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>
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_var.h>
85 #include <net/if_types.h>
86 #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>
106 #include <sys/rman.h>
107 #include <sys/mutex.h>
108 #include <sys/errno.h>
109 #include <sys/types.h>
110 #include <machine/atomic.h>
112 #include <machine/intr_machdep.h>
114 #include <machine/in_cksum.h>
116 #include <dev/hyperv/include/hyperv.h>
117 #include <dev/hyperv/include/hyperv_busdma.h>
119 #include "hv_net_vsc.h"
120 #include "hv_rndis.h"
121 #include "hv_rndis_filter.h"
122 #include "vmbus_if.h"
124 /* Short for Hyper-V network interface */
125 #define NETVSC_DEVNAME "hn"
128 * It looks like offset 0 of buf is reserved to hold the softc pointer.
129 * The sc pointer evidently not needed, and is not presently populated.
130 * The packet offset is where the netvsc_packet starts in the buffer.
132 #define HV_NV_SC_PTR_OFFSET_IN_BUF 0
133 #define HV_NV_PACKET_OFFSET_IN_BUF 16
135 /* YYY should get it from the underlying channel */
136 #define HN_TX_DESC_CNT 512
138 #define HN_LROENT_CNT_DEF 128
140 #define HN_RING_CNT_DEF_MAX 8
142 #define HN_RNDIS_MSG_LEN \
143 (sizeof(rndis_msg) + \
144 RNDIS_HASHVAL_PPI_SIZE + \
145 RNDIS_VLAN_PPI_SIZE + \
146 RNDIS_TSO_PPI_SIZE + \
148 #define HN_RNDIS_MSG_BOUNDARY PAGE_SIZE
149 #define HN_RNDIS_MSG_ALIGN CACHE_LINE_SIZE
151 #define HN_TX_DATA_BOUNDARY PAGE_SIZE
152 #define HN_TX_DATA_MAXSIZE IP_MAXPACKET
153 #define HN_TX_DATA_SEGSIZE PAGE_SIZE
154 #define HN_TX_DATA_SEGCNT_MAX \
155 (NETVSC_PACKET_MAXPAGE - HV_RF_NUM_TX_RESERVED_PAGE_BUFS)
157 #define HN_DIRECT_TX_SIZE_DEF 128
159 #define HN_EARLY_TXEOF_THRESH 8
162 #ifndef HN_USE_TXDESC_BUFRING
163 SLIST_ENTRY(hn_txdesc) link;
166 struct hn_tx_ring *txr;
168 uint32_t flags; /* HN_TXD_FLAG_ */
169 netvsc_packet netvsc_pkt; /* XXX to be removed */
171 bus_dmamap_t data_dmap;
173 bus_addr_t rndis_msg_paddr;
174 rndis_msg *rndis_msg;
175 bus_dmamap_t rndis_msg_dmap;
178 #define HN_TXD_FLAG_ONLIST 0x1
179 #define HN_TXD_FLAG_DMAMAP 0x2
182 * Only enable UDP checksum offloading when it is on 2012R2 or
183 * later. UDP checksum offloading doesn't work on earlier
186 #define HN_CSUM_ASSIST_WIN8 (CSUM_IP | CSUM_TCP)
187 #define HN_CSUM_ASSIST (CSUM_IP | CSUM_UDP | CSUM_TCP)
189 #define HN_LRO_LENLIM_MULTIRX_DEF (12 * ETHERMTU)
190 #define HN_LRO_LENLIM_DEF (25 * ETHERMTU)
191 /* YYY 2*MTU is a bit rough, but should be good enough. */
192 #define HN_LRO_LENLIM_MIN(ifp) (2 * (ifp)->if_mtu)
194 #define HN_LRO_ACKCNT_DEF 1
197 * Be aware that this sleepable mutex will exhibit WITNESS errors when
198 * certain TCP and ARP code paths are taken. This appears to be a
199 * well-known condition, as all other drivers checked use a sleeping
200 * mutex to protect their transmit paths.
201 * Also Be aware that mutexes do not play well with semaphores, and there
202 * is a conflicting semaphore in a certain channel code path.
204 #define NV_LOCK_INIT(_sc, _name) \
205 mtx_init(&(_sc)->hn_lock, _name, MTX_NETWORK_LOCK, MTX_DEF)
206 #define NV_LOCK(_sc) mtx_lock(&(_sc)->hn_lock)
207 #define NV_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->hn_lock, MA_OWNED)
208 #define NV_UNLOCK(_sc) mtx_unlock(&(_sc)->hn_lock)
209 #define NV_LOCK_DESTROY(_sc) mtx_destroy(&(_sc)->hn_lock)
216 int hv_promisc_mode = 0; /* normal mode by default */
218 SYSCTL_NODE(_hw, OID_AUTO, hn, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
219 "Hyper-V network interface");
221 /* Trust tcp segements verification on host side. */
222 static int hn_trust_hosttcp = 1;
223 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hosttcp, CTLFLAG_RDTUN,
224 &hn_trust_hosttcp, 0,
225 "Trust tcp segement verification on host side, "
226 "when csum info is missing (global setting)");
228 /* Trust udp datagrams verification on host side. */
229 static int hn_trust_hostudp = 1;
230 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostudp, CTLFLAG_RDTUN,
231 &hn_trust_hostudp, 0,
232 "Trust udp datagram verification on host side, "
233 "when csum info is missing (global setting)");
235 /* Trust ip packets verification on host side. */
236 static int hn_trust_hostip = 1;
237 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostip, CTLFLAG_RDTUN,
239 "Trust ip packet verification on host side, "
240 "when csum info is missing (global setting)");
242 #if __FreeBSD_version >= 1100045
243 /* Limit TSO burst size */
244 static int hn_tso_maxlen = 0;
245 SYSCTL_INT(_hw_hn, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN,
246 &hn_tso_maxlen, 0, "TSO burst limit");
249 /* Limit chimney send size */
250 static int hn_tx_chimney_size = 0;
251 SYSCTL_INT(_hw_hn, OID_AUTO, tx_chimney_size, CTLFLAG_RDTUN,
252 &hn_tx_chimney_size, 0, "Chimney send packet size limit");
254 /* Limit the size of packet for direct transmission */
255 static int hn_direct_tx_size = HN_DIRECT_TX_SIZE_DEF;
256 SYSCTL_INT(_hw_hn, OID_AUTO, direct_tx_size, CTLFLAG_RDTUN,
257 &hn_direct_tx_size, 0, "Size of the packet for direct transmission");
259 #if defined(INET) || defined(INET6)
260 #if __FreeBSD_version >= 1100095
261 static int hn_lro_entry_count = HN_LROENT_CNT_DEF;
262 SYSCTL_INT(_hw_hn, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN,
263 &hn_lro_entry_count, 0, "LRO entry count");
267 static int hn_share_tx_taskq = 0;
268 SYSCTL_INT(_hw_hn, OID_AUTO, share_tx_taskq, CTLFLAG_RDTUN,
269 &hn_share_tx_taskq, 0, "Enable shared TX taskqueue");
271 static struct taskqueue *hn_tx_taskq;
273 #ifndef HN_USE_TXDESC_BUFRING
274 static int hn_use_txdesc_bufring = 0;
276 static int hn_use_txdesc_bufring = 1;
278 SYSCTL_INT(_hw_hn, OID_AUTO, use_txdesc_bufring, CTLFLAG_RD,
279 &hn_use_txdesc_bufring, 0, "Use buf_ring for TX descriptors");
281 static int hn_bind_tx_taskq = -1;
282 SYSCTL_INT(_hw_hn, OID_AUTO, bind_tx_taskq, CTLFLAG_RDTUN,
283 &hn_bind_tx_taskq, 0, "Bind TX taskqueue to the specified cpu");
285 static int hn_use_if_start = 0;
286 SYSCTL_INT(_hw_hn, OID_AUTO, use_if_start, CTLFLAG_RDTUN,
287 &hn_use_if_start, 0, "Use if_start TX method");
289 static int hn_chan_cnt = 0;
290 SYSCTL_INT(_hw_hn, OID_AUTO, chan_cnt, CTLFLAG_RDTUN,
292 "# of channels to use; each channel has one RX ring and one TX ring");
294 static int hn_tx_ring_cnt = 0;
295 SYSCTL_INT(_hw_hn, OID_AUTO, tx_ring_cnt, CTLFLAG_RDTUN,
296 &hn_tx_ring_cnt, 0, "# of TX rings to use");
298 static int hn_tx_swq_depth = 0;
299 SYSCTL_INT(_hw_hn, OID_AUTO, tx_swq_depth, CTLFLAG_RDTUN,
300 &hn_tx_swq_depth, 0, "Depth of IFQ or BUFRING");
302 #if __FreeBSD_version >= 1100095
303 static u_int hn_lro_mbufq_depth = 0;
304 SYSCTL_UINT(_hw_hn, OID_AUTO, lro_mbufq_depth, CTLFLAG_RDTUN,
305 &hn_lro_mbufq_depth, 0, "Depth of LRO mbuf queue");
308 static u_int hn_cpu_index;
311 * Forward declarations
313 static void hn_stop(hn_softc_t *sc);
314 static void hn_ifinit_locked(hn_softc_t *sc);
315 static void hn_ifinit(void *xsc);
316 static int hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data);
317 static int hn_start_locked(struct hn_tx_ring *txr, int len);
