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 if (hn_tx_taskq == NULL) {
441 sc->hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
442 taskqueue_thread_enqueue, &sc->hn_tx_taskq);
443 taskqueue_start_threads(&sc->hn_tx_taskq, 1, PI_NET, "%s tx",
444 device_get_nameunit(dev));
445 if (hn_bind_tx_taskq >= 0) {
446 int cpu = hn_bind_tx_taskq;
447 struct task cpuset_task;
450 if (cpu > mp_ncpus - 1)
452 CPU_SETOF(cpu, &cpu_set);
453 TASK_INIT(&cpuset_task, 0, hn_cpuset_setthread_task,
455 taskqueue_enqueue(sc->hn_tx_taskq, &cpuset_task);
456 taskqueue_drain(sc->hn_tx_taskq, &cpuset_task);
459 sc->hn_tx_taskq = hn_tx_taskq;
461 NV_LOCK_INIT(sc, "NetVSCLock");
463 sc->hn_dev_obj = device_ctx;
465 ifp = sc->hn_ifp = sc->arpcom.ac_ifp = if_alloc(IFT_ETHER);
467 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
470 * Figure out the # of RX rings (ring_cnt) and the # of TX rings
471 * to use (tx_ring_cnt).
474 * The # of RX rings to use is same as the # of channels to use.
476 ring_cnt = hn_chan_cnt;
480 if (ring_cnt > HN_RING_CNT_DEF_MAX)
481 ring_cnt = HN_RING_CNT_DEF_MAX;
482 } else if (ring_cnt > mp_ncpus) {
486 tx_ring_cnt = hn_tx_ring_cnt;
487 if (tx_ring_cnt <= 0 || tx_ring_cnt > ring_cnt)
488 tx_ring_cnt = ring_cnt;
489 if (hn_use_if_start) {
490 /* ifnet.if_start only needs one TX ring. */
495 * Set the leader CPU for channels.
497 sc->hn_cpu = atomic_fetchadd_int(&hn_cpu_index, ring_cnt) % mp_ncpus;
499 error = hn_create_tx_data(sc, tx_ring_cnt);
502 hn_create_rx_data(sc, ring_cnt);
505 * Associate the first TX/RX ring w/ the primary channel.
507 chan = device_ctx->channel;
508 KASSERT(HV_VMBUS_CHAN_ISPRIMARY(chan), ("not primary channel"));
509 KASSERT(chan->offer_msg.offer.sub_channel_index == 0,
510 ("primary channel subidx %u",
511 chan->offer_msg.offer.sub_channel_index));
512 hn_channel_attach(sc, chan);
514 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
515 ifp->if_ioctl = hn_ioctl;
516 ifp->if_init = hn_ifinit;
517 /* needed by hv_rf_on_device_add() code */
518 ifp->if_mtu = ETHERMTU;
519 if (hn_use_if_start) {
520 ifp->if_start = hn_start;
521 IFQ_SET_MAXLEN(&ifp->if_snd, 512);
522 ifp->if_snd.ifq_drv_maxlen = 511;
523 IFQ_SET_READY(&ifp->if_snd);
525 ifp->if_transmit = hn_transmit;
526 ifp->if_qflush = hn_xmit_qflush;
529 ifmedia_init(&sc->hn_media, 0, hn_ifmedia_upd, hn_ifmedia_sts);
530 ifmedia_add(&sc->hn_media, IFM_ETHER | IFM_AUTO, 0, NULL);
531 ifmedia_set(&sc->hn_media, IFM_ETHER | IFM_AUTO);
532 /* XXX ifmedia_set really should do this for us */
533 sc->hn_media.ifm_media = sc->hn_media.ifm_cur->ifm_media;
536 * Tell upper layers that we support full VLAN capability.
538 ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
539 ifp->if_capabilities |=
540 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
543 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
545 ifp->if_hwassist = sc->hn_tx_ring[0].hn_csum_assist | CSUM_TSO;
547 error = hv_rf_on_device_add(device_ctx, &device_info, ring_cnt);
550 KASSERT(sc->net_dev->num_channel > 0 &&
551 sc->net_dev->num_channel <= sc->hn_rx_ring_inuse,
552 ("invalid channel count %u, should be less than %d",
553 sc->net_dev->num_channel, sc->hn_rx_ring_inuse));
556 * Set the # of TX/RX rings that could be used according to
557 * the # of channels that host offered.
559 if (sc->hn_tx_ring_inuse > sc->net_dev->num_channel)
560 sc->hn_tx_ring_inuse = sc->net_dev->num_channel;
561 sc->hn_rx_ring_inuse = sc->net_dev->num_channel;
562 device_printf(dev, "%d TX ring, %d RX ring\n",
563 sc->hn_tx_ring_inuse, sc->hn_rx_ring_inuse);
565 #if __FreeBSD_version >= 1100099
566 if (sc->hn_rx_ring_inuse > 1) {
568 * Reduce TCP segment aggregation limit for multiple
569 * RX rings to increase ACK timeliness.
571 hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MULTIRX_DEF);
575 if (device_info.link_state == 0) {
579 tso_maxlen = hn_tso_maxlen;
580 if (tso_maxlen <= 0 || tso_maxlen > IP_MAXPACKET)
581 tso_maxlen = IP_MAXPACKET;
583 ifp->if_hw_tsomaxsegcount = HN_TX_DATA_SEGCNT_MAX;
584 ifp->if_hw_tsomaxsegsize = PAGE_SIZE;
585 ifp->if_hw_tsomax = tso_maxlen -
586 (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
588 ether_ifattach(ifp, device_info.mac_addr);
590 if_printf(ifp, "TSO: %u/%u/%u\n", ifp->if_hw_tsomax,
591 ifp->if_hw_tsomaxsegcount, ifp->if_hw_tsomaxsegsize);
593 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
594 hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max);
595 if (hn_tx_chimney_size > 0 &&
596 hn_tx_chimney_size < sc->hn_tx_chimney_max)
597 hn_set_tx_chimney_size(sc, hn_tx_chimney_size);
601 hn_destroy_tx_data(sc);
608 * Standard detach entry point
611 netvsc_detach(device_t dev)
613 struct hn_softc *sc = device_get_softc(dev);
614 struct hv_device *hv_device = vmbus_get_devctx(dev);
617 printf("netvsc_detach\n");
620 * XXXKYS: Need to clean up all our
621 * driver state; this is the driver
626 * XXXKYS: Need to stop outgoing traffic and unregister
630 hv_rf_on_device_remove(hv_device, HV_RF_NV_DESTROY_CHANNEL);
632 hn_stop_tx_tasks(sc);
634 ifmedia_removeall(&sc->hn_media);
635 hn_destroy_rx_data(sc);
636 hn_destroy_tx_data(sc);
638 if (sc->hn_tx_taskq != hn_tx_taskq)
639 taskqueue_free(sc->hn_tx_taskq);
645 * Standard shutdown entry point
648 netvsc_shutdown(device_t dev)
654 hn_txdesc_dmamap_load(struct hn_tx_ring *txr, struct hn_txdesc *txd,
655 struct mbuf **m_head, bus_dma_segment_t *segs, int *nsegs)
657 struct mbuf *m = *m_head;
660 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag, txd->data_dmap,
661 m, segs, nsegs, BUS_DMA_NOWAIT);
662 if (error == EFBIG) {
665 m_new = m_collapse(m, M_NOWAIT, HN_TX_DATA_SEGCNT_MAX);
670 txr->hn_tx_collapsed++;
672 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag,
673 txd->data_dmap, m, segs, nsegs, BUS_DMA_NOWAIT);
676 bus_dmamap_sync(txr->hn_tx_data_dtag, txd->data_dmap,
677 BUS_DMASYNC_PREWRITE);
678 txd->flags |= HN_TXD_FLAG_DMAMAP;
684 hn_txdesc_dmamap_unload(struct hn_tx_ring *txr, struct hn_txdesc *txd)
687 if (txd->flags & HN_TXD_FLAG_DMAMAP) {
688 bus_dmamap_sync(txr->hn_tx_data_dtag,
689 txd->data_dmap, BUS_DMASYNC_POSTWRITE);
690 bus_dmamap_unload(txr->hn_tx_data_dtag,
692 txd->flags &= ~HN_TXD_FLAG_DMAMAP;
697 hn_txdesc_put(struct hn_tx_ring *txr, struct hn_txdesc *txd)
700 KASSERT((txd->flags & HN_TXD_FLAG_ONLIST) == 0,
701 ("put an onlist txd %#x", txd->flags));
703 KASSERT(txd->refs > 0, ("invalid txd refs %d", txd->refs));
704 if (atomic_fetchadd_int(&txd->refs, -1) != 1)
707 hn_txdesc_dmamap_unload(txr, txd);
708 if (txd->m != NULL) {
713 txd->flags |= HN_TXD_FLAG_ONLIST;
715 #ifndef HN_USE_TXDESC_BUFRING
716 mtx_lock_spin(&txr->hn_txlist_spin);
717 KASSERT(txr->hn_txdesc_avail >= 0 &&
718 txr->hn_txdesc_avail < txr->hn_txdesc_cnt,
719 ("txdesc_put: invalid txd avail %d", txr->hn_txdesc_avail));
720 txr->hn_txdesc_avail++;
721 SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
722 mtx_unlock_spin(&txr->hn_txlist_spin);
724 atomic_add_int(&txr->hn_txdesc_avail, 1);
725 buf_ring_enqueue(txr->hn_txdesc_br, txd);
731 static __inline struct hn_txdesc *
