2 * Copyright (c) 2010-2012 Citrix Inc.
3 * Copyright (c) 2009-2012,2016 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>
70 #include <sys/queue.h>
73 #include <sys/sysctl.h>
74 #include <sys/buf_ring.h>
77 #include <net/if_arp.h>
78 #include <net/ethernet.h>
79 #include <net/if_dl.h>
80 #include <net/if_media.h>
84 #include <net/if_types.h>
85 #include <net/if_vlan_var.h>
88 #include <netinet/in_systm.h>
89 #include <netinet/in.h>
90 #include <netinet/ip.h>
91 #include <netinet/if_ether.h>
92 #include <netinet/tcp.h>
93 #include <netinet/udp.h>
94 #include <netinet/ip6.h>
97 #include <vm/vm_param.h>
98 #include <vm/vm_kern.h>
101 #include <machine/bus.h>
102 #include <machine/resource.h>
103 #include <machine/frame.h>
104 #include <machine/vmparam.h>
107 #include <sys/rman.h>
108 #include <sys/mutex.h>
109 #include <sys/errno.h>
110 #include <sys/types.h>
111 #include <machine/atomic.h>
113 #include <machine/intr_machdep.h>
115 #include <machine/in_cksum.h>
117 #include <dev/hyperv/include/hyperv.h>
118 #include "hv_net_vsc.h"
119 #include "hv_rndis.h"
120 #include "hv_rndis_filter.h"
122 #define hv_chan_rxr hv_chan_priv1
123 #define hv_chan_txr hv_chan_priv2
125 /* Short for Hyper-V network interface */
126 #define NETVSC_DEVNAME "hn"
129 * It looks like offset 0 of buf is reserved to hold the softc pointer.
130 * The sc pointer evidently not needed, and is not presently populated.
131 * The packet offset is where the netvsc_packet starts in the buffer.
133 #define HV_NV_SC_PTR_OFFSET_IN_BUF 0
134 #define HV_NV_PACKET_OFFSET_IN_BUF 16
136 /* YYY should get it from the underlying channel */
137 #define HN_TX_DESC_CNT 512
139 #define HN_LROENT_CNT_DEF 128
141 #define HN_RING_CNT_DEF_MAX 8
143 #define HN_RNDIS_MSG_LEN \
144 (sizeof(rndis_msg) + \
145 RNDIS_HASH_PPI_SIZE + \
146 RNDIS_VLAN_PPI_SIZE + \
147 RNDIS_TSO_PPI_SIZE + \
149 #define HN_RNDIS_MSG_BOUNDARY PAGE_SIZE
150 #define HN_RNDIS_MSG_ALIGN CACHE_LINE_SIZE
152 #define HN_TX_DATA_BOUNDARY PAGE_SIZE
153 #define HN_TX_DATA_MAXSIZE IP_MAXPACKET
154 #define HN_TX_DATA_SEGSIZE PAGE_SIZE
155 #define HN_TX_DATA_SEGCNT_MAX \
156 (NETVSC_PACKET_MAXPAGE - HV_RF_NUM_TX_RESERVED_PAGE_BUFS)
158 #define HN_DIRECT_TX_SIZE_DEF 128
160 #define HN_EARLY_TXEOF_THRESH 8
163 #ifndef HN_USE_TXDESC_BUFRING
164 SLIST_ENTRY(hn_txdesc) link;
167 struct hn_tx_ring *txr;
169 uint32_t flags; /* HN_TXD_FLAG_ */
170 netvsc_packet netvsc_pkt; /* XXX to be removed */
172 bus_dmamap_t data_dmap;
174 bus_addr_t rndis_msg_paddr;
175 rndis_msg *rndis_msg;
176 bus_dmamap_t rndis_msg_dmap;
179 #define HN_TXD_FLAG_ONLIST 0x1
180 #define HN_TXD_FLAG_DMAMAP 0x2
183 * Only enable UDP checksum offloading when it is on 2012R2 or
184 * later. UDP checksum offloading doesn't work on earlier
187 #define HN_CSUM_ASSIST_WIN8 (CSUM_IP | CSUM_TCP)
188 #define HN_CSUM_ASSIST (CSUM_IP | CSUM_UDP | CSUM_TCP)
190 #define HN_LRO_LENLIM_MULTIRX_DEF (12 * ETHERMTU)
191 #define HN_LRO_LENLIM_DEF (25 * ETHERMTU)
192 /* YYY 2*MTU is a bit rough, but should be good enough. */
193 #define HN_LRO_LENLIM_MIN(ifp) (2 * (ifp)->if_mtu)
195 #define HN_LRO_ACKCNT_DEF 1
198 * Be aware that this sleepable mutex will exhibit WITNESS errors when
199 * certain TCP and ARP code paths are taken. This appears to be a
200 * well-known condition, as all other drivers checked use a sleeping
201 * mutex to protect their transmit paths.
202 * Also Be aware that mutexes do not play well with semaphores, and there
203 * is a conflicting semaphore in a certain channel code path.
205 #define NV_LOCK_INIT(_sc, _name) \
206 mtx_init(&(_sc)->hn_lock, _name, MTX_NETWORK_LOCK, MTX_DEF)
207 #define NV_LOCK(_sc) mtx_lock(&(_sc)->hn_lock)
208 #define NV_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->hn_lock, MA_OWNED)
209 #define NV_UNLOCK(_sc) mtx_unlock(&(_sc)->hn_lock)
210 #define NV_LOCK_DESTROY(_sc) mtx_destroy(&(_sc)->hn_lock)
217 int hv_promisc_mode = 0; /* normal mode by default */
219 SYSCTL_NODE(_hw, OID_AUTO, hn, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
220 "Hyper-V network interface");
222 /* Trust tcp segements verification on host side. */
223 static int hn_trust_hosttcp = 1;
224 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hosttcp, CTLFLAG_RDTUN,
225 &hn_trust_hosttcp, 0,
226 "Trust tcp segement verification on host side, "
227 "when csum info is missing (global setting)");
229 /* Trust udp datagrams verification on host side. */
230 static int hn_trust_hostudp = 1;
231 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostudp, CTLFLAG_RDTUN,
232 &hn_trust_hostudp, 0,
233 "Trust udp datagram verification on host side, "
234 "when csum info is missing (global setting)");
236 /* Trust ip packets verification on host side. */
237 static int hn_trust_hostip = 1;
238 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostip, CTLFLAG_RDTUN,
240 "Trust ip packet verification on host side, "
241 "when csum info is missing (global setting)");
243 /* Limit TSO burst size */
244 static int hn_tso_maxlen = 0;
245 SYSCTL_INT(_hw_hn, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN,
246 &hn_tso_maxlen, 0, "TSO burst limit");
248 /* Limit chimney send size */
249 static int hn_tx_chimney_size = 0;
250 SYSCTL_INT(_hw_hn, OID_AUTO, tx_chimney_size, CTLFLAG_RDTUN,
251 &hn_tx_chimney_size, 0, "Chimney send packet size limit");
253 /* Limit the size of packet for direct transmission */
254 static int hn_direct_tx_size = HN_DIRECT_TX_SIZE_DEF;
255 SYSCTL_INT(_hw_hn, OID_AUTO, direct_tx_size, CTLFLAG_RDTUN,
256 &hn_direct_tx_size, 0, "Size of the packet for direct transmission");
258 #if defined(INET) || defined(INET6)
259 #if __FreeBSD_version >= 1100095
260 static int hn_lro_entry_count = HN_LROENT_CNT_DEF;
261 SYSCTL_INT(_hw_hn, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN,
262 &hn_lro_entry_count, 0, "LRO entry count");
266 static int hn_share_tx_taskq = 0;
267 SYSCTL_INT(_hw_hn, OID_AUTO, share_tx_taskq, CTLFLAG_RDTUN,
268 &hn_share_tx_taskq, 0, "Enable shared TX taskqueue");
270 static struct taskqueue *hn_tx_taskq;
272 #ifndef HN_USE_TXDESC_BUFRING
273 static int hn_use_txdesc_bufring = 0;
275 static int hn_use_txdesc_bufring = 1;
277 SYSCTL_INT(_hw_hn, OID_AUTO, use_txdesc_bufring, CTLFLAG_RD,
278 &hn_use_txdesc_bufring, 0, "Use buf_ring for TX descriptors");
280 static int hn_bind_tx_taskq = -1;
281 SYSCTL_INT(_hw_hn, OID_AUTO, bind_tx_taskq, CTLFLAG_RDTUN,
282 &hn_bind_tx_taskq, 0, "Bind TX taskqueue to the specified cpu");
284 static int hn_use_if_start = 0;
285 SYSCTL_INT(_hw_hn, OID_AUTO, use_if_start, CTLFLAG_RDTUN,
286 &hn_use_if_start, 0, "Use if_start TX method");
288 static int hn_chan_cnt = 0;
289 SYSCTL_INT(_hw_hn, OID_AUTO, chan_cnt, CTLFLAG_RDTUN,
291 "# of channels to use; each channel has one RX ring and one TX ring");
293 static int hn_tx_ring_cnt = 0;
294 SYSCTL_INT(_hw_hn, OID_AUTO, tx_ring_cnt, CTLFLAG_RDTUN,
295 &hn_tx_ring_cnt, 0, "# of TX rings to use");
297 static int hn_tx_swq_depth = 0;
298 SYSCTL_INT(_hw_hn, OID_AUTO, tx_swq_depth, CTLFLAG_RDTUN,
299 &hn_tx_swq_depth, 0, "Depth of IFQ or BUFRING");
301 static u_int hn_cpu_index;
304 * Forward declarations
306 static void hn_stop(hn_softc_t *sc);
307 static void hn_ifinit_locked(hn_softc_t *sc);
308 static void hn_ifinit(void *xsc);
309 static int hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data);
310 static int hn_start_locked(struct hn_tx_ring *txr, int len);
311 static void hn_start(struct ifnet *ifp);
312 static void hn_start_txeof(struct hn_tx_ring *);
313 static int hn_ifmedia_upd(struct ifnet *ifp);
314 static void hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr);
315 #if __FreeBSD_version >= 1100099
316 static int hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS);
317 static int hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS);
319 static int hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS);
320 static int hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS);
321 #if __FreeBSD_version < 1100095
322 static int hn_rx_stat_int_sysctl(SYSCTL_HANDLER_ARGS);
324 static int hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS);
326 static int hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS);
327 static int hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS);
328 static int hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS);
329 static int hn_check_iplen(const struct mbuf *, int);
330 static int hn_create_tx_ring(struct hn_softc *, int);
331 static void hn_destroy_tx_ring(struct hn_tx_ring *);
332 static int hn_create_tx_data(struct hn_softc *, int);
333 static void hn_destroy_tx_data(struct hn_softc *);
334 static void hn_start_taskfunc(void *, int);
335 static void hn_start_txeof_taskfunc(void *, int);
336 static void hn_stop_tx_tasks(struct hn_softc *);
337 static int hn_encap(struct hn_tx_ring *, struct hn_txdesc *, struct mbuf **);
338 static void hn_create_rx_data(struct hn_softc *sc, int);
339 static void hn_destroy_rx_data(struct hn_softc *sc);
340 static void hn_set_tx_chimney_size(struct hn_softc *, int);
341 static void hn_channel_attach(struct hn_softc *, struct hv_vmbus_channel *);
342 static void hn_subchan_attach(struct hn_softc *, struct hv_vmbus_channel *);
344 static int hn_transmit(struct ifnet *, struct mbuf *);
345 static void hn_xmit_qflush(struct ifnet *);
346 static int hn_xmit(struct hn_tx_ring *, int);
347 static void hn_xmit_txeof(struct hn_tx_ring *);
348 static void hn_xmit_taskfunc(void *, int);
349 static void hn_xmit_txeof_taskfunc(void *, int);
351 #if __FreeBSD_version >= 1100099
353 hn_set_lro_lenlim(struct hn_softc *sc, int lenlim)
357 for (i = 0; i < sc->hn_rx_ring_inuse; ++i)
358 sc->hn_rx_ring[i].hn_lro.lro_length_lim = lenlim;
363 hn_get_txswq_depth(const struct hn_tx_ring *txr)
366 KASSERT(txr->hn_txdesc_cnt > 0, ("tx ring is not setup yet"));
367 if (hn_tx_swq_depth < txr->hn_txdesc_cnt)
368 return txr->hn_txdesc_cnt;
369 return hn_tx_swq_depth;
373 hn_ifmedia_upd(struct ifnet *ifp __unused)
380 hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
382 struct hn_softc *sc = ifp->if_softc;
384 ifmr->ifm_status = IFM_AVALID;
385 ifmr->ifm_active = IFM_ETHER;
387 if (!sc->hn_carrier) {
388 ifmr->ifm_active |= IFM_NONE;
391 ifmr->ifm_status |= IFM_ACTIVE;
392 ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
395 /* {F8615163-DF3E-46c5-913F-F2D2F965ED0E} */
396 static const hv_guid g_net_vsc_device_type = {
397 .data = {0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
398 0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E}
402 * Standard probe entry point.
