetVSCLock");
380 sc->hn_dev_obj = device_ctx;
382 ifp = sc->hn_ifp = sc->arpcom.ac_ifp = if_alloc(IFT_ETHER);
385 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
386 ifp->if_dunit = unit;
387 ifp->if_dname = NETVSC_DEVNAME;
389 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
390 ifp->if_ioctl = hn_ioctl;
391 ifp->if_start = hn_start;
392 ifp->if_init = hn_ifinit;
393 /* needed by hv_rf_on_device_add() code */
394 ifp->if_mtu = ETHERMTU;
395 IFQ_SET_MAXLEN(&ifp->if_snd, 512);
396 ifp->if_snd.ifq_drv_maxlen = 511;
397 IFQ_SET_READY(&ifp->if_snd);
399 ifmedia_init(&sc->hn_media, 0, hn_ifmedia_upd, hn_ifmedia_sts);
400 ifmedia_add(&sc->hn_media, IFM_ETHER | IFM_AUTO, 0, NULL);
401 ifmedia_set(&sc->hn_media, IFM_ETHER | IFM_AUTO);
402 /* XXX ifmedia_set really should do this for us */
403 sc->hn_media.ifm_media = sc->hn_media.ifm_cur->ifm_media;
406 * Tell upper layers that we support full VLAN capability.
408 ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
409 ifp->if_capabilities |=
410 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
413 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
416 if (hv_vmbus_protocal_version >= HV_VMBUS_VERSION_WIN8_1)
417 sc->hn_csum_assist = HN_CSUM_ASSIST;
419 sc->hn_csum_assist = HN_CSUM_ASSIST_WIN8;
420 ifp->if_hwassist = sc->hn_csum_assist | CSUM_TSO;
422 error = hv_rf_on_device_add(device_ctx, &device_info);
426 if (device_info.link_state == 0) {
430 #if defined(INET) || defined(INET6)
431 #if __FreeBSD_version >= 1100095
432 lroent_cnt = hn_lro_entry_count;
433 if (lroent_cnt < TCP_LRO_ENTRIES)
434 lroent_cnt = TCP_LRO_ENTRIES;
435 tcp_lro_init_args(&sc->hn_lro, ifp, lroent_cnt, 0);
436 device_printf(dev, "LRO: entry count %d\n", lroent_cnt);
438 tcp_lro_init(&sc->hn_lro);
439 /* Driver private LRO settings */
440 sc->hn_lro.ifp = ifp;
442 #if __FreeBSD_version >= 1100099
443 sc->hn_lro.lro_length_lim = HN_LRO_LENLIM_DEF;
444 sc->hn_lro.lro_ackcnt_lim = HN_LRO_ACKCNT_DEF;
446 #endif /* INET || INET6 */
448 #if __FreeBSD_version >= 1100045
449 tso_maxlen = hn_tso_maxlen;
450 if (tso_maxlen <= 0 || tso_maxlen > IP_MAXPACKET)
451 tso_maxlen = IP_MAXPACKET;
453 ifp->if_hw_tsomaxsegcount = HN_TX_DATA_SEGCNT_MAX;
454 ifp->if_hw_tsomaxsegsize = PAGE_SIZE;
455 ifp->if_hw_tsomax = tso_maxlen -
456 (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
459 ether_ifattach(ifp, device_info.mac_addr);
461 #if __FreeBSD_version >= 1100045
462 if_printf(ifp, "TSO: %u/%u/%u\n", ifp->if_hw_tsomax,
463 ifp->if_hw_tsomaxsegcount, ifp->if_hw_tsomaxsegsize);
466 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
467 sc->hn_tx_chimney_size = sc->hn_tx_chimney_max;
468 if (hn_tx_chimney_size > 0 &&
469 hn_tx_chimney_size < sc->hn_tx_chimney_max)
470 sc->hn_tx_chimney_size = hn_tx_chimney_size;
473 * Always schedule transmission instead of trying
474 * to do direct transmission. This one gives the
475 * best performance so far.
479 ctx = device_get_sysctl_ctx(dev);
480 child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
482 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "lro_queued",
483 CTLFLAG_RW, &sc->hn_lro.lro_queued, 0, "LRO queued");
484 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "lro_flushed",
485 CTLFLAG_RW, &sc->hn_lro.lro_flushed, 0, "LRO flushed");
486 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "lro_tried",
487 CTLFLAG_RW, &sc->hn_lro_tried, "# of LRO tries");
488 #if __FreeBSD_version >= 1100099
489 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_length_lim",
490 CTLTYPE_UINT | CTLFLAG_RW, sc, 0, hn_lro_lenlim_sysctl, "IU",
491 "Max # of data bytes to be aggregated by LRO");
492 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_ackcnt_lim",
493 CTLTYPE_INT | CTLFLAG_RW, sc, 0, hn_lro_ackcnt_sysctl, "I",
494 "Max # of ACKs to be aggregated by LRO");
496 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hosttcp",
497 CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_TCP,
498 hn_trust_hcsum_sysctl, "I",
499 "Trust tcp segement verification on host side, "
500 "when csum info is missing");
501 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostudp",
502 CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_UDP,
503 hn_trust_hcsum_sysctl, "I",
504 "Trust udp datagram verification on host side, "
505 "when csum info is missing");
506 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostip",
507 CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_IP,
508 hn_trust_hcsum_sysctl, "I",
509 "Trust ip packet verification on host side, "
510 "when csum info is missing");
511 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "csum_ip",
512 CTLFLAG_RW, &sc->hn_csum_ip, "RXCSUM IP");
513 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "csum_tcp",
514 CTLFLAG_RW, &sc->hn_csum_tcp, "RXCSUM TCP");
515 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "csum_udp",
516 CTLFLAG_RW, &sc->hn_csum_udp, "RXCSUM UDP");
517 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "csum_trusted",
518 CTLFLAG_RW, &sc->hn_csum_trusted,
519 "# of packets that we trust host's csum verification");
520 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "small_pkts",
521 CTLFLAG_RW, &sc->hn_small_pkts, "# of small packets received");
522 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "no_txdescs",
523 CTLFLAG_RW, &sc->hn_no_txdescs, "# of times short of TX descs");
524 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "send_failed",
525 CTLFLAG_RW, &sc->hn_send_failed, "# of hyper-v sending failure");
526 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "txdma_failed",
527 CTLFLAG_RW, &sc->hn_txdma_failed, "# of TX DMA failure");
528 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_collapsed",
529 CTLFLAG_RW, &sc->hn_tx_collapsed, "# of TX mbuf collapsed");
530 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_chimney",
531 CTLFLAG_RW, &sc->hn_tx_chimney, "# of chimney send");