732 hn_txdesc_get(struct hn_tx_ring *txr)
734 struct hn_txdesc *txd;
736 #ifndef HN_USE_TXDESC_BUFRING
737 mtx_lock_spin(&txr->hn_txlist_spin);
738 txd = SLIST_FIRST(&txr->hn_txlist);
740 KASSERT(txr->hn_txdesc_avail > 0,
741 ("txdesc_get: invalid txd avail %d", txr->hn_txdesc_avail));
742 txr->hn_txdesc_avail--;
743 SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
745 mtx_unlock_spin(&txr->hn_txlist_spin);
747 txd = buf_ring_dequeue_sc(txr->hn_txdesc_br);
751 #ifdef HN_USE_TXDESC_BUFRING
752 atomic_subtract_int(&txr->hn_txdesc_avail, 1);
754 KASSERT(txd->m == NULL && txd->refs == 0 &&
755 (txd->flags & HN_TXD_FLAG_ONLIST), ("invalid txd"));
756 txd->flags &= ~HN_TXD_FLAG_ONLIST;
763 hn_txdesc_hold(struct hn_txdesc *txd)
766 /* 0->1 transition will never work */
767 KASSERT(txd->refs > 0, ("invalid refs %d", txd->refs));
768 atomic_add_int(&txd->refs, 1);
772 hn_tx_done(struct hv_vmbus_channel *chan, void *xpkt)
774 netvsc_packet *packet = xpkt;
775 struct hn_txdesc *txd;
776 struct hn_tx_ring *txr;
778 txd = (struct hn_txdesc *)(uintptr_t)
779 packet->compl.send.send_completion_tid;
782 KASSERT(txr->hn_chan == chan,
783 ("channel mismatch, on channel%u, should be channel%u",
784 chan->offer_msg.offer.sub_channel_index,
785 txr->hn_chan->offer_msg.offer.sub_channel_index));
787 txr->hn_has_txeof = 1;
788 hn_txdesc_put(txr, txd);
792 netvsc_channel_rollup(struct hv_vmbus_channel *chan)
794 struct hn_tx_ring *txr = chan->hv_chan_txr;
795 #if defined(INET) || defined(INET6)
796 struct hn_rx_ring *rxr = chan->hv_chan_rxr;
797 struct lro_ctrl *lro = &rxr->hn_lro;
798 struct lro_entry *queued;
800 while ((queued = SLIST_FIRST(&lro->lro_active)) != NULL) {
801 SLIST_REMOVE_HEAD(&lro->lro_active, next);
802 tcp_lro_flush(lro, queued);
808 * 'txr' could be NULL, if multiple channels and
809 * ifnet.if_start method are enabled.
811 if (txr == NULL || !txr->hn_has_txeof)
814 txr->hn_has_txeof = 0;
820 * If this function fails, then both txd and m_head0 will be freed.
823 hn_encap(struct hn_tx_ring *txr, struct hn_txdesc *txd, struct mbuf **m_head0)
825 bus_dma_segment_t segs[HN_TX_DATA_SEGCNT_MAX];
827 struct mbuf *m_head = *m_head0;
828 netvsc_packet *packet;
829 rndis_msg *rndis_mesg;
830 rndis_packet *rndis_pkt;
831 rndis_per_packet_info *rppi;
832 struct ndis_hash_info *hash_info;
833 uint32_t rndis_msg_size;
835 packet = &txd->netvsc_pkt;
836 packet->is_data_pkt = TRUE;
837 packet->tot_data_buf_len = m_head->m_pkthdr.len;
840 * extension points to the area reserved for the
841 * rndis_filter_packet, which is placed just after
842 * the netvsc_packet (and rppi struct, if present;
843 * length is updated later).
845 rndis_mesg = txd->rndis_msg;
846 /* XXX not necessary */
847 memset(rndis_mesg, 0, HN_RNDIS_MSG_LEN);
848 rndis_mesg->ndis_msg_type = REMOTE_NDIS_PACKET_MSG;
850 rndis_pkt = &rndis_mesg->msg.packet;
851 rndis_pkt->data_offset = sizeof(rndis_packet);
852 rndis_pkt->data_length = packet->tot_data_buf_len;
853 rndis_pkt->per_pkt_info_offset = sizeof(rndis_packet);
855 rndis_msg_size = RNDIS_MESSAGE_SIZE(rndis_packet);
858 * Set the hash info for this packet, so that the host could
859 * dispatch the TX done event for this packet back to this TX
862 rndis_msg_size += RNDIS_HASH_PPI_SIZE;
863 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_HASH_PPI_SIZE,
865 hash_info = (struct ndis_hash_info *)((uint8_t *)rppi +
866 rppi->per_packet_info_offset);
867 hash_info->hash = txr->hn_tx_idx;
869 if (m_head->m_flags & M_VLANTAG) {
870 ndis_8021q_info *rppi_vlan_info;
872 rndis_msg_size += RNDIS_VLAN_PPI_SIZE;
873 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_VLAN_PPI_SIZE,
876 rppi_vlan_info = (ndis_8021q_info *)((uint8_t *)rppi +
877 rppi->per_packet_info_offset);
878 rppi_vlan_info->u1.s1.vlan_id =
879 m_head->m_pkthdr.ether_vtag & 0xfff;
882 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
883 rndis_tcp_tso_info *tso_info;
884 struct ether_vlan_header *eh;
888 * XXX need m_pullup and use mtodo
890 eh = mtod(m_head, struct ether_vlan_header*);
891 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN))
892 ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
894 ether_len = ETHER_HDR_LEN;
896 rndis_msg_size += RNDIS_TSO_PPI_SIZE;
897 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_TSO_PPI_SIZE,
898 tcp_large_send_info);
900 tso_info = (rndis_tcp_tso_info *)((uint8_t *)rppi +
901 rppi->per_packet_info_offset);
902 tso_info->lso_v2_xmit.type =
903 RNDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
906 if (m_head->m_pkthdr.csum_flags & CSUM_IP_TSO) {
908 (struct ip *)(m_head->m_data + ether_len);
909 unsigned long iph_len = ip->ip_hl << 2;
911 (struct tcphdr *)((caddr_t)ip + iph_len);
913 tso_info->lso_v2_xmit.ip_version =
914 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
918 th->th_sum = in_pseudo(ip->ip_src.s_addr,
919 ip->ip_dst.s_addr, htons(IPPROTO_TCP));
922 #if defined(INET6) && defined(INET)
927 struct ip6_hdr *ip6 = (struct ip6_hdr *)
928 (m_head->m_data + ether_len);
929 struct tcphdr *th = (struct tcphdr *)(ip6 + 1);
931 tso_info->lso_v2_xmit.ip_version =
932 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
934 th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
937 tso_info->lso_v2_xmit.tcp_header_offset = 0;
938 tso_info->lso_v2_xmit.mss = m_head->m_pkthdr.tso_segsz;
939 } else if (m_head->m_pkthdr.csum_flags & txr->hn_csum_assist) {
940 rndis_tcp_ip_csum_info *csum_info;
942 rndis_msg_size += RNDIS_CSUM_PPI_SIZE;
943 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_CSUM_PPI_SIZE,
945 csum_info = (rndis_tcp_ip_csum_info *)((uint8_t *)rppi +
946 rppi->per_packet_info_offset);
948 csum_info->xmit.is_ipv4 = 1;
949 if (m_head->m_pkthdr.csum_flags & CSUM_IP)
950 csum_info->xmit.ip_header_csum = 1;
952 if (m_head->m_pkthdr.csum_flags & CSUM_TCP) {
953 csum_info->xmit.tcp_csum = 1;
954 csum_info->xmit.tcp_header_offset = 0;
955 } else if (m_head->m_pkthdr.csum_flags & CSUM_UDP) {
956 csum_info->xmit.udp_csum = 1;
960 rndis_mesg->msg_len = packet->tot_data_buf_len + rndis_msg_size;
961 packet->tot_data_buf_len = rndis_mesg->msg_len;
964 * Chimney send, if the packet could fit into one chimney buffer.
966 * TODO: vRSS, chimney buffer should be per-channel.
968 if (packet->tot_data_buf_len < txr->hn_tx_chimney_size) {
969 netvsc_dev *net_dev = txr->hn_sc->net_dev;
970 uint32_t send_buf_section_idx;
972 send_buf_section_idx =
973 hv_nv_get_next_send_section(net_dev);
974 if (send_buf_section_idx !=
975 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX) {
976 uint8_t *dest = ((uint8_t *)net_dev->send_buf +
977 (send_buf_section_idx *
978 net_dev->send_section_size));
980 memcpy(dest, rndis_mesg, rndis_msg_size);
981 dest += rndis_msg_size;
982 m_copydata(m_head, 0, m_head->m_pkthdr.len, dest);
984 packet->send_buf_section_idx = send_buf_section_idx;
985 packet->send_buf_section_size =
986 packet->tot_data_buf_len;
987 packet->page_buf_count = 0;
988 txr->hn_tx_chimney++;
993 error = hn_txdesc_dmamap_load(txr, txd, &m_head, segs, &nsegs);
998 * This mbuf is not linked w/ the txd yet, so free it now.
