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>
69 #include <sys/queue.h>
72 #include <sys/sysctl.h>
73 #include <sys/buf_ring.h>
76 #include <net/if_arp.h>
77 #include <net/ethernet.h>
78 #include <net/if_dl.h>
79 #include <net/if_media.h>
83 #include <net/if_var.h>
84 #include <net/if_types.h>
85 #include <net/if_vlan_var.h>
87 #include <netinet/in_systm.h>
88 #include <netinet/in.h>
89 #include <netinet/ip.h>
90 #include <netinet/if_ether.h>
91 #include <netinet/tcp.h>
92 #include <netinet/udp.h>
93 #include <netinet/ip6.h>
96 #include <vm/vm_param.h>
97 #include <vm/vm_kern.h>
100 #include <machine/bus.h>
101 #include <machine/resource.h>
102 #include <machine/frame.h>
105 #include <sys/rman.h>
106 #include <sys/mutex.h>
107 #include <sys/errno.h>
108 #include <sys/types.h>
109 #include <machine/atomic.h>
111 #include <machine/intr_machdep.h>
113 #include <machine/in_cksum.h>
115 #include <dev/hyperv/include/hyperv.h>
116 #include "hv_net_vsc.h"
117 #include "hv_rndis.h"
118 #include "hv_rndis_filter.h"
120 #define hv_chan_rxr hv_chan_priv1
121 #define hv_chan_txr hv_chan_priv2
123 /* Short for Hyper-V network interface */
124 #define NETVSC_DEVNAME "hn"
127 * It looks like offset 0 of buf is reserved to hold the softc pointer.
128 * The sc pointer evidently not needed, and is not presently populated.
129 * The packet offset is where the netvsc_packet starts in the buffer.
131 #define HV_NV_SC_PTR_OFFSET_IN_BUF 0
132 #define HV_NV_PACKET_OFFSET_IN_BUF 16
134 /* YYY should get it from the underlying channel */
135 #define HN_TX_DESC_CNT 512
137 #define HN_LROENT_CNT_DEF 128
139 #define HN_RING_CNT_DEF_MAX 8
141 #define HN_RNDIS_MSG_LEN \
142 (sizeof(rndis_msg) + \
143 RNDIS_HASH_PPI_SIZE + \
144 RNDIS_VLAN_PPI_SIZE + \
145 RNDIS_TSO_PPI_SIZE + \
147 #define HN_RNDIS_MSG_BOUNDARY PAGE_SIZE
148 #define HN_RNDIS_MSG_ALIGN CACHE_LINE_SIZE
150 #define HN_TX_DATA_BOUNDARY PAGE_SIZE
151 #define HN_TX_DATA_MAXSIZE IP_MAXPACKET
152 #define HN_TX_DATA_SEGSIZE PAGE_SIZE
153 #define HN_TX_DATA_SEGCNT_MAX \
154 (NETVSC_PACKET_MAXPAGE - HV_RF_NUM_TX_RESERVED_PAGE_BUFS)
156 #define HN_DIRECT_TX_SIZE_DEF 128
158 #define HN_EARLY_TXEOF_THRESH 8
161 #ifndef HN_USE_TXDESC_BUFRING
162 SLIST_ENTRY(hn_txdesc) link;
165 struct hn_tx_ring *txr;
167 uint32_t flags; /* HN_TXD_FLAG_ */
168 netvsc_packet netvsc_pkt; /* XXX to be removed */
170 bus_dmamap_t data_dmap;
172 bus_addr_t rndis_msg_paddr;
173 rndis_msg *rndis_msg;
174 bus_dmamap_t rndis_msg_dmap;
177 #define HN_TXD_FLAG_ONLIST 0x1
178 #define HN_TXD_FLAG_DMAMAP 0x2
181 * Only enable UDP checksum offloading when it is on 2012R2 or
182 * later. UDP checksum offloading doesn't work on earlier
185 #define HN_CSUM_ASSIST_WIN8 (CSUM_IP | CSUM_TCP)
186 #define HN_CSUM_ASSIST (CSUM_IP | CSUM_UDP | CSUM_TCP)
188 #define HN_LRO_LENLIM_MULTIRX_DEF (12 * ETHERMTU)
189 #define HN_LRO_LENLIM_DEF (25 * ETHERMTU)
190 /* YYY 2*MTU is a bit rough, but should be good enough. */
191 #define HN_LRO_LENLIM_MIN(ifp) (2 * (ifp)->if_mtu)
193 #define HN_LRO_ACKCNT_DEF 1
196 * Be aware that this sleepable mutex will exhibit WITNESS errors when
197 * certain TCP and ARP code paths are taken. This appears to be a
198 * well-known condition, as all other drivers checked use a sleeping
199 * mutex to protect their transmit paths.
200 * Also Be aware that mutexes do not play well with semaphores, and there
201 * is a conflicting semaphore in a certain channel code path.
203 #define NV_LOCK_INIT(_sc, _name) \
204 mtx_init(&(_sc)->hn_lock, _name, MTX_NETWORK_LOCK, MTX_DEF)
205 #define NV_LOCK(_sc) mtx_lock(&(_sc)->hn_lock)
206 #define NV_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->hn_lock, MA_OWNED)
207 #define NV_UNLOCK(_sc) mtx_unlock(&(_sc)->hn_lock)
208 #define NV_LOCK_DESTROY(_sc) mtx_destroy(&(_sc)->hn_lock)
215 int hv_promisc_mode = 0; /* normal mode by default */
217 SYSCTL_NODE(_hw, OID_AUTO, hn, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
218 "Hyper-V network interface");
220 /* Trust tcp segements verification on host side. */
221 static int hn_trust_hosttcp = 1;
222 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hosttcp, CTLFLAG_RDTUN,
223 &hn_trust_hosttcp, 0,
224 "Trust tcp segement verification on host side, "
225 "when csum info is missing (global setting)");
227 /* Trust udp datagrams verification on host side. */
228 static int hn_trust_hostudp = 1;
229 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostudp, CTLFLAG_RDTUN,
230 &hn_trust_hostudp, 0,
231 "Trust udp datagram verification on host side, "
232 "when csum info is missing (global setting)");
234 /* Trust ip packets verification on host side. */
235 static int hn_trust_hostip = 1;
236 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostip, CTLFLAG_RDTUN,
238 "Trust ip packet verification on host side, "
239 "when csum info is missing (global setting)");
241 #if __FreeBSD_version >= 1100045
242 /* Limit TSO burst size */
243 static int hn_tso_maxlen = 0;
244 SYSCTL_INT(_hw_hn, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN,
245 &hn_tso_maxlen, 0, "TSO burst limit");
248 /* Limit chimney send size */
249 static int hn_tx_chimney_size = 0;
250 SYSCTL_INT(_hw_hn, OID_AUTO, tx_chimney_size, CTLFLAG_RDTUN,
251 &hn_tx_chimney_size, 0, "Chimney send packet size limit");
253 /* Limit the size of packet for direct transmission */
254 static int hn_direct_tx_size = HN_DIRECT_TX_SIZE_DEF;
255 SYSCTL_INT(_hw_hn, OID_AUTO, direct_tx_size, CTLFLAG_RDTUN,
256 &hn_direct_tx_size, 0, "Size of the packet for direct transmission");
258 #if defined(INET) || defined(INET6)
259 #if __FreeBSD_version >= 1100095
260 static int hn_lro_entry_count = HN_LROENT_CNT_DEF;
261 SYSCTL_INT(_hw_hn, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN,
262 &hn_lro_entry_count, 0, "LRO entry count");
266 static int hn_share_tx_taskq = 0;
267 SYSCTL_INT(_hw_hn, OID_AUTO, share_tx_taskq, CTLFLAG_RDTUN,
268 &hn_share_tx_taskq, 0, "Enable shared TX taskqueue");
270 static struct taskqueue *hn_tx_taskq;
272 #ifndef HN_USE_TXDESC_BUFRING
273 static int hn_use_txdesc_bufring = 0;
275 static int hn_use_txdesc_bufring = 1;
277 SYSCTL_INT(_hw_hn, OID_AUTO, use_txdesc_bufring, CTLFLAG_RD,
278 &hn_use_txdesc_bufring, 0, "Use buf_ring for TX descriptors");
280 static int hn_bind_tx_taskq = -1;
281 SYSCTL_INT(_hw_hn, OID_AUTO, bind_tx_taskq, CTLFLAG_RDTUN,
282 &hn_bind_tx_taskq, 0, "Bind TX taskqueue to the specified cpu");
284 static int hn_use_if_start = 0;
285 SYSCTL_INT(_hw_hn, OID_AUTO, use_if_start, CTLFLAG_RDTUN,
286 &hn_use_if_start, 0, "Use if_start TX method");
288 static int hn_chan_cnt = 0;
289 SYSCTL_INT(_hw_hn, OID_AUTO, chan_cnt, CTLFLAG_RDTUN,
291 "# of channels to use; each channel has one RX ring and one TX ring");
293 static int hn_tx_ring_cnt = 0;
294 SYSCTL_INT(_hw_hn, OID_AUTO, tx_ring_cnt, CTLFLAG_RDTUN,
295 &hn_tx_ring_cnt, 0, "# of TX rings to use");
297 static int hn_tx_swq_depth = 0;
298 SYSCTL_INT(_hw_hn, OID_AUTO, tx_swq_depth, CTLFLAG_RDTUN,
299 &hn_tx_swq_depth, 0, "Depth of IFQ or BUFRING");
301 static u_int hn_cpu_index;
304 * Forward declarations
306 static void hn_stop(hn_softc_t *sc);
307 static void hn_ifinit_locked(hn_softc_t *sc);
308 static void hn_ifinit(void *xsc);
309 static int hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data);
310 static int hn_start_locked(struct hn_tx_ring *txr, int len);
311 static void hn_start(struct ifnet *ifp);
312 static void hn_start_txeof(struct hn_tx_ring *);
313 static int hn_ifmedia_upd(struct ifnet *ifp);
314 static void hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr);
315 #if __FreeBSD_version >= 1100099
316 static int hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS);
317 static int hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS);
319 static int hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS);
320 static int hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS);
321 static int hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS);
322 static int hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS);
323 static int hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS);
324 static int hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS);
325 static int hn_check_iplen(const struct mbuf *, int);
326 static int hn_create_tx_ring(struct hn_softc *, int);
327 static void hn_destroy_tx_ring(struct hn_tx_ring *);
328 static int hn_create_tx_data(struct hn_softc *, int);
329 static void hn_destroy_tx_data(struct hn_softc *);
330 static void hn_start_taskfunc(void *, int);
331 static void hn_start_txeof_taskfunc(void *, int);
332 static void hn_stop_tx_tasks(struct hn_softc *);
333 static int hn_encap(struct hn_tx_ring *, struct hn_txdesc *, struct mbuf **);
334 static void hn_create_rx_data(struct hn_softc *sc, int);
335 static void hn_destroy_rx_data(struct hn_softc *sc);
336 static void hn_set_tx_chimney_size(struct hn_softc *, int);
337 static void hn_channel_attach(struct hn_softc *, struct hv_vmbus_channel *);
338 static void hn_subchan_attach(struct hn_softc *, struct hv_vmbus_channel *);
340 static int hn_transmit(struct ifnet *, struct mbuf *);
341 static void hn_xmit_qflush(struct ifnet *);
342 static int hn_xmit(struct hn_tx_ring *, int);
343 static void hn_xmit_txeof(struct hn_tx_ring *);
344 static void hn_xmit_taskfunc(void *, int);
345 static void hn_xmit_txeof_taskfunc(void *, int);
347 #if __FreeBSD_version >= 1100099
349 hn_set_lro_lenlim(struct hn_softc *sc, int lenlim)
353 for (i = 0; i < sc->hn_rx_ring_inuse; ++i)
354 sc->hn_rx_ring[i].hn_lro.lro_length_lim = lenlim;
359 hn_get_txswq_depth(const struct hn_tx_ring *txr)
362 KASSERT(txr->hn_txdesc_cnt > 0, ("tx ring is not setup yet"));
363 if (hn_tx_swq_depth < txr->hn_txdesc_cnt)
364 return txr->hn_txdesc_cnt;
365 return hn_tx_swq_depth;
369 hn_ifmedia_upd(struct ifnet *ifp __unused)
376 hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
378 struct hn_softc *sc = ifp->if_softc;
380 ifmr->ifm_status = IFM_AVALID;
381 ifmr->ifm_active = IFM_ETHER;
383 if (!sc->hn_carrier) {
384 ifmr->ifm_active |= IFM_NONE;
387 ifmr->ifm_status |= IFM_ACTIVE;
388 ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
391 /* {F8615163-DF3E-46c5-913F-F2D2F965ED0E} */
392 static const hv_guid g_net_vsc_device_type = {
393 .data = {0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
394 0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E}
398 * Standard probe entry point.