532 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_cnt",
533 CTLFLAG_RD, &sc->hn_txdesc_cnt, 0, "# of total TX descs");
534 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_avail",
535 CTLFLAG_RD, &sc->hn_txdesc_avail, 0, "# of available TX descs");
536 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_chimney_max",
537 CTLFLAG_RD, &sc->hn_tx_chimney_max, 0,
538 "Chimney send packet size upper boundary");
539 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_size",
540 CTLTYPE_INT | CTLFLAG_RW, sc, 0, hn_tx_chimney_size_sysctl,
541 "I", "Chimney send packet size limit");
542 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "direct_tx_size",
543 CTLFLAG_RW, &sc->hn_direct_tx_size, 0,
544 "Size of the packet for direct transmission");
545 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "sched_tx",
546 CTLFLAG_RW, &sc->hn_sched_tx, 0,
547 "Always schedule transmission "
548 "instead of doing direct transmission");
551 struct sysctl_ctx_list *dc_ctx;
552 struct sysctl_oid_list *dc_child;
556 * Add sysctl nodes for devclass
558 dc = device_get_devclass(dev);
559 dc_ctx = devclass_get_sysctl_ctx(dc);
560 dc_child = SYSCTL_CHILDREN(devclass_get_sysctl_tree(dc));
562 SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "trust_hosttcp",
563 CTLFLAG_RD, &hn_trust_hosttcp, 0,
564 "Trust tcp segement verification on host side, "
565 "when csum info is missing (global setting)");
566 SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "trust_hostudp",
567 CTLFLAG_RD, &hn_trust_hostudp, 0,
568 "Trust udp datagram verification on host side, "
569 "when csum info is missing (global setting)");
570 SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "trust_hostip",
571 CTLFLAG_RD, &hn_trust_hostip, 0,
572 "Trust ip packet verification on host side, "
573 "when csum info is missing (global setting)");
574 SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "tx_chimney_size",
575 CTLFLAG_RD, &hn_tx_chimney_size, 0,
576 "Chimney send packet size limit");
577 #if __FreeBSD_version >= 1100045
578 SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "tso_maxlen",
579 CTLFLAG_RD, &hn_tso_maxlen, 0, "TSO burst limit");
581 SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "direct_tx_size",
582 CTLFLAG_RD, &hn_direct_tx_size, 0,
583 "Size of the packet for direct transmission");
584 #if defined(INET) || defined(INET6)
585 #if __FreeBSD_version >= 1100095
586 SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "lro_entry_count",
587 CTLFLAG_RD, &hn_lro_entry_count, 0, "LRO entry count");
594 hn_destroy_tx_ring(sc);
601 * Standard detach entry point
604 netvsc_detach(device_t dev)
606 struct hn_softc *sc = device_get_softc(dev);
607 struct hv_device *hv_device = vmbus_get_devctx(dev);
610 printf("netvsc_detach\n");
613 * XXXKYS: Need to clean up all our
614 * driver state; this is the driver
619 * XXXKYS: Need to stop outgoing traffic and unregister
623 hv_rf_on_device_remove(hv_device, HV_RF_NV_DESTROY_CHANNEL);
625 taskqueue_drain(sc->hn_tx_taskq, &sc->hn_start_task);
626 taskqueue_drain(sc->hn_tx_taskq, &sc->hn_txeof_task);
627 if (sc->hn_tx_taskq != hn_tx_taskq)
628 taskqueue_free(sc->hn_tx_taskq);
630 ifmedia_removeall(&sc->hn_media);
631 #if defined(INET) || defined(INET6)
632 tcp_lro_free(&sc->hn_lro);
634 hn_destroy_tx_ring(sc);
640 * Standard shutdown entry point
643 netvsc_shutdown(device_t dev)
649 hn_txdesc_dmamap_load(struct hn_softc *sc, struct hn_txdesc *txd,
650 struct mbuf **m_head, bus_dma_segment_t *segs, int *nsegs)
652 struct mbuf *m = *m_head;
655 error = bus_dmamap_load_mbuf_sg(sc->hn_tx_data_dtag, txd->data_dmap,
656 m, segs, nsegs, BUS_DMA_NOWAIT);
657 if (error == EFBIG) {
660 m_new = m_collapse(m, M_NOWAIT, HN_TX_DATA_SEGCNT_MAX);
665 sc->hn_tx_collapsed++;
667 error = bus_dmamap_load_mbuf_sg(sc->hn_tx_data_dtag,
668 txd->data_dmap, m, segs, nsegs, BUS_DMA_NOWAIT);
671 bus_dmamap_sync(sc->hn_tx_data_dtag, txd->data_dmap,
672 BUS_DMASYNC_PREWRITE);
673 txd->flags |= HN_TXD_FLAG_DMAMAP;
679 hn_txdesc_dmamap_unload(struct hn_softc *sc, struct hn_txdesc *txd)
682 if (txd->flags & HN_TXD_FLAG_DMAMAP) {
683 bus_dmamap_sync(sc->hn_tx_data_dtag,
684 txd->data_dmap, BUS_DMASYNC_POSTWRITE);
685 bus_dmamap_unload(sc->hn_tx_data_dtag,
687 txd->flags &= ~HN_TXD_FLAG_DMAMAP;
692 hn_txdesc_put(struct hn_softc *sc, struct hn_txdesc *txd)
695 KASSERT((txd->flags & HN_TXD_FLAG_ONLIST) == 0,
696 ("put an onlist txd %#x", txd->flags));
698 KASSERT(txd->refs > 0, ("invalid txd refs %d", txd->refs));
699 if (atomic_fetchadd_int(&txd->refs, -1) != 1)
702 hn_txdesc_dmamap_unload(sc, txd);
703 if (txd->m != NULL) {
708 txd->flags |= HN_TXD_FLAG_ONLIST;
710 mtx_lock_spin(&sc->hn_txlist_spin);
711 KASSERT(sc->hn_txdesc_avail >= 0 &&
712 sc->hn_txdesc_avail < sc->hn_txdesc_cnt,
713 ("txdesc_put: invalid txd avail %d", sc->hn_txdesc_avail));
714 sc->hn_txdesc_avail++;
715 SLIST_INSERT_HEAD(&sc->hn_txlist, txd, link);
716 mtx_unlock_spin(&sc->hn_txlist_spin);
721 static __inline struct hn_txdesc *
722 hn_txdesc_get(struct hn_softc *sc)
724 struct hn_txdesc *txd;
726 mtx_lock_spin(&sc->hn_txlist_spin);
727 txd = SLIST_FIRST(&sc->hn_txlist);
729 KASSERT(sc->hn_txdesc_avail > 0,
730 ("txdesc_get: invalid txd avail %d", sc->hn_txdesc_avail));
731 sc->hn_txdesc_avail--;
732 SLIST_REMOVE_HEAD(&sc->hn_txlist, link);
734 mtx_unlock_spin(&sc->hn_txlist_spin);
737 KASSERT(txd->m == NULL && txd->refs == 0 &&
738 (txd->flags & HN_TXD_FLAG_ONLIST), ("invalid txd"));
739 txd->flags &= ~HN_TXD_FLAG_ONLIST;
746 hn_txdesc_hold(struct hn_txdesc *txd)
749 /* 0->1 transition will never work */
750 KASSERT(txd->refs > 0, ("invalid refs %d", txd->refs));
751 atomic_add_int(&txd->refs, 1);
755 * Send completion processing
757 * Note: It looks like offset 0 of buf is reserved to hold the softc
758 * pointer. The sc pointer is not currently needed in this function, and
759 * it is not presently populated by the TX function.
