2 * Copyright (c) 2010-2012 Citrix Inc.
3 * Copyright (c) 2009-2012,2016 Microsoft Corp.
4 * Copyright (c) 2012 NetApp Inc.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice unmodified, this list of conditions, and the following
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30 * Copyright (c) 2004-2006 Kip Macy
31 * All rights reserved.
33 * Redistribution and use in source and binary forms, with or without
34 * modification, are permitted provided that the following conditions
36 * 1. Redistributions of source code must retain the above copyright
37 * notice, this list of conditions and the following disclaimer.
38 * 2. Redistributions in binary form must reproduce the above copyright
39 * notice, this list of conditions and the following disclaimer in the
40 * documentation and/or other materials provided with the distribution.
42 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
43 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
44 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
45 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
46 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
47 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
48 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
49 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
50 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
51 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
55 #include <sys/cdefs.h>
56 __FBSDID("$FreeBSD$");
58 #include "opt_inet6.h"
61 #include <sys/param.h>
62 #include <sys/systm.h>
63 #include <sys/sockio.h>
65 #include <sys/malloc.h>
66 #include <sys/module.h>
67 #include <sys/kernel.h>
68 #include <sys/socket.h>
70 #include <sys/queue.h>
73 #include <sys/sysctl.h>
74 #include <sys/buf_ring.h>
77 #include <net/if_arp.h>
78 #include <net/ethernet.h>
79 #include <net/if_dl.h>
80 #include <net/if_media.h>
84 #include <net/if_types.h>
85 #include <net/if_vlan_var.h>
88 #include <netinet/in_systm.h>
89 #include <netinet/in.h>
90 #include <netinet/ip.h>
91 #include <netinet/if_ether.h>
92 #include <netinet/tcp.h>
93 #include <netinet/udp.h>
94 #include <netinet/ip6.h>
97 #include <vm/vm_param.h>
98 #include <vm/vm_kern.h>
101 #include <machine/bus.h>
102 #include <machine/resource.h>
103 #include <machine/frame.h>
104 #include <machine/vmparam.h>
107 #include <sys/rman.h>
108 #include <sys/mutex.h>
109 #include <sys/errno.h>
110 #include <sys/types.h>
111 #include <machine/atomic.h>
113 #include <machine/intr_machdep.h>
115 #include <machine/in_cksum.h>
117 #include <dev/hyperv/include/hyperv.h>
118 #include <dev/hyperv/include/hyperv_busdma.h>
119 #include "hv_net_vsc.h"
120 #include "hv_rndis.h"
121 #include "hv_rndis_filter.h"
123 #define hv_chan_rxr hv_chan_priv1
124 #define hv_chan_txr hv_chan_priv2
126 /* Short for Hyper-V network interface */
127 #define NETVSC_DEVNAME "hn"
130 * It looks like offset 0 of buf is reserved to hold the softc pointer.
131 * The sc pointer evidently not needed, and is not presently populated.
132 * The packet offset is where the netvsc_packet starts in the buffer.
134 #define HV_NV_SC_PTR_OFFSET_IN_BUF 0
135 #define HV_NV_PACKET_OFFSET_IN_BUF 16
137 /* YYY should get it from the underlying channel */
138 #define HN_TX_DESC_CNT 512
140 #define HN_LROENT_CNT_DEF 128
142 #define HN_RING_CNT_DEF_MAX 8
144 #define HN_RNDIS_MSG_LEN \
145 (sizeof(rndis_msg) + \
146 RNDIS_HASHVAL_PPI_SIZE + \
147 RNDIS_VLAN_PPI_SIZE + \
148 RNDIS_TSO_PPI_SIZE + \
150 #define HN_RNDIS_MSG_BOUNDARY PAGE_SIZE
151 #define HN_RNDIS_MSG_ALIGN CACHE_LINE_SIZE
153 #define HN_TX_DATA_BOUNDARY PAGE_SIZE
154 #define HN_TX_DATA_MAXSIZE IP_MAXPACKET
155 #define HN_TX_DATA_SEGSIZE PAGE_SIZE
156 #define HN_TX_DATA_SEGCNT_MAX \
157 (NETVSC_PACKET_MAXPAGE - HV_RF_NUM_TX_RESERVED_PAGE_BUFS)
159 #define HN_DIRECT_TX_SIZE_DEF 128
161 #define HN_EARLY_TXEOF_THRESH 8
164 #ifndef HN_USE_TXDESC_BUFRING
165 SLIST_ENTRY(hn_txdesc) link;
168 struct hn_tx_ring *txr;
170 uint32_t flags; /* HN_TXD_FLAG_ */
171 netvsc_packet netvsc_pkt; /* XXX to be removed */
173 bus_dmamap_t data_dmap;
175 bus_addr_t rndis_msg_paddr;
176 rndis_msg *rndis_msg;
177 bus_dmamap_t rndis_msg_dmap;
180 #define HN_TXD_FLAG_ONLIST 0x1
181 #define HN_TXD_FLAG_DMAMAP 0x2
184 * Only enable UDP checksum offloading when it is on 2012R2 or
185 * later. UDP checksum offloading doesn't work on earlier
188 #define HN_CSUM_ASSIST_WIN8 (CSUM_IP | CSUM_TCP)
189 #define HN_CSUM_ASSIST (CSUM_IP | CSUM_UDP | CSUM_TCP)
191 #define HN_LRO_LENLIM_MULTIRX_DEF (12 * ETHERMTU)
192 #define HN_LRO_LENLIM_DEF (25 * ETHERMTU)
193 /* YYY 2*MTU is a bit rough, but should be good enough. */
194 #define HN_LRO_LENLIM_MIN(ifp) (2 * (ifp)->if_mtu)
196 #define HN_LRO_ACKCNT_DEF 1
199 * Be aware that this sleepable mutex will exhibit WITNESS errors when
200 * certain TCP and ARP code paths are taken. This appears to be a
201 * well-known condition, as all other drivers checked use a sleeping
202 * mutex to protect their transmit paths.
203 * Also Be aware that mutexes do not play well with semaphores, and there
204 * is a conflicting semaphore in a certain channel code path.
206 #define NV_LOCK_INIT(_sc, _name) \
207 mtx_init(&(_sc)->hn_lock, _name, MTX_NETWORK_LOCK, MTX_DEF)
208 #define NV_LOCK(_sc) mtx_lock(&(_sc)->hn_lock)
209 #define NV_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->hn_lock, MA_OWNED)
210 #define NV_UNLOCK(_sc) mtx_unlock(&(_sc)->hn_lock)
211 #define NV_LOCK_DESTROY(_sc) mtx_destroy(&(_sc)->hn_lock)
218 int hv_promisc_mode = 0; /* normal mode by default */
220 SYSCTL_NODE(_hw, OID_AUTO, hn, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
221 "Hyper-V network interface");
223 /* Trust tcp segements verification on host side. */
224 static int hn_trust_hosttcp = 1;
225 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hosttcp, CTLFLAG_RDTUN,
226 &hn_trust_hosttcp, 0,
227 "Trust tcp segement verification on host side, "
228 "when csum info is missing (global setting)");
230 /* Trust udp datagrams verification on host side. */
231 static int hn_trust_hostudp = 1;
232 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostudp, CTLFLAG_RDTUN,
233 &hn_trust_hostudp, 0,
234 "Trust udp datagram verification on host side, "
235 "when csum info is missing (global setting)");
237 /* Trust ip packets verification on host side. */
238 static int hn_trust_hostip = 1;
239 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostip, CTLFLAG_RDTUN,
241 "Trust ip packet verification on host side, "
242 "when csum info is missing (global setting)");
244 /* Limit TSO burst size */
245 static int hn_tso_maxlen = 0;
246 SYSCTL_INT(_hw_hn, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN,
247 &hn_tso_maxlen, 0, "TSO burst limit");
249 /* Limit chimney send size */
250 static int hn_tx_chimney_size = 0;
251 SYSCTL_INT(_hw_hn, OID_AUTO, tx_chimney_size, CTLFLAG_RDTUN,
252 &hn_tx_chimney_size, 0, "Chimney send packet size limit");
254 /* Limit the size of packet for direct transmission */
255 static int hn_direct_tx_size = HN_DIRECT_TX_SIZE_DEF;
256 SYSCTL_INT(_hw_hn, OID_AUTO, direct_tx_size, CTLFLAG_RDTUN,
257 &hn_direct_tx_size, 0, "Size of the packet for direct transmission");
259 #if defined(INET) || defined(INET6)
260 #if __FreeBSD_version >= 1100095
261 static int hn_lro_entry_count = HN_LROENT_CNT_DEF;
262 SYSCTL_INT(_hw_hn, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN,
263 &hn_lro_entry_count, 0, "LRO entry count");
267 static int hn_share_tx_taskq = 0;
268 SYSCTL_INT(_hw_hn, OID_AUTO, share_tx_taskq, CTLFLAG_RDTUN,
269 &hn_share_tx_taskq, 0, "Enable shared TX taskqueue");
271 static struct taskqueue *hn_tx_taskq;
273 #ifndef HN_USE_TXDESC_BUFRING
274 static int hn_use_txdesc_bufring = 0;
276 static int hn_use_txdesc_bufring = 1;
278 SYSCTL_INT(_hw_hn, OID_AUTO, use_txdesc_bufring, CTLFLAG_RD,
279 &hn_use_txdesc_bufring, 0, "Use buf_ring for TX descriptors");
281 static int hn_bind_tx_taskq = -1;
282 SYSCTL_INT(_hw_hn, OID_AUTO, bind_tx_taskq, CTLFLAG_RDTUN,
283 &hn_bind_tx_taskq, 0, "Bind TX taskqueue to the specified cpu");
285 static int hn_use_if_start = 0;
286 SYSCTL_INT(_hw_hn, OID_AUTO, use_if_start, CTLFLAG_RDTUN,
287 &hn_use_if_start, 0, "Use if_start TX method");
289 static int hn_chan_cnt = 0;
290 SYSCTL_INT(_hw_hn, OID_AUTO, chan_cnt, CTLFLAG_RDTUN,
292 "# of channels to use; each channel has one RX ring and one TX ring");
294 static int hn_tx_ring_cnt = 0;
295 SYSCTL_INT(_hw_hn, OID_AUTO, tx_ring_cnt, CTLFLAG_RDTUN,
296 &hn_tx_ring_cnt, 0, "# of TX rings to use");
298 static int hn_tx_swq_depth = 0;
299 SYSCTL_INT(_hw_hn, OID_AUTO, tx_swq_depth, CTLFLAG_RDTUN,
300 &hn_tx_swq_depth, 0, "Depth of IFQ or BUFRING");
302 static u_int hn_cpu_index;
305 * Forward declarations
307 static void hn_stop(hn_softc_t *sc);
308 static void hn_ifinit_locked(hn_softc_t *sc);
309 static void hn_ifinit(void *xsc);
310 static int hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data);
311 static int hn_start_locked(struct hn_tx_ring *txr, int len);
312 static void hn_start(struct ifnet *ifp);
313 static void hn_start_txeof(struct hn_tx_ring *);
314 static int hn_ifmedia_upd(struct ifnet *ifp);
315 static void hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr);
316 #if __FreeBSD_version >= 1100099
317 static int hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS);
318 static int hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS);
320 static int hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS);
321 static int hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS);
322 #if __FreeBSD_version < 1100095
323 static int hn_rx_stat_int_sysctl(SYSCTL_HANDLER_ARGS);
325 static int hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS);
327 static int hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS);
328 static int hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS);
329 static int hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS);
330 static int hn_check_iplen(const struct mbuf *, int);
331 static int hn_create_tx_ring(struct hn_softc *, int);
332 static void hn_destroy_tx_ring(struct hn_tx_ring *);
333 static int hn_create_tx_data(struct hn_softc *, int);
334 static void hn_destroy_tx_data(struct hn_softc *);
335 static void hn_start_taskfunc(void *, int);
336 static void hn_start_txeof_taskfunc(void *, int);
337 static void hn_stop_tx_tasks(struct hn_softc *);
338 static int hn_encap(struct hn_tx_ring *, struct hn_txdesc *, struct mbuf **);
339 static void hn_create_rx_data(struct hn_softc *sc, int);
340 static void hn_destroy_rx_data(struct hn_softc *sc);
341 static void hn_set_tx_chimney_size(struct hn_softc *, int);
342 static void hn_channel_attach(struct hn_softc *, struct hv_vmbus_channel *);
343 static void hn_subchan_attach(struct hn_softc *, struct hv_vmbus_channel *);
345 static int hn_transmit(struct ifnet *, struct mbuf *);
346 static void hn_xmit_qflush(struct ifnet *);
347 static int hn_xmit(struct hn_tx_ring *, int);
348 static void hn_xmit_txeof(struct hn_tx_ring *);
349 static void hn_xmit_taskfunc(void *, int);
350 static void hn_xmit_txeof_taskfunc(void *, int);
352 #if __FreeBSD_version >= 1100099
354 hn_set_lro_lenlim(struct hn_softc *sc, int lenlim)
358 for (i = 0; i < sc->hn_rx_ring_inuse; ++i)
359 sc->hn_rx_ring[i].hn_lro.lro_length_lim = lenlim;
364 hn_get_txswq_depth(const struct hn_tx_ring *txr)
367 KASSERT(txr->hn_txdesc_cnt > 0, ("tx ring is not setup yet"));
368 if (hn_tx_swq_depth < txr->hn_txdesc_cnt)
369 return txr->hn_txdesc_cnt;
370 return hn_tx_swq_depth;
374 hn_ifmedia_upd(struct ifnet *ifp __unused)
381 hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
383 struct hn_softc *sc = ifp->if_softc;
385 ifmr->ifm_status = IFM_AVALID;
386 ifmr->ifm_active = IFM_ETHER;
388 if (!sc->hn_carrier) {
389 ifmr->ifm_active |= IFM_NONE;
392 ifmr->ifm_status |= IFM_ACTIVE;
393 ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
396 /* {F8615163-DF3E-46c5-913F-F2D2F965ED0E} */
397 static const hv_guid g_net_vsc_device_type = {
398 .data = {0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
399 0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E}
403 * Standard probe entry point.