402 netvsc_probe(device_t dev)
406 p = vmbus_get_type(dev);
407 if (!memcmp(p, &g_net_vsc_device_type.data, sizeof(hv_guid))) {
408 device_set_desc(dev, "Hyper-V Network Interface");
410 printf("Netvsc probe... DONE \n");
412 return (BUS_PROBE_DEFAULT);
419 * Standard attach entry point.
421 * Called when the driver is loaded. It allocates needed resources,
422 * and initializes the "hardware" and software.
425 netvsc_attach(device_t dev)
427 struct hv_device *device_ctx = vmbus_get_devctx(dev);
428 struct hv_vmbus_channel *pri_chan;
429 netvsc_device_info device_info;
431 int unit = device_get_unit(dev);
432 struct ifnet *ifp = NULL;
433 int error, ring_cnt, tx_ring_cnt;
434 #if __FreeBSD_version >= 1100045
438 sc = device_get_softc(dev);
443 if (hn_tx_taskq == NULL) {
444 sc->hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
445 taskqueue_thread_enqueue, &sc->hn_tx_taskq);
446 if (hn_bind_tx_taskq >= 0) {
447 int cpu = hn_bind_tx_taskq;
450 if (cpu > mp_ncpus - 1)
452 CPU_SETOF(cpu, &cpu_set);
453 taskqueue_start_threads_cpuset(&sc->hn_tx_taskq, 1,
454 PI_NET, &cpu_set, "%s tx",
455 device_get_nameunit(dev));
457 taskqueue_start_threads(&sc->hn_tx_taskq, 1, PI_NET,
458 "%s tx", device_get_nameunit(dev));
461 sc->hn_tx_taskq = hn_tx_taskq;
463 NV_LOCK_INIT(sc, "NetVSCLock");
465 sc->hn_dev_obj = device_ctx;
467 ifp = sc->hn_ifp = if_alloc(IFT_ETHER);
469 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
472 * Figure out the # of RX rings (ring_cnt) and the # of TX rings
473 * to use (tx_ring_cnt).
476 * The # of RX rings to use is same as the # of channels to use.
478 ring_cnt = hn_chan_cnt;
482 if (ring_cnt > HN_RING_CNT_DEF_MAX)
483 ring_cnt = HN_RING_CNT_DEF_MAX;
484 } else if (ring_cnt > mp_ncpus) {
488 tx_ring_cnt = hn_tx_ring_cnt;
489 if (tx_ring_cnt <= 0 || tx_ring_cnt > ring_cnt)
490 tx_ring_cnt = ring_cnt;
491 if (hn_use_if_start) {
492 /* ifnet.if_start only needs one TX ring. */
497 * Set the leader CPU for channels.
499 sc->hn_cpu = atomic_fetchadd_int(&hn_cpu_index, ring_cnt) % mp_ncpus;
501 error = hn_create_tx_data(sc, tx_ring_cnt);
504 hn_create_rx_data(sc, ring_cnt);
507 * Associate the first TX/RX ring w/ the primary channel.
509 pri_chan = device_ctx->channel;
510 KASSERT(HV_VMBUS_CHAN_ISPRIMARY(pri_chan), ("not primary channel"));
511 KASSERT(pri_chan->offer_msg.offer.sub_channel_index == 0,
512 ("primary channel subidx %u",
513 pri_chan->offer_msg.offer.sub_channel_index));
514 hn_channel_attach(sc, pri_chan);
516 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
517 ifp->if_ioctl = hn_ioctl;
518 ifp->if_init = hn_ifinit;
519 /* needed by hv_rf_on_device_add() code */
520 ifp->if_mtu = ETHERMTU;
521 if (hn_use_if_start) {
522 int qdepth = hn_get_txswq_depth(&sc->hn_tx_ring[0]);
524 ifp->if_start = hn_start;
525 IFQ_SET_MAXLEN(&ifp->if_snd, qdepth);
526 ifp->if_snd.ifq_drv_maxlen = qdepth - 1;
527 IFQ_SET_READY(&ifp->if_snd);
529 ifp->if_transmit = hn_transmit;
530 ifp->if_qflush = hn_xmit_qflush;
533 ifmedia_init(&sc->hn_media, 0, hn_ifmedia_upd, hn_ifmedia_sts);
534 ifmedia_add(&sc->hn_media, IFM_ETHER | IFM_AUTO, 0, NULL);
535 ifmedia_set(&sc->hn_media, IFM_ETHER | IFM_AUTO);
536 /* XXX ifmedia_set really should do this for us */
537 sc->hn_media.ifm_media = sc->hn_media.ifm_cur->ifm_media;
540 * Tell upper layers that we support full VLAN capability.
542 ifp->if_hdrlen = sizeof(struct ether_vlan_header);
543 ifp->if_capabilities |=
544 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
547 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
549 ifp->if_hwassist = sc->hn_tx_ring[0].hn_csum_assist | CSUM_TSO;
551 error = hv_rf_on_device_add(device_ctx, &device_info, ring_cnt);
554 KASSERT(sc->net_dev->num_channel > 0 &&
555 sc->net_dev->num_channel <= sc->hn_rx_ring_inuse,
556 ("invalid channel count %u, should be less than %d",
557 sc->net_dev->num_channel, sc->hn_rx_ring_inuse));
560 * Set the # of TX/RX rings that could be used according to
561 * the # of channels that host offered.
563 if (sc->hn_tx_ring_inuse > sc->net_dev->num_channel)
564 sc->hn_tx_ring_inuse = sc->net_dev->num_channel;
565 sc->hn_rx_ring_inuse = sc->net_dev->num_channel;
566 device_printf(dev, "%d TX ring, %d RX ring\n",
567 sc->hn_tx_ring_inuse, sc->hn_rx_ring_inuse);
569 if (sc->net_dev->num_channel > 1) {
570 struct hv_vmbus_channel **subchan;
571 int subchan_cnt = sc->net_dev->num_channel - 1;
574 /* Wait for sub-channels setup to complete. */
575 subchan = vmbus_get_subchan(pri_chan, subchan_cnt);
577 /* Attach the sub-channels. */
578 for (i = 0; i < subchan_cnt; ++i) {
579 /* NOTE: Calling order is critical. */
580 hn_subchan_attach(sc, subchan[i]);
581 hv_nv_subchan_attach(subchan[i]);
584 /* Release the sub-channels */
585 vmbus_rel_subchan(subchan, subchan_cnt);
586 device_printf(dev, "%d sub-channels setup done\n", subchan_cnt);
589 #if __FreeBSD_version >= 1100099
590 if (sc->hn_rx_ring_inuse > 1) {
592 * Reduce TCP segment aggregation limit for multiple
593 * RX rings to increase ACK timeliness.