762 netvsc_xmit_completion(void *context)
764 netvsc_packet *packet = context;
765 struct hn_txdesc *txd;
768 txd = (struct hn_txdesc *)(uintptr_t)
769 packet->compl.send.send_completion_tid;
773 hn_txdesc_put(sc, txd);
777 netvsc_channel_rollup(struct hv_device *device_ctx)
779 struct hn_softc *sc = device_get_softc(device_ctx->device);
780 #if defined(INET) || defined(INET6)
781 struct lro_ctrl *lro = &sc->hn_lro;
782 struct lro_entry *queued;
784 while ((queued = SLIST_FIRST(&lro->lro_active)) != NULL) {
785 SLIST_REMOVE_HEAD(&lro->lro_active, next);
786 tcp_lro_flush(lro, queued);
794 hn_start_txeof(sc->hn_ifp);
799 * If this function fails, then both txd and m_head0 will be freed.
802 hn_encap(struct hn_softc *sc, struct hn_txdesc *txd, struct mbuf **m_head0)
804 bus_dma_segment_t segs[HN_TX_DATA_SEGCNT_MAX];
806 struct mbuf *m_head = *m_head0;
807 netvsc_packet *packet;
808 rndis_msg *rndis_mesg;
809 rndis_packet *rndis_pkt;
810 rndis_per_packet_info *rppi;
811 uint32_t rndis_msg_size;
813 packet = &txd->netvsc_pkt;
814 packet->is_data_pkt = TRUE;
815 packet->tot_data_buf_len = m_head->m_pkthdr.len;
818 * extension points to the area reserved for the
819 * rndis_filter_packet, which is placed just after
820 * the netvsc_packet (and rppi struct, if present;
821 * length is updated later).
823 rndis_mesg = txd->rndis_msg;
824 /* XXX not necessary */
825 memset(rndis_mesg, 0, HN_RNDIS_MSG_LEN);
826 rndis_mesg->ndis_msg_type = REMOTE_NDIS_PACKET_MSG;
828 rndis_pkt = &rndis_mesg->msg.packet;
829 rndis_pkt->data_offset = sizeof(rndis_packet);
830 rndis_pkt->data_length = packet->tot_data_buf_len;
831 rndis_pkt->per_pkt_info_offset = sizeof(rndis_packet);
833 rndis_msg_size = RNDIS_MESSAGE_SIZE(rndis_packet);
835 if (m_head->m_flags & M_VLANTAG) {
836 ndis_8021q_info *rppi_vlan_info;
838 rndis_msg_size += RNDIS_VLAN_PPI_SIZE;
839 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_VLAN_PPI_SIZE,
842 rppi_vlan_info = (ndis_8021q_info *)((uint8_t *)rppi +
843 rppi->per_packet_info_offset);
844 rppi_vlan_info->u1.s1.vlan_id =
845 m_head->m_pkthdr.ether_vtag & 0xfff;
848 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
849 rndis_tcp_tso_info *tso_info;
850 struct ether_vlan_header *eh;
854 * XXX need m_pullup and use mtodo
856 eh = mtod(m_head, struct ether_vlan_header*);
857 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN))
858 ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
860 ether_len = ETHER_HDR_LEN;
862 rndis_msg_size += RNDIS_TSO_PPI_SIZE;
863 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_TSO_PPI_SIZE,
864 tcp_large_send_info);
866 tso_info = (rndis_tcp_tso_info *)((uint8_t *)rppi +
867 rppi->per_packet_info_offset);
868 tso_info->lso_v2_xmit.type =
869 RNDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
872 if (m_head->m_pkthdr.csum_flags & CSUM_IP_TSO) {
874 (struct ip *)(m_head->m_data + ether_len);
875 unsigned long iph_len = ip->ip_hl << 2;
877 (struct tcphdr *)((caddr_t)ip + iph_len);
879 tso_info->lso_v2_xmit.ip_version =
880 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
884 th->th_sum = in_pseudo(ip->ip_src.s_addr,
885 ip->ip_dst.s_addr, htons(IPPROTO_TCP));
888 #if defined(INET6) && defined(INET)
893 struct ip6_hdr *ip6 = (struct ip6_hdr *)
894 (m_head->m_data + ether_len);
895 struct tcphdr *th = (struct tcphdr *)(ip6 + 1);
897 tso_info->lso_v2_xmit.ip_version =
898 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
900 th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
903 tso_info->lso_v2_xmit.tcp_header_offset = 0;
904 tso_info->lso_v2_xmit.mss = m_head->m_pkthdr.tso_segsz;
905 } else if (m_head->m_pkthdr.csum_flags & sc->hn_csum_assist) {
906 rndis_tcp_ip_csum_info *csum_info;
908 rndis_msg_size += RNDIS_CSUM_PPI_SIZE;
909 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_CSUM_PPI_SIZE,
911 csum_info = (rndis_tcp_ip_csum_info *)((uint8_t *)rppi +
912 rppi->per_packet_info_offset);
914 csum_info->xmit.is_ipv4 = 1;
915 if (m_head->m_pkthdr.csum_flags & CSUM_IP)
916 csum_info->xmit.ip_header_csum = 1;
918 if (m_head->m_pkthdr.csum_flags & CSUM_TCP) {
919 csum_info->xmit.tcp_csum = 1;
920 csum_info->xmit.tcp_header_offset = 0;
921 } else if (m_head->m_pkthdr.csum_flags & CSUM_UDP) {
922 csum_info->xmit.udp_csum = 1;
926 rndis_mesg->msg_len = packet->tot_data_buf_len + rndis_msg_size;
927 packet->tot_data_buf_len = rndis_mesg->msg_len;
930 * Chimney send, if the packet could fit into one chimney buffer.
