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>
120 #include "hv_net_vsc.h"
121 #include "hv_rndis.h"
122 #include "hv_rndis_filter.h"
123 #include "vmbus_if.h"
125 #define hv_chan_rxr hv_chan_priv1
126 #define hv_chan_txr hv_chan_priv2
128 /* Short for Hyper-V network interface */
129 #define NETVSC_DEVNAME "hn"
132 * It looks like offset 0 of buf is reserved to hold the softc pointer.
133 * The sc pointer evidently not needed, and is not presently populated.
134 * The packet offset is where the netvsc_packet starts in the buffer.
136 #define HV_NV_SC_PTR_OFFSET_IN_BUF 0
137 #define HV_NV_PACKET_OFFSET_IN_BUF 16
139 /* YYY should get it from the underlying channel */
140 #define HN_TX_DESC_CNT 512
142 #define HN_LROENT_CNT_DEF 128
144 #define HN_RING_CNT_DEF_MAX 8
146 #define HN_RNDIS_MSG_LEN \
147 (sizeof(rndis_msg) + \
148 RNDIS_HASHVAL_PPI_SIZE + \
149 RNDIS_VLAN_PPI_SIZE + \
150 RNDIS_TSO_PPI_SIZE + \
152 #define HN_RNDIS_MSG_BOUNDARY PAGE_SIZE
153 #define HN_RNDIS_MSG_ALIGN CACHE_LINE_SIZE
155 #define HN_TX_DATA_BOUNDARY PAGE_SIZE
156 #define HN_TX_DATA_MAXSIZE IP_MAXPACKET
157 #define HN_TX_DATA_SEGSIZE PAGE_SIZE
158 #define HN_TX_DATA_SEGCNT_MAX \
159 (NETVSC_PACKET_MAXPAGE - HV_RF_NUM_TX_RESERVED_PAGE_BUFS)
161 #define HN_DIRECT_TX_SIZE_DEF 128
163 #define HN_EARLY_TXEOF_THRESH 8
166 #ifndef HN_USE_TXDESC_BUFRING
167 SLIST_ENTRY(hn_txdesc) link;
170 struct hn_tx_ring *txr;
172 uint32_t flags; /* HN_TXD_FLAG_ */
173 netvsc_packet netvsc_pkt; /* XXX to be removed */
175 bus_dmamap_t data_dmap;
177 bus_addr_t rndis_msg_paddr;
178 rndis_msg *rndis_msg;
179 bus_dmamap_t rndis_msg_dmap;
182 #define HN_TXD_FLAG_ONLIST 0x1
183 #define HN_TXD_FLAG_DMAMAP 0x2
186 * Only enable UDP checksum offloading when it is on 2012R2 or
187 * later. UDP checksum offloading doesn't work on earlier
190 #define HN_CSUM_ASSIST_WIN8 (CSUM_IP | CSUM_TCP)
191 #define HN_CSUM_ASSIST (CSUM_IP | CSUM_UDP | CSUM_TCP)
193 #define HN_LRO_LENLIM_MULTIRX_DEF (12 * ETHERMTU)
194 #define HN_LRO_LENLIM_DEF (25 * ETHERMTU)
195 /* YYY 2*MTU is a bit rough, but should be good enough. */
196 #define HN_LRO_LENLIM_MIN(ifp) (2 * (ifp)->if_mtu)
198 #define HN_LRO_ACKCNT_DEF 1
201 * Be aware that this sleepable mutex will exhibit WITNESS errors when
202 * certain TCP and ARP code paths are taken. This appears to be a
203 * well-known condition, as all other drivers checked use a sleeping
204 * mutex to protect their transmit paths.
205 * Also Be aware that mutexes do not play well with semaphores, and there
206 * is a conflicting semaphore in a certain channel code path.
208 #define NV_LOCK_INIT(_sc, _name) \
209 mtx_init(&(_sc)->hn_lock, _name, MTX_NETWORK_LOCK, MTX_DEF)
210 #define NV_LOCK(_sc) mtx_lock(&(_sc)->hn_lock)
211 #define NV_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->hn_lock, MA_OWNED)
212 #define NV_UNLOCK(_sc) mtx_unlock(&(_sc)->hn_lock)
213 #define NV_LOCK_DESTROY(_sc) mtx_destroy(&(_sc)->hn_lock)
220 int hv_promisc_mode = 0; /* normal mode by default */
222 SYSCTL_NODE(_hw, OID_AUTO, hn, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
223 "Hyper-V network interface");
225 /* Trust tcp segements verification on host side. */
226 static int hn_trust_hosttcp = 1;
227 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hosttcp, CTLFLAG_RDTUN,
228 &hn_trust_hosttcp, 0,
229 "Trust tcp segement verification on host side, "
230 "when csum info is missing (global setting)");
232 /* Trust udp datagrams verification on host side. */
233 static int hn_trust_hostudp = 1;
234 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostudp, CTLFLAG_RDTUN,
235 &hn_trust_hostudp, 0,
236 "Trust udp datagram verification on host side, "
237 "when csum info is missing (global setting)");
239 /* Trust ip packets verification on host side. */
240 static int hn_trust_hostip = 1;
241 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostip, CTLFLAG_RDTUN,
243 "Trust ip packet verification on host side, "
244 "when csum info is missing (global setting)");
246 /* Limit TSO burst size */
247 static int hn_tso_maxlen = 0;
248 SYSCTL_INT(_hw_hn, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN,
249 &hn_tso_maxlen, 0, "TSO burst limit");
251 /* Limit chimney send size */
252 static int hn_tx_chimney_size = 0;
253 SYSCTL_INT(_hw_hn, OID_AUTO, tx_chimney_size, CTLFLAG_RDTUN,
254 &hn_tx_chimney_size, 0, "Chimney send packet size limit");
256 /* Limit the size of packet for direct transmission */
257 static int hn_direct_tx_size = HN_DIRECT_TX_SIZE_DEF;
258 SYSCTL_INT(_hw_hn, OID_AUTO, direct_tx_size, CTLFLAG_RDTUN,
259 &hn_direct_tx_size, 0, "Size of the packet for direct transmission");
261 #if defined(INET) || defined(INET6)
262 #if __FreeBSD_version >= 1100095
263 static int hn_lro_entry_count = HN_LROENT_CNT_DEF;
264 SYSCTL_INT(_hw_hn, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN,
265 &hn_lro_entry_count, 0, "LRO entry count");
269 static int hn_share_tx_taskq = 0;
270 SYSCTL_INT(_hw_hn, OID_AUTO, share_tx_taskq, CTLFLAG_RDTUN,
271 &hn_share_tx_taskq, 0, "Enable shared TX taskqueue");
273 static struct taskqueue *hn_tx_taskq;
275 #ifndef HN_USE_TXDESC_BUFRING
276 static int hn_use_txdesc_bufring = 0;
278 static int hn_use_txdesc_bufring = 1;
280 SYSCTL_INT(_hw_hn, OID_AUTO, use_txdesc_bufring, CTLFLAG_RD,
281 &hn_use_txdesc_bufring, 0, "Use buf_ring for TX descriptors");
283 static int hn_bind_tx_taskq = -1;
284 SYSCTL_INT(_hw_hn, OID_AUTO, bind_tx_taskq, CTLFLAG_RDTUN,
285 &hn_bind_tx_taskq, 0, "Bind TX taskqueue to the specified cpu");
287 static int hn_use_if_start = 0;
288 SYSCTL_INT(_hw_hn, OID_AUTO, use_if_start, CTLFLAG_RDTUN,
289 &hn_use_if_start, 0, "Use if_start TX method");
291 static int hn_chan_cnt = 0;
292 SYSCTL_INT(_hw_hn, OID_AUTO, chan_cnt, CTLFLAG_RDTUN,
294 "# of channels to use; each channel has one RX ring and one TX ring");
296 static int hn_tx_ring_cnt = 0;
297 SYSCTL_INT(_hw_hn, OID_AUTO, tx_ring_cnt, CTLFLAG_RDTUN,
298 &hn_tx_ring_cnt, 0, "# of TX rings to use");
300 static int hn_tx_swq_depth = 0;
301 SYSCTL_INT(_hw_hn, OID_AUTO, tx_swq_depth, CTLFLAG_RDTUN,
302 &hn_tx_swq_depth, 0, "Depth of IFQ or BUFRING");
304 #if __FreeBSD_version >= 1100095
305 static u_int hn_lro_mbufq_depth = 0;
306 SYSCTL_UINT(_hw_hn, OID_AUTO, lro_mbufq_depth, CTLFLAG_RDTUN,
307 &hn_lro_mbufq_depth, 0, "Depth of LRO mbuf queue");
310 static u_int hn_cpu_index;
313 * Forward declarations
315 static void hn_stop(hn_softc_t *sc);
316 static void hn_ifinit_locked(hn_softc_t *sc);
317 static void hn_ifinit(void *xsc);
318 static int hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data);
319 static int hn_start_locked(struct hn_tx_ring *txr, int len);
320 static void hn_start(struct ifnet *ifp);
321 static void hn_start_txeof(struct hn_tx_ring *);
322 static int hn_ifmedia_upd(struct ifnet *ifp);
323 static void hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr);
324 #if __FreeBSD_version >= 1100099
325 static int hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS);
326 static int hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS);
328 static int hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS);
329 static int hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS);
330 #if __FreeBSD_version < 1100095
331 static int hn_rx_stat_int_sysctl(SYSCTL_HANDLER_ARGS);
333 static int hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS);
335 static int hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS);
336 static int hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS);
337 static int hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS);
338 static int hn_check_iplen(const struct mbuf *, int);
339 static int hn_create_tx_ring(struct hn_softc *, int);
340 static void hn_destroy_tx_ring(struct hn_tx_ring *);
341 static int hn_create_tx_data(struct hn_softc *, int);
342 static void hn_destroy_tx_data(struct hn_softc *);
343 static void hn_start_taskfunc(void *, int);
344 static void hn_start_txeof_taskfunc(void *, int);
345 static void hn_stop_tx_tasks(struct hn_softc *);
346 static int hn_encap(struct hn_tx_ring *, struct hn_txdesc *, struct mbuf **);
347 static void hn_create_rx_data(struct hn_softc *sc, int);
348 static void hn_destroy_rx_data(struct hn_softc *sc);
349 static void hn_set_tx_chimney_size(struct hn_softc *, int);
350 static void hn_channel_attach(struct hn_softc *, struct hv_vmbus_channel *);
351 static void hn_subchan_attach(struct hn_softc *, struct hv_vmbus_channel *);
352 static void hn_subchan_setup(struct hn_softc *);
354 static int hn_transmit(struct ifnet *, struct mbuf *);
355 static void hn_xmit_qflush(struct ifnet *);
356 static int hn_xmit(struct hn_tx_ring *, int);
357 static void hn_xmit_txeof(struct hn_tx_ring *);
358 static void hn_xmit_taskfunc(void *, int);
359 static void hn_xmit_txeof_taskfunc(void *, int);
361 #if __FreeBSD_version >= 1100099
363 hn_set_lro_lenlim(struct hn_softc *sc, int lenlim)
367 for (i = 0; i < sc->hn_rx_ring_inuse; ++i)
368 sc->hn_rx_ring[i].hn_lro.lro_length_lim = lenlim;
373 hn_get_txswq_depth(const struct hn_tx_ring *txr)
376 KASSERT(txr->hn_txdesc_cnt > 0, ("tx ring is not setup yet"));
377 if (hn_tx_swq_depth < txr->hn_txdesc_cnt)
378 return txr->hn_txdesc_cnt;
379 return hn_tx_swq_depth;
383 hn_ifmedia_upd(struct ifnet *ifp __unused)
390 hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
392 struct hn_softc *sc = ifp->if_softc;
394 ifmr->ifm_status = IFM_AVALID;
395 ifmr->ifm_active = IFM_ETHER;
397 if (!sc->hn_carrier) {
398 ifmr->ifm_active |= IFM_NONE;
401 ifmr->ifm_status |= IFM_ACTIVE;
402 ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
405 /* {F8615163-DF3E-46c5-913F-F2D2F965ED0E} */
406 static const hv_guid g_net_vsc_device_type = {
407 .data = {0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
408 0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E}
412 * Standard probe entry point.