406 netvsc_probe(device_t dev)
410 p = vmbus_get_type(dev);
411 if (!memcmp(p, &g_net_vsc_device_type.data, sizeof(hv_guid))) {
412 device_set_desc(dev, "Hyper-V Network Interface");
414 printf("Netvsc probe... DONE \n");
416 return (BUS_PROBE_DEFAULT);
423 hn_cpuset_setthread_task(void *xmask, int pending __unused)
425 cpuset_t *mask = xmask;
428 error = cpuset_setthread(curthread->td_tid, mask);
430 panic("curthread=%ju: can't pin; error=%d",
431 (uintmax_t)curthread->td_tid, error);
436 * Standard attach entry point.
438 * Called when the driver is loaded. It allocates needed resources,
439 * and initializes the "hardware" and software.
442 netvsc_attach(device_t dev)
444 struct hv_device *device_ctx = vmbus_get_devctx(dev);
445 struct hv_vmbus_channel *pri_chan;
446 netvsc_device_info device_info;
448 int unit = device_get_unit(dev);
449 struct ifnet *ifp = NULL;
450 int error, ring_cnt, tx_ring_cnt;
453 sc = device_get_softc(dev);
458 if (hn_tx_taskq == NULL) {
459 sc->hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
460 taskqueue_thread_enqueue, &sc->hn_tx_taskq);
461 taskqueue_start_threads(&sc->hn_tx_taskq, 1, PI_NET, "%s tx",
462 device_get_nameunit(dev));
463 if (hn_bind_tx_taskq >= 0) {
464 int cpu = hn_bind_tx_taskq;
465 struct task cpuset_task;
468 if (cpu > mp_ncpus - 1)
470 CPU_SETOF(cpu, &cpu_set);
471 TASK_INIT(&cpuset_task, 0, hn_cpuset_setthread_task,
473 taskqueue_enqueue(sc->hn_tx_taskq, &cpuset_task);
474 taskqueue_drain(sc->hn_tx_taskq, &cpuset_task);
477 sc->hn_tx_taskq = hn_tx_taskq;
479 NV_LOCK_INIT(sc, "NetVSCLock");
481 sc->hn_dev_obj = device_ctx;
483 ifp = sc->hn_ifp = sc->arpcom.ac_ifp = if_alloc(IFT_ETHER);
485 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
488 * Figure out the # of RX rings (ring_cnt) and the # of TX rings
489 * to use (tx_ring_cnt).
492 * The # of RX rings to use is same as the # of channels to use.
494 ring_cnt = hn_chan_cnt;
498 if (ring_cnt > HN_RING_CNT_DEF_MAX)
499 ring_cnt = HN_RING_CNT_DEF_MAX;
500 } else if (ring_cnt > mp_ncpus) {
504 tx_ring_cnt = hn_tx_ring_cnt;
505 if (tx_ring_cnt <= 0 || tx_ring_cnt > ring_cnt)
506 tx_ring_cnt = ring_cnt;
507 if (hn_use_if_start) {
508 /* ifnet.if_start only needs one TX ring. */
513 * Set the leader CPU for channels.
515 sc->hn_cpu = atomic_fetchadd_int(&hn_cpu_index, ring_cnt) % mp_ncpus;
517 error = hn_create_tx_data(sc, tx_ring_cnt);
520 hn_create_rx_data(sc, ring_cnt);
523 * Associate the first TX/RX ring w/ the primary channel.
525 pri_chan = device_ctx->channel;
526 KASSERT(HV_VMBUS_CHAN_ISPRIMARY(pri_chan), ("not primary channel"));
527 KASSERT(pri_chan->offer_msg.offer.sub_channel_index == 0,
528 ("primary channel subidx %u",
529 pri_chan->offer_msg.offer.sub_channel_index));
530 hn_channel_attach(sc, pri_chan);
532 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
533 ifp->if_ioctl = hn_ioctl;
534 ifp->if_init = hn_ifinit;
535 /* needed by hv_rf_on_device_add() code */
536 ifp->if_mtu = ETHERMTU;
537 if (hn_use_if_start) {
538 int qdepth = hn_get_txswq_depth(&sc->hn_tx_ring[0]);
540 ifp->if_start = hn_start;
541 IFQ_SET_MAXLEN(&ifp->if_snd, qdepth);
542 ifp->if_snd.ifq_drv_maxlen = qdepth - 1;
543 IFQ_SET_READY(&ifp->if_snd);
545 ifp->if_transmit = hn_transmit;
546 ifp->if_qflush = hn_xmit_qflush;
549 ifmedia_init(&sc->hn_media, 0, hn_ifmedia_upd, hn_ifmedia_sts);
550 ifmedia_add(&sc->hn_media, IFM_ETHER | IFM_AUTO, 0, NULL);
551 ifmedia_set(&sc->hn_media, IFM_ETHER | IFM_AUTO);
552 /* XXX ifmedia_set really should do this for us */
553 sc->hn_media.ifm_media = sc->hn_media.ifm_cur->ifm_media;
556 * Tell upper layers that we support full VLAN capability.
558 ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
559 ifp->if_capabilities |=
560 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
563 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
565 ifp->if_hwassist = sc->hn_tx_ring[0].hn_csum_assist | CSUM_TSO;
567 error = hv_rf_on_device_add(device_ctx, &device_info, ring_cnt);
570 KASSERT(sc->net_dev->num_channel > 0 &&
571 sc->net_dev->num_channel <= sc->hn_rx_ring_inuse,
572 ("invalid channel count %u, should be less than %d",
573 sc->net_dev->num_channel, sc->hn_rx_ring_inuse));
576 * Set the # of TX/RX rings that could be used according to
577 * the # of channels that host offered.
579 if (sc->hn_tx_ring_inuse > sc->net_dev->num_channel)
580 sc->hn_tx_ring_inuse = sc->net_dev->num_channel;
581 sc->hn_rx_ring_inuse = sc->net_dev->num_channel;
582 device_printf(dev, "%d TX ring, %d RX ring\n",
583 sc->hn_tx_ring_inuse, sc->hn_rx_ring_inuse);
585 if (sc->net_dev->num_channel > 1) {
586 struct hv_vmbus_channel **subchan;
587 int subchan_cnt = sc->net_dev->num_channel - 1;
590 /* Wait for sub-channels setup to complete. */
591 subchan = vmbus_get_subchan(pri_chan, subchan_cnt);
593 /* Attach the sub-channels. */
594 for (i = 0; i < subchan_cnt; ++i) {
595 /* NOTE: Calling order is critical. */
596 hn_subchan_attach(sc, subchan[i]);
597 hv_nv_subchan_attach(subchan[i]);
600 /* Release the sub-channels */
601 vmbus_rel_subchan(subchan, subchan_cnt);
602 device_printf(dev, "%d sub-channels setup done\n", subchan_cnt);
605 #if __FreeBSD_version >= 1100099
606 if (sc->hn_rx_ring_inuse > 1) {
608 * Reduce TCP segment aggregation limit for multiple
609 * RX rings to increase ACK timeliness.
