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>
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
161 #ifndef HN_USE_TXDESC_BUFRING
162 SLIST_ENTRY(hn_txdesc) link;
165 struct hn_tx_ring *txr;
167 uint32_t flags; /* HN_TXD_FLAG_ */
168 netvsc_packet netvsc_pkt; /* XXX to be removed */
170 bus_dmamap_t data_dmap;
172 bus_addr_t rndis_msg_paddr;
173 rndis_msg *rndis_msg;
174 bus_dmamap_t rndis_msg_dmap;
177 #define HN_TXD_FLAG_ONLIST 0x1
178 #define HN_TXD_FLAG_DMAMAP 0x2
181 * Only enable UDP checksum offloading when it is on 2012R2 or
182 * later. UDP checksum offloading doesn't work on earlier
185 #define HN_CSUM_ASSIST_WIN8 (CSUM_IP | CSUM_TCP)
186 #define HN_CSUM_ASSIST (CSUM_IP | CSUM_UDP | CSUM_TCP)
188 #define HN_LRO_LENLIM_MULTIRX_DEF (12 * ETHERMTU)
189 #define HN_LRO_LENLIM_DEF (25 * ETHERMTU)
190 /* YYY 2*MTU is a bit rough, but should be good enough. */
191 #define HN_LRO_LENLIM_MIN(ifp) (2 * (ifp)->if_mtu)
193 #define HN_LRO_ACKCNT_DEF 1
196 * Be aware that this sleepable mutex will exhibit WITNESS errors when
197 * certain TCP and ARP code paths are taken. This appears to be a
198 * well-known condition, as all other drivers checked use a sleeping
199 * mutex to protect their transmit paths.
200 * Also Be aware that mutexes do not play well with semaphores, and there
201 * is a conflicting semaphore in a certain channel code path.
203 #define NV_LOCK_INIT(_sc, _name) \
204 mtx_init(&(_sc)->hn_lock, _name, MTX_NETWORK_LOCK, MTX_DEF)
205 #define NV_LOCK(_sc) mtx_lock(&(_sc)->hn_lock)
206 #define NV_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->hn_lock, MA_OWNED)
207 #define NV_UNLOCK(_sc) mtx_unlock(&(_sc)->hn_lock)
208 #define NV_LOCK_DESTROY(_sc) mtx_destroy(&(_sc)->hn_lock)
215 int hv_promisc_mode = 0; /* normal mode by default */
217 SYSCTL_NODE(_hw, OID_AUTO, hn, CTLFLAG_RD, NULL, "Hyper-V network interface");
219 /* Trust tcp segements verification on host side. */
220 static int hn_trust_hosttcp = 1;
221 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hosttcp, CTLFLAG_RDTUN,
222 &hn_trust_hosttcp, 0,
223 "Trust tcp segement verification on host side, "
224 "when csum info is missing (global setting)");
226 /* Trust udp datagrams verification on host side. */
227 static int hn_trust_hostudp = 1;
228 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostudp, CTLFLAG_RDTUN,
229 &hn_trust_hostudp, 0,
230 "Trust udp datagram verification on host side, "
231 "when csum info is missing (global setting)");
233 /* Trust ip packets verification on host side. */
234 static int hn_trust_hostip = 1;
235 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostip, CTLFLAG_RDTUN,
237 "Trust ip packet verification on host side, "
238 "when csum info is missing (global setting)");
240 /* Limit TSO burst size */
241 static int hn_tso_maxlen = 0;
242 SYSCTL_INT(_hw_hn, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN,
243 &hn_tso_maxlen, 0, "TSO burst limit");
245 /* Limit chimney send size */
246 static int hn_tx_chimney_size = 0;
247 SYSCTL_INT(_hw_hn, OID_AUTO, tx_chimney_size, CTLFLAG_RDTUN,
248 &hn_tx_chimney_size, 0, "Chimney send packet size limit");
250 /* Limit the size of packet for direct transmission */
251 static int hn_direct_tx_size = HN_DIRECT_TX_SIZE_DEF;
252 SYSCTL_INT(_hw_hn, OID_AUTO, direct_tx_size, CTLFLAG_RDTUN,
253 &hn_direct_tx_size, 0, "Size of the packet for direct transmission");
255 #if defined(INET) || defined(INET6)
256 #if __FreeBSD_version >= 1100095
257 static int hn_lro_entry_count = HN_LROENT_CNT_DEF;
258 SYSCTL_INT(_hw_hn, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN,
259 &hn_lro_entry_count, 0, "LRO entry count");
263 static int hn_share_tx_taskq = 0;
264 SYSCTL_INT(_hw_hn, OID_AUTO, share_tx_taskq, CTLFLAG_RDTUN,
265 &hn_share_tx_taskq, 0, "Enable shared TX taskqueue");
267 static struct taskqueue *hn_tx_taskq;
269 #ifndef HN_USE_TXDESC_BUFRING
270 static int hn_use_txdesc_bufring = 0;
272 static int hn_use_txdesc_bufring = 1;
274 SYSCTL_INT(_hw_hn, OID_AUTO, use_txdesc_bufring, CTLFLAG_RD,
275 &hn_use_txdesc_bufring, 0, "Use buf_ring for TX descriptors");
277 static int hn_bind_tx_taskq = -1;
278 SYSCTL_INT(_hw_hn, OID_AUTO, bind_tx_taskq, CTLFLAG_RDTUN,
279 &hn_bind_tx_taskq, 0, "Bind TX taskqueue to the specified cpu");
281 static int hn_use_if_start = 0;
282 SYSCTL_INT(_hw_hn, OID_AUTO, use_if_start, CTLFLAG_RDTUN,
283 &hn_use_if_start, 0, "Use if_start TX method");
285 static int hn_chan_cnt = 0;
286 SYSCTL_INT(_hw_hn, OID_AUTO, chan_cnt, CTLFLAG_RDTUN,
288 "# of channels to use; each channel has one RX ring and one TX ring");
290 static int hn_tx_ring_cnt = 0;
291 SYSCTL_INT(_hw_hn, OID_AUTO, tx_ring_cnt, CTLFLAG_RDTUN,
292 &hn_tx_ring_cnt, 0, "# of TX rings to use");
294 static u_int hn_cpu_index;
297 * Forward declarations
299 static void hn_stop(hn_softc_t *sc);
300 static void hn_ifinit_locked(hn_softc_t *sc);
301 static void hn_ifinit(void *xsc);
302 static int hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data);
303 static int hn_start_locked(struct hn_tx_ring *txr, int len);
304 static void hn_start(struct ifnet *ifp);
305 static void hn_start_txeof(struct hn_tx_ring *);
306 static int hn_ifmedia_upd(struct ifnet *ifp);
307 static void hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr);
308 #if __FreeBSD_version >= 1100099
309 static int hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS);
310 static int hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS);
312 static int hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS);
313 static int hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS);
314 #if __FreeBSD_version < 1100095
315 static int hn_rx_stat_int_sysctl(SYSCTL_HANDLER_ARGS);
317 static int hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS);
319 static int hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS);
320 static int hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS);
321 static int hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS);
322 static int hn_check_iplen(const struct mbuf *, int);
323 static int hn_create_tx_ring(struct hn_softc *, int);
324 static void hn_destroy_tx_ring(struct hn_tx_ring *);
325 static int hn_create_tx_data(struct hn_softc *, int);
326 static void hn_destroy_tx_data(struct hn_softc *);
327 static void hn_start_taskfunc(void *, int);
328 static void hn_start_txeof_taskfunc(void *, int);
329 static void hn_stop_tx_tasks(struct hn_softc *);
330 static int hn_encap(struct hn_tx_ring *, struct hn_txdesc *, struct mbuf **);
331 static void hn_create_rx_data(struct hn_softc *sc, int);
332 static void hn_destroy_rx_data(struct hn_softc *sc);
333 static void hn_set_tx_chimney_size(struct hn_softc *, int);
334 static void hn_channel_attach(struct hn_softc *, struct hv_vmbus_channel *);
336 static int hn_transmit(struct ifnet *, struct mbuf *);
337 static void hn_xmit_qflush(struct ifnet *);
338 static int hn_xmit(struct hn_tx_ring *, int);
339 static void hn_xmit_txeof(struct hn_tx_ring *);
340 static void hn_xmit_taskfunc(void *, int);
341 static void hn_xmit_txeof_taskfunc(void *, int);
343 #if __FreeBSD_version >= 1100099
345 hn_set_lro_lenlim(struct hn_softc *sc, int lenlim)
349 for (i = 0; i < sc->hn_rx_ring_inuse; ++i)
350 sc->hn_rx_ring[i].hn_lro.lro_length_lim = lenlim;
355 hn_ifmedia_upd(struct ifnet *ifp __unused)
362 hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
364 struct hn_softc *sc = ifp->if_softc;
366 ifmr->ifm_status = IFM_AVALID;
367 ifmr->ifm_active = IFM_ETHER;
369 if (!sc->hn_carrier) {
370 ifmr->ifm_active |= IFM_NONE;
373 ifmr->ifm_status |= IFM_ACTIVE;
374 ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
377 /* {F8615163-DF3E-46c5-913F-F2D2F965ED0E} */
378 static const hv_guid g_net_vsc_device_type = {
379 .data = {0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
380 0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E}
384 * Standard probe entry point.
388 netvsc_probe(device_t dev)
392 p = vmbus_get_type(dev);
393 if (!memcmp(p, &g_net_vsc_device_type.data, sizeof(hv_guid))) {
394 device_set_desc(dev, "Synthetic Network Interface");
396 printf("Netvsc probe... DONE \n");
398 return (BUS_PROBE_DEFAULT);
405 hn_cpuset_setthread_task(void *xmask, int pending __unused)
407 cpuset_t *mask = xmask;
410 error = cpuset_setthread(curthread->td_tid, mask);
412 panic("curthread=%ju: can't pin; error=%d",
413 (uintmax_t)curthread->td_tid, error);
418 * Standard attach entry point.
420 * Called when the driver is loaded. It allocates needed resources,
421 * and initializes the "hardware" and software.