416 netvsc_probe(device_t dev)
420 p = vmbus_get_type(dev);
421 if (!memcmp(p, &g_net_vsc_device_type.data, sizeof(hv_guid))) {
422 device_set_desc(dev, "Hyper-V Network Interface");
424 printf("Netvsc probe... DONE \n");
426 return (BUS_PROBE_DEFAULT);
433 hn_cpuset_setthread_task(void *xmask, int pending __unused)
435 cpuset_t *mask = xmask;
438 error = cpuset_setthread(curthread->td_tid, mask);
440 panic("curthread=%ju: can't pin; error=%d",
441 (uintmax_t)curthread->td_tid, error);
446 * Standard attach entry point.
448 * Called when the driver is loaded. It allocates needed resources,
449 * and initializes the "hardware" and software.
452 netvsc_attach(device_t dev)
454 struct hv_device *device_ctx = vmbus_get_devctx(dev);
455 struct hv_vmbus_channel *pri_chan;
456 netvsc_device_info device_info;
458 int unit = device_get_unit(dev);
459 struct ifnet *ifp = NULL;
460 int error, ring_cnt, tx_ring_cnt;
463 sc = device_get_softc(dev);
468 if (hn_tx_taskq == NULL) {
469 sc->hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
470 taskqueue_thread_enqueue, &sc->hn_tx_taskq);
471 taskqueue_start_threads(&sc->hn_tx_taskq, 1, PI_NET, "%s tx",
472 device_get_nameunit(dev));
473 if (hn_bind_tx_taskq >= 0) {
474 int cpu = hn_bind_tx_taskq;
475 struct task cpuset_task;
478 if (cpu > mp_ncpus - 1)
480 CPU_SETOF(cpu, &cpu_set);
481 TASK_INIT(&cpuset_task, 0, hn_cpuset_setthread_task,
483 taskqueue_enqueue(sc->hn_tx_taskq, &cpuset_task);
484 taskqueue_drain(sc->hn_tx_taskq, &cpuset_task);
487 sc->hn_tx_taskq = hn_tx_taskq;
489 NV_LOCK_INIT(sc, "NetVSCLock");
491 sc->hn_dev_obj = device_ctx;
493 ifp = sc->hn_ifp = sc->arpcom.ac_ifp = if_alloc(IFT_ETHER);
495 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
498 * Figure out the # of RX rings (ring_cnt) and the # of TX rings
499 * to use (tx_ring_cnt).
502 * The # of RX rings to use is same as the # of channels to use.
504 ring_cnt = hn_chan_cnt;
508 if (ring_cnt > HN_RING_CNT_DEF_MAX)
509 ring_cnt = HN_RING_CNT_DEF_MAX;
510 } else if (ring_cnt > mp_ncpus) {
514 tx_ring_cnt = hn_tx_ring_cnt;
515 if (tx_ring_cnt <= 0 || tx_ring_cnt > ring_cnt)
516 tx_ring_cnt = ring_cnt;
517 if (hn_use_if_start) {
518 /* ifnet.if_start only needs one TX ring. */
523 * Set the leader CPU for channels.
525 sc->hn_cpu = atomic_fetchadd_int(&hn_cpu_index, ring_cnt) % mp_ncpus;
527 error = hn_create_tx_data(sc, tx_ring_cnt);
530 hn_create_rx_data(sc, ring_cnt);
533 * Associate the first TX/RX ring w/ the primary channel.
535 pri_chan = device_ctx->channel;
536 KASSERT(HV_VMBUS_CHAN_ISPRIMARY(pri_chan), ("not primary channel"));
537 KASSERT(pri_chan->offer_msg.offer.sub_channel_index == 0,
538 ("primary channel subidx %u",
539 pri_chan->offer_msg.offer.sub_channel_index));
540 hn_channel_attach(sc, pri_chan);
542 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
543 ifp->if_ioctl = hn_ioctl;
544 ifp->if_init = hn_ifinit;
545 /* needed by hv_rf_on_device_add() code */
546 ifp->if_mtu = ETHERMTU;
547 if (hn_use_if_start) {
548 int qdepth = hn_get_txswq_depth(&sc->hn_tx_ring[0]);
550 ifp->if_start = hn_start;
551 IFQ_SET_MAXLEN(&ifp->if_snd, qdepth);
552 ifp->if_snd.ifq_drv_maxlen = qdepth - 1;
553 IFQ_SET_READY(&ifp->if_snd);
555 ifp->if_transmit = hn_transmit;
556 ifp->if_qflush = hn_xmit_qflush;
559 ifmedia_init(&sc->hn_media, 0, hn_ifmedia_upd, hn_ifmedia_sts);
560 ifmedia_add(&sc->hn_media, IFM_ETHER | IFM_AUTO, 0, NULL);
561 ifmedia_set(&sc->hn_media, IFM_ETHER | IFM_AUTO);
562 /* XXX ifmedia_set really should do this for us */
563 sc->hn_media.ifm_media = sc->hn_media.ifm_cur->ifm_media;
566 * Tell upper layers that we support full VLAN capability.
568 ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
569 ifp->if_capabilities |=
570 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
573 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
575 ifp->if_hwassist = sc->hn_tx_ring[0].hn_csum_assist | CSUM_TSO;
577 error = hv_rf_on_device_add(device_ctx, &device_info, ring_cnt);
580 KASSERT(sc->net_dev->num_channel > 0 &&
581 sc->net_dev->num_channel <= sc->hn_rx_ring_inuse,
582 ("invalid channel count %u, should be less than %d",
583 sc->net_dev->num_channel, sc->hn_rx_ring_inuse));
586 * Set the # of TX/RX rings that could be used according to
587 * the # of channels that host offered.
589 if (sc->hn_tx_ring_inuse > sc->net_dev->num_channel)
590 sc->hn_tx_ring_inuse = sc->net_dev->num_channel;
591 sc->hn_rx_ring_inuse = sc->net_dev->num_channel;
592 device_printf(dev, "%d TX ring, %d RX ring\n",
593 sc->hn_tx_ring_inuse, sc->hn_rx_ring_inuse);
595 if (sc->net_dev->num_channel > 1)
596 hn_subchan_setup(sc);
598 #if __FreeBSD_version >= 1100099
599 if (sc->hn_rx_ring_inuse > 1) {
601 * Reduce TCP segment aggregation limit for multiple
602 * RX rings to increase ACK timeliness.
604 hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MULTIRX_DEF);
608 if (device_info.link_state == 0) {
612 tso_maxlen = hn_tso_maxlen;
613 if (tso_maxlen <= 0 || tso_maxlen > IP_MAXPACKET)
614 tso_maxlen = IP_MAXPACKET;
616 ifp->if_hw_tsomaxsegcount = HN_TX_DATA_SEGCNT_MAX;
617 ifp->if_hw_tsomaxsegsize = PAGE_SIZE;
618 ifp->if_hw_tsomax = tso_maxlen -
619 (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
621 ether_ifattach(ifp, device_info.mac_addr);
623 if_printf(ifp, "TSO: %u/%u/%u\n", ifp->if_hw_tsomax,
624 ifp->if_hw_tsomaxsegcount, ifp->if_hw_tsomaxsegsize);
626 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
627 hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max);
628 if (hn_tx_chimney_size > 0 &&
629 hn_tx_chimney_size < sc->hn_tx_chimney_max)
630 hn_set_tx_chimney_size(sc, hn_tx_chimney_size);
634 hn_destroy_tx_data(sc);
641 * Standard detach entry point
644 netvsc_detach(device_t dev)
646 struct hn_softc *sc = device_get_softc(dev);
647 struct hv_device *hv_device = vmbus_get_devctx(dev);
650 printf("netvsc_detach\n");
653 * XXXKYS: Need to clean up all our
654 * driver state; this is the driver
659 * XXXKYS: Need to stop outgoing traffic and unregister
663 hv_rf_on_device_remove(hv_device, HV_RF_NV_DESTROY_CHANNEL);
665 hn_stop_tx_tasks(sc);
667 ifmedia_removeall(&sc->hn_media);
668 hn_destroy_rx_data(sc);
669 hn_destroy_tx_data(sc);
671 if (sc->hn_tx_taskq != hn_tx_taskq)
672 taskqueue_free(sc->hn_tx_taskq);
678 * Standard shutdown entry point
681 netvsc_shutdown(device_t dev)
687 hn_txdesc_dmamap_load(struct hn_tx_ring *txr, struct hn_txdesc *txd,
688 struct mbuf **m_head, bus_dma_segment_t *segs, int *nsegs)
690 struct mbuf *m = *m_head;
693 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag, txd->data_dmap,
694 m, segs, nsegs, BUS_DMA_NOWAIT);
695 if (error == EFBIG) {
698 m_new = m_collapse(m, M_NOWAIT, HN_TX_DATA_SEGCNT_MAX);
703 txr->hn_tx_collapsed++;
705 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag,
706 txd->data_dmap, m, segs, nsegs, BUS_DMA_NOWAIT);
709 bus_dmamap_sync(txr->hn_tx_data_dtag, txd->data_dmap,
710 BUS_DMASYNC_PREWRITE);
711 txd->flags |= HN_TXD_FLAG_DMAMAP;
717 hn_txdesc_dmamap_unload(struct hn_tx_ring *txr, struct hn_txdesc *txd)
720 if (txd->flags & HN_TXD_FLAG_DMAMAP) {
721 bus_dmamap_sync(txr->hn_tx_data_dtag,
722 txd->data_dmap, BUS_DMASYNC_POSTWRITE);
723 bus_dmamap_unload(txr->hn_tx_data_dtag,
725 txd->flags &= ~HN_TXD_FLAG_DMAMAP;
730 hn_txdesc_put(struct hn_tx_ring *txr, struct hn_txdesc *txd)
733 KASSERT((txd->flags & HN_TXD_FLAG_ONLIST) == 0,
734 ("put an onlist txd %#x", txd->flags));
736 KASSERT(txd->refs > 0, ("invalid txd refs %d", txd->refs));
737 if (atomic_fetchadd_int(&txd->refs, -1) != 1)
740 hn_txdesc_dmamap_unload(txr, txd);
741 if (txd->m != NULL) {
746 txd->flags |= HN_TXD_FLAG_ONLIST;
748 #ifndef HN_USE_TXDESC_BUFRING
749 mtx_lock_spin(&txr->hn_txlist_spin);
750 KASSERT(txr->hn_txdesc_avail >= 0 &&
751 txr->hn_txdesc_avail < txr->hn_txdesc_cnt,
752 ("txdesc_put: invalid txd avail %d", txr->hn_txdesc_avail));
753 txr->hn_txdesc_avail++;
754 SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
755 mtx_unlock_spin(&txr->hn_txlist_spin);
757 atomic_add_int(&txr->hn_txdesc_avail, 1);
758 buf_ring_enqueue(txr->hn_txdesc_br, txd);
764 static __inline struct hn_txdesc *
765 hn_txdesc_get(struct hn_tx_ring *txr)
767 struct hn_txdesc *txd;
769 #ifndef HN_USE_TXDESC_BUFRING
770 mtx_lock_spin(&txr->hn_txlist_spin);
771 txd = SLIST_FIRST(&txr->hn_txlist);
773 KASSERT(txr->hn_txdesc_avail > 0,
774 ("txdesc_get: invalid txd avail %d", txr->hn_txdesc_avail));
775 txr->hn_txdesc_avail--;
776 SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
778 mtx_unlock_spin(&txr->hn_txlist_spin);
780 txd = buf_ring_dequeue_sc(txr->hn_txdesc_br);
784 #ifdef HN_USE_TXDESC_BUFRING
785 atomic_subtract_int(&txr->hn_txdesc_avail, 1);
787 KASSERT(txd->m == NULL && txd->refs == 0 &&
788 (txd->flags & HN_TXD_FLAG_ONLIST), ("invalid txd"));
789 txd->flags &= ~HN_TXD_FLAG_ONLIST;
796 hn_txdesc_hold(struct hn_txdesc *txd)
799 /* 0->1 transition will never work */
800 KASSERT(txd->refs > 0, ("invalid refs %d", txd->refs));
801 atomic_add_int(&txd->refs, 1);
805 hn_txeof(struct hn_tx_ring *txr)
807 txr->hn_has_txeof = 0;
812 hn_tx_done(struct hv_vmbus_channel *chan, void *xpkt)
814 netvsc_packet *packet = xpkt;
815 struct hn_txdesc *txd;
816 struct hn_tx_ring *txr;
818 txd = (struct hn_txdesc *)(uintptr_t)
819 packet->compl.send.send_completion_tid;
822 KASSERT(txr->hn_chan == chan,
823 ("channel mismatch, on channel%u, should be channel%u",
824 chan->offer_msg.offer.sub_channel_index,
825 txr->hn_chan->offer_msg.offer.sub_channel_index));
827 txr->hn_has_txeof = 1;
828 hn_txdesc_put(txr, txd);
830 ++txr->hn_txdone_cnt;
831 if (txr->hn_txdone_cnt >= HN_EARLY_TXEOF_THRESH) {
832 txr->hn_txdone_cnt = 0;
839 netvsc_channel_rollup(struct hv_vmbus_channel *chan)
841 struct hn_tx_ring *txr = chan->hv_chan_txr;
842 #if defined(INET) || defined(INET6)
843 struct hn_rx_ring *rxr = chan->hv_chan_rxr;
844 struct lro_ctrl *lro = &rxr->hn_lro;
845 struct lro_entry *queued;
847 while ((queued = SLIST_FIRST(&lro->lro_active)) != NULL) {
848 SLIST_REMOVE_HEAD(&lro->lro_active, next);
849 tcp_lro_flush(lro, queued);
855 * 'txr' could be NULL, if multiple channels and
856 * ifnet.if_start method are enabled.