318 static void hn_start(struct ifnet *ifp);
319 static void hn_start_txeof(struct hn_tx_ring *);
320 static int hn_ifmedia_upd(struct ifnet *ifp);
321 static void hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr);
322 #if __FreeBSD_version >= 1100099
323 static int hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS);
324 static int hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS);
326 static int hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS);
327 static int hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS);
328 static int hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS);
329 static int hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS);
330 static int hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS);
331 static int hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS);
332 static int hn_check_iplen(const struct mbuf *, int);
333 static int hn_create_tx_ring(struct hn_softc *, int);
334 static void hn_destroy_tx_ring(struct hn_tx_ring *);
335 static int hn_create_tx_data(struct hn_softc *, int);
336 static void hn_destroy_tx_data(struct hn_softc *);
337 static void hn_start_taskfunc(void *, int);
338 static void hn_start_txeof_taskfunc(void *, int);
339 static void hn_stop_tx_tasks(struct hn_softc *);
340 static int hn_encap(struct hn_tx_ring *, struct hn_txdesc *, struct mbuf **);
341 static void hn_create_rx_data(struct hn_softc *sc, int);
342 static void hn_destroy_rx_data(struct hn_softc *sc);
343 static void hn_set_tx_chimney_size(struct hn_softc *, int);
344 static void hn_channel_attach(struct hn_softc *, struct vmbus_channel *);
345 static void hn_subchan_attach(struct hn_softc *, struct vmbus_channel *);
346 static void hn_subchan_setup(struct hn_softc *);
348 static int hn_transmit(struct ifnet *, struct mbuf *);
349 static void hn_xmit_qflush(struct ifnet *);
350 static int hn_xmit(struct hn_tx_ring *, int);
351 static void hn_xmit_txeof(struct hn_tx_ring *);
352 static void hn_xmit_taskfunc(void *, int);
353 static void hn_xmit_txeof_taskfunc(void *, int);
355 #if __FreeBSD_version >= 1100099
357 hn_set_lro_lenlim(struct hn_softc *sc, int lenlim)
361 for (i = 0; i < sc->hn_rx_ring_inuse; ++i)
362 sc->hn_rx_ring[i].hn_lro.lro_length_lim = lenlim;
367 hn_get_txswq_depth(const struct hn_tx_ring *txr)
370 KASSERT(txr->hn_txdesc_cnt > 0, ("tx ring is not setup yet"));
371 if (hn_tx_swq_depth < txr->hn_txdesc_cnt)
372 return txr->hn_txdesc_cnt;
373 return hn_tx_swq_depth;
377 hn_ifmedia_upd(struct ifnet *ifp __unused)
384 hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
386 struct hn_softc *sc = ifp->if_softc;
388 ifmr->ifm_status = IFM_AVALID;
389 ifmr->ifm_active = IFM_ETHER;
391 if (!sc->hn_carrier) {
392 ifmr->ifm_active |= IFM_NONE;
395 ifmr->ifm_status |= IFM_ACTIVE;
396 ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
399 /* {F8615163-DF3E-46c5-913F-F2D2F965ED0E} */
400 static const struct hyperv_guid g_net_vsc_device_type = {
401 .hv_guid = {0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
402 0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E}
406 * Standard probe entry point.
410 netvsc_probe(device_t dev)
412 if (VMBUS_PROBE_GUID(device_get_parent(dev), dev,
413 &g_net_vsc_device_type) == 0) {
414 device_set_desc(dev, "Hyper-V Network Interface");
415 return BUS_PROBE_DEFAULT;
421 * Standard attach entry point.
423 * Called when the driver is loaded. It allocates needed resources,
424 * and initializes the "hardware" and software.
427 netvsc_attach(device_t dev)
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);
442 sc->hn_prichan = vmbus_get_channel(dev);
444 if (hn_tx_taskq == NULL) {
445 sc->hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
446 taskqueue_thread_enqueue, &sc->hn_tx_taskq);
447 if (hn_bind_tx_taskq >= 0) {
448 int cpu = hn_bind_tx_taskq;
451 if (cpu > mp_ncpus - 1)
453 CPU_SETOF(cpu, &cpu_set);
454 taskqueue_start_threads_cpuset(&sc->hn_tx_taskq, 1,
455 PI_NET, &cpu_set, "%s tx",
456 device_get_nameunit(dev));
458 taskqueue_start_threads(&sc->hn_tx_taskq, 1, PI_NET,
459 "%s tx", device_get_nameunit(dev));
462 sc->hn_tx_taskq = hn_tx_taskq;
464 NV_LOCK_INIT(sc, "NetVSCLock");
466 ifp = sc->hn_ifp = if_alloc(IFT_ETHER);
468 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
471 * Figure out the # of RX rings (ring_cnt) and the # of TX rings
472 * to use (tx_ring_cnt).
475 * The # of RX rings to use is same as the # of channels to use.
477 ring_cnt = hn_chan_cnt;
481 if (ring_cnt > HN_RING_CNT_DEF_MAX)
482 ring_cnt = HN_RING_CNT_DEF_MAX;
483 } else if (ring_cnt > mp_ncpus) {
487 tx_ring_cnt = hn_tx_ring_cnt;
488 if (tx_ring_cnt <= 0 || tx_ring_cnt > ring_cnt)
489 tx_ring_cnt = ring_cnt;
490 if (hn_use_if_start) {
491 /* ifnet.if_start only needs one TX ring. */
496 * Set the leader CPU for channels.
498 sc->hn_cpu = atomic_fetchadd_int(&hn_cpu_index, ring_cnt) % mp_ncpus;
500 error = hn_create_tx_data(sc, tx_ring_cnt);
503 hn_create_rx_data(sc, ring_cnt);
506 * Associate the first TX/RX ring w/ the primary channel.
508 hn_channel_attach(sc, sc->hn_prichan);
510 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
511 ifp->if_ioctl = hn_ioctl;
512 ifp->if_init = hn_ifinit;
513 /* needed by hv_rf_on_device_add() code */
514 ifp->if_mtu = ETHERMTU;
515 if (hn_use_if_start) {
516 int qdepth = hn_get_txswq_depth(&sc->hn_tx_ring[0]);
518 ifp->if_start = hn_start;
519 IFQ_SET_MAXLEN(&ifp->if_snd, qdepth);
520 ifp->if_snd.ifq_drv_maxlen = qdepth - 1;
521 IFQ_SET_READY(&ifp->if_snd);
523 ifp->if_transmit = hn_transmit;
524 ifp->if_qflush = hn_xmit_qflush;
527 ifmedia_init(&sc->hn_media, 0, hn_ifmedia_upd, hn_ifmedia_sts);
528 ifmedia_add(&sc->hn_media, IFM_ETHER | IFM_AUTO, 0, NULL);
529 ifmedia_set(&sc->hn_media, IFM_ETHER | IFM_AUTO);
530 /* XXX ifmedia_set really should do this for us */
531 sc->hn_media.ifm_media = sc->hn_media.ifm_cur->ifm_media;
534 * Tell upper layers that we support full VLAN capability.
536 ifp->if_hdrlen = sizeof(struct ether_vlan_header);
537 ifp->if_capabilities |=
538 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
541 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
543 ifp->if_hwassist = sc->hn_tx_ring[0].hn_csum_assist | CSUM_TSO;
545 error = hv_rf_on_device_add(sc, &device_info, ring_cnt,
549 KASSERT(sc->net_dev->num_channel > 0 &&
550 sc->net_dev->num_channel <= sc->hn_rx_ring_inuse,
551 ("invalid channel count %u, should be less than %d",
552 sc->net_dev->num_channel, sc->hn_rx_ring_inuse));
555 * Set the # of TX/RX rings that could be used according to
556 * the # of channels that host offered.
558 if (sc->hn_tx_ring_inuse > sc->net_dev->num_channel)
559 sc->hn_tx_ring_inuse = sc->net_dev->num_channel;
560 sc->hn_rx_ring_inuse = sc->net_dev->num_channel;
561 device_printf(dev, "%d TX ring, %d RX ring\n",
562 sc->hn_tx_ring_inuse, sc->hn_rx_ring_inuse);
564 if (sc->net_dev->num_channel > 1)
565 hn_subchan_setup(sc);
567 #if __FreeBSD_version >= 1100099
568 if (sc->hn_rx_ring_inuse > 1) {
570 * Reduce TCP segment aggregation limit for multiple
571 * RX rings to increase ACK timeliness.