932 if (packet->tot_data_buf_len < sc->hn_tx_chimney_size) {
933 netvsc_dev *net_dev = sc->net_dev;
934 uint32_t send_buf_section_idx;
936 send_buf_section_idx =
937 hv_nv_get_next_send_section(net_dev);
938 if (send_buf_section_idx !=
939 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX) {
940 uint8_t *dest = ((uint8_t *)net_dev->send_buf +
941 (send_buf_section_idx *
942 net_dev->send_section_size));
944 memcpy(dest, rndis_mesg, rndis_msg_size);
945 dest += rndis_msg_size;
946 m_copydata(m_head, 0, m_head->m_pkthdr.len, dest);
948 packet->send_buf_section_idx = send_buf_section_idx;
949 packet->send_buf_section_size =
950 packet->tot_data_buf_len;
951 packet->page_buf_count = 0;
957 error = hn_txdesc_dmamap_load(sc, txd, &m_head, segs, &nsegs);
962 * This mbuf is not linked w/ the txd yet, so free it now.
967 freed = hn_txdesc_put(sc, txd);
969 ("fail to free txd upon txdma error"));
971 sc->hn_txdma_failed++;
972 if_inc_counter(sc->hn_ifp, IFCOUNTER_OERRORS, 1);
977 packet->page_buf_count = nsegs + HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
979 /* send packet with page buffer */
980 packet->page_buffers[0].pfn = atop(txd->rndis_msg_paddr);
981 packet->page_buffers[0].offset = txd->rndis_msg_paddr & PAGE_MASK;
982 packet->page_buffers[0].length = rndis_msg_size;
985 * Fill the page buffers with mbuf info starting at index
986 * HV_RF_NUM_TX_RESERVED_PAGE_BUFS.
988 for (i = 0; i < nsegs; ++i) {
989 hv_vmbus_page_buffer *pb = &packet->page_buffers[
990 i + HV_RF_NUM_TX_RESERVED_PAGE_BUFS];
992 pb->pfn = atop(segs[i].ds_addr);
993 pb->offset = segs[i].ds_addr & PAGE_MASK;
994 pb->length = segs[i].ds_len;
997 packet->send_buf_section_idx =
998 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX;
999 packet->send_buf_section_size = 0;
1003 /* Set the completion routine */
1004 packet->compl.send.on_send_completion = netvsc_xmit_completion;
1005 packet->compl.send.send_completion_context = packet;
1006 packet->compl.send.send_completion_tid = (uint64_t)(uintptr_t)txd;
1012 * Start a transmit of one or more packets
1015 hn_start_locked(struct ifnet *ifp, int len)
1017 struct hn_softc *sc = ifp->if_softc;
1018 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1020 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1024 while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
1025 int error, send_failed = 0;
1026 struct hn_txdesc *txd;
1027 struct mbuf *m_head;
1029 IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
1033 if (len > 0 && m_head->m_pkthdr.len > len) {
1035 * This sending could be time consuming; let callers
1036 * dispatch this packet sending (and sending of any
1037 * following up packets) to tx taskqueue.
1039 IF_PREPEND(&ifp->if_snd, m_head);
1043 txd = hn_txdesc_get(sc);
1045 sc->hn_no_txdescs++;
1046 IF_PREPEND(&ifp->if_snd, m_head);
1047 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1051 error = hn_encap(sc, txd, &m_head);
1053 /* Both txd and m_head are freed */
1058 * Make sure that txd is not freed before ETHER_BPF_MTAP.
1060 hn_txdesc_hold(txd);
1061 error = hv_nv_on_send(device_ctx, &txd->netvsc_pkt);
1063 ETHER_BPF_MTAP(ifp, m_head);
1064 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1066 hn_txdesc_put(sc, txd);
1068 if (__predict_false(error)) {
1072 * This should "really rarely" happen.
1074 * XXX Too many RX to be acked or too many sideband
1075 * commands to run? Ask netvsc_channel_rollup()
1076 * to kick start later.
1080 sc->hn_send_failed++;
1083 * Try sending again after set hn_txeof;
1084 * in case that we missed the last
1085 * netvsc_channel_rollup().
1089 if_printf(ifp, "send failed\n");
1092 * This mbuf will be prepended, don't free it
1093 * in hn_txdesc_put(); only unload it from the
1094 * DMA map in hn_txdesc_put(), if it was loaded.
1097 freed = hn_txdesc_put(sc, txd);
1099 ("fail to free txd upon send error"));
1101 sc->hn_send_failed++;
1102 IF_PREPEND(&ifp->if_snd, m_head);
1103 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1111 * Link up/down notification
1114 netvsc_linkstatus_callback(struct hv_device *device_obj, uint32_t status)
1116 hn_softc_t *sc = device_get_softc(device_obj->device);
1130 * Append the specified data to the indicated mbuf chain,
1131 * Extend the mbuf chain if the new data does not fit in
1134 * This is a minor rewrite of m_append() from sys/kern/uipc_mbuf.c.
1135 * There should be an equivalent in the kernel mbuf code,
1136 * but there does not appear to be one yet.
1138 * Differs from m_append() in that additional mbufs are
1139 * allocated with cluster size MJUMPAGESIZE, and filled
1142 * Return 1 if able to complete the job; otherwise 0.
1145 hv_m_append(struct mbuf *m0, int len, c_caddr_t cp)
1148 int remainder, space;
1150 for (m = m0; m->m_next != NULL; m = m->m_next)
1153 space = M_TRAILINGSPACE(m);
1156 * Copy into available space.
1158 if (space > remainder)
1160 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1165 while (remainder > 0) {
1167 * Allocate a new mbuf; could check space
1168 * and allocate a cluster instead.