407 netvsc_probe(device_t dev)
411 p = vmbus_get_type(dev);
412 if (!memcmp(p, &g_net_vsc_device_type.data, sizeof(hv_guid))) {
413 device_set_desc(dev, "Hyper-V Network Interface");
415 printf("Netvsc probe... DONE \n");
417 return (BUS_PROBE_DEFAULT);
424 hn_cpuset_setthread_task(void *xmask, int pending __unused)
426 cpuset_t *mask = xmask;
429 error = cpuset_setthread(curthread->td_tid, mask);
431 panic("curthread=%ju: can't pin; error=%d",
432 (uintmax_t)curthread->td_tid, error);
437 * Standard attach entry point.
439 * Called when the driver is loaded. It allocates needed resources,
440 * and initializes the "hardware" and software.
443 netvsc_attach(device_t dev)
445 struct hv_device *device_ctx = vmbus_get_devctx(dev);
446 struct hv_vmbus_channel *pri_chan;
447 netvsc_device_info device_info;
449 int unit = device_get_unit(dev);
450 struct ifnet *ifp = NULL;
451 int error, ring_cnt, tx_ring_cnt;
454 sc = device_get_softc(dev);
459 if (hn_tx_taskq == NULL) {
460 sc->hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
461 taskqueue_thread_enqueue, &sc->hn_tx_taskq);
462 taskqueue_start_threads(&sc->hn_tx_taskq, 1, PI_NET, "%s tx",
463 device_get_nameunit(dev));
464 if (hn_bind_tx_taskq >= 0) {
465 int cpu = hn_bind_tx_taskq;
466 struct task cpuset_task;
469 if (cpu > mp_ncpus - 1)
471 CPU_SETOF(cpu, &cpu_set);
472 TASK_INIT(&cpuset_task, 0, hn_cpuset_setthread_task,
474 taskqueue_enqueue(sc->hn_tx_taskq, &cpuset_task);
475 taskqueue_drain(sc->hn_tx_taskq, &cpuset_task);
478 sc->hn_tx_taskq = hn_tx_taskq;
480 NV_LOCK_INIT(sc, "NetVSCLock");
482 sc->hn_dev_obj = device_ctx;
484 ifp = sc->hn_ifp = sc->arpcom.ac_ifp = if_alloc(IFT_ETHER);
486 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
489 * Figure out the # of RX rings (ring_cnt) and the # of TX rings
490 * to use (tx_ring_cnt).
493 * The # of RX rings to use is same as the # of channels to use.
495 ring_cnt = hn_chan_cnt;
499 if (ring_cnt > HN_RING_CNT_DEF_MAX)
500 ring_cnt = HN_RING_CNT_DEF_MAX;
501 } else if (ring_cnt > mp_ncpus) {
505 tx_ring_cnt = hn_tx_ring_cnt;
506 if (tx_ring_cnt <= 0 || tx_ring_cnt > ring_cnt)
507 tx_ring_cnt = ring_cnt;
508 if (hn_use_if_start) {
509 /* ifnet.if_start only needs one TX ring. */
514 * Set the leader CPU for channels.
516 sc->hn_cpu = atomic_fetchadd_int(&hn_cpu_index, ring_cnt) % mp_ncpus;
518 error = hn_create_tx_data(sc, tx_ring_cnt);
521 hn_create_rx_data(sc, ring_cnt);
524 * Associate the first TX/RX ring w/ the primary channel.
526 pri_chan = device_ctx->channel;
527 KASSERT(HV_VMBUS_CHAN_ISPRIMARY(pri_chan), ("not primary channel"));
528 KASSERT(pri_chan->offer_msg.offer.sub_channel_index == 0,
529 ("primary channel subidx %u",
530 pri_chan->offer_msg.offer.sub_channel_index));
531 hn_channel_attach(sc, pri_chan);
533 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
534 ifp->if_ioctl = hn_ioctl;
535 ifp->if_init = hn_ifinit;
536 /* needed by hv_rf_on_device_add() code */
537 ifp->if_mtu = ETHERMTU;
538 if (hn_use_if_start) {
539 int qdepth = hn_get_txswq_depth(&sc->hn_tx_ring[0]);
541 ifp->if_start = hn_start;
542 IFQ_SET_MAXLEN(&ifp->if_snd, qdepth);
543 ifp->if_snd.ifq_drv_maxlen = qdepth - 1;
544 IFQ_SET_READY(&ifp->if_snd);
546 ifp->if_transmit = hn_transmit;
547 ifp->if_qflush = hn_xmit_qflush;
550 ifmedia_init(&sc->hn_media, 0, hn_ifmedia_upd, hn_ifmedia_sts);
551 ifmedia_add(&sc->hn_media, IFM_ETHER | IFM_AUTO, 0, NULL);
552 ifmedia_set(&sc->hn_media, IFM_ETHER | IFM_AUTO);
553 /* XXX ifmedia_set really should do this for us */
554 sc->hn_media.ifm_media = sc->hn_media.ifm_cur->ifm_media;
557 * Tell upper layers that we support full VLAN capability.
559 ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
560 ifp->if_capabilities |=
561 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
564 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
566 ifp->if_hwassist = sc->hn_tx_ring[0].hn_csum_assist | CSUM_TSO;
568 error = hv_rf_on_device_add(device_ctx, &device_info, ring_cnt);
571 KASSERT(sc->net_dev->num_channel > 0 &&
572 sc->net_dev->num_channel <= sc->hn_rx_ring_inuse,
573 ("invalid channel count %u, should be less than %d",
574 sc->net_dev->num_channel, sc->hn_rx_ring_inuse));
577 * Set the # of TX/RX rings that could be used according to
578 * the # of channels that host offered.
580 if (sc->hn_tx_ring_inuse > sc->net_dev->num_channel)
581 sc->hn_tx_ring_inuse = sc->net_dev->num_channel;
582 sc->hn_rx_ring_inuse = sc->net_dev->num_channel;
583 device_printf(dev, "%d TX ring, %d RX ring\n",
584 sc->hn_tx_ring_inuse, sc->hn_rx_ring_inuse);
586 if (sc->net_dev->num_channel > 1) {
587 struct hv_vmbus_channel **subchan;
588 int subchan_cnt = sc->net_dev->num_channel - 1;
591 /* Wait for sub-channels setup to complete. */
592 subchan = vmbus_get_subchan(pri_chan, subchan_cnt);
594 /* Attach the sub-channels. */
595 for (i = 0; i < subchan_cnt; ++i) {
596 /* NOTE: Calling order is critical. */
597 hn_subchan_attach(sc, subchan[i]);
598 hv_nv_subchan_attach(subchan[i]);
601 /* Release the sub-channels */
602 vmbus_rel_subchan(subchan, subchan_cnt);
603 device_printf(dev, "%d sub-channels setup done\n", subchan_cnt);
606 #if __FreeBSD_version >= 1100099
607 if (sc->hn_rx_ring_inuse > 1) {
609 * Reduce TCP segment aggregation limit for multiple
610 * RX rings to increase ACK timeliness.