611 hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MULTIRX_DEF);
615 if (device_info.link_state == 0) {
619 tso_maxlen = hn_tso_maxlen;
620 if (tso_maxlen <= 0 || tso_maxlen > IP_MAXPACKET)
621 tso_maxlen = IP_MAXPACKET;
623 ifp->if_hw_tsomaxsegcount = HN_TX_DATA_SEGCNT_MAX;
624 ifp->if_hw_tsomaxsegsize = PAGE_SIZE;
625 ifp->if_hw_tsomax = tso_maxlen -
626 (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
628 ether_ifattach(ifp, device_info.mac_addr);
630 if_printf(ifp, "TSO: %u/%u/%u\n", ifp->if_hw_tsomax,
631 ifp->if_hw_tsomaxsegcount, ifp->if_hw_tsomaxsegsize);
633 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
634 hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max);
635 if (hn_tx_chimney_size > 0 &&
636 hn_tx_chimney_size < sc->hn_tx_chimney_max)
637 hn_set_tx_chimney_size(sc, hn_tx_chimney_size);
641 hn_destroy_tx_data(sc);
648 * Standard detach entry point
651 netvsc_detach(device_t dev)
653 struct hn_softc *sc = device_get_softc(dev);
654 struct hv_device *hv_device = vmbus_get_devctx(dev);
657 printf("netvsc_detach\n");
660 * XXXKYS: Need to clean up all our
661 * driver state; this is the driver
666 * XXXKYS: Need to stop outgoing traffic and unregister
670 hv_rf_on_device_remove(hv_device, HV_RF_NV_DESTROY_CHANNEL);
672 hn_stop_tx_tasks(sc);
674 ifmedia_removeall(&sc->hn_media);
675 hn_destroy_rx_data(sc);
676 hn_destroy_tx_data(sc);
678 if (sc->hn_tx_taskq != hn_tx_taskq)
679 taskqueue_free(sc->hn_tx_taskq);
685 * Standard shutdown entry point
688 netvsc_shutdown(device_t dev)
694 hn_txdesc_dmamap_load(struct hn_tx_ring *txr, struct hn_txdesc *txd,
695 struct mbuf **m_head, bus_dma_segment_t *segs, int *nsegs)
697 struct mbuf *m = *m_head;
700 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag, txd->data_dmap,
701 m, segs, nsegs, BUS_DMA_NOWAIT);
702 if (error == EFBIG) {
705 m_new = m_collapse(m, M_NOWAIT, HN_TX_DATA_SEGCNT_MAX);
710 txr->hn_tx_collapsed++;
712 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag,
713 txd->data_dmap, m, segs, nsegs, BUS_DMA_NOWAIT);
716 bus_dmamap_sync(txr->hn_tx_data_dtag, txd->data_dmap,
717 BUS_DMASYNC_PREWRITE);
718 txd->flags |= HN_TXD_FLAG_DMAMAP;
724 hn_txdesc_dmamap_unload(struct hn_tx_ring *txr, struct hn_txdesc *txd)
727 if (txd->flags & HN_TXD_FLAG_DMAMAP) {
728 bus_dmamap_sync(txr->hn_tx_data_dtag,
729 txd->data_dmap, BUS_DMASYNC_POSTWRITE);
730 bus_dmamap_unload(txr->hn_tx_data_dtag,
732 txd->flags &= ~HN_TXD_FLAG_DMAMAP;
737 hn_txdesc_put(struct hn_tx_ring *txr, struct hn_txdesc *txd)
740 KASSERT((txd->flags & HN_TXD_FLAG_ONLIST) == 0,
741 ("put an onlist txd %#x", txd->flags));
743 KASSERT(txd->refs > 0, ("invalid txd refs %d", txd->refs));
744 if (atomic_fetchadd_int(&txd->refs, -1) != 1)
747 hn_txdesc_dmamap_unload(txr, txd);
748 if (txd->m != NULL) {
753 txd->flags |= HN_TXD_FLAG_ONLIST;
755 #ifndef HN_USE_TXDESC_BUFRING
756 mtx_lock_spin(&txr->hn_txlist_spin);
757 KASSERT(txr->hn_txdesc_avail >= 0 &&
758 txr->hn_txdesc_avail < txr->hn_txdesc_cnt,
759 ("txdesc_put: invalid txd avail %d", txr->hn_txdesc_avail));
760 txr->hn_txdesc_avail++;
761 SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
762 mtx_unlock_spin(&txr->hn_txlist_spin);
764 atomic_add_int(&txr->hn_txdesc_avail, 1);
765 buf_ring_enqueue(txr->hn_txdesc_br, txd);
771 static __inline struct hn_txdesc *
772 hn_txdesc_get(struct hn_tx_ring *txr)
774 struct hn_txdesc *txd;
776 #ifndef HN_USE_TXDESC_BUFRING
777 mtx_lock_spin(&txr->hn_txlist_spin);
778 txd = SLIST_FIRST(&txr->hn_txlist);
780 KASSERT(txr->hn_txdesc_avail > 0,
781 ("txdesc_get: invalid txd avail %d", txr->hn_txdesc_avail));
782 txr->hn_txdesc_avail--;
783 SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
785 mtx_unlock_spin(&txr->hn_txlist_spin);
787 txd = buf_ring_dequeue_sc(txr->hn_txdesc_br);
791 #ifdef HN_USE_TXDESC_BUFRING
792 atomic_subtract_int(&txr->hn_txdesc_avail, 1);
794 KASSERT(txd->m == NULL && txd->refs == 0 &&
795 (txd->flags & HN_TXD_FLAG_ONLIST), ("invalid txd"));
796 txd->flags &= ~HN_TXD_FLAG_ONLIST;
803 hn_txdesc_hold(struct hn_txdesc *txd)
806 /* 0->1 transition will never work */
807 KASSERT(txd->refs > 0, ("invalid refs %d", txd->refs));
808 atomic_add_int(&txd->refs, 1);
812 hn_txeof(struct hn_tx_ring *txr)
814 txr->hn_has_txeof = 0;
819 hn_tx_done(struct hv_vmbus_channel *chan, void *xpkt)
821 netvsc_packet *packet = xpkt;
822 struct hn_txdesc *txd;
823 struct hn_tx_ring *txr;
825 txd = (struct hn_txdesc *)(uintptr_t)
826 packet->compl.send.send_completion_tid;
829 KASSERT(txr->hn_chan == chan,
830 ("channel mismatch, on channel%u, should be channel%u",
831 chan->offer_msg.offer.sub_channel_index,
832 txr->hn_chan->offer_msg.offer.sub_channel_index));
834 txr->hn_has_txeof = 1;
835 hn_txdesc_put(txr, txd);
837 ++txr->hn_txdone_cnt;
838 if (txr->hn_txdone_cnt >= HN_EARLY_TXEOF_THRESH) {
839 txr->hn_txdone_cnt = 0;
846 netvsc_channel_rollup(struct hv_vmbus_channel *chan)
848 struct hn_tx_ring *txr = chan->hv_chan_txr;
849 #if defined(INET) || defined(INET6)
850 struct hn_rx_ring *rxr = chan->hv_chan_rxr;
851 struct lro_ctrl *lro = &rxr->hn_lro;
852 struct lro_entry *queued;
854 while ((queued = SLIST_FIRST(&lro->lro_active)) != NULL) {
855 SLIST_REMOVE_HEAD(&lro->lro_active, next);
856 tcp_lro_flush(lro, queued);
862 * 'txr' could be NULL, if multiple channels and
863 * ifnet.if_start method are enabled.
865 if (txr == NULL || !txr->hn_has_txeof)
868 txr->hn_txdone_cnt = 0;
874 * If this function fails, then both txd and m_head0 will be freed.
877 hn_encap(struct hn_tx_ring *txr, struct hn_txdesc *txd, struct mbuf **m_head0)
879 bus_dma_segment_t segs[HN_TX_DATA_SEGCNT_MAX];
881 struct mbuf *m_head = *m_head0;
882 netvsc_packet *packet;
883 rndis_msg *rndis_mesg;
884 rndis_packet *rndis_pkt;
885 rndis_per_packet_info *rppi;
886 struct ndis_hash_info *hash_info;
887 uint32_t rndis_msg_size;
889 packet = &txd->netvsc_pkt;
890 packet->is_data_pkt = TRUE;
891 packet->tot_data_buf_len = m_head->m_pkthdr.len;
894 * extension points to the area reserved for the
895 * rndis_filter_packet, which is placed just after
896 * the netvsc_packet (and rppi struct, if present;
897 * length is updated later).
899 rndis_mesg = txd->rndis_msg;
900 /* XXX not necessary */
901 memset(rndis_mesg, 0, HN_RNDIS_MSG_LEN);
902 rndis_mesg->ndis_msg_type = REMOTE_NDIS_PACKET_MSG;
904 rndis_pkt = &rndis_mesg->msg.packet;
905 rndis_pkt->data_offset = sizeof(rndis_packet);
906 rndis_pkt->data_length = packet->tot_data_buf_len;
907 rndis_pkt->per_pkt_info_offset = sizeof(rndis_packet);
909 rndis_msg_size = RNDIS_MESSAGE_SIZE(rndis_packet);
912 * Set the hash info for this packet, so that the host could
913 * dispatch the TX done event for this packet back to this TX
916 rndis_msg_size += RNDIS_HASH_PPI_SIZE;
917 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_HASH_PPI_SIZE,
919 hash_info = (struct ndis_hash_info *)((uint8_t *)rppi +
920 rppi->per_packet_info_offset);
921 hash_info->hash = txr->hn_tx_idx;
923 if (m_head->m_flags & M_VLANTAG) {
924 ndis_8021q_info *rppi_vlan_info;
926 rndis_msg_size += RNDIS_VLAN_PPI_SIZE;
927 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_VLAN_PPI_SIZE,
930 rppi_vlan_info = (ndis_8021q_info *)((uint8_t *)rppi +
931 rppi->per_packet_info_offset);
932 rppi_vlan_info->u1.s1.vlan_id =
933 m_head->m_pkthdr.ether_vtag & 0xfff;
936 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
937 rndis_tcp_tso_info *tso_info;
938 struct ether_vlan_header *eh;
942 * XXX need m_pullup and use mtodo
944 eh = mtod(m_head, struct ether_vlan_header*);
945 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN))
946 ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
948 ether_len = ETHER_HDR_LEN;
950 rndis_msg_size += RNDIS_TSO_PPI_SIZE;
951 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_TSO_PPI_SIZE,
952 tcp_large_send_info);
954 tso_info = (rndis_tcp_tso_info *)((uint8_t *)rppi +
955 rppi->per_packet_info_offset);
956 tso_info->lso_v2_xmit.type =
957 RNDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
960 if (m_head->m_pkthdr.csum_flags & CSUM_IP_TSO) {
962 (struct ip *)(m_head->m_data + ether_len);
963 unsigned long iph_len = ip->ip_hl << 2;
965 (struct tcphdr *)((caddr_t)ip + iph_len);
967 tso_info->lso_v2_xmit.ip_version =
968 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
972 th->th_sum = in_pseudo(ip->ip_src.s_addr,
973 ip->ip_dst.s_addr, htons(IPPROTO_TCP));
976 #if defined(INET6) && defined(INET)
981 struct ip6_hdr *ip6 = (struct ip6_hdr *)
982 (m_head->m_data + ether_len);
983 struct tcphdr *th = (struct tcphdr *)(ip6 + 1);
985 tso_info->lso_v2_xmit.ip_version =
986 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
988 th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
991 tso_info->lso_v2_xmit.tcp_header_offset = 0;
992 tso_info->lso_v2_xmit.mss = m_head->m_pkthdr.tso_segsz;
993 } else if (m_head->m_pkthdr.csum_flags & txr->hn_csum_assist) {
994 rndis_tcp_ip_csum_info *csum_info;
996 rndis_msg_size += RNDIS_CSUM_PPI_SIZE;
997 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_CSUM_PPI_SIZE,
999 csum_info = (rndis_tcp_ip_csum_info *)((uint8_t *)rppi +
1000 rppi->per_packet_info_offset);
1002 csum_info->xmit.is_ipv4 = 1;
1003 if (m_head->m_pkthdr.csum_flags & CSUM_IP)
1004 csum_info->xmit.ip_header_csum = 1;
1006 if (m_head->m_pkthdr.csum_flags & CSUM_TCP) {
1007 csum_info->xmit.tcp_csum = 1;
1008 csum_info->xmit.tcp_header_offset = 0;
1009 } else if (m_head->m_pkthdr.csum_flags & CSUM_UDP) {
1010 csum_info->xmit.udp_csum = 1;
1014 rndis_mesg->msg_len = packet->tot_data_buf_len + rndis_msg_size;
1015 packet->tot_data_buf_len = rndis_mesg->msg_len;
1018 * Chimney send, if the packet could fit into one chimney buffer.
1020 if (packet->tot_data_buf_len < txr->hn_tx_chimney_size) {
1021 netvsc_dev *net_dev = txr->hn_sc->net_dev;
1022 uint32_t send_buf_section_idx;
1024 txr->hn_tx_chimney_tried++;
1025 send_buf_section_idx =
1026 hv_nv_get_next_send_section(net_dev);
1027 if (send_buf_section_idx !=
1028 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX) {
1029 uint8_t *dest = ((uint8_t *)net_dev->send_buf +
1030 (send_buf_section_idx *
1031 net_dev->send_section_size));
1033 memcpy(dest, rndis_mesg, rndis_msg_size);
1034 dest += rndis_msg_size;
1035 m_copydata(m_head, 0, m_head->m_pkthdr.len, dest);
1037 packet->send_buf_section_idx = send_buf_section_idx;
1038 packet->send_buf_section_size =
1039 packet->tot_data_buf_len;
1040 packet->page_buf_count = 0;
1041 txr->hn_tx_chimney++;
1046 error = hn_txdesc_dmamap_load(txr, txd, &m_head, segs, &nsegs);
1051 * This mbuf is not linked w/ the txd yet, so free it now.