858 if (txr == NULL || !txr->hn_has_txeof)
861 txr->hn_txdone_cnt = 0;
867 * If this function fails, then both txd and m_head0 will be freed.
870 hn_encap(struct hn_tx_ring *txr, struct hn_txdesc *txd, struct mbuf **m_head0)
872 bus_dma_segment_t segs[HN_TX_DATA_SEGCNT_MAX];
874 struct mbuf *m_head = *m_head0;
875 netvsc_packet *packet;
876 rndis_msg *rndis_mesg;
877 rndis_packet *rndis_pkt;
878 rndis_per_packet_info *rppi;
879 struct rndis_hash_value *hash_value;
880 uint32_t rndis_msg_size;
882 packet = &txd->netvsc_pkt;
883 packet->is_data_pkt = TRUE;
884 packet->tot_data_buf_len = m_head->m_pkthdr.len;
887 * extension points to the area reserved for the
888 * rndis_filter_packet, which is placed just after
889 * the netvsc_packet (and rppi struct, if present;
890 * length is updated later).
892 rndis_mesg = txd->rndis_msg;
893 /* XXX not necessary */
894 memset(rndis_mesg, 0, HN_RNDIS_MSG_LEN);
895 rndis_mesg->ndis_msg_type = REMOTE_NDIS_PACKET_MSG;
897 rndis_pkt = &rndis_mesg->msg.packet;
898 rndis_pkt->data_offset = sizeof(rndis_packet);
899 rndis_pkt->data_length = packet->tot_data_buf_len;
900 rndis_pkt->per_pkt_info_offset = sizeof(rndis_packet);
902 rndis_msg_size = RNDIS_MESSAGE_SIZE(rndis_packet);
905 * Set the hash value for this packet, so that the host could
906 * dispatch the TX done event for this packet back to this TX
909 rndis_msg_size += RNDIS_HASHVAL_PPI_SIZE;
910 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_HASHVAL_PPI_SIZE,
912 hash_value = (struct rndis_hash_value *)((uint8_t *)rppi +
913 rppi->per_packet_info_offset);
914 hash_value->hash_value = txr->hn_tx_idx;
916 if (m_head->m_flags & M_VLANTAG) {
917 ndis_8021q_info *rppi_vlan_info;
919 rndis_msg_size += RNDIS_VLAN_PPI_SIZE;
920 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_VLAN_PPI_SIZE,
923 rppi_vlan_info = (ndis_8021q_info *)((uint8_t *)rppi +
924 rppi->per_packet_info_offset);
925 rppi_vlan_info->u1.s1.vlan_id =
926 m_head->m_pkthdr.ether_vtag & 0xfff;
929 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
930 rndis_tcp_tso_info *tso_info;
931 struct ether_vlan_header *eh;
935 * XXX need m_pullup and use mtodo
937 eh = mtod(m_head, struct ether_vlan_header*);
938 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN))
939 ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
941 ether_len = ETHER_HDR_LEN;
943 rndis_msg_size += RNDIS_TSO_PPI_SIZE;
944 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_TSO_PPI_SIZE,
945 tcp_large_send_info);
947 tso_info = (rndis_tcp_tso_info *)((uint8_t *)rppi +
948 rppi->per_packet_info_offset);
949 tso_info->lso_v2_xmit.type =
950 RNDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
953 if (m_head->m_pkthdr.csum_flags & CSUM_IP_TSO) {
955 (struct ip *)(m_head->m_data + ether_len);
956 unsigned long iph_len = ip->ip_hl << 2;
958 (struct tcphdr *)((caddr_t)ip + iph_len);
960 tso_info->lso_v2_xmit.ip_version =
961 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
965 th->th_sum = in_pseudo(ip->ip_src.s_addr,
966 ip->ip_dst.s_addr, htons(IPPROTO_TCP));
969 #if defined(INET6) && defined(INET)
974 struct ip6_hdr *ip6 = (struct ip6_hdr *)
975 (m_head->m_data + ether_len);
976 struct tcphdr *th = (struct tcphdr *)(ip6 + 1);
978 tso_info->lso_v2_xmit.ip_version =
979 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
981 th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
984 tso_info->lso_v2_xmit.tcp_header_offset = 0;
985 tso_info->lso_v2_xmit.mss = m_head->m_pkthdr.tso_segsz;
986 } else if (m_head->m_pkthdr.csum_flags & txr->hn_csum_assist) {
987 rndis_tcp_ip_csum_info *csum_info;
989 rndis_msg_size += RNDIS_CSUM_PPI_SIZE;
990 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_CSUM_PPI_SIZE,
992 csum_info = (rndis_tcp_ip_csum_info *)((uint8_t *)rppi +
993 rppi->per_packet_info_offset);
995 csum_info->xmit.is_ipv4 = 1;
996 if (m_head->m_pkthdr.csum_flags & CSUM_IP)
997 csum_info->xmit.ip_header_csum = 1;
999 if (m_head->m_pkthdr.csum_flags & CSUM_TCP) {
1000 csum_info->xmit.tcp_csum = 1;
1001 csum_info->xmit.tcp_header_offset = 0;
1002 } else if (m_head->m_pkthdr.csum_flags & CSUM_UDP) {
1003 csum_info->xmit.udp_csum = 1;
1007 rndis_mesg->msg_len = packet->tot_data_buf_len + rndis_msg_size;
1008 packet->tot_data_buf_len = rndis_mesg->msg_len;
1011 * Chimney send, if the packet could fit into one chimney buffer.
1013 if (packet->tot_data_buf_len < txr->hn_tx_chimney_size) {
1014 netvsc_dev *net_dev = txr->hn_sc->net_dev;
1015 uint32_t send_buf_section_idx;
1017 txr->hn_tx_chimney_tried++;
1018 send_buf_section_idx =
1019 hv_nv_get_next_send_section(net_dev);
1020 if (send_buf_section_idx !=
1021 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX) {
1022 uint8_t *dest = ((uint8_t *)net_dev->send_buf +
1023 (send_buf_section_idx *
1024 net_dev->send_section_size));
1026 memcpy(dest, rndis_mesg, rndis_msg_size);
1027 dest += rndis_msg_size;
1028 m_copydata(m_head, 0, m_head->m_pkthdr.len, dest);
1030 packet->send_buf_section_idx = send_buf_section_idx;
1031 packet->send_buf_section_size =
1032 packet->tot_data_buf_len;
1033 packet->page_buf_count = 0;
1034 txr->hn_tx_chimney++;
1039 error = hn_txdesc_dmamap_load(txr, txd, &m_head, segs, &nsegs);
1044 * This mbuf is not linked w/ the txd yet, so free it now.
1049 freed = hn_txdesc_put(txr, txd);
1051 ("fail to free txd upon txdma error"));
1053 txr->hn_txdma_failed++;
1054 if_inc_counter(txr->hn_sc->hn_ifp, IFCOUNTER_OERRORS, 1);
1059 packet->page_buf_count = nsegs + HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
1061 /* send packet with page buffer */
1062 packet->page_buffers[0].pfn = atop(txd->rndis_msg_paddr);
1063 packet->page_buffers[0].offset = txd->rndis_msg_paddr & PAGE_MASK;
1064 packet->page_buffers[0].length = rndis_msg_size;
1067 * Fill the page buffers with mbuf info starting at index
1068 * HV_RF_NUM_TX_RESERVED_PAGE_BUFS.
1070 for (i = 0; i < nsegs; ++i) {
1071 hv_vmbus_page_buffer *pb = &packet->page_buffers[
1072 i + HV_RF_NUM_TX_RESERVED_PAGE_BUFS];
1074 pb->pfn = atop(segs[i].ds_addr);
1075 pb->offset = segs[i].ds_addr & PAGE_MASK;
1076 pb->length = segs[i].ds_len;
1079 packet->send_buf_section_idx =
1080 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX;
1081 packet->send_buf_section_size = 0;
1085 /* Set the completion routine */
1086 packet->compl.send.on_send_completion = hn_tx_done;
1087 packet->compl.send.send_completion_context = packet;
1088 packet->compl.send.send_completion_tid = (uint64_t)(uintptr_t)txd;
1095 * If this function fails, then txd will be freed, but the mbuf
1096 * associated w/ the txd will _not_ be freed.
1099 hn_send_pkt(struct ifnet *ifp, struct hn_tx_ring *txr, struct hn_txdesc *txd)
1101 int error, send_failed = 0;
1105 * Make sure that txd is not freed before ETHER_BPF_MTAP.