612 hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MULTIRX_DEF);
616 if (device_info.link_state == 0) {
620 tso_maxlen = hn_tso_maxlen;
621 if (tso_maxlen <= 0 || tso_maxlen > IP_MAXPACKET)
622 tso_maxlen = IP_MAXPACKET;
624 ifp->if_hw_tsomaxsegcount = HN_TX_DATA_SEGCNT_MAX;
625 ifp->if_hw_tsomaxsegsize = PAGE_SIZE;
626 ifp->if_hw_tsomax = tso_maxlen -
627 (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
629 ether_ifattach(ifp, device_info.mac_addr);
631 if_printf(ifp, "TSO: %u/%u/%u\n", ifp->if_hw_tsomax,
632 ifp->if_hw_tsomaxsegcount, ifp->if_hw_tsomaxsegsize);
634 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
635 hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max);
636 if (hn_tx_chimney_size > 0 &&
637 hn_tx_chimney_size < sc->hn_tx_chimney_max)
638 hn_set_tx_chimney_size(sc, hn_tx_chimney_size);
642 hn_destroy_tx_data(sc);
649 * Standard detach entry point
652 netvsc_detach(device_t dev)
654 struct hn_softc *sc = device_get_softc(dev);
655 struct hv_device *hv_device = vmbus_get_devctx(dev);
658 printf("netvsc_detach\n");
661 * XXXKYS: Need to clean up all our
662 * driver state; this is the driver
667 * XXXKYS: Need to stop outgoing traffic and unregister
671 hv_rf_on_device_remove(hv_device, HV_RF_NV_DESTROY_CHANNEL);
673 hn_stop_tx_tasks(sc);
675 ifmedia_removeall(&sc->hn_media);
676 hn_destroy_rx_data(sc);
677 hn_destroy_tx_data(sc);
679 if (sc->hn_tx_taskq != hn_tx_taskq)
680 taskqueue_free(sc->hn_tx_taskq);
686 * Standard shutdown entry point
689 netvsc_shutdown(device_t dev)
695 hn_txdesc_dmamap_load(struct hn_tx_ring *txr, struct hn_txdesc *txd,
696 struct mbuf **m_head, bus_dma_segment_t *segs, int *nsegs)
698 struct mbuf *m = *m_head;
701 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag, txd->data_dmap,
702 m, segs, nsegs, BUS_DMA_NOWAIT);
703 if (error == EFBIG) {
706 m_new = m_collapse(m, M_NOWAIT, HN_TX_DATA_SEGCNT_MAX);
711 txr->hn_tx_collapsed++;
713 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag,
714 txd->data_dmap, m, segs, nsegs, BUS_DMA_NOWAIT);
717 bus_dmamap_sync(txr->hn_tx_data_dtag, txd->data_dmap,
718 BUS_DMASYNC_PREWRITE);
719 txd->flags |= HN_TXD_FLAG_DMAMAP;
725 hn_txdesc_dmamap_unload(struct hn_tx_ring *txr, struct hn_txdesc *txd)
728 if (txd->flags & HN_TXD_FLAG_DMAMAP) {
729 bus_dmamap_sync(txr->hn_tx_data_dtag,
730 txd->data_dmap, BUS_DMASYNC_POSTWRITE);
731 bus_dmamap_unload(txr->hn_tx_data_dtag,
733 txd->flags &= ~HN_TXD_FLAG_DMAMAP;
738 hn_txdesc_put(struct hn_tx_ring *txr, struct hn_txdesc *txd)
741 KASSERT((txd->flags & HN_TXD_FLAG_ONLIST) == 0,
742 ("put an onlist txd %#x", txd->flags));
744 KASSERT(txd->refs > 0, ("invalid txd refs %d", txd->refs));
745 if (atomic_fetchadd_int(&txd->refs, -1) != 1)
748 hn_txdesc_dmamap_unload(txr, txd);
749 if (txd->m != NULL) {
754 txd->flags |= HN_TXD_FLAG_ONLIST;
756 #ifndef HN_USE_TXDESC_BUFRING
757 mtx_lock_spin(&txr->hn_txlist_spin);
758 KASSERT(txr->hn_txdesc_avail >= 0 &&
759 txr->hn_txdesc_avail < txr->hn_txdesc_cnt,
760 ("txdesc_put: invalid txd avail %d", txr->hn_txdesc_avail));
761 txr->hn_txdesc_avail++;
762 SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
763 mtx_unlock_spin(&txr->hn_txlist_spin);
765 atomic_add_int(&txr->hn_txdesc_avail, 1);
766 buf_ring_enqueue(txr->hn_txdesc_br, txd);
772 static __inline struct hn_txdesc *
773 hn_txdesc_get(struct hn_tx_ring *txr)
775 struct hn_txdesc *txd;
777 #ifndef HN_USE_TXDESC_BUFRING
778 mtx_lock_spin(&txr->hn_txlist_spin);
779 txd = SLIST_FIRST(&txr->hn_txlist);
781 KASSERT(txr->hn_txdesc_avail > 0,
782 ("txdesc_get: invalid txd avail %d", txr->hn_txdesc_avail));
783 txr->hn_txdesc_avail--;
784 SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
786 mtx_unlock_spin(&txr->hn_txlist_spin);
788 txd = buf_ring_dequeue_sc(txr->hn_txdesc_br);
792 #ifdef HN_USE_TXDESC_BUFRING
793 atomic_subtract_int(&txr->hn_txdesc_avail, 1);
795 KASSERT(txd->m == NULL && txd->refs == 0 &&
796 (txd->flags & HN_TXD_FLAG_ONLIST), ("invalid txd"));
797 txd->flags &= ~HN_TXD_FLAG_ONLIST;
804 hn_txdesc_hold(struct hn_txdesc *txd)
807 /* 0->1 transition will never work */
808 KASSERT(txd->refs > 0, ("invalid refs %d", txd->refs));
809 atomic_add_int(&txd->refs, 1);
813 hn_txeof(struct hn_tx_ring *txr)
815 txr->hn_has_txeof = 0;
820 hn_tx_done(struct hv_vmbus_channel *chan, void *xpkt)
822 netvsc_packet *packet = xpkt;
823 struct hn_txdesc *txd;
824 struct hn_tx_ring *txr;
826 txd = (struct hn_txdesc *)(uintptr_t)
827 packet->compl.send.send_completion_tid;
830 KASSERT(txr->hn_chan == chan,
831 ("channel mismatch, on channel%u, should be channel%u",
832 chan->offer_msg.offer.sub_channel_index,
833 txr->hn_chan->offer_msg.offer.sub_channel_index));
835 txr->hn_has_txeof = 1;
836 hn_txdesc_put(txr, txd);
838 ++txr->hn_txdone_cnt;
839 if (txr->hn_txdone_cnt >= HN_EARLY_TXEOF_THRESH) {
840 txr->hn_txdone_cnt = 0;
847 netvsc_channel_rollup(struct hv_vmbus_channel *chan)
849 struct hn_tx_ring *txr = chan->hv_chan_txr;
850 #if defined(INET) || defined(INET6)
851 struct hn_rx_ring *rxr = chan->hv_chan_rxr;
852 struct lro_ctrl *lro = &rxr->hn_lro;
853 struct lro_entry *queued;
855 while ((queued = SLIST_FIRST(&lro->lro_active)) != NULL) {
856 SLIST_REMOVE_HEAD(&lro->lro_active, next);
857 tcp_lro_flush(lro, queued);
863 * 'txr' could be NULL, if multiple channels and
864 * ifnet.if_start method are enabled.
866 if (txr == NULL || !txr->hn_has_txeof)
869 txr->hn_txdone_cnt = 0;
875 * If this function fails, then both txd and m_head0 will be freed.
878 hn_encap(struct hn_tx_ring *txr, struct hn_txdesc *txd, struct mbuf **m_head0)
880 bus_dma_segment_t segs[HN_TX_DATA_SEGCNT_MAX];
882 struct mbuf *m_head = *m_head0;
883 netvsc_packet *packet;
884 rndis_msg *rndis_mesg;
885 rndis_packet *rndis_pkt;
886 rndis_per_packet_info *rppi;
887 struct rndis_hash_value *hash_value;
888 uint32_t rndis_msg_size;
890 packet = &txd->netvsc_pkt;
891 packet->is_data_pkt = TRUE;
892 packet->tot_data_buf_len = m_head->m_pkthdr.len;
895 * extension points to the area reserved for the
896 * rndis_filter_packet, which is placed just after
897 * the netvsc_packet (and rppi struct, if present;
898 * length is updated later).
900 rndis_mesg = txd->rndis_msg;
901 /* XXX not necessary */
902 memset(rndis_mesg, 0, HN_RNDIS_MSG_LEN);
903 rndis_mesg->ndis_msg_type = REMOTE_NDIS_PACKET_MSG;
905 rndis_pkt = &rndis_mesg->msg.packet;
906 rndis_pkt->data_offset = sizeof(rndis_packet);
907 rndis_pkt->data_length = packet->tot_data_buf_len;
908 rndis_pkt->per_pkt_info_offset = sizeof(rndis_packet);
910 rndis_msg_size = RNDIS_MESSAGE_SIZE(rndis_packet);
913 * Set the hash value for this packet, so that the host could
914 * dispatch the TX done event for this packet back to this TX
917 rndis_msg_size += RNDIS_HASHVAL_PPI_SIZE;
918 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_HASHVAL_PPI_SIZE,
920 hash_value = (struct rndis_hash_value *)((uint8_t *)rppi +
921 rppi->per_packet_info_offset);
922 hash_value->hash_value = txr->hn_tx_idx;
924 if (m_head->m_flags & M_VLANTAG) {
925 ndis_8021q_info *rppi_vlan_info;
927 rndis_msg_size += RNDIS_VLAN_PPI_SIZE;
928 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_VLAN_PPI_SIZE,
931 rppi_vlan_info = (ndis_8021q_info *)((uint8_t *)rppi +
932 rppi->per_packet_info_offset);
933 rppi_vlan_info->u1.s1.vlan_id =
934 m_head->m_pkthdr.ether_vtag & 0xfff;
937 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
938 rndis_tcp_tso_info *tso_info;
939 struct ether_vlan_header *eh;
943 * XXX need m_pullup and use mtodo
945 eh = mtod(m_head, struct ether_vlan_header*);
946 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN))
947 ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
949 ether_len = ETHER_HDR_LEN;
951 rndis_msg_size += RNDIS_TSO_PPI_SIZE;
952 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_TSO_PPI_SIZE,
953 tcp_large_send_info);
955 tso_info = (rndis_tcp_tso_info *)((uint8_t *)rppi +
956 rppi->per_packet_info_offset);
957 tso_info->lso_v2_xmit.type =
958 RNDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
961 if (m_head->m_pkthdr.csum_flags & CSUM_IP_TSO) {
963 (struct ip *)(m_head->m_data + ether_len);
964 unsigned long iph_len = ip->ip_hl << 2;
966 (struct tcphdr *)((caddr_t)ip + iph_len);
968 tso_info->lso_v2_xmit.ip_version =
969 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
973 th->th_sum = in_pseudo(ip->ip_src.s_addr,
974 ip->ip_dst.s_addr, htons(IPPROTO_TCP));
977 #if defined(INET6) && defined(INET)
982 struct ip6_hdr *ip6 = (struct ip6_hdr *)
983 (m_head->m_data + ether_len);
984 struct tcphdr *th = (struct tcphdr *)(ip6 + 1);
986 tso_info->lso_v2_xmit.ip_version =
987 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
989 th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
992 tso_info->lso_v2_xmit.tcp_header_offset = 0;
993 tso_info->lso_v2_xmit.mss = m_head->m_pkthdr.tso_segsz;
994 } else if (m_head->m_pkthdr.csum_flags & txr->hn_csum_assist) {
995 rndis_tcp_ip_csum_info *csum_info;
997 rndis_msg_size += RNDIS_CSUM_PPI_SIZE;
998 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_CSUM_PPI_SIZE,
1000 csum_info = (rndis_tcp_ip_csum_info *)((uint8_t *)rppi +
1001 rppi->per_packet_info_offset);
1003 csum_info->xmit.is_ipv4 = 1;
1004 if (m_head->m_pkthdr.csum_flags & CSUM_IP)
1005 csum_info->xmit.ip_header_csum = 1;
1007 if (m_head->m_pkthdr.csum_flags & CSUM_TCP) {
1008 csum_info->xmit.tcp_csum = 1;
1009 csum_info->xmit.tcp_header_offset = 0;
1010 } else if (m_head->m_pkthdr.csum_flags & CSUM_UDP) {
1011 csum_info->xmit.udp_csum = 1;
1015 rndis_mesg->msg_len = packet->tot_data_buf_len + rndis_msg_size;
1016 packet->tot_data_buf_len = rndis_mesg->msg_len;
1019 * Chimney send, if the packet could fit into one chimney buffer.