424 netvsc_attach(device_t dev)
426 struct hv_device *device_ctx = vmbus_get_devctx(dev);
427 struct hv_vmbus_channel *chan;
428 netvsc_device_info device_info;
430 int unit = device_get_unit(dev);
431 struct ifnet *ifp = NULL;
432 int error, ring_cnt, tx_ring_cnt;
435 sc = device_get_softc(dev);
440 bzero(sc, sizeof(hn_softc_t));
444 if (hn_tx_taskq == NULL) {
445 sc->hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
446 taskqueue_thread_enqueue, &sc->hn_tx_taskq);
447 taskqueue_start_threads(&sc->hn_tx_taskq, 1, PI_NET, "%s tx",
448 device_get_nameunit(dev));
449 if (hn_bind_tx_taskq >= 0) {
450 int cpu = hn_bind_tx_taskq;
451 struct task cpuset_task;
454 if (cpu > mp_ncpus - 1)
456 CPU_SETOF(cpu, &cpu_set);
457 TASK_INIT(&cpuset_task, 0, hn_cpuset_setthread_task,
459 taskqueue_enqueue(sc->hn_tx_taskq, &cpuset_task);
460 taskqueue_drain(sc->hn_tx_taskq, &cpuset_task);
463 sc->hn_tx_taskq = hn_tx_taskq;
465 NV_LOCK_INIT(sc, "NetVSCLock");
467 sc->hn_dev_obj = device_ctx;
469 ifp = sc->hn_ifp = sc->arpcom.ac_ifp = if_alloc(IFT_ETHER);
471 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
474 * Figure out the # of RX rings (ring_cnt) and the # of TX rings
475 * to use (tx_ring_cnt).
478 * The # of RX rings to use is same as the # of channels to use.
480 ring_cnt = hn_chan_cnt;
484 if (ring_cnt > HN_RING_CNT_DEF_MAX)
485 ring_cnt = HN_RING_CNT_DEF_MAX;
486 } else if (ring_cnt > mp_ncpus) {
490 tx_ring_cnt = hn_tx_ring_cnt;
491 if (tx_ring_cnt <= 0 || tx_ring_cnt > ring_cnt)
492 tx_ring_cnt = ring_cnt;
493 if (hn_use_if_start) {
494 /* ifnet.if_start only needs one TX ring. */
499 * Set the leader CPU for channels.
501 sc->hn_cpu = atomic_fetchadd_int(&hn_cpu_index, ring_cnt) % mp_ncpus;
503 error = hn_create_tx_data(sc, tx_ring_cnt);
506 hn_create_rx_data(sc, ring_cnt);
509 * Associate the first TX/RX ring w/ the primary channel.
511 chan = device_ctx->channel;
512 KASSERT(HV_VMBUS_CHAN_ISPRIMARY(chan), ("not primary channel"));
513 KASSERT(chan->offer_msg.offer.sub_channel_index == 0,
514 ("primary channel subidx %u",
515 chan->offer_msg.offer.sub_channel_index));
516 hn_channel_attach(sc, chan);
518 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
519 ifp->if_ioctl = hn_ioctl;
520 ifp->if_init = hn_ifinit;
521 /* needed by hv_rf_on_device_add() code */
522 ifp->if_mtu = ETHERMTU;
523 if (hn_use_if_start) {
524 ifp->if_start = hn_start;
525 IFQ_SET_MAXLEN(&ifp->if_snd, 512);
526 ifp->if_snd.ifq_drv_maxlen = 511;
527 IFQ_SET_READY(&ifp->if_snd);
529 ifp->if_transmit = hn_transmit;
530 ifp->if_qflush = hn_xmit_qflush;
533 ifmedia_init(&sc->hn_media, 0, hn_ifmedia_upd, hn_ifmedia_sts);
534 ifmedia_add(&sc->hn_media, IFM_ETHER | IFM_AUTO, 0, NULL);
535 ifmedia_set(&sc->hn_media, IFM_ETHER | IFM_AUTO);
536 /* XXX ifmedia_set really should do this for us */
537 sc->hn_media.ifm_media = sc->hn_media.ifm_cur->ifm_media;
540 * Tell upper layers that we support full VLAN capability.
542 ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
543 ifp->if_capabilities |=
544 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
547 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
549 ifp->if_hwassist = sc->hn_tx_ring[0].hn_csum_assist | CSUM_TSO;
551 error = hv_rf_on_device_add(device_ctx, &device_info, ring_cnt);
554 KASSERT(sc->net_dev->num_channel > 0 &&
555 sc->net_dev->num_channel <= sc->hn_rx_ring_inuse,
556 ("invalid channel count %u, should be less than %d",
557 sc->net_dev->num_channel, sc->hn_rx_ring_inuse));
560 * Set the # of TX/RX rings that could be used according to
561 * the # of channels that host offered.
563 if (sc->hn_tx_ring_inuse > sc->net_dev->num_channel)
564 sc->hn_tx_ring_inuse = sc->net_dev->num_channel;
565 sc->hn_rx_ring_inuse = sc->net_dev->num_channel;
566 device_printf(dev, "%d TX ring, %d RX ring\n",
567 sc->hn_tx_ring_inuse, sc->hn_rx_ring_inuse);
569 #if __FreeBSD_version >= 1100099
570 if (sc->hn_rx_ring_inuse > 1) {
572 * Reduce TCP segment aggregation limit for multiple
573 * RX rings to increase ACK timeliness.
575 hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MULTIRX_DEF);
579 if (device_info.link_state == 0) {
583 tso_maxlen = hn_tso_maxlen;
584 if (tso_maxlen <= 0 || tso_maxlen > IP_MAXPACKET)
585 tso_maxlen = IP_MAXPACKET;
587 ifp->if_hw_tsomaxsegcount = HN_TX_DATA_SEGCNT_MAX;
588 ifp->if_hw_tsomaxsegsize = PAGE_SIZE;
589 ifp->if_hw_tsomax = tso_maxlen -
590 (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
592 ether_ifattach(ifp, device_info.mac_addr);
594 if_printf(ifp, "TSO: %u/%u/%u\n", ifp->if_hw_tsomax,
595 ifp->if_hw_tsomaxsegcount, ifp->if_hw_tsomaxsegsize);
597 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
598 hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max);
599 if (hn_tx_chimney_size > 0 &&
600 hn_tx_chimney_size < sc->hn_tx_chimney_max)
601 hn_set_tx_chimney_size(sc, hn_tx_chimney_size);
605 hn_destroy_tx_data(sc);
612 * Standard detach entry point
615 netvsc_detach(device_t dev)
617 struct hn_softc *sc = device_get_softc(dev);
618 struct hv_device *hv_device = vmbus_get_devctx(dev);
621 printf("netvsc_detach\n");
624 * XXXKYS: Need to clean up all our
625 * driver state; this is the driver
630 * XXXKYS: Need to stop outgoing traffic and unregister
634 hv_rf_on_device_remove(hv_device, HV_RF_NV_DESTROY_CHANNEL);
636 hn_stop_tx_tasks(sc);
638 ifmedia_removeall(&sc->hn_media);
639 hn_destroy_rx_data(sc);
640 hn_destroy_tx_data(sc);
642 if (sc->hn_tx_taskq != hn_tx_taskq)
643 taskqueue_free(sc->hn_tx_taskq);
649 * Standard shutdown entry point
652 netvsc_shutdown(device_t dev)
658 hn_txdesc_dmamap_load(struct hn_tx_ring *txr, struct hn_txdesc *txd,
659 struct mbuf **m_head, bus_dma_segment_t *segs, int *nsegs)
661 struct mbuf *m = *m_head;
664 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag, txd->data_dmap,
665 m, segs, nsegs, BUS_DMA_NOWAIT);
666 if (error == EFBIG) {
669 m_new = m_collapse(m, M_NOWAIT, HN_TX_DATA_SEGCNT_MAX);
674 txr->hn_tx_collapsed++;
676 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag,
677 txd->data_dmap, m, segs, nsegs, BUS_DMA_NOWAIT);
680 bus_dmamap_sync(txr->hn_tx_data_dtag, txd->data_dmap,
681 BUS_DMASYNC_PREWRITE);
682 txd->flags |= HN_TXD_FLAG_DMAMAP;
688 hn_txdesc_dmamap_unload(struct hn_tx_ring *txr, struct hn_txdesc *txd)
691 if (txd->flags & HN_TXD_FLAG_DMAMAP) {
692 bus_dmamap_sync(txr->hn_tx_data_dtag,
693 txd->data_dmap, BUS_DMASYNC_POSTWRITE);
694 bus_dmamap_unload(txr->hn_tx_data_dtag,
696 txd->flags &= ~HN_TXD_FLAG_DMAMAP;
701 hn_txdesc_put(struct hn_tx_ring *txr, struct hn_txdesc *txd)
704 KASSERT((txd->flags & HN_TXD_FLAG_ONLIST) == 0,
705 ("put an onlist txd %#x", txd->flags));
707 KASSERT(txd->refs > 0, ("invalid txd refs %d", txd->refs));
708 if (atomic_fetchadd_int(&txd->refs, -1) != 1)
711 hn_txdesc_dmamap_unload(txr, txd);
712 if (txd->m != NULL) {
717 txd->flags |= HN_TXD_FLAG_ONLIST;
719 #ifndef HN_USE_TXDESC_BUFRING
720 mtx_lock_spin(&txr->hn_txlist_spin);
721 KASSERT(txr->hn_txdesc_avail >= 0 &&
722 txr->hn_txdesc_avail < txr->hn_txdesc_cnt,
723 ("txdesc_put: invalid txd avail %d", txr->hn_txdesc_avail));
724 txr->hn_txdesc_avail++;
725 SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
726 mtx_unlock_spin(&txr->hn_txlist_spin);
728 atomic_add_int(&txr->hn_txdesc_avail, 1);
729 buf_ring_enqueue(txr->hn_txdesc_br, txd);
735 static __inline struct hn_txdesc *
736 hn_txdesc_get(struct hn_tx_ring *txr)
738 struct hn_txdesc *txd;
740 #ifndef HN_USE_TXDESC_BUFRING
741 mtx_lock_spin(&txr->hn_txlist_spin);
742 txd = SLIST_FIRST(&txr->hn_txlist);
744 KASSERT(txr->hn_txdesc_avail > 0,
745 ("txdesc_get: invalid txd avail %d", txr->hn_txdesc_avail));
746 txr->hn_txdesc_avail--;
747 SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
749 mtx_unlock_spin(&txr->hn_txlist_spin);
751 txd = buf_ring_dequeue_sc(txr->hn_txdesc_br);
755 #ifdef HN_USE_TXDESC_BUFRING
756 atomic_subtract_int(&txr->hn_txdesc_avail, 1);
758 KASSERT(txd->m == NULL && txd->refs == 0 &&
759 (txd->flags & HN_TXD_FLAG_ONLIST), ("invalid txd"));
760 txd->flags &= ~HN_TXD_FLAG_ONLIST;
767 hn_txdesc_hold(struct hn_txdesc *txd)
770 /* 0->1 transition will never work */
771 KASSERT(txd->refs > 0, ("invalid refs %d", txd->refs));
772 atomic_add_int(&txd->refs, 1);
776 hn_tx_done(struct hv_vmbus_channel *chan, void *xpkt)
778 netvsc_packet *packet = xpkt;
779 struct hn_txdesc *txd;
780 struct hn_tx_ring *txr;
782 txd = (struct hn_txdesc *)(uintptr_t)
783 packet->compl.send.send_completion_tid;
786 KASSERT(txr->hn_chan == chan,
787 ("channel mismatch, on channel%u, should be channel%u",
788 chan->offer_msg.offer.sub_channel_index,
789 txr->hn_chan->offer_msg.offer.sub_channel_index));
791 txr->hn_has_txeof = 1;
792 hn_txdesc_put(txr, txd);
796 netvsc_channel_rollup(struct hv_vmbus_channel *chan)
798 struct hn_tx_ring *txr = chan->hv_chan_txr;
799 #if defined(INET) || defined(INET6)
800 struct hn_rx_ring *rxr = chan->hv_chan_rxr;
801 struct lro_ctrl *lro = &rxr->hn_lro;
802 struct lro_entry *queued;
804 while ((queued = SLIST_FIRST(&lro->lro_active)) != NULL) {
805 SLIST_REMOVE_HEAD(&lro->lro_active, next);
806 tcp_lro_flush(lro, queued);
812 * 'txr' could be NULL, if multiple channels and
813 * ifnet.if_start method are enabled.