595 hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MULTIRX_DEF);
599 if (device_info.link_state == 0) {
603 #if __FreeBSD_version >= 1100045
604 tso_maxlen = hn_tso_maxlen;
605 if (tso_maxlen <= 0 || tso_maxlen > IP_MAXPACKET)
606 tso_maxlen = IP_MAXPACKET;
608 ifp->if_hw_tsomaxsegcount = HN_TX_DATA_SEGCNT_MAX;
609 ifp->if_hw_tsomaxsegsize = PAGE_SIZE;
610 ifp->if_hw_tsomax = tso_maxlen -
611 (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
614 ether_ifattach(ifp, device_info.mac_addr);
616 #if __FreeBSD_version >= 1100045
617 if_printf(ifp, "TSO: %u/%u/%u\n", ifp->if_hw_tsomax,
618 ifp->if_hw_tsomaxsegcount, ifp->if_hw_tsomaxsegsize);
621 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
622 hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max);
623 if (hn_tx_chimney_size > 0 &&
624 hn_tx_chimney_size < sc->hn_tx_chimney_max)
625 hn_set_tx_chimney_size(sc, hn_tx_chimney_size);
629 hn_destroy_tx_data(sc);
636 * Standard detach entry point
639 netvsc_detach(device_t dev)
641 struct hn_softc *sc = device_get_softc(dev);
642 struct hv_device *hv_device = vmbus_get_devctx(dev);
645 printf("netvsc_detach\n");
648 * XXXKYS: Need to clean up all our
649 * driver state; this is the driver
654 * XXXKYS: Need to stop outgoing traffic and unregister
658 hv_rf_on_device_remove(hv_device, HV_RF_NV_DESTROY_CHANNEL);
660 hn_stop_tx_tasks(sc);
662 ifmedia_removeall(&sc->hn_media);
663 hn_destroy_rx_data(sc);
664 hn_destroy_tx_data(sc);
666 if (sc->hn_tx_taskq != hn_tx_taskq)
667 taskqueue_free(sc->hn_tx_taskq);
673 * Standard shutdown entry point
676 netvsc_shutdown(device_t dev)
682 hn_txdesc_dmamap_load(struct hn_tx_ring *txr, struct hn_txdesc *txd,
683 struct mbuf **m_head, bus_dma_segment_t *segs, int *nsegs)
685 struct mbuf *m = *m_head;
688 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag, txd->data_dmap,
689 m, segs, nsegs, BUS_DMA_NOWAIT);
690 if (error == EFBIG) {
693 m_new = m_collapse(m, M_NOWAIT, HN_TX_DATA_SEGCNT_MAX);
698 txr->hn_tx_collapsed++;
700 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag,
701 txd->data_dmap, m, segs, nsegs, BUS_DMA_NOWAIT);
704 bus_dmamap_sync(txr->hn_tx_data_dtag, txd->data_dmap,
705 BUS_DMASYNC_PREWRITE);
706 txd->flags |= HN_TXD_FLAG_DMAMAP;
712 hn_txdesc_dmamap_unload(struct hn_tx_ring *txr, struct hn_txdesc *txd)
715 if (txd->flags & HN_TXD_FLAG_DMAMAP) {
716 bus_dmamap_sync(txr->hn_tx_data_dtag,
717 txd->data_dmap, BUS_DMASYNC_POSTWRITE);
718 bus_dmamap_unload(txr->hn_tx_data_dtag,
720 txd->flags &= ~HN_TXD_FLAG_DMAMAP;
725 hn_txdesc_put(struct hn_tx_ring *txr, struct hn_txdesc *txd)
728 KASSERT((txd->flags & HN_TXD_FLAG_ONLIST) == 0,
729 ("put an onlist txd %#x", txd->flags));
731 KASSERT(txd->refs > 0, ("invalid txd refs %d", txd->refs));
732 if (atomic_fetchadd_int(&txd->refs, -1) != 1)
735 hn_txdesc_dmamap_unload(txr, txd);
736 if (txd->m != NULL) {
741 txd->flags |= HN_TXD_FLAG_ONLIST;
743 #ifndef HN_USE_TXDESC_BUFRING
744 mtx_lock_spin(&txr->hn_txlist_spin);
745 KASSERT(txr->hn_txdesc_avail >= 0 &&
746 txr->hn_txdesc_avail < txr->hn_txdesc_cnt,
747 ("txdesc_put: invalid txd avail %d", txr->hn_txdesc_avail));
748 txr->hn_txdesc_avail++;
749 SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
750 mtx_unlock_spin(&txr->hn_txlist_spin);
752 atomic_add_int(&txr->hn_txdesc_avail, 1);
753 buf_ring_enqueue(txr->hn_txdesc_br, txd);
759 static __inline struct hn_txdesc *
760 hn_txdesc_get(struct hn_tx_ring *txr)
762 struct hn_txdesc *txd;
764 #ifndef HN_USE_TXDESC_BUFRING
765 mtx_lock_spin(&txr->hn_txlist_spin);
766 txd = SLIST_FIRST(&txr->hn_txlist);
768 KASSERT(txr->hn_txdesc_avail > 0,
769 ("txdesc_get: invalid txd avail %d", txr->hn_txdesc_avail));
770 txr->hn_txdesc_avail--;
771 SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
773 mtx_unlock_spin(&txr->hn_txlist_spin);
775 txd = buf_ring_dequeue_sc(txr->hn_txdesc_br);
779 #ifdef HN_USE_TXDESC_BUFRING
780 atomic_subtract_int(&txr->hn_txdesc_avail, 1);
782 KASSERT(txd->m == NULL && txd->refs == 0 &&
783 (txd->flags & HN_TXD_FLAG_ONLIST), ("invalid txd"));
784 txd->flags &= ~HN_TXD_FLAG_ONLIST;
791 hn_txdesc_hold(struct hn_txdesc *txd)
794 /* 0->1 transition will never work */
795 KASSERT(txd->refs > 0, ("invalid refs %d", txd->refs));
796 atomic_add_int(&txd->refs, 1);
800 hn_txeof(struct hn_tx_ring *txr)
802 txr->hn_has_txeof = 0;
807 hn_tx_done(struct hv_vmbus_channel *chan, void *xpkt)
809 netvsc_packet *packet = xpkt;
810 struct hn_txdesc *txd;
811 struct hn_tx_ring *txr;
813 txd = (struct hn_txdesc *)(uintptr_t)
814 packet->compl.send.send_completion_tid;
817 KASSERT(txr->hn_chan == chan,
818 ("channel mismatch, on channel%u, should be channel%u",
819 chan->offer_msg.offer.sub_channel_index,
820 txr->hn_chan->offer_msg.offer.sub_channel_index));
822 txr->hn_has_txeof = 1;
823 hn_txdesc_put(txr, txd);
825 ++txr->hn_txdone_cnt;
826 if (txr->hn_txdone_cnt >= HN_EARLY_TXEOF_THRESH) {
827 txr->hn_txdone_cnt = 0;
834 netvsc_channel_rollup(struct hv_vmbus_channel *chan)
836 struct hn_tx_ring *txr = chan->hv_chan_txr;
837 #if defined(INET) || defined(INET6)
838 struct hn_rx_ring *rxr = chan->hv_chan_rxr;
840 tcp_lro_flush_all(&rxr->hn_lro);
845 * 'txr' could be NULL, if multiple channels and
846 * ifnet.if_start method are enabled.
848 if (txr == NULL || !txr->hn_has_txeof)
851 txr->hn_txdone_cnt = 0;
857 * If this function fails, then both txd and m_head0 will be freed.
860 hn_encap(struct hn_tx_ring *txr, struct hn_txdesc *txd, struct mbuf **m_head0)
862 bus_dma_segment_t segs[HN_TX_DATA_SEGCNT_MAX];
864 struct mbuf *m_head = *m_head0;
865 netvsc_packet *packet;
866 rndis_msg *rndis_mesg;
867 rndis_packet *rndis_pkt;
868 rndis_per_packet_info *rppi;
869 struct ndis_hash_info *hash_info;
870 uint32_t rndis_msg_size;
872 packet = &txd->netvsc_pkt;
873 packet->is_data_pkt = TRUE;
874 packet->tot_data_buf_len = m_head->m_pkthdr.len;
877 * extension points to the area reserved for the
878 * rndis_filter_packet, which is placed just after
879 * the netvsc_packet (and rppi struct, if present;
880 * length is updated later).
882 rndis_mesg = txd->rndis_msg;
883 /* XXX not necessary */
884 memset(rndis_mesg, 0, HN_RNDIS_MSG_LEN);
885 rndis_mesg->ndis_msg_type = REMOTE_NDIS_PACKET_MSG;
887 rndis_pkt = &rndis_mesg->msg.packet;
888 rndis_pkt->data_offset = sizeof(rndis_packet);
889 rndis_pkt->data_length = packet->tot_data_buf_len;
890 rndis_pkt->per_pkt_info_offset = sizeof(rndis_packet);
892 rndis_msg_size = RNDIS_MESSAGE_SIZE(rndis_packet);
895 * Set the hash info for this packet, so that the host could
896 * dispatch the TX done event for this packet back to this TX
899 rndis_msg_size += RNDIS_HASH_PPI_SIZE;
900 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_HASH_PPI_SIZE,
902 hash_info = (struct ndis_hash_info *)((uint8_t *)rppi +
903 rppi->per_packet_info_offset);
904 hash_info->hash = txr->hn_tx_idx;
906 if (m_head->m_flags & M_VLANTAG) {
907 ndis_8021q_info *rppi_vlan_info;
909 rndis_msg_size += RNDIS_VLAN_PPI_SIZE;
910 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_VLAN_PPI_SIZE,
913 rppi_vlan_info = (ndis_8021q_info *)((uint8_t *)rppi +
914 rppi->per_packet_info_offset);
915 rppi_vlan_info->u1.s1.vlan_id =
916 m_head->m_pkthdr.ether_vtag & 0xfff;
919 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
920 rndis_tcp_tso_info *tso_info;
921 struct ether_vlan_header *eh;
925 * XXX need m_pullup and use mtodo
927 eh = mtod(m_head, struct ether_vlan_header*);
928 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN))
929 ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
931 ether_len = ETHER_HDR_LEN;
933 rndis_msg_size += RNDIS_TSO_PPI_SIZE;
934 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_TSO_PPI_SIZE,
935 tcp_large_send_info);
937 tso_info = (rndis_tcp_tso_info *)((uint8_t *)rppi +
938 rppi->per_packet_info_offset);
939 tso_info->lso_v2_xmit.type =
940 RNDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
943 if (m_head->m_pkthdr.csum_flags & CSUM_IP_TSO) {
945 (struct ip *)(m_head->m_data + ether_len);
946 unsigned long iph_len = ip->ip_hl << 2;
948 (struct tcphdr *)((caddr_t)ip + iph_len);
950 tso_info->lso_v2_xmit.ip_version =
951 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
955 th->th_sum = in_pseudo(ip->ip_src.s_addr,
956 ip->ip_dst.s_addr, htons(IPPROTO_TCP));
959 #if defined(INET6) && defined(INET)
964 struct ip6_hdr *ip6 = (struct ip6_hdr *)
965 (m_head->m_data + ether_len);
966 struct tcphdr *th = (struct tcphdr *)(ip6 + 1);
968 tso_info->lso_v2_xmit.ip_version =
969 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
971 th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
974 tso_info->lso_v2_xmit.tcp_header_offset = 0;
975 tso_info->lso_v2_xmit.mss = m_head->m_pkthdr.tso_segsz;
976 } else if (m_head->m_pkthdr.csum_flags & txr->hn_csum_assist) {
977 rndis_tcp_ip_csum_info *csum_info;
979 rndis_msg_size += RNDIS_CSUM_PPI_SIZE;
980 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_CSUM_PPI_SIZE,
982 csum_info = (rndis_tcp_ip_csum_info *)((uint8_t *)rppi +
983 rppi->per_packet_info_offset);
985 csum_info->xmit.is_ipv4 = 1;
986 if (m_head->m_pkthdr.csum_flags & CSUM_IP)
987 csum_info->xmit.ip_header_csum = 1;
989 if (m_head->m_pkthdr.csum_flags & CSUM_TCP) {
990 csum_info->xmit.tcp_csum = 1;
991 csum_info->xmit.tcp_header_offset = 0;
992 } else if (m_head->m_pkthdr.csum_flags & CSUM_UDP) {
993 csum_info->xmit.udp_csum = 1;
997 rndis_mesg->msg_len = packet->tot_data_buf_len + rndis_msg_size;
998 packet->tot_data_buf_len = rndis_mesg->msg_len;
1001 * Chimney send, if the packet could fit into one chimney buffer.
1003 if (packet->tot_data_buf_len < txr->hn_tx_chimney_size) {
1004 netvsc_dev *net_dev = txr->hn_sc->net_dev;
1005 uint32_t send_buf_section_idx;
1007 txr->hn_tx_chimney_tried++;
1008 send_buf_section_idx =
1009 hv_nv_get_next_send_section(net_dev);
1010 if (send_buf_section_idx !=
1011 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX) {
1012 uint8_t *dest = ((uint8_t *)net_dev->send_buf +
1013 (send_buf_section_idx *
1014 net_dev->send_section_size));
1016 memcpy(dest, rndis_mesg, rndis_msg_size);
1017 dest += rndis_msg_size;
1018 m_copydata(m_head, 0, m_head->m_pkthdr.len, dest);
1020 packet->send_buf_section_idx = send_buf_section_idx;
1021 packet->send_buf_section_size =
1022 packet->tot_data_buf_len;
1023 packet->page_buf_count = 0;
1024 txr->hn_tx_chimney++;
1029 error = hn_txdesc_dmamap_load(txr, txd, &m_head, segs, &nsegs);
1034 * This mbuf is not linked w/ the txd yet, so free it now.