573 hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MULTIRX_DEF);
577 if (device_info.link_state == 0) {
581 #if __FreeBSD_version >= 1100045
582 tso_maxlen = hn_tso_maxlen;
583 if (tso_maxlen <= 0 || tso_maxlen > IP_MAXPACKET)
584 tso_maxlen = IP_MAXPACKET;
586 ifp->if_hw_tsomaxsegcount = HN_TX_DATA_SEGCNT_MAX;
587 ifp->if_hw_tsomaxsegsize = PAGE_SIZE;
588 ifp->if_hw_tsomax = tso_maxlen -
589 (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
592 ether_ifattach(ifp, device_info.mac_addr);
594 #if __FreeBSD_version >= 1100045
595 if_printf(ifp, "TSO: %u/%u/%u\n", ifp->if_hw_tsomax,
596 ifp->if_hw_tsomaxsegcount, ifp->if_hw_tsomaxsegsize);
599 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
600 hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max);
601 if (hn_tx_chimney_size > 0 &&
602 hn_tx_chimney_size < sc->hn_tx_chimney_max)
603 hn_set_tx_chimney_size(sc, hn_tx_chimney_size);
607 hn_destroy_tx_data(sc);
614 * Standard detach entry point
617 netvsc_detach(device_t dev)
619 struct hn_softc *sc = device_get_softc(dev);
622 printf("netvsc_detach\n");
625 * XXXKYS: Need to clean up all our
626 * driver state; this is the driver
631 * XXXKYS: Need to stop outgoing traffic and unregister
635 hv_rf_on_device_remove(sc, HV_RF_NV_DESTROY_CHANNEL);
637 hn_stop_tx_tasks(sc);
639 ifmedia_removeall(&sc->hn_media);
640 hn_destroy_rx_data(sc);
641 hn_destroy_tx_data(sc);
643 if (sc->hn_tx_taskq != hn_tx_taskq)
644 taskqueue_free(sc->hn_tx_taskq);
650 * Standard shutdown entry point
653 netvsc_shutdown(device_t dev)
659 hn_txdesc_dmamap_load(struct hn_tx_ring *txr, struct hn_txdesc *txd,
660 struct mbuf **m_head, bus_dma_segment_t *segs, int *nsegs)
662 struct mbuf *m = *m_head;
665 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag, txd->data_dmap,
666 m, segs, nsegs, BUS_DMA_NOWAIT);
667 if (error == EFBIG) {
670 m_new = m_collapse(m, M_NOWAIT, HN_TX_DATA_SEGCNT_MAX);
675 txr->hn_tx_collapsed++;
677 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag,
678 txd->data_dmap, m, segs, nsegs, BUS_DMA_NOWAIT);
681 bus_dmamap_sync(txr->hn_tx_data_dtag, txd->data_dmap,
682 BUS_DMASYNC_PREWRITE);
683 txd->flags |= HN_TXD_FLAG_DMAMAP;
689 hn_txdesc_dmamap_unload(struct hn_tx_ring *txr, struct hn_txdesc *txd)
692 if (txd->flags & HN_TXD_FLAG_DMAMAP) {
693 bus_dmamap_sync(txr->hn_tx_data_dtag,
694 txd->data_dmap, BUS_DMASYNC_POSTWRITE);
695 bus_dmamap_unload(txr->hn_tx_data_dtag,
697 txd->flags &= ~HN_TXD_FLAG_DMAMAP;
702 hn_txdesc_put(struct hn_tx_ring *txr, struct hn_txdesc *txd)
705 KASSERT((txd->flags & HN_TXD_FLAG_ONLIST) == 0,
706 ("put an onlist txd %#x", txd->flags));
708 KASSERT(txd->refs > 0, ("invalid txd refs %d", txd->refs));
709 if (atomic_fetchadd_int(&txd->refs, -1) != 1)
712 hn_txdesc_dmamap_unload(txr, txd);
713 if (txd->m != NULL) {
718 txd->flags |= HN_TXD_FLAG_ONLIST;
720 #ifndef HN_USE_TXDESC_BUFRING
721 mtx_lock_spin(&txr->hn_txlist_spin);
722 KASSERT(txr->hn_txdesc_avail >= 0 &&
723 txr->hn_txdesc_avail < txr->hn_txdesc_cnt,
724 ("txdesc_put: invalid txd avail %d", txr->hn_txdesc_avail));
725 txr->hn_txdesc_avail++;
726 SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
727 mtx_unlock_spin(&txr->hn_txlist_spin);
729 atomic_add_int(&txr->hn_txdesc_avail, 1);
730 buf_ring_enqueue(txr->hn_txdesc_br, txd);
736 static __inline struct hn_txdesc *
737 hn_txdesc_get(struct hn_tx_ring *txr)
739 struct hn_txdesc *txd;
741 #ifndef HN_USE_TXDESC_BUFRING
742 mtx_lock_spin(&txr->hn_txlist_spin);
743 txd = SLIST_FIRST(&txr->hn_txlist);
745 KASSERT(txr->hn_txdesc_avail > 0,
746 ("txdesc_get: invalid txd avail %d", txr->hn_txdesc_avail));
747 txr->hn_txdesc_avail--;
748 SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
750 mtx_unlock_spin(&txr->hn_txlist_spin);
752 txd = buf_ring_dequeue_sc(txr->hn_txdesc_br);
756 #ifdef HN_USE_TXDESC_BUFRING
757 atomic_subtract_int(&txr->hn_txdesc_avail, 1);
759 KASSERT(txd->m == NULL && txd->refs == 0 &&
760 (txd->flags & HN_TXD_FLAG_ONLIST), ("invalid txd"));
761 txd->flags &= ~HN_TXD_FLAG_ONLIST;
768 hn_txdesc_hold(struct hn_txdesc *txd)
771 /* 0->1 transition will never work */
772 KASSERT(txd->refs > 0, ("invalid refs %d", txd->refs));
773 atomic_add_int(&txd->refs, 1);
777 hn_txeof(struct hn_tx_ring *txr)
779 txr->hn_has_txeof = 0;
784 hn_tx_done(struct vmbus_channel *chan, void *xpkt)
786 netvsc_packet *packet = xpkt;
787 struct hn_txdesc *txd;
788 struct hn_tx_ring *txr;
790 txd = (struct hn_txdesc *)(uintptr_t)
791 packet->compl.send.send_completion_tid;
794 KASSERT(txr->hn_chan == chan,
795 ("channel mismatch, on chan%u, should be chan%u",
796 vmbus_chan_subidx(chan), vmbus_chan_subidx(txr->hn_chan)));
798 txr->hn_has_txeof = 1;
799 hn_txdesc_put(txr, txd);
801 ++txr->hn_txdone_cnt;
802 if (txr->hn_txdone_cnt >= HN_EARLY_TXEOF_THRESH) {
803 txr->hn_txdone_cnt = 0;
810 netvsc_channel_rollup(struct hn_rx_ring *rxr, struct hn_tx_ring *txr)
812 #if defined(INET) || defined(INET6)
813 tcp_lro_flush_all(&rxr->hn_lro);
818 * 'txr' could be NULL, if multiple channels and
819 * ifnet.if_start method are enabled.
821 if (txr == NULL || !txr->hn_has_txeof)
824 txr->hn_txdone_cnt = 0;
830 * If this function fails, then both txd and m_head0 will be freed.
833 hn_encap(struct hn_tx_ring *txr, struct hn_txdesc *txd, struct mbuf **m_head0)
835 bus_dma_segment_t segs[HN_TX_DATA_SEGCNT_MAX];
837 struct mbuf *m_head = *m_head0;
838 netvsc_packet *packet;
839 rndis_msg *rndis_mesg;
840 rndis_packet *rndis_pkt;
841 rndis_per_packet_info *rppi;
842 struct rndis_hash_value *hash_value;
843 uint32_t rndis_msg_size;
845 packet = &txd->netvsc_pkt;
846 packet->is_data_pkt = TRUE;
847 packet->tot_data_buf_len = m_head->m_pkthdr.len;
850 * extension points to the area reserved for the
851 * rndis_filter_packet, which is placed just after
852 * the netvsc_packet (and rppi struct, if present;
853 * length is updated later).