1021 if (packet->tot_data_buf_len < txr->hn_tx_chimney_size) {
1022 netvsc_dev *net_dev = txr->hn_sc->net_dev;
1023 uint32_t send_buf_section_idx;
1025 txr->hn_tx_chimney_tried++;
1026 send_buf_section_idx =
1027 hv_nv_get_next_send_section(net_dev);
1028 if (send_buf_section_idx !=
1029 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX) {
1030 uint8_t *dest = ((uint8_t *)net_dev->send_buf +
1031 (send_buf_section_idx *
1032 net_dev->send_section_size));
1034 memcpy(dest, rndis_mesg, rndis_msg_size);
1035 dest += rndis_msg_size;
1036 m_copydata(m_head, 0, m_head->m_pkthdr.len, dest);
1038 packet->send_buf_section_idx = send_buf_section_idx;
1039 packet->send_buf_section_size =
1040 packet->tot_data_buf_len;
1041 packet->page_buf_count = 0;
1042 txr->hn_tx_chimney++;
1047 error = hn_txdesc_dmamap_load(txr, txd, &m_head, segs, &nsegs);
1052 * This mbuf is not linked w/ the txd yet, so free it now.
1057 freed = hn_txdesc_put(txr, txd);
1059 ("fail to free txd upon txdma error"));
1061 txr->hn_txdma_failed++;
1062 if_inc_counter(txr->hn_sc->hn_ifp, IFCOUNTER_OERRORS, 1);
1067 packet->page_buf_count = nsegs + HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
1069 /* send packet with page buffer */
1070 packet->page_buffers[0].pfn = atop(txd->rndis_msg_paddr);
1071 packet->page_buffers[0].offset = txd->rndis_msg_paddr & PAGE_MASK;
1072 packet->page_buffers[0].length = rndis_msg_size;
1075 * Fill the page buffers with mbuf info starting at index
1076 * HV_RF_NUM_TX_RESERVED_PAGE_BUFS.
1078 for (i = 0; i < nsegs; ++i) {
1079 hv_vmbus_page_buffer *pb = &packet->page_buffers[
1080 i + HV_RF_NUM_TX_RESERVED_PAGE_BUFS];
1082 pb->pfn = atop(segs[i].ds_addr);
1083 pb->offset = segs[i].ds_addr & PAGE_MASK;
1084 pb->length = segs[i].ds_len;
1087 packet->send_buf_section_idx =
1088 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX;
1089 packet->send_buf_section_size = 0;
1093 /* Set the completion routine */
1094 packet->compl.send.on_send_completion = hn_tx_done;
1095 packet->compl.send.send_completion_context = packet;
1096 packet->compl.send.send_completion_tid = (uint64_t)(uintptr_t)txd;
1103 * If this function fails, then txd will be freed, but the mbuf
1104 * associated w/ the txd will _not_ be freed.
1107 hn_send_pkt(struct ifnet *ifp, struct hn_tx_ring *txr, struct hn_txdesc *txd)
1109 int error, send_failed = 0;
1113 * Make sure that txd is not freed before ETHER_BPF_MTAP.
1115 hn_txdesc_hold(txd);
1116 error = hv_nv_on_send(txr->hn_chan, &txd->netvsc_pkt);
1118 ETHER_BPF_MTAP(ifp, txd->m);
1119 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1120 if (!hn_use_if_start) {
1121 if_inc_counter(ifp, IFCOUNTER_OBYTES,
1122 txd->m->m_pkthdr.len);
1123 if (txd->m->m_flags & M_MCAST)
1124 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
1128 hn_txdesc_put(txr, txd);
1130 if (__predict_false(error)) {
1134 * This should "really rarely" happen.
1136 * XXX Too many RX to be acked or too many sideband
1137 * commands to run? Ask netvsc_channel_rollup()
1138 * to kick start later.
1140 txr->hn_has_txeof = 1;
1142 txr->hn_send_failed++;
1145 * Try sending again after set hn_has_txeof;
1146 * in case that we missed the last
1147 * netvsc_channel_rollup().
1151 if_printf(ifp, "send failed\n");
1154 * Caller will perform further processing on the
1155 * associated mbuf, so don't free it in hn_txdesc_put();
1156 * only unload it from the DMA map in hn_txdesc_put(),
1160 freed = hn_txdesc_put(txr, txd);
1162 ("fail to free txd upon send error"));
1164 txr->hn_send_failed++;
1170 * Start a transmit of one or more packets
1173 hn_start_locked(struct hn_tx_ring *txr, int len)
1175 struct hn_softc *sc = txr->hn_sc;
1176 struct ifnet *ifp = sc->hn_ifp;
1178 KASSERT(hn_use_if_start,
1179 ("hn_start_locked is called, when if_start is disabled"));
1180 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1181 mtx_assert(&txr->hn_tx_lock, MA_OWNED);
1183 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1187 while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
1188 struct hn_txdesc *txd;
1189 struct mbuf *m_head;
1192 IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
1196 if (len > 0 && m_head->m_pkthdr.len > len) {
1198 * This sending could be time consuming; let callers
1199 * dispatch this packet sending (and sending of any
1200 * following up packets) to tx taskqueue.
1202 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1206 txd = hn_txdesc_get(txr);
1208 txr->hn_no_txdescs++;
1209 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1210 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1214 error = hn_encap(txr, txd, &m_head);
1216 /* Both txd and m_head are freed */
1220 error = hn_send_pkt(ifp, txr, txd);
1221 if (__predict_false(error)) {
1222 /* txd is freed, but m_head is not */
1223 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1224 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1232 * Link up/down notification
1235 netvsc_linkstatus_callback(struct hv_device *device_obj, uint32_t status)
1237 hn_softc_t *sc = device_get_softc(device_obj->device);
1247 * Append the specified data to the indicated mbuf chain,
1248 * Extend the mbuf chain if the new data does not fit in
1251 * This is a minor rewrite of m_append() from sys/kern/uipc_mbuf.c.
1252 * There should be an equivalent in the kernel mbuf code,
1253 * but there does not appear to be one yet.
1255 * Differs from m_append() in that additional mbufs are
1256 * allocated with cluster size MJUMPAGESIZE, and filled
1259 * Return 1 if able to complete the job; otherwise 0.
1262 hv_m_append(struct mbuf *m0, int len, c_caddr_t cp)
1265 int remainder, space;
1267 for (m = m0; m->m_next != NULL; m = m->m_next)
1270 space = M_TRAILINGSPACE(m);
1273 * Copy into available space.
1275 if (space > remainder)
1277 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1282 while (remainder > 0) {
1284 * Allocate a new mbuf; could check space
1285 * and allocate a cluster instead.
1287 n = m_getjcl(M_DONTWAIT, m->m_type, 0, MJUMPAGESIZE);
1290 n->m_len = min(MJUMPAGESIZE, remainder);
1291 bcopy(cp, mtod(n, caddr_t), n->m_len);
1293 remainder -= n->m_len;
1297 if (m0->m_flags & M_PKTHDR)
1298 m0->m_pkthdr.len += len - remainder;
1300 return (remainder == 0);
1305 * Called when we receive a data packet from the "wire" on the
1308 * Note: This is no longer used as a callback
1311 netvsc_recv(struct hv_vmbus_channel *chan, netvsc_packet *packet,
1312 const rndis_tcp_ip_csum_info *csum_info,
1313 const struct rndis_hash_info *hash_info,
1314 const struct rndis_hash_value *hash_value)
1316 struct hn_rx_ring *rxr = chan->hv_chan_rxr;
1317 struct ifnet *ifp = rxr->hn_ifp;
1319 int size, do_lro = 0, do_csum = 1;
1321 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1325 * Bail out if packet contains more data than configured MTU.
1327 if (packet->tot_data_buf_len > (ifp->if_mtu + ETHER_HDR_LEN)) {
1329 } else if (packet->tot_data_buf_len <= MHLEN) {
1330 m_new = m_gethdr(M_NOWAIT, MT_DATA);
1331 if (m_new == NULL) {
1332 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1335 memcpy(mtod(m_new, void *), packet->data,
1336 packet->tot_data_buf_len);
1337 m_new->m_pkthdr.len = m_new->m_len = packet->tot_data_buf_len;
1338 rxr->hn_small_pkts++;
1341 * Get an mbuf with a cluster. For packets 2K or less,
1342 * get a standard 2K cluster. For anything larger, get a
1343 * 4K cluster. Any buffers larger than 4K can cause problems
1344 * if looped around to the Hyper-V TX channel, so avoid them.
1347 if (packet->tot_data_buf_len > MCLBYTES) {
1349 size = MJUMPAGESIZE;
1352 m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, size);
1353 if (m_new == NULL) {
1354 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1358 hv_m_append(m_new, packet->tot_data_buf_len, packet->data);
1360 m_new->m_pkthdr.rcvif = ifp;
1362 if (__predict_false((ifp->if_capenable & IFCAP_RXCSUM) == 0))
1365 /* receive side checksum offload */
1366 if (csum_info != NULL) {
1367 /* IP csum offload */
1368 if (csum_info->receive.ip_csum_succeeded && do_csum) {
1369 m_new->m_pkthdr.csum_flags |=
1370 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1374 /* TCP/UDP csum offload */
1375 if ((csum_info->receive.tcp_csum_succeeded ||
1376 csum_info->receive.udp_csum_succeeded) && do_csum) {
1377 m_new->m_pkthdr.csum_flags |=
1378 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1379 m_new->m_pkthdr.csum_data = 0xffff;
1380 if (csum_info->receive.tcp_csum_succeeded)
1386 if (csum_info->receive.ip_csum_succeeded &&
1387 csum_info->receive.tcp_csum_succeeded)
1390 const struct ether_header *eh;
1395 if (m_new->m_len < hoff)
1397 eh = mtod(m_new, struct ether_header *);
1398 etype = ntohs(eh->ether_type);
1399 if (etype == ETHERTYPE_VLAN) {
1400 const struct ether_vlan_header *evl;
1402 hoff = sizeof(*evl);
1403 if (m_new->m_len < hoff)
1405 evl = mtod(m_new, struct ether_vlan_header *);
1406 etype = ntohs(evl->evl_proto);
1409 if (etype == ETHERTYPE_IP) {
1412 pr = hn_check_iplen(m_new, hoff);
1413 if (pr == IPPROTO_TCP) {
1415 (rxr->hn_trust_hcsum &
1416 HN_TRUST_HCSUM_TCP)) {
1417 rxr->hn_csum_trusted++;
1418 m_new->m_pkthdr.csum_flags |=
1419 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1420 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1421 m_new->m_pkthdr.csum_data = 0xffff;
1424 } else if (pr == IPPROTO_UDP) {
1426 (rxr->hn_trust_hcsum &
1427 HN_TRUST_HCSUM_UDP)) {
1428 rxr->hn_csum_trusted++;
1429 m_new->m_pkthdr.csum_flags |=
1430 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1431 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1432 m_new->m_pkthdr.csum_data = 0xffff;
1434 } else if (pr != IPPROTO_DONE && do_csum &&
1435 (rxr->hn_trust_hcsum & HN_TRUST_HCSUM_IP)) {
1436 rxr->hn_csum_trusted++;
1437 m_new->m_pkthdr.csum_flags |=
1438 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1443 if ((packet->vlan_tci != 0) &&
1444 (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0) {
1445 m_new->m_pkthdr.ether_vtag = packet->vlan_tci;
1446 m_new->m_flags |= M_VLANTAG;
1449 if (hash_info != NULL && hash_value != NULL) {
1450 int hash_type = M_HASHTYPE_OPAQUE;
1453 m_new->m_pkthdr.flowid = hash_value->hash_value;
1454 if ((hash_info->hash_info & NDIS_HASH_FUNCTION_MASK) ==
1455 NDIS_HASH_FUNCTION_TOEPLITZ) {
1457 (hash_info->hash_info & NDIS_HASH_TYPE_MASK);
1460 case NDIS_HASH_IPV4:
1461 hash_type = M_HASHTYPE_RSS_IPV4;
1464 case NDIS_HASH_TCP_IPV4:
1465 hash_type = M_HASHTYPE_RSS_TCP_IPV4;
1468 case NDIS_HASH_IPV6:
1469 hash_type = M_HASHTYPE_RSS_IPV6;
1472 case NDIS_HASH_IPV6_EX:
1473 hash_type = M_HASHTYPE_RSS_IPV6_EX;
1476 case NDIS_HASH_TCP_IPV6:
1477 hash_type = M_HASHTYPE_RSS_TCP_IPV6;
1480 case NDIS_HASH_TCP_IPV6_EX:
1481 hash_type = M_HASHTYPE_RSS_TCP_IPV6_EX;
1485 M_HASHTYPE_SET(m_new, hash_type);
1487 if (hash_value != NULL)
1488 m_new->m_pkthdr.flowid = hash_value->hash_value;
1490 m_new->m_pkthdr.flowid = rxr->hn_rx_idx;
1491 M_HASHTYPE_SET(m_new, M_HASHTYPE_OPAQUE);
1495 * Note: Moved RX completion back to hv_nv_on_receive() so all
1496 * messages (not just data messages) will trigger a response.