1170 n = m_getjcl(M_DONTWAIT, m->m_type, 0, MJUMPAGESIZE);
1173 n->m_len = min(MJUMPAGESIZE, remainder);
1174 bcopy(cp, mtod(n, caddr_t), n->m_len);
1176 remainder -= n->m_len;
1180 if (m0->m_flags & M_PKTHDR)
1181 m0->m_pkthdr.len += len - remainder;
1183 return (remainder == 0);
1188 * Called when we receive a data packet from the "wire" on the
1191 * Note: This is no longer used as a callback
1194 netvsc_recv(struct hv_device *device_ctx, netvsc_packet *packet,
1195 rndis_tcp_ip_csum_info *csum_info)
1197 hn_softc_t *sc = (hn_softc_t *)device_get_softc(device_ctx->device);
1200 device_t dev = device_ctx->device;
1201 int size, do_lro = 0, do_csum = 1;
1204 return (0); /* TODO: KYS how can this be! */
1209 ifp = sc->arpcom.ac_ifp;
1211 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1216 * Bail out if packet contains more data than configured MTU.
1218 if (packet->tot_data_buf_len > (ifp->if_mtu + ETHER_HDR_LEN)) {
1220 } else if (packet->tot_data_buf_len <= MHLEN) {
1221 m_new = m_gethdr(M_NOWAIT, MT_DATA);
1224 memcpy(mtod(m_new, void *), packet->data,
1225 packet->tot_data_buf_len);
1226 m_new->m_pkthdr.len = m_new->m_len = packet->tot_data_buf_len;
1227 sc->hn_small_pkts++;
1230 * Get an mbuf with a cluster. For packets 2K or less,
1231 * get a standard 2K cluster. For anything larger, get a
1232 * 4K cluster. Any buffers larger than 4K can cause problems
1233 * if looped around to the Hyper-V TX channel, so avoid them.
1236 if (packet->tot_data_buf_len > MCLBYTES) {
1238 size = MJUMPAGESIZE;
1241 m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, size);
1242 if (m_new == NULL) {
1243 device_printf(dev, "alloc mbuf failed.\n");
1247 hv_m_append(m_new, packet->tot_data_buf_len, packet->data);
1249 m_new->m_pkthdr.rcvif = ifp;
1251 if (__predict_false((ifp->if_capenable & IFCAP_RXCSUM) == 0))
1254 /* receive side checksum offload */
1255 if (csum_info != NULL) {
1256 /* IP csum offload */
1257 if (csum_info->receive.ip_csum_succeeded && do_csum) {
1258 m_new->m_pkthdr.csum_flags |=
1259 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1263 /* TCP/UDP csum offload */
1264 if ((csum_info->receive.tcp_csum_succeeded ||
1265 csum_info->receive.udp_csum_succeeded) && do_csum) {
1266 m_new->m_pkthdr.csum_flags |=
1267 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1268 m_new->m_pkthdr.csum_data = 0xffff;
1269 if (csum_info->receive.tcp_csum_succeeded)
1275 if (csum_info->receive.ip_csum_succeeded &&
1276 csum_info->receive.tcp_csum_succeeded)
1279 const struct ether_header *eh;
1284 if (m_new->m_len < hoff)
1286 eh = mtod(m_new, struct ether_header *);
1287 etype = ntohs(eh->ether_type);
1288 if (etype == ETHERTYPE_VLAN) {
1289 const struct ether_vlan_header *evl;
1291 hoff = sizeof(*evl);
1292 if (m_new->m_len < hoff)
1294 evl = mtod(m_new, struct ether_vlan_header *);
1295 etype = ntohs(evl->evl_proto);
1298 if (etype == ETHERTYPE_IP) {
1301 pr = hn_check_iplen(m_new, hoff);
1302 if (pr == IPPROTO_TCP) {
1304 (sc->hn_trust_hcsum & HN_TRUST_HCSUM_TCP)) {
1305 sc->hn_csum_trusted++;
1306 m_new->m_pkthdr.csum_flags |=
1307 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1308 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1309 m_new->m_pkthdr.csum_data = 0xffff;
1311 /* Rely on SW csum verification though... */
1313 } else if (pr == IPPROTO_UDP) {
1315 (sc->hn_trust_hcsum & HN_TRUST_HCSUM_UDP)) {
1316 sc->hn_csum_trusted++;
1317 m_new->m_pkthdr.csum_flags |=
1318 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1319 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1320 m_new->m_pkthdr.csum_data = 0xffff;
1322 } else if (pr != IPPROTO_DONE && do_csum &&
1323 (sc->hn_trust_hcsum & HN_TRUST_HCSUM_IP)) {
1324 sc->hn_csum_trusted++;
1325 m_new->m_pkthdr.csum_flags |=
1326 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1331 if ((packet->vlan_tci != 0) &&
1332 (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0) {
1333 m_new->m_pkthdr.ether_vtag = packet->vlan_tci;
1334 m_new->m_flags |= M_VLANTAG;
1338 * Note: Moved RX completion back to hv_nv_on_receive() so all
1339 * messages (not just data messages) will trigger a response.
1344 if ((ifp->if_capenable & IFCAP_LRO) && do_lro) {
1345 #if defined(INET) || defined(INET6)
1346 struct lro_ctrl *lro = &sc->hn_lro;
1350 if (tcp_lro_rx(lro, m_new, 0) == 0) {
1358 /* We're not holding the lock here, so don't release it */
1359 (*ifp->if_input)(ifp, m_new);
1365 netvsc_recv_rollup(struct hv_device *device_ctx __unused)
1370 * Rules for using sc->temp_unusable:
1371 * 1. sc->temp_unusable can only be read or written while holding NV_LOCK()
1372 * 2. code reading sc->temp_unusable under NV_LOCK(), and finding
1373 * sc->temp_unusable set, must release NV_LOCK() and exit
1374 * 3. to retain exclusive control of the interface,
1375 * sc->temp_unusable must be set by code before releasing NV_LOCK()
1376 * 4. only code setting sc->temp_unusable can clear sc->temp_unusable
1377 * 5. code setting sc->temp_unusable must eventually clear sc->temp_unusable
1381 * Standard ioctl entry point. Called when the user wants to configure
1385 hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1387 hn_softc_t *sc = ifp->if_softc;
1388 struct ifreq *ifr = (struct ifreq *)data;
1390 struct ifaddr *ifa = (struct ifaddr *)data;
1392 netvsc_device_info device_info;
1393 struct hv_device *hn_dev;
1394 int mask, error = 0;
1395 int retry_cnt = 500;
1401 if (ifa->ifa_addr->sa_family == AF_INET) {
1402 ifp->if_flags |= IFF_UP;
1403 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1405 arp_ifinit(ifp, ifa);
1408 error = ether_ioctl(ifp, cmd, data);
1411 hn_dev = vmbus_get_devctx(sc->hn_dev);
1413 /* Check MTU value change */
1414 if (ifp->if_mtu == ifr->ifr_mtu)
1417 if (ifr->ifr_mtu > NETVSC_MAX_CONFIGURABLE_MTU) {
1422 /* Obtain and record requested MTU */
1423 ifp->if_mtu = ifr->ifr_mtu;
1425 #if __FreeBSD_version >= 1100099
1427 * Make sure that LRO aggregation length limit is still
1428 * valid, after the MTU change.