1107 hn_txdesc_hold(txd);
1108 error = hv_nv_on_send(txr->hn_chan, &txd->netvsc_pkt);
1110 ETHER_BPF_MTAP(ifp, txd->m);
1111 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1112 if (!hn_use_if_start) {
1113 if_inc_counter(ifp, IFCOUNTER_OBYTES,
1114 txd->m->m_pkthdr.len);
1115 if (txd->m->m_flags & M_MCAST)
1116 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
1120 hn_txdesc_put(txr, txd);
1122 if (__predict_false(error)) {
1126 * This should "really rarely" happen.
1128 * XXX Too many RX to be acked or too many sideband
1129 * commands to run? Ask netvsc_channel_rollup()
1130 * to kick start later.
1132 txr->hn_has_txeof = 1;
1134 txr->hn_send_failed++;
1137 * Try sending again after set hn_has_txeof;
1138 * in case that we missed the last
1139 * netvsc_channel_rollup().
1143 if_printf(ifp, "send failed\n");
1146 * Caller will perform further processing on the
1147 * associated mbuf, so don't free it in hn_txdesc_put();
1148 * only unload it from the DMA map in hn_txdesc_put(),
1152 freed = hn_txdesc_put(txr, txd);
1154 ("fail to free txd upon send error"));
1156 txr->hn_send_failed++;
1162 * Start a transmit of one or more packets
1165 hn_start_locked(struct hn_tx_ring *txr, int len)
1167 struct hn_softc *sc = txr->hn_sc;
1168 struct ifnet *ifp = sc->hn_ifp;
1170 KASSERT(hn_use_if_start,
1171 ("hn_start_locked is called, when if_start is disabled"));
1172 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1173 mtx_assert(&txr->hn_tx_lock, MA_OWNED);
1175 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1179 while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
1180 struct hn_txdesc *txd;
1181 struct mbuf *m_head;
1184 IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
1188 if (len > 0 && m_head->m_pkthdr.len > len) {
1190 * This sending could be time consuming; let callers
1191 * dispatch this packet sending (and sending of any
1192 * following up packets) to tx taskqueue.
1194 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1198 txd = hn_txdesc_get(txr);
1200 txr->hn_no_txdescs++;
1201 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1202 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1206 error = hn_encap(txr, txd, &m_head);
1208 /* Both txd and m_head are freed */
1212 error = hn_send_pkt(ifp, txr, txd);
1213 if (__predict_false(error)) {
1214 /* txd is freed, but m_head is not */
1215 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1216 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1224 * Link up/down notification
1227 netvsc_linkstatus_callback(struct hv_device *device_obj, uint32_t status)
1229 hn_softc_t *sc = device_get_softc(device_obj->device);
1239 * Append the specified data to the indicated mbuf chain,
1240 * Extend the mbuf chain if the new data does not fit in
1243 * This is a minor rewrite of m_append() from sys/kern/uipc_mbuf.c.
1244 * There should be an equivalent in the kernel mbuf code,
1245 * but there does not appear to be one yet.
1247 * Differs from m_append() in that additional mbufs are
1248 * allocated with cluster size MJUMPAGESIZE, and filled
1251 * Return 1 if able to complete the job; otherwise 0.
1254 hv_m_append(struct mbuf *m0, int len, c_caddr_t cp)
1257 int remainder, space;
1259 for (m = m0; m->m_next != NULL; m = m->m_next)
1262 space = M_TRAILINGSPACE(m);
1265 * Copy into available space.
1267 if (space > remainder)
1269 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1274 while (remainder > 0) {
1276 * Allocate a new mbuf; could check space
1277 * and allocate a cluster instead.
1279 n = m_getjcl(M_DONTWAIT, m->m_type, 0, MJUMPAGESIZE);
1282 n->m_len = min(MJUMPAGESIZE, remainder);
1283 bcopy(cp, mtod(n, caddr_t), n->m_len);
1285 remainder -= n->m_len;
1289 if (m0->m_flags & M_PKTHDR)
1290 m0->m_pkthdr.len += len - remainder;
1292 return (remainder == 0);
1295 #if defined(INET) || defined(INET6)
1297 hn_lro_rx(struct lro_ctrl *lc, struct mbuf *m)
1299 #if __FreeBSD_version >= 1100095
1300 if (hn_lro_mbufq_depth) {
1301 tcp_lro_queue_mbuf(lc, m);
1305 return tcp_lro_rx(lc, m, 0);
1310 * Called when we receive a data packet from the "wire" on the
1313 * Note: This is no longer used as a callback
1316 netvsc_recv(struct hv_vmbus_channel *chan, netvsc_packet *packet,
1317 const rndis_tcp_ip_csum_info *csum_info,
1318 const struct rndis_hash_info *hash_info,
1319 const struct rndis_hash_value *hash_value)
1321 struct hn_rx_ring *rxr = chan->hv_chan_rxr;
1322 struct ifnet *ifp = rxr->hn_ifp;
1324 int size, do_lro = 0, do_csum = 1;
1326 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1330 * Bail out if packet contains more data than configured MTU.
1332 if (packet->tot_data_buf_len > (ifp->if_mtu + ETHER_HDR_LEN)) {
1334 } else if (packet->tot_data_buf_len <= MHLEN) {
1335 m_new = m_gethdr(M_NOWAIT, MT_DATA);
1336 if (m_new == NULL) {
1337 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1340 memcpy(mtod(m_new, void *), packet->data,
1341 packet->tot_data_buf_len);
1342 m_new->m_pkthdr.len = m_new->m_len = packet->tot_data_buf_len;
1343 rxr->hn_small_pkts++;
1346 * Get an mbuf with a cluster. For packets 2K or less,
1347 * get a standard 2K cluster. For anything larger, get a
1348 * 4K cluster. Any buffers larger than 4K can cause problems
1349 * if looped around to the Hyper-V TX channel, so avoid them.
1352 if (packet->tot_data_buf_len > MCLBYTES) {
1354 size = MJUMPAGESIZE;
1357 m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, size);
1358 if (m_new == NULL) {
1359 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1363 hv_m_append(m_new, packet->tot_data_buf_len, packet->data);
1365 m_new->m_pkthdr.rcvif = ifp;
1367 if (__predict_false((ifp->if_capenable & IFCAP_RXCSUM) == 0))
1370 /* receive side checksum offload */
1371 if (csum_info != NULL) {
1372 /* IP csum offload */
1373 if (csum_info->receive.ip_csum_succeeded && do_csum) {
1374 m_new->m_pkthdr.csum_flags |=
1375 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1379 /* TCP/UDP csum offload */
1380 if ((csum_info->receive.tcp_csum_succeeded ||
1381 csum_info->receive.udp_csum_succeeded) && do_csum) {
1382 m_new->m_pkthdr.csum_flags |=
1383 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1384 m_new->m_pkthdr.csum_data = 0xffff;
1385 if (csum_info->receive.tcp_csum_succeeded)
1391 if (csum_info->receive.ip_csum_succeeded &&
1392 csum_info->receive.tcp_csum_succeeded)
1395 const struct ether_header *eh;
1400 if (m_new->m_len < hoff)
1402 eh = mtod(m_new, struct ether_header *);
1403 etype = ntohs(eh->ether_type);
1404 if (etype == ETHERTYPE_VLAN) {
1405 const struct ether_vlan_header *evl;
1407 hoff = sizeof(*evl);
1408 if (m_new->m_len < hoff)
1410 evl = mtod(m_new, struct ether_vlan_header *);
1411 etype = ntohs(evl->evl_proto);
1414 if (etype == ETHERTYPE_IP) {
1417 pr = hn_check_iplen(m_new, hoff);
1418 if (pr == IPPROTO_TCP) {
1420 (rxr->hn_trust_hcsum &
1421 HN_TRUST_HCSUM_TCP)) {
1422 rxr->hn_csum_trusted++;
1423 m_new->m_pkthdr.csum_flags |=
1424 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1425 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1426 m_new->m_pkthdr.csum_data = 0xffff;
1429 } else if (pr == IPPROTO_UDP) {
1431 (rxr->hn_trust_hcsum &
1432 HN_TRUST_HCSUM_UDP)) {
1433 rxr->hn_csum_trusted++;
1434 m_new->m_pkthdr.csum_flags |=
1435 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1436 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1437 m_new->m_pkthdr.csum_data = 0xffff;
1439 } else if (pr != IPPROTO_DONE && do_csum &&
1440 (rxr->hn_trust_hcsum & HN_TRUST_HCSUM_IP)) {
1441 rxr->hn_csum_trusted++;
1442 m_new->m_pkthdr.csum_flags |=
1443 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1448 if ((packet->vlan_tci != 0) &&
1449 (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0) {
1450 m_new->m_pkthdr.ether_vtag = packet->vlan_tci;
1451 m_new->m_flags |= M_VLANTAG;
1454 if (hash_info != NULL && hash_value != NULL) {
1455 int hash_type = M_HASHTYPE_OPAQUE;
1458 m_new->m_pkthdr.flowid = hash_value->hash_value;
1459 if ((hash_info->hash_info & NDIS_HASH_FUNCTION_MASK) ==
1460 NDIS_HASH_FUNCTION_TOEPLITZ) {
1462 (hash_info->hash_info & NDIS_HASH_TYPE_MASK);
1465 case NDIS_HASH_IPV4:
1466 hash_type = M_HASHTYPE_RSS_IPV4;
1469 case NDIS_HASH_TCP_IPV4:
1470 hash_type = M_HASHTYPE_RSS_TCP_IPV4;
1473 case NDIS_HASH_IPV6:
1474 hash_type = M_HASHTYPE_RSS_IPV6;
1477 case NDIS_HASH_IPV6_EX:
1478 hash_type = M_HASHTYPE_RSS_IPV6_EX;
1481 case NDIS_HASH_TCP_IPV6:
1482 hash_type = M_HASHTYPE_RSS_TCP_IPV6;
1485 case NDIS_HASH_TCP_IPV6_EX:
1486 hash_type = M_HASHTYPE_RSS_TCP_IPV6_EX;
1490 M_HASHTYPE_SET(m_new, hash_type);
1492 if (hash_value != NULL)
1493 m_new->m_pkthdr.flowid = hash_value->hash_value;
1495 m_new->m_pkthdr.flowid = rxr->hn_rx_idx;
1496 M_HASHTYPE_SET(m_new, M_HASHTYPE_OPAQUE);
1500 * Note: Moved RX completion back to hv_nv_on_receive() so all
1501 * messages (not just data messages) will trigger a response.