1502 if ((ifp->if_capenable & IFCAP_LRO) && do_lro) {
1503 #if defined(INET) || defined(INET6)
1504 struct lro_ctrl *lro = &rxr->hn_lro;
1507 rxr->hn_lro_tried++;
1508 if (tcp_lro_rx(lro, m_new, 0) == 0) {
1516 /* We're not holding the lock here, so don't release it */
1517 (*ifp->if_input)(ifp, m_new);
1523 * Rules for using sc->temp_unusable:
1524 * 1. sc->temp_unusable can only be read or written while holding NV_LOCK()
1525 * 2. code reading sc->temp_unusable under NV_LOCK(), and finding
1526 * sc->temp_unusable set, must release NV_LOCK() and exit
1527 * 3. to retain exclusive control of the interface,
1528 * sc->temp_unusable must be set by code before releasing NV_LOCK()
1529 * 4. only code setting sc->temp_unusable can clear sc->temp_unusable
1530 * 5. code setting sc->temp_unusable must eventually clear sc->temp_unusable
1534 * Standard ioctl entry point. Called when the user wants to configure
1538 hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1540 hn_softc_t *sc = ifp->if_softc;
1541 struct ifreq *ifr = (struct ifreq *)data;
1543 struct ifaddr *ifa = (struct ifaddr *)data;
1545 netvsc_device_info device_info;
1546 struct hv_device *hn_dev;
1547 int mask, error = 0;
1548 int retry_cnt = 500;
1554 if (ifa->ifa_addr->sa_family == AF_INET) {
1555 ifp->if_flags |= IFF_UP;
1556 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1558 arp_ifinit(ifp, ifa);
1561 error = ether_ioctl(ifp, cmd, data);
1564 hn_dev = vmbus_get_devctx(sc->hn_dev);
1566 /* Check MTU value change */
1567 if (ifp->if_mtu == ifr->ifr_mtu)
1570 if (ifr->ifr_mtu > NETVSC_MAX_CONFIGURABLE_MTU) {
1575 /* Obtain and record requested MTU */
1576 ifp->if_mtu = ifr->ifr_mtu;
1578 #if __FreeBSD_version >= 1100099
1580 * Make sure that LRO aggregation length limit is still
1581 * valid, after the MTU change.
1584 if (sc->hn_rx_ring[0].hn_lro.lro_length_lim <
1585 HN_LRO_LENLIM_MIN(ifp))
1586 hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MIN(ifp));
1592 if (!sc->temp_unusable) {
1593 sc->temp_unusable = TRUE;
1597 if (retry_cnt > 0) {
1601 } while (retry_cnt > 0);
1603 if (retry_cnt == 0) {
1608 /* We must remove and add back the device to cause the new
1609 * MTU to take effect. This includes tearing down, but not
1610 * deleting the channel, then bringing it back up.
1612 error = hv_rf_on_device_remove(hn_dev, HV_RF_NV_RETAIN_CHANNEL);
1615 sc->temp_unusable = FALSE;
1619 error = hv_rf_on_device_add(hn_dev, &device_info,
1620 sc->hn_rx_ring_inuse);
1623 sc->temp_unusable = FALSE;
1628 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
1629 if (sc->hn_tx_ring[0].hn_tx_chimney_size >
1630 sc->hn_tx_chimney_max)
1631 hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max);
1633 hn_ifinit_locked(sc);
1636 sc->temp_unusable = FALSE;
1642 if (!sc->temp_unusable) {
1643 sc->temp_unusable = TRUE;
1647 if (retry_cnt > 0) {
1651 } while (retry_cnt > 0);
1653 if (retry_cnt == 0) {
1658 if (ifp->if_flags & IFF_UP) {
1660 * If only the state of the PROMISC flag changed,
1661 * then just use the 'set promisc mode' command
1662 * instead of reinitializing the entire NIC. Doing
1663 * a full re-init means reloading the firmware and
1664 * waiting for it to start up, which may take a
1668 /* Fixme: Promiscuous mode? */
1669 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1670 ifp->if_flags & IFF_PROMISC &&
1671 !(sc->hn_if_flags & IFF_PROMISC)) {
1672 /* do something here for Hyper-V */
1673 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1674 !(ifp->if_flags & IFF_PROMISC) &&
1675 sc->hn_if_flags & IFF_PROMISC) {
1676 /* do something here for Hyper-V */
1679 hn_ifinit_locked(sc);
1681 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1686 sc->temp_unusable = FALSE;
1688 sc->hn_if_flags = ifp->if_flags;
1694 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1695 if (mask & IFCAP_TXCSUM) {
1696 ifp->if_capenable ^= IFCAP_TXCSUM;
1697 if (ifp->if_capenable & IFCAP_TXCSUM) {
1699 sc->hn_tx_ring[0].hn_csum_assist;
1702 ~sc->hn_tx_ring[0].hn_csum_assist;
1706 if (mask & IFCAP_RXCSUM)
1707 ifp->if_capenable ^= IFCAP_RXCSUM;
1709 if (mask & IFCAP_LRO)
1710 ifp->if_capenable ^= IFCAP_LRO;
1712 if (mask & IFCAP_TSO4) {
1713 ifp->if_capenable ^= IFCAP_TSO4;
1714 if (ifp->if_capenable & IFCAP_TSO4)
1715 ifp->if_hwassist |= CSUM_IP_TSO;
1717 ifp->if_hwassist &= ~CSUM_IP_TSO;
1720 if (mask & IFCAP_TSO6) {
1721 ifp->if_capenable ^= IFCAP_TSO6;
1722 if (ifp->if_capenable & IFCAP_TSO6)
1723 ifp->if_hwassist |= CSUM_IP6_TSO;
1725 ifp->if_hwassist &= ~CSUM_IP6_TSO;
1734 /* Fixme: Multicast mode? */
1735 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1737 netvsc_setmulti(sc);
1746 error = ifmedia_ioctl(ifp, ifr, &sc->hn_media, cmd);
1749 error = ether_ioctl(ifp, cmd, data);
1760 hn_stop(hn_softc_t *sc)
1764 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1769 printf(" Closing Device ...\n");
1771 atomic_clear_int(&ifp->if_drv_flags,
1772 (IFF_DRV_RUNNING | IFF_DRV_OACTIVE));
1773 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
1774 sc->hn_tx_ring[i].hn_oactive = 0;
1776 if_link_state_change(ifp, LINK_STATE_DOWN);
1777 sc->hn_initdone = 0;
1779 ret = hv_rf_on_close(device_ctx);
1783 * FreeBSD transmit entry point
1786 hn_start(struct ifnet *ifp)
1788 struct hn_softc *sc = ifp->if_softc;
1789 struct hn_tx_ring *txr = &sc->hn_tx_ring[0];
1791 if (txr->hn_sched_tx)
1794 if (mtx_trylock(&txr->hn_tx_lock)) {
1797 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1798 mtx_unlock(&txr->hn_tx_lock);
1803 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
1807 hn_start_txeof(struct hn_tx_ring *txr)
1809 struct hn_softc *sc = txr->hn_sc;
1810 struct ifnet *ifp = sc->hn_ifp;
1812 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1814 if (txr->hn_sched_tx)
1817 if (mtx_trylock(&txr->hn_tx_lock)) {
1820 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1821 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1822 mtx_unlock(&txr->hn_tx_lock);
1824 taskqueue_enqueue(txr->hn_tx_taskq,
1830 * Release the OACTIVE earlier, with the hope, that
1831 * others could catch up. The task will clear the
1832 * flag again with the hn_tx_lock to avoid possible
1835 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1836 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
1844 hn_ifinit_locked(hn_softc_t *sc)
1847 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1852 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1856 hv_promisc_mode = 1;
1858 ret = hv_rf_on_open(device_ctx);
1862 sc->hn_initdone = 1;
1865 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1866 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
1867 sc->hn_tx_ring[i].hn_oactive = 0;
1869 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_RUNNING);
1870 if_link_state_change(ifp, LINK_STATE_UP);
1877 hn_ifinit(void *xsc)
1879 hn_softc_t *sc = xsc;
1882 if (sc->temp_unusable) {
1886 sc->temp_unusable = TRUE;
1889 hn_ifinit_locked(sc);
1892 sc->temp_unusable = FALSE;
1901 hn_watchdog(struct ifnet *ifp)
1906 printf("hn%d: watchdog timeout -- resetting\n", sc->hn_unit);
1907 hn_ifinit(sc); /*???*/
1912 #if __FreeBSD_version >= 1100099
1915 hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS)
1917 struct hn_softc *sc = arg1;
1918 unsigned int lenlim;
1921 lenlim = sc->hn_rx_ring[0].hn_lro.lro_length_lim;
1922 error = sysctl_handle_int(oidp, &lenlim, 0, req);
1923 if (error || req->newptr == NULL)
1926 if (lenlim < HN_LRO_LENLIM_MIN(sc->hn_ifp) ||
1927 lenlim > TCP_LRO_LENGTH_MAX)
1931 hn_set_lro_lenlim(sc, lenlim);
1937 hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS)
1939 struct hn_softc *sc = arg1;
1940 int ackcnt, error, i;
1943 * lro_ackcnt_lim is append count limit,
1944 * +1 to turn it into aggregation limit.