1039 freed = hn_txdesc_put(txr, txd);
1041 ("fail to free txd upon txdma error"));
1043 txr->hn_txdma_failed++;
1044 if_inc_counter(txr->hn_sc->hn_ifp, IFCOUNTER_OERRORS, 1);
1049 packet->page_buf_count = nsegs + HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
1051 /* send packet with page buffer */
1052 packet->page_buffers[0].pfn = atop(txd->rndis_msg_paddr);
1053 packet->page_buffers[0].offset = txd->rndis_msg_paddr & PAGE_MASK;
1054 packet->page_buffers[0].length = rndis_msg_size;
1057 * Fill the page buffers with mbuf info starting at index
1058 * HV_RF_NUM_TX_RESERVED_PAGE_BUFS.
1060 for (i = 0; i < nsegs; ++i) {
1061 hv_vmbus_page_buffer *pb = &packet->page_buffers[
1062 i + HV_RF_NUM_TX_RESERVED_PAGE_BUFS];
1064 pb->pfn = atop(segs[i].ds_addr);
1065 pb->offset = segs[i].ds_addr & PAGE_MASK;
1066 pb->length = segs[i].ds_len;
1069 packet->send_buf_section_idx =
1070 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX;
1071 packet->send_buf_section_size = 0;
1075 /* Set the completion routine */
1076 packet->compl.send.on_send_completion = hn_tx_done;
1077 packet->compl.send.send_completion_context = packet;
1078 packet->compl.send.send_completion_tid = (uint64_t)(uintptr_t)txd;
1085 * If this function fails, then txd will be freed, but the mbuf
1086 * associated w/ the txd will _not_ be freed.
1089 hn_send_pkt(struct ifnet *ifp, struct hn_tx_ring *txr, struct hn_txdesc *txd)
1091 int error, send_failed = 0;
1095 * Make sure that txd is not freed before ETHER_BPF_MTAP.
1097 hn_txdesc_hold(txd);
1098 error = hv_nv_on_send(txr->hn_chan, &txd->netvsc_pkt);
1100 ETHER_BPF_MTAP(ifp, txd->m);
1101 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1102 if (!hn_use_if_start) {
1103 if_inc_counter(ifp, IFCOUNTER_OBYTES,
1104 txd->m->m_pkthdr.len);
1105 if (txd->m->m_flags & M_MCAST)
1106 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
1110 hn_txdesc_put(txr, txd);
1112 if (__predict_false(error)) {
1116 * This should "really rarely" happen.
1118 * XXX Too many RX to be acked or too many sideband
1119 * commands to run? Ask netvsc_channel_rollup()
1120 * to kick start later.
1122 txr->hn_has_txeof = 1;
1124 txr->hn_send_failed++;
1127 * Try sending again after set hn_has_txeof;
1128 * in case that we missed the last
1129 * netvsc_channel_rollup().
1133 if_printf(ifp, "send failed\n");
1136 * Caller will perform further processing on the
1137 * associated mbuf, so don't free it in hn_txdesc_put();
1138 * only unload it from the DMA map in hn_txdesc_put(),
1142 freed = hn_txdesc_put(txr, txd);
1144 ("fail to free txd upon send error"));
1146 txr->hn_send_failed++;
1152 * Start a transmit of one or more packets
1155 hn_start_locked(struct hn_tx_ring *txr, int len)
1157 struct hn_softc *sc = txr->hn_sc;
1158 struct ifnet *ifp = sc->hn_ifp;
1160 KASSERT(hn_use_if_start,
1161 ("hn_start_locked is called, when if_start is disabled"));
1162 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1163 mtx_assert(&txr->hn_tx_lock, MA_OWNED);
1165 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1169 while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
1170 struct hn_txdesc *txd;
1171 struct mbuf *m_head;
1174 IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
1178 if (len > 0 && m_head->m_pkthdr.len > len) {
1180 * This sending could be time consuming; let callers
1181 * dispatch this packet sending (and sending of any
1182 * following up packets) to tx taskqueue.
1184 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1188 txd = hn_txdesc_get(txr);
1190 txr->hn_no_txdescs++;
1191 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1192 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1196 error = hn_encap(txr, txd, &m_head);
1198 /* Both txd and m_head are freed */
1202 error = hn_send_pkt(ifp, txr, txd);
1203 if (__predict_false(error)) {
1204 /* txd is freed, but m_head is not */
1205 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1206 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1214 * Link up/down notification
1217 netvsc_linkstatus_callback(struct hv_device *device_obj, uint32_t status)
1219 hn_softc_t *sc = device_get_softc(device_obj->device);
1229 * Append the specified data to the indicated mbuf chain,
1230 * Extend the mbuf chain if the new data does not fit in
1233 * This is a minor rewrite of m_append() from sys/kern/uipc_mbuf.c.
1234 * There should be an equivalent in the kernel mbuf code,
1235 * but there does not appear to be one yet.
1237 * Differs from m_append() in that additional mbufs are
1238 * allocated with cluster size MJUMPAGESIZE, and filled
1241 * Return 1 if able to complete the job; otherwise 0.
1244 hv_m_append(struct mbuf *m0, int len, c_caddr_t cp)
1247 int remainder, space;
1249 for (m = m0; m->m_next != NULL; m = m->m_next)
1252 space = M_TRAILINGSPACE(m);
1255 * Copy into available space.
1257 if (space > remainder)
1259 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1264 while (remainder > 0) {
1266 * Allocate a new mbuf; could check space
1267 * and allocate a cluster instead.
1269 n = m_getjcl(M_NOWAIT, m->m_type, 0, MJUMPAGESIZE);
1272 n->m_len = min(MJUMPAGESIZE, remainder);
1273 bcopy(cp, mtod(n, caddr_t), n->m_len);
1275 remainder -= n->m_len;
1279 if (m0->m_flags & M_PKTHDR)
1280 m0->m_pkthdr.len += len - remainder;
1282 return (remainder == 0);
1287 * Called when we receive a data packet from the "wire" on the
1290 * Note: This is no longer used as a callback
1293 netvsc_recv(struct hv_vmbus_channel *chan, netvsc_packet *packet,
1294 rndis_tcp_ip_csum_info *csum_info)
1296 struct hn_rx_ring *rxr = chan->hv_chan_rxr;
1297 struct ifnet *ifp = rxr->hn_ifp;
1299 int size, do_lro = 0, do_csum = 1;
1301 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1305 * Bail out if packet contains more data than configured MTU.
1307 if (packet->tot_data_buf_len > (ifp->if_mtu + ETHER_HDR_LEN)) {
1309 } else if (packet->tot_data_buf_len <= MHLEN) {
1310 m_new = m_gethdr(M_NOWAIT, MT_DATA);
1311 if (m_new == NULL) {
1312 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1315 memcpy(mtod(m_new, void *), packet->data,
1316 packet->tot_data_buf_len);
1317 m_new->m_pkthdr.len = m_new->m_len = packet->tot_data_buf_len;
1318 rxr->hn_small_pkts++;
1321 * Get an mbuf with a cluster. For packets 2K or less,
1322 * get a standard 2K cluster. For anything larger, get a
1323 * 4K cluster. Any buffers larger than 4K can cause problems
1324 * if looped around to the Hyper-V TX channel, so avoid them.
1327 if (packet->tot_data_buf_len > MCLBYTES) {
1329 size = MJUMPAGESIZE;
1332 m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, size);
1333 if (m_new == NULL) {
1334 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1338 hv_m_append(m_new, packet->tot_data_buf_len, packet->data);
1340 m_new->m_pkthdr.rcvif = ifp;
1342 if (__predict_false((ifp->if_capenable & IFCAP_RXCSUM) == 0))
1345 /* receive side checksum offload */
1346 if (csum_info != NULL) {
1347 /* IP csum offload */
1348 if (csum_info->receive.ip_csum_succeeded && do_csum) {
1349 m_new->m_pkthdr.csum_flags |=
1350 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1354 /* TCP/UDP csum offload */
1355 if ((csum_info->receive.tcp_csum_succeeded ||
1356 csum_info->receive.udp_csum_succeeded) && do_csum) {
1357 m_new->m_pkthdr.csum_flags |=
1358 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1359 m_new->m_pkthdr.csum_data = 0xffff;
1360 if (csum_info->receive.tcp_csum_succeeded)
1366 if (csum_info->receive.ip_csum_succeeded &&
1367 csum_info->receive.tcp_csum_succeeded)
1370 const struct ether_header *eh;
1375 if (m_new->m_len < hoff)
1377 eh = mtod(m_new, struct ether_header *);
1378 etype = ntohs(eh->ether_type);
1379 if (etype == ETHERTYPE_VLAN) {
1380 const struct ether_vlan_header *evl;
1382 hoff = sizeof(*evl);
1383 if (m_new->m_len < hoff)
1385 evl = mtod(m_new, struct ether_vlan_header *);
1386 etype = ntohs(evl->evl_proto);
1389 if (etype == ETHERTYPE_IP) {
1392 pr = hn_check_iplen(m_new, hoff);
1393 if (pr == IPPROTO_TCP) {
1395 (rxr->hn_trust_hcsum &
1396 HN_TRUST_HCSUM_TCP)) {
1397 rxr->hn_csum_trusted++;
1398 m_new->m_pkthdr.csum_flags |=
1399 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1400 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1401 m_new->m_pkthdr.csum_data = 0xffff;
1403 /* Rely on SW csum verification though... */
1405 } else if (pr == IPPROTO_UDP) {
1407 (rxr->hn_trust_hcsum &
1408 HN_TRUST_HCSUM_UDP)) {
1409 rxr->hn_csum_trusted++;
1410 m_new->m_pkthdr.csum_flags |=
1411 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1412 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1413 m_new->m_pkthdr.csum_data = 0xffff;
1415 } else if (pr != IPPROTO_DONE && do_csum &&
1416 (rxr->hn_trust_hcsum & HN_TRUST_HCSUM_IP)) {
1417 rxr->hn_csum_trusted++;
1418 m_new->m_pkthdr.csum_flags |=
1419 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1424 if ((packet->vlan_tci != 0) &&
1425 (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0) {
1426 m_new->m_pkthdr.ether_vtag = packet->vlan_tci;
1427 m_new->m_flags |= M_VLANTAG;
1430 m_new->m_pkthdr.flowid = rxr->hn_rx_idx;
1431 M_HASHTYPE_SET(m_new, M_HASHTYPE_OPAQUE);
1434 * Note: Moved RX completion back to hv_nv_on_receive() so all
1435 * messages (not just data messages) will trigger a response.