1003 freed = hn_txdesc_put(txr, txd);
1005 ("fail to free txd upon txdma error"));
1007 txr->hn_txdma_failed++;
1008 if_inc_counter(txr->hn_sc->hn_ifp, IFCOUNTER_OERRORS, 1);
1013 packet->page_buf_count = nsegs + HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
1015 /* send packet with page buffer */
1016 packet->page_buffers[0].pfn = atop(txd->rndis_msg_paddr);
1017 packet->page_buffers[0].offset = txd->rndis_msg_paddr & PAGE_MASK;
1018 packet->page_buffers[0].length = rndis_msg_size;
1021 * Fill the page buffers with mbuf info starting at index
1022 * HV_RF_NUM_TX_RESERVED_PAGE_BUFS.
1024 for (i = 0; i < nsegs; ++i) {
1025 hv_vmbus_page_buffer *pb = &packet->page_buffers[
1026 i + HV_RF_NUM_TX_RESERVED_PAGE_BUFS];
1028 pb->pfn = atop(segs[i].ds_addr);
1029 pb->offset = segs[i].ds_addr & PAGE_MASK;
1030 pb->length = segs[i].ds_len;
1033 packet->send_buf_section_idx =
1034 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX;
1035 packet->send_buf_section_size = 0;
1039 /* Set the completion routine */
1040 packet->compl.send.on_send_completion = hn_tx_done;
1041 packet->compl.send.send_completion_context = packet;
1042 packet->compl.send.send_completion_tid = (uint64_t)(uintptr_t)txd;
1049 * If this function fails, then txd will be freed, but the mbuf
1050 * associated w/ the txd will _not_ be freed.
1053 hn_send_pkt(struct ifnet *ifp, struct hn_tx_ring *txr, struct hn_txdesc *txd)
1055 int error, send_failed = 0;
1059 * Make sure that txd is not freed before ETHER_BPF_MTAP.
1061 hn_txdesc_hold(txd);
1062 error = hv_nv_on_send(txr->hn_chan, &txd->netvsc_pkt);
1064 ETHER_BPF_MTAP(ifp, txd->m);
1065 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1066 if (!hn_use_if_start) {
1067 if_inc_counter(ifp, IFCOUNTER_OBYTES,
1068 txd->m->m_pkthdr.len);
1069 if (txd->m->m_flags & M_MCAST)
1070 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
1074 hn_txdesc_put(txr, txd);
1076 if (__predict_false(error)) {
1080 * This should "really rarely" happen.
1082 * XXX Too many RX to be acked or too many sideband
1083 * commands to run? Ask netvsc_channel_rollup()
1084 * to kick start later.
1086 txr->hn_has_txeof = 1;
1088 txr->hn_send_failed++;
1091 * Try sending again after set hn_has_txeof;
1092 * in case that we missed the last
1093 * netvsc_channel_rollup().
1097 if_printf(ifp, "send failed\n");
1100 * Caller will perform further processing on the
1101 * associated mbuf, so don't free it in hn_txdesc_put();
1102 * only unload it from the DMA map in hn_txdesc_put(),
1106 freed = hn_txdesc_put(txr, txd);
1108 ("fail to free txd upon send error"));
1110 txr->hn_send_failed++;
1116 * Start a transmit of one or more packets
1119 hn_start_locked(struct hn_tx_ring *txr, int len)
1121 struct hn_softc *sc = txr->hn_sc;
1122 struct ifnet *ifp = sc->hn_ifp;
1124 KASSERT(hn_use_if_start,
1125 ("hn_start_locked is called, when if_start is disabled"));
1126 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1127 mtx_assert(&txr->hn_tx_lock, MA_OWNED);
1129 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1133 while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
1134 struct hn_txdesc *txd;
1135 struct mbuf *m_head;
1138 IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
1142 if (len > 0 && m_head->m_pkthdr.len > len) {
1144 * This sending could be time consuming; let callers
1145 * dispatch this packet sending (and sending of any
1146 * following up packets) to tx taskqueue.
1148 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1152 txd = hn_txdesc_get(txr);
1154 txr->hn_no_txdescs++;
1155 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1156 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1160 error = hn_encap(txr, txd, &m_head);
1162 /* Both txd and m_head are freed */
1166 error = hn_send_pkt(ifp, txr, txd);
1167 if (__predict_false(error)) {
1168 /* txd is freed, but m_head is not */
1169 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1170 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1178 * Link up/down notification
1181 netvsc_linkstatus_callback(struct hv_device *device_obj, uint32_t status)
1183 hn_softc_t *sc = device_get_softc(device_obj->device);
1193 * Append the specified data to the indicated mbuf chain,
1194 * Extend the mbuf chain if the new data does not fit in
1197 * This is a minor rewrite of m_append() from sys/kern/uipc_mbuf.c.
1198 * There should be an equivalent in the kernel mbuf code,
1199 * but there does not appear to be one yet.
1201 * Differs from m_append() in that additional mbufs are
1202 * allocated with cluster size MJUMPAGESIZE, and filled
1205 * Return 1 if able to complete the job; otherwise 0.
1208 hv_m_append(struct mbuf *m0, int len, c_caddr_t cp)
1211 int remainder, space;
1213 for (m = m0; m->m_next != NULL; m = m->m_next)
1216 space = M_TRAILINGSPACE(m);
1219 * Copy into available space.
1221 if (space > remainder)
1223 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1228 while (remainder > 0) {
1230 * Allocate a new mbuf; could check space
1231 * and allocate a cluster instead.
1233 n = m_getjcl(M_DONTWAIT, m->m_type, 0, MJUMPAGESIZE);
1236 n->m_len = min(MJUMPAGESIZE, remainder);
1237 bcopy(cp, mtod(n, caddr_t), n->m_len);
1239 remainder -= n->m_len;
1243 if (m0->m_flags & M_PKTHDR)
1244 m0->m_pkthdr.len += len - remainder;
1246 return (remainder == 0);
1251 * Called when we receive a data packet from the "wire" on the
1254 * Note: This is no longer used as a callback
1257 netvsc_recv(struct hv_vmbus_channel *chan, netvsc_packet *packet,
1258 rndis_tcp_ip_csum_info *csum_info)
1260 struct hn_rx_ring *rxr = chan->hv_chan_rxr;
1261 struct ifnet *ifp = rxr->hn_ifp;
1263 int size, do_lro = 0, do_csum = 1;
1265 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1269 * Bail out if packet contains more data than configured MTU.
1271 if (packet->tot_data_buf_len > (ifp->if_mtu + ETHER_HDR_LEN)) {
1273 } else if (packet->tot_data_buf_len <= MHLEN) {
1274 m_new = m_gethdr(M_NOWAIT, MT_DATA);
1275 if (m_new == NULL) {
1276 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1279 memcpy(mtod(m_new, void *), packet->data,
1280 packet->tot_data_buf_len);
1281 m_new->m_pkthdr.len = m_new->m_len = packet->tot_data_buf_len;
1282 rxr->hn_small_pkts++;
1285 * Get an mbuf with a cluster. For packets 2K or less,
1286 * get a standard 2K cluster. For anything larger, get a
1287 * 4K cluster. Any buffers larger than 4K can cause problems
1288 * if looped around to the Hyper-V TX channel, so avoid them.
1291 if (packet->tot_data_buf_len > MCLBYTES) {
1293 size = MJUMPAGESIZE;
1296 m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, size);
1297 if (m_new == NULL) {
1298 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1302 hv_m_append(m_new, packet->tot_data_buf_len, packet->data);
1304 m_new->m_pkthdr.rcvif = ifp;
1306 if (__predict_false((ifp->if_capenable & IFCAP_RXCSUM) == 0))
1309 /* receive side checksum offload */
1310 if (csum_info != NULL) {
1311 /* IP csum offload */
1312 if (csum_info->receive.ip_csum_succeeded && do_csum) {
1313 m_new->m_pkthdr.csum_flags |=
1314 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1318 /* TCP/UDP csum offload */
1319 if ((csum_info->receive.tcp_csum_succeeded ||
1320 csum_info->receive.udp_csum_succeeded) && do_csum) {
1321 m_new->m_pkthdr.csum_flags |=
1322 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1323 m_new->m_pkthdr.csum_data = 0xffff;
1324 if (csum_info->receive.tcp_csum_succeeded)
1330 if (csum_info->receive.ip_csum_succeeded &&
1331 csum_info->receive.tcp_csum_succeeded)
1334 const struct ether_header *eh;
1339 if (m_new->m_len < hoff)
1341 eh = mtod(m_new, struct ether_header *);
1342 etype = ntohs(eh->ether_type);
1343 if (etype == ETHERTYPE_VLAN) {
1344 const struct ether_vlan_header *evl;
1346 hoff = sizeof(*evl);
1347 if (m_new->m_len < hoff)
1349 evl = mtod(m_new, struct ether_vlan_header *);
1350 etype = ntohs(evl->evl_proto);
1353 if (etype == ETHERTYPE_IP) {
1356 pr = hn_check_iplen(m_new, hoff);
1357 if (pr == IPPROTO_TCP) {
1359 (rxr->hn_trust_hcsum &
1360 HN_TRUST_HCSUM_TCP)) {
1361 rxr->hn_csum_trusted++;
1362 m_new->m_pkthdr.csum_flags |=
1363 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1364 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1365 m_new->m_pkthdr.csum_data = 0xffff;
1367 /* Rely on SW csum verification though... */
1369 } else if (pr == IPPROTO_UDP) {
1371 (rxr->hn_trust_hcsum &
1372 HN_TRUST_HCSUM_UDP)) {
1373 rxr->hn_csum_trusted++;
1374 m_new->m_pkthdr.csum_flags |=
1375 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1376 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1377 m_new->m_pkthdr.csum_data = 0xffff;
1379 } else if (pr != IPPROTO_DONE && do_csum &&
1380 (rxr->hn_trust_hcsum & HN_TRUST_HCSUM_IP)) {
1381 rxr->hn_csum_trusted++;
1382 m_new->m_pkthdr.csum_flags |=
1383 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1388 if ((packet->vlan_tci != 0) &&
1389 (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0) {
1390 m_new->m_pkthdr.ether_vtag = packet->vlan_tci;
1391 m_new->m_flags |= M_VLANTAG;
1394 m_new->m_pkthdr.flowid = rxr->hn_rx_idx;
1395 M_HASHTYPE_SET(m_new, M_HASHTYPE_OPAQUE);
1398 * Note: Moved RX completion back to hv_nv_on_receive() so all
1399 * messages (not just data messages) will trigger a response.