1946 ackcnt = sc->hn_rx_ring[0].hn_lro.lro_ackcnt_lim + 1;
1947 error = sysctl_handle_int(oidp, &ackcnt, 0, req);
1948 if (error || req->newptr == NULL)
1951 if (ackcnt < 2 || ackcnt > (TCP_LRO_ACKCNT_MAX + 1))
1955 * Convert aggregation limit back to append
1960 for (i = 0; i < sc->hn_rx_ring_inuse; ++i)
1961 sc->hn_rx_ring[i].hn_lro.lro_ackcnt_lim = ackcnt;
1969 hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS)
1971 struct hn_softc *sc = arg1;
1976 if (sc->hn_rx_ring[0].hn_trust_hcsum & hcsum)
1979 error = sysctl_handle_int(oidp, &on, 0, req);
1980 if (error || req->newptr == NULL)
1984 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1985 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
1988 rxr->hn_trust_hcsum |= hcsum;
1990 rxr->hn_trust_hcsum &= ~hcsum;
1997 hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS)
1999 struct hn_softc *sc = arg1;
2000 int chimney_size, error;
2002 chimney_size = sc->hn_tx_ring[0].hn_tx_chimney_size;
2003 error = sysctl_handle_int(oidp, &chimney_size, 0, req);
2004 if (error || req->newptr == NULL)
2007 if (chimney_size > sc->hn_tx_chimney_max || chimney_size <= 0)
2010 hn_set_tx_chimney_size(sc, chimney_size);
2014 #if __FreeBSD_version < 1100095
2016 hn_rx_stat_int_sysctl(SYSCTL_HANDLER_ARGS)
2018 struct hn_softc *sc = arg1;
2019 int ofs = arg2, i, error;
2020 struct hn_rx_ring *rxr;
2024 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2025 rxr = &sc->hn_rx_ring[i];
2026 stat += *((int *)((uint8_t *)rxr + ofs));
2029 error = sysctl_handle_64(oidp, &stat, 0, req);
2030 if (error || req->newptr == NULL)
2033 /* Zero out this stat. */
2034 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2035 rxr = &sc->hn_rx_ring[i];
2036 *((int *)((uint8_t *)rxr + ofs)) = 0;
2042 hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS)
2044 struct hn_softc *sc = arg1;
2045 int ofs = arg2, i, error;
2046 struct hn_rx_ring *rxr;
2050 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2051 rxr = &sc->hn_rx_ring[i];
2052 stat += *((uint64_t *)((uint8_t *)rxr + ofs));
2055 error = sysctl_handle_64(oidp, &stat, 0, req);
2056 if (error || req->newptr == NULL)
2059 /* Zero out this stat. */
2060 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2061 rxr = &sc->hn_rx_ring[i];
2062 *((uint64_t *)((uint8_t *)rxr + ofs)) = 0;
2070 hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
2072 struct hn_softc *sc = arg1;
2073 int ofs = arg2, i, error;
2074 struct hn_rx_ring *rxr;
2078 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2079 rxr = &sc->hn_rx_ring[i];
2080 stat += *((u_long *)((uint8_t *)rxr + ofs));
2083 error = sysctl_handle_long(oidp, &stat, 0, req);
2084 if (error || req->newptr == NULL)
2087 /* Zero out this stat. */
2088 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2089 rxr = &sc->hn_rx_ring[i];
2090 *((u_long *)((uint8_t *)rxr + ofs)) = 0;
2096 hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
2098 struct hn_softc *sc = arg1;
2099 int ofs = arg2, i, error;
2100 struct hn_tx_ring *txr;
2104 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2105 txr = &sc->hn_tx_ring[i];
2106 stat += *((u_long *)((uint8_t *)txr + ofs));
2109 error = sysctl_handle_long(oidp, &stat, 0, req);
2110 if (error || req->newptr == NULL)
2113 /* Zero out this stat. */
2114 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2115 txr = &sc->hn_tx_ring[i];
2116 *((u_long *)((uint8_t *)txr + ofs)) = 0;
2122 hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS)
2124 struct hn_softc *sc = arg1;
2125 int ofs = arg2, i, error, conf;
2126 struct hn_tx_ring *txr;
2128 txr = &sc->hn_tx_ring[0];
2129 conf = *((int *)((uint8_t *)txr + ofs));
2131 error = sysctl_handle_int(oidp, &conf, 0, req);
2132 if (error || req->newptr == NULL)
2136 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2137 txr = &sc->hn_tx_ring[i];
2138 *((int *)((uint8_t *)txr + ofs)) = conf;
2146 hn_check_iplen(const struct mbuf *m, int hoff)
2148 const struct ip *ip;
2149 int len, iphlen, iplen;
2150 const struct tcphdr *th;
2151 int thoff; /* TCP data offset */
2153 len = hoff + sizeof(struct ip);
2155 /* The packet must be at least the size of an IP header. */
2156 if (m->m_pkthdr.len < len)
2157 return IPPROTO_DONE;
2159 /* The fixed IP header must reside completely in the first mbuf. */
2161 return IPPROTO_DONE;
2163 ip = mtodo(m, hoff);
2165 /* Bound check the packet's stated IP header length. */
2166 iphlen = ip->ip_hl << 2;
2167 if (iphlen < sizeof(struct ip)) /* minimum header length */
2168 return IPPROTO_DONE;
2170 /* The full IP header must reside completely in the one mbuf. */
2171 if (m->m_len < hoff + iphlen)
2172 return IPPROTO_DONE;
2174 iplen = ntohs(ip->ip_len);
2177 * Check that the amount of data in the buffers is as
2178 * at least much as the IP header would have us expect.
2180 if (m->m_pkthdr.len < hoff + iplen)
2181 return IPPROTO_DONE;
2184 * Ignore IP fragments.
2186 if (ntohs(ip->ip_off) & (IP_OFFMASK | IP_MF))
2187 return IPPROTO_DONE;
2190 * The TCP/IP or UDP/IP header must be entirely contained within
2191 * the first fragment of a packet.
2195 if (iplen < iphlen + sizeof(struct tcphdr))
2196 return IPPROTO_DONE;
2197 if (m->m_len < hoff + iphlen + sizeof(struct tcphdr))
2198 return IPPROTO_DONE;
2199 th = (const struct tcphdr *)((const uint8_t *)ip + iphlen);
2200 thoff = th->th_off << 2;
2201 if (thoff < sizeof(struct tcphdr) || thoff + iphlen > iplen)
2202 return IPPROTO_DONE;
2203 if (m->m_len < hoff + iphlen + thoff)
2204 return IPPROTO_DONE;
2207 if (iplen < iphlen + sizeof(struct udphdr))
2208 return IPPROTO_DONE;
2209 if (m->m_len < hoff + iphlen + sizeof(struct udphdr))
2210 return IPPROTO_DONE;
2214 return IPPROTO_DONE;
2221 hn_create_rx_data(struct hn_softc *sc, int ring_cnt)
2223 struct sysctl_oid_list *child;
2224 struct sysctl_ctx_list *ctx;
2225 device_t dev = sc->hn_dev;
2226 #if defined(INET) || defined(INET6)
2227 #if __FreeBSD_version >= 1100095
2233 sc->hn_rx_ring_cnt = ring_cnt;
2234 sc->hn_rx_ring_inuse = sc->hn_rx_ring_cnt;
2236 sc->hn_rx_ring = malloc(sizeof(struct hn_rx_ring) * sc->hn_rx_ring_cnt,
2237 M_NETVSC, M_WAITOK | M_ZERO);
2239 #if defined(INET) || defined(INET6)
2240 #if __FreeBSD_version >= 1100095
2241 lroent_cnt = hn_lro_entry_count;
2242 if (lroent_cnt < TCP_LRO_ENTRIES)
2243 lroent_cnt = TCP_LRO_ENTRIES;
2244 device_printf(dev, "LRO: entry count %d\n", lroent_cnt);
2246 #endif /* INET || INET6 */
2248 ctx = device_get_sysctl_ctx(dev);
2249 child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
2251 /* Create dev.hn.UNIT.rx sysctl tree */
2252 sc->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "rx",
2253 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2255 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2256 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
2258 if (hn_trust_hosttcp)
2259 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_TCP;
2260 if (hn_trust_hostudp)
2261 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_UDP;
2262 if (hn_trust_hostip)
2263 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_IP;
2264 rxr->hn_ifp = sc->hn_ifp;
2270 #if defined(INET) || defined(INET6)
2271 #if __FreeBSD_version >= 1100095
2272 tcp_lro_init_args(&rxr->hn_lro, sc->hn_ifp, lroent_cnt, 0);
2274 tcp_lro_init(&rxr->hn_lro);
2275 rxr->hn_lro.ifp = sc->hn_ifp;
2277 #if __FreeBSD_version >= 1100099
2278 rxr->hn_lro.lro_length_lim = HN_LRO_LENLIM_DEF;
2279 rxr->hn_lro.lro_ackcnt_lim = HN_LRO_ACKCNT_DEF;
2281 #endif /* INET || INET6 */
2283 if (sc->hn_rx_sysctl_tree != NULL) {
2287 * Create per RX ring sysctl tree:
2288 * dev.hn.UNIT.rx.RINGID
2290 snprintf(name, sizeof(name), "%d", i);
2291 rxr->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx,
2292 SYSCTL_CHILDREN(sc->hn_rx_sysctl_tree),
2293 OID_AUTO, name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2295 if (rxr->hn_rx_sysctl_tree != NULL) {
2296 SYSCTL_ADD_ULONG(ctx,
2297 SYSCTL_CHILDREN(rxr->hn_rx_sysctl_tree),
2298 OID_AUTO, "packets", CTLFLAG_RW,
2299 &rxr->hn_pkts, "# of packets received");
2300 SYSCTL_ADD_ULONG(ctx,
2301 SYSCTL_CHILDREN(rxr->hn_rx_sysctl_tree),
2302 OID_AUTO, "rss_pkts", CTLFLAG_RW,
2304 "# of packets w/ RSS info received");
2309 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_queued",
2310 CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2311 __offsetof(struct hn_rx_ring, hn_lro.lro_queued),
2312 #if __FreeBSD_version < 1100095
2313 hn_rx_stat_int_sysctl,
2315 hn_rx_stat_u64_sysctl,
2317 "LU", "LRO queued");
2318 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_flushed",
2319 CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2320 __offsetof(struct hn_rx_ring, hn_lro.lro_flushed),
2321 #if __FreeBSD_version < 1100095
2322 hn_rx_stat_int_sysctl,
2324 hn_rx_stat_u64_sysctl,
2326 "LU", "LRO flushed");
2327 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_tried",
2328 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2329 __offsetof(struct hn_rx_ring, hn_lro_tried),
2330 hn_rx_stat_ulong_sysctl, "LU", "# of LRO tries");
2331 #if __FreeBSD_version >= 1100099
2332 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_length_lim",
2333 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2334 hn_lro_lenlim_sysctl, "IU",
2335 "Max # of data bytes to be aggregated by LRO");
2336 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_ackcnt_lim",
2337 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2338 hn_lro_ackcnt_sysctl, "I",
2339 "Max # of ACKs to be aggregated by LRO");
2341 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hosttcp",
2342 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_TCP,
2343 hn_trust_hcsum_sysctl, "I",
2344 "Trust tcp segement verification on host side, "
2345 "when csum info is missing");
2346 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostudp",
2347 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_UDP,
2348 hn_trust_hcsum_sysctl, "I",
2349 "Trust udp datagram verification on host side, "
2350 "when csum info is missing");
2351 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostip",
2352 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_IP,
2353 hn_trust_hcsum_sysctl, "I",
2354 "Trust ip packet verification on host side, "
2355 "when csum info is missing");
2356 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_ip",
2357 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2358 __offsetof(struct hn_rx_ring, hn_csum_ip),
2359 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM IP");
2360 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_tcp",
2361 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2362 __offsetof(struct hn_rx_ring, hn_csum_tcp),
2363 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM TCP");
2364 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_udp",
2365 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2366 __offsetof(struct hn_rx_ring, hn_csum_udp),
2367 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM UDP");
2368 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_trusted",
2369 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2370 __offsetof(struct hn_rx_ring, hn_csum_trusted),
2371 hn_rx_stat_ulong_sysctl, "LU",
2372 "# of packets that we trust host's csum verification");
2373 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "small_pkts",
2374 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2375 __offsetof(struct hn_rx_ring, hn_small_pkts),
2376 hn_rx_stat_ulong_sysctl, "LU", "# of small packets received");
2377 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_cnt",
2378 CTLFLAG_RD, &sc->hn_rx_ring_cnt, 0, "# created RX rings");
2379 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_inuse",
2380 CTLFLAG_RD, &sc->hn_rx_ring_inuse, 0, "# used RX rings");
2384 hn_destroy_rx_data(struct hn_softc *sc)
2386 #if defined(INET) || defined(INET6)
2390 if (sc->hn_rx_ring_cnt == 0)
2393 #if defined(INET) || defined(INET6)
2394 for (i = 0; i < sc->hn_rx_ring_cnt; ++i)
2395 tcp_lro_free(&sc->hn_rx_ring[i].hn_lro);
2397 free(sc->hn_rx_ring, M_NETVSC);
2398 sc->hn_rx_ring = NULL;
2400 sc->hn_rx_ring_cnt = 0;
2401 sc->hn_rx_ring_inuse = 0;
2405 hn_create_tx_ring(struct hn_softc *sc, int id)
2407 struct hn_tx_ring *txr = &sc->hn_tx_ring[id];
2408 bus_dma_tag_t parent_dtag;
2412 txr->hn_tx_idx = id;
2414 #ifndef HN_USE_TXDESC_BUFRING
2415 mtx_init(&txr->hn_txlist_spin, "hn txlist", NULL, MTX_SPIN);
2417 mtx_init(&txr->hn_tx_lock, "hn tx", NULL, MTX_DEF);
2419 txr->hn_txdesc_cnt = HN_TX_DESC_CNT;
2420 txr->hn_txdesc = malloc(sizeof(struct hn_txdesc) * txr->hn_txdesc_cnt,
2421 M_NETVSC, M_WAITOK | M_ZERO);
2422 #ifndef HN_USE_TXDESC_BUFRING
2423 SLIST_INIT(&txr->hn_txlist);
2425 txr->hn_txdesc_br = buf_ring_alloc(txr->hn_txdesc_cnt, M_NETVSC,
2426 M_WAITOK, &txr->hn_tx_lock);
2429 txr->hn_tx_taskq = sc->hn_tx_taskq;
2431 if (hn_use_if_start) {
2432 txr->hn_txeof = hn_start_txeof;
2433 TASK_INIT(&txr->hn_tx_task, 0, hn_start_taskfunc, txr);
2434 TASK_INIT(&txr->hn_txeof_task, 0, hn_start_txeof_taskfunc, txr);
2438 txr->hn_txeof = hn_xmit_txeof;
2439 TASK_INIT(&txr->hn_tx_task, 0, hn_xmit_taskfunc, txr);
2440 TASK_INIT(&txr->hn_txeof_task, 0, hn_xmit_txeof_taskfunc, txr);
2442 br_depth = hn_get_txswq_depth(txr);
2443 txr->hn_mbuf_br = buf_ring_alloc(br_depth, M_NETVSC,
2444 M_WAITOK, &txr->hn_tx_lock);
2447 txr->hn_direct_tx_size = hn_direct_tx_size;
2448 if (hv_vmbus_protocal_version >= HV_VMBUS_VERSION_WIN8_1)
2449 txr->hn_csum_assist = HN_CSUM_ASSIST;
2451 txr->hn_csum_assist = HN_CSUM_ASSIST_WIN8;
2454 * Always schedule transmission instead of trying to do direct
2455 * transmission. This one gives the best performance so far.