1056 freed = hn_txdesc_put(txr, txd);
1058 ("fail to free txd upon txdma error"));
1060 txr->hn_txdma_failed++;
1061 if_inc_counter(txr->hn_sc->hn_ifp, IFCOUNTER_OERRORS, 1);
1066 packet->page_buf_count = nsegs + HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
1068 /* send packet with page buffer */
1069 packet->page_buffers[0].pfn = atop(txd->rndis_msg_paddr);
1070 packet->page_buffers[0].offset = txd->rndis_msg_paddr & PAGE_MASK;
1071 packet->page_buffers[0].length = rndis_msg_size;
1074 * Fill the page buffers with mbuf info starting at index
1075 * HV_RF_NUM_TX_RESERVED_PAGE_BUFS.
1077 for (i = 0; i < nsegs; ++i) {
1078 hv_vmbus_page_buffer *pb = &packet->page_buffers[
1079 i + HV_RF_NUM_TX_RESERVED_PAGE_BUFS];
1081 pb->pfn = atop(segs[i].ds_addr);
1082 pb->offset = segs[i].ds_addr & PAGE_MASK;
1083 pb->length = segs[i].ds_len;
1086 packet->send_buf_section_idx =
1087 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX;
1088 packet->send_buf_section_size = 0;
1092 /* Set the completion routine */
1093 packet->compl.send.on_send_completion = hn_tx_done;
1094 packet->compl.send.send_completion_context = packet;
1095 packet->compl.send.send_completion_tid = (uint64_t)(uintptr_t)txd;
1102 * If this function fails, then txd will be freed, but the mbuf
1103 * associated w/ the txd will _not_ be freed.
1106 hn_send_pkt(struct ifnet *ifp, struct hn_tx_ring *txr, struct hn_txdesc *txd)
1108 int error, send_failed = 0;
1112 * Make sure that txd is not freed before ETHER_BPF_MTAP.
1114 hn_txdesc_hold(txd);
1115 error = hv_nv_on_send(txr->hn_chan, &txd->netvsc_pkt);
1117 ETHER_BPF_MTAP(ifp, txd->m);
1118 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1119 if (!hn_use_if_start) {
1120 if_inc_counter(ifp, IFCOUNTER_OBYTES,
1121 txd->m->m_pkthdr.len);
1122 if (txd->m->m_flags & M_MCAST)
1123 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
1127 hn_txdesc_put(txr, txd);
1129 if (__predict_false(error)) {
1133 * This should "really rarely" happen.
1135 * XXX Too many RX to be acked or too many sideband
1136 * commands to run? Ask netvsc_channel_rollup()
1137 * to kick start later.
1139 txr->hn_has_txeof = 1;
1141 txr->hn_send_failed++;
1144 * Try sending again after set hn_has_txeof;
1145 * in case that we missed the last
1146 * netvsc_channel_rollup().
1150 if_printf(ifp, "send failed\n");
1153 * Caller will perform further processing on the
1154 * associated mbuf, so don't free it in hn_txdesc_put();
1155 * only unload it from the DMA map in hn_txdesc_put(),
1159 freed = hn_txdesc_put(txr, txd);
1161 ("fail to free txd upon send error"));
1163 txr->hn_send_failed++;
1169 * Start a transmit of one or more packets
1172 hn_start_locked(struct hn_tx_ring *txr, int len)
1174 struct hn_softc *sc = txr->hn_sc;
1175 struct ifnet *ifp = sc->hn_ifp;
1177 KASSERT(hn_use_if_start,
1178 ("hn_start_locked is called, when if_start is disabled"));
1179 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1180 mtx_assert(&txr->hn_tx_lock, MA_OWNED);
1182 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1186 while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
1187 struct hn_txdesc *txd;
1188 struct mbuf *m_head;
1191 IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
1195 if (len > 0 && m_head->m_pkthdr.len > len) {
1197 * This sending could be time consuming; let callers
1198 * dispatch this packet sending (and sending of any
1199 * following up packets) to tx taskqueue.
1201 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1205 txd = hn_txdesc_get(txr);
1207 txr->hn_no_txdescs++;
1208 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1209 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1213 error = hn_encap(txr, txd, &m_head);
1215 /* Both txd and m_head are freed */
1219 error = hn_send_pkt(ifp, txr, txd);
1220 if (__predict_false(error)) {
1221 /* txd is freed, but m_head is not */
1222 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1223 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1231 * Link up/down notification
1234 netvsc_linkstatus_callback(struct hv_device *device_obj, uint32_t status)
1236 hn_softc_t *sc = device_get_softc(device_obj->device);
1246 * Append the specified data to the indicated mbuf chain,
1247 * Extend the mbuf chain if the new data does not fit in
1250 * This is a minor rewrite of m_append() from sys/kern/uipc_mbuf.c.
1251 * There should be an equivalent in the kernel mbuf code,
1252 * but there does not appear to be one yet.
1254 * Differs from m_append() in that additional mbufs are
1255 * allocated with cluster size MJUMPAGESIZE, and filled
1258 * Return 1 if able to complete the job; otherwise 0.
1261 hv_m_append(struct mbuf *m0, int len, c_caddr_t cp)
1264 int remainder, space;
1266 for (m = m0; m->m_next != NULL; m = m->m_next)
1269 space = M_TRAILINGSPACE(m);
1272 * Copy into available space.
1274 if (space > remainder)
1276 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1281 while (remainder > 0) {
1283 * Allocate a new mbuf; could check space
1284 * and allocate a cluster instead.
1286 n = m_getjcl(M_DONTWAIT, m->m_type, 0, MJUMPAGESIZE);
1289 n->m_len = min(MJUMPAGESIZE, remainder);
1290 bcopy(cp, mtod(n, caddr_t), n->m_len);
1292 remainder -= n->m_len;
1296 if (m0->m_flags & M_PKTHDR)
1297 m0->m_pkthdr.len += len - remainder;
1299 return (remainder == 0);
1304 * Called when we receive a data packet from the "wire" on the
1307 * Note: This is no longer used as a callback
1310 netvsc_recv(struct hv_vmbus_channel *chan, netvsc_packet *packet,
1311 rndis_tcp_ip_csum_info *csum_info)
1313 struct hn_rx_ring *rxr = chan->hv_chan_rxr;
1314 struct ifnet *ifp = rxr->hn_ifp;
1316 int size, do_lro = 0, do_csum = 1;
1318 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1322 * Bail out if packet contains more data than configured MTU.
1324 if (packet->tot_data_buf_len > (ifp->if_mtu + ETHER_HDR_LEN)) {
1326 } else if (packet->tot_data_buf_len <= MHLEN) {
1327 m_new = m_gethdr(M_NOWAIT, MT_DATA);
1328 if (m_new == NULL) {
1329 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1332 memcpy(mtod(m_new, void *), packet->data,
1333 packet->tot_data_buf_len);
1334 m_new->m_pkthdr.len = m_new->m_len = packet->tot_data_buf_len;
1335 rxr->hn_small_pkts++;
1338 * Get an mbuf with a cluster. For packets 2K or less,
1339 * get a standard 2K cluster. For anything larger, get a
1340 * 4K cluster. Any buffers larger than 4K can cause problems
1341 * if looped around to the Hyper-V TX channel, so avoid them.
1344 if (packet->tot_data_buf_len > MCLBYTES) {
1346 size = MJUMPAGESIZE;
1349 m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, size);
1350 if (m_new == NULL) {
1351 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1355 hv_m_append(m_new, packet->tot_data_buf_len, packet->data);
1357 m_new->m_pkthdr.rcvif = ifp;
1359 if (__predict_false((ifp->if_capenable & IFCAP_RXCSUM) == 0))
1362 /* receive side checksum offload */
1363 if (csum_info != NULL) {
1364 /* IP csum offload */
1365 if (csum_info->receive.ip_csum_succeeded && do_csum) {
1366 m_new->m_pkthdr.csum_flags |=
1367 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1371 /* TCP/UDP csum offload */
1372 if ((csum_info->receive.tcp_csum_succeeded ||
1373 csum_info->receive.udp_csum_succeeded) && do_csum) {
1374 m_new->m_pkthdr.csum_flags |=
1375 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1376 m_new->m_pkthdr.csum_data = 0xffff;
1377 if (csum_info->receive.tcp_csum_succeeded)
1383 if (csum_info->receive.ip_csum_succeeded &&
1384 csum_info->receive.tcp_csum_succeeded)
1387 const struct ether_header *eh;
1392 if (m_new->m_len < hoff)
1394 eh = mtod(m_new, struct ether_header *);
1395 etype = ntohs(eh->ether_type);
1396 if (etype == ETHERTYPE_VLAN) {
1397 const struct ether_vlan_header *evl;
1399 hoff = sizeof(*evl);
1400 if (m_new->m_len < hoff)
1402 evl = mtod(m_new, struct ether_vlan_header *);
1403 etype = ntohs(evl->evl_proto);
1406 if (etype == ETHERTYPE_IP) {
1409 pr = hn_check_iplen(m_new, hoff);
1410 if (pr == IPPROTO_TCP) {
1412 (rxr->hn_trust_hcsum &
1413 HN_TRUST_HCSUM_TCP)) {
1414 rxr->hn_csum_trusted++;
1415 m_new->m_pkthdr.csum_flags |=
1416 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1417 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1418 m_new->m_pkthdr.csum_data = 0xffff;
1420 /* Rely on SW csum verification though... */
1422 } else if (pr == IPPROTO_UDP) {
1424 (rxr->hn_trust_hcsum &
1425 HN_TRUST_HCSUM_UDP)) {
1426 rxr->hn_csum_trusted++;
1427 m_new->m_pkthdr.csum_flags |=
1428 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1429 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1430 m_new->m_pkthdr.csum_data = 0xffff;
1432 } else if (pr != IPPROTO_DONE && do_csum &&
1433 (rxr->hn_trust_hcsum & HN_TRUST_HCSUM_IP)) {
1434 rxr->hn_csum_trusted++;
1435 m_new->m_pkthdr.csum_flags |=
1436 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1441 if ((packet->vlan_tci != 0) &&
1442 (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0) {
1443 m_new->m_pkthdr.ether_vtag = packet->vlan_tci;
1444 m_new->m_flags |= M_VLANTAG;
1447 m_new->m_pkthdr.flowid = rxr->hn_rx_idx;
1448 M_HASHTYPE_SET(m_new, M_HASHTYPE_OPAQUE);
1451 * Note: Moved RX completion back to hv_nv_on_receive() so all
1452 * messages (not just data messages) will trigger a response.