855 rndis_mesg = txd->rndis_msg;
856 /* XXX not necessary */
857 memset(rndis_mesg, 0, HN_RNDIS_MSG_LEN);
858 rndis_mesg->ndis_msg_type = REMOTE_NDIS_PACKET_MSG;
860 rndis_pkt = &rndis_mesg->msg.packet;
861 rndis_pkt->data_offset = sizeof(rndis_packet);
862 rndis_pkt->data_length = packet->tot_data_buf_len;
863 rndis_pkt->per_pkt_info_offset = sizeof(rndis_packet);
865 rndis_msg_size = RNDIS_MESSAGE_SIZE(rndis_packet);
868 * Set the hash value for this packet, so that the host could
869 * dispatch the TX done event for this packet back to this TX
872 rndis_msg_size += RNDIS_HASHVAL_PPI_SIZE;
873 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_HASHVAL_PPI_SIZE,
875 hash_value = (struct rndis_hash_value *)((uint8_t *)rppi +
876 rppi->per_packet_info_offset);
877 hash_value->hash_value = txr->hn_tx_idx;
879 if (m_head->m_flags & M_VLANTAG) {
880 ndis_8021q_info *rppi_vlan_info;
882 rndis_msg_size += RNDIS_VLAN_PPI_SIZE;
883 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_VLAN_PPI_SIZE,
886 rppi_vlan_info = (ndis_8021q_info *)((uint8_t *)rppi +
887 rppi->per_packet_info_offset);
888 rppi_vlan_info->u1.s1.vlan_id =
889 m_head->m_pkthdr.ether_vtag & 0xfff;
892 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
893 rndis_tcp_tso_info *tso_info;
894 struct ether_vlan_header *eh;
898 * XXX need m_pullup and use mtodo
900 eh = mtod(m_head, struct ether_vlan_header*);
901 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN))
902 ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
904 ether_len = ETHER_HDR_LEN;
906 rndis_msg_size += RNDIS_TSO_PPI_SIZE;
907 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_TSO_PPI_SIZE,
908 tcp_large_send_info);
910 tso_info = (rndis_tcp_tso_info *)((uint8_t *)rppi +
911 rppi->per_packet_info_offset);
912 tso_info->lso_v2_xmit.type =
913 RNDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
916 if (m_head->m_pkthdr.csum_flags & CSUM_IP_TSO) {
918 (struct ip *)(m_head->m_data + ether_len);
919 unsigned long iph_len = ip->ip_hl << 2;
921 (struct tcphdr *)((caddr_t)ip + iph_len);
923 tso_info->lso_v2_xmit.ip_version =
924 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
928 th->th_sum = in_pseudo(ip->ip_src.s_addr,
929 ip->ip_dst.s_addr, htons(IPPROTO_TCP));
932 #if defined(INET6) && defined(INET)
937 struct ip6_hdr *ip6 = (struct ip6_hdr *)
938 (m_head->m_data + ether_len);
939 struct tcphdr *th = (struct tcphdr *)(ip6 + 1);
941 tso_info->lso_v2_xmit.ip_version =
942 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
944 th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
947 tso_info->lso_v2_xmit.tcp_header_offset = 0;
948 tso_info->lso_v2_xmit.mss = m_head->m_pkthdr.tso_segsz;
949 } else if (m_head->m_pkthdr.csum_flags & txr->hn_csum_assist) {
950 rndis_tcp_ip_csum_info *csum_info;
952 rndis_msg_size += RNDIS_CSUM_PPI_SIZE;
953 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_CSUM_PPI_SIZE,
955 csum_info = (rndis_tcp_ip_csum_info *)((uint8_t *)rppi +
956 rppi->per_packet_info_offset);
958 csum_info->xmit.is_ipv4 = 1;
959 if (m_head->m_pkthdr.csum_flags & CSUM_IP)
960 csum_info->xmit.ip_header_csum = 1;
962 if (m_head->m_pkthdr.csum_flags & CSUM_TCP) {
963 csum_info->xmit.tcp_csum = 1;
964 csum_info->xmit.tcp_header_offset = 0;
965 } else if (m_head->m_pkthdr.csum_flags & CSUM_UDP) {
966 csum_info->xmit.udp_csum = 1;
970 rndis_mesg->msg_len = packet->tot_data_buf_len + rndis_msg_size;
971 packet->tot_data_buf_len = rndis_mesg->msg_len;
974 * Chimney send, if the packet could fit into one chimney buffer.
976 if (packet->tot_data_buf_len < txr->hn_tx_chimney_size) {
977 netvsc_dev *net_dev = txr->hn_sc->net_dev;
978 uint32_t send_buf_section_idx;
980 txr->hn_tx_chimney_tried++;
981 send_buf_section_idx =
982 hv_nv_get_next_send_section(net_dev);
983 if (send_buf_section_idx !=
984 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX) {
985 uint8_t *dest = ((uint8_t *)net_dev->send_buf +
986 (send_buf_section_idx *
987 net_dev->send_section_size));
989 memcpy(dest, rndis_mesg, rndis_msg_size);
990 dest += rndis_msg_size;
991 m_copydata(m_head, 0, m_head->m_pkthdr.len, dest);
993 packet->send_buf_section_idx = send_buf_section_idx;
994 packet->send_buf_section_size =
995 packet->tot_data_buf_len;
997 txr->hn_tx_chimney++;
1002 error = hn_txdesc_dmamap_load(txr, txd, &m_head, segs, &nsegs);
1007 * This mbuf is not linked w/ the txd yet, so free it now.
1012 freed = hn_txdesc_put(txr, txd);
1014 ("fail to free txd upon txdma error"));
1016 txr->hn_txdma_failed++;
1017 if_inc_counter(txr->hn_sc->hn_ifp, IFCOUNTER_OERRORS, 1);
1022 txr->hn_gpa_cnt = nsegs + HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
1024 /* send packet with page buffer */
1025 txr->hn_gpa[0].gpa_page = atop(txd->rndis_msg_paddr);
1026 txr->hn_gpa[0].gpa_ofs = txd->rndis_msg_paddr & PAGE_MASK;
1027 txr->hn_gpa[0].gpa_len = rndis_msg_size;
1030 * Fill the page buffers with mbuf info starting at index
1031 * HV_RF_NUM_TX_RESERVED_PAGE_BUFS.
1033 for (i = 0; i < nsegs; ++i) {
1034 struct vmbus_gpa *gpa = &txr->hn_gpa[
1035 i + HV_RF_NUM_TX_RESERVED_PAGE_BUFS];
1037 gpa->gpa_page = atop(segs[i].ds_addr);
1038 gpa->gpa_ofs = segs[i].ds_addr & PAGE_MASK;
1039 gpa->gpa_len = segs[i].ds_len;
1042 packet->send_buf_section_idx =
1043 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX;
1044 packet->send_buf_section_size = 0;
1048 /* Set the completion routine */
1049 packet->compl.send.on_send_completion = hn_tx_done;
1050 packet->compl.send.send_completion_context = packet;
1051 packet->compl.send.send_completion_tid = (uint64_t)(uintptr_t)txd;
1058 * If this function fails, then txd will be freed, but the mbuf
1059 * associated w/ the txd will _not_ be freed.
1062 hn_send_pkt(struct ifnet *ifp, struct hn_tx_ring *txr, struct hn_txdesc *txd)
1064 int error, send_failed = 0;
1068 * Make sure that txd is not freed before ETHER_BPF_MTAP.
1070 hn_txdesc_hold(txd);
1071 error = hv_nv_on_send(txr->hn_chan, &txd->netvsc_pkt,
1072 txr->hn_gpa, txr->hn_gpa_cnt);
1074 ETHER_BPF_MTAP(ifp, txd->m);
1075 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1076 if (!hn_use_if_start) {
1077 if_inc_counter(ifp, IFCOUNTER_OBYTES,
1078 txd->m->m_pkthdr.len);
1079 if (txd->m->m_flags & M_MCAST)
1080 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
1084 hn_txdesc_put(txr, txd);
1086 if (__predict_false(error)) {
1090 * This should "really rarely" happen.
1092 * XXX Too many RX to be acked or too many sideband
1093 * commands to run? Ask netvsc_channel_rollup()
1094 * to kick start later.
1096 txr->hn_has_txeof = 1;
1098 txr->hn_send_failed++;
1101 * Try sending again after set hn_has_txeof;
1102 * in case that we missed the last
1103 * netvsc_channel_rollup().
1107 if_printf(ifp, "send failed\n");
1110 * Caller will perform further processing on the
1111 * associated mbuf, so don't free it in hn_txdesc_put();
1112 * only unload it from the DMA map in hn_txdesc_put(),
1116 freed = hn_txdesc_put(txr, txd);
1118 ("fail to free txd upon send error"));
1120 txr->hn_send_failed++;
1126 * Start a transmit of one or more packets
1129 hn_start_locked(struct hn_tx_ring *txr, int len)
1131 struct hn_softc *sc = txr->hn_sc;
1132 struct ifnet *ifp = sc->hn_ifp;
1134 KASSERT(hn_use_if_start,
1135 ("hn_start_locked is called, when if_start is disabled"));
1136 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1137 mtx_assert(&txr->hn_tx_lock, MA_OWNED);
1139 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1143 while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
1144 struct hn_txdesc *txd;
1145 struct mbuf *m_head;
1148 IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
1152 if (len > 0 && m_head->m_pkthdr.len > len) {
1154 * This sending could be time consuming; let callers
1155 * dispatch this packet sending (and sending of any
1156 * following up packets) to tx taskqueue.
1158 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1162 txd = hn_txdesc_get(txr);
1164 txr->hn_no_txdescs++;
1165 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1166 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1170 error = hn_encap(txr, txd, &m_head);
1172 /* Both txd and m_head are freed */
1176 error = hn_send_pkt(ifp, txr, txd);
1177 if (__predict_false(error)) {
1178 /* txd is freed, but m_head is not */
1179 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1180 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1188 * Link up/down notification
1191 netvsc_linkstatus_callback(struct hn_softc *sc, uint32_t status)
1201 * Append the specified data to the indicated mbuf chain,
1202 * Extend the mbuf chain if the new data does not fit in
1205 * This is a minor rewrite of m_append() from sys/kern/uipc_mbuf.c.
1206 * There should be an equivalent in the kernel mbuf code,
1207 * but there does not appear to be one yet.
1209 * Differs from m_append() in that additional mbufs are
1210 * allocated with cluster size MJUMPAGESIZE, and filled
1213 * Return 1 if able to complete the job; otherwise 0.
1216 hv_m_append(struct mbuf *m0, int len, c_caddr_t cp)
1219 int remainder, space;
1221 for (m = m0; m->m_next != NULL; m = m->m_next)
1224 space = M_TRAILINGSPACE(m);
1227 * Copy into available space.
1229 if (space > remainder)
1231 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1236 while (remainder > 0) {
1238 * Allocate a new mbuf; could check space
1239 * and allocate a cluster instead.