2457 txr->hn_sched_tx = 1;
2459 parent_dtag = bus_get_dma_tag(sc->hn_dev);
2461 /* DMA tag for RNDIS messages. */
2462 error = bus_dma_tag_create(parent_dtag, /* parent */
2463 HN_RNDIS_MSG_ALIGN, /* alignment */
2464 HN_RNDIS_MSG_BOUNDARY, /* boundary */
2465 BUS_SPACE_MAXADDR, /* lowaddr */
2466 BUS_SPACE_MAXADDR, /* highaddr */
2467 NULL, NULL, /* filter, filterarg */
2468 HN_RNDIS_MSG_LEN, /* maxsize */
2470 HN_RNDIS_MSG_LEN, /* maxsegsize */
2472 NULL, /* lockfunc */
2473 NULL, /* lockfuncarg */
2474 &txr->hn_tx_rndis_dtag);
2476 device_printf(sc->hn_dev, "failed to create rndis dmatag\n");
2480 /* DMA tag for data. */
2481 error = bus_dma_tag_create(parent_dtag, /* parent */
2483 HN_TX_DATA_BOUNDARY, /* boundary */
2484 BUS_SPACE_MAXADDR, /* lowaddr */
2485 BUS_SPACE_MAXADDR, /* highaddr */
2486 NULL, NULL, /* filter, filterarg */
2487 HN_TX_DATA_MAXSIZE, /* maxsize */
2488 HN_TX_DATA_SEGCNT_MAX, /* nsegments */
2489 HN_TX_DATA_SEGSIZE, /* maxsegsize */
2491 NULL, /* lockfunc */
2492 NULL, /* lockfuncarg */
2493 &txr->hn_tx_data_dtag);
2495 device_printf(sc->hn_dev, "failed to create data dmatag\n");
2499 for (i = 0; i < txr->hn_txdesc_cnt; ++i) {
2500 struct hn_txdesc *txd = &txr->hn_txdesc[i];
2505 * Allocate and load RNDIS messages.
2507 error = bus_dmamem_alloc(txr->hn_tx_rndis_dtag,
2508 (void **)&txd->rndis_msg,
2509 BUS_DMA_WAITOK | BUS_DMA_COHERENT,
2510 &txd->rndis_msg_dmap);
2512 device_printf(sc->hn_dev,
2513 "failed to allocate rndis_msg, %d\n", i);
2517 error = bus_dmamap_load(txr->hn_tx_rndis_dtag,
2518 txd->rndis_msg_dmap,
2519 txd->rndis_msg, HN_RNDIS_MSG_LEN,
2520 hyperv_dma_map_paddr, &txd->rndis_msg_paddr,
2523 device_printf(sc->hn_dev,
2524 "failed to load rndis_msg, %d\n", i);
2525 bus_dmamem_free(txr->hn_tx_rndis_dtag,
2526 txd->rndis_msg, txd->rndis_msg_dmap);
2530 /* DMA map for TX data. */
2531 error = bus_dmamap_create(txr->hn_tx_data_dtag, 0,
2534 device_printf(sc->hn_dev,
2535 "failed to allocate tx data dmamap\n");
2536 bus_dmamap_unload(txr->hn_tx_rndis_dtag,
2537 txd->rndis_msg_dmap);
2538 bus_dmamem_free(txr->hn_tx_rndis_dtag,
2539 txd->rndis_msg, txd->rndis_msg_dmap);
2543 /* All set, put it to list */
2544 txd->flags |= HN_TXD_FLAG_ONLIST;
2545 #ifndef HN_USE_TXDESC_BUFRING
2546 SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
2548 buf_ring_enqueue(txr->hn_txdesc_br, txd);
2551 txr->hn_txdesc_avail = txr->hn_txdesc_cnt;
2553 if (sc->hn_tx_sysctl_tree != NULL) {
2554 struct sysctl_oid_list *child;
2555 struct sysctl_ctx_list *ctx;
2559 * Create per TX ring sysctl tree:
2560 * dev.hn.UNIT.tx.RINGID
2562 ctx = device_get_sysctl_ctx(sc->hn_dev);
2563 child = SYSCTL_CHILDREN(sc->hn_tx_sysctl_tree);
2565 snprintf(name, sizeof(name), "%d", id);
2566 txr->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO,
2567 name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2569 if (txr->hn_tx_sysctl_tree != NULL) {
2570 child = SYSCTL_CHILDREN(txr->hn_tx_sysctl_tree);
2572 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_avail",
2573 CTLFLAG_RD, &txr->hn_txdesc_avail, 0,
2574 "# of available TX descs");
2575 if (!hn_use_if_start) {
2576 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "oactive",
2577 CTLFLAG_RD, &txr->hn_oactive, 0,
2580 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "packets",
2581 CTLFLAG_RW, &txr->hn_pkts,
2582 "# of packets transmitted");
2590 hn_txdesc_dmamap_destroy(struct hn_txdesc *txd)
2592 struct hn_tx_ring *txr = txd->txr;
2594 KASSERT(txd->m == NULL, ("still has mbuf installed"));
2595 KASSERT((txd->flags & HN_TXD_FLAG_DMAMAP) == 0, ("still dma mapped"));
2597 bus_dmamap_unload(txr->hn_tx_rndis_dtag, txd->rndis_msg_dmap);
2598 bus_dmamem_free(txr->hn_tx_rndis_dtag, txd->rndis_msg,
2599 txd->rndis_msg_dmap);
2600 bus_dmamap_destroy(txr->hn_tx_data_dtag, txd->data_dmap);
2604 hn_destroy_tx_ring(struct hn_tx_ring *txr)
2606 struct hn_txdesc *txd;
2608 if (txr->hn_txdesc == NULL)
2611 #ifndef HN_USE_TXDESC_BUFRING
2612 while ((txd = SLIST_FIRST(&txr->hn_txlist)) != NULL) {
2613 SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
2614 hn_txdesc_dmamap_destroy(txd);
2617 mtx_lock(&txr->hn_tx_lock);
2618 while ((txd = buf_ring_dequeue_sc(txr->hn_txdesc_br)) != NULL)
2619 hn_txdesc_dmamap_destroy(txd);
2620 mtx_unlock(&txr->hn_tx_lock);
2623 if (txr->hn_tx_data_dtag != NULL)
2624 bus_dma_tag_destroy(txr->hn_tx_data_dtag);
2625 if (txr->hn_tx_rndis_dtag != NULL)
2626 bus_dma_tag_destroy(txr->hn_tx_rndis_dtag);
2628 #ifdef HN_USE_TXDESC_BUFRING
2629 buf_ring_free(txr->hn_txdesc_br, M_NETVSC);
2632 free(txr->hn_txdesc, M_NETVSC);
2633 txr->hn_txdesc = NULL;
2635 if (txr->hn_mbuf_br != NULL)
2636 buf_ring_free(txr->hn_mbuf_br, M_NETVSC);
2638 #ifndef HN_USE_TXDESC_BUFRING
2639 mtx_destroy(&txr->hn_txlist_spin);
2641 mtx_destroy(&txr->hn_tx_lock);
2645 hn_create_tx_data(struct hn_softc *sc, int ring_cnt)
2647 struct sysctl_oid_list *child;
2648 struct sysctl_ctx_list *ctx;
2651 sc->hn_tx_ring_cnt = ring_cnt;
2652 sc->hn_tx_ring_inuse = sc->hn_tx_ring_cnt;
2654 sc->hn_tx_ring = malloc(sizeof(struct hn_tx_ring) * sc->hn_tx_ring_cnt,
2655 M_NETVSC, M_WAITOK | M_ZERO);
2657 ctx = device_get_sysctl_ctx(sc->hn_dev);
2658 child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->hn_dev));
2660 /* Create dev.hn.UNIT.tx sysctl tree */
2661 sc->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "tx",
2662 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2664 for (i = 0; i < sc->hn_tx_ring_cnt; ++i) {
2667 error = hn_create_tx_ring(sc, i);
2672 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "no_txdescs",
2673 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2674 __offsetof(struct hn_tx_ring, hn_no_txdescs),
2675 hn_tx_stat_ulong_sysctl, "LU", "# of times short of TX descs");
2676 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "send_failed",
2677 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2678 __offsetof(struct hn_tx_ring, hn_send_failed),
2679 hn_tx_stat_ulong_sysctl, "LU", "# of hyper-v sending failure");
2680 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "txdma_failed",
2681 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2682 __offsetof(struct hn_tx_ring, hn_txdma_failed),
2683 hn_tx_stat_ulong_sysctl, "LU", "# of TX DMA failure");
2684 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_collapsed",
2685 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2686 __offsetof(struct hn_tx_ring, hn_tx_collapsed),
2687 hn_tx_stat_ulong_sysctl, "LU", "# of TX mbuf collapsed");
2688 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney",
2689 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2690 __offsetof(struct hn_tx_ring, hn_tx_chimney),
2691 hn_tx_stat_ulong_sysctl, "LU", "# of chimney send");
2692 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_tried",
2693 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2694 __offsetof(struct hn_tx_ring, hn_tx_chimney_tried),
2695 hn_tx_stat_ulong_sysctl, "LU", "# of chimney send tries");
2696 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_cnt",
2697 CTLFLAG_RD, &sc->hn_tx_ring[0].hn_txdesc_cnt, 0,
2698 "# of total TX descs");
2699 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_chimney_max",
2700 CTLFLAG_RD, &sc->hn_tx_chimney_max, 0,
2701 "Chimney send packet size upper boundary");
2702 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_size",
2703 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2704 hn_tx_chimney_size_sysctl,
2705 "I", "Chimney send packet size limit");
2706 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "direct_tx_size",
2707 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2708 __offsetof(struct hn_tx_ring, hn_direct_tx_size),
2709 hn_tx_conf_int_sysctl, "I",