1458 if ((ifp->if_capenable & IFCAP_LRO) && do_lro) {
1459 #if defined(INET) || defined(INET6)
1460 struct lro_ctrl *lro = &rxr->hn_lro;
1463 rxr->hn_lro_tried++;
1464 if (tcp_lro_rx(lro, m_new, 0) == 0) {
1472 /* We're not holding the lock here, so don't release it */
1473 (*ifp->if_input)(ifp, m_new);
1479 * Rules for using sc->temp_unusable:
1480 * 1. sc->temp_unusable can only be read or written while holding NV_LOCK()
1481 * 2. code reading sc->temp_unusable under NV_LOCK(), and finding
1482 * sc->temp_unusable set, must release NV_LOCK() and exit
1483 * 3. to retain exclusive control of the interface,
1484 * sc->temp_unusable must be set by code before releasing NV_LOCK()
1485 * 4. only code setting sc->temp_unusable can clear sc->temp_unusable
1486 * 5. code setting sc->temp_unusable must eventually clear sc->temp_unusable
1490 * Standard ioctl entry point. Called when the user wants to configure
1494 hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1496 hn_softc_t *sc = ifp->if_softc;
1497 struct ifreq *ifr = (struct ifreq *)data;
1499 struct ifaddr *ifa = (struct ifaddr *)data;
1501 netvsc_device_info device_info;
1502 struct hv_device *hn_dev;
1503 int mask, error = 0;
1504 int retry_cnt = 500;
1510 if (ifa->ifa_addr->sa_family == AF_INET) {
1511 ifp->if_flags |= IFF_UP;
1512 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1514 arp_ifinit(ifp, ifa);
1517 error = ether_ioctl(ifp, cmd, data);
1520 hn_dev = vmbus_get_devctx(sc->hn_dev);
1522 /* Check MTU value change */
1523 if (ifp->if_mtu == ifr->ifr_mtu)
1526 if (ifr->ifr_mtu > NETVSC_MAX_CONFIGURABLE_MTU) {
1531 /* Obtain and record requested MTU */
1532 ifp->if_mtu = ifr->ifr_mtu;
1534 #if __FreeBSD_version >= 1100099
1536 * Make sure that LRO aggregation length limit is still
1537 * valid, after the MTU change.
1540 if (sc->hn_rx_ring[0].hn_lro.lro_length_lim <
1541 HN_LRO_LENLIM_MIN(ifp))
1542 hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MIN(ifp));
1548 if (!sc->temp_unusable) {
1549 sc->temp_unusable = TRUE;
1553 if (retry_cnt > 0) {
1557 } while (retry_cnt > 0);
1559 if (retry_cnt == 0) {
1564 /* We must remove and add back the device to cause the new
1565 * MTU to take effect. This includes tearing down, but not
1566 * deleting the channel, then bringing it back up.
1568 error = hv_rf_on_device_remove(hn_dev, HV_RF_NV_RETAIN_CHANNEL);
1571 sc->temp_unusable = FALSE;
1575 error = hv_rf_on_device_add(hn_dev, &device_info,
1576 sc->hn_rx_ring_inuse);
1579 sc->temp_unusable = FALSE;
1584 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
1585 if (sc->hn_tx_ring[0].hn_tx_chimney_size >
1586 sc->hn_tx_chimney_max)
1587 hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max);
1589 hn_ifinit_locked(sc);
1592 sc->temp_unusable = FALSE;
1598 if (!sc->temp_unusable) {
1599 sc->temp_unusable = TRUE;
1603 if (retry_cnt > 0) {
1607 } while (retry_cnt > 0);
1609 if (retry_cnt == 0) {
1614 if (ifp->if_flags & IFF_UP) {
1616 * If only the state of the PROMISC flag changed,
1617 * then just use the 'set promisc mode' command
1618 * instead of reinitializing the entire NIC. Doing
1619 * a full re-init means reloading the firmware and
1620 * waiting for it to start up, which may take a
1624 /* Fixme: Promiscuous mode? */
1625 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1626 ifp->if_flags & IFF_PROMISC &&
1627 !(sc->hn_if_flags & IFF_PROMISC)) {
1628 /* do something here for Hyper-V */
1629 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1630 !(ifp->if_flags & IFF_PROMISC) &&
1631 sc->hn_if_flags & IFF_PROMISC) {
1632 /* do something here for Hyper-V */
1635 hn_ifinit_locked(sc);
1637 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1642 sc->temp_unusable = FALSE;
1644 sc->hn_if_flags = ifp->if_flags;
1650 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1651 if (mask & IFCAP_TXCSUM) {
1652 ifp->if_capenable ^= IFCAP_TXCSUM;
1653 if (ifp->if_capenable & IFCAP_TXCSUM) {
1655 sc->hn_tx_ring[0].hn_csum_assist;
1658 ~sc->hn_tx_ring[0].hn_csum_assist;
1662 if (mask & IFCAP_RXCSUM)
1663 ifp->if_capenable ^= IFCAP_RXCSUM;
1665 if (mask & IFCAP_LRO)
1666 ifp->if_capenable ^= IFCAP_LRO;
1668 if (mask & IFCAP_TSO4) {
1669 ifp->if_capenable ^= IFCAP_TSO4;
1670 if (ifp->if_capenable & IFCAP_TSO4)
1671 ifp->if_hwassist |= CSUM_IP_TSO;
1673 ifp->if_hwassist &= ~CSUM_IP_TSO;
1676 if (mask & IFCAP_TSO6) {
1677 ifp->if_capenable ^= IFCAP_TSO6;
1678 if (ifp->if_capenable & IFCAP_TSO6)
1679 ifp->if_hwassist |= CSUM_IP6_TSO;
1681 ifp->if_hwassist &= ~CSUM_IP6_TSO;
1690 /* Fixme: Multicast mode? */
1691 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1693 netvsc_setmulti(sc);
1702 error = ifmedia_ioctl(ifp, ifr, &sc->hn_media, cmd);
1705 error = ether_ioctl(ifp, cmd, data);
1716 hn_stop(hn_softc_t *sc)
1720 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1725 printf(" Closing Device ...\n");
1727 atomic_clear_int(&ifp->if_drv_flags,
1728 (IFF_DRV_RUNNING | IFF_DRV_OACTIVE));
1729 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
1730 sc->hn_tx_ring[i].hn_oactive = 0;
1732 if_link_state_change(ifp, LINK_STATE_DOWN);
1733 sc->hn_initdone = 0;
1735 ret = hv_rf_on_close(device_ctx);
1739 * FreeBSD transmit entry point
1742 hn_start(struct ifnet *ifp)
1744 struct hn_softc *sc = ifp->if_softc;
1745 struct hn_tx_ring *txr = &sc->hn_tx_ring[0];
1747 if (txr->hn_sched_tx)
1750 if (mtx_trylock(&txr->hn_tx_lock)) {
1753 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1754 mtx_unlock(&txr->hn_tx_lock);
1759 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
1763 hn_start_txeof(struct hn_tx_ring *txr)
1765 struct hn_softc *sc = txr->hn_sc;
1766 struct ifnet *ifp = sc->hn_ifp;
1768 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1770 if (txr->hn_sched_tx)
1773 if (mtx_trylock(&txr->hn_tx_lock)) {
1776 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1777 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1778 mtx_unlock(&txr->hn_tx_lock);
1780 taskqueue_enqueue(txr->hn_tx_taskq,
1786 * Release the OACTIVE earlier, with the hope, that
1787 * others could catch up. The task will clear the
1788 * flag again with the hn_tx_lock to avoid possible
1791 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1792 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
1800 hn_ifinit_locked(hn_softc_t *sc)
1803 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1808 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1812 hv_promisc_mode = 1;
1814 ret = hv_rf_on_open(device_ctx);
1818 sc->hn_initdone = 1;
1821 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1822 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
1823 sc->hn_tx_ring[i].hn_oactive = 0;
1825 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_RUNNING);
1826 if_link_state_change(ifp, LINK_STATE_UP);
1833 hn_ifinit(void *xsc)
1835 hn_softc_t *sc = xsc;
1838 if (sc->temp_unusable) {
1842 sc->temp_unusable = TRUE;
1845 hn_ifinit_locked(sc);
1848 sc->temp_unusable = FALSE;
1857 hn_watchdog(struct ifnet *ifp)
1862 printf("hn%d: watchdog timeout -- resetting\n", sc->hn_unit);
1863 hn_ifinit(sc); /*???*/
1868 #if __FreeBSD_version >= 1100099
1871 hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS)
1873 struct hn_softc *sc = arg1;
1874 unsigned int lenlim;
1877 lenlim = sc->hn_rx_ring[0].hn_lro.lro_length_lim;
1878 error = sysctl_handle_int(oidp, &lenlim, 0, req);
1879 if (error || req->newptr == NULL)
1882 if (lenlim < HN_LRO_LENLIM_MIN(sc->hn_ifp) ||
1883 lenlim > TCP_LRO_LENGTH_MAX)
1887 hn_set_lro_lenlim(sc, lenlim);
1893 hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS)
1895 struct hn_softc *sc = arg1;
1896 int ackcnt, error, i;
1899 * lro_ackcnt_lim is append count limit,
1900 * +1 to turn it into aggregation limit.