1405 if ((ifp->if_capenable & IFCAP_LRO) && do_lro) {
1406 #if defined(INET) || defined(INET6)
1407 struct lro_ctrl *lro = &rxr->hn_lro;
1410 rxr->hn_lro_tried++;
1411 if (tcp_lro_rx(lro, m_new, 0) == 0) {
1419 /* We're not holding the lock here, so don't release it */
1420 (*ifp->if_input)(ifp, m_new);
1426 * Rules for using sc->temp_unusable:
1427 * 1. sc->temp_unusable can only be read or written while holding NV_LOCK()
1428 * 2. code reading sc->temp_unusable under NV_LOCK(), and finding
1429 * sc->temp_unusable set, must release NV_LOCK() and exit
1430 * 3. to retain exclusive control of the interface,
1431 * sc->temp_unusable must be set by code before releasing NV_LOCK()
1432 * 4. only code setting sc->temp_unusable can clear sc->temp_unusable
1433 * 5. code setting sc->temp_unusable must eventually clear sc->temp_unusable
1437 * Standard ioctl entry point. Called when the user wants to configure
1441 hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1443 hn_softc_t *sc = ifp->if_softc;
1444 struct ifreq *ifr = (struct ifreq *)data;
1446 struct ifaddr *ifa = (struct ifaddr *)data;
1448 netvsc_device_info device_info;
1449 struct hv_device *hn_dev;
1450 int mask, error = 0;
1451 int retry_cnt = 500;
1457 if (ifa->ifa_addr->sa_family == AF_INET) {
1458 ifp->if_flags |= IFF_UP;
1459 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1461 arp_ifinit(ifp, ifa);
1464 error = ether_ioctl(ifp, cmd, data);
1467 hn_dev = vmbus_get_devctx(sc->hn_dev);
1469 /* Check MTU value change */
1470 if (ifp->if_mtu == ifr->ifr_mtu)
1473 if (ifr->ifr_mtu > NETVSC_MAX_CONFIGURABLE_MTU) {
1478 /* Obtain and record requested MTU */
1479 ifp->if_mtu = ifr->ifr_mtu;
1481 #if __FreeBSD_version >= 1100099
1483 * Make sure that LRO aggregation length limit is still
1484 * valid, after the MTU change.
1487 if (sc->hn_rx_ring[0].hn_lro.lro_length_lim <
1488 HN_LRO_LENLIM_MIN(ifp))
1489 hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MIN(ifp));
1495 if (!sc->temp_unusable) {
1496 sc->temp_unusable = TRUE;
1500 if (retry_cnt > 0) {
1504 } while (retry_cnt > 0);
1506 if (retry_cnt == 0) {
1511 /* We must remove and add back the device to cause the new
1512 * MTU to take effect. This includes tearing down, but not
1513 * deleting the channel, then bringing it back up.
1515 error = hv_rf_on_device_remove(hn_dev, HV_RF_NV_RETAIN_CHANNEL);
1518 sc->temp_unusable = FALSE;
1522 error = hv_rf_on_device_add(hn_dev, &device_info,
1523 sc->hn_rx_ring_inuse);
1526 sc->temp_unusable = FALSE;
1531 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
1532 if (sc->hn_tx_ring[0].hn_tx_chimney_size >
1533 sc->hn_tx_chimney_max)
1534 hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max);
1536 hn_ifinit_locked(sc);
1539 sc->temp_unusable = FALSE;
1545 if (!sc->temp_unusable) {
1546 sc->temp_unusable = TRUE;
1550 if (retry_cnt > 0) {
1554 } while (retry_cnt > 0);
1556 if (retry_cnt == 0) {
1561 if (ifp->if_flags & IFF_UP) {
1563 * If only the state of the PROMISC flag changed,
1564 * then just use the 'set promisc mode' command
1565 * instead of reinitializing the entire NIC. Doing
1566 * a full re-init means reloading the firmware and
1567 * waiting for it to start up, which may take a
1571 /* Fixme: Promiscuous mode? */
1572 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1573 ifp->if_flags & IFF_PROMISC &&
1574 !(sc->hn_if_flags & IFF_PROMISC)) {
1575 /* do something here for Hyper-V */
1576 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1577 !(ifp->if_flags & IFF_PROMISC) &&
1578 sc->hn_if_flags & IFF_PROMISC) {
1579 /* do something here for Hyper-V */
1582 hn_ifinit_locked(sc);
1584 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1589 sc->temp_unusable = FALSE;
1591 sc->hn_if_flags = ifp->if_flags;
1597 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1598 if (mask & IFCAP_TXCSUM) {
1599 ifp->if_capenable ^= IFCAP_TXCSUM;
1600 if (ifp->if_capenable & IFCAP_TXCSUM) {
1602 sc->hn_tx_ring[0].hn_csum_assist;
1605 ~sc->hn_tx_ring[0].hn_csum_assist;
1609 if (mask & IFCAP_RXCSUM)
1610 ifp->if_capenable ^= IFCAP_RXCSUM;
1612 if (mask & IFCAP_LRO)
1613 ifp->if_capenable ^= IFCAP_LRO;
1615 if (mask & IFCAP_TSO4) {
1616 ifp->if_capenable ^= IFCAP_TSO4;
1617 if (ifp->if_capenable & IFCAP_TSO4)
1618 ifp->if_hwassist |= CSUM_IP_TSO;
1620 ifp->if_hwassist &= ~CSUM_IP_TSO;
1623 if (mask & IFCAP_TSO6) {
1624 ifp->if_capenable ^= IFCAP_TSO6;
1625 if (ifp->if_capenable & IFCAP_TSO6)
1626 ifp->if_hwassist |= CSUM_IP6_TSO;
1628 ifp->if_hwassist &= ~CSUM_IP6_TSO;
1637 /* Fixme: Multicast mode? */
1638 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1640 netvsc_setmulti(sc);
1649 error = ifmedia_ioctl(ifp, ifr, &sc->hn_media, cmd);
1652 error = ether_ioctl(ifp, cmd, data);
1663 hn_stop(hn_softc_t *sc)
1667 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1672 printf(" Closing Device ...\n");
1674 atomic_clear_int(&ifp->if_drv_flags,
1675 (IFF_DRV_RUNNING | IFF_DRV_OACTIVE));
1676 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
1677 sc->hn_tx_ring[i].hn_oactive = 0;
1679 if_link_state_change(ifp, LINK_STATE_DOWN);
1680 sc->hn_initdone = 0;
1682 ret = hv_rf_on_close(device_ctx);
1686 * FreeBSD transmit entry point
1689 hn_start(struct ifnet *ifp)
1691 struct hn_softc *sc = ifp->if_softc;
1692 struct hn_tx_ring *txr = &sc->hn_tx_ring[0];
1694 if (txr->hn_sched_tx)
1697 if (mtx_trylock(&txr->hn_tx_lock)) {
1700 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1701 mtx_unlock(&txr->hn_tx_lock);
1706 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
1710 hn_start_txeof(struct hn_tx_ring *txr)
1712 struct hn_softc *sc = txr->hn_sc;
1713 struct ifnet *ifp = sc->hn_ifp;
1715 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1717 if (txr->hn_sched_tx)
1720 if (mtx_trylock(&txr->hn_tx_lock)) {
1723 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1724 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1725 mtx_unlock(&txr->hn_tx_lock);
1727 taskqueue_enqueue(txr->hn_tx_taskq,
1733 * Release the OACTIVE earlier, with the hope, that
1734 * others could catch up. The task will clear the
1735 * flag again with the hn_tx_lock to avoid possible
1738 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1739 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
1747 hn_ifinit_locked(hn_softc_t *sc)
1750 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1755 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1759 hv_promisc_mode = 1;
1761 ret = hv_rf_on_open(device_ctx);
1765 sc->hn_initdone = 1;
1768 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1769 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
1770 sc->hn_tx_ring[i].hn_oactive = 0;
1772 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_RUNNING);
1773 if_link_state_change(ifp, LINK_STATE_UP);
1780 hn_ifinit(void *xsc)
1782 hn_softc_t *sc = xsc;
1785 if (sc->temp_unusable) {
1789 sc->temp_unusable = TRUE;
1792 hn_ifinit_locked(sc);
1795 sc->temp_unusable = FALSE;
1804 hn_watchdog(struct ifnet *ifp)
1809 printf("hn%d: watchdog timeout -- resetting\n", sc->hn_unit);
1810 hn_ifinit(sc); /*???*/
1815 #if __FreeBSD_version >= 1100099
1818 hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS)
1820 struct hn_softc *sc = arg1;
1821 unsigned int lenlim;
1824 lenlim = sc->hn_rx_ring[0].hn_lro.lro_length_lim;
1825 error = sysctl_handle_int(oidp, &lenlim, 0, req);
1826 if (error || req->newptr == NULL)
1829 if (lenlim < HN_LRO_LENLIM_MIN(sc->hn_ifp) ||
1830 lenlim > TCP_LRO_LENGTH_MAX)
1834 hn_set_lro_lenlim(sc, lenlim);
1840 hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS)
1842 struct hn_softc *sc = arg1;
1843 int ackcnt, error, i;
1846 * lro_ackcnt_lim is append count limit,
1847 * +1 to turn it into aggregation limit.