1438 if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1441 if ((ifp->if_capenable & IFCAP_LRO) && do_lro) {
1442 #if defined(INET) || defined(INET6)
1443 struct lro_ctrl *lro = &rxr->hn_lro;
1446 rxr->hn_lro_tried++;
1447 if (tcp_lro_rx(lro, m_new, 0) == 0) {
1455 /* We're not holding the lock here, so don't release it */
1456 (*ifp->if_input)(ifp, m_new);
1462 * Rules for using sc->temp_unusable:
1463 * 1. sc->temp_unusable can only be read or written while holding NV_LOCK()
1464 * 2. code reading sc->temp_unusable under NV_LOCK(), and finding
1465 * sc->temp_unusable set, must release NV_LOCK() and exit
1466 * 3. to retain exclusive control of the interface,
1467 * sc->temp_unusable must be set by code before releasing NV_LOCK()
1468 * 4. only code setting sc->temp_unusable can clear sc->temp_unusable
1469 * 5. code setting sc->temp_unusable must eventually clear sc->temp_unusable
1473 * Standard ioctl entry point. Called when the user wants to configure
1477 hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1479 hn_softc_t *sc = ifp->if_softc;
1480 struct ifreq *ifr = (struct ifreq *)data;
1482 struct ifaddr *ifa = (struct ifaddr *)data;
1484 netvsc_device_info device_info;
1485 struct hv_device *hn_dev;
1486 int mask, error = 0;
1487 int retry_cnt = 500;
1493 if (ifa->ifa_addr->sa_family == AF_INET) {
1494 ifp->if_flags |= IFF_UP;
1495 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1497 arp_ifinit(ifp, ifa);
1500 error = ether_ioctl(ifp, cmd, data);
1503 hn_dev = vmbus_get_devctx(sc->hn_dev);
1505 /* Check MTU value change */
1506 if (ifp->if_mtu == ifr->ifr_mtu)
1509 if (ifr->ifr_mtu > NETVSC_MAX_CONFIGURABLE_MTU) {
1514 /* Obtain and record requested MTU */
1515 ifp->if_mtu = ifr->ifr_mtu;
1517 #if __FreeBSD_version >= 1100099
1519 * Make sure that LRO aggregation length limit is still
1520 * valid, after the MTU change.
1523 if (sc->hn_rx_ring[0].hn_lro.lro_length_lim <
1524 HN_LRO_LENLIM_MIN(ifp))
1525 hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MIN(ifp));
1531 if (!sc->temp_unusable) {
1532 sc->temp_unusable = TRUE;
1536 if (retry_cnt > 0) {
1540 } while (retry_cnt > 0);
1542 if (retry_cnt == 0) {
1547 /* We must remove and add back the device to cause the new
1548 * MTU to take effect. This includes tearing down, but not
1549 * deleting the channel, then bringing it back up.
1551 error = hv_rf_on_device_remove(hn_dev, HV_RF_NV_RETAIN_CHANNEL);
1554 sc->temp_unusable = FALSE;
1558 error = hv_rf_on_device_add(hn_dev, &device_info,
1559 sc->hn_rx_ring_inuse);
1562 sc->temp_unusable = FALSE;
1567 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
1568 if (sc->hn_tx_ring[0].hn_tx_chimney_size >
1569 sc->hn_tx_chimney_max)
1570 hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max);
1572 hn_ifinit_locked(sc);
1575 sc->temp_unusable = FALSE;
1581 if (!sc->temp_unusable) {
1582 sc->temp_unusable = TRUE;
1586 if (retry_cnt > 0) {
1590 } while (retry_cnt > 0);
1592 if (retry_cnt == 0) {
1597 if (ifp->if_flags & IFF_UP) {
1599 * If only the state of the PROMISC flag changed,
1600 * then just use the 'set promisc mode' command
1601 * instead of reinitializing the entire NIC. Doing
1602 * a full re-init means reloading the firmware and
1603 * waiting for it to start up, which may take a
1607 /* Fixme: Promiscuous mode? */
1608 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1609 ifp->if_flags & IFF_PROMISC &&
1610 !(sc->hn_if_flags & IFF_PROMISC)) {
1611 /* do something here for Hyper-V */
1612 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1613 !(ifp->if_flags & IFF_PROMISC) &&
1614 sc->hn_if_flags & IFF_PROMISC) {
1615 /* do something here for Hyper-V */
1618 hn_ifinit_locked(sc);
1620 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1625 sc->temp_unusable = FALSE;
1627 sc->hn_if_flags = ifp->if_flags;
1633 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1634 if (mask & IFCAP_TXCSUM) {
1635 ifp->if_capenable ^= IFCAP_TXCSUM;
1636 if (ifp->if_capenable & IFCAP_TXCSUM) {
1638 sc->hn_tx_ring[0].hn_csum_assist;
1641 ~sc->hn_tx_ring[0].hn_csum_assist;
1645 if (mask & IFCAP_RXCSUM)
1646 ifp->if_capenable ^= IFCAP_RXCSUM;
1648 if (mask & IFCAP_LRO)
1649 ifp->if_capenable ^= IFCAP_LRO;
1651 if (mask & IFCAP_TSO4) {
1652 ifp->if_capenable ^= IFCAP_TSO4;
1653 if (ifp->if_capenable & IFCAP_TSO4)
1654 ifp->if_hwassist |= CSUM_IP_TSO;
1656 ifp->if_hwassist &= ~CSUM_IP_TSO;
1659 if (mask & IFCAP_TSO6) {
1660 ifp->if_capenable ^= IFCAP_TSO6;
1661 if (ifp->if_capenable & IFCAP_TSO6)
1662 ifp->if_hwassist |= CSUM_IP6_TSO;
1664 ifp->if_hwassist &= ~CSUM_IP6_TSO;
1673 /* Fixme: Multicast mode? */
1674 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1676 netvsc_setmulti(sc);
1685 error = ifmedia_ioctl(ifp, ifr, &sc->hn_media, cmd);
1688 error = ether_ioctl(ifp, cmd, data);
1699 hn_stop(hn_softc_t *sc)
1703 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1708 printf(" Closing Device ...\n");
1710 atomic_clear_int(&ifp->if_drv_flags,
1711 (IFF_DRV_RUNNING | IFF_DRV_OACTIVE));
1712 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
1713 sc->hn_tx_ring[i].hn_oactive = 0;
1715 if_link_state_change(ifp, LINK_STATE_DOWN);
1716 sc->hn_initdone = 0;
1718 ret = hv_rf_on_close(device_ctx);
1722 * FreeBSD transmit entry point
1725 hn_start(struct ifnet *ifp)
1727 struct hn_softc *sc = ifp->if_softc;
1728 struct hn_tx_ring *txr = &sc->hn_tx_ring[0];
1730 if (txr->hn_sched_tx)
1733 if (mtx_trylock(&txr->hn_tx_lock)) {
1736 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1737 mtx_unlock(&txr->hn_tx_lock);
1742 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
1746 hn_start_txeof(struct hn_tx_ring *txr)
1748 struct hn_softc *sc = txr->hn_sc;
1749 struct ifnet *ifp = sc->hn_ifp;
1751 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1753 if (txr->hn_sched_tx)
1756 if (mtx_trylock(&txr->hn_tx_lock)) {
1759 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1760 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1761 mtx_unlock(&txr->hn_tx_lock);
1763 taskqueue_enqueue(txr->hn_tx_taskq,
1769 * Release the OACTIVE earlier, with the hope, that
1770 * others could catch up. The task will clear the
1771 * flag again with the hn_tx_lock to avoid possible
1774 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1775 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
1783 hn_ifinit_locked(hn_softc_t *sc)
1786 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1791 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1795 hv_promisc_mode = 1;
1797 ret = hv_rf_on_open(device_ctx);
1801 sc->hn_initdone = 1;
1804 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1805 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
1806 sc->hn_tx_ring[i].hn_oactive = 0;
1808 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_RUNNING);
1809 if_link_state_change(ifp, LINK_STATE_UP);
1816 hn_ifinit(void *xsc)
1818 hn_softc_t *sc = xsc;
1821 if (sc->temp_unusable) {
1825 sc->temp_unusable = TRUE;
1828 hn_ifinit_locked(sc);
1831 sc->temp_unusable = FALSE;
1840 hn_watchdog(struct ifnet *ifp)
1845 printf("hn%d: watchdog timeout -- resetting\n", sc->hn_unit);
1846 hn_ifinit(sc); /*???*/
1847 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1851 #if __FreeBSD_version >= 1100099
1854 hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS)
1856 struct hn_softc *sc = arg1;
1857 unsigned int lenlim;
1860 lenlim = sc->hn_rx_ring[0].hn_lro.lro_length_lim;
1861 error = sysctl_handle_int(oidp, &lenlim, 0, req);
1862 if (error || req->newptr == NULL)
1865 if (lenlim < HN_LRO_LENLIM_MIN(sc->hn_ifp) ||
1866 lenlim > TCP_LRO_LENGTH_MAX)
1870 hn_set_lro_lenlim(sc, lenlim);
1876 hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS)
1878 struct hn_softc *sc = arg1;
1879 int ackcnt, error, i;
1882 * lro_ackcnt_lim is append count limit,
1883 * +1 to turn it into aggregation limit.