1241 n = m_getjcl(M_NOWAIT, m->m_type, 0, MJUMPAGESIZE);
1244 n->m_len = min(MJUMPAGESIZE, remainder);
1245 bcopy(cp, mtod(n, caddr_t), n->m_len);
1247 remainder -= n->m_len;
1251 if (m0->m_flags & M_PKTHDR)
1252 m0->m_pkthdr.len += len - remainder;
1254 return (remainder == 0);
1257 #if defined(INET) || defined(INET6)
1259 hn_lro_rx(struct lro_ctrl *lc, struct mbuf *m)
1261 #if __FreeBSD_version >= 1100095
1262 if (hn_lro_mbufq_depth) {
1263 tcp_lro_queue_mbuf(lc, m);
1267 return tcp_lro_rx(lc, m, 0);
1272 * Called when we receive a data packet from the "wire" on the
1275 * Note: This is no longer used as a callback
1278 netvsc_recv(struct hn_rx_ring *rxr, netvsc_packet *packet,
1279 const rndis_tcp_ip_csum_info *csum_info,
1280 const struct rndis_hash_info *hash_info,
1281 const struct rndis_hash_value *hash_value)
1283 struct ifnet *ifp = rxr->hn_ifp;
1285 int size, do_lro = 0, do_csum = 1;
1286 int hash_type = M_HASHTYPE_OPAQUE_HASH;
1288 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1292 * Bail out if packet contains more data than configured MTU.
1294 if (packet->tot_data_buf_len > (ifp->if_mtu + ETHER_HDR_LEN)) {
1296 } else if (packet->tot_data_buf_len <= MHLEN) {
1297 m_new = m_gethdr(M_NOWAIT, MT_DATA);
1298 if (m_new == NULL) {
1299 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1302 memcpy(mtod(m_new, void *), packet->data,
1303 packet->tot_data_buf_len);
1304 m_new->m_pkthdr.len = m_new->m_len = packet->tot_data_buf_len;
1305 rxr->hn_small_pkts++;
1308 * Get an mbuf with a cluster. For packets 2K or less,
1309 * get a standard 2K cluster. For anything larger, get a
1310 * 4K cluster. Any buffers larger than 4K can cause problems
1311 * if looped around to the Hyper-V TX channel, so avoid them.
1314 if (packet->tot_data_buf_len > MCLBYTES) {
1316 size = MJUMPAGESIZE;
1319 m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, size);
1320 if (m_new == NULL) {
1321 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1325 hv_m_append(m_new, packet->tot_data_buf_len, packet->data);
1327 m_new->m_pkthdr.rcvif = ifp;
1329 if (__predict_false((ifp->if_capenable & IFCAP_RXCSUM) == 0))
1332 /* receive side checksum offload */
1333 if (csum_info != NULL) {
1334 /* IP csum offload */
1335 if (csum_info->receive.ip_csum_succeeded && do_csum) {
1336 m_new->m_pkthdr.csum_flags |=
1337 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1341 /* TCP/UDP csum offload */
1342 if ((csum_info->receive.tcp_csum_succeeded ||
1343 csum_info->receive.udp_csum_succeeded) && do_csum) {
1344 m_new->m_pkthdr.csum_flags |=
1345 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1346 m_new->m_pkthdr.csum_data = 0xffff;
1347 if (csum_info->receive.tcp_csum_succeeded)
1353 if (csum_info->receive.ip_csum_succeeded &&
1354 csum_info->receive.tcp_csum_succeeded)
1357 const struct ether_header *eh;
1362 if (m_new->m_len < hoff)
1364 eh = mtod(m_new, struct ether_header *);
1365 etype = ntohs(eh->ether_type);
1366 if (etype == ETHERTYPE_VLAN) {
1367 const struct ether_vlan_header *evl;
1369 hoff = sizeof(*evl);
1370 if (m_new->m_len < hoff)
1372 evl = mtod(m_new, struct ether_vlan_header *);
1373 etype = ntohs(evl->evl_proto);
1376 if (etype == ETHERTYPE_IP) {
1379 pr = hn_check_iplen(m_new, hoff);
1380 if (pr == IPPROTO_TCP) {
1382 (rxr->hn_trust_hcsum &
1383 HN_TRUST_HCSUM_TCP)) {
1384 rxr->hn_csum_trusted++;
1385 m_new->m_pkthdr.csum_flags |=
1386 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1387 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1388 m_new->m_pkthdr.csum_data = 0xffff;
1391 } else if (pr == IPPROTO_UDP) {
1393 (rxr->hn_trust_hcsum &
1394 HN_TRUST_HCSUM_UDP)) {
1395 rxr->hn_csum_trusted++;
1396 m_new->m_pkthdr.csum_flags |=
1397 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1398 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1399 m_new->m_pkthdr.csum_data = 0xffff;
1401 } else if (pr != IPPROTO_DONE && do_csum &&
1402 (rxr->hn_trust_hcsum & HN_TRUST_HCSUM_IP)) {
1403 rxr->hn_csum_trusted++;
1404 m_new->m_pkthdr.csum_flags |=
1405 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1410 if ((packet->vlan_tci != 0) &&
1411 (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0) {
1412 m_new->m_pkthdr.ether_vtag = packet->vlan_tci;
1413 m_new->m_flags |= M_VLANTAG;
1416 if (hash_info != NULL && hash_value != NULL) {
1418 m_new->m_pkthdr.flowid = hash_value->hash_value;
1419 if ((hash_info->hash_info & NDIS_HASH_FUNCTION_MASK) ==
1420 NDIS_HASH_FUNCTION_TOEPLITZ) {
1422 (hash_info->hash_info & NDIS_HASH_TYPE_MASK);
1425 case NDIS_HASH_IPV4:
1426 hash_type = M_HASHTYPE_RSS_IPV4;
1429 case NDIS_HASH_TCP_IPV4:
1430 hash_type = M_HASHTYPE_RSS_TCP_IPV4;
1433 case NDIS_HASH_IPV6:
1434 hash_type = M_HASHTYPE_RSS_IPV6;
1437 case NDIS_HASH_IPV6_EX:
1438 hash_type = M_HASHTYPE_RSS_IPV6_EX;
1441 case NDIS_HASH_TCP_IPV6:
1442 hash_type = M_HASHTYPE_RSS_TCP_IPV6;
1445 case NDIS_HASH_TCP_IPV6_EX:
1446 hash_type = M_HASHTYPE_RSS_TCP_IPV6_EX;
1451 if (hash_value != NULL) {
1452 m_new->m_pkthdr.flowid = hash_value->hash_value;
1454 m_new->m_pkthdr.flowid = rxr->hn_rx_idx;
1455 hash_type = M_HASHTYPE_OPAQUE;
1458 M_HASHTYPE_SET(m_new, hash_type);
1461 * Note: Moved RX completion back to hv_nv_on_receive() so all
1462 * messages (not just data messages) will trigger a response.
1465 if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1468 if ((ifp->if_capenable & IFCAP_LRO) && do_lro) {
1469 #if defined(INET) || defined(INET6)
1470 struct lro_ctrl *lro = &rxr->hn_lro;
1473 rxr->hn_lro_tried++;
1474 if (hn_lro_rx(lro, m_new) == 0) {
1482 /* We're not holding the lock here, so don't release it */
1483 (*ifp->if_input)(ifp, m_new);
1489 * Rules for using sc->temp_unusable:
1490 * 1. sc->temp_unusable can only be read or written while holding NV_LOCK()
1491 * 2. code reading sc->temp_unusable under NV_LOCK(), and finding
1492 * sc->temp_unusable set, must release NV_LOCK() and exit
1493 * 3. to retain exclusive control of the interface,
1494 * sc->temp_unusable must be set by code before releasing NV_LOCK()
1495 * 4. only code setting sc->temp_unusable can clear sc->temp_unusable
1496 * 5. code setting sc->temp_unusable must eventually clear sc->temp_unusable
1500 * Standard ioctl entry point. Called when the user wants to configure
1504 hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1506 hn_softc_t *sc = ifp->if_softc;
1507 struct ifreq *ifr = (struct ifreq *)data;
1509 struct ifaddr *ifa = (struct ifaddr *)data;
1511 netvsc_device_info device_info;
1512 int mask, error = 0;
1513 int retry_cnt = 500;
1519 if (ifa->ifa_addr->sa_family == AF_INET) {
1520 ifp->if_flags |= IFF_UP;
1521 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1523 arp_ifinit(ifp, ifa);
1526 error = ether_ioctl(ifp, cmd, data);
1529 /* Check MTU value change */
1530 if (ifp->if_mtu == ifr->ifr_mtu)
1533 if (ifr->ifr_mtu > NETVSC_MAX_CONFIGURABLE_MTU) {
1538 /* Obtain and record requested MTU */
1539 ifp->if_mtu = ifr->ifr_mtu;
1541 #if __FreeBSD_version >= 1100099
1543 * Make sure that LRO aggregation length limit is still
1544 * valid, after the MTU change.
1547 if (sc->hn_rx_ring[0].hn_lro.lro_length_lim <
1548 HN_LRO_LENLIM_MIN(ifp))
1549 hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MIN(ifp));
1555 if (!sc->temp_unusable) {
1556 sc->temp_unusable = TRUE;
1560 if (retry_cnt > 0) {
1564 } while (retry_cnt > 0);
1566 if (retry_cnt == 0) {
1571 /* We must remove and add back the device to cause the new
1572 * MTU to take effect. This includes tearing down, but not
1573 * deleting the channel, then bringing it back up.
1575 error = hv_rf_on_device_remove(sc, HV_RF_NV_RETAIN_CHANNEL);
1578 sc->temp_unusable = FALSE;
1583 /* Wait for subchannels to be destroyed */
1584 vmbus_subchan_drain(sc->hn_prichan);
1586 error = hv_rf_on_device_add(sc, &device_info,
1587 sc->hn_rx_ring_inuse, &sc->hn_rx_ring[0]);
1590 sc->temp_unusable = FALSE;
1594 KASSERT(sc->hn_rx_ring_cnt == sc->net_dev->num_channel,
1595 ("RX ring count %d and channel count %u mismatch",
1596 sc->hn_rx_ring_cnt, sc->net_dev->num_channel));
1597 if (sc->net_dev->num_channel > 1) {
1601 * Skip the rings on primary channel; they are
1602 * handled by the hv_rf_on_device_add() above.