1849 ackcnt = sc->hn_rx_ring[0].hn_lro.lro_ackcnt_lim + 1;
1850 error = sysctl_handle_int(oidp, &ackcnt, 0, req);
1851 if (error || req->newptr == NULL)
1854 if (ackcnt < 2 || ackcnt > (TCP_LRO_ACKCNT_MAX + 1))
1858 * Convert aggregation limit back to append
1863 for (i = 0; i < sc->hn_rx_ring_inuse; ++i)
1864 sc->hn_rx_ring[i].hn_lro.lro_ackcnt_lim = ackcnt;
1872 hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS)
1874 struct hn_softc *sc = arg1;
1879 if (sc->hn_rx_ring[0].hn_trust_hcsum & hcsum)
1882 error = sysctl_handle_int(oidp, &on, 0, req);
1883 if (error || req->newptr == NULL)
1887 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1888 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
1891 rxr->hn_trust_hcsum |= hcsum;
1893 rxr->hn_trust_hcsum &= ~hcsum;
1900 hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS)
1902 struct hn_softc *sc = arg1;
1903 int chimney_size, error;
1905 chimney_size = sc->hn_tx_ring[0].hn_tx_chimney_size;
1906 error = sysctl_handle_int(oidp, &chimney_size, 0, req);
1907 if (error || req->newptr == NULL)
1910 if (chimney_size > sc->hn_tx_chimney_max || chimney_size <= 0)
1913 hn_set_tx_chimney_size(sc, chimney_size);
1917 #if __FreeBSD_version < 1100095
1919 hn_rx_stat_int_sysctl(SYSCTL_HANDLER_ARGS)
1921 struct hn_softc *sc = arg1;
1922 int ofs = arg2, i, error;
1923 struct hn_rx_ring *rxr;
1927 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1928 rxr = &sc->hn_rx_ring[i];
1929 stat += *((int *)((uint8_t *)rxr + ofs));
1932 error = sysctl_handle_64(oidp, &stat, 0, req);
1933 if (error || req->newptr == NULL)
1936 /* Zero out this stat. */
1937 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1938 rxr = &sc->hn_rx_ring[i];
1939 *((int *)((uint8_t *)rxr + ofs)) = 0;
1945 hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS)
1947 struct hn_softc *sc = arg1;
1948 int ofs = arg2, i, error;
1949 struct hn_rx_ring *rxr;
1953 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1954 rxr = &sc->hn_rx_ring[i];
1955 stat += *((uint64_t *)((uint8_t *)rxr + ofs));
1958 error = sysctl_handle_64(oidp, &stat, 0, req);
1959 if (error || req->newptr == NULL)
1962 /* Zero out this stat. */
1963 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1964 rxr = &sc->hn_rx_ring[i];
1965 *((uint64_t *)((uint8_t *)rxr + ofs)) = 0;
1973 hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
1975 struct hn_softc *sc = arg1;
1976 int ofs = arg2, i, error;
1977 struct hn_rx_ring *rxr;
1981 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
1982 rxr = &sc->hn_rx_ring[i];
1983 stat += *((u_long *)((uint8_t *)rxr + ofs));
1986 error = sysctl_handle_long(oidp, &stat, 0, req);
1987 if (error || req->newptr == NULL)
1990 /* Zero out this stat. */
1991 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
1992 rxr = &sc->hn_rx_ring[i];
1993 *((u_long *)((uint8_t *)rxr + ofs)) = 0;
1999 hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
2001 struct hn_softc *sc = arg1;
2002 int ofs = arg2, i, error;
2003 struct hn_tx_ring *txr;
2007 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2008 txr = &sc->hn_tx_ring[i];
2009 stat += *((u_long *)((uint8_t *)txr + ofs));
2012 error = sysctl_handle_long(oidp, &stat, 0, req);
2013 if (error || req->newptr == NULL)
2016 /* Zero out this stat. */
2017 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2018 txr = &sc->hn_tx_ring[i];
2019 *((u_long *)((uint8_t *)txr + ofs)) = 0;
2025 hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS)
2027 struct hn_softc *sc = arg1;
2028 int ofs = arg2, i, error, conf;
2029 struct hn_tx_ring *txr;
2031 txr = &sc->hn_tx_ring[0];
2032 conf = *((int *)((uint8_t *)txr + ofs));
2034 error = sysctl_handle_int(oidp, &conf, 0, req);
2035 if (error || req->newptr == NULL)
2039 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2040 txr = &sc->hn_tx_ring[i];
2041 *((int *)((uint8_t *)txr + ofs)) = conf;
2049 hn_check_iplen(const struct mbuf *m, int hoff)
2051 const struct ip *ip;
2052 int len, iphlen, iplen;
2053 const struct tcphdr *th;
2054 int thoff; /* TCP data offset */
2056 len = hoff + sizeof(struct ip);
2058 /* The packet must be at least the size of an IP header. */
2059 if (m->m_pkthdr.len < len)
2060 return IPPROTO_DONE;
2062 /* The fixed IP header must reside completely in the first mbuf. */
2064 return IPPROTO_DONE;
2066 ip = mtodo(m, hoff);
2068 /* Bound check the packet's stated IP header length. */
2069 iphlen = ip->ip_hl << 2;
2070 if (iphlen < sizeof(struct ip)) /* minimum header length */
2071 return IPPROTO_DONE;
2073 /* The full IP header must reside completely in the one mbuf. */
2074 if (m->m_len < hoff + iphlen)
2075 return IPPROTO_DONE;
2077 iplen = ntohs(ip->ip_len);
2080 * Check that the amount of data in the buffers is as
2081 * at least much as the IP header would have us expect.
2083 if (m->m_pkthdr.len < hoff + iplen)
2084 return IPPROTO_DONE;
2087 * Ignore IP fragments.
2089 if (ntohs(ip->ip_off) & (IP_OFFMASK | IP_MF))
2090 return IPPROTO_DONE;
2093 * The TCP/IP or UDP/IP header must be entirely contained within
2094 * the first fragment of a packet.