1430 if (sc->hn_lro.lro_length_lim < HN_LRO_LENLIM_MIN(ifp))
1431 sc->hn_lro.lro_length_lim = HN_LRO_LENLIM_MIN(ifp);
1436 if (!sc->temp_unusable) {
1437 sc->temp_unusable = TRUE;
1441 if (retry_cnt > 0) {
1445 } while (retry_cnt > 0);
1447 if (retry_cnt == 0) {
1452 /* We must remove and add back the device to cause the new
1453 * MTU to take effect. This includes tearing down, but not
1454 * deleting the channel, then bringing it back up.
1456 error = hv_rf_on_device_remove(hn_dev, HV_RF_NV_RETAIN_CHANNEL);
1459 sc->temp_unusable = FALSE;
1463 error = hv_rf_on_device_add(hn_dev, &device_info);
1466 sc->temp_unusable = FALSE;
1471 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
1472 if (sc->hn_tx_chimney_size > sc->hn_tx_chimney_max)
1473 sc->hn_tx_chimney_size = sc->hn_tx_chimney_max;
1474 hn_ifinit_locked(sc);
1477 sc->temp_unusable = FALSE;
1483 if (!sc->temp_unusable) {
1484 sc->temp_unusable = TRUE;
1488 if (retry_cnt > 0) {
1492 } while (retry_cnt > 0);
1494 if (retry_cnt == 0) {
1499 if (ifp->if_flags & IFF_UP) {
1501 * If only the state of the PROMISC flag changed,
1502 * then just use the 'set promisc mode' command
1503 * instead of reinitializing the entire NIC. Doing
1504 * a full re-init means reloading the firmware and
1505 * waiting for it to start up, which may take a
1509 /* Fixme: Promiscuous mode? */
1510 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1511 ifp->if_flags & IFF_PROMISC &&
1512 !(sc->hn_if_flags & IFF_PROMISC)) {
1513 /* do something here for Hyper-V */
1514 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1515 !(ifp->if_flags & IFF_PROMISC) &&
1516 sc->hn_if_flags & IFF_PROMISC) {
1517 /* do something here for Hyper-V */
1520 hn_ifinit_locked(sc);
1522 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1527 sc->temp_unusable = FALSE;
1529 sc->hn_if_flags = ifp->if_flags;
1535 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1536 if (mask & IFCAP_TXCSUM) {
1537 ifp->if_capenable ^= IFCAP_TXCSUM;
1538 if (ifp->if_capenable & IFCAP_TXCSUM)
1539 ifp->if_hwassist |= sc->hn_csum_assist;
1541 ifp->if_hwassist &= ~sc->hn_csum_assist;
1544 if (mask & IFCAP_RXCSUM)
1545 ifp->if_capenable ^= IFCAP_RXCSUM;
1547 if (mask & IFCAP_LRO)
1548 ifp->if_capenable ^= IFCAP_LRO;
1550 if (mask & IFCAP_TSO4) {
1551 ifp->if_capenable ^= IFCAP_TSO4;
1552 if (ifp->if_capenable & IFCAP_TSO4)
1553 ifp->if_hwassist |= CSUM_IP_TSO;
1555 ifp->if_hwassist &= ~CSUM_IP_TSO;
1558 if (mask & IFCAP_TSO6) {
1559 ifp->if_capenable ^= IFCAP_TSO6;
1560 if (ifp->if_capenable & IFCAP_TSO6)
1561 ifp->if_hwassist |= CSUM_IP6_TSO;
1563 ifp->if_hwassist &= ~CSUM_IP6_TSO;
1572 /* Fixme: Multicast mode? */
1573 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1575 netvsc_setmulti(sc);
1584 error = ifmedia_ioctl(ifp, ifr, &sc->hn_media, cmd);
1587 error = ether_ioctl(ifp, cmd, data);
1598 hn_stop(hn_softc_t *sc)
1602 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1607 printf(" Closing Device ...\n");
1609 atomic_clear_int(&ifp->if_drv_flags,
1610 (IFF_DRV_RUNNING | IFF_DRV_OACTIVE));
1611 if_link_state_change(ifp, LINK_STATE_DOWN);
1612 sc->hn_initdone = 0;
1614 ret = hv_rf_on_close(device_ctx);
1618 * FreeBSD transmit entry point
1621 hn_start(struct ifnet *ifp)
1623 struct hn_softc *sc = ifp->if_softc;
1625 if (sc->hn_sched_tx)
1628 if (NV_TRYLOCK(sc)) {
1631 sched = hn_start_locked(ifp, sc->hn_direct_tx_size);
1637 taskqueue_enqueue_fast(sc->hn_tx_taskq, &sc->hn_start_task);
1641 hn_start_txeof(struct ifnet *ifp)
1643 struct hn_softc *sc = ifp->if_softc;
1645 if (sc->hn_sched_tx)
1648 if (NV_TRYLOCK(sc)) {
1651 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1652 sched = hn_start_locked(ifp, sc->hn_direct_tx_size);
1655 taskqueue_enqueue_fast(sc->hn_tx_taskq,
1656 &sc->hn_start_task);
1661 * Release the OACTIVE earlier, with the hope, that
1662 * others could catch up. The task will clear the
1663 * flag again with the NV_LOCK to avoid possible
1666 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1667 taskqueue_enqueue_fast(sc->hn_tx_taskq, &sc->hn_txeof_task);
1675 hn_ifinit_locked(hn_softc_t *sc)
1678 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1683 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1687 hv_promisc_mode = 1;
1689 ret = hv_rf_on_open(device_ctx);
1693 sc->hn_initdone = 1;
1695 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1696 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_RUNNING);
1697 if_link_state_change(ifp, LINK_STATE_UP);
1704 hn_ifinit(void *xsc)
1706 hn_softc_t *sc = xsc;
1709 if (sc->temp_unusable) {
1713 sc->temp_unusable = TRUE;
1716 hn_ifinit_locked(sc);
1719 sc->temp_unusable = FALSE;
1728 hn_watchdog(struct ifnet *ifp)
1733 printf("hn%d: watchdog timeout -- resetting\n", sc->hn_unit);
1734 hn_ifinit(sc); /*???*/
1739 #if __FreeBSD_version >= 1100099
1742 hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS)
1744 struct hn_softc *sc = arg1;
1745 unsigned int lenlim;
1748 lenlim = sc->hn_lro.lro_length_lim;
1749 error = sysctl_handle_int(oidp, &lenlim, 0, req);
1750 if (error || req->newptr == NULL)
1753 if (lenlim < HN_LRO_LENLIM_MIN(sc->hn_ifp) ||
1754 lenlim > TCP_LRO_LENGTH_MAX)
1757 sc->hn_lro.lro_length_lim = lenlim;
1762 hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS)
1764 struct hn_softc *sc = arg1;
1768 * lro_ackcnt_lim is append count limit,
1769 * +1 to turn it into aggregation limit.