815 if (txr == NULL || !txr->hn_has_txeof)
818 txr->hn_has_txeof = 0;
824 * If this function fails, then both txd and m_head0 will be freed.
827 hn_encap(struct hn_tx_ring *txr, struct hn_txdesc *txd, struct mbuf **m_head0)
829 bus_dma_segment_t segs[HN_TX_DATA_SEGCNT_MAX];
831 struct mbuf *m_head = *m_head0;
832 netvsc_packet *packet;
833 rndis_msg *rndis_mesg;
834 rndis_packet *rndis_pkt;
835 rndis_per_packet_info *rppi;
836 struct ndis_hash_info *hash_info;
837 uint32_t rndis_msg_size;
839 packet = &txd->netvsc_pkt;
840 packet->is_data_pkt = TRUE;
841 packet->tot_data_buf_len = m_head->m_pkthdr.len;
844 * extension points to the area reserved for the
845 * rndis_filter_packet, which is placed just after
846 * the netvsc_packet (and rppi struct, if present;
847 * length is updated later).
849 rndis_mesg = txd->rndis_msg;
850 /* XXX not necessary */
851 memset(rndis_mesg, 0, HN_RNDIS_MSG_LEN);
852 rndis_mesg->ndis_msg_type = REMOTE_NDIS_PACKET_MSG;
854 rndis_pkt = &rndis_mesg->msg.packet;
855 rndis_pkt->data_offset = sizeof(rndis_packet);
856 rndis_pkt->data_length = packet->tot_data_buf_len;
857 rndis_pkt->per_pkt_info_offset = sizeof(rndis_packet);
859 rndis_msg_size = RNDIS_MESSAGE_SIZE(rndis_packet);
862 * Set the hash info for this packet, so that the host could
863 * dispatch the TX done event for this packet back to this TX
866 rndis_msg_size += RNDIS_HASH_PPI_SIZE;
867 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_HASH_PPI_SIZE,
869 hash_info = (struct ndis_hash_info *)((uint8_t *)rppi +
870 rppi->per_packet_info_offset);
871 hash_info->hash = txr->hn_tx_idx;
873 if (m_head->m_flags & M_VLANTAG) {
874 ndis_8021q_info *rppi_vlan_info;
876 rndis_msg_size += RNDIS_VLAN_PPI_SIZE;
877 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_VLAN_PPI_SIZE,
880 rppi_vlan_info = (ndis_8021q_info *)((uint8_t *)rppi +
881 rppi->per_packet_info_offset);
882 rppi_vlan_info->u1.s1.vlan_id =
883 m_head->m_pkthdr.ether_vtag & 0xfff;
886 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
887 rndis_tcp_tso_info *tso_info;
888 struct ether_vlan_header *eh;
892 * XXX need m_pullup and use mtodo
894 eh = mtod(m_head, struct ether_vlan_header*);
895 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN))
896 ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
898 ether_len = ETHER_HDR_LEN;
900 rndis_msg_size += RNDIS_TSO_PPI_SIZE;
901 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_TSO_PPI_SIZE,
902 tcp_large_send_info);
904 tso_info = (rndis_tcp_tso_info *)((uint8_t *)rppi +
905 rppi->per_packet_info_offset);
906 tso_info->lso_v2_xmit.type =
907 RNDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
910 if (m_head->m_pkthdr.csum_flags & CSUM_IP_TSO) {
912 (struct ip *)(m_head->m_data + ether_len);
913 unsigned long iph_len = ip->ip_hl << 2;
915 (struct tcphdr *)((caddr_t)ip + iph_len);
917 tso_info->lso_v2_xmit.ip_version =
918 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
922 th->th_sum = in_pseudo(ip->ip_src.s_addr,
923 ip->ip_dst.s_addr, htons(IPPROTO_TCP));
926 #if defined(INET6) && defined(INET)
931 struct ip6_hdr *ip6 = (struct ip6_hdr *)
932 (m_head->m_data + ether_len);
933 struct tcphdr *th = (struct tcphdr *)(ip6 + 1);
935 tso_info->lso_v2_xmit.ip_version =
936 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
938 th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
941 tso_info->lso_v2_xmit.tcp_header_offset = 0;
942 tso_info->lso_v2_xmit.mss = m_head->m_pkthdr.tso_segsz;
943 } else if (m_head->m_pkthdr.csum_flags & txr->hn_csum_assist) {
944 rndis_tcp_ip_csum_info *csum_info;
946 rndis_msg_size += RNDIS_CSUM_PPI_SIZE;
947 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_CSUM_PPI_SIZE,
949 csum_info = (rndis_tcp_ip_csum_info *)((uint8_t *)rppi +
950 rppi->per_packet_info_offset);
952 csum_info->xmit.is_ipv4 = 1;
953 if (m_head->m_pkthdr.csum_flags & CSUM_IP)
954 csum_info->xmit.ip_header_csum = 1;
956 if (m_head->m_pkthdr.csum_flags & CSUM_TCP) {
957 csum_info->xmit.tcp_csum = 1;
958 csum_info->xmit.tcp_header_offset = 0;
959 } else if (m_head->m_pkthdr.csum_flags & CSUM_UDP) {
960 csum_info->xmit.udp_csum = 1;
964 rndis_mesg->msg_len = packet->tot_data_buf_len + rndis_msg_size;
965 packet->tot_data_buf_len = rndis_mesg->msg_len;
968 * Chimney send, if the packet could fit into one chimney buffer.
970 * TODO: vRSS, chimney buffer should be per-channel.
972 if (packet->tot_data_buf_len < txr->hn_tx_chimney_size) {
973 netvsc_dev *net_dev = txr->hn_sc->net_dev;
974 uint32_t send_buf_section_idx;
976 send_buf_section_idx =
977 hv_nv_get_next_send_section(net_dev);
978 if (send_buf_section_idx !=
979 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX) {
980 uint8_t *dest = ((uint8_t *)net_dev->send_buf +
981 (send_buf_section_idx *
982 net_dev->send_section_size));
984 memcpy(dest, rndis_mesg, rndis_msg_size);
985 dest += rndis_msg_size;
986 m_copydata(m_head, 0, m_head->m_pkthdr.len, dest);
988 packet->send_buf_section_idx = send_buf_section_idx;
989 packet->send_buf_section_size =
990 packet->tot_data_buf_len;
991 packet->page_buf_count = 0;
992 txr->hn_tx_chimney++;
997 error = hn_txdesc_dmamap_load(txr, txd, &m_head, segs, &nsegs);
1002 * This mbuf is not linked w/ the txd yet, so free it now.
1007 freed = hn_txdesc_put(txr, txd);
1009 ("fail to free txd upon txdma error"));
1011 txr->hn_txdma_failed++;
1012 if_inc_counter(txr->hn_sc->hn_ifp, IFCOUNTER_OERRORS, 1);
1017 packet->page_buf_count = nsegs + HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
1019 /* send packet with page buffer */
1020 packet->page_buffers[0].pfn = atop(txd->rndis_msg_paddr);
1021 packet->page_buffers[0].offset = txd->rndis_msg_paddr & PAGE_MASK;
1022 packet->page_buffers[0].length = rndis_msg_size;
1025 * Fill the page buffers with mbuf info starting at index
1026 * HV_RF_NUM_TX_RESERVED_PAGE_BUFS.
1028 for (i = 0; i < nsegs; ++i) {
1029 hv_vmbus_page_buffer *pb = &packet->page_buffers[
1030 i + HV_RF_NUM_TX_RESERVED_PAGE_BUFS];
1032 pb->pfn = atop(segs[i].ds_addr);
1033 pb->offset = segs[i].ds_addr & PAGE_MASK;
1034 pb->length = segs[i].ds_len;
1037 packet->send_buf_section_idx =
1038 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX;
1039 packet->send_buf_section_size = 0;
1043 /* Set the completion routine */
1044 packet->compl.send.on_send_completion = hn_tx_done;
1045 packet->compl.send.send_completion_context = packet;
1046 packet->compl.send.send_completion_tid = (uint64_t)(uintptr_t)txd;
1053 * If this function fails, then txd will be freed, but the mbuf
1054 * associated w/ the txd will _not_ be freed.
1057 hn_send_pkt(struct ifnet *ifp, struct hn_tx_ring *txr, struct hn_txdesc *txd)
1059 int error, send_failed = 0;
1063 * Make sure that txd is not freed before ETHER_BPF_MTAP.
1065 hn_txdesc_hold(txd);
1066 error = hv_nv_on_send(txr->hn_chan, &txd->netvsc_pkt);
1068 ETHER_BPF_MTAP(ifp, txd->m);
1069 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1070 if (!hn_use_if_start) {
1071 if_inc_counter(ifp, IFCOUNTER_OBYTES,
1072 txd->m->m_pkthdr.len);
1073 if (txd->m->m_flags & M_MCAST)
1074 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
1078 hn_txdesc_put(txr, txd);
1080 if (__predict_false(error)) {
1084 * This should "really rarely" happen.
1086 * XXX Too many RX to be acked or too many sideband
1087 * commands to run? Ask netvsc_channel_rollup()
1088 * to kick start later.
1090 txr->hn_has_txeof = 1;
1092 txr->hn_send_failed++;
1095 * Try sending again after set hn_has_txeof;
1096 * in case that we missed the last
1097 * netvsc_channel_rollup().