1604 for (r = 1; r < sc->hn_rx_ring_cnt; ++r) {
1605 sc->hn_rx_ring[r].hn_rx_flags &=
1606 ~HN_RX_FLAG_ATTACHED;
1608 for (r = 1; r < sc->hn_tx_ring_cnt; ++r) {
1609 sc->hn_tx_ring[r].hn_tx_flags &=
1610 ~HN_TX_FLAG_ATTACHED;
1612 hn_subchan_setup(sc);
1615 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
1616 if (sc->hn_tx_ring[0].hn_tx_chimney_size >
1617 sc->hn_tx_chimney_max)
1618 hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max);
1620 hn_ifinit_locked(sc);
1623 sc->temp_unusable = FALSE;
1629 if (!sc->temp_unusable) {
1630 sc->temp_unusable = TRUE;
1634 if (retry_cnt > 0) {
1638 } while (retry_cnt > 0);
1640 if (retry_cnt == 0) {
1645 if (ifp->if_flags & IFF_UP) {
1647 * If only the state of the PROMISC flag changed,
1648 * then just use the 'set promisc mode' command
1649 * instead of reinitializing the entire NIC. Doing
1650 * a full re-init means reloading the firmware and
1651 * waiting for it to start up, which may take a
1655 /* Fixme: Promiscuous mode? */
1656 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1657 ifp->if_flags & IFF_PROMISC &&
1658 !(sc->hn_if_flags & IFF_PROMISC)) {
1659 /* do something here for Hyper-V */
1660 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1661 !(ifp->if_flags & IFF_PROMISC) &&
1662 sc->hn_if_flags & IFF_PROMISC) {
1663 /* do something here for Hyper-V */
1666 hn_ifinit_locked(sc);
1668 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1673 sc->temp_unusable = FALSE;
1675 sc->hn_if_flags = ifp->if_flags;
1681 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1682 if (mask & IFCAP_TXCSUM) {
1683 ifp->if_capenable ^= IFCAP_TXCSUM;
1684 if (ifp->if_capenable & IFCAP_TXCSUM) {
1686 sc->hn_tx_ring[0].hn_csum_assist;
1689 ~sc->hn_tx_ring[0].hn_csum_assist;
1693 if (mask & IFCAP_RXCSUM)
1694 ifp->if_capenable ^= IFCAP_RXCSUM;
1696 if (mask & IFCAP_LRO)
1697 ifp->if_capenable ^= IFCAP_LRO;
1699 if (mask & IFCAP_TSO4) {
1700 ifp->if_capenable ^= IFCAP_TSO4;
1701 if (ifp->if_capenable & IFCAP_TSO4)
1702 ifp->if_hwassist |= CSUM_IP_TSO;
1704 ifp->if_hwassist &= ~CSUM_IP_TSO;
1707 if (mask & IFCAP_TSO6) {
1708 ifp->if_capenable ^= IFCAP_TSO6;
1709 if (ifp->if_capenable & IFCAP_TSO6)
1710 ifp->if_hwassist |= CSUM_IP6_TSO;
1712 ifp->if_hwassist &= ~CSUM_IP6_TSO;
1721 /* Fixme: Multicast mode? */
1722 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1724 netvsc_setmulti(sc);
1733 error = ifmedia_ioctl(ifp, ifr, &sc->hn_media, cmd);
1736 error = ether_ioctl(ifp, cmd, data);
1747 hn_stop(hn_softc_t *sc)
1755 printf(" Closing Device ...\n");
1757 atomic_clear_int(&ifp->if_drv_flags,
1758 (IFF_DRV_RUNNING | IFF_DRV_OACTIVE));
1759 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
1760 sc->hn_tx_ring[i].hn_oactive = 0;
1762 if_link_state_change(ifp, LINK_STATE_DOWN);
1763 sc->hn_initdone = 0;
1765 ret = hv_rf_on_close(sc);
1769 * FreeBSD transmit entry point
1772 hn_start(struct ifnet *ifp)
1774 struct hn_softc *sc = ifp->if_softc;
1775 struct hn_tx_ring *txr = &sc->hn_tx_ring[0];
1777 if (txr->hn_sched_tx)
1780 if (mtx_trylock(&txr->hn_tx_lock)) {
1783 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1784 mtx_unlock(&txr->hn_tx_lock);
1789 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
1793 hn_start_txeof(struct hn_tx_ring *txr)
1795 struct hn_softc *sc = txr->hn_sc;
1796 struct ifnet *ifp = sc->hn_ifp;
1798 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1800 if (txr->hn_sched_tx)
1803 if (mtx_trylock(&txr->hn_tx_lock)) {
1806 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1807 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1808 mtx_unlock(&txr->hn_tx_lock);
1810 taskqueue_enqueue(txr->hn_tx_taskq,
1816 * Release the OACTIVE earlier, with the hope, that
1817 * others could catch up. The task will clear the
1818 * flag again with the hn_tx_lock to avoid possible
1821 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1822 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
1830 hn_ifinit_locked(hn_softc_t *sc)
1837 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1841 hv_promisc_mode = 1;
1843 ret = hv_rf_on_open(sc);
1847 sc->hn_initdone = 1;
1850 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1851 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
1852 sc->hn_tx_ring[i].hn_oactive = 0;
1854 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_RUNNING);
1855 if_link_state_change(ifp, LINK_STATE_UP);
1862 hn_ifinit(void *xsc)
1864 hn_softc_t *sc = xsc;
1867 if (sc->temp_unusable) {
1871 sc->temp_unusable = TRUE;
1874 hn_ifinit_locked(sc);
1877 sc->temp_unusable = FALSE;
1886 hn_watchdog(struct ifnet *ifp)
1891 printf("hn%d: watchdog timeout -- resetting\n", sc->hn_unit);
1892 hn_ifinit(sc); /*???*/
1893 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1897 #if __FreeBSD_version >= 1100099
1900 hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS)
1902 struct hn_softc *sc = arg1;
1903 unsigned int lenlim;
1906 lenlim = sc->hn_rx_ring[0].hn_lro.lro_length_lim;
1907 error = sysctl_handle_int(oidp, &lenlim, 0, req);
1908 if (error || req->newptr == NULL)
1911 if (lenlim < HN_LRO_LENLIM_MIN(sc->hn_ifp) ||
1912 lenlim > TCP_LRO_LENGTH_MAX)
1916 hn_set_lro_lenlim(sc, lenlim);
1922 hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS)
1924 struct hn_softc *sc = arg1;
1925 int ackcnt, error, i;
1928 * lro_ackcnt_lim is append count limit,
1929 * +1 to turn it into aggregation limit.