1771 ackcnt = sc->hn_lro.lro_ackcnt_lim + 1;
1772 error = sysctl_handle_int(oidp, &ackcnt, 0, req);
1773 if (error || req->newptr == NULL)
1776 if (ackcnt < 2 || ackcnt > (TCP_LRO_ACKCNT_MAX + 1))
1780 * Convert aggregation limit back to append
1783 sc->hn_lro.lro_ackcnt_lim = ackcnt - 1;
1790 hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS)
1792 struct hn_softc *sc = arg1;
1797 if (sc->hn_trust_hcsum & hcsum)
1800 error = sysctl_handle_int(oidp, &on, 0, req);
1801 if (error || req->newptr == NULL)
1806 sc->hn_trust_hcsum |= hcsum;
1808 sc->hn_trust_hcsum &= ~hcsum;
1814 hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS)
1816 struct hn_softc *sc = arg1;
1817 int chimney_size, error;
1819 chimney_size = sc->hn_tx_chimney_size;
1820 error = sysctl_handle_int(oidp, &chimney_size, 0, req);
1821 if (error || req->newptr == NULL)
1824 if (chimney_size > sc->hn_tx_chimney_max || chimney_size <= 0)
1827 if (sc->hn_tx_chimney_size != chimney_size)
1828 sc->hn_tx_chimney_size = chimney_size;
1833 hn_check_iplen(const struct mbuf *m, int hoff)
1835 const struct ip *ip;
1836 int len, iphlen, iplen;
1837 const struct tcphdr *th;
1838 int thoff; /* TCP data offset */
1840 len = hoff + sizeof(struct ip);
1842 /* The packet must be at least the size of an IP header. */
1843 if (m->m_pkthdr.len < len)
1844 return IPPROTO_DONE;
1846 /* The fixed IP header must reside completely in the first mbuf. */
1848 return IPPROTO_DONE;
1850 ip = mtodo(m, hoff);
1852 /* Bound check the packet's stated IP header length. */
1853 iphlen = ip->ip_hl << 2;
1854 if (iphlen < sizeof(struct ip)) /* minimum header length */
1855 return IPPROTO_DONE;
1857 /* The full IP header must reside completely in the one mbuf. */
1858 if (m->m_len < hoff + iphlen)
1859 return IPPROTO_DONE;
1861 iplen = ntohs(ip->ip_len);
1864 * Check that the amount of data in the buffers is as
1865 * at least much as the IP header would have us expect.
1867 if (m->m_pkthdr.len < hoff + iplen)
1868 return IPPROTO_DONE;
1871 * Ignore IP fragments.
1873 if (ntohs(ip->ip_off) & (IP_OFFMASK | IP_MF))
1874 return IPPROTO_DONE;
1877 * The TCP/IP or UDP/IP header must be entirely contained within
1878 * the first fragment of a packet.
1882 if (iplen < iphlen + sizeof(struct tcphdr))
1883 return IPPROTO_DONE;
1884 if (m->m_len < hoff + iphlen + sizeof(struct tcphdr))
1885 return IPPROTO_DONE;
1886 th = (const struct tcphdr *)((const uint8_t *)ip + iphlen);
1887 thoff = th->th_off << 2;
1888 if (thoff < sizeof(struct tcphdr) || thoff + iphlen > iplen)
1889 return IPPROTO_DONE;
1890 if (m->m_len < hoff + iphlen + thoff)
1891 return IPPROTO_DONE;
1894 if (iplen < iphlen + sizeof(struct udphdr))
1895 return IPPROTO_DONE;
1896 if (m->m_len < hoff + iphlen + sizeof(struct udphdr))
1897 return IPPROTO_DONE;
1901 return IPPROTO_DONE;
1908 hn_dma_map_paddr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
1910 bus_addr_t *paddr = arg;
1915 KASSERT(nseg == 1, ("too many segments %d!", nseg));
1916 *paddr = segs->ds_addr;
1920 hn_create_tx_ring(struct hn_softc *sc)
1922 bus_dma_tag_t parent_dtag;
1925 sc->hn_txdesc_cnt = HN_TX_DESC_CNT;
1926 sc->hn_txdesc = malloc(sizeof(struct hn_txdesc) * sc->hn_txdesc_cnt,
1927 M_NETVSC, M_WAITOK | M_ZERO);
1928 SLIST_INIT(&sc->hn_txlist);
1929 mtx_init(&sc->hn_txlist_spin, "hn txlist", NULL, MTX_SPIN);
1931 parent_dtag = bus_get_dma_tag(sc->hn_dev);
1933 /* DMA tag for RNDIS messages. */
1934 error = bus_dma_tag_create(parent_dtag, /* parent */
1935 HN_RNDIS_MSG_ALIGN, /* alignment */
1936 HN_RNDIS_MSG_BOUNDARY, /* boundary */
1937 BUS_SPACE_MAXADDR, /* lowaddr */
1938 BUS_SPACE_MAXADDR, /* highaddr */
1939 NULL, NULL, /* filter, filterarg */
1940 HN_RNDIS_MSG_LEN, /* maxsize */
1942 HN_RNDIS_MSG_LEN, /* maxsegsize */
1944 NULL, /* lockfunc */
1945 NULL, /* lockfuncarg */
1946 &sc->hn_tx_rndis_dtag);
1948 device_printf(sc->hn_dev, "failed to create rndis dmatag\n");
1952 /* DMA tag for data. */
1953 error = bus_dma_tag_create(parent_dtag, /* parent */
1955 HN_TX_DATA_BOUNDARY, /* boundary */
1956 BUS_SPACE_MAXADDR, /* lowaddr */
1957 BUS_SPACE_MAXADDR, /* highaddr */
1958 NULL, NULL, /* filter, filterarg */
1959 HN_TX_DATA_MAXSIZE, /* maxsize */
1960 HN_TX_DATA_SEGCNT_MAX, /* nsegments */
1961 HN_TX_DATA_SEGSIZE, /* maxsegsize */
1963 NULL, /* lockfunc */
1964 NULL, /* lockfuncarg */
1965 &sc->hn_tx_data_dtag);
1967 device_printf(sc->hn_dev, "failed to create data dmatag\n");
1971 for (i = 0; i < sc->hn_txdesc_cnt; ++i) {
1972 struct hn_txdesc *txd = &sc->hn_txdesc[i];
1977 * Allocate and load RNDIS messages.