1902 ackcnt = sc->hn_rx_ring[0].hn_lro.lro_ackcnt_lim + 1;
1903 error = sysctl_handle_int(oidp, &ackcnt, 0, req);
1904 if (error || req->newptr == NULL)
1907 if (ackcnt < 2 || ackcnt > (TCP_LRO_ACKCNT_MAX + 1))
1911 * Convert aggregation limit back to append
1916 for (i = 0; i < sc->hn_rx_ring_inuse; ++i)
1917 sc->hn_rx_ring[i].hn_lro.lro_ackcnt_lim = ackcnt;
1925 hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS)
1927 struct hn_softc *sc = arg1;
1932 if (sc->hn_rx_ring[0].hn_trust_hcsum & hcsum)
1935 error = sysctl_handle_int(oidp, &on, 0, req);
1936 if (error || req->newptr == NULL)
1940 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1941 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
1944 rxr->hn_trust_hcsum |= hcsum;
1946 rxr->hn_trust_hcsum &= ~hcsum;
1953 hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS)
1955 struct hn_softc *sc = arg1;
1956 int chimney_size, error;
1958 chimney_size = sc->hn_tx_ring[0].hn_tx_chimney_size;
1959 error = sysctl_handle_int(oidp, &chimney_size, 0, req);
1960 if (error || req->newptr == NULL)
1963 if (chimney_size > sc->hn_tx_chimney_max || chimney_size <= 0)
1966 hn_set_tx_chimney_size(sc, chimney_size);
1970 #if __FreeBSD_version < 1100095
1972 hn_rx_stat_int_sysctl(SYSCTL_HANDLER_ARGS)
1974 struct hn_softc *sc = arg1;
1975 int ofs = arg2, i, error;
1976 struct hn_rx_ring *rxr;
1980 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1981 rxr = &sc->hn_rx_ring[i];
1982 stat += *((int *)((uint8_t *)rxr + ofs));
1985 error = sysctl_handle_64(oidp, &stat, 0, req);
1986 if (error || req->newptr == NULL)
1989 /* Zero out this stat. */
1990 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1991 rxr = &sc->hn_rx_ring[i];
1992 *((int *)((uint8_t *)rxr + ofs)) = 0;
1998 hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS)
2000 struct hn_softc *sc = arg1;
2001 int ofs = arg2, i, error;
2002 struct hn_rx_ring *rxr;
2006 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2007 rxr = &sc->hn_rx_ring[i];
2008 stat += *((uint64_t *)((uint8_t *)rxr + ofs));
2011 error = sysctl_handle_64(oidp, &stat, 0, req);
2012 if (error || req->newptr == NULL)
2015 /* Zero out this stat. */
2016 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2017 rxr = &sc->hn_rx_ring[i];
2018 *((uint64_t *)((uint8_t *)rxr + ofs)) = 0;
2026 hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
2028 struct hn_softc *sc = arg1;
2029 int ofs = arg2, i, error;
2030 struct hn_rx_ring *rxr;
2034 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2035 rxr = &sc->hn_rx_ring[i];
2036 stat += *((u_long *)((uint8_t *)rxr + ofs));
2039 error = sysctl_handle_long(oidp, &stat, 0, req);
2040 if (error || req->newptr == NULL)
2043 /* Zero out this stat. */
2044 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2045 rxr = &sc->hn_rx_ring[i];
2046 *((u_long *)((uint8_t *)rxr + ofs)) = 0;
2052 hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
2054 struct hn_softc *sc = arg1;
2055 int ofs = arg2, i, error;
2056 struct hn_tx_ring *txr;
2060 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2061 txr = &sc->hn_tx_ring[i];
2062 stat += *((u_long *)((uint8_t *)txr + ofs));
2065 error = sysctl_handle_long(oidp, &stat, 0, req);
2066 if (error || req->newptr == NULL)
2069 /* Zero out this stat. */
2070 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2071 txr = &sc->hn_tx_ring[i];
2072 *((u_long *)((uint8_t *)txr + ofs)) = 0;
2078 hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS)
2080 struct hn_softc *sc = arg1;
2081 int ofs = arg2, i, error, conf;
2082 struct hn_tx_ring *txr;
2084 txr = &sc->hn_tx_ring[0];
2085 conf = *((int *)((uint8_t *)txr + ofs));
2087 error = sysctl_handle_int(oidp, &conf, 0, req);
2088 if (error || req->newptr == NULL)
2092 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2093 txr = &sc->hn_tx_ring[i];
2094 *((int *)((uint8_t *)txr + ofs)) = conf;
2102 hn_check_iplen(const struct mbuf *m, int hoff)
2104 const struct ip *ip;
2105 int len, iphlen, iplen;
2106 const struct tcphdr *th;
2107 int thoff; /* TCP data offset */
2109 len = hoff + sizeof(struct ip);
2111 /* The packet must be at least the size of an IP header. */
2112 if (m->m_pkthdr.len < len)
2113 return IPPROTO_DONE;
2115 /* The fixed IP header must reside completely in the first mbuf. */
2117 return IPPROTO_DONE;
2119 ip = mtodo(m, hoff);
2121 /* Bound check the packet's stated IP header length. */
2122 iphlen = ip->ip_hl << 2;
2123 if (iphlen < sizeof(struct ip)) /* minimum header length */
2124 return IPPROTO_DONE;
2126 /* The full IP header must reside completely in the one mbuf. */
2127 if (m->m_len < hoff + iphlen)
2128 return IPPROTO_DONE;
2130 iplen = ntohs(ip->ip_len);
2133 * Check that the amount of data in the buffers is as
2134 * at least much as the IP header would have us expect.
2136 if (m->m_pkthdr.len < hoff + iplen)
2137 return IPPROTO_DONE;
2140 * Ignore IP fragments.
2142 if (ntohs(ip->ip_off) & (IP_OFFMASK | IP_MF))
2143 return IPPROTO_DONE;
2146 * The TCP/IP or UDP/IP header must be entirely contained within
2147 * the first fragment of a packet.
2151 if (iplen < iphlen + sizeof(struct tcphdr))
2152 return IPPROTO_DONE;
2153 if (m->m_len < hoff + iphlen + sizeof(struct tcphdr))
2154 return IPPROTO_DONE;
2155 th = (const struct tcphdr *)((const uint8_t *)ip + iphlen);
2156 thoff = th->th_off << 2;
2157 if (thoff < sizeof(struct tcphdr) || thoff + iphlen > iplen)
2158 return IPPROTO_DONE;
2159 if (m->m_len < hoff + iphlen + thoff)
2160 return IPPROTO_DONE;
2163 if (iplen < iphlen + sizeof(struct udphdr))
2164 return IPPROTO_DONE;
2165 if (m->m_len < hoff + iphlen + sizeof(struct udphdr))
2166 return IPPROTO_DONE;
2170 return IPPROTO_DONE;
2177 hn_dma_map_paddr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
2179 bus_addr_t *paddr = arg;
2184 KASSERT(nseg == 1, ("too many segments %d!", nseg));
2185 *paddr = segs->ds_addr;
2189 hn_create_rx_data(struct hn_softc *sc, int ring_cnt)
2191 struct sysctl_oid_list *child;
2192 struct sysctl_ctx_list *ctx;
2193 device_t dev = sc->hn_dev;
2194 #if defined(INET) || defined(INET6)
2195 #if __FreeBSD_version >= 1100095
2201 sc->hn_rx_ring_cnt = ring_cnt;
2202 sc->hn_rx_ring_inuse = sc->hn_rx_ring_cnt;
2204 sc->hn_rx_ring = malloc(sizeof(struct hn_rx_ring) * sc->hn_rx_ring_cnt,
2205 M_NETVSC, M_WAITOK | M_ZERO);
2207 #if defined(INET) || defined(INET6)
2208 #if __FreeBSD_version >= 1100095
2209 lroent_cnt = hn_lro_entry_count;
2210 if (lroent_cnt < TCP_LRO_ENTRIES)
2211 lroent_cnt = TCP_LRO_ENTRIES;
2212 device_printf(dev, "LRO: entry count %d\n", lroent_cnt);
2214 #endif /* INET || INET6 */
2216 ctx = device_get_sysctl_ctx(dev);
2217 child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
2219 /* Create dev.hn.UNIT.rx sysctl tree */
2220 sc->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "rx",
2221 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2223 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2224 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
2226 if (hn_trust_hosttcp)
2227 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_TCP;
2228 if (hn_trust_hostudp)
2229 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_UDP;
2230 if (hn_trust_hostip)
2231 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_IP;
2232 rxr->hn_ifp = sc->hn_ifp;
2238 #if defined(INET) || defined(INET6)
2239 #if __FreeBSD_version >= 1100095
2240 tcp_lro_init_args(&rxr->hn_lro, sc->hn_ifp, lroent_cnt, 0);
2242 tcp_lro_init(&rxr->hn_lro);
2243 rxr->hn_lro.ifp = sc->hn_ifp;
2245 #if __FreeBSD_version >= 1100099
2246 rxr->hn_lro.lro_length_lim = HN_LRO_LENLIM_DEF;
2247 rxr->hn_lro.lro_ackcnt_lim = HN_LRO_ACKCNT_DEF;
2249 #endif /* INET || INET6 */
2251 if (sc->hn_rx_sysctl_tree != NULL) {
2255 * Create per RX ring sysctl tree:
2256 * dev.hn.UNIT.rx.RINGID
2258 snprintf(name, sizeof(name), "%d", i);
2259 rxr->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx,
2260 SYSCTL_CHILDREN(sc->hn_rx_sysctl_tree),
2261 OID_AUTO, name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2263 if (rxr->hn_rx_sysctl_tree != NULL) {
2264 SYSCTL_ADD_ULONG(ctx,
2265 SYSCTL_CHILDREN(rxr->hn_rx_sysctl_tree),
2266 OID_AUTO, "packets", CTLFLAG_RW,
2267 &rxr->hn_pkts, "# of packets received");
2272 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_queued",
2273 CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2274 __offsetof(struct hn_rx_ring, hn_lro.lro_queued),
2275 #if __FreeBSD_version < 1100095
2276 hn_rx_stat_int_sysctl,
2278 hn_rx_stat_u64_sysctl,
2280 "LU", "LRO queued");
2281 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_flushed",
2282 CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2283 __offsetof(struct hn_rx_ring, hn_lro.lro_flushed),
2284 #if __FreeBSD_version < 1100095
2285 hn_rx_stat_int_sysctl,
2287 hn_rx_stat_u64_sysctl,
2289 "LU", "LRO flushed");
2290 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_tried",
2291 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2292 __offsetof(struct hn_rx_ring, hn_lro_tried),
2293 hn_rx_stat_ulong_sysctl, "LU", "# of LRO tries");
2294 #if __FreeBSD_version >= 1100099
2295 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_length_lim",
2296 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2297 hn_lro_lenlim_sysctl, "IU",
2298 "Max # of data bytes to be aggregated by LRO");
2299 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_ackcnt_lim",
2300 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2301 hn_lro_ackcnt_sysctl, "I",
2302 "Max # of ACKs to be aggregated by LRO");
2304 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hosttcp",
2305 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_TCP,
2306 hn_trust_hcsum_sysctl, "I",
2307 "Trust tcp segement verification on host side, "
2308 "when csum info is missing");
2309 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostudp",
2310 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_UDP,
2311 hn_trust_hcsum_sysctl, "I",
2312 "Trust udp datagram verification on host side, "
2313 "when csum info is missing");
2314 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostip",
2315 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_IP,
2316 hn_trust_hcsum_sysctl, "I",
2317 "Trust ip packet verification on host side, "
2318 "when csum info is missing");
2319 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_ip",
2320 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2321 __offsetof(struct hn_rx_ring, hn_csum_ip),
2322 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM IP");
2323 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_tcp",
2324 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2325 __offsetof(struct hn_rx_ring, hn_csum_tcp),
2326 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM TCP");
2327 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_udp",
2328 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2329 __offsetof(struct hn_rx_ring, hn_csum_udp),
2330 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM UDP");
2331 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_trusted",
2332 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2333 __offsetof(struct hn_rx_ring, hn_csum_trusted),
2334 hn_rx_stat_ulong_sysctl, "LU",
2335 "# of packets that we trust host's csum verification");
2336 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "small_pkts",
2337 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2338 __offsetof(struct hn_rx_ring, hn_small_pkts),
2339 hn_rx_stat_ulong_sysctl, "LU", "# of small packets received");
2340 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_cnt",
2341 CTLFLAG_RD, &sc->hn_rx_ring_cnt, 0, "# created RX rings");
2342 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_inuse",
2343 CTLFLAG_RD, &sc->hn_rx_ring_inuse, 0, "# used RX rings");
2347 hn_destroy_rx_data(struct hn_softc *sc)
2349 #if defined(INET) || defined(INET6)
2353 if (sc->hn_rx_ring_cnt == 0)
2356 #if defined(INET) || defined(INET6)
2357 for (i = 0; i < sc->hn_rx_ring_cnt; ++i)
2358 tcp_lro_free(&sc->hn_rx_ring[i].hn_lro);
2360 free(sc->hn_rx_ring, M_NETVSC);
2361 sc->hn_rx_ring = NULL;
2363 sc->hn_rx_ring_cnt = 0;
2364 sc->hn_rx_ring_inuse = 0;
2368 hn_create_tx_ring(struct hn_softc *sc, int id)
2370 struct hn_tx_ring *txr = &sc->hn_tx_ring[id];
2371 bus_dma_tag_t parent_dtag;
2375 txr->hn_tx_idx = id;
2377 #ifndef HN_USE_TXDESC_BUFRING
2378 mtx_init(&txr->hn_txlist_spin, "hn txlist", NULL, MTX_SPIN);
2380 mtx_init(&txr->hn_tx_lock, "hn tx", NULL, MTX_DEF);
2382 txr->hn_txdesc_cnt = HN_TX_DESC_CNT;
2383 txr->hn_txdesc = malloc(sizeof(struct hn_txdesc) * txr->hn_txdesc_cnt,
2384 M_NETVSC, M_WAITOK | M_ZERO);
2385 #ifndef HN_USE_TXDESC_BUFRING
2386 SLIST_INIT(&txr->hn_txlist);
2388 txr->hn_txdesc_br = buf_ring_alloc(txr->hn_txdesc_cnt, M_NETVSC,
2389 M_WAITOK, &txr->hn_tx_lock);
2392 txr->hn_tx_taskq = sc->hn_tx_taskq;
2394 if (hn_use_if_start) {
2395 txr->hn_txeof = hn_start_txeof;
2396 TASK_INIT(&txr->hn_tx_task, 0, hn_start_taskfunc, txr);
2397 TASK_INIT(&txr->hn_txeof_task, 0, hn_start_txeof_taskfunc, txr);
2401 txr->hn_txeof = hn_xmit_txeof;
2402 TASK_INIT(&txr->hn_tx_task, 0, hn_xmit_taskfunc, txr);
2403 TASK_INIT(&txr->hn_txeof_task, 0, hn_xmit_txeof_taskfunc, txr);
2405 br_depth = hn_get_txswq_depth(txr);
2406 txr->hn_mbuf_br = buf_ring_alloc(br_depth, M_NETVSC,
2407 M_WAITOK, &txr->hn_tx_lock);
2410 txr->hn_direct_tx_size = hn_direct_tx_size;
2411 if (hv_vmbus_protocal_version >= HV_VMBUS_VERSION_WIN8_1)
2412 txr->hn_csum_assist = HN_CSUM_ASSIST;
2414 txr->hn_csum_assist = HN_CSUM_ASSIST_WIN8;
2417 * Always schedule transmission instead of trying to do direct
2418 * transmission. This one gives the best performance so far.