1931 ackcnt = sc->hn_rx_ring[0].hn_lro.lro_ackcnt_lim + 1;
1932 error = sysctl_handle_int(oidp, &ackcnt, 0, req);
1933 if (error || req->newptr == NULL)
1936 if (ackcnt < 2 || ackcnt > (TCP_LRO_ACKCNT_MAX + 1))
1940 * Convert aggregation limit back to append
1945 for (i = 0; i < sc->hn_rx_ring_inuse; ++i)
1946 sc->hn_rx_ring[i].hn_lro.lro_ackcnt_lim = ackcnt;
1954 hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS)
1956 struct hn_softc *sc = arg1;
1961 if (sc->hn_rx_ring[0].hn_trust_hcsum & hcsum)
1964 error = sysctl_handle_int(oidp, &on, 0, req);
1965 if (error || req->newptr == NULL)
1969 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1970 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
1973 rxr->hn_trust_hcsum |= hcsum;
1975 rxr->hn_trust_hcsum &= ~hcsum;
1982 hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS)
1984 struct hn_softc *sc = arg1;
1985 int chimney_size, error;
1987 chimney_size = sc->hn_tx_ring[0].hn_tx_chimney_size;
1988 error = sysctl_handle_int(oidp, &chimney_size, 0, req);
1989 if (error || req->newptr == NULL)
1992 if (chimney_size > sc->hn_tx_chimney_max || chimney_size <= 0)
1995 hn_set_tx_chimney_size(sc, chimney_size);
2000 hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
2002 struct hn_softc *sc = arg1;
2003 int ofs = arg2, i, error;
2004 struct hn_rx_ring *rxr;
2008 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2009 rxr = &sc->hn_rx_ring[i];
2010 stat += *((u_long *)((uint8_t *)rxr + ofs));
2013 error = sysctl_handle_long(oidp, &stat, 0, req);
2014 if (error || req->newptr == NULL)
2017 /* Zero out this stat. */
2018 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2019 rxr = &sc->hn_rx_ring[i];
2020 *((u_long *)((uint8_t *)rxr + ofs)) = 0;
2026 hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS)
2028 struct hn_softc *sc = arg1;
2029 int ofs = arg2, i, error;
2030 struct hn_rx_ring *rxr;
2034 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2035 rxr = &sc->hn_rx_ring[i];
2036 stat += *((uint64_t *)((uint8_t *)rxr + ofs));
2039 error = sysctl_handle_64(oidp, &stat, 0, req);
2040 if (error || req->newptr == NULL)
2043 /* Zero out this stat. */
2044 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2045 rxr = &sc->hn_rx_ring[i];
2046 *((uint64_t *)((uint8_t *)rxr + ofs)) = 0;
2052 hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
2054 struct hn_softc *sc = arg1;
2055 int ofs = arg2, i, error;
2056 struct hn_tx_ring *txr;
2060 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2061 txr = &sc->hn_tx_ring[i];
2062 stat += *((u_long *)((uint8_t *)txr + ofs));
2065 error = sysctl_handle_long(oidp, &stat, 0, req);
2066 if (error || req->newptr == NULL)
2069 /* Zero out this stat. */
2070 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2071 txr = &sc->hn_tx_ring[i];
2072 *((u_long *)((uint8_t *)txr + ofs)) = 0;
2078 hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS)
2080 struct hn_softc *sc = arg1;
2081 int ofs = arg2, i, error, conf;
2082 struct hn_tx_ring *txr;
2084 txr = &sc->hn_tx_ring[0];
2085 conf = *((int *)((uint8_t *)txr + ofs));
2087 error = sysctl_handle_int(oidp, &conf, 0, req);
2088 if (error || req->newptr == NULL)
2092 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2093 txr = &sc->hn_tx_ring[i];
2094 *((int *)((uint8_t *)txr + ofs)) = conf;
2102 hn_check_iplen(const struct mbuf *m, int hoff)
2104 const struct ip *ip;
2105 int len, iphlen, iplen;
2106 const struct tcphdr *th;
2107 int thoff; /* TCP data offset */
2109 len = hoff + sizeof(struct ip);
2111 /* The packet must be at least the size of an IP header. */
2112 if (m->m_pkthdr.len < len)
2113 return IPPROTO_DONE;
2115 /* The fixed IP header must reside completely in the first mbuf. */
2117 return IPPROTO_DONE;
2119 ip = mtodo(m, hoff);
2121 /* Bound check the packet's stated IP header length. */
2122 iphlen = ip->ip_hl << 2;
2123 if (iphlen < sizeof(struct ip)) /* minimum header length */
2124 return IPPROTO_DONE;
2126 /* The full IP header must reside completely in the one mbuf. */
2127 if (m->m_len < hoff + iphlen)
2128 return IPPROTO_DONE;
2130 iplen = ntohs(ip->ip_len);
2133 * Check that the amount of data in the buffers is as
2134 * at least much as the IP header would have us expect.
2136 if (m->m_pkthdr.len < hoff + iplen)
2137 return IPPROTO_DONE;
2140 * Ignore IP fragments.
2142 if (ntohs(ip->ip_off) & (IP_OFFMASK | IP_MF))
2143 return IPPROTO_DONE;
2146 * The TCP/IP or UDP/IP header must be entirely contained within
2147 * the first fragment of a packet.
2151 if (iplen < iphlen + sizeof(struct tcphdr))
2152 return IPPROTO_DONE;
2153 if (m->m_len < hoff + iphlen + sizeof(struct tcphdr))
2154 return IPPROTO_DONE;
2155 th = (const struct tcphdr *)((const uint8_t *)ip + iphlen);
2156 thoff = th->th_off << 2;
2157 if (thoff < sizeof(struct tcphdr) || thoff + iphlen > iplen)
2158 return IPPROTO_DONE;
2159 if (m->m_len < hoff + iphlen + thoff)
2160 return IPPROTO_DONE;
2163 if (iplen < iphlen + sizeof(struct udphdr))
2164 return IPPROTO_DONE;
2165 if (m->m_len < hoff + iphlen + sizeof(struct udphdr))
2166 return IPPROTO_DONE;
2170 return IPPROTO_DONE;
2177 hn_create_rx_data(struct hn_softc *sc, int ring_cnt)
2179 struct sysctl_oid_list *child;
2180 struct sysctl_ctx_list *ctx;
2181 device_t dev = sc->hn_dev;
2182 #if defined(INET) || defined(INET6)
2183 #if __FreeBSD_version >= 1100095
2189 sc->hn_rx_ring_cnt = ring_cnt;
2190 sc->hn_rx_ring_inuse = sc->hn_rx_ring_cnt;
2192 sc->hn_rx_ring = malloc(sizeof(struct hn_rx_ring) * sc->hn_rx_ring_cnt,
2193 M_NETVSC, M_WAITOK | M_ZERO);
2195 #if defined(INET) || defined(INET6)
2196 #if __FreeBSD_version >= 1100095
2197 lroent_cnt = hn_lro_entry_count;
2198 if (lroent_cnt < TCP_LRO_ENTRIES)
2199 lroent_cnt = TCP_LRO_ENTRIES;
2200 device_printf(dev, "LRO: entry count %d\n", lroent_cnt);
2202 #endif /* INET || INET6 */
2204 ctx = device_get_sysctl_ctx(dev);
2205 child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
2207 /* Create dev.hn.UNIT.rx sysctl tree */
2208 sc->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "rx",
2209 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2211 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2212 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
2214 if (hn_trust_hosttcp)
2215 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_TCP;
2216 if (hn_trust_hostudp)
2217 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_UDP;
2218 if (hn_trust_hostip)
2219 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_IP;
2220 rxr->hn_ifp = sc->hn_ifp;
2221 if (i < sc->hn_tx_ring_cnt)
2222 rxr->hn_txr = &sc->hn_tx_ring[i];
2223 rxr->hn_rdbuf = malloc(NETVSC_PACKET_SIZE, M_NETVSC, M_WAITOK);
2229 #if defined(INET) || defined(INET6)
2230 #if __FreeBSD_version >= 1100095
2231 tcp_lro_init_args(&rxr->hn_lro, sc->hn_ifp, lroent_cnt,
2232 hn_lro_mbufq_depth);
2234 tcp_lro_init(&rxr->hn_lro);
2235 rxr->hn_lro.ifp = sc->hn_ifp;
2237 #if __FreeBSD_version >= 1100099
2238 rxr->hn_lro.lro_length_lim = HN_LRO_LENLIM_DEF;
2239 rxr->hn_lro.lro_ackcnt_lim = HN_LRO_ACKCNT_DEF;
2241 #endif /* INET || INET6 */
2243 if (sc->hn_rx_sysctl_tree != NULL) {
2247 * Create per RX ring sysctl tree:
2248 * dev.hn.UNIT.rx.RINGID
2250 snprintf(name, sizeof(name), "%d", i);
2251 rxr->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx,
2252 SYSCTL_CHILDREN(sc->hn_rx_sysctl_tree),
2253 OID_AUTO, name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2255 if (rxr->hn_rx_sysctl_tree != NULL) {
2256 SYSCTL_ADD_ULONG(ctx,
2257 SYSCTL_CHILDREN(rxr->hn_rx_sysctl_tree),
2258 OID_AUTO, "packets", CTLFLAG_RW,
2259 &rxr->hn_pkts, "# of packets received");
2260 SYSCTL_ADD_ULONG(ctx,
2261 SYSCTL_CHILDREN(rxr->hn_rx_sysctl_tree),
2262 OID_AUTO, "rss_pkts", CTLFLAG_RW,
2264 "# of packets w/ RSS info received");
2269 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_queued",
2270 CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2271 __offsetof(struct hn_rx_ring, hn_lro.lro_queued),
2272 hn_rx_stat_u64_sysctl, "LU", "LRO queued");
2273 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_flushed",
2274 CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2275 __offsetof(struct hn_rx_ring, hn_lro.lro_flushed),
2276 hn_rx_stat_u64_sysctl, "LU", "LRO flushed");
2277 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_tried",
2278 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2279 __offsetof(struct hn_rx_ring, hn_lro_tried),
2280 hn_rx_stat_ulong_sysctl, "LU", "# of LRO tries");
2281 #if __FreeBSD_version >= 1100099
2282 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_length_lim",
2283 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2284 hn_lro_lenlim_sysctl, "IU",