1979 error = bus_dmamem_alloc(sc->hn_tx_rndis_dtag,
1980 (void **)&txd->rndis_msg,
1981 BUS_DMA_WAITOK | BUS_DMA_COHERENT,
1982 &txd->rndis_msg_dmap);
1984 device_printf(sc->hn_dev,
1985 "failed to allocate rndis_msg, %d\n", i);
1989 error = bus_dmamap_load(sc->hn_tx_rndis_dtag,
1990 txd->rndis_msg_dmap,
1991 txd->rndis_msg, HN_RNDIS_MSG_LEN,
1992 hn_dma_map_paddr, &txd->rndis_msg_paddr,
1995 device_printf(sc->hn_dev,
1996 "failed to load rndis_msg, %d\n", i);
1997 bus_dmamem_free(sc->hn_tx_rndis_dtag,
1998 txd->rndis_msg, txd->rndis_msg_dmap);
2002 /* DMA map for TX data. */
2003 error = bus_dmamap_create(sc->hn_tx_data_dtag, 0,
2006 device_printf(sc->hn_dev,
2007 "failed to allocate tx data dmamap\n");
2008 bus_dmamap_unload(sc->hn_tx_rndis_dtag,
2009 txd->rndis_msg_dmap);
2010 bus_dmamem_free(sc->hn_tx_rndis_dtag,
2011 txd->rndis_msg, txd->rndis_msg_dmap);
2015 /* All set, put it to list */
2016 txd->flags |= HN_TXD_FLAG_ONLIST;
2017 SLIST_INSERT_HEAD(&sc->hn_txlist, txd, link);
2019 sc->hn_txdesc_avail = sc->hn_txdesc_cnt;
2025 hn_destroy_tx_ring(struct hn_softc *sc)
2027 struct hn_txdesc *txd;
2029 while ((txd = SLIST_FIRST(&sc->hn_txlist)) != NULL) {
2030 KASSERT(txd->m == NULL, ("still has mbuf installed"));
2031 KASSERT((txd->flags & HN_TXD_FLAG_DMAMAP) == 0,
2032 ("still dma mapped"));
2033 SLIST_REMOVE_HEAD(&sc->hn_txlist, link);
2035 bus_dmamap_unload(sc->hn_tx_rndis_dtag,
2036 txd->rndis_msg_dmap);
2037 bus_dmamem_free(sc->hn_tx_rndis_dtag,
2038 txd->rndis_msg, txd->rndis_msg_dmap);
2040 bus_dmamap_destroy(sc->hn_tx_data_dtag, txd->data_dmap);
2043 if (sc->hn_tx_data_dtag != NULL)
2044 bus_dma_tag_destroy(sc->hn_tx_data_dtag);
2045 if (sc->hn_tx_rndis_dtag != NULL)
2046 bus_dma_tag_destroy(sc->hn_tx_rndis_dtag);
2047 free(sc->hn_txdesc, M_NETVSC);
2048 mtx_destroy(&sc->hn_txlist_spin);
2052 hn_start_taskfunc(void *xsc, int pending __unused)
2054 struct hn_softc *sc = xsc;
2057 hn_start_locked(sc->hn_ifp, 0);
2062 hn_txeof_taskfunc(void *xsc, int pending __unused)
2064 struct hn_softc *sc = xsc;
2065 struct ifnet *ifp = sc->hn_ifp;
2068 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
2069 hn_start_locked(ifp, 0);
2074 hn_tx_taskq_create(void *arg __unused)
2076 if (!hn_share_tx_taskq)
2079 hn_tx_taskq = taskqueue_create_fast("hn_tx", M_WAITOK,
2080 taskqueue_thread_enqueue, &hn_tx_taskq);
2081 taskqueue_start_threads(&hn_tx_taskq, 1, PI_NET, "hn tx");
2083 SYSINIT(hn_txtq_create, SI_SUB_DRIVERS, SI_ORDER_FIRST,
2084 hn_tx_taskq_create, NULL);
2087 hn_tx_taskq_destroy(void *arg __unused)
2089 if (hn_tx_taskq != NULL)
2090 taskqueue_free(hn_tx_taskq);
2092 SYSUNINIT(hn_txtq_destroy, SI_SUB_DRIVERS, SI_ORDER_FIRST,
2093 hn_tx_taskq_destroy, NULL);
2095 static device_method_t netvsc_methods[] = {
2096 /* Device interface */
2097 DEVMETHOD(device_probe, netvsc_probe),
2098 DEVMETHOD(device_attach, netvsc_attach),
2099 DEVMETHOD(device_detach, netvsc_detach),
2100 DEVMETHOD(device_shutdown, netvsc_shutdown),
2105 static driver_t netvsc_driver = {
2111 static devclass_t netvsc_devclass;
2113 DRIVER_MODULE(hn, vmbus, netvsc_driver, netvsc_devclass, 0, 0);
2114 MODULE_VERSION(hn, 1);
2115 MODULE_DEPEND(hn, vmbus, 1, 1, 1);