1101 if_printf(ifp, "send failed\n");
1104 * Caller will perform further processing on the
1105 * associated mbuf, so don't free it in hn_txdesc_put();
1106 * only unload it from the DMA map in hn_txdesc_put(),
1110 freed = hn_txdesc_put(txr, txd);
1112 ("fail to free txd upon send error"));
1114 txr->hn_send_failed++;
1120 * Start a transmit of one or more packets
1123 hn_start_locked(struct hn_tx_ring *txr, int len)
1125 struct hn_softc *sc = txr->hn_sc;
1126 struct ifnet *ifp = sc->hn_ifp;
1128 KASSERT(hn_use_if_start,
1129 ("hn_start_locked is called, when if_start is disabled"));
1130 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1131 mtx_assert(&txr->hn_tx_lock, MA_OWNED);
1133 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1137 while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
1138 struct hn_txdesc *txd;
1139 struct mbuf *m_head;
1142 IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
1146 if (len > 0 && m_head->m_pkthdr.len > len) {
1148 * This sending could be time consuming; let callers
1149 * dispatch this packet sending (and sending of any
1150 * following up packets) to tx taskqueue.
1152 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1156 txd = hn_txdesc_get(txr);
1158 txr->hn_no_txdescs++;
1159 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1160 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1164 error = hn_encap(txr, txd, &m_head);
1166 /* Both txd and m_head are freed */
1170 error = hn_send_pkt(ifp, txr, txd);
1171 if (__predict_false(error)) {
1172 /* txd is freed, but m_head is not */
1173 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1174 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1182 * Link up/down notification
1185 netvsc_linkstatus_callback(struct hv_device *device_obj, uint32_t status)
1187 hn_softc_t *sc = device_get_softc(device_obj->device);
1201 * Append the specified data to the indicated mbuf chain,
1202 * Extend the mbuf chain if the new data does not fit in
1205 * This is a minor rewrite of m_append() from sys/kern/uipc_mbuf.c.
1206 * There should be an equivalent in the kernel mbuf code,
1207 * but there does not appear to be one yet.
1209 * Differs from m_append() in that additional mbufs are
1210 * allocated with cluster size MJUMPAGESIZE, and filled
1213 * Return 1 if able to complete the job; otherwise 0.
1216 hv_m_append(struct mbuf *m0, int len, c_caddr_t cp)
1219 int remainder, space;
1221 for (m = m0; m->m_next != NULL; m = m->m_next)
1224 space = M_TRAILINGSPACE(m);
1227 * Copy into available space.
1229 if (space > remainder)
1231 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1236 while (remainder > 0) {
1238 * Allocate a new mbuf; could check space
1239 * and allocate a cluster instead.
1241 n = m_getjcl(M_DONTWAIT, m->m_type, 0, MJUMPAGESIZE);
1244 n->m_len = min(MJUMPAGESIZE, remainder);
1245 bcopy(cp, mtod(n, caddr_t), n->m_len);
1247 remainder -= n->m_len;
1251 if (m0->m_flags & M_PKTHDR)
1252 m0->m_pkthdr.len += len - remainder;
1254 return (remainder == 0);
1259 * Called when we receive a data packet from the "wire" on the
1262 * Note: This is no longer used as a callback
1265 netvsc_recv(struct hv_vmbus_channel *chan, netvsc_packet *packet,
1266 rndis_tcp_ip_csum_info *csum_info)
1268 struct hn_rx_ring *rxr = chan->hv_chan_rxr;
1269 struct ifnet *ifp = rxr->hn_ifp;
1271 int size, do_lro = 0, do_csum = 1;
1273 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1277 * Bail out if packet contains more data than configured MTU.
1279 if (packet->tot_data_buf_len > (ifp->if_mtu + ETHER_HDR_LEN)) {
1281 } else if (packet->tot_data_buf_len <= MHLEN) {
1282 m_new = m_gethdr(M_NOWAIT, MT_DATA);
1283 if (m_new == NULL) {
1284 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1287 memcpy(mtod(m_new, void *), packet->data,
1288 packet->tot_data_buf_len);
1289 m_new->m_pkthdr.len = m_new->m_len = packet->tot_data_buf_len;
1290 rxr->hn_small_pkts++;
1293 * Get an mbuf with a cluster. For packets 2K or less,
1294 * get a standard 2K cluster. For anything larger, get a
1295 * 4K cluster. Any buffers larger than 4K can cause problems
1296 * if looped around to the Hyper-V TX channel, so avoid them.
1299 if (packet->tot_data_buf_len > MCLBYTES) {
1301 size = MJUMPAGESIZE;
1304 m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, size);
1305 if (m_new == NULL) {
1306 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1310 hv_m_append(m_new, packet->tot_data_buf_len, packet->data);
1312 m_new->m_pkthdr.rcvif = ifp;
1314 if (__predict_false((ifp->if_capenable & IFCAP_RXCSUM) == 0))
1317 /* receive side checksum offload */
1318 if (csum_info != NULL) {
1319 /* IP csum offload */
1320 if (csum_info->receive.ip_csum_succeeded && do_csum) {
1321 m_new->m_pkthdr.csum_flags |=
1322 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1326 /* TCP/UDP csum offload */
1327 if ((csum_info->receive.tcp_csum_succeeded ||
1328 csum_info->receive.udp_csum_succeeded) && do_csum) {
1329 m_new->m_pkthdr.csum_flags |=
1330 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1331 m_new->m_pkthdr.csum_data = 0xffff;
1332 if (csum_info->receive.tcp_csum_succeeded)
1338 if (csum_info->receive.ip_csum_succeeded &&
1339 csum_info->receive.tcp_csum_succeeded)
1342 const struct ether_header *eh;
1347 if (m_new->m_len < hoff)
1349 eh = mtod(m_new, struct ether_header *);
1350 etype = ntohs(eh->ether_type);
1351 if (etype == ETHERTYPE_VLAN) {
1352 const struct ether_vlan_header *evl;
1354 hoff = sizeof(*evl);
1355 if (m_new->m_len < hoff)
1357 evl = mtod(m_new, struct ether_vlan_header *);
1358 etype = ntohs(evl->evl_proto);
1361 if (etype == ETHERTYPE_IP) {
1364 pr = hn_check_iplen(m_new, hoff);
1365 if (pr == IPPROTO_TCP) {
1367 (rxr->hn_trust_hcsum &
1368 HN_TRUST_HCSUM_TCP)) {
1369 rxr->hn_csum_trusted++;
1370 m_new->m_pkthdr.csum_flags |=
1371 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1372 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1373 m_new->m_pkthdr.csum_data = 0xffff;
1375 /* Rely on SW csum verification though... */
1377 } else if (pr == IPPROTO_UDP) {
1379 (rxr->hn_trust_hcsum &
1380 HN_TRUST_HCSUM_UDP)) {
1381 rxr->hn_csum_trusted++;
1382 m_new->m_pkthdr.csum_flags |=
1383 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1384 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1385 m_new->m_pkthdr.csum_data = 0xffff;
1387 } else if (pr != IPPROTO_DONE && do_csum &&
1388 (rxr->hn_trust_hcsum & HN_TRUST_HCSUM_IP)) {
1389 rxr->hn_csum_trusted++;
1390 m_new->m_pkthdr.csum_flags |=
1391 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1396 if ((packet->vlan_tci != 0) &&
1397 (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0) {
1398 m_new->m_pkthdr.ether_vtag = packet->vlan_tci;
1399 m_new->m_flags |= M_VLANTAG;
1402 m_new->m_pkthdr.flowid = rxr->hn_rx_idx;
1403 M_HASHTYPE_SET(m_new, M_HASHTYPE_OPAQUE);
1406 * Note: Moved RX completion back to hv_nv_on_receive() so all
1407 * messages (not just data messages) will trigger a response.
1413 if ((ifp->if_capenable & IFCAP_LRO) && do_lro) {
1414 #if defined(INET) || defined(INET6)
1415 struct lro_ctrl *lro = &rxr->hn_lro;
1418 rxr->hn_lro_tried++;
1419 if (tcp_lro_rx(lro, m_new, 0) == 0) {
1427 /* We're not holding the lock here, so don't release it */
1428 (*ifp->if_input)(ifp, m_new);
1434 * Rules for using sc->temp_unusable:
1435 * 1. sc->temp_unusable can only be read or written while holding NV_LOCK()
1436 * 2. code reading sc->temp_unusable under NV_LOCK(), and finding
1437 * sc->temp_unusable set, must release NV_LOCK() and exit
1438 * 3. to retain exclusive control of the interface,
1439 * sc->temp_unusable must be set by code before releasing NV_LOCK()
1440 * 4. only code setting sc->temp_unusable can clear sc->temp_unusable
1441 * 5. code setting sc->temp_unusable must eventually clear sc->temp_unusable
1445 * Standard ioctl entry point. Called when the user wants to configure
1449 hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1451 hn_softc_t *sc = ifp->if_softc;
1452 struct ifreq *ifr = (struct ifreq *)data;
1454 struct ifaddr *ifa = (struct ifaddr *)data;
1456 netvsc_device_info device_info;
1457 struct hv_device *hn_dev;
1458 int mask, error = 0;
1459 int retry_cnt = 500;
1465 if (ifa->ifa_addr->sa_family == AF_INET) {
1466 ifp->if_flags |= IFF_UP;
1467 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1469 arp_ifinit(ifp, ifa);
1472 error = ether_ioctl(ifp, cmd, data);
1475 hn_dev = vmbus_get_devctx(sc->hn_dev);
1477 /* Check MTU value change */
1478 if (ifp->if_mtu == ifr->ifr_mtu)
1481 if (ifr->ifr_mtu > NETVSC_MAX_CONFIGURABLE_MTU) {
1486 /* Obtain and record requested MTU */
1487 ifp->if_mtu = ifr->ifr_mtu;
1489 #if __FreeBSD_version >= 1100099
1491 * Make sure that LRO aggregation length limit is still
1492 * valid, after the MTU change.
1495 if (sc->hn_rx_ring[0].hn_lro.lro_length_lim <
1496 HN_LRO_LENLIM_MIN(ifp))
1497 hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MIN(ifp));
1503 if (!sc->temp_unusable) {
1504 sc->temp_unusable = TRUE;
1508 if (retry_cnt > 0) {
1512 } while (retry_cnt > 0);
1514 if (retry_cnt == 0) {
1519 /* We must remove and add back the device to cause the new
1520 * MTU to take effect. This includes tearing down, but not
1521 * deleting the channel, then bringing it back up.