1885 ackcnt = sc->hn_rx_ring[0].hn_lro.lro_ackcnt_lim + 1;
1886 error = sysctl_handle_int(oidp, &ackcnt, 0, req);
1887 if (error || req->newptr == NULL)
1890 if (ackcnt < 2 || ackcnt > (TCP_LRO_ACKCNT_MAX + 1))
1894 * Convert aggregation limit back to append
1899 for (i = 0; i < sc->hn_rx_ring_inuse; ++i)
1900 sc->hn_rx_ring[i].hn_lro.lro_ackcnt_lim = ackcnt;
1908 hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS)
1910 struct hn_softc *sc = arg1;
1915 if (sc->hn_rx_ring[0].hn_trust_hcsum & hcsum)
1918 error = sysctl_handle_int(oidp, &on, 0, req);
1919 if (error || req->newptr == NULL)
1923 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1924 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
1927 rxr->hn_trust_hcsum |= hcsum;
1929 rxr->hn_trust_hcsum &= ~hcsum;
1936 hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS)
1938 struct hn_softc *sc = arg1;
1939 int chimney_size, error;
1941 chimney_size = sc->hn_tx_ring[0].hn_tx_chimney_size;
1942 error = sysctl_handle_int(oidp, &chimney_size, 0, req);
1943 if (error || req->newptr == NULL)
1946 if (chimney_size > sc->hn_tx_chimney_max || chimney_size <= 0)
1949 hn_set_tx_chimney_size(sc, chimney_size);
1954 hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
1956 struct hn_softc *sc = arg1;
1957 int ofs = arg2, i, error;
1958 struct hn_rx_ring *rxr;
1962 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1963 rxr = &sc->hn_rx_ring[i];
1964 stat += *((u_long *)((uint8_t *)rxr + ofs));
1967 error = sysctl_handle_long(oidp, &stat, 0, req);
1968 if (error || req->newptr == NULL)
1971 /* Zero out this stat. */
1972 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1973 rxr = &sc->hn_rx_ring[i];
1974 *((u_long *)((uint8_t *)rxr + ofs)) = 0;
1980 hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS)
1982 struct hn_softc *sc = arg1;
1983 int ofs = arg2, i, error;
1984 struct hn_rx_ring *rxr;
1988 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1989 rxr = &sc->hn_rx_ring[i];
1990 stat += *((uint64_t *)((uint8_t *)rxr + ofs));
1993 error = sysctl_handle_64(oidp, &stat, 0, req);
1994 if (error || req->newptr == NULL)
1997 /* Zero out this stat. */
1998 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1999 rxr = &sc->hn_rx_ring[i];
2000 *((uint64_t *)((uint8_t *)rxr + ofs)) = 0;
2006 hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
2008 struct hn_softc *sc = arg1;
2009 int ofs = arg2, i, error;
2010 struct hn_tx_ring *txr;
2014 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2015 txr = &sc->hn_tx_ring[i];
2016 stat += *((u_long *)((uint8_t *)txr + ofs));
2019 error = sysctl_handle_long(oidp, &stat, 0, req);
2020 if (error || req->newptr == NULL)
2023 /* Zero out this stat. */
2024 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2025 txr = &sc->hn_tx_ring[i];
2026 *((u_long *)((uint8_t *)txr + ofs)) = 0;
2032 hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS)
2034 struct hn_softc *sc = arg1;
2035 int ofs = arg2, i, error, conf;
2036 struct hn_tx_ring *txr;
2038 txr = &sc->hn_tx_ring[0];
2039 conf = *((int *)((uint8_t *)txr + ofs));
2041 error = sysctl_handle_int(oidp, &conf, 0, req);
2042 if (error || req->newptr == NULL)
2046 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2047 txr = &sc->hn_tx_ring[i];
2048 *((int *)((uint8_t *)txr + ofs)) = conf;
2056 hn_check_iplen(const struct mbuf *m, int hoff)
2058 const struct ip *ip;
2059 int len, iphlen, iplen;
2060 const struct tcphdr *th;
2061 int thoff; /* TCP data offset */
2063 len = hoff + sizeof(struct ip);
2065 /* The packet must be at least the size of an IP header. */
2066 if (m->m_pkthdr.len < len)
2067 return IPPROTO_DONE;
2069 /* The fixed IP header must reside completely in the first mbuf. */
2071 return IPPROTO_DONE;
2073 ip = mtodo(m, hoff);
2075 /* Bound check the packet's stated IP header length. */
2076 iphlen = ip->ip_hl << 2;
2077 if (iphlen < sizeof(struct ip)) /* minimum header length */
2078 return IPPROTO_DONE;
2080 /* The full IP header must reside completely in the one mbuf. */
2081 if (m->m_len < hoff + iphlen)
2082 return IPPROTO_DONE;
2084 iplen = ntohs(ip->ip_len);
2087 * Check that the amount of data in the buffers is as
2088 * at least much as the IP header would have us expect.
2090 if (m->m_pkthdr.len < hoff + iplen)
2091 return IPPROTO_DONE;
2094 * Ignore IP fragments.
2096 if (ntohs(ip->ip_off) & (IP_OFFMASK | IP_MF))
2097 return IPPROTO_DONE;
2100 * The TCP/IP or UDP/IP header must be entirely contained within
2101 * the first fragment of a packet.
2105 if (iplen < iphlen + sizeof(struct tcphdr))
2106 return IPPROTO_DONE;
2107 if (m->m_len < hoff + iphlen + sizeof(struct tcphdr))
2108 return IPPROTO_DONE;
2109 th = (const struct tcphdr *)((const uint8_t *)ip + iphlen);
2110 thoff = th->th_off << 2;
2111 if (thoff < sizeof(struct tcphdr) || thoff + iphlen > iplen)
2112 return IPPROTO_DONE;
2113 if (m->m_len < hoff + iphlen + thoff)
2114 return IPPROTO_DONE;
2117 if (iplen < iphlen + sizeof(struct udphdr))
2118 return IPPROTO_DONE;
2119 if (m->m_len < hoff + iphlen + sizeof(struct udphdr))
2120 return IPPROTO_DONE;
2124 return IPPROTO_DONE;
2131 hn_dma_map_paddr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
2133 bus_addr_t *paddr = arg;
2138 KASSERT(nseg == 1, ("too many segments %d!", nseg));
2139 *paddr = segs->ds_addr;
2143 hn_create_rx_data(struct hn_softc *sc, int ring_cnt)
2145 struct sysctl_oid_list *child;
2146 struct sysctl_ctx_list *ctx;
2147 device_t dev = sc->hn_dev;
2148 #if defined(INET) || defined(INET6)
2149 #if __FreeBSD_version >= 1100095
2155 sc->hn_rx_ring_cnt = ring_cnt;
2156 sc->hn_rx_ring_inuse = sc->hn_rx_ring_cnt;
2158 sc->hn_rx_ring = malloc(sizeof(struct hn_rx_ring) * sc->hn_rx_ring_cnt,
2159 M_NETVSC, M_WAITOK | M_ZERO);
2161 #if defined(INET) || defined(INET6)
2162 #if __FreeBSD_version >= 1100095
2163 lroent_cnt = hn_lro_entry_count;
2164 if (lroent_cnt < TCP_LRO_ENTRIES)
2165 lroent_cnt = TCP_LRO_ENTRIES;
2166 device_printf(dev, "LRO: entry count %d\n", lroent_cnt);
2168 #endif /* INET || INET6 */
2170 ctx = device_get_sysctl_ctx(dev);
2171 child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
2173 /* Create dev.hn.UNIT.rx sysctl tree */
2174 sc->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "rx",
2175 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2177 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2178 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
2180 if (hn_trust_hosttcp)
2181 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_TCP;
2182 if (hn_trust_hostudp)
2183 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_UDP;
2184 if (hn_trust_hostip)
2185 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_IP;
2186 rxr->hn_ifp = sc->hn_ifp;
2192 #if defined(INET) || defined(INET6)
2193 #if __FreeBSD_version >= 1100095
2194 tcp_lro_init_args(&rxr->hn_lro, sc->hn_ifp, lroent_cnt, 0);
2196 tcp_lro_init(&rxr->hn_lro);
2197 rxr->hn_lro.ifp = sc->hn_ifp;
2199 #if __FreeBSD_version >= 1100099
2200 rxr->hn_lro.lro_length_lim = HN_LRO_LENLIM_DEF;
2201 rxr->hn_lro.lro_ackcnt_lim = HN_LRO_ACKCNT_DEF;
2203 #endif /* INET || INET6 */
2205 if (sc->hn_rx_sysctl_tree != NULL) {
2209 * Create per RX ring sysctl tree:
2210 * dev.hn.UNIT.rx.RINGID
2212 snprintf(name, sizeof(name), "%d", i);
2213 rxr->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx,
2214 SYSCTL_CHILDREN(sc->hn_rx_sysctl_tree),
2215 OID_AUTO, name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2217 if (rxr->hn_rx_sysctl_tree != NULL) {
2218 SYSCTL_ADD_ULONG(ctx,
2219 SYSCTL_CHILDREN(rxr->hn_rx_sysctl_tree),
2220 OID_AUTO, "packets", CTLFLAG_RW,
2221 &rxr->hn_pkts, "# of packets received");
2226 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_queued",
2227 CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2228 __offsetof(struct hn_rx_ring, hn_lro.lro_queued),
2229 hn_rx_stat_u64_sysctl, "LU", "LRO queued");
2230 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_flushed",
2231 CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2232 __offsetof(struct hn_rx_ring, hn_lro.lro_flushed),
2233 hn_rx_stat_u64_sysctl, "LU", "LRO flushed");
2234 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_tried",
2235 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2236 __offsetof(struct hn_rx_ring, hn_lro_tried),
2237 hn_rx_stat_ulong_sysctl, "LU", "# of LRO tries");
2238 #if __FreeBSD_version >= 1100099
2239 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_length_lim",
2240 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2241 hn_lro_lenlim_sysctl, "IU",
2242 "Max # of data bytes to be aggregated by LRO");
2243 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_ackcnt_lim",
2244 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2245 hn_lro_ackcnt_sysctl, "I",
2246 "Max # of ACKs to be aggregated by LRO");
2248 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hosttcp",
2249 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_TCP,
2250 hn_trust_hcsum_sysctl, "I",
2251 "Trust tcp segement verification on host side, "
2252 "when csum info is missing");
2253 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostudp",
2254 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_UDP,
2255 hn_trust_hcsum_sysctl, "I",
2256 "Trust udp datagram verification on host side, "
2257 "when csum info is missing");
2258 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostip",
2259 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_IP,
2260 hn_trust_hcsum_sysctl, "I",
2261 "Trust ip packet verification on host side, "
2262 "when csum info is missing");
2263 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_ip",
2264 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2265 __offsetof(struct hn_rx_ring, hn_csum_ip),
2266 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM IP");
2267 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_tcp",
2268 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2269 __offsetof(struct hn_rx_ring, hn_csum_tcp),
2270 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM TCP");
2271 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_udp",
2272 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2273 __offsetof(struct hn_rx_ring, hn_csum_udp),
2274 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM UDP");
2275 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_trusted",
2276 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2277 __offsetof(struct hn_rx_ring, hn_csum_trusted),
2278 hn_rx_stat_ulong_sysctl, "LU",
2279 "# of packets that we trust host's csum verification");
2280 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "small_pkts",
2281 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2282 __offsetof(struct hn_rx_ring, hn_small_pkts),
2283 hn_rx_stat_ulong_sysctl, "LU", "# of small packets received");
2284 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_cnt",
2285 CTLFLAG_RD, &sc->hn_rx_ring_cnt, 0, "# created RX rings");
2286 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_inuse",
2287 CTLFLAG_RD, &sc->hn_rx_ring_inuse, 0, "# used RX rings");
2291 hn_destroy_rx_data(struct hn_softc *sc)
2293 #if defined(INET) || defined(INET6)
2297 if (sc->hn_rx_ring_cnt == 0)
2300 #if defined(INET) || defined(INET6)
2301 for (i = 0; i < sc->hn_rx_ring_cnt; ++i)
2302 tcp_lro_free(&sc->hn_rx_ring[i].hn_lro);
2304 free(sc->hn_rx_ring, M_NETVSC);
2305 sc->hn_rx_ring = NULL;
2307 sc->hn_rx_ring_cnt = 0;
2308 sc->hn_rx_ring_inuse = 0;
2312 hn_create_tx_ring(struct hn_softc *sc, int id)
2314 struct hn_tx_ring *txr = &sc->hn_tx_ring[id];
2315 bus_dma_tag_t parent_dtag;
2319 txr->hn_tx_idx = id;
2321 #ifndef HN_USE_TXDESC_BUFRING
2322 mtx_init(&txr->hn_txlist_spin, "hn txlist", NULL, MTX_SPIN);
2324 mtx_init(&txr->hn_tx_lock, "hn tx", NULL, MTX_DEF);
2326 txr->hn_txdesc_cnt = HN_TX_DESC_CNT;
2327 txr->hn_txdesc = malloc(sizeof(struct hn_txdesc) * txr->hn_txdesc_cnt,
2328 M_NETVSC, M_WAITOK | M_ZERO);
2329 #ifndef HN_USE_TXDESC_BUFRING
2330 SLIST_INIT(&txr->hn_txlist);
2332 txr->hn_txdesc_br = buf_ring_alloc(txr->hn_txdesc_cnt, M_NETVSC,
2333 M_WAITOK, &txr->hn_tx_lock);
2336 txr->hn_tx_taskq = sc->hn_tx_taskq;
2338 if (hn_use_if_start) {
2339 txr->hn_txeof = hn_start_txeof;
2340 TASK_INIT(&txr->hn_tx_task, 0, hn_start_taskfunc, txr);
2341 TASK_INIT(&txr->hn_txeof_task, 0, hn_start_txeof_taskfunc, txr);
2345 txr->hn_txeof = hn_xmit_txeof;
2346 TASK_INIT(&txr->hn_tx_task, 0, hn_xmit_taskfunc, txr);
2347 TASK_INIT(&txr->hn_txeof_task, 0, hn_xmit_txeof_taskfunc, txr);
2349 br_depth = hn_get_txswq_depth(txr);
2350 txr->hn_mbuf_br = buf_ring_alloc(br_depth, M_NETVSC,
2351 M_WAITOK, &txr->hn_tx_lock);
2354 txr->hn_direct_tx_size = hn_direct_tx_size;
2355 if (hv_vmbus_protocal_version >= HV_VMBUS_VERSION_WIN8_1)
2356 txr->hn_csum_assist = HN_CSUM_ASSIST;
2358 txr->hn_csum_assist = HN_CSUM_ASSIST_WIN8;
2361 * Always schedule transmission instead of trying to do direct
2362 * transmission. This one gives the best performance so far.