2420 txr->hn_sched_tx = 1;
2422 parent_dtag = bus_get_dma_tag(sc->hn_dev);
2424 /* DMA tag for RNDIS messages. */
2425 error = bus_dma_tag_create(parent_dtag, /* parent */
2426 HN_RNDIS_MSG_ALIGN, /* alignment */
2427 HN_RNDIS_MSG_BOUNDARY, /* boundary */
2428 BUS_SPACE_MAXADDR, /* lowaddr */
2429 BUS_SPACE_MAXADDR, /* highaddr */
2430 NULL, NULL, /* filter, filterarg */
2431 HN_RNDIS_MSG_LEN, /* maxsize */
2433 HN_RNDIS_MSG_LEN, /* maxsegsize */
2435 NULL, /* lockfunc */
2436 NULL, /* lockfuncarg */
2437 &txr->hn_tx_rndis_dtag);
2439 device_printf(sc->hn_dev, "failed to create rndis dmatag\n");
2443 /* DMA tag for data. */
2444 error = bus_dma_tag_create(parent_dtag, /* parent */
2446 HN_TX_DATA_BOUNDARY, /* boundary */
2447 BUS_SPACE_MAXADDR, /* lowaddr */
2448 BUS_SPACE_MAXADDR, /* highaddr */
2449 NULL, NULL, /* filter, filterarg */
2450 HN_TX_DATA_MAXSIZE, /* maxsize */
2451 HN_TX_DATA_SEGCNT_MAX, /* nsegments */
2452 HN_TX_DATA_SEGSIZE, /* maxsegsize */
2454 NULL, /* lockfunc */
2455 NULL, /* lockfuncarg */
2456 &txr->hn_tx_data_dtag);
2458 device_printf(sc->hn_dev, "failed to create data dmatag\n");
2462 for (i = 0; i < txr->hn_txdesc_cnt; ++i) {
2463 struct hn_txdesc *txd = &txr->hn_txdesc[i];
2468 * Allocate and load RNDIS messages.
2470 error = bus_dmamem_alloc(txr->hn_tx_rndis_dtag,
2471 (void **)&txd->rndis_msg,
2472 BUS_DMA_WAITOK | BUS_DMA_COHERENT,
2473 &txd->rndis_msg_dmap);
2475 device_printf(sc->hn_dev,
2476 "failed to allocate rndis_msg, %d\n", i);
2480 error = bus_dmamap_load(txr->hn_tx_rndis_dtag,
2481 txd->rndis_msg_dmap,
2482 txd->rndis_msg, HN_RNDIS_MSG_LEN,
2483 hn_dma_map_paddr, &txd->rndis_msg_paddr,
2486 device_printf(sc->hn_dev,
2487 "failed to load rndis_msg, %d\n", i);
2488 bus_dmamem_free(txr->hn_tx_rndis_dtag,
2489 txd->rndis_msg, txd->rndis_msg_dmap);
2493 /* DMA map for TX data. */
2494 error = bus_dmamap_create(txr->hn_tx_data_dtag, 0,
2497 device_printf(sc->hn_dev,
2498 "failed to allocate tx data dmamap\n");
2499 bus_dmamap_unload(txr->hn_tx_rndis_dtag,
2500 txd->rndis_msg_dmap);
2501 bus_dmamem_free(txr->hn_tx_rndis_dtag,
2502 txd->rndis_msg, txd->rndis_msg_dmap);
2506 /* All set, put it to list */
2507 txd->flags |= HN_TXD_FLAG_ONLIST;
2508 #ifndef HN_USE_TXDESC_BUFRING
2509 SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
2511 buf_ring_enqueue(txr->hn_txdesc_br, txd);
2514 txr->hn_txdesc_avail = txr->hn_txdesc_cnt;
2516 if (sc->hn_tx_sysctl_tree != NULL) {
2517 struct sysctl_oid_list *child;
2518 struct sysctl_ctx_list *ctx;
2522 * Create per TX ring sysctl tree:
2523 * dev.hn.UNIT.tx.RINGID
2525 ctx = device_get_sysctl_ctx(sc->hn_dev);
2526 child = SYSCTL_CHILDREN(sc->hn_tx_sysctl_tree);
2528 snprintf(name, sizeof(name), "%d", id);
2529 txr->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO,
2530 name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2532 if (txr->hn_tx_sysctl_tree != NULL) {
2533 child = SYSCTL_CHILDREN(txr->hn_tx_sysctl_tree);
2535 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_avail",
2536 CTLFLAG_RD, &txr->hn_txdesc_avail, 0,
2537 "# of available TX descs");
2538 if (!hn_use_if_start) {
2539 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "oactive",
2540 CTLFLAG_RD, &txr->hn_oactive, 0,
2543 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "packets",
2544 CTLFLAG_RW, &txr->hn_pkts,
2545 "# of packets transmitted");
2553 hn_txdesc_dmamap_destroy(struct hn_txdesc *txd)
2555 struct hn_tx_ring *txr = txd->txr;
2557 KASSERT(txd->m == NULL, ("still has mbuf installed"));
2558 KASSERT((txd->flags & HN_TXD_FLAG_DMAMAP) == 0, ("still dma mapped"));
2560 bus_dmamap_unload(txr->hn_tx_rndis_dtag, txd->rndis_msg_dmap);
2561 bus_dmamem_free(txr->hn_tx_rndis_dtag, txd->rndis_msg,
2562 txd->rndis_msg_dmap);
2563 bus_dmamap_destroy(txr->hn_tx_data_dtag, txd->data_dmap);
2567 hn_destroy_tx_ring(struct hn_tx_ring *txr)
2569 struct hn_txdesc *txd;
2571 if (txr->hn_txdesc == NULL)
2574 #ifndef HN_USE_TXDESC_BUFRING
2575 while ((txd = SLIST_FIRST(&txr->hn_txlist)) != NULL) {
2576 SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
2577 hn_txdesc_dmamap_destroy(txd);
2580 mtx_lock(&txr->hn_tx_lock);
2581 while ((txd = buf_ring_dequeue_sc(txr->hn_txdesc_br)) != NULL)
2582 hn_txdesc_dmamap_destroy(txd);
2583 mtx_unlock(&txr->hn_tx_lock);
2586 if (txr->hn_tx_data_dtag != NULL)
2587 bus_dma_tag_destroy(txr->hn_tx_data_dtag);
2588 if (txr->hn_tx_rndis_dtag != NULL)
2589 bus_dma_tag_destroy(txr->hn_tx_rndis_dtag);
2591 #ifdef HN_USE_TXDESC_BUFRING
2592 buf_ring_free(txr->hn_txdesc_br, M_NETVSC);
2595 free(txr->hn_txdesc, M_NETVSC);
2596 txr->hn_txdesc = NULL;
2598 if (txr->hn_mbuf_br != NULL)
2599 buf_ring_free(txr->hn_mbuf_br, M_NETVSC);
2601 #ifndef HN_USE_TXDESC_BUFRING
2602 mtx_destroy(&txr->hn_txlist_spin);
2604 mtx_destroy(&txr->hn_tx_lock);
2608 hn_create_tx_data(struct hn_softc *sc, int ring_cnt)
2610 struct sysctl_oid_list *child;
2611 struct sysctl_ctx_list *ctx;
2614 sc->hn_tx_ring_cnt = ring_cnt;
2615 sc->hn_tx_ring_inuse = sc->hn_tx_ring_cnt;
2617 sc->hn_tx_ring = malloc(sizeof(struct hn_tx_ring) * sc->hn_tx_ring_cnt,
2618 M_NETVSC, M_WAITOK | M_ZERO);
2620 ctx = device_get_sysctl_ctx(sc->hn_dev);
2621 child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->hn_dev));
2623 /* Create dev.hn.UNIT.tx sysctl tree */
2624 sc->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "tx",
2625 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2627 for (i = 0; i < sc->hn_tx_ring_cnt; ++i) {
2630 error = hn_create_tx_ring(sc, i);
2635 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "no_txdescs",
2636 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2637 __offsetof(struct hn_tx_ring, hn_no_txdescs),
2638 hn_tx_stat_ulong_sysctl, "LU", "# of times short of TX descs");
2639 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "send_failed",
2640 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2641 __offsetof(struct hn_tx_ring, hn_send_failed),
2642 hn_tx_stat_ulong_sysctl, "LU", "# of hyper-v sending failure");
2643 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "txdma_failed",
2644 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2645 __offsetof(struct hn_tx_ring, hn_txdma_failed),
2646 hn_tx_stat_ulong_sysctl, "LU", "# of TX DMA failure");
2647 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_collapsed",
2648 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2649 __offsetof(struct hn_tx_ring, hn_tx_collapsed),
2650 hn_tx_stat_ulong_sysctl, "LU", "# of TX mbuf collapsed");
2651 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney",
2652 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2653 __offsetof(struct hn_tx_ring, hn_tx_chimney),
2654 hn_tx_stat_ulong_sysctl, "LU", "# of chimney send");
2655 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_tried",
2656 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2657 __offsetof(struct hn_tx_ring, hn_tx_chimney_tried),
2658 hn_tx_stat_ulong_sysctl, "LU", "# of chimney send tries");
2659 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_cnt",
2660 CTLFLAG_RD, &sc->hn_tx_ring[0].hn_txdesc_cnt, 0,
2661 "# of total TX descs");
2662 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_chimney_max",
2663 CTLFLAG_RD, &sc->hn_tx_chimney_max, 0,
2664 "Chimney send packet size upper boundary");
2665 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_size",
2666 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2667 hn_tx_chimney_size_sysctl,
2668 "I", "Chimney send packet size limit");
2669 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "direct_tx_size",
2670 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2671 __offsetof(struct hn_tx_ring, hn_direct_tx_size),
2672 hn_tx_conf_int_sysctl, "I",