1523 error = hv_rf_on_device_remove(hn_dev, HV_RF_NV_RETAIN_CHANNEL);
1526 sc->temp_unusable = FALSE;
1530 error = hv_rf_on_device_add(hn_dev, &device_info,
1531 sc->hn_rx_ring_inuse);
1534 sc->temp_unusable = FALSE;
1539 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
1540 if (sc->hn_tx_ring[0].hn_tx_chimney_size >
1541 sc->hn_tx_chimney_max)
1542 hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max);
1544 hn_ifinit_locked(sc);
1547 sc->temp_unusable = FALSE;
1553 if (!sc->temp_unusable) {
1554 sc->temp_unusable = TRUE;
1558 if (retry_cnt > 0) {
1562 } while (retry_cnt > 0);
1564 if (retry_cnt == 0) {
1569 if (ifp->if_flags & IFF_UP) {
1571 * If only the state of the PROMISC flag changed,
1572 * then just use the 'set promisc mode' command
1573 * instead of reinitializing the entire NIC. Doing
1574 * a full re-init means reloading the firmware and
1575 * waiting for it to start up, which may take a
1579 /* Fixme: Promiscuous mode? */
1580 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1581 ifp->if_flags & IFF_PROMISC &&
1582 !(sc->hn_if_flags & IFF_PROMISC)) {
1583 /* do something here for Hyper-V */
1584 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1585 !(ifp->if_flags & IFF_PROMISC) &&
1586 sc->hn_if_flags & IFF_PROMISC) {
1587 /* do something here for Hyper-V */
1590 hn_ifinit_locked(sc);
1592 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1597 sc->temp_unusable = FALSE;
1599 sc->hn_if_flags = ifp->if_flags;
1605 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1606 if (mask & IFCAP_TXCSUM) {
1607 ifp->if_capenable ^= IFCAP_TXCSUM;
1608 if (ifp->if_capenable & IFCAP_TXCSUM) {
1610 sc->hn_tx_ring[0].hn_csum_assist;
1613 ~sc->hn_tx_ring[0].hn_csum_assist;
1617 if (mask & IFCAP_RXCSUM)
1618 ifp->if_capenable ^= IFCAP_RXCSUM;
1620 if (mask & IFCAP_LRO)
1621 ifp->if_capenable ^= IFCAP_LRO;
1623 if (mask & IFCAP_TSO4) {
1624 ifp->if_capenable ^= IFCAP_TSO4;
1625 if (ifp->if_capenable & IFCAP_TSO4)
1626 ifp->if_hwassist |= CSUM_IP_TSO;
1628 ifp->if_hwassist &= ~CSUM_IP_TSO;
1631 if (mask & IFCAP_TSO6) {
1632 ifp->if_capenable ^= IFCAP_TSO6;
1633 if (ifp->if_capenable & IFCAP_TSO6)
1634 ifp->if_hwassist |= CSUM_IP6_TSO;
1636 ifp->if_hwassist &= ~CSUM_IP6_TSO;
1645 /* Fixme: Multicast mode? */
1646 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1648 netvsc_setmulti(sc);
1657 error = ifmedia_ioctl(ifp, ifr, &sc->hn_media, cmd);
1660 error = ether_ioctl(ifp, cmd, data);
1671 hn_stop(hn_softc_t *sc)
1675 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1680 printf(" Closing Device ...\n");
1682 atomic_clear_int(&ifp->if_drv_flags,
1683 (IFF_DRV_RUNNING | IFF_DRV_OACTIVE));
1684 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
1685 sc->hn_tx_ring[i].hn_oactive = 0;
1687 if_link_state_change(ifp, LINK_STATE_DOWN);
1688 sc->hn_initdone = 0;
1690 ret = hv_rf_on_close(device_ctx);
1694 * FreeBSD transmit entry point
1697 hn_start(struct ifnet *ifp)
1699 struct hn_softc *sc = ifp->if_softc;
1700 struct hn_tx_ring *txr = &sc->hn_tx_ring[0];
1702 if (txr->hn_sched_tx)
1705 if (mtx_trylock(&txr->hn_tx_lock)) {
1708 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1709 mtx_unlock(&txr->hn_tx_lock);
1714 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
1718 hn_start_txeof(struct hn_tx_ring *txr)
1720 struct hn_softc *sc = txr->hn_sc;
1721 struct ifnet *ifp = sc->hn_ifp;
1723 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1725 if (txr->hn_sched_tx)
1728 if (mtx_trylock(&txr->hn_tx_lock)) {
1731 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1732 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1733 mtx_unlock(&txr->hn_tx_lock);
1735 taskqueue_enqueue(txr->hn_tx_taskq,
1741 * Release the OACTIVE earlier, with the hope, that
1742 * others could catch up. The task will clear the
1743 * flag again with the hn_tx_lock to avoid possible
1746 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1747 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
1755 hn_ifinit_locked(hn_softc_t *sc)
1758 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1763 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1767 hv_promisc_mode = 1;
1769 ret = hv_rf_on_open(device_ctx);
1773 sc->hn_initdone = 1;
1776 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1777 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
1778 sc->hn_tx_ring[i].hn_oactive = 0;
1780 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_RUNNING);
1781 if_link_state_change(ifp, LINK_STATE_UP);
1788 hn_ifinit(void *xsc)
1790 hn_softc_t *sc = xsc;
1793 if (sc->temp_unusable) {
1797 sc->temp_unusable = TRUE;
1800 hn_ifinit_locked(sc);
1803 sc->temp_unusable = FALSE;
1812 hn_watchdog(struct ifnet *ifp)
1817 printf("hn%d: watchdog timeout -- resetting\n", sc->hn_unit);
1818 hn_ifinit(sc); /*???*/
1823 #if __FreeBSD_version >= 1100099
1826 hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS)
1828 struct hn_softc *sc = arg1;
1829 unsigned int lenlim;
1832 lenlim = sc->hn_rx_ring[0].hn_lro.lro_length_lim;
1833 error = sysctl_handle_int(oidp, &lenlim, 0, req);
1834 if (error || req->newptr == NULL)
1837 if (lenlim < HN_LRO_LENLIM_MIN(sc->hn_ifp) ||
1838 lenlim > TCP_LRO_LENGTH_MAX)
1842 hn_set_lro_lenlim(sc, lenlim);
1848 hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS)
1850 struct hn_softc *sc = arg1;
1851 int ackcnt, error, i;
1854 * lro_ackcnt_lim is append count limit,
1855 * +1 to turn it into aggregation limit.
1857 ackcnt = sc->hn_rx_ring[0].hn_lro.lro_ackcnt_lim + 1;
1858 error = sysctl_handle_int(oidp, &ackcnt, 0, req);
1859 if (error || req->newptr == NULL)
1862 if (ackcnt < 2 || ackcnt > (TCP_LRO_ACKCNT_MAX + 1))
1866 * Convert aggregation limit back to append
1871 for (i = 0; i < sc->hn_rx_ring_inuse; ++i)
1872 sc->hn_rx_ring[i].hn_lro.lro_ackcnt_lim = ackcnt;
1880 hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS)
1882 struct hn_softc *sc = arg1;
1887 if (sc->hn_rx_ring[0].hn_trust_hcsum & hcsum)
1890 error = sysctl_handle_int(oidp, &on, 0, req);
1891 if (error || req->newptr == NULL)
1895 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1896 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
1899 rxr->hn_trust_hcsum |= hcsum;
1901 rxr->hn_trust_hcsum &= ~hcsum;
1908 hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS)
1910 struct hn_softc *sc = arg1;
1911 int chimney_size, error;
1913 chimney_size = sc->hn_tx_ring[0].hn_tx_chimney_size;
1914 error = sysctl_handle_int(oidp, &chimney_size, 0, req);
1915 if (error || req->newptr == NULL)
1918 if (chimney_size > sc->hn_tx_chimney_max || chimney_size <= 0)
1921 hn_set_tx_chimney_size(sc, chimney_size);
1925 #if __FreeBSD_version < 1100095
1927 hn_rx_stat_int_sysctl(SYSCTL_HANDLER_ARGS)
1929 struct hn_softc *sc = arg1;
1930 int ofs = arg2, i, error;
1931 struct hn_rx_ring *rxr;
1935 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1936 rxr = &sc->hn_rx_ring[i];
1937 stat += *((int *)((uint8_t *)rxr + ofs));
1940 error = sysctl_handle_64(oidp, &stat, 0, req);
1941 if (error || req->newptr == NULL)
1944 /* Zero out this stat. */
1945 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1946 rxr = &sc->hn_rx_ring[i];
1947 *((int *)((uint8_t *)rxr + ofs)) = 0;
1953 hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS)
1955 struct hn_softc *sc = arg1;
1956 int ofs = arg2, i, error;
1957 struct hn_rx_ring *rxr;
1961 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1962 rxr = &sc->hn_rx_ring[i];
1963 stat += *((uint64_t *)((uint8_t *)rxr + ofs));
1966 error = sysctl_handle_64(oidp, &stat, 0, req);
1967 if (error || req->newptr == NULL)
1970 /* Zero out this stat. */
1971 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1972 rxr = &sc->hn_rx_ring[i];
1973 *((uint64_t *)((uint8_t *)rxr + ofs)) = 0;
1981 hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
1983 struct hn_softc *sc = arg1;
1984 int ofs = arg2, i, error;
1985 struct hn_rx_ring *rxr;
1989 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
1990 rxr = &sc->hn_rx_ring[i];
1991 stat += *((u_long *)((uint8_t *)rxr + ofs));
1994 error = sysctl_handle_long(oidp, &stat, 0, req);
1995 if (error || req->newptr == NULL)
1998 /* Zero out this stat. */
1999 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2000 rxr = &sc->hn_rx_ring[i];
2001 *((u_long *)((uint8_t *)rxr + ofs)) = 0;
2007 hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
2009 struct hn_softc *sc = arg1;
2010 int ofs = arg2, i, error;
2011 struct hn_tx_ring *txr;
2015 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2016 txr = &sc->hn_tx_ring[i];
2017 stat += *((u_long *)((uint8_t *)txr + ofs));
2020 error = sysctl_handle_long(oidp, &stat, 0, req);
2021 if (error || req->newptr == NULL)
2024 /* Zero out this stat. */
2025 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2026 txr = &sc->hn_tx_ring[i];
2027 *((u_long *)((uint8_t *)txr + ofs)) = 0;
2033 hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS)
2035 struct hn_softc *sc = arg1;
2036 int ofs = arg2, i, error, conf;
2037 struct hn_tx_ring *txr;
2039 txr = &sc->hn_tx_ring[0];
2040 conf = *((int *)((uint8_t *)txr + ofs));
2042 error = sysctl_handle_int(oidp, &conf, 0, req);
2043 if (error || req->newptr == NULL)
2047 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2048 txr = &sc->hn_tx_ring[i];
2049 *((int *)((uint8_t *)txr + ofs)) = conf;
2057 hn_check_iplen(const struct mbuf *m, int hoff)
2059 const struct ip *ip;
2060 int len, iphlen, iplen;
2061 const struct tcphdr *th;
2062 int thoff; /* TCP data offset */
2064 len = hoff + sizeof(struct ip);
2066 /* The packet must be at least the size of an IP header. */
2067 if (m->m_pkthdr.len < len)
2068 return IPPROTO_DONE;
2070 /* The fixed IP header must reside completely in the first mbuf. */
2072 return IPPROTO_DONE;
2074 ip = mtodo(m, hoff);
2076 /* Bound check the packet's stated IP header length. */
2077 iphlen = ip->ip_hl << 2;
2078 if (iphlen < sizeof(struct ip)) /* minimum header length */
2079 return IPPROTO_DONE;
2081 /* The full IP header must reside completely in the one mbuf. */
2082 if (m->m_len < hoff + iphlen)
2083 return IPPROTO_DONE;
2085 iplen = ntohs(ip->ip_len);
2088 * Check that the amount of data in the buffers is as
2089 * at least much as the IP header would have us expect.