1507 if ((ifp->if_capenable & IFCAP_LRO) && do_lro) {
1508 #if defined(INET) || defined(INET6)
1509 struct lro_ctrl *lro = &rxr->hn_lro;
1512 rxr->hn_lro_tried++;
1513 if (hn_lro_rx(lro, m_new) == 0) {
1521 /* We're not holding the lock here, so don't release it */
1522 (*ifp->if_input)(ifp, m_new);
1528 * Rules for using sc->temp_unusable:
1529 * 1. sc->temp_unusable can only be read or written while holding NV_LOCK()
1530 * 2. code reading sc->temp_unusable under NV_LOCK(), and finding
1531 * sc->temp_unusable set, must release NV_LOCK() and exit
1532 * 3. to retain exclusive control of the interface,
1533 * sc->temp_unusable must be set by code before releasing NV_LOCK()
1534 * 4. only code setting sc->temp_unusable can clear sc->temp_unusable
1535 * 5. code setting sc->temp_unusable must eventually clear sc->temp_unusable
1539 * Standard ioctl entry point. Called when the user wants to configure
1543 hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1545 hn_softc_t *sc = ifp->if_softc;
1546 struct ifreq *ifr = (struct ifreq *)data;
1548 struct ifaddr *ifa = (struct ifaddr *)data;
1550 netvsc_device_info device_info;
1551 struct hv_device *hn_dev;
1552 int mask, error = 0;
1553 int retry_cnt = 500;
1559 if (ifa->ifa_addr->sa_family == AF_INET) {
1560 ifp->if_flags |= IFF_UP;
1561 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1563 arp_ifinit(ifp, ifa);
1566 error = ether_ioctl(ifp, cmd, data);
1569 hn_dev = vmbus_get_devctx(sc->hn_dev);
1571 /* Check MTU value change */
1572 if (ifp->if_mtu == ifr->ifr_mtu)
1575 if (ifr->ifr_mtu > NETVSC_MAX_CONFIGURABLE_MTU) {
1580 /* Obtain and record requested MTU */
1581 ifp->if_mtu = ifr->ifr_mtu;
1583 #if __FreeBSD_version >= 1100099
1585 * Make sure that LRO aggregation length limit is still
1586 * valid, after the MTU change.
1589 if (sc->hn_rx_ring[0].hn_lro.lro_length_lim <
1590 HN_LRO_LENLIM_MIN(ifp))
1591 hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MIN(ifp));
1597 if (!sc->temp_unusable) {
1598 sc->temp_unusable = TRUE;
1602 if (retry_cnt > 0) {
1606 } while (retry_cnt > 0);
1608 if (retry_cnt == 0) {
1613 /* We must remove and add back the device to cause the new
1614 * MTU to take effect. This includes tearing down, but not
1615 * deleting the channel, then bringing it back up.
1617 error = hv_rf_on_device_remove(hn_dev, HV_RF_NV_RETAIN_CHANNEL);
1620 sc->temp_unusable = FALSE;
1625 /* Wait for subchannels to be destroyed */
1626 vmbus_drain_subchan(hn_dev->channel);
1628 error = hv_rf_on_device_add(hn_dev, &device_info,
1629 sc->hn_rx_ring_inuse);
1632 sc->temp_unusable = FALSE;
1636 KASSERT(sc->hn_rx_ring_cnt == sc->net_dev->num_channel,
1637 ("RX ring count %d and channel count %u mismatch",
1638 sc->hn_rx_ring_cnt, sc->net_dev->num_channel));
1639 if (sc->net_dev->num_channel > 1) {
1643 * Skip the rings on primary channel; they are
1644 * handled by the hv_rf_on_device_add() above.
1646 for (r = 1; r < sc->hn_rx_ring_cnt; ++r) {
1647 sc->hn_rx_ring[r].hn_rx_flags &=
1648 ~HN_RX_FLAG_ATTACHED;
1650 for (r = 1; r < sc->hn_tx_ring_cnt; ++r) {
1651 sc->hn_tx_ring[r].hn_tx_flags &=
1652 ~HN_TX_FLAG_ATTACHED;
1654 hn_subchan_setup(sc);
1657 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
1658 if (sc->hn_tx_ring[0].hn_tx_chimney_size >
1659 sc->hn_tx_chimney_max)
1660 hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max);
1662 hn_ifinit_locked(sc);
1665 sc->temp_unusable = FALSE;
1671 if (!sc->temp_unusable) {
1672 sc->temp_unusable = TRUE;
1676 if (retry_cnt > 0) {
1680 } while (retry_cnt > 0);
1682 if (retry_cnt == 0) {
1687 if (ifp->if_flags & IFF_UP) {
1689 * If only the state of the PROMISC flag changed,
1690 * then just use the 'set promisc mode' command
1691 * instead of reinitializing the entire NIC. Doing
1692 * a full re-init means reloading the firmware and
1693 * waiting for it to start up, which may take a
1697 /* Fixme: Promiscuous mode? */
1698 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1699 ifp->if_flags & IFF_PROMISC &&
1700 !(sc->hn_if_flags & IFF_PROMISC)) {
1701 /* do something here for Hyper-V */
1702 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1703 !(ifp->if_flags & IFF_PROMISC) &&
1704 sc->hn_if_flags & IFF_PROMISC) {
1705 /* do something here for Hyper-V */
1708 hn_ifinit_locked(sc);
1710 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1715 sc->temp_unusable = FALSE;
1717 sc->hn_if_flags = ifp->if_flags;
1723 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1724 if (mask & IFCAP_TXCSUM) {
1725 ifp->if_capenable ^= IFCAP_TXCSUM;
1726 if (ifp->if_capenable & IFCAP_TXCSUM) {
1728 sc->hn_tx_ring[0].hn_csum_assist;
1731 ~sc->hn_tx_ring[0].hn_csum_assist;
1735 if (mask & IFCAP_RXCSUM)
1736 ifp->if_capenable ^= IFCAP_RXCSUM;
1738 if (mask & IFCAP_LRO)
1739 ifp->if_capenable ^= IFCAP_LRO;
1741 if (mask & IFCAP_TSO4) {
1742 ifp->if_capenable ^= IFCAP_TSO4;
1743 if (ifp->if_capenable & IFCAP_TSO4)
1744 ifp->if_hwassist |= CSUM_IP_TSO;
1746 ifp->if_hwassist &= ~CSUM_IP_TSO;
1749 if (mask & IFCAP_TSO6) {
1750 ifp->if_capenable ^= IFCAP_TSO6;
1751 if (ifp->if_capenable & IFCAP_TSO6)
1752 ifp->if_hwassist |= CSUM_IP6_TSO;
1754 ifp->if_hwassist &= ~CSUM_IP6_TSO;
1763 /* Fixme: Multicast mode? */
1764 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1766 netvsc_setmulti(sc);
1775 error = ifmedia_ioctl(ifp, ifr, &sc->hn_media, cmd);
1778 error = ether_ioctl(ifp, cmd, data);
1789 hn_stop(hn_softc_t *sc)
1793 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1798 printf(" Closing Device ...\n");
1800 atomic_clear_int(&ifp->if_drv_flags,
1801 (IFF_DRV_RUNNING | IFF_DRV_OACTIVE));
1802 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
1803 sc->hn_tx_ring[i].hn_oactive = 0;
1805 if_link_state_change(ifp, LINK_STATE_DOWN);
1806 sc->hn_initdone = 0;
1808 ret = hv_rf_on_close(device_ctx);
1812 * FreeBSD transmit entry point
1815 hn_start(struct ifnet *ifp)
1817 struct hn_softc *sc = ifp->if_softc;
1818 struct hn_tx_ring *txr = &sc->hn_tx_ring[0];
1820 if (txr->hn_sched_tx)
1823 if (mtx_trylock(&txr->hn_tx_lock)) {
1826 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1827 mtx_unlock(&txr->hn_tx_lock);
1832 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
1836 hn_start_txeof(struct hn_tx_ring *txr)
1838 struct hn_softc *sc = txr->hn_sc;
1839 struct ifnet *ifp = sc->hn_ifp;
1841 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1843 if (txr->hn_sched_tx)
1846 if (mtx_trylock(&txr->hn_tx_lock)) {
1849 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1850 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1851 mtx_unlock(&txr->hn_tx_lock);
1853 taskqueue_enqueue(txr->hn_tx_taskq,
1859 * Release the OACTIVE earlier, with the hope, that
1860 * others could catch up. The task will clear the
1861 * flag again with the hn_tx_lock to avoid possible
1864 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1865 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
1873 hn_ifinit_locked(hn_softc_t *sc)
1876 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1881 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1885 hv_promisc_mode = 1;
1887 ret = hv_rf_on_open(device_ctx);
1891 sc->hn_initdone = 1;
1894 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1895 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
1896 sc->hn_tx_ring[i].hn_oactive = 0;
1898 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_RUNNING);
1899 if_link_state_change(ifp, LINK_STATE_UP);
1906 hn_ifinit(void *xsc)
1908 hn_softc_t *sc = xsc;
1911 if (sc->temp_unusable) {
1915 sc->temp_unusable = TRUE;
1918 hn_ifinit_locked(sc);
1921 sc->temp_unusable = FALSE;
1930 hn_watchdog(struct ifnet *ifp)
1935 printf("hn%d: watchdog timeout -- resetting\n", sc->hn_unit);
1936 hn_ifinit(sc); /*???*/
1941 #if __FreeBSD_version >= 1100099
1944 hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS)
1946 struct hn_softc *sc = arg1;
1947 unsigned int lenlim;
1950 lenlim = sc->hn_rx_ring[0].hn_lro.lro_length_lim;
1951 error = sysctl_handle_int(oidp, &lenlim, 0, req);
1952 if (error || req->newptr == NULL)
1955 if (lenlim < HN_LRO_LENLIM_MIN(sc->hn_ifp) ||
1956 lenlim > TCP_LRO_LENGTH_MAX)
1960 hn_set_lro_lenlim(sc, lenlim);
1966 hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS)
1968 struct hn_softc *sc = arg1;
1969 int ackcnt, error, i;
1972 * lro_ackcnt_lim is append count limit,
1973 * +1 to turn it into aggregation limit.