2364 txr->hn_sched_tx = 1;
2366 parent_dtag = bus_get_dma_tag(sc->hn_dev);
2368 /* DMA tag for RNDIS messages. */
2369 error = bus_dma_tag_create(parent_dtag, /* parent */
2370 HN_RNDIS_MSG_ALIGN, /* alignment */
2371 HN_RNDIS_MSG_BOUNDARY, /* boundary */
2372 BUS_SPACE_MAXADDR, /* lowaddr */
2373 BUS_SPACE_MAXADDR, /* highaddr */
2374 NULL, NULL, /* filter, filterarg */
2375 HN_RNDIS_MSG_LEN, /* maxsize */
2377 HN_RNDIS_MSG_LEN, /* maxsegsize */
2379 NULL, /* lockfunc */
2380 NULL, /* lockfuncarg */
2381 &txr->hn_tx_rndis_dtag);
2383 device_printf(sc->hn_dev, "failed to create rndis dmatag\n");
2387 /* DMA tag for data. */
2388 error = bus_dma_tag_create(parent_dtag, /* parent */
2390 HN_TX_DATA_BOUNDARY, /* boundary */
2391 BUS_SPACE_MAXADDR, /* lowaddr */
2392 BUS_SPACE_MAXADDR, /* highaddr */
2393 NULL, NULL, /* filter, filterarg */
2394 HN_TX_DATA_MAXSIZE, /* maxsize */
2395 HN_TX_DATA_SEGCNT_MAX, /* nsegments */
2396 HN_TX_DATA_SEGSIZE, /* maxsegsize */
2398 NULL, /* lockfunc */
2399 NULL, /* lockfuncarg */
2400 &txr->hn_tx_data_dtag);
2402 device_printf(sc->hn_dev, "failed to create data dmatag\n");
2406 for (i = 0; i < txr->hn_txdesc_cnt; ++i) {
2407 struct hn_txdesc *txd = &txr->hn_txdesc[i];
2412 * Allocate and load RNDIS messages.
2414 error = bus_dmamem_alloc(txr->hn_tx_rndis_dtag,
2415 (void **)&txd->rndis_msg,
2416 BUS_DMA_WAITOK | BUS_DMA_COHERENT,
2417 &txd->rndis_msg_dmap);
2419 device_printf(sc->hn_dev,
2420 "failed to allocate rndis_msg, %d\n", i);
2424 error = bus_dmamap_load(txr->hn_tx_rndis_dtag,
2425 txd->rndis_msg_dmap,
2426 txd->rndis_msg, HN_RNDIS_MSG_LEN,
2427 hn_dma_map_paddr, &txd->rndis_msg_paddr,
2430 device_printf(sc->hn_dev,
2431 "failed to load rndis_msg, %d\n", i);
2432 bus_dmamem_free(txr->hn_tx_rndis_dtag,
2433 txd->rndis_msg, txd->rndis_msg_dmap);
2437 /* DMA map for TX data. */
2438 error = bus_dmamap_create(txr->hn_tx_data_dtag, 0,
2441 device_printf(sc->hn_dev,
2442 "failed to allocate tx data dmamap\n");
2443 bus_dmamap_unload(txr->hn_tx_rndis_dtag,
2444 txd->rndis_msg_dmap);
2445 bus_dmamem_free(txr->hn_tx_rndis_dtag,
2446 txd->rndis_msg, txd->rndis_msg_dmap);
2450 /* All set, put it to list */
2451 txd->flags |= HN_TXD_FLAG_ONLIST;
2452 #ifndef HN_USE_TXDESC_BUFRING
2453 SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
2455 buf_ring_enqueue(txr->hn_txdesc_br, txd);
2458 txr->hn_txdesc_avail = txr->hn_txdesc_cnt;
2460 if (sc->hn_tx_sysctl_tree != NULL) {
2461 struct sysctl_oid_list *child;
2462 struct sysctl_ctx_list *ctx;
2466 * Create per TX ring sysctl tree:
2467 * dev.hn.UNIT.tx.RINGID
2469 ctx = device_get_sysctl_ctx(sc->hn_dev);
2470 child = SYSCTL_CHILDREN(sc->hn_tx_sysctl_tree);
2472 snprintf(name, sizeof(name), "%d", id);
2473 txr->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO,
2474 name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2476 if (txr->hn_tx_sysctl_tree != NULL) {
2477 child = SYSCTL_CHILDREN(txr->hn_tx_sysctl_tree);
2479 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_avail",
2480 CTLFLAG_RD, &txr->hn_txdesc_avail, 0,
2481 "# of available TX descs");
2482 if (!hn_use_if_start) {
2483 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "oactive",
2484 CTLFLAG_RD, &txr->hn_oactive, 0,
2487 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "packets",
2488 CTLFLAG_RW, &txr->hn_pkts,
2489 "# of packets transmitted");
2497 hn_txdesc_dmamap_destroy(struct hn_txdesc *txd)
2499 struct hn_tx_ring *txr = txd->txr;
2501 KASSERT(txd->m == NULL, ("still has mbuf installed"));
2502 KASSERT((txd->flags & HN_TXD_FLAG_DMAMAP) == 0, ("still dma mapped"));
2504 bus_dmamap_unload(txr->hn_tx_rndis_dtag, txd->rndis_msg_dmap);
2505 bus_dmamem_free(txr->hn_tx_rndis_dtag, txd->rndis_msg,
2506 txd->rndis_msg_dmap);
2507 bus_dmamap_destroy(txr->hn_tx_data_dtag, txd->data_dmap);
2511 hn_destroy_tx_ring(struct hn_tx_ring *txr)
2513 struct hn_txdesc *txd;
2515 if (txr->hn_txdesc == NULL)
2518 #ifndef HN_USE_TXDESC_BUFRING
2519 while ((txd = SLIST_FIRST(&txr->hn_txlist)) != NULL) {
2520 SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
2521 hn_txdesc_dmamap_destroy(txd);
2524 mtx_lock(&txr->hn_tx_lock);
2525 while ((txd = buf_ring_dequeue_sc(txr->hn_txdesc_br)) != NULL)
2526 hn_txdesc_dmamap_destroy(txd);
2527 mtx_unlock(&txr->hn_tx_lock);
2530 if (txr->hn_tx_data_dtag != NULL)
2531 bus_dma_tag_destroy(txr->hn_tx_data_dtag);
2532 if (txr->hn_tx_rndis_dtag != NULL)
2533 bus_dma_tag_destroy(txr->hn_tx_rndis_dtag);
2535 #ifdef HN_USE_TXDESC_BUFRING
2536 buf_ring_free(txr->hn_txdesc_br, M_NETVSC);
2539 free(txr->hn_txdesc, M_NETVSC);
2540 txr->hn_txdesc = NULL;
2542 if (txr->hn_mbuf_br != NULL)
2543 buf_ring_free(txr->hn_mbuf_br, M_NETVSC);
2545 #ifndef HN_USE_TXDESC_BUFRING
2546 mtx_destroy(&txr->hn_txlist_spin);
2548 mtx_destroy(&txr->hn_tx_lock);
2552 hn_create_tx_data(struct hn_softc *sc, int ring_cnt)
2554 struct sysctl_oid_list *child;
2555 struct sysctl_ctx_list *ctx;
2558 sc->hn_tx_ring_cnt = ring_cnt;
2559 sc->hn_tx_ring_inuse = sc->hn_tx_ring_cnt;
2561 sc->hn_tx_ring = malloc(sizeof(struct hn_tx_ring) * sc->hn_tx_ring_cnt,
2562 M_NETVSC, M_WAITOK | M_ZERO);
2564 ctx = device_get_sysctl_ctx(sc->hn_dev);
2565 child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->hn_dev));
2567 /* Create dev.hn.UNIT.tx sysctl tree */
2568 sc->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "tx",
2569 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2571 for (i = 0; i < sc->hn_tx_ring_cnt; ++i) {
2574 error = hn_create_tx_ring(sc, i);
2579 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "no_txdescs",
2580 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2581 __offsetof(struct hn_tx_ring, hn_no_txdescs),
2582 hn_tx_stat_ulong_sysctl, "LU", "# of times short of TX descs");
2583 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "send_failed",
2584 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2585 __offsetof(struct hn_tx_ring, hn_send_failed),
2586 hn_tx_stat_ulong_sysctl, "LU", "# of hyper-v sending failure");
2587 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "txdma_failed",
2588 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2589 __offsetof(struct hn_tx_ring, hn_txdma_failed),
2590 hn_tx_stat_ulong_sysctl, "LU", "# of TX DMA failure");
2591 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_collapsed",
2592 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2593 __offsetof(struct hn_tx_ring, hn_tx_collapsed),
2594 hn_tx_stat_ulong_sysctl, "LU", "# of TX mbuf collapsed");
2595 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney",
2596 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2597 __offsetof(struct hn_tx_ring, hn_tx_chimney),
2598 hn_tx_stat_ulong_sysctl, "LU", "# of chimney send");
2599 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_tried",
2600 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2601 __offsetof(struct hn_tx_ring, hn_tx_chimney_tried),
2602 hn_tx_stat_ulong_sysctl, "LU", "# of chimney send tries");
2603 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_cnt",
2604 CTLFLAG_RD, &sc->hn_tx_ring[0].hn_txdesc_cnt, 0,
2605 "# of total TX descs");
2606 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_chimney_max",
2607 CTLFLAG_RD, &sc->hn_tx_chimney_max, 0,
2608 "Chimney send packet size upper boundary");
2609 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_size",
2610 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2611 hn_tx_chimney_size_sysctl,
2612 "I", "Chimney send packet size limit");
2613 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "direct_tx_size",
2614 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2615 __offsetof(struct hn_tx_ring, hn_direct_tx_size),