2091 if (m->m_pkthdr.len < hoff + iplen)
2092 return IPPROTO_DONE;
2095 * Ignore IP fragments.
2097 if (ntohs(ip->ip_off) & (IP_OFFMASK | IP_MF))
2098 return IPPROTO_DONE;
2101 * The TCP/IP or UDP/IP header must be entirely contained within
2102 * the first fragment of a packet.
2106 if (iplen < iphlen + sizeof(struct tcphdr))
2107 return IPPROTO_DONE;
2108 if (m->m_len < hoff + iphlen + sizeof(struct tcphdr))
2109 return IPPROTO_DONE;
2110 th = (const struct tcphdr *)((const uint8_t *)ip + iphlen);
2111 thoff = th->th_off << 2;
2112 if (thoff < sizeof(struct tcphdr) || thoff + iphlen > iplen)
2113 return IPPROTO_DONE;
2114 if (m->m_len < hoff + iphlen + thoff)
2115 return IPPROTO_DONE;
2118 if (iplen < iphlen + sizeof(struct udphdr))
2119 return IPPROTO_DONE;
2120 if (m->m_len < hoff + iphlen + sizeof(struct udphdr))
2121 return IPPROTO_DONE;
2125 return IPPROTO_DONE;
2132 hn_dma_map_paddr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
2134 bus_addr_t *paddr = arg;
2139 KASSERT(nseg == 1, ("too many segments %d!", nseg));
2140 *paddr = segs->ds_addr;
2144 hn_create_rx_data(struct hn_softc *sc, int ring_cnt)
2146 struct sysctl_oid_list *child;
2147 struct sysctl_ctx_list *ctx;
2148 device_t dev = sc->hn_dev;
2149 #if defined(INET) || defined(INET6)
2150 #if __FreeBSD_version >= 1100095
2156 sc->hn_rx_ring_cnt = ring_cnt;
2157 sc->hn_rx_ring_inuse = sc->hn_rx_ring_cnt;
2159 sc->hn_rx_ring = malloc(sizeof(struct hn_rx_ring) * sc->hn_rx_ring_cnt,
2160 M_NETVSC, M_WAITOK | M_ZERO);
2162 #if defined(INET) || defined(INET6)
2163 #if __FreeBSD_version >= 1100095
2164 lroent_cnt = hn_lro_entry_count;
2165 if (lroent_cnt < TCP_LRO_ENTRIES)
2166 lroent_cnt = TCP_LRO_ENTRIES;
2167 device_printf(dev, "LRO: entry count %d\n", lroent_cnt);
2169 #endif /* INET || INET6 */
2171 ctx = device_get_sysctl_ctx(dev);
2172 child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
2174 /* Create dev.hn.UNIT.rx sysctl tree */
2175 sc->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "rx",
2178 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2179 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
2181 if (hn_trust_hosttcp)
2182 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_TCP;
2183 if (hn_trust_hostudp)
2184 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_UDP;
2185 if (hn_trust_hostip)
2186 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_IP;
2187 rxr->hn_ifp = sc->hn_ifp;
2193 #if defined(INET) || defined(INET6)
2194 #if __FreeBSD_version >= 1100095
2195 tcp_lro_init_args(&rxr->hn_lro, sc->hn_ifp, lroent_cnt, 0);
2197 tcp_lro_init(&rxr->hn_lro);
2198 rxr->hn_lro.ifp = sc->hn_ifp;
2200 #if __FreeBSD_version >= 1100099
2201 rxr->hn_lro.lro_length_lim = HN_LRO_LENLIM_DEF;
2202 rxr->hn_lro.lro_ackcnt_lim = HN_LRO_ACKCNT_DEF;
2204 #endif /* INET || INET6 */
2206 if (sc->hn_rx_sysctl_tree != NULL) {
2210 * Create per RX ring sysctl tree:
2211 * dev.hn.UNIT.rx.RINGID
2213 snprintf(name, sizeof(name), "%d", i);
2214 rxr->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx,
2215 SYSCTL_CHILDREN(sc->hn_rx_sysctl_tree),
2216 OID_AUTO, name, CTLFLAG_RD, 0, "");
2218 if (rxr->hn_rx_sysctl_tree != NULL) {
2219 SYSCTL_ADD_ULONG(ctx,
2220 SYSCTL_CHILDREN(rxr->hn_rx_sysctl_tree),
2221 OID_AUTO, "packets", CTLFLAG_RW,
2222 &rxr->hn_pkts, "# of packets received");
2227 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_queued",
2228 CTLTYPE_U64 | CTLFLAG_RW, sc,
2229 __offsetof(struct hn_rx_ring, hn_lro.lro_queued),
2230 #if __FreeBSD_version < 1100095
2231 hn_rx_stat_int_sysctl,
2233 hn_rx_stat_u64_sysctl,
2235 "LU", "LRO queued");
2236 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_flushed",
2237 CTLTYPE_U64 | CTLFLAG_RW, sc,
2238 __offsetof(struct hn_rx_ring, hn_lro.lro_flushed),
2239 #if __FreeBSD_version < 1100095
2240 hn_rx_stat_int_sysctl,
2242 hn_rx_stat_u64_sysctl,
2244 "LU", "LRO flushed");
2245 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_tried",
2246 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2247 __offsetof(struct hn_rx_ring, hn_lro_tried),
2248 hn_rx_stat_ulong_sysctl, "LU", "# of LRO tries");
2249 #if __FreeBSD_version >= 1100099
2250 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_length_lim",
2251 CTLTYPE_UINT | CTLFLAG_RW, sc, 0, hn_lro_lenlim_sysctl, "IU",
2252 "Max # of data bytes to be aggregated by LRO");
2253 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_ackcnt_lim",
2254 CTLTYPE_INT | CTLFLAG_RW, sc, 0, hn_lro_ackcnt_sysctl, "I",
2255 "Max # of ACKs to be aggregated by LRO");
2257 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hosttcp",
2258 CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_TCP,
2259 hn_trust_hcsum_sysctl, "I",
2260 "Trust tcp segement verification on host side, "
2261 "when csum info is missing");
2262 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostudp",
2263 CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_UDP,
2264 hn_trust_hcsum_sysctl, "I",
2265 "Trust udp datagram verification on host side, "
2266 "when csum info is missing");
2267 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostip",
2268 CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_IP,
2269 hn_trust_hcsum_sysctl, "I",
2270 "Trust ip packet verification on host side, "
2271 "when csum info is missing");
2272 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_ip",
2273 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2274 __offsetof(struct hn_rx_ring, hn_csum_ip),
2275 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM IP");
2276 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_tcp",
2277 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2278 __offsetof(struct hn_rx_ring, hn_csum_tcp),
2279 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM TCP");
2280 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_udp",
2281 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2282 __offsetof(struct hn_rx_ring, hn_csum_udp),
2283 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM UDP");
2284 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_trusted",
2285 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2286 __offsetof(struct hn_rx_ring, hn_csum_trusted),
2287 hn_rx_stat_ulong_sysctl, "LU",
2288 "# of packets that we trust host's csum verification");
2289 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "small_pkts",
2290 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2291 __offsetof(struct hn_rx_ring, hn_small_pkts),
2292 hn_rx_stat_ulong_sysctl, "LU", "# of small packets received");
2293 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_cnt",
2294 CTLFLAG_RD, &sc->hn_rx_ring_cnt, 0, "# created RX rings");
2295 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_inuse",
2296 CTLFLAG_RD, &sc->hn_rx_ring_inuse, 0, "# used RX rings");
2300 hn_destroy_rx_data(struct hn_softc *sc)
2302 #if defined(INET) || defined(INET6)
2306 if (sc->hn_rx_ring_cnt == 0)
2309 #if defined(INET) || defined(INET6)
2310 for (i = 0; i < sc->hn_rx_ring_cnt; ++i)
2311 tcp_lro_free(&sc->hn_rx_ring[i].hn_lro);
2313 free(sc->hn_rx_ring, M_NETVSC);
2314 sc->hn_rx_ring = NULL;
2316 sc->hn_rx_ring_cnt = 0;
2317 sc->hn_rx_ring_inuse = 0;
2321 hn_create_tx_ring(struct hn_softc *sc, int id)
2323 struct hn_tx_ring *txr = &sc->hn_tx_ring[id];
2324 bus_dma_tag_t parent_dtag;
2328 txr->hn_tx_idx = id;
2330 #ifndef HN_USE_TXDESC_BUFRING
2331 mtx_init(&txr->hn_txlist_spin, "hn txlist", NULL, MTX_SPIN);
2333 mtx_init(&txr->hn_tx_lock, "hn tx", NULL, MTX_DEF);
2335 txr->hn_txdesc_cnt = HN_TX_DESC_CNT;
2336 txr->hn_txdesc = malloc(sizeof(struct hn_txdesc) * txr->hn_txdesc_cnt,
2337 M_NETVSC, M_WAITOK | M_ZERO);
2338 #ifndef HN_USE_TXDESC_BUFRING
2339 SLIST_INIT(&txr->hn_txlist);
2341 txr->hn_txdesc_br = buf_ring_alloc(txr->hn_txdesc_cnt, M_NETVSC,
2342 M_WAITOK, &txr->hn_tx_lock);
2345 txr->hn_tx_taskq = sc->hn_tx_taskq;
2347 if (hn_use_if_start) {
2348 txr->hn_txeof = hn_start_txeof;
2349 TASK_INIT(&txr->hn_tx_task, 0, hn_start_taskfunc, txr);
2350 TASK_INIT(&txr->hn_txeof_task, 0, hn_start_txeof_taskfunc, txr);
2352 txr->hn_txeof = hn_xmit_txeof;
2353 TASK_INIT(&txr->hn_tx_task, 0, hn_xmit_taskfunc, txr);
2354 TASK_INIT(&txr->hn_txeof_task, 0, hn_xmit_txeof_taskfunc, txr);
2355 txr->hn_mbuf_br = buf_ring_alloc(txr->hn_txdesc_cnt, M_NETVSC,
2356 M_WAITOK, &txr->hn_tx_lock);
2359 txr->hn_direct_tx_size = hn_direct_tx_size;
2360 if (hv_vmbus_protocal_version >= HV_VMBUS_VERSION_WIN8_1)
2361 txr->hn_csum_assist = HN_CSUM_ASSIST;
2363 txr->hn_csum_assist = HN_CSUM_ASSIST_WIN8;
2366 * Always schedule transmission instead of trying to do direct
2367 * transmission. This one gives the best performance so far.