1975 ackcnt = sc->hn_rx_ring[0].hn_lro.lro_ackcnt_lim + 1;
1976 error = sysctl_handle_int(oidp, &ackcnt, 0, req);
1977 if (error || req->newptr == NULL)
1980 if (ackcnt < 2 || ackcnt > (TCP_LRO_ACKCNT_MAX + 1))
1984 * Convert aggregation limit back to append
1989 for (i = 0; i < sc->hn_rx_ring_inuse; ++i)
1990 sc->hn_rx_ring[i].hn_lro.lro_ackcnt_lim = ackcnt;
1998 hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS)
2000 struct hn_softc *sc = arg1;
2005 if (sc->hn_rx_ring[0].hn_trust_hcsum & hcsum)
2008 error = sysctl_handle_int(oidp, &on, 0, req);
2009 if (error || req->newptr == NULL)
2013 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2014 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
2017 rxr->hn_trust_hcsum |= hcsum;
2019 rxr->hn_trust_hcsum &= ~hcsum;
2026 hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS)
2028 struct hn_softc *sc = arg1;
2029 int chimney_size, error;
2031 chimney_size = sc->hn_tx_ring[0].hn_tx_chimney_size;
2032 error = sysctl_handle_int(oidp, &chimney_size, 0, req);
2033 if (error || req->newptr == NULL)
2036 if (chimney_size > sc->hn_tx_chimney_max || chimney_size <= 0)
2039 hn_set_tx_chimney_size(sc, chimney_size);
2043 #if __FreeBSD_version < 1100095
2045 hn_rx_stat_int_sysctl(SYSCTL_HANDLER_ARGS)
2047 struct hn_softc *sc = arg1;
2048 int ofs = arg2, i, error;
2049 struct hn_rx_ring *rxr;
2053 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2054 rxr = &sc->hn_rx_ring[i];
2055 stat += *((int *)((uint8_t *)rxr + ofs));
2058 error = sysctl_handle_64(oidp, &stat, 0, req);
2059 if (error || req->newptr == NULL)
2062 /* Zero out this stat. */
2063 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2064 rxr = &sc->hn_rx_ring[i];
2065 *((int *)((uint8_t *)rxr + ofs)) = 0;
2071 hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS)
2073 struct hn_softc *sc = arg1;
2074 int ofs = arg2, i, error;
2075 struct hn_rx_ring *rxr;
2079 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2080 rxr = &sc->hn_rx_ring[i];
2081 stat += *((uint64_t *)((uint8_t *)rxr + ofs));
2084 error = sysctl_handle_64(oidp, &stat, 0, req);
2085 if (error || req->newptr == NULL)
2088 /* Zero out this stat. */
2089 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2090 rxr = &sc->hn_rx_ring[i];
2091 *((uint64_t *)((uint8_t *)rxr + ofs)) = 0;
2099 hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
2101 struct hn_softc *sc = arg1;
2102 int ofs = arg2, i, error;
2103 struct hn_rx_ring *rxr;
2107 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2108 rxr = &sc->hn_rx_ring[i];
2109 stat += *((u_long *)((uint8_t *)rxr + ofs));
2112 error = sysctl_handle_long(oidp, &stat, 0, req);
2113 if (error || req->newptr == NULL)
2116 /* Zero out this stat. */
2117 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2118 rxr = &sc->hn_rx_ring[i];
2119 *((u_long *)((uint8_t *)rxr + ofs)) = 0;
2125 hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
2127 struct hn_softc *sc = arg1;
2128 int ofs = arg2, i, error;
2129 struct hn_tx_ring *txr;
2133 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2134 txr = &sc->hn_tx_ring[i];
2135 stat += *((u_long *)((uint8_t *)txr + ofs));
2138 error = sysctl_handle_long(oidp, &stat, 0, req);
2139 if (error || req->newptr == NULL)
2142 /* Zero out this stat. */
2143 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2144 txr = &sc->hn_tx_ring[i];
2145 *((u_long *)((uint8_t *)txr + ofs)) = 0;
2151 hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS)
2153 struct hn_softc *sc = arg1;
2154 int ofs = arg2, i, error, conf;
2155 struct hn_tx_ring *txr;
2157 txr = &sc->hn_tx_ring[0];
2158 conf = *((int *)((uint8_t *)txr + ofs));
2160 error = sysctl_handle_int(oidp, &conf, 0, req);
2161 if (error || req->newptr == NULL)
2165 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2166 txr = &sc->hn_tx_ring[i];
2167 *((int *)((uint8_t *)txr + ofs)) = conf;
2175 hn_check_iplen(const struct mbuf *m, int hoff)
2177 const struct ip *ip;
2178 int len, iphlen, iplen;
2179 const struct tcphdr *th;
2180 int thoff; /* TCP data offset */
2182 len = hoff + sizeof(struct ip);
2184 /* The packet must be at least the size of an IP header. */
2185 if (m->m_pkthdr.len < len)
2186 return IPPROTO_DONE;
2188 /* The fixed IP header must reside completely in the first mbuf. */
2190 return IPPROTO_DONE;
2192 ip = mtodo(m, hoff);
2194 /* Bound check the packet's stated IP header length. */
2195 iphlen = ip->ip_hl << 2;
2196 if (iphlen < sizeof(struct ip)) /* minimum header length */
2197 return IPPROTO_DONE;
2199 /* The full IP header must reside completely in the one mbuf. */
2200 if (m->m_len < hoff + iphlen)
2201 return IPPROTO_DONE;
2203 iplen = ntohs(ip->ip_len);
2206 * Check that the amount of data in the buffers is as
2207 * at least much as the IP header would have us expect.
2209 if (m->m_pkthdr.len < hoff + iplen)
2210 return IPPROTO_DONE;
2213 * Ignore IP fragments.
2215 if (ntohs(ip->ip_off) & (IP_OFFMASK | IP_MF))
2216 return IPPROTO_DONE;
2219 * The TCP/IP or UDP/IP header must be entirely contained within
2220 * the first fragment of a packet.
2224 if (iplen < iphlen + sizeof(struct tcphdr))
2225 return IPPROTO_DONE;
2226 if (m->m_len < hoff + iphlen + sizeof(struct tcphdr))
2227 return IPPROTO_DONE;
2228 th = (const struct tcphdr *)((const uint8_t *)ip + iphlen);
2229 thoff = th->th_off << 2;
2230 if (thoff < sizeof(struct tcphdr) || thoff + iphlen > iplen)
2231 return IPPROTO_DONE;
2232 if (m->m_len < hoff + iphlen + thoff)
2233 return IPPROTO_DONE;
2236 if (iplen < iphlen + sizeof(struct udphdr))
2237 return IPPROTO_DONE;
2238 if (m->m_len < hoff + iphlen + sizeof(struct udphdr))
2239 return IPPROTO_DONE;
2243 return IPPROTO_DONE;
2250 hn_create_rx_data(struct hn_softc *sc, int ring_cnt)
2252 struct sysctl_oid_list *child;
2253 struct sysctl_ctx_list *ctx;
2254 device_t dev = sc->hn_dev;
2255 #if defined(INET) || defined(INET6)
2256 #if __FreeBSD_version >= 1100095
2262 sc->hn_rx_ring_cnt = ring_cnt;
2263 sc->hn_rx_ring_inuse = sc->hn_rx_ring_cnt;
2265 sc->hn_rx_ring = malloc(sizeof(struct hn_rx_ring) * sc->hn_rx_ring_cnt,
2266 M_NETVSC, M_WAITOK | M_ZERO);
2268 #if defined(INET) || defined(INET6)
2269 #if __FreeBSD_version >= 1100095
2270 lroent_cnt = hn_lro_entry_count;
2271 if (lroent_cnt < TCP_LRO_ENTRIES)
2272 lroent_cnt = TCP_LRO_ENTRIES;
2273 device_printf(dev, "LRO: entry count %d\n", lroent_cnt);
2275 #endif /* INET || INET6 */
2277 ctx = device_get_sysctl_ctx(dev);
2278 child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
2280 /* Create dev.hn.UNIT.rx sysctl tree */
2281 sc->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "rx",
2282 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2284 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2285 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
2287 if (hn_trust_hosttcp)
2288 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_TCP;
2289 if (hn_trust_hostudp)
2290 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_UDP;
2291 if (hn_trust_hostip)
2292 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_IP;
2293 rxr->hn_ifp = sc->hn_ifp;
2299 #if defined(INET) || defined(INET6)
2300 #if __FreeBSD_version >= 1100095
2301 tcp_lro_init_args(&rxr->hn_lro, sc->hn_ifp, lroent_cnt,
2302 hn_lro_mbufq_depth);
2304 tcp_lro_init(&rxr->hn_lro);
2305 rxr->hn_lro.ifp = sc->hn_ifp;
2307 #if __FreeBSD_version >= 1100099
2308 rxr->hn_lro.lro_length_lim = HN_LRO_LENLIM_DEF;
2309 rxr->hn_lro.lro_ackcnt_lim = HN_LRO_ACKCNT_DEF;
2311 #endif /* INET || INET6 */
2313 if (sc->hn_rx_sysctl_tree != NULL) {
2317 * Create per RX ring sysctl tree:
2318 * dev.hn.UNIT.rx.RINGID
2320 snprintf(name, sizeof(name), "%d", i);
2321 rxr->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx,
2322 SYSCTL_CHILDREN(sc->hn_rx_sysctl_tree),
2323 OID_AUTO, name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2325 if (rxr->hn_rx_sysctl_tree != NULL) {
2326 SYSCTL_ADD_ULONG(ctx,
2327 SYSCTL_CHILDREN(rxr->hn_rx_sysctl_tree),
2328 OID_AUTO, "packets", CTLFLAG_RW,
2329 &rxr->hn_pkts, "# of packets received");
2330 SYSCTL_ADD_ULONG(ctx,
2331 SYSCTL_CHILDREN(rxr->hn_rx_sysctl_tree),
2332 OID_AUTO, "rss_pkts", CTLFLAG_RW,
2334 "# of packets w/ RSS info received");
2339 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_queued",
2340 CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2341 __offsetof(struct hn_rx_ring, hn_lro.lro_queued),
2342 #if __FreeBSD_version < 1100095
2343 hn_rx_stat_int_sysctl,
2345 hn_rx_stat_u64_sysctl,
2347 "LU", "LRO queued");
2348 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_flushed",
2349 CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2350 __offsetof(struct hn_rx_ring, hn_lro.lro_flushed),
2351 #if __FreeBSD_version < 1100095
2352 hn_rx_stat_int_sysctl,
2354 hn_rx_stat_u64_sysctl,
2356 "LU", "LRO flushed");
2357 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_tried",
2358 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2359 __offsetof(struct hn_rx_ring, hn_lro_tried),
2360 hn_rx_stat_ulong_sysctl, "LU", "# of LRO tries");
2361 #if __FreeBSD_version >= 1100099
2362 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_length_lim",
2363 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2364 hn_lro_lenlim_sysctl, "IU",
2365 "Max # of data bytes to be aggregated by LRO");
2366 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_ackcnt_lim",
2367 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2368 hn_lro_ackcnt_sysctl, "I",
2369 "Max # of ACKs to be aggregated by LRO");
2371 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hosttcp",
2372 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_TCP,
2373 hn_trust_hcsum_sysctl, "I",
2374 "Trust tcp segement verification on host side, "
2375 "when csum info is missing");
2376 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostudp",
2377 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_UDP,
2378 hn_trust_hcsum_sysctl, "I",
2379 "Trust udp datagram verification on host side, "
2380 "when csum info is missing");
2381 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostip",
2382 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_IP,
2383 hn_trust_hcsum_sysctl, "I",
2384 "Trust ip packet verification on host side, "
2385 "when csum info is missing");
2386 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_ip",
2387 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2388 __offsetof(struct hn_rx_ring, hn_csum_ip),
2389 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM IP");
2390 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_tcp",
2391 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2392 __offsetof(struct hn_rx_ring, hn_csum_tcp),
2393 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM TCP");
2394 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_udp",
2395 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2396 __offsetof(struct hn_rx_ring, hn_csum_udp),
2397 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM UDP");
2398 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_trusted",
2399 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2400 __offsetof(struct hn_rx_ring, hn_csum_trusted),
2401 hn_rx_stat_ulong_sysctl, "LU",
2402 "# of packets that we trust host's csum verification");
2403 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "small_pkts",
2404 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2405 __offsetof(struct hn_rx_ring, hn_small_pkts),
2406 hn_rx_stat_ulong_sysctl, "LU", "# of small packets received");
2407 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_cnt",
2408 CTLFLAG_RD, &sc->hn_rx_ring_cnt, 0, "# created RX rings");
2409 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_inuse",
2410 CTLFLAG_RD, &sc->hn_rx_ring_inuse, 0, "# used RX rings");
2414 hn_destroy_rx_data(struct hn_softc *sc)
2416 #if defined(INET) || defined(INET6)
2420 if (sc->hn_rx_ring_cnt == 0)
2423 #if defined(INET) || defined(INET6)
2424 for (i = 0; i < sc->hn_rx_ring_cnt; ++i)
2425 tcp_lro_free(&sc->hn_rx_ring[i].hn_lro);
2427 free(sc->hn_rx_ring, M_NETVSC);
2428 sc->hn_rx_ring = NULL;
2430 sc->hn_rx_ring_cnt = 0;
2431 sc->hn_rx_ring_inuse = 0;
2435 hn_create_tx_ring(struct hn_softc *sc, int id)
2437 struct hn_tx_ring *txr = &sc->hn_tx_ring[id];
2438 device_t dev = sc->hn_dev;
2439 bus_dma_tag_t parent_dtag;
2444 txr->hn_tx_idx = id;
2446 #ifndef HN_USE_TXDESC_BUFRING
2447 mtx_init(&txr->hn_txlist_spin, "hn txlist", NULL, MTX_SPIN);
2449 mtx_init(&txr->hn_tx_lock, "hn tx", NULL, MTX_DEF);
2451 txr->hn_txdesc_cnt = HN_TX_DESC_CNT;
2452 txr->hn_txdesc = malloc(sizeof(struct hn_txdesc) * txr->hn_txdesc_cnt,
2453 M_NETVSC, M_WAITOK | M_ZERO);
2454 #ifndef HN_USE_TXDESC_BUFRING
2455 SLIST_INIT(&txr->hn_txlist);
2457 txr->hn_txdesc_br = buf_ring_alloc(txr->hn_txdesc_cnt, M_NETVSC,
2458 M_WAITOK, &txr->hn_tx_lock);
2461 txr->hn_tx_taskq = sc->hn_tx_taskq;
2463 if (hn_use_if_start) {
2464 txr->hn_txeof = hn_start_txeof;
2465 TASK_INIT(&txr->hn_tx_task, 0, hn_start_taskfunc, txr);
2466 TASK_INIT(&txr->hn_txeof_task, 0, hn_start_txeof_taskfunc, txr);
2470 txr->hn_txeof = hn_xmit_txeof;
2471 TASK_INIT(&txr->hn_tx_task, 0, hn_xmit_taskfunc, txr);
2472 TASK_INIT(&txr->hn_txeof_task, 0, hn_xmit_txeof_taskfunc, txr);
2474 br_depth = hn_get_txswq_depth(txr);
2475 txr->hn_mbuf_br = buf_ring_alloc(br_depth, M_NETVSC,
2476 M_WAITOK, &txr->hn_tx_lock);
2479 txr->hn_direct_tx_size = hn_direct_tx_size;
2480 version = VMBUS_GET_VERSION(device_get_parent(dev), dev);
2481 if (version >= VMBUS_VERSION_WIN8_1) {
2482 txr->hn_csum_assist = HN_CSUM_ASSIST;
2484 txr->hn_csum_assist = HN_CSUM_ASSIST_WIN8;
2486 device_printf(dev, "bus version %u.%u, "
2487 "no UDP checksum offloading\n",
2488 VMBUS_VERSION_MAJOR(version),
2489 VMBUS_VERSION_MINOR(version));
2494 * Always schedule transmission instead of trying to do direct
2495 * transmission. This one gives the best performance so far.