2098 if (iplen < iphlen + sizeof(struct tcphdr))
2099 return IPPROTO_DONE;
2100 if (m->m_len < hoff + iphlen + sizeof(struct tcphdr))
2101 return IPPROTO_DONE;
2102 th = (const struct tcphdr *)((const uint8_t *)ip + iphlen);
2103 thoff = th->th_off << 2;
2104 if (thoff < sizeof(struct tcphdr) || thoff + iphlen > iplen)
2105 return IPPROTO_DONE;
2106 if (m->m_len < hoff + iphlen + thoff)
2107 return IPPROTO_DONE;
2110 if (iplen < iphlen + sizeof(struct udphdr))
2111 return IPPROTO_DONE;
2112 if (m->m_len < hoff + iphlen + sizeof(struct udphdr))
2113 return IPPROTO_DONE;
2117 return IPPROTO_DONE;
2124 hn_dma_map_paddr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
2126 bus_addr_t *paddr = arg;
2131 KASSERT(nseg == 1, ("too many segments %d!", nseg));
2132 *paddr = segs->ds_addr;
2136 hn_create_rx_data(struct hn_softc *sc, int ring_cnt)
2138 struct sysctl_oid_list *child;
2139 struct sysctl_ctx_list *ctx;
2140 device_t dev = sc->hn_dev;
2141 #if defined(INET) || defined(INET6)
2142 #if __FreeBSD_version >= 1100095
2148 sc->hn_rx_ring_cnt = ring_cnt;
2149 sc->hn_rx_ring_inuse = sc->hn_rx_ring_cnt;
2151 sc->hn_rx_ring = malloc(sizeof(struct hn_rx_ring) * sc->hn_rx_ring_cnt,
2152 M_NETVSC, M_WAITOK | M_ZERO);
2154 #if defined(INET) || defined(INET6)
2155 #if __FreeBSD_version >= 1100095
2156 lroent_cnt = hn_lro_entry_count;
2157 if (lroent_cnt < TCP_LRO_ENTRIES)
2158 lroent_cnt = TCP_LRO_ENTRIES;
2159 device_printf(dev, "LRO: entry count %d\n", lroent_cnt);
2161 #endif /* INET || INET6 */
2163 ctx = device_get_sysctl_ctx(dev);
2164 child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
2166 /* Create dev.hn.UNIT.rx sysctl tree */
2167 sc->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "rx",
2170 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2171 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
2173 if (hn_trust_hosttcp)
2174 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_TCP;
2175 if (hn_trust_hostudp)
2176 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_UDP;
2177 if (hn_trust_hostip)
2178 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_IP;
2179 rxr->hn_ifp = sc->hn_ifp;
2185 #if defined(INET) || defined(INET6)
2186 #if __FreeBSD_version >= 1100095
2187 tcp_lro_init_args(&rxr->hn_lro, sc->hn_ifp, lroent_cnt, 0);
2189 tcp_lro_init(&rxr->hn_lro);
2190 rxr->hn_lro.ifp = sc->hn_ifp;
2192 #if __FreeBSD_version >= 1100099
2193 rxr->hn_lro.lro_length_lim = HN_LRO_LENLIM_DEF;
2194 rxr->hn_lro.lro_ackcnt_lim = HN_LRO_ACKCNT_DEF;
2196 #endif /* INET || INET6 */
2198 if (sc->hn_rx_sysctl_tree != NULL) {
2202 * Create per RX ring sysctl tree:
2203 * dev.hn.UNIT.rx.RINGID
2205 snprintf(name, sizeof(name), "%d", i);
2206 rxr->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx,
2207 SYSCTL_CHILDREN(sc->hn_rx_sysctl_tree),
2208 OID_AUTO, name, CTLFLAG_RD, 0, "");
2210 if (rxr->hn_rx_sysctl_tree != NULL) {
2211 SYSCTL_ADD_ULONG(ctx,
2212 SYSCTL_CHILDREN(rxr->hn_rx_sysctl_tree),
2213 OID_AUTO, "packets", CTLFLAG_RW,
2214 &rxr->hn_pkts, "# of packets received");
2219 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_queued",
2220 CTLTYPE_U64 | CTLFLAG_RW, sc,
2221 __offsetof(struct hn_rx_ring, hn_lro.lro_queued),
2222 #if __FreeBSD_version < 1100095
2223 hn_rx_stat_int_sysctl,
2225 hn_rx_stat_u64_sysctl,
2227 "LU", "LRO queued");
2228 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_flushed",
2229 CTLTYPE_U64 | CTLFLAG_RW, sc,
2230 __offsetof(struct hn_rx_ring, hn_lro.lro_flushed),
2231 #if __FreeBSD_version < 1100095
2232 hn_rx_stat_int_sysctl,
2234 hn_rx_stat_u64_sysctl,
2236 "LU", "LRO flushed");
2237 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_tried",
2238 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2239 __offsetof(struct hn_rx_ring, hn_lro_tried),
2240 hn_rx_stat_ulong_sysctl, "LU", "# of LRO tries");
2241 #if __FreeBSD_version >= 1100099
2242 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_length_lim",
2243 CTLTYPE_UINT | CTLFLAG_RW, sc, 0, hn_lro_lenlim_sysctl, "IU",
2244 "Max # of data bytes to be aggregated by LRO");
2245 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_ackcnt_lim",
2246 CTLTYPE_INT | CTLFLAG_RW, sc, 0, hn_lro_ackcnt_sysctl, "I",
2247 "Max # of ACKs to be aggregated by LRO");
2249 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hosttcp",
2250 CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_TCP,
2251 hn_trust_hcsum_sysctl, "I",
2252 "Trust tcp segement verification on host side, "
2253 "when csum info is missing");
2254 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostudp",
2255 CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_UDP,
2256 hn_trust_hcsum_sysctl, "I",
2257 "Trust udp datagram verification on host side, "
2258 "when csum info is missing");
2259 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostip",
2260 CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_IP,
2261 hn_trust_hcsum_sysctl, "I",
2262 "Trust ip packet verification on host side, "
2263 "when csum info is missing");
2264 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_ip",
2265 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2266 __offsetof(struct hn_rx_ring, hn_csum_ip),
2267 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM IP");
2268 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_tcp",
2269 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2270 __offsetof(struct hn_rx_ring, hn_csum_tcp),
2271 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM TCP");
2272 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_udp",
2273 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2274 __offsetof(struct hn_rx_ring, hn_csum_udp),
2275 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM UDP");
2276 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_trusted",
2277 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2278 __offsetof(struct hn_rx_ring, hn_csum_trusted),
2279 hn_rx_stat_ulong_sysctl, "LU",
2280 "# of packets that we trust host's csum verification");
2281 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "small_pkts",
2282 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2283 __offsetof(struct hn_rx_ring, hn_small_pkts),
2284 hn_rx_stat_ulong_sysctl, "LU", "# of small packets received");
2285 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_cnt",
2286 CTLFLAG_RD, &sc->hn_rx_ring_cnt, 0, "# created RX rings");
2287 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_inuse",
2288 CTLFLAG_RD, &sc->hn_rx_ring_inuse, 0, "# used RX rings");
2292 hn_destroy_rx_data(struct hn_softc *sc)
2294 #if defined(INET) || defined(INET6)
2298 if (sc->hn_rx_ring_cnt == 0)
2301 #if defined(INET) || defined(INET6)
2302 for (i = 0; i < sc->hn_rx_ring_cnt; ++i)
2303 tcp_lro_free(&sc->hn_rx_ring[i].hn_lro);
2305 free(sc->hn_rx_ring, M_NETVSC);
2306 sc->hn_rx_ring = NULL;
2308 sc->hn_rx_ring_cnt = 0;
2309 sc->hn_rx_ring_inuse = 0;
2313 hn_create_tx_ring(struct hn_softc *sc, int id)
2315 struct hn_tx_ring *txr = &sc->hn_tx_ring[id];
2316 bus_dma_tag_t parent_dtag;
2320 txr->hn_tx_idx = id;
2322 #ifndef HN_USE_TXDESC_BUFRING
2323 mtx_init(&txr->hn_txlist_spin, "hn txlist", NULL, MTX_SPIN);
2325 mtx_init(&txr->hn_tx_lock, "hn tx", NULL, MTX_DEF);
2327 txr->hn_txdesc_cnt = HN_TX_DESC_CNT;
2328 txr->hn_txdesc = malloc(sizeof(struct hn_txdesc) * txr->hn_txdesc_cnt,
2329 M_NETVSC, M_WAITOK | M_ZERO);
2330 #ifndef HN_USE_TXDESC_BUFRING
2331 SLIST_INIT(&txr->hn_txlist);
2333 txr->hn_txdesc_br = buf_ring_alloc(txr->hn_txdesc_cnt, M_NETVSC,
2334 M_WAITOK, &txr->hn_tx_lock);
2337 txr->hn_tx_taskq = sc->hn_tx_taskq;
2339 if (hn_use_if_start) {
2340 txr->hn_txeof = hn_start_txeof;
2341 TASK_INIT(&txr->hn_tx_task, 0, hn_start_taskfunc, txr);
2342 TASK_INIT(&txr->hn_txeof_task, 0, hn_start_txeof_taskfunc, txr);
2344 txr->hn_txeof = hn_xmit_txeof;
2345 TASK_INIT(&txr->hn_tx_task, 0, hn_xmit_taskfunc, txr);
2346 TASK_INIT(&txr->hn_txeof_task, 0, hn_xmit_txeof_taskfunc, txr);
2347 txr->hn_mbuf_br = buf_ring_alloc(txr->hn_txdesc_cnt, M_NETVSC,
2348 M_WAITOK, &txr->hn_tx_lock);
2351 txr->hn_direct_tx_size = hn_direct_tx_size;
2352 if (hv_vmbus_protocal_version >= HV_VMBUS_VERSION_WIN8_1)
2353 txr->hn_csum_assist = HN_CSUM_ASSIST;
2355 txr->hn_csum_assist = HN_CSUM_ASSIST_WIN8;
2358 * Always schedule transmission instead of trying to do direct
2359 * transmission. This one gives the best performance so far.
2361 txr->hn_sched_tx = 1;
2363 parent_dtag = bus_get_dma_tag(sc->hn_dev);
2365 /* DMA tag for RNDIS messages. */
2366 error = bus_dma_tag_create(parent_dtag, /* parent */
2367 HN_RNDIS_MSG_ALIGN, /* alignment */
2368 HN_RNDIS_MSG_BOUNDARY, /* boundary */
2369 BUS_SPACE_MAXADDR, /* lowaddr */
2370 BUS_SPACE_MAXADDR, /* highaddr */
2371 NULL, NULL, /* filter, filterarg */
2372 HN_RNDIS_MSG_LEN, /* maxsize */
2374 HN_RNDIS_MSG_LEN, /* maxsegsize */
2376 NULL, /* lockfunc */
2377 NULL, /* lockfuncarg */
2378 &txr->hn_tx_rndis_dtag);
2380 device_printf(sc->hn_dev, "failed to create rndis dmatag\n");
2384 /* DMA tag for data. */
2385 error = bus_dma_tag_create(parent_dtag, /* parent */
2387 HN_TX_DATA_BOUNDARY, /* boundary */
2388 BUS_SPACE_MAXADDR, /* lowaddr */
2389 BUS_SPACE_MAXADDR, /* highaddr */
2390 NULL, NULL, /* filter, filterarg */
2391 HN_TX_DATA_MAXSIZE, /* maxsize */
2392 HN_TX_DATA_SEGCNT_MAX, /* nsegments */
2393 HN_TX_DATA_SEGSIZE, /* maxsegsize */
2395 NULL, /* lockfunc */
2396 NULL, /* lockfuncarg */
2397 &txr->hn_tx_data_dtag);
2399 device_printf(sc->hn_dev, "failed to create data dmatag\n");
2403 for (i = 0; i < txr->hn_txdesc_cnt; ++i) {
2404 struct hn_txdesc *txd = &txr->hn_txdesc[i];
2409 * Allocate and load RNDIS messages.