2710 "Size of the packet for direct transmission");
2711 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "sched_tx",
2712 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2713 __offsetof(struct hn_tx_ring, hn_sched_tx),
2714 hn_tx_conf_int_sysctl, "I",
2715 "Always schedule transmission "
2716 "instead of doing direct transmission");
2717 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_cnt",
2718 CTLFLAG_RD, &sc->hn_tx_ring_cnt, 0, "# created TX rings");
2719 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_inuse",
2720 CTLFLAG_RD, &sc->hn_tx_ring_inuse, 0, "# used TX rings");
2726 hn_set_tx_chimney_size(struct hn_softc *sc, int chimney_size)
2731 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
2732 sc->hn_tx_ring[i].hn_tx_chimney_size = chimney_size;
2737 hn_destroy_tx_data(struct hn_softc *sc)
2741 if (sc->hn_tx_ring_cnt == 0)
2744 for (i = 0; i < sc->hn_tx_ring_cnt; ++i)
2745 hn_destroy_tx_ring(&sc->hn_tx_ring[i]);
2747 free(sc->hn_tx_ring, M_NETVSC);
2748 sc->hn_tx_ring = NULL;
2750 sc->hn_tx_ring_cnt = 0;
2751 sc->hn_tx_ring_inuse = 0;
2755 hn_start_taskfunc(void *xtxr, int pending __unused)
2757 struct hn_tx_ring *txr = xtxr;
2759 mtx_lock(&txr->hn_tx_lock);
2760 hn_start_locked(txr, 0);
2761 mtx_unlock(&txr->hn_tx_lock);
2765 hn_start_txeof_taskfunc(void *xtxr, int pending __unused)
2767 struct hn_tx_ring *txr = xtxr;
2769 mtx_lock(&txr->hn_tx_lock);
2770 atomic_clear_int(&txr->hn_sc->hn_ifp->if_drv_flags, IFF_DRV_OACTIVE);
2771 hn_start_locked(txr, 0);
2772 mtx_unlock(&txr->hn_tx_lock);
2776 hn_stop_tx_tasks(struct hn_softc *sc)
2780 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2781 struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
2783 taskqueue_drain(txr->hn_tx_taskq, &txr->hn_tx_task);
2784 taskqueue_drain(txr->hn_tx_taskq, &txr->hn_txeof_task);
2789 hn_xmit(struct hn_tx_ring *txr, int len)
2791 struct hn_softc *sc = txr->hn_sc;
2792 struct ifnet *ifp = sc->hn_ifp;
2793 struct mbuf *m_head;
2795 mtx_assert(&txr->hn_tx_lock, MA_OWNED);
2796 KASSERT(hn_use_if_start == 0,
2797 ("hn_xmit is called, when if_start is enabled"));
2799 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || txr->hn_oactive)
2802 while ((m_head = drbr_peek(ifp, txr->hn_mbuf_br)) != NULL) {
2803 struct hn_txdesc *txd;
2806 if (len > 0 && m_head->m_pkthdr.len > len) {
2808 * This sending could be time consuming; let callers
2809 * dispatch this packet sending (and sending of any
2810 * following up packets) to tx taskqueue.
2812 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2816 txd = hn_txdesc_get(txr);
2818 txr->hn_no_txdescs++;
2819 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2820 txr->hn_oactive = 1;
2824 error = hn_encap(txr, txd, &m_head);
2826 /* Both txd and m_head are freed; discard */
2827 drbr_advance(ifp, txr->hn_mbuf_br);
2831 error = hn_send_pkt(ifp, txr, txd);
2832 if (__predict_false(error)) {
2833 /* txd is freed, but m_head is not */
2834 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2835 txr->hn_oactive = 1;
2840 drbr_advance(ifp, txr->hn_mbuf_br);
2846 hn_transmit(struct ifnet *ifp, struct mbuf *m)
2848 struct hn_softc *sc = ifp->if_softc;
2849 struct hn_tx_ring *txr;
2853 * Select the TX ring based on flowid
2855 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2856 idx = m->m_pkthdr.flowid % sc->hn_tx_ring_inuse;
2857 txr = &sc->hn_tx_ring[idx];
2859 error = drbr_enqueue(ifp, txr->hn_mbuf_br, m);
2861 if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1);
2865 if (txr->hn_oactive)
2868 if (txr->hn_sched_tx)
2871 if (mtx_trylock(&txr->hn_tx_lock)) {
2874 sched = hn_xmit(txr, txr->hn_direct_tx_size);
2875 mtx_unlock(&txr->hn_tx_lock);
2880 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
2885 hn_xmit_qflush(struct ifnet *ifp)
2887 struct hn_softc *sc = ifp->if_softc;
2890 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2891 struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
2894 mtx_lock(&txr->hn_tx_lock);
2895 while ((m = buf_ring_dequeue_sc(txr->hn_mbuf_br)) != NULL)
2897 mtx_unlock(&txr->hn_tx_lock);
2903 hn_xmit_txeof(struct hn_tx_ring *txr)
2906 if (txr->hn_sched_tx)
2909 if (mtx_trylock(&txr->hn_tx_lock)) {
2912 txr->hn_oactive = 0;
2913 sched = hn_xmit(txr, txr->hn_direct_tx_size);
2914 mtx_unlock(&txr->hn_tx_lock);
2916 taskqueue_enqueue(txr->hn_tx_taskq,
2922 * Release the oactive earlier, with the hope, that
2923 * others could catch up. The task will clear the
2924 * oactive again with the hn_tx_lock to avoid possible
2927 txr->hn_oactive = 0;
2928 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
2933 hn_xmit_taskfunc(void *xtxr, int pending __unused)
2935 struct hn_tx_ring *txr = xtxr;
2937 mtx_lock(&txr->hn_tx_lock);
2939 mtx_unlock(&txr->hn_tx_lock);
2943 hn_xmit_txeof_taskfunc(void *xtxr, int pending __unused)
2945 struct hn_tx_ring *txr = xtxr;
2947 mtx_lock(&txr->hn_tx_lock);
2948 txr->hn_oactive = 0;
2950 mtx_unlock(&txr->hn_tx_lock);
2954 hn_channel_attach(struct hn_softc *sc, struct hv_vmbus_channel *chan)
2956 struct hn_rx_ring *rxr;
2959 idx = chan->offer_msg.offer.sub_channel_index;
2961 KASSERT(idx >= 0 && idx < sc->hn_rx_ring_inuse,
2962 ("invalid channel index %d, should > 0 && < %d",
2963 idx, sc->hn_rx_ring_inuse));
2964 rxr = &sc->hn_rx_ring[idx];
2965 KASSERT((rxr->hn_rx_flags & HN_RX_FLAG_ATTACHED) == 0,
2966 ("RX ring %d already attached", idx));
2967 rxr->hn_rx_flags |= HN_RX_FLAG_ATTACHED;
2969 chan->hv_chan_rxr = rxr;
2971 if_printf(sc->hn_ifp, "link RX ring %d to channel%u\n",
2972 idx, chan->offer_msg.child_rel_id);
2975 if (idx < sc->hn_tx_ring_inuse) {
2976 struct hn_tx_ring *txr = &sc->hn_tx_ring[idx];
2978 KASSERT((txr->hn_tx_flags & HN_TX_FLAG_ATTACHED) == 0,
2979 ("TX ring %d already attached", idx));
2980 txr->hn_tx_flags |= HN_TX_FLAG_ATTACHED;
2982 chan->hv_chan_txr = txr;
2983 txr->hn_chan = chan;
2985 if_printf(sc->hn_ifp, "link TX ring %d to channel%u\n",
2986 idx, chan->offer_msg.child_rel_id);
2990 /* Bind channel to a proper CPU */
2991 vmbus_channel_cpu_set(chan, (sc->hn_cpu + idx) % mp_ncpus);
2995 hn_subchan_attach(struct hn_softc *sc, struct hv_vmbus_channel *chan)
2998 KASSERT(!HV_VMBUS_CHAN_ISPRIMARY(chan),
2999 ("subchannel callback on primary channel"));
3000 KASSERT(chan->offer_msg.offer.sub_channel_index > 0,
3001 ("invalid channel subidx %u",
3002 chan->offer_msg.offer.sub_channel_index));
3003 hn_channel_attach(sc, chan);
3007 hn_tx_taskq_create(void *arg __unused)
3009 if (!hn_share_tx_taskq)
3012 hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
3013 taskqueue_thread_enqueue, &hn_tx_taskq);
3014 taskqueue_start_threads(&hn_tx_taskq, 1, PI_NET, "hn tx");
3015 if (hn_bind_tx_taskq >= 0) {
3016 int cpu = hn_bind_tx_taskq;
3017 struct task cpuset_task;
3020 if (cpu > mp_ncpus - 1)
3022 CPU_SETOF(cpu, &cpu_set);
3023 TASK_INIT(&cpuset_task, 0, hn_cpuset_setthread_task, &cpu_set);
3024 taskqueue_enqueue(hn_tx_taskq, &cpuset_task);
3025 taskqueue_drain(hn_tx_taskq, &cpuset_task);
3028 SYSINIT(hn_txtq_create, SI_SUB_DRIVERS, SI_ORDER_FIRST,
3029 hn_tx_taskq_create, NULL);
3032 hn_tx_taskq_destroy(void *arg __unused)
3034 if (hn_tx_taskq != NULL)
3035 taskqueue_free(hn_tx_taskq);
3037 SYSUNINIT(hn_txtq_destroy, SI_SUB_DRIVERS, SI_ORDER_FIRST,
3038 hn_tx_taskq_destroy, NULL);
3040 static device_method_t netvsc_methods[] = {
3041 /* Device interface */
3042 DEVMETHOD(device_probe, netvsc_probe),
3043 DEVMETHOD(device_attach, netvsc_attach),
3044 DEVMETHOD(device_detach, netvsc_detach),
3045 DEVMETHOD(device_shutdown, netvsc_shutdown),
3050 static driver_t netvsc_driver = {
3056 static devclass_t netvsc_devclass;
3058 DRIVER_MODULE(hn, vmbus, netvsc_driver, netvsc_devclass, 0, 0);
3059 MODULE_VERSION(hn, 1);
3060 MODULE_DEPEND(hn, vmbus, 1, 1, 1);