2497 txr->hn_sched_tx = 1;
2499 parent_dtag = bus_get_dma_tag(dev);
2501 /* DMA tag for RNDIS messages. */
2502 error = bus_dma_tag_create(parent_dtag, /* parent */
2503 HN_RNDIS_MSG_ALIGN, /* alignment */
2504 HN_RNDIS_MSG_BOUNDARY, /* boundary */
2505 BUS_SPACE_MAXADDR, /* lowaddr */
2506 BUS_SPACE_MAXADDR, /* highaddr */
2507 NULL, NULL, /* filter, filterarg */
2508 HN_RNDIS_MSG_LEN, /* maxsize */
2510 HN_RNDIS_MSG_LEN, /* maxsegsize */
2512 NULL, /* lockfunc */
2513 NULL, /* lockfuncarg */
2514 &txr->hn_tx_rndis_dtag);
2516 device_printf(dev, "failed to create rndis dmatag\n");
2520 /* DMA tag for data. */
2521 error = bus_dma_tag_create(parent_dtag, /* parent */
2523 HN_TX_DATA_BOUNDARY, /* boundary */
2524 BUS_SPACE_MAXADDR, /* lowaddr */
2525 BUS_SPACE_MAXADDR, /* highaddr */
2526 NULL, NULL, /* filter, filterarg */
2527 HN_TX_DATA_MAXSIZE, /* maxsize */
2528 HN_TX_DATA_SEGCNT_MAX, /* nsegments */
2529 HN_TX_DATA_SEGSIZE, /* maxsegsize */
2531 NULL, /* lockfunc */
2532 NULL, /* lockfuncarg */
2533 &txr->hn_tx_data_dtag);
2535 device_printf(dev, "failed to create data dmatag\n");
2539 for (i = 0; i < txr->hn_txdesc_cnt; ++i) {
2540 struct hn_txdesc *txd = &txr->hn_txdesc[i];
2545 * Allocate and load RNDIS messages.
2547 error = bus_dmamem_alloc(txr->hn_tx_rndis_dtag,
2548 (void **)&txd->rndis_msg,
2549 BUS_DMA_WAITOK | BUS_DMA_COHERENT,
2550 &txd->rndis_msg_dmap);
2553 "failed to allocate rndis_msg, %d\n", i);
2557 error = bus_dmamap_load(txr->hn_tx_rndis_dtag,
2558 txd->rndis_msg_dmap,
2559 txd->rndis_msg, HN_RNDIS_MSG_LEN,
2560 hyperv_dma_map_paddr, &txd->rndis_msg_paddr,
2564 "failed to load rndis_msg, %d\n", i);
2565 bus_dmamem_free(txr->hn_tx_rndis_dtag,
2566 txd->rndis_msg, txd->rndis_msg_dmap);
2570 /* DMA map for TX data. */
2571 error = bus_dmamap_create(txr->hn_tx_data_dtag, 0,
2575 "failed to allocate tx data dmamap\n");
2576 bus_dmamap_unload(txr->hn_tx_rndis_dtag,
2577 txd->rndis_msg_dmap);
2578 bus_dmamem_free(txr->hn_tx_rndis_dtag,
2579 txd->rndis_msg, txd->rndis_msg_dmap);
2583 /* All set, put it to list */
2584 txd->flags |= HN_TXD_FLAG_ONLIST;
2585 #ifndef HN_USE_TXDESC_BUFRING
2586 SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
2588 buf_ring_enqueue(txr->hn_txdesc_br, txd);
2591 txr->hn_txdesc_avail = txr->hn_txdesc_cnt;
2593 if (sc->hn_tx_sysctl_tree != NULL) {
2594 struct sysctl_oid_list *child;
2595 struct sysctl_ctx_list *ctx;
2599 * Create per TX ring sysctl tree:
2600 * dev.hn.UNIT.tx.RINGID
2602 ctx = device_get_sysctl_ctx(dev);
2603 child = SYSCTL_CHILDREN(sc->hn_tx_sysctl_tree);
2605 snprintf(name, sizeof(name), "%d", id);
2606 txr->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO,
2607 name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2609 if (txr->hn_tx_sysctl_tree != NULL) {
2610 child = SYSCTL_CHILDREN(txr->hn_tx_sysctl_tree);
2612 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_avail",
2613 CTLFLAG_RD, &txr->hn_txdesc_avail, 0,
2614 "# of available TX descs");
2615 if (!hn_use_if_start) {
2616 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "oactive",
2617 CTLFLAG_RD, &txr->hn_oactive, 0,
2620 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "packets",
2621 CTLFLAG_RW, &txr->hn_pkts,
2622 "# of packets transmitted");
2630 hn_txdesc_dmamap_destroy(struct hn_txdesc *txd)
2632 struct hn_tx_ring *txr = txd->txr;
2634 KASSERT(txd->m == NULL, ("still has mbuf installed"));
2635 KASSERT((txd->flags & HN_TXD_FLAG_DMAMAP) == 0, ("still dma mapped"));
2637 bus_dmamap_unload(txr->hn_tx_rndis_dtag, txd->rndis_msg_dmap);
2638 bus_dmamem_free(txr->hn_tx_rndis_dtag, txd->rndis_msg,
2639 txd->rndis_msg_dmap);
2640 bus_dmamap_destroy(txr->hn_tx_data_dtag, txd->data_dmap);
2644 hn_destroy_tx_ring(struct hn_tx_ring *txr)
2646 struct hn_txdesc *txd;
2648 if (txr->hn_txdesc == NULL)
2651 #ifndef HN_USE_TXDESC_BUFRING
2652 while ((txd = SLIST_FIRST(&txr->hn_txlist)) != NULL) {
2653 SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
2654 hn_txdesc_dmamap_destroy(txd);
2657 mtx_lock(&txr->hn_tx_lock);
2658 while ((txd = buf_ring_dequeue_sc(txr->hn_txdesc_br)) != NULL)
2659 hn_txdesc_dmamap_destroy(txd);
2660 mtx_unlock(&txr->hn_tx_lock);
2663 if (txr->hn_tx_data_dtag != NULL)
2664 bus_dma_tag_destroy(txr->hn_tx_data_dtag);
2665 if (txr->hn_tx_rndis_dtag != NULL)
2666 bus_dma_tag_destroy(txr->hn_tx_rndis_dtag);
2668 #ifdef HN_USE_TXDESC_BUFRING
2669 buf_ring_free(txr->hn_txdesc_br, M_NETVSC);
2672 free(txr->hn_txdesc, M_NETVSC);
2673 txr->hn_txdesc = NULL;
2675 if (txr->hn_mbuf_br != NULL)
2676 buf_ring_free(txr->hn_mbuf_br, M_NETVSC);
2678 #ifndef HN_USE_TXDESC_BUFRING
2679 mtx_destroy(&txr->hn_txlist_spin);
2681 mtx_destroy(&txr->hn_tx_lock);
2685 hn_create_tx_data(struct hn_softc *sc, int ring_cnt)
2687 struct sysctl_oid_list *child;
2688 struct sysctl_ctx_list *ctx;
2691 sc->hn_tx_ring_cnt = ring_cnt;
2692 sc->hn_tx_ring_inuse = sc->hn_tx_ring_cnt;
2694 sc->hn_tx_ring = malloc(sizeof(struct hn_tx_ring) * sc->hn_tx_ring_cnt,
2695 M_NETVSC, M_WAITOK | M_ZERO);
2697 ctx = device_get_sysctl_ctx(sc->hn_dev);
2698 child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->hn_dev));
2700 /* Create dev.hn.UNIT.tx sysctl tree */
2701 sc->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "tx",
2702 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2704 for (i = 0; i < sc->hn_tx_ring_cnt; ++i) {
2707 error = hn_create_tx_ring(sc, i);
2712 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "no_txdescs",
2713 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2714 __offsetof(struct hn_tx_ring, hn_no_txdescs),
2715 hn_tx_stat_ulong_sysctl, "LU", "# of times short of TX descs");
2716 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "send_failed",
2717 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2718 __offsetof(struct hn_tx_ring, hn_send_failed),
2719 hn_tx_stat_ulong_sysctl, "LU", "# of hyper-v sending failure");
2720 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "txdma_failed",
2721 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2722 __offsetof(struct hn_tx_ring, hn_txdma_failed),
2723 hn_tx_stat_ulong_sysctl, "LU", "# of TX DMA failure");
2724 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_collapsed",
2725 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2726 __offsetof(struct hn_tx_ring, hn_tx_collapsed),
2727 hn_tx_stat_ulong_sysctl, "LU", "# of TX mbuf collapsed");
2728 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney",
2729 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2730 __offsetof(struct hn_tx_ring, hn_tx_chimney),
2731 hn_tx_stat_ulong_sysctl, "LU", "# of chimney send");
2732 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_tried",
2733 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2734 __offsetof(struct hn_tx_ring, hn_tx_chimney_tried),
2735 hn_tx_stat_ulong_sysctl, "LU", "# of chimney send tries");
2736 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_cnt",
2737 CTLFLAG_RD, &sc->hn_tx_ring[0].hn_txdesc_cnt, 0,
2738 "# of total TX descs");
2739 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_chimney_max",
2740 CTLFLAG_RD, &sc->hn_tx_chimney_max, 0,
2741 "Chimney send packet size upper boundary");
2742 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_size",
2743 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2744 hn_tx_chimney_size_sysctl,
2745 "I", "Chimney send packet size limit");
2746 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "direct_tx_size",
2747 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2748 __offsetof(struct hn_tx_ring, hn_direct_tx_size),
2749 hn_tx_conf_int_sysctl, "I",
2750 "Size of the packet for direct transmission");
2751 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "sched_tx",
2752 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2753 __offsetof(struct hn_tx_ring, hn_sched_tx),
2754 hn_tx_conf_int_sysctl, "I",
2755 "Always schedule transmission "
2756 "instead of doing direct transmission");
2757 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_cnt",