2369 txr->hn_sched_tx = 1;
2371 parent_dtag = bus_get_dma_tag(sc->hn_dev);
2373 /* DMA tag for RNDIS messages. */
2374 error = bus_dma_tag_create(parent_dtag, /* parent */
2375 HN_RNDIS_MSG_ALIGN, /* alignment */
2376 HN_RNDIS_MSG_BOUNDARY, /* boundary */
2377 BUS_SPACE_MAXADDR, /* lowaddr */
2378 BUS_SPACE_MAXADDR, /* highaddr */
2379 NULL, NULL, /* filter, filterarg */
2380 HN_RNDIS_MSG_LEN, /* maxsize */
2382 HN_RNDIS_MSG_LEN, /* maxsegsize */
2384 NULL, /* lockfunc */
2385 NULL, /* lockfuncarg */
2386 &txr->hn_tx_rndis_dtag);
2388 device_printf(sc->hn_dev, "failed to create rndis dmatag\n");
2392 /* DMA tag for data. */
2393 error = bus_dma_tag_create(parent_dtag, /* parent */
2395 HN_TX_DATA_BOUNDARY, /* boundary */
2396 BUS_SPACE_MAXADDR, /* lowaddr */
2397 BUS_SPACE_MAXADDR, /* highaddr */
2398 NULL, NULL, /* filter, filterarg */
2399 HN_TX_DATA_MAXSIZE, /* maxsize */
2400 HN_TX_DATA_SEGCNT_MAX, /* nsegments */
2401 HN_TX_DATA_SEGSIZE, /* maxsegsize */
2403 NULL, /* lockfunc */
2404 NULL, /* lockfuncarg */
2405 &txr->hn_tx_data_dtag);
2407 device_printf(sc->hn_dev, "failed to create data dmatag\n");
2411 for (i = 0; i < txr->hn_txdesc_cnt; ++i) {
2412 struct hn_txdesc *txd = &txr->hn_txdesc[i];
2417 * Allocate and load RNDIS messages.
2419 error = bus_dmamem_alloc(txr->hn_tx_rndis_dtag,
2420 (void **)&txd->rndis_msg,
2421 BUS_DMA_WAITOK | BUS_DMA_COHERENT,
2422 &txd->rndis_msg_dmap);
2424 device_printf(sc->hn_dev,
2425 "failed to allocate rndis_msg, %d\n", i);
2429 error = bus_dmamap_load(txr->hn_tx_rndis_dtag,
2430 txd->rndis_msg_dmap,
2431 txd->rndis_msg, HN_RNDIS_MSG_LEN,
2432 hn_dma_map_paddr, &txd->rndis_msg_paddr,
2435 device_printf(sc->hn_dev,
2436 "failed to load rndis_msg, %d\n", i);
2437 bus_dmamem_free(txr->hn_tx_rndis_dtag,
2438 txd->rndis_msg, txd->rndis_msg_dmap);
2442 /* DMA map for TX data. */
2443 error = bus_dmamap_create(txr->hn_tx_data_dtag, 0,
2446 device_printf(sc->hn_dev,
2447 "failed to allocate tx data dmamap\n");
2448 bus_dmamap_unload(txr->hn_tx_rndis_dtag,
2449 txd->rndis_msg_dmap);
2450 bus_dmamem_free(txr->hn_tx_rndis_dtag,
2451 txd->rndis_msg, txd->rndis_msg_dmap);
2455 /* All set, put it to list */
2456 txd->flags |= HN_TXD_FLAG_ONLIST;
2457 #ifndef HN_USE_TXDESC_BUFRING
2458 SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
2460 buf_ring_enqueue(txr->hn_txdesc_br, txd);
2463 txr->hn_txdesc_avail = txr->hn_txdesc_cnt;
2465 if (sc->hn_tx_sysctl_tree != NULL) {
2466 struct sysctl_oid_list *child;
2467 struct sysctl_ctx_list *ctx;
2471 * Create per TX ring sysctl tree:
2472 * dev.hn.UNIT.tx.RINGID
2474 ctx = device_get_sysctl_ctx(sc->hn_dev);
2475 child = SYSCTL_CHILDREN(sc->hn_tx_sysctl_tree);
2477 snprintf(name, sizeof(name), "%d", id);
2478 txr->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO,
2479 name, CTLFLAG_RD, 0, "");
2481 if (txr->hn_tx_sysctl_tree != NULL) {
2482 child = SYSCTL_CHILDREN(txr->hn_tx_sysctl_tree);
2484 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_avail",
2485 CTLFLAG_RD, &txr->hn_txdesc_avail, 0,
2486 "# of available TX descs");
2487 if (!hn_use_if_start) {
2488 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "oactive",
2489 CTLFLAG_RD, &txr->hn_oactive, 0,
2492 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "packets",
2493 CTLFLAG_RW, &txr->hn_pkts,
2494 "# of packets transmitted");
2502 hn_txdesc_dmamap_destroy(struct hn_txdesc *txd)
2504 struct hn_tx_ring *txr = txd->txr;
2506 KASSERT(txd->m == NULL, ("still has mbuf installed"));
2507 KASSERT((txd->flags & HN_TXD_FLAG_DMAMAP) == 0, ("still dma mapped"));
2509 bus_dmamap_unload(txr->hn_tx_rndis_dtag, txd->rndis_msg_dmap);
2510 bus_dmamem_free(txr->hn_tx_rndis_dtag, txd->rndis_msg,
2511 txd->rndis_msg_dmap);
2512 bus_dmamap_destroy(txr->hn_tx_data_dtag, txd->data_dmap);
2516 hn_destroy_tx_ring(struct hn_tx_ring *txr)
2518 struct hn_txdesc *txd;
2520 if (txr->hn_txdesc == NULL)
2523 #ifndef HN_USE_TXDESC_BUFRING
2524 while ((txd = SLIST_FIRST(&txr->hn_txlist)) != NULL) {
2525 SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
2526 hn_txdesc_dmamap_destroy(txd);
2529 mtx_lock(&txr->hn_tx_lock);
2530 while ((txd = buf_ring_dequeue_sc(txr->hn_txdesc_br)) != NULL)
2531 hn_txdesc_dmamap_destroy(txd);
2532 mtx_unlock(&txr->hn_tx_lock);
2535 if (txr->hn_tx_data_dtag != NULL)
2536 bus_dma_tag_destroy(txr->hn_tx_data_dtag);
2537 if (txr->hn_tx_rndis_dtag != NULL)
2538 bus_dma_tag_destroy(txr->hn_tx_rndis_dtag);
2540 #ifdef HN_USE_TXDESC_BUFRING
2541 buf_ring_free(txr->hn_txdesc_br, M_NETVSC);
2544 free(txr->hn_txdesc, M_NETVSC);
2545 txr->hn_txdesc = NULL;
2547 if (txr->hn_mbuf_br != NULL)
2548 buf_ring_free(txr->hn_mbuf_br, M_NETVSC);
2550 #ifndef HN_USE_TXDESC_BUFRING
2551 mtx_destroy(&txr->hn_txlist_spin);
2553 mtx_destroy(&txr->hn_tx_lock);
2557 hn_create_tx_data(struct hn_softc *sc, int ring_cnt)
2559 struct sysctl_oid_list *child;
2560 struct sysctl_ctx_list *ctx;
2563 sc->hn_tx_ring_cnt = ring_cnt;
2564 sc->hn_tx_ring_inuse = sc->hn_tx_ring_cnt;
2566 sc->hn_tx_ring = malloc(sizeof(struct hn_tx_ring) * sc->hn_tx_ring_cnt,
2567 M_NETVSC, M_WAITOK | M_ZERO);
2569 ctx = device_get_sysctl_ctx(sc->hn_dev);
2570 child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->hn_dev));
2572 /* Create dev.hn.UNIT.tx sysctl tree */
2573 sc->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "tx",
2576 for (i = 0; i < sc->hn_tx_ring_cnt; ++i) {
2579 error = hn_create_tx_ring(sc, i);
2584 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "no_txdescs",
2585 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2586 __offsetof(struct hn_tx_ring, hn_no_txdescs),
2587 hn_tx_stat_ulong_sysctl, "LU", "# of times short of TX descs");
2588 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "send_failed",
2589 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2590 __offsetof(struct hn_tx_ring, hn_send_failed),
2591 hn_tx_stat_ulong_sysctl, "LU", "# of hyper-v sending failure");
2592 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "txdma_failed",
2593 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2594 __offsetof(struct hn_tx_ring, hn_txdma_failed),
2595 hn_tx_stat_ulong_sysctl, "LU", "# of TX DMA failure");
2596 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_collapsed",
2597 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2598 __offsetof(struct hn_tx_ring, hn_tx_collapsed),
2599 hn_tx_stat_ulong_sysctl, "LU", "# of TX mbuf collapsed");
2600 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney",
2601 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2602 __offsetof(struct hn_tx_ring, hn_tx_chimney),
2603 hn_tx_stat_ulong_sysctl, "LU", "# of chimney send");
2604 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_cnt",
2605 CTLFLAG_RD, &sc->hn_tx_ring[0].hn_txdesc_cnt, 0,
2606 "# of total TX descs");
2607 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_chimney_max",
2608 CTLFLAG_RD, &sc->hn_tx_chimney_max, 0,
2609 "Chimney send packet size upper boundary");
2610 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_size",
2611 CTLTYPE_INT | CTLFLAG_RW, sc, 0, hn_tx_chimney_size_sysctl,
2612 "I", "Chimney send packet size limit");
2613 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "direct_tx_size",
2614 CTLTYPE_INT | CTLFLAG_RW, sc,
2615 __offsetof(struct hn_tx_ring, hn_direct_tx_size),
2616 hn_tx_conf_int_sysctl, "I",
2617 "Size of the packet for direct transmission");
2618 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "sched_tx",
2619 CTLTYPE_INT | CTLFLAG_RW, sc,