2411 error = bus_dmamem_alloc(txr->hn_tx_rndis_dtag,
2412 (void **)&txd->rndis_msg,
2413 BUS_DMA_WAITOK | BUS_DMA_COHERENT,
2414 &txd->rndis_msg_dmap);
2416 device_printf(sc->hn_dev,
2417 "failed to allocate rndis_msg, %d\n", i);
2421 error = bus_dmamap_load(txr->hn_tx_rndis_dtag,
2422 txd->rndis_msg_dmap,
2423 txd->rndis_msg, HN_RNDIS_MSG_LEN,
2424 hn_dma_map_paddr, &txd->rndis_msg_paddr,
2427 device_printf(sc->hn_dev,
2428 "failed to load rndis_msg, %d\n", i);
2429 bus_dmamem_free(txr->hn_tx_rndis_dtag,
2430 txd->rndis_msg, txd->rndis_msg_dmap);
2434 /* DMA map for TX data. */
2435 error = bus_dmamap_create(txr->hn_tx_data_dtag, 0,
2438 device_printf(sc->hn_dev,
2439 "failed to allocate tx data dmamap\n");
2440 bus_dmamap_unload(txr->hn_tx_rndis_dtag,
2441 txd->rndis_msg_dmap);
2442 bus_dmamem_free(txr->hn_tx_rndis_dtag,
2443 txd->rndis_msg, txd->rndis_msg_dmap);
2447 /* All set, put it to list */
2448 txd->flags |= HN_TXD_FLAG_ONLIST;
2449 #ifndef HN_USE_TXDESC_BUFRING
2450 SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
2452 buf_ring_enqueue(txr->hn_txdesc_br, txd);
2455 txr->hn_txdesc_avail = txr->hn_txdesc_cnt;
2457 if (sc->hn_tx_sysctl_tree != NULL) {
2458 struct sysctl_oid_list *child;
2459 struct sysctl_ctx_list *ctx;
2463 * Create per TX ring sysctl tree:
2464 * dev.hn.UNIT.tx.RINGID
2466 ctx = device_get_sysctl_ctx(sc->hn_dev);
2467 child = SYSCTL_CHILDREN(sc->hn_tx_sysctl_tree);
2469 snprintf(name, sizeof(name), "%d", id);
2470 txr->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO,
2471 name, CTLFLAG_RD, 0, "");
2473 if (txr->hn_tx_sysctl_tree != NULL) {
2474 child = SYSCTL_CHILDREN(txr->hn_tx_sysctl_tree);
2476 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_avail",
2477 CTLFLAG_RD, &txr->hn_txdesc_avail, 0,
2478 "# of available TX descs");
2479 if (!hn_use_if_start) {
2480 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "oactive",
2481 CTLFLAG_RD, &txr->hn_oactive, 0,
2484 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "packets",
2485 CTLFLAG_RW, &txr->hn_pkts,
2486 "# of packets transmitted");
2494 hn_txdesc_dmamap_destroy(struct hn_txdesc *txd)
2496 struct hn_tx_ring *txr = txd->txr;
2498 KASSERT(txd->m == NULL, ("still has mbuf installed"));
2499 KASSERT((txd->flags & HN_TXD_FLAG_DMAMAP) == 0, ("still dma mapped"));
2501 bus_dmamap_unload(txr->hn_tx_rndis_dtag, txd->rndis_msg_dmap);
2502 bus_dmamem_free(txr->hn_tx_rndis_dtag, txd->rndis_msg,
2503 txd->rndis_msg_dmap);
2504 bus_dmamap_destroy(txr->hn_tx_data_dtag, txd->data_dmap);
2508 hn_destroy_tx_ring(struct hn_tx_ring *txr)
2510 struct hn_txdesc *txd;
2512 if (txr->hn_txdesc == NULL)
2515 #ifndef HN_USE_TXDESC_BUFRING
2516 while ((txd = SLIST_FIRST(&txr->hn_txlist)) != NULL) {
2517 SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
2518 hn_txdesc_dmamap_destroy(txd);
2521 mtx_lock(&txr->hn_tx_lock);
2522 while ((txd = buf_ring_dequeue_sc(txr->hn_txdesc_br)) != NULL)
2523 hn_txdesc_dmamap_destroy(txd);
2524 mtx_unlock(&txr->hn_tx_lock);
2527 if (txr->hn_tx_data_dtag != NULL)
2528 bus_dma_tag_destroy(txr->hn_tx_data_dtag);
2529 if (txr->hn_tx_rndis_dtag != NULL)
2530 bus_dma_tag_destroy(txr->hn_tx_rndis_dtag);
2532 #ifdef HN_USE_TXDESC_BUFRING
2533 buf_ring_free(txr->hn_txdesc_br, M_NETVSC);
2536 free(txr->hn_txdesc, M_NETVSC);
2537 txr->hn_txdesc = NULL;
2539 if (txr->hn_mbuf_br != NULL)
2540 buf_ring_free(txr->hn_mbuf_br, M_NETVSC);
2542 #ifndef HN_USE_TXDESC_BUFRING
2543 mtx_destroy(&txr->hn_txlist_spin);
2545 mtx_destroy(&txr->hn_tx_lock);
2549 hn_create_tx_data(struct hn_softc *sc, int ring_cnt)
2551 struct sysctl_oid_list *child;
2552 struct sysctl_ctx_list *ctx;
2555 sc->hn_tx_ring_cnt = ring_cnt;
2556 sc->hn_tx_ring_inuse = sc->hn_tx_ring_cnt;
2558 sc->hn_tx_ring = malloc(sizeof(struct hn_tx_ring) * sc->hn_tx_ring_cnt,
2559 M_NETVSC, M_WAITOK | M_ZERO);
2561 ctx = device_get_sysctl_ctx(sc->hn_dev);
2562 child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->hn_dev));
2564 /* Create dev.hn.UNIT.tx sysctl tree */
2565 sc->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "tx",
2568 for (i = 0; i < sc->hn_tx_ring_cnt; ++i) {
2571 error = hn_create_tx_ring(sc, i);
2576 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "no_txdescs",
2577 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2578 __offsetof(struct hn_tx_ring, hn_no_txdescs),
2579 hn_tx_stat_ulong_sysctl, "LU", "# of times short of TX descs");
2580 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "send_failed",
2581 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2582 __offsetof(struct hn_tx_ring, hn_send_failed),
2583 hn_tx_stat_ulong_sysctl, "LU", "# of hyper-v sending failure");
2584 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "txdma_failed",
2585 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2586 __offsetof(struct hn_tx_ring, hn_txdma_failed),
2587 hn_tx_stat_ulong_sysctl, "LU", "# of TX DMA failure");
2588 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_collapsed",
2589 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2590 __offsetof(struct hn_tx_ring, hn_tx_collapsed),
2591 hn_tx_stat_ulong_sysctl, "LU", "# of TX mbuf collapsed");
2592 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney",
2593 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2594 __offsetof(struct hn_tx_ring, hn_tx_chimney),
2595 hn_tx_stat_ulong_sysctl, "LU", "# of chimney send");
2596 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_cnt",
2597 CTLFLAG_RD, &sc->hn_tx_ring[0].hn_txdesc_cnt, 0,
2598 "# of total TX descs");
2599 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_chimney_max",
2600 CTLFLAG_RD, &sc->hn_tx_chimney_max, 0,
2601 "Chimney send packet size upper boundary");
2602 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_size",
2603 CTLTYPE_INT | CTLFLAG_RW, sc, 0, hn_tx_chimney_size_sysctl,
2604 "I", "Chimney send packet size limit");
2605 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "direct_tx_size",
2606 CTLTYPE_INT | CTLFLAG_RW, sc,
2607 __offsetof(struct hn_tx_ring, hn_direct_tx_size),
2608 hn_tx_conf_int_sysctl, "I",
2609 "Size of the packet for direct transmission");
2610 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "sched_tx",
2611 CTLTYPE_INT | CTLFLAG_RW, sc,
2612 __offsetof(struct hn_tx_ring, hn_sched_tx),
2613 hn_tx_conf_int_sysctl, "I",
2614 "Always schedule transmission "
2615 "instead of doing direct transmission");
2616 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_cnt",
2617 CTLFLAG_RD, &sc->hn_tx_ring_cnt, 0, "# created TX rings");
2618 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_inuse",
2619 CTLFLAG_RD, &sc->hn_tx_ring_inuse, 0, "# used TX rings");
2625 hn_set_tx_chimney_size(struct hn_softc *sc, int chimney_size)
2630 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
2631 sc->hn_tx_ring[i].hn_tx_chimney_size = chimney_size;
2636 hn_destroy_tx_data(struct hn_softc *sc)
2640 if (sc->hn_tx_ring_cnt == 0)
2643 for (i = 0; i < sc->hn_tx_ring_cnt; ++i)
2644 hn_destroy_tx_ring(&sc->hn_tx_ring[i]);
2646 free(sc->hn_tx_ring, M_NETVSC);
2647 sc->hn_tx_ring = NULL;
2649 sc->hn_tx_ring_cnt = 0;
2650 sc->hn_tx_ring_inuse = 0;
2654 hn_start_taskfunc(void *xtxr, int pending __unused)
2656 struct hn_tx_ring *txr = xtxr;
2658 mtx_lock(&txr->hn_tx_lock);
2659 hn_start_locked(txr, 0);
2660 mtx_unlock(&txr->hn_tx_lock);
2664 hn_start_txeof_taskfunc(void *xtxr, int pending __unused)
2666 struct hn_tx_ring *txr = xtxr;
2668 mtx_lock(&txr->hn_tx_lock);
2669 atomic_clear_int(&txr->hn_sc->hn_ifp->if_drv_flags, IFF_DRV_OACTIVE);
2670 hn_start_locked(txr, 0);
2671 mtx_unlock(&txr->hn_tx_lock);
2675 hn_stop_tx_tasks(struct hn_softc *sc)
2679 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2680 struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
2682 taskqueue_drain(txr->hn_tx_taskq, &txr->hn_tx_task);
2683 taskqueue_drain(txr->hn_tx_taskq, &txr->hn_txeof_task);
2688 hn_xmit(struct hn_tx_ring *txr, int len)
2690 struct hn_softc *sc = txr->hn_sc;
2691 struct ifnet *ifp = sc->hn_ifp;
2692 struct mbuf *m_head;
2694 mtx_assert(&txr->hn_tx_lock, MA_OWNED);
2695 KASSERT(hn_use_if_start == 0,
2696 ("hn_xmit is called, when if_start is enabled"));
2698 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || txr->hn_oactive)
2701 while ((m_head = drbr_peek(ifp, txr->hn_mbuf_br)) != NULL) {
2702 struct hn_txdesc *txd;
2705 if (len > 0 && m_head->m_pkthdr.len > len) {
2707 * This sending could be time consuming; let callers
2708 * dispatch this packet sending (and sending of any
2709 * following up packets) to tx taskqueue.