2758 CTLFLAG_RD, &sc->hn_tx_ring_cnt, 0, "# created TX rings");
2759 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_inuse",
2760 CTLFLAG_RD, &sc->hn_tx_ring_inuse, 0, "# used TX rings");
2766 hn_set_tx_chimney_size(struct hn_softc *sc, int chimney_size)
2771 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
2772 sc->hn_tx_ring[i].hn_tx_chimney_size = chimney_size;
2777 hn_destroy_tx_data(struct hn_softc *sc)
2781 if (sc->hn_tx_ring_cnt == 0)
2784 for (i = 0; i < sc->hn_tx_ring_cnt; ++i)
2785 hn_destroy_tx_ring(&sc->hn_tx_ring[i]);
2787 free(sc->hn_tx_ring, M_NETVSC);
2788 sc->hn_tx_ring = NULL;
2790 sc->hn_tx_ring_cnt = 0;
2791 sc->hn_tx_ring_inuse = 0;
2795 hn_start_taskfunc(void *xtxr, int pending __unused)
2797 struct hn_tx_ring *txr = xtxr;
2799 mtx_lock(&txr->hn_tx_lock);
2800 hn_start_locked(txr, 0);
2801 mtx_unlock(&txr->hn_tx_lock);
2805 hn_start_txeof_taskfunc(void *xtxr, int pending __unused)
2807 struct hn_tx_ring *txr = xtxr;
2809 mtx_lock(&txr->hn_tx_lock);
2810 atomic_clear_int(&txr->hn_sc->hn_ifp->if_drv_flags, IFF_DRV_OACTIVE);
2811 hn_start_locked(txr, 0);
2812 mtx_unlock(&txr->hn_tx_lock);
2816 hn_stop_tx_tasks(struct hn_softc *sc)
2820 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2821 struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
2823 taskqueue_drain(txr->hn_tx_taskq, &txr->hn_tx_task);
2824 taskqueue_drain(txr->hn_tx_taskq, &txr->hn_txeof_task);
2829 hn_xmit(struct hn_tx_ring *txr, int len)
2831 struct hn_softc *sc = txr->hn_sc;
2832 struct ifnet *ifp = sc->hn_ifp;
2833 struct mbuf *m_head;
2835 mtx_assert(&txr->hn_tx_lock, MA_OWNED);
2836 KASSERT(hn_use_if_start == 0,
2837 ("hn_xmit is called, when if_start is enabled"));
2839 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || txr->hn_oactive)
2842 while ((m_head = drbr_peek(ifp, txr->hn_mbuf_br)) != NULL) {
2843 struct hn_txdesc *txd;
2846 if (len > 0 && m_head->m_pkthdr.len > len) {
2848 * This sending could be time consuming; let callers
2849 * dispatch this packet sending (and sending of any
2850 * following up packets) to tx taskqueue.
2852 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2856 txd = hn_txdesc_get(txr);
2858 txr->hn_no_txdescs++;
2859 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2860 txr->hn_oactive = 1;
2864 error = hn_encap(txr, txd, &m_head);
2866 /* Both txd and m_head are freed; discard */
2867 drbr_advance(ifp, txr->hn_mbuf_br);
2871 error = hn_send_pkt(ifp, txr, txd);
2872 if (__predict_false(error)) {
2873 /* txd is freed, but m_head is not */
2874 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2875 txr->hn_oactive = 1;
2880 drbr_advance(ifp, txr->hn_mbuf_br);
2886 hn_transmit(struct ifnet *ifp, struct mbuf *m)
2888 struct hn_softc *sc = ifp->if_softc;
2889 struct hn_tx_ring *txr;
2893 * Select the TX ring based on flowid
2895 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2896 idx = m->m_pkthdr.flowid % sc->hn_tx_ring_inuse;
2897 txr = &sc->hn_tx_ring[idx];
2899 error = drbr_enqueue(ifp, txr->hn_mbuf_br, m);
2901 if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1);
2905 if (txr->hn_oactive)
2908 if (txr->hn_sched_tx)
2911 if (mtx_trylock(&txr->hn_tx_lock)) {
2914 sched = hn_xmit(txr, txr->hn_direct_tx_size);
2915 mtx_unlock(&txr->hn_tx_lock);
2920 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
2925 hn_xmit_qflush(struct ifnet *ifp)
2927 struct hn_softc *sc = ifp->if_softc;
2930 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2931 struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
2934 mtx_lock(&txr->hn_tx_lock);
2935 while ((m = buf_ring_dequeue_sc(txr->hn_mbuf_br)) != NULL)
2937 mtx_unlock(&txr->hn_tx_lock);
2943 hn_xmit_txeof(struct hn_tx_ring *txr)
2946 if (txr->hn_sched_tx)
2949 if (mtx_trylock(&txr->hn_tx_lock)) {
2952 txr->hn_oactive = 0;
2953 sched = hn_xmit(txr, txr->hn_direct_tx_size);
2954 mtx_unlock(&txr->hn_tx_lock);
2956 taskqueue_enqueue(txr->hn_tx_taskq,
2962 * Release the oactive earlier, with the hope, that
2963 * others could catch up. The task will clear the
2964 * oactive again with the hn_tx_lock to avoid possible
2967 txr->hn_oactive = 0;
2968 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
2973 hn_xmit_taskfunc(void *xtxr, int pending __unused)
2975 struct hn_tx_ring *txr = xtxr;
2977 mtx_lock(&txr->hn_tx_lock);
2979 mtx_unlock(&txr->hn_tx_lock);
2983 hn_xmit_txeof_taskfunc(void *xtxr, int pending __unused)
2985 struct hn_tx_ring *txr = xtxr;
2987 mtx_lock(&txr->hn_tx_lock);
2988 txr->hn_oactive = 0;
2990 mtx_unlock(&txr->hn_tx_lock);
2994 hn_channel_attach(struct hn_softc *sc, struct hv_vmbus_channel *chan)
2996 struct hn_rx_ring *rxr;
2999 idx = chan->offer_msg.offer.sub_channel_index;
3001 KASSERT(idx >= 0 && idx < sc->hn_rx_ring_inuse,
3002 ("invalid channel index %d, should > 0 && < %d",
3003 idx, sc->hn_rx_ring_inuse));
3004 rxr = &sc->hn_rx_ring[idx];
3005 KASSERT((rxr->hn_rx_flags & HN_RX_FLAG_ATTACHED) == 0,
3006 ("RX ring %d already attached", idx));
3007 rxr->hn_rx_flags |= HN_RX_FLAG_ATTACHED;
3009 chan->hv_chan_rxr = rxr;
3011 if_printf(sc->hn_ifp, "link RX ring %d to channel%u\n",
3012 idx, chan->offer_msg.child_rel_id);
3015 if (idx < sc->hn_tx_ring_inuse) {
3016 struct hn_tx_ring *txr = &sc->hn_tx_ring[idx];
3018 KASSERT((txr->hn_tx_flags & HN_TX_FLAG_ATTACHED) == 0,
3019 ("TX ring %d already attached", idx));
3020 txr->hn_tx_flags |= HN_TX_FLAG_ATTACHED;
3022 chan->hv_chan_txr = txr;
3023 txr->hn_chan = chan;
3025 if_printf(sc->hn_ifp, "link TX ring %d to channel%u\n",
3026 idx, chan->offer_msg.child_rel_id);
3030 /* Bind channel to a proper CPU */
3031 vmbus_channel_cpu_set(chan, (sc->hn_cpu + idx) % mp_ncpus);
3035 hn_subchan_attach(struct hn_softc *sc, struct hv_vmbus_channel *chan)
3038 KASSERT(!HV_VMBUS_CHAN_ISPRIMARY(chan),
3039 ("subchannel callback on primary channel"));
3040 KASSERT(chan->offer_msg.offer.sub_channel_index > 0,
3041 ("invalid channel subidx %u",
3042 chan->offer_msg.offer.sub_channel_index));
3043 hn_channel_attach(sc, chan);
3047 hn_subchan_setup(struct hn_softc *sc)
3049 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
3050 struct hv_vmbus_channel **subchan;
3051 int subchan_cnt = sc->net_dev->num_channel - 1;
3054 /* Wait for sub-channels setup to complete. */
3055 subchan = vmbus_get_subchan(device_ctx->channel, subchan_cnt);
3057 /* Attach the sub-channels. */
3058 for (i = 0; i < subchan_cnt; ++i) {
3059 /* NOTE: Calling order is critical. */
3060 hn_subchan_attach(sc, subchan[i]);
3061 hv_nv_subchan_attach(subchan[i]);
3064 /* Release the sub-channels */
3065 vmbus_rel_subchan(subchan, subchan_cnt);
3066 if_printf(sc->hn_ifp, "%d sub-channels setup done\n", subchan_cnt);
3070 hn_tx_taskq_create(void *arg __unused)
3072 if (!hn_share_tx_taskq)
3075 hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
3076 taskqueue_thread_enqueue, &hn_tx_taskq);
3077 taskqueue_start_threads(&hn_tx_taskq, 1, PI_NET, "hn tx");
3078 if (hn_bind_tx_taskq >= 0) {
3079 int cpu = hn_bind_tx_taskq;
3080 struct task cpuset_task;
3083 if (cpu > mp_ncpus - 1)
3085 CPU_SETOF(cpu, &cpu_set);
3086 TASK_INIT(&cpuset_task, 0, hn_cpuset_setthread_task, &cpu_set);
3087 taskqueue_enqueue(hn_tx_taskq, &cpuset_task);
3088 taskqueue_drain(hn_tx_taskq, &cpuset_task);
3091 SYSINIT(hn_txtq_create, SI_SUB_DRIVERS, SI_ORDER_FIRST,
3092 hn_tx_taskq_create, NULL);
3095 hn_tx_taskq_destroy(void *arg __unused)
3097 if (hn_tx_taskq != NULL)
3098 taskqueue_free(hn_tx_taskq);
3100 SYSUNINIT(hn_txtq_destroy, SI_SUB_DRIVERS, SI_ORDER_FIRST,
3101 hn_tx_taskq_destroy, NULL);
3103 static device_method_t netvsc_methods[] = {
3104 /* Device interface */
3105 DEVMETHOD(device_probe, netvsc_probe),
3106 DEVMETHOD(device_attach, netvsc_attach),
3107 DEVMETHOD(device_detach, netvsc_detach),
3108 DEVMETHOD(device_shutdown, netvsc_shutdown),
3113 static driver_t netvsc_driver = {
3119 static devclass_t netvsc_devclass;
3121 DRIVER_MODULE(hn, vmbus, netvsc_driver, netvsc_devclass, 0, 0);
3122 MODULE_VERSION(hn, 1);
3123 MODULE_DEPEND(hn, vmbus, 1, 1, 1);