2620 __offsetof(struct hn_tx_ring, hn_sched_tx),
2621 hn_tx_conf_int_sysctl, "I",
2622 "Always schedule transmission "
2623 "instead of doing direct transmission");
2624 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_cnt",
2625 CTLFLAG_RD, &sc->hn_tx_ring_cnt, 0, "# created TX rings");
2626 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_inuse",
2627 CTLFLAG_RD, &sc->hn_tx_ring_inuse, 0, "# used TX rings");
2633 hn_set_tx_chimney_size(struct hn_softc *sc, int chimney_size)
2638 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
2639 sc->hn_tx_ring[i].hn_tx_chimney_size = chimney_size;
2644 hn_destroy_tx_data(struct hn_softc *sc)
2648 if (sc->hn_tx_ring_cnt == 0)
2651 for (i = 0; i < sc->hn_tx_ring_cnt; ++i)
2652 hn_destroy_tx_ring(&sc->hn_tx_ring[i]);
2654 free(sc->hn_tx_ring, M_NETVSC);
2655 sc->hn_tx_ring = NULL;
2657 sc->hn_tx_ring_cnt = 0;
2658 sc->hn_tx_ring_inuse = 0;
2662 hn_start_taskfunc(void *xtxr, int pending __unused)
2664 struct hn_tx_ring *txr = xtxr;
2666 mtx_lock(&txr->hn_tx_lock);
2667 hn_start_locked(txr, 0);
2668 mtx_unlock(&txr->hn_tx_lock);
2672 hn_start_txeof_taskfunc(void *xtxr, int pending __unused)
2674 struct hn_tx_ring *txr = xtxr;
2676 mtx_lock(&txr->hn_tx_lock);
2677 atomic_clear_int(&txr->hn_sc->hn_ifp->if_drv_flags, IFF_DRV_OACTIVE);
2678 hn_start_locked(txr, 0);
2679 mtx_unlock(&txr->hn_tx_lock);
2683 hn_stop_tx_tasks(struct hn_softc *sc)
2687 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2688 struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
2690 taskqueue_drain(txr->hn_tx_taskq, &txr->hn_tx_task);
2691 taskqueue_drain(txr->hn_tx_taskq, &txr->hn_txeof_task);
2696 hn_xmit(struct hn_tx_ring *txr, int len)
2698 struct hn_softc *sc = txr->hn_sc;
2699 struct ifnet *ifp = sc->hn_ifp;
2700 struct mbuf *m_head;
2702 mtx_assert(&txr->hn_tx_lock, MA_OWNED);
2703 KASSERT(hn_use_if_start == 0,
2704 ("hn_xmit is called, when if_start is enabled"));
2706 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || txr->hn_oactive)
2709 while ((m_head = drbr_peek(ifp, txr->hn_mbuf_br)) != NULL) {
2710 struct hn_txdesc *txd;
2713 if (len > 0 && m_head->m_pkthdr.len > len) {
2715 * This sending could be time consuming; let callers
2716 * dispatch this packet sending (and sending of any
2717 * following up packets) to tx taskqueue.
2719 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2723 txd = hn_txdesc_get(txr);
2725 txr->hn_no_txdescs++;
2726 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2727 txr->hn_oactive = 1;
2731 error = hn_encap(txr, txd, &m_head);
2733 /* Both txd and m_head are freed; discard */
2734 drbr_advance(ifp, txr->hn_mbuf_br);
2738 error = hn_send_pkt(ifp, txr, txd);
2739 if (__predict_false(error)) {
2740 /* txd is freed, but m_head is not */
2741 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2742 txr->hn_oactive = 1;
2747 drbr_advance(ifp, txr->hn_mbuf_br);
2753 hn_transmit(struct ifnet *ifp, struct mbuf *m)
2755 struct hn_softc *sc = ifp->if_softc;
2756 struct hn_tx_ring *txr;
2760 * Select the TX ring based on flowid
2762 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2763 idx = m->m_pkthdr.flowid % sc->hn_tx_ring_inuse;
2764 txr = &sc->hn_tx_ring[idx];
2766 error = drbr_enqueue(ifp, txr->hn_mbuf_br, m);
2770 if (txr->hn_oactive)
2773 if (txr->hn_sched_tx)
2776 if (mtx_trylock(&txr->hn_tx_lock)) {
2779 sched = hn_xmit(txr, txr->hn_direct_tx_size);
2780 mtx_unlock(&txr->hn_tx_lock);
2785 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
2790 hn_xmit_qflush(struct ifnet *ifp)
2792 struct hn_softc *sc = ifp->if_softc;
2795 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2796 struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
2799 mtx_lock(&txr->hn_tx_lock);
2800 while ((m = buf_ring_dequeue_sc(txr->hn_mbuf_br)) != NULL)
2802 mtx_unlock(&txr->hn_tx_lock);
2808 hn_xmit_txeof(struct hn_tx_ring *txr)
2811 if (txr->hn_sched_tx)
2814 if (mtx_trylock(&txr->hn_tx_lock)) {
2817 txr->hn_oactive = 0;
2818 sched = hn_xmit(txr, txr->hn_direct_tx_size);
2819 mtx_unlock(&txr->hn_tx_lock);
2821 taskqueue_enqueue(txr->hn_tx_taskq,
2827 * Release the oactive earlier, with the hope, that
2828 * others could catch up. The task will clear the
2829 * oactive again with the hn_tx_lock to avoid possible
2832 txr->hn_oactive = 0;
2833 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
2838 hn_xmit_taskfunc(void *xtxr, int pending __unused)
2840 struct hn_tx_ring *txr = xtxr;
2842 mtx_lock(&txr->hn_tx_lock);
2844 mtx_unlock(&txr->hn_tx_lock);
2848 hn_xmit_txeof_taskfunc(void *xtxr, int pending __unused)
2850 struct hn_tx_ring *txr = xtxr;
2852 mtx_lock(&txr->hn_tx_lock);
2853 txr->hn_oactive = 0;
2855 mtx_unlock(&txr->hn_tx_lock);
2859 hn_channel_attach(struct hn_softc *sc, struct hv_vmbus_channel *chan)
2861 struct hn_rx_ring *rxr;
2864 idx = chan->offer_msg.offer.sub_channel_index;
2866 KASSERT(idx >= 0 && idx < sc->hn_rx_ring_inuse,
2867 ("invalid channel index %d, should > 0 && < %d",
2868 idx, sc->hn_rx_ring_inuse));
2869 rxr = &sc->hn_rx_ring[idx];
2870 KASSERT((rxr->hn_rx_flags & HN_RX_FLAG_ATTACHED) == 0,
2871 ("RX ring %d already attached", idx));
2872 rxr->hn_rx_flags |= HN_RX_FLAG_ATTACHED;
2874 chan->hv_chan_rxr = rxr;
2875 if_printf(sc->hn_ifp, "link RX ring %d to channel%u\n",
2876 idx, chan->offer_msg.child_rel_id);
2878 if (idx < sc->hn_tx_ring_inuse) {
2879 struct hn_tx_ring *txr = &sc->hn_tx_ring[idx];
2881 KASSERT((txr->hn_tx_flags & HN_TX_FLAG_ATTACHED) == 0,
2882 ("TX ring %d already attached", idx));
2883 txr->hn_tx_flags |= HN_TX_FLAG_ATTACHED;
2885 chan->hv_chan_txr = txr;
2886 txr->hn_chan = chan;
2887 if_printf(sc->hn_ifp, "link TX ring %d to channel%u\n",
2888 idx, chan->offer_msg.child_rel_id);
2891 /* Bind channel to a proper CPU */
2892 vmbus_channel_cpu_set(chan, (sc->hn_cpu + idx) % mp_ncpus);
2896 netvsc_subchan_callback(struct hn_softc *sc, struct hv_vmbus_channel *chan)
2899 KASSERT(!HV_VMBUS_CHAN_ISPRIMARY(chan),
2900 ("subchannel callback on primary channel"));
2901 KASSERT(chan->offer_msg.offer.sub_channel_index > 0,
2902 ("invalid channel subidx %u",
2903 chan->offer_msg.offer.sub_channel_index));
2904 hn_channel_attach(sc, chan);
2908 hn_tx_taskq_create(void *arg __unused)
2910 if (!hn_share_tx_taskq)
2913 hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
2914 taskqueue_thread_enqueue, &hn_tx_taskq);
2915 taskqueue_start_threads(&hn_tx_taskq, 1, PI_NET, "hn tx");
2916 if (hn_bind_tx_taskq >= 0) {
2917 int cpu = hn_bind_tx_taskq;
2918 struct task cpuset_task;
2921 if (cpu > mp_ncpus - 1)
2923 CPU_SETOF(cpu, &cpu_set);
2924 TASK_INIT(&cpuset_task, 0, hn_cpuset_setthread_task, &cpu_set);
2925 taskqueue_enqueue(hn_tx_taskq, &cpuset_task);
2926 taskqueue_drain(hn_tx_taskq, &cpuset_task);
2929 SYSINIT(hn_txtq_create, SI_SUB_DRIVERS, SI_ORDER_FIRST,
2930 hn_tx_taskq_create, NULL);
2933 hn_tx_taskq_destroy(void *arg __unused)
2935 if (hn_tx_taskq != NULL)
2936 taskqueue_free(hn_tx_taskq);
2938 SYSUNINIT(hn_txtq_destroy, SI_SUB_DRIVERS, SI_ORDER_FIRST,
2939 hn_tx_taskq_destroy, NULL);
2941 static device_method_t netvsc_methods[] = {
2942 /* Device interface */
2943 DEVMETHOD(device_probe, netvsc_probe),
2944 DEVMETHOD(device_attach, netvsc_attach),
2945 DEVMETHOD(device_detach, netvsc_detach),
2946 DEVMETHOD(device_shutdown, netvsc_shutdown),
2951 static driver_t netvsc_driver = {
2957 static devclass_t netvsc_devclass;
2959 DRIVER_MODULE(hn, vmbus, netvsc_driver, netvsc_devclass, 0, 0);
2960 MODULE_VERSION(hn, 1);
2961 MODULE_DEPEND(hn, vmbus, 1, 1, 1);