2285 "Max # of data bytes to be aggregated by LRO");
2286 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_ackcnt_lim",
2287 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2288 hn_lro_ackcnt_sysctl, "I",
2289 "Max # of ACKs to be aggregated by LRO");
2291 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hosttcp",
2292 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_TCP,
2293 hn_trust_hcsum_sysctl, "I",
2294 "Trust tcp segement verification on host side, "
2295 "when csum info is missing");
2296 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostudp",
2297 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_UDP,
2298 hn_trust_hcsum_sysctl, "I",
2299 "Trust udp datagram verification on host side, "
2300 "when csum info is missing");
2301 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostip",
2302 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_IP,
2303 hn_trust_hcsum_sysctl, "I",
2304 "Trust ip packet verification on host side, "
2305 "when csum info is missing");
2306 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_ip",
2307 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2308 __offsetof(struct hn_rx_ring, hn_csum_ip),
2309 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM IP");
2310 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_tcp",
2311 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2312 __offsetof(struct hn_rx_ring, hn_csum_tcp),
2313 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM TCP");
2314 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_udp",
2315 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2316 __offsetof(struct hn_rx_ring, hn_csum_udp),
2317 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM UDP");
2318 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_trusted",
2319 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2320 __offsetof(struct hn_rx_ring, hn_csum_trusted),
2321 hn_rx_stat_ulong_sysctl, "LU",
2322 "# of packets that we trust host's csum verification");
2323 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "small_pkts",
2324 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2325 __offsetof(struct hn_rx_ring, hn_small_pkts),
2326 hn_rx_stat_ulong_sysctl, "LU", "# of small packets received");
2327 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_cnt",
2328 CTLFLAG_RD, &sc->hn_rx_ring_cnt, 0, "# created RX rings");
2329 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_inuse",
2330 CTLFLAG_RD, &sc->hn_rx_ring_inuse, 0, "# used RX rings");
2334 hn_destroy_rx_data(struct hn_softc *sc)
2338 if (sc->hn_rx_ring_cnt == 0)
2341 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2342 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
2344 #if defined(INET) || defined(INET6)
2345 tcp_lro_free(&rxr->hn_lro);
2347 free(rxr->hn_rdbuf, M_NETVSC);
2349 free(sc->hn_rx_ring, M_NETVSC);
2350 sc->hn_rx_ring = NULL;
2352 sc->hn_rx_ring_cnt = 0;
2353 sc->hn_rx_ring_inuse = 0;
2357 hn_create_tx_ring(struct hn_softc *sc, int id)
2359 struct hn_tx_ring *txr = &sc->hn_tx_ring[id];
2360 device_t dev = sc->hn_dev;
2361 bus_dma_tag_t parent_dtag;
2366 txr->hn_tx_idx = id;
2368 #ifndef HN_USE_TXDESC_BUFRING
2369 mtx_init(&txr->hn_txlist_spin, "hn txlist", NULL, MTX_SPIN);
2371 mtx_init(&txr->hn_tx_lock, "hn tx", NULL, MTX_DEF);
2373 txr->hn_txdesc_cnt = HN_TX_DESC_CNT;
2374 txr->hn_txdesc = malloc(sizeof(struct hn_txdesc) * txr->hn_txdesc_cnt,
2375 M_NETVSC, M_WAITOK | M_ZERO);
2376 #ifndef HN_USE_TXDESC_BUFRING
2377 SLIST_INIT(&txr->hn_txlist);
2379 txr->hn_txdesc_br = buf_ring_alloc(txr->hn_txdesc_cnt, M_NETVSC,
2380 M_WAITOK, &txr->hn_tx_lock);
2383 txr->hn_tx_taskq = sc->hn_tx_taskq;
2385 if (hn_use_if_start) {
2386 txr->hn_txeof = hn_start_txeof;
2387 TASK_INIT(&txr->hn_tx_task, 0, hn_start_taskfunc, txr);
2388 TASK_INIT(&txr->hn_txeof_task, 0, hn_start_txeof_taskfunc, txr);
2392 txr->hn_txeof = hn_xmit_txeof;
2393 TASK_INIT(&txr->hn_tx_task, 0, hn_xmit_taskfunc, txr);
2394 TASK_INIT(&txr->hn_txeof_task, 0, hn_xmit_txeof_taskfunc, txr);
2396 br_depth = hn_get_txswq_depth(txr);
2397 txr->hn_mbuf_br = buf_ring_alloc(br_depth, M_NETVSC,
2398 M_WAITOK, &txr->hn_tx_lock);
2401 txr->hn_direct_tx_size = hn_direct_tx_size;
2402 version = VMBUS_GET_VERSION(device_get_parent(dev), dev);
2403 if (version >= VMBUS_VERSION_WIN8_1) {
2404 txr->hn_csum_assist = HN_CSUM_ASSIST;
2406 txr->hn_csum_assist = HN_CSUM_ASSIST_WIN8;
2408 device_printf(dev, "bus version %u.%u, "
2409 "no UDP checksum offloading\n",
2410 VMBUS_VERSION_MAJOR(version),
2411 VMBUS_VERSION_MINOR(version));
2416 * Always schedule transmission instead of trying to do direct
2417 * transmission. This one gives the best performance so far.
2419 txr->hn_sched_tx = 1;
2421 parent_dtag = bus_get_dma_tag(dev);
2423 /* DMA tag for RNDIS messages. */
2424 error = bus_dma_tag_create(parent_dtag, /* parent */
2425 HN_RNDIS_MSG_ALIGN, /* alignment */
2426 HN_RNDIS_MSG_BOUNDARY, /* boundary */
2427 BUS_SPACE_MAXADDR, /* lowaddr */
2428 BUS_SPACE_MAXADDR, /* highaddr */
2429 NULL, NULL, /* filter, filterarg */
2430 HN_RNDIS_MSG_LEN, /* maxsize */
2432 HN_RNDIS_MSG_LEN, /* maxsegsize */
2434 NULL, /* lockfunc */
2435 NULL, /* lockfuncarg */
2436 &txr->hn_tx_rndis_dtag);
2438 device_printf(dev, "failed to create rndis dmatag\n");
2442 /* DMA tag for data. */
2443 error = bus_dma_tag_create(parent_dtag, /* parent */
2445 HN_TX_DATA_BOUNDARY, /* boundary */
2446 BUS_SPACE_MAXADDR, /* lowaddr */
2447 BUS_SPACE_MAXADDR, /* highaddr */
2448 NULL, NULL, /* filter, filterarg */
2449 HN_TX_DATA_MAXSIZE, /* maxsize */
2450 HN_TX_DATA_SEGCNT_MAX, /* nsegments */
2451 HN_TX_DATA_SEGSIZE, /* maxsegsize */
2453 NULL, /* lockfunc */
2454 NULL, /* lockfuncarg */
2455 &txr->hn_tx_data_dtag);
2457 device_printf(dev, "failed to create data dmatag\n");
2461 for (i = 0; i < txr->hn_txdesc_cnt; ++i) {
2462 struct hn_txdesc *txd = &txr->hn_txdesc[i];
2467 * Allocate and load RNDIS messages.
2469 error = bus_dmamem_alloc(txr->hn_tx_rndis_dtag,
2470 (void **)&txd->rndis_msg,
2471 BUS_DMA_WAITOK | BUS_DMA_COHERENT,
2472 &txd->rndis_msg_dmap);
2475 "failed to allocate rndis_msg, %d\n", i);
2479 error = bus_dmamap_load(txr->hn_tx_rndis_dtag,
2480 txd->rndis_msg_dmap,
2481 txd->rndis_msg, HN_RNDIS_MSG_LEN,
2482 hyperv_dma_map_paddr, &txd->rndis_msg_paddr,
2486 "failed to load rndis_msg, %d\n", i);
2487 bus_dmamem_free(txr->hn_tx_rndis_dtag,
2488 txd->rndis_msg, txd->rndis_msg_dmap);
2492 /* DMA map for TX data. */
2493 error = bus_dmamap_create(txr->hn_tx_data_dtag, 0,
2497 "failed to allocate tx data dmamap\n");
2498 bus_dmamap_unload(txr->hn_tx_rndis_dtag,
2499 txd->rndis_msg_dmap);
2500 bus_dmamem_free(txr->hn_tx_rndis_dtag,
2501 txd->rndis_msg, txd->rndis_msg_dmap);
2505 /* All set, put it to list */
2506 txd->flags |= HN_TXD_FLAG_ONLIST;
2507 #ifndef HN_USE_TXDESC_BUFRING
2508 SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
2510 buf_ring_enqueue(txr->hn_txdesc_br, txd);
2513 txr->hn_txdesc_avail = txr->hn_txdesc_cnt;
2515 if (sc->hn_tx_sysctl_tree != NULL) {
2516 struct sysctl_oid_list *child;
2517 struct sysctl_ctx_list *ctx;
2521 * Create per TX ring sysctl tree:
2522 * dev.hn.UNIT.tx.RINGID
2524 ctx = device_get_sysctl_ctx(dev);
2525 child = SYSCTL_CHILDREN(sc->hn_tx_sysctl_tree);
2527 snprintf(name, sizeof(name), "%d", id);
2528 txr->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO,
2529 name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2531 if (txr->hn_tx_sysctl_tree != NULL) {
2532 child = SYSCTL_CHILDREN(txr->hn_tx_sysctl_tree);
2534 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_avail",
2535 CTLFLAG_RD, &txr->hn_txdesc_avail, 0,
2536 "# of available TX descs");
2537 if (!hn_use_if_start) {
2538 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "oactive",
2539 CTLFLAG_RD, &txr->hn_oactive, 0,
2542 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "packets",
2543 CTLFLAG_RW, &txr->hn_pkts,
2544 "# of packets transmitted");
2552 hn_txdesc_dmamap_destroy(struct hn_txdesc *txd)
2554 struct hn_tx_ring *txr = txd->txr;
2556 KASSERT(txd->m == NULL, ("still has mbuf installed"));
2557 KASSERT((txd->flags & HN_TXD_FLAG_DMAMAP) == 0, ("still dma mapped"));
2559 bus_dmamap_unload(txr->hn_tx_rndis_dtag, txd->rndis_msg_dmap);
2560 bus_dmamem_free(txr->hn_tx_rndis_dtag, txd->rndis_msg,
2561 txd->rndis_msg_dmap);
2562 bus_dmamap_destroy(txr->hn_tx_data_dtag, txd->data_dmap);
2566 hn_destroy_tx_ring(struct hn_tx_ring *txr)
2568 struct hn_txdesc *txd;
2570 if (txr->hn_txdesc == NULL)
2573 #ifndef HN_USE_TXDESC_BUFRING
2574 while ((txd = SLIST_FIRST(&txr->hn_txlist)) != NULL) {
2575 SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