2616 hn_tx_conf_int_sysctl, "I",
2617 "Size of the packet for direct transmission");
2618 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "sched_tx",
2619 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2620 __offsetof(struct hn_tx_ring, hn_sched_tx),
2621 hn_tx_conf_int_sysctl, "I",
2622 "Always schedule transmission "
2623 "instead of doing direct transmission");
2624 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_cnt",
2625 CTLFLAG_RD, &sc->hn_tx_ring_cnt, 0, "# created TX rings");
2626 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_inuse",
2627 CTLFLAG_RD, &sc->hn_tx_ring_inuse, 0, "# used TX rings");
2633 hn_set_tx_chimney_size(struct hn_softc *sc, int chimney_size)
2638 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
2639 sc->hn_tx_ring[i].hn_tx_chimney_size = chimney_size;
2644 hn_destroy_tx_data(struct hn_softc *sc)
2648 if (sc->hn_tx_ring_cnt == 0)
2651 for (i = 0; i < sc->hn_tx_ring_cnt; ++i)
2652 hn_destroy_tx_ring(&sc->hn_tx_ring[i]);
2654 free(sc->hn_tx_ring, M_NETVSC);
2655 sc->hn_tx_ring = NULL;
2657 sc->hn_tx_ring_cnt = 0;
2658 sc->hn_tx_ring_inuse = 0;
2662 hn_start_taskfunc(void *xtxr, int pending __unused)
2664 struct hn_tx_ring *txr = xtxr;
2666 mtx_lock(&txr->hn_tx_lock);
2667 hn_start_locked(txr, 0);
2668 mtx_unlock(&txr->hn_tx_lock);
2672 hn_start_txeof_taskfunc(void *xtxr, int pending __unused)
2674 struct hn_tx_ring *txr = xtxr;
2676 mtx_lock(&txr->hn_tx_lock);
2677 atomic_clear_int(&txr->hn_sc->hn_ifp->if_drv_flags, IFF_DRV_OACTIVE);
2678 hn_start_locked(txr, 0);
2679 mtx_unlock(&txr->hn_tx_lock);
2683 hn_stop_tx_tasks(struct hn_softc *sc)
2687 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2688 struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
2690 taskqueue_drain(txr->hn_tx_taskq, &txr->hn_tx_task);
2691 taskqueue_drain(txr->hn_tx_taskq, &txr->hn_txeof_task);
2696 hn_xmit(struct hn_tx_ring *txr, int len)
2698 struct hn_softc *sc = txr->hn_sc;
2699 struct ifnet *ifp = sc->hn_ifp;
2700 struct mbuf *m_head;
2702 mtx_assert(&txr->hn_tx_lock, MA_OWNED);
2703 KASSERT(hn_use_if_start == 0,
2704 ("hn_xmit is called, when if_start is enabled"));
2706 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || txr->hn_oactive)
2709 while ((m_head = drbr_peek(ifp, txr->hn_mbuf_br)) != NULL) {
2710 struct hn_txdesc *txd;
2713 if (len > 0 && m_head->m_pkthdr.len > len) {
2715 * This sending could be time consuming; let callers
2716 * dispatch this packet sending (and sending of any
2717 * following up packets) to tx taskqueue.
2719 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2723 txd = hn_txdesc_get(txr);
2725 txr->hn_no_txdescs++;
2726 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2727 txr->hn_oactive = 1;
2731 error = hn_encap(txr, txd, &m_head);
2733 /* Both txd and m_head are freed; discard */
2734 drbr_advance(ifp, txr->hn_mbuf_br);
2738 error = hn_send_pkt(ifp, txr, txd);
2739 if (__predict_false(error)) {
2740 /* txd is freed, but m_head is not */
2741 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2742 txr->hn_oactive = 1;
2747 drbr_advance(ifp, txr->hn_mbuf_br);
2753 hn_transmit(struct ifnet *ifp, struct mbuf *m)
2755 struct hn_softc *sc = ifp->if_softc;
2756 struct hn_tx_ring *txr;
2760 * Select the TX ring based on flowid
2762 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2763 idx = m->m_pkthdr.flowid % sc->hn_tx_ring_inuse;
2764 txr = &sc->hn_tx_ring[idx];
2766 error = drbr_enqueue(ifp, txr->hn_mbuf_br, m);
2768 if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1);
2772 if (txr->hn_oactive)
2775 if (txr->hn_sched_tx)
2778 if (mtx_trylock(&txr->hn_tx_lock)) {
2781 sched = hn_xmit(txr, txr->hn_direct_tx_size);
2782 mtx_unlock(&txr->hn_tx_lock);
2787 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
2792 hn_xmit_qflush(struct ifnet *ifp)
2794 struct hn_softc *sc = ifp->if_softc;
2797 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2798 struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
2801 mtx_lock(&txr->hn_tx_lock);
2802 while ((m = buf_ring_dequeue_sc(txr->hn_mbuf_br)) != NULL)
2804 mtx_unlock(&txr->hn_tx_lock);
2810 hn_xmit_txeof(struct hn_tx_ring *txr)
2813 if (txr->hn_sched_tx)
2816 if (mtx_trylock(&txr->hn_tx_lock)) {
2819 txr->hn_oactive = 0;
2820 sched = hn_xmit(txr, txr->hn_direct_tx_size);
2821 mtx_unlock(&txr->hn_tx_lock);
2823 taskqueue_enqueue(txr->hn_tx_taskq,
2829 * Release the oactive earlier, with the hope, that
2830 * others could catch up. The task will clear the
2831 * oactive again with the hn_tx_lock to avoid possible
2834 txr->hn_oactive = 0;
2835 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
2840 hn_xmit_taskfunc(void *xtxr, int pending __unused)
2842 struct hn_tx_ring *txr = xtxr;
2844 mtx_lock(&txr->hn_tx_lock);
2846 mtx_unlock(&txr->hn_tx_lock);
2850 hn_xmit_txeof_taskfunc(void *xtxr, int pending __unused)
2852 struct hn_tx_ring *txr = xtxr;
2854 mtx_lock(&txr->hn_tx_lock);
2855 txr->hn_oactive = 0;
2857 mtx_unlock(&txr->hn_tx_lock);
2861 hn_channel_attach(struct hn_softc *sc, struct hv_vmbus_channel *chan)
2863 struct hn_rx_ring *rxr;
2866 idx = chan->offer_msg.offer.sub_channel_index;
2868 KASSERT(idx >= 0 && idx < sc->hn_rx_ring_inuse,
2869 ("invalid channel index %d, should > 0 && < %d",
2870 idx, sc->hn_rx_ring_inuse));
2871 rxr = &sc->hn_rx_ring[idx];
2872 KASSERT((rxr->hn_rx_flags & HN_RX_FLAG_ATTACHED) == 0,
2873 ("RX ring %d already attached", idx));
2874 rxr->hn_rx_flags |= HN_RX_FLAG_ATTACHED;
2876 chan->hv_chan_rxr = rxr;
2878 if_printf(sc->hn_ifp, "link RX ring %d to channel%u\n",
2879 idx, chan->offer_msg.child_rel_id);
2882 if (idx < sc->hn_tx_ring_inuse) {
2883 struct hn_tx_ring *txr = &sc->hn_tx_ring[idx];
2885 KASSERT((txr->hn_tx_flags & HN_TX_FLAG_ATTACHED) == 0,
2886 ("TX ring %d already attached", idx));
2887 txr->hn_tx_flags |= HN_TX_FLAG_ATTACHED;
2889 chan->hv_chan_txr = txr;
2890 txr->hn_chan = chan;
2892 if_printf(sc->hn_ifp, "link TX ring %d to channel%u\n",
2893 idx, chan->offer_msg.child_rel_id);
2897 /* Bind channel to a proper CPU */
2898 vmbus_channel_cpu_set(chan, (sc->hn_cpu + idx) % mp_ncpus);
2902 hn_subchan_attach(struct hn_softc *sc, struct hv_vmbus_channel *chan)
2905 KASSERT(!HV_VMBUS_CHAN_ISPRIMARY(chan),
2906 ("subchannel callback on primary channel"));
2907 KASSERT(chan->offer_msg.offer.sub_channel_index > 0,
2908 ("invalid channel subidx %u",
2909 chan->offer_msg.offer.sub_channel_index));
2910 hn_channel_attach(sc, chan);
2914 hn_tx_taskq_create(void *arg __unused)
2916 if (!hn_share_tx_taskq)
2919 hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
2920 taskqueue_thread_enqueue, &hn_tx_taskq);
2921 if (hn_bind_tx_taskq >= 0) {
2922 int cpu = hn_bind_tx_taskq;
2925 if (cpu > mp_ncpus - 1)
2927 CPU_SETOF(cpu, &cpu_set);
2928 taskqueue_start_threads_cpuset(&hn_tx_taskq, 1, PI_NET,
2931 taskqueue_start_threads(&hn_tx_taskq, 1, PI_NET, "hn tx");
2934 SYSINIT(hn_txtq_create, SI_SUB_DRIVERS, SI_ORDER_FIRST,
2935 hn_tx_taskq_create, NULL);
2938 hn_tx_taskq_destroy(void *arg __unused)
2940 if (hn_tx_taskq != NULL)
2941 taskqueue_free(hn_tx_taskq);
2943 SYSUNINIT(hn_txtq_destroy, SI_SUB_DRIVERS, SI_ORDER_FIRST,
2944 hn_tx_taskq_destroy, NULL);
2946 static device_method_t netvsc_methods[] = {
2947 /* Device interface */
2948 DEVMETHOD(device_probe, netvsc_probe),
2949 DEVMETHOD(device_attach, netvsc_attach),
2950 DEVMETHOD(device_detach, netvsc_detach),
2951 DEVMETHOD(device_shutdown, netvsc_shutdown),
2956 static driver_t netvsc_driver = {
2962 static devclass_t netvsc_devclass;
2964 DRIVER_MODULE(hn, vmbus, netvsc_driver, netvsc_devclass, 0, 0);
2965 MODULE_VERSION(hn, 1);
2966 MODULE_DEPEND(hn, vmbus, 1, 1, 1);