2711 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2715 txd = hn_txdesc_get(txr);
2717 txr->hn_no_txdescs++;
2718 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2719 txr->hn_oactive = 1;
2723 error = hn_encap(txr, txd, &m_head);
2725 /* Both txd and m_head are freed; discard */
2726 drbr_advance(ifp, txr->hn_mbuf_br);
2730 error = hn_send_pkt(ifp, txr, txd);
2731 if (__predict_false(error)) {
2732 /* txd is freed, but m_head is not */
2733 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2734 txr->hn_oactive = 1;
2739 drbr_advance(ifp, txr->hn_mbuf_br);
2745 hn_transmit(struct ifnet *ifp, struct mbuf *m)
2747 struct hn_softc *sc = ifp->if_softc;
2748 struct hn_tx_ring *txr;
2752 * Select the TX ring based on flowid
2754 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2755 idx = m->m_pkthdr.flowid % sc->hn_tx_ring_inuse;
2756 txr = &sc->hn_tx_ring[idx];
2758 error = drbr_enqueue(ifp, txr->hn_mbuf_br, m);
2762 if (txr->hn_oactive)
2765 if (txr->hn_sched_tx)
2768 if (mtx_trylock(&txr->hn_tx_lock)) {
2771 sched = hn_xmit(txr, txr->hn_direct_tx_size);
2772 mtx_unlock(&txr->hn_tx_lock);
2777 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
2782 hn_xmit_qflush(struct ifnet *ifp)
2784 struct hn_softc *sc = ifp->if_softc;
2787 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2788 struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
2791 mtx_lock(&txr->hn_tx_lock);
2792 while ((m = buf_ring_dequeue_sc(txr->hn_mbuf_br)) != NULL)
2794 mtx_unlock(&txr->hn_tx_lock);
2800 hn_xmit_txeof(struct hn_tx_ring *txr)
2803 if (txr->hn_sched_tx)
2806 if (mtx_trylock(&txr->hn_tx_lock)) {
2809 txr->hn_oactive = 0;
2810 sched = hn_xmit(txr, txr->hn_direct_tx_size);
2811 mtx_unlock(&txr->hn_tx_lock);
2813 taskqueue_enqueue(txr->hn_tx_taskq,
2819 * Release the oactive earlier, with the hope, that
2820 * others could catch up. The task will clear the
2821 * oactive again with the hn_tx_lock to avoid possible
2824 txr->hn_oactive = 0;
2825 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
2830 hn_xmit_taskfunc(void *xtxr, int pending __unused)
2832 struct hn_tx_ring *txr = xtxr;
2834 mtx_lock(&txr->hn_tx_lock);
2836 mtx_unlock(&txr->hn_tx_lock);
2840 hn_xmit_txeof_taskfunc(void *xtxr, int pending __unused)
2842 struct hn_tx_ring *txr = xtxr;
2844 mtx_lock(&txr->hn_tx_lock);
2845 txr->hn_oactive = 0;
2847 mtx_unlock(&txr->hn_tx_lock);
2851 hn_channel_attach(struct hn_softc *sc, struct hv_vmbus_channel *chan)
2853 struct hn_rx_ring *rxr;
2856 idx = chan->offer_msg.offer.sub_channel_index;
2858 KASSERT(idx >= 0 && idx < sc->hn_rx_ring_inuse,
2859 ("invalid channel index %d, should > 0 && < %d",
2860 idx, sc->hn_rx_ring_inuse));
2861 rxr = &sc->hn_rx_ring[idx];
2862 KASSERT((rxr->hn_rx_flags & HN_RX_FLAG_ATTACHED) == 0,
2863 ("RX ring %d already attached", idx));
2864 rxr->hn_rx_flags |= HN_RX_FLAG_ATTACHED;
2866 chan->hv_chan_rxr = rxr;
2868 if_printf(sc->hn_ifp, "link RX ring %d to channel%u\n",
2869 idx, chan->offer_msg.child_rel_id);
2872 if (idx < sc->hn_tx_ring_inuse) {
2873 struct hn_tx_ring *txr = &sc->hn_tx_ring[idx];
2875 KASSERT((txr->hn_tx_flags & HN_TX_FLAG_ATTACHED) == 0,
2876 ("TX ring %d already attached", idx));
2877 txr->hn_tx_flags |= HN_TX_FLAG_ATTACHED;
2879 chan->hv_chan_txr = txr;
2880 txr->hn_chan = chan;
2882 if_printf(sc->hn_ifp, "link TX ring %d to channel%u\n",
2883 idx, chan->offer_msg.child_rel_id);
2887 /* Bind channel to a proper CPU */
2888 vmbus_channel_cpu_set(chan, (sc->hn_cpu + idx) % mp_ncpus);
2892 netvsc_subchan_callback(struct hn_softc *sc, struct hv_vmbus_channel *chan)
2895 KASSERT(!HV_VMBUS_CHAN_ISPRIMARY(chan),
2896 ("subchannel callback on primary channel"));
2897 KASSERT(chan->offer_msg.offer.sub_channel_index > 0,
2898 ("invalid channel subidx %u",
2899 chan->offer_msg.offer.sub_channel_index));
2900 hn_channel_attach(sc, chan);
2904 hn_tx_taskq_create(void *arg __unused)
2906 if (!hn_share_tx_taskq)
2909 hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
2910 taskqueue_thread_enqueue, &hn_tx_taskq);
2911 taskqueue_start_threads(&hn_tx_taskq, 1, PI_NET, "hn tx");
2912 if (hn_bind_tx_taskq >= 0) {
2913 int cpu = hn_bind_tx_taskq;
2914 struct task cpuset_task;
2917 if (cpu > mp_ncpus - 1)
2919 CPU_SETOF(cpu, &cpu_set);
2920 TASK_INIT(&cpuset_task, 0, hn_cpuset_setthread_task, &cpu_set);
2921 taskqueue_enqueue(hn_tx_taskq, &cpuset_task);
2922 taskqueue_drain(hn_tx_taskq, &cpuset_task);
2925 SYSINIT(hn_txtq_create, SI_SUB_DRIVERS, SI_ORDER_FIRST,
2926 hn_tx_taskq_create, NULL);
2929 hn_tx_taskq_destroy(void *arg __unused)
2931 if (hn_tx_taskq != NULL)
2932 taskqueue_free(hn_tx_taskq);
2934 SYSUNINIT(hn_txtq_destroy, SI_SUB_DRIVERS, SI_ORDER_FIRST,
2935 hn_tx_taskq_destroy, NULL);
2937 static device_method_t netvsc_methods[] = {
2938 /* Device interface */
2939 DEVMETHOD(device_probe, netvsc_probe),
2940 DEVMETHOD(device_attach, netvsc_attach),
2941 DEVMETHOD(device_detach, netvsc_detach),
2942 DEVMETHOD(device_shutdown, netvsc_shutdown),
2947 static driver_t netvsc_driver = {
2953 static devclass_t netvsc_devclass;
2955 DRIVER_MODULE(hn, vmbus, netvsc_driver, netvsc_devclass, 0, 0);
2956 MODULE_VERSION(hn, 1);
2957 MODULE_DEPEND(hn, vmbus, 1, 1, 1);