2673 "Size of the packet for direct transmission");
2674 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "sched_tx",
2675 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2676 __offsetof(struct hn_tx_ring, hn_sched_tx),
2677 hn_tx_conf_int_sysctl, "I",
2678 "Always schedule transmission "
2679 "instead of doing direct transmission");
2680 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_cnt",
2681 CTLFLAG_RD, &sc->hn_tx_ring_cnt, 0, "# created TX rings");
2682 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_inuse",
2683 CTLFLAG_RD, &sc->hn_tx_ring_inuse, 0, "# used TX rings");
2689 hn_set_tx_chimney_size(struct hn_softc *sc, int chimney_size)
2694 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
2695 sc->hn_tx_ring[i].hn_tx_chimney_size = chimney_size;
2700 hn_destroy_tx_data(struct hn_softc *sc)
2704 if (sc->hn_tx_ring_cnt == 0)
2707 for (i = 0; i < sc->hn_tx_ring_cnt; ++i)
2708 hn_destroy_tx_ring(&sc->hn_tx_ring[i]);
2710 free(sc->hn_tx_ring, M_NETVSC);
2711 sc->hn_tx_ring = NULL;
2713 sc->hn_tx_ring_cnt = 0;
2714 sc->hn_tx_ring_inuse = 0;
2718 hn_start_taskfunc(void *xtxr, int pending __unused)
2720 struct hn_tx_ring *txr = xtxr;
2722 mtx_lock(&txr->hn_tx_lock);
2723 hn_start_locked(txr, 0);
2724 mtx_unlock(&txr->hn_tx_lock);
2728 hn_start_txeof_taskfunc(void *xtxr, int pending __unused)
2730 struct hn_tx_ring *txr = xtxr;
2732 mtx_lock(&txr->hn_tx_lock);
2733 atomic_clear_int(&txr->hn_sc->hn_ifp->if_drv_flags, IFF_DRV_OACTIVE);
2734 hn_start_locked(txr, 0);
2735 mtx_unlock(&txr->hn_tx_lock);
2739 hn_stop_tx_tasks(struct hn_softc *sc)
2743 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2744 struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
2746 taskqueue_drain(txr->hn_tx_taskq, &txr->hn_tx_task);
2747 taskqueue_drain(txr->hn_tx_taskq, &txr->hn_txeof_task);
2752 hn_xmit(struct hn_tx_ring *txr, int len)
2754 struct hn_softc *sc = txr->hn_sc;
2755 struct ifnet *ifp = sc->hn_ifp;
2756 struct mbuf *m_head;
2758 mtx_assert(&txr->hn_tx_lock, MA_OWNED);
2759 KASSERT(hn_use_if_start == 0,
2760 ("hn_xmit is called, when if_start is enabled"));
2762 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || txr->hn_oactive)
2765 while ((m_head = drbr_peek(ifp, txr->hn_mbuf_br)) != NULL) {
2766 struct hn_txdesc *txd;
2769 if (len > 0 && m_head->m_pkthdr.len > len) {
2771 * This sending could be time consuming; let callers
2772 * dispatch this packet sending (and sending of any
2773 * following up packets) to tx taskqueue.
2775 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2779 txd = hn_txdesc_get(txr);
2781 txr->hn_no_txdescs++;
2782 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2783 txr->hn_oactive = 1;
2787 error = hn_encap(txr, txd, &m_head);
2789 /* Both txd and m_head are freed; discard */
2790 drbr_advance(ifp, txr->hn_mbuf_br);
2794 error = hn_send_pkt(ifp, txr, txd);
2795 if (__predict_false(error)) {
2796 /* txd is freed, but m_head is not */
2797 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2798 txr->hn_oactive = 1;
2803 drbr_advance(ifp, txr->hn_mbuf_br);
2809 hn_transmit(struct ifnet *ifp, struct mbuf *m)
2811 struct hn_softc *sc = ifp->if_softc;
2812 struct hn_tx_ring *txr;
2816 * Select the TX ring based on flowid
2818 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2819 idx = m->m_pkthdr.flowid % sc->hn_tx_ring_inuse;
2820 txr = &sc->hn_tx_ring[idx];
2822 error = drbr_enqueue(ifp, txr->hn_mbuf_br, m);
2824 if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1);
2828 if (txr->hn_oactive)
2831 if (txr->hn_sched_tx)
2834 if (mtx_trylock(&txr->hn_tx_lock)) {
2837 sched = hn_xmit(txr, txr->hn_direct_tx_size);
2838 mtx_unlock(&txr->hn_tx_lock);
2843 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
2848 hn_xmit_qflush(struct ifnet *ifp)
2850 struct hn_softc *sc = ifp->if_softc;
2853 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2854 struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
2857 mtx_lock(&txr->hn_tx_lock);
2858 while ((m = buf_ring_dequeue_sc(txr->hn_mbuf_br)) != NULL)
2860 mtx_unlock(&txr->hn_tx_lock);
2866 hn_xmit_txeof(struct hn_tx_ring *txr)
2869 if (txr->hn_sched_tx)
2872 if (mtx_trylock(&txr->hn_tx_lock)) {
2875 txr->hn_oactive = 0;
2876 sched = hn_xmit(txr, txr->hn_direct_tx_size);
2877 mtx_unlock(&txr->hn_tx_lock);
2879 taskqueue_enqueue(txr->hn_tx_taskq,
2885 * Release the oactive earlier, with the hope, that
2886 * others could catch up. The task will clear the
2887 * oactive again with the hn_tx_lock to avoid possible
2890 txr->hn_oactive = 0;
2891 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
2896 hn_xmit_taskfunc(void *xtxr, int pending __unused)
2898 struct hn_tx_ring *txr = xtxr;
2900 mtx_lock(&txr->hn_tx_lock);
2902 mtx_unlock(&txr->hn_tx_lock);
2906 hn_xmit_txeof_taskfunc(void *xtxr, int pending __unused)
2908 struct hn_tx_ring *txr = xtxr;
2910 mtx_lock(&txr->hn_tx_lock);
2911 txr->hn_oactive = 0;
2913 mtx_unlock(&txr->hn_tx_lock);
2917 hn_channel_attach(struct hn_softc *sc, struct hv_vmbus_channel *chan)
2919 struct hn_rx_ring *rxr;
2922 idx = chan->offer_msg.offer.sub_channel_index;
2924 KASSERT(idx >= 0 && idx < sc->hn_rx_ring_inuse,
2925 ("invalid channel index %d, should > 0 && < %d",
2926 idx, sc->hn_rx_ring_inuse));
2927 rxr = &sc->hn_rx_ring[idx];
2928 KASSERT((rxr->hn_rx_flags & HN_RX_FLAG_ATTACHED) == 0,
2929 ("RX ring %d already attached", idx));
2930 rxr->hn_rx_flags |= HN_RX_FLAG_ATTACHED;
2932 chan->hv_chan_rxr = rxr;
2934 if_printf(sc->hn_ifp, "link RX ring %d to channel%u\n",
2935 idx, chan->offer_msg.child_rel_id);
2938 if (idx < sc->hn_tx_ring_inuse) {
2939 struct hn_tx_ring *txr = &sc->hn_tx_ring[idx];
2941 KASSERT((txr->hn_tx_flags & HN_TX_FLAG_ATTACHED) == 0,
2942 ("TX ring %d already attached", idx));
2943 txr->hn_tx_flags |= HN_TX_FLAG_ATTACHED;
2945 chan->hv_chan_txr = txr;
2946 txr->hn_chan = chan;
2948 if_printf(sc->hn_ifp, "link TX ring %d to channel%u\n",
2949 idx, chan->offer_msg.child_rel_id);
2953 /* Bind channel to a proper CPU */
2954 vmbus_channel_cpu_set(chan, (sc->hn_cpu + idx) % mp_ncpus);
2958 hn_subchan_attach(struct hn_softc *sc, struct hv_vmbus_channel *chan)
2961 KASSERT(!HV_VMBUS_CHAN_ISPRIMARY(chan),
2962 ("subchannel callback on primary channel"));
2963 KASSERT(chan->offer_msg.offer.sub_channel_index > 0,
2964 ("invalid channel subidx %u",
2965 chan->offer_msg.offer.sub_channel_index));
2966 hn_channel_attach(sc, chan);
2970 hn_tx_taskq_create(void *arg __unused)
2972 if (!hn_share_tx_taskq)
2975 hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
2976 taskqueue_thread_enqueue, &hn_tx_taskq);
2977 taskqueue_start_threads(&hn_tx_taskq, 1, PI_NET, "hn tx");
2978 if (hn_bind_tx_taskq >= 0) {
2979 int cpu = hn_bind_tx_taskq;
2980 struct task cpuset_task;
2983 if (cpu > mp_ncpus - 1)
2985 CPU_SETOF(cpu, &cpu_set);
2986 TASK_INIT(&cpuset_task, 0, hn_cpuset_setthread_task, &cpu_set);
2987 taskqueue_enqueue(hn_tx_taskq, &cpuset_task);
2988 taskqueue_drain(hn_tx_taskq, &cpuset_task);
2991 SYSINIT(hn_txtq_create, SI_SUB_DRIVERS, SI_ORDER_FIRST,
2992 hn_tx_taskq_create, NULL);
2995 hn_tx_taskq_destroy(void *arg __unused)
2997 if (hn_tx_taskq != NULL)
2998 taskqueue_free(hn_tx_taskq);
3000 SYSUNINIT(hn_txtq_destroy, SI_SUB_DRIVERS, SI_ORDER_FIRST,
3001 hn_tx_taskq_destroy, NULL);
3003 static device_method_t netvsc_methods[] = {
3004 /* Device interface */
3005 DEVMETHOD(device_probe, netvsc_probe),
3006 DEVMETHOD(device_attach, netvsc_attach),
3007 DEVMETHOD(device_detach, netvsc_detach),
3008 DEVMETHOD(device_shutdown, netvsc_shutdown),
3013 static driver_t netvsc_driver = {
3019 static devclass_t netvsc_devclass;
3021 DRIVER_MODULE(hn, vmbus, netvsc_driver, netvsc_devclass, 0, 0);
3022 MODULE_VERSION(hn, 1);
3023 MODULE_DEPEND(hn, vmbus, 1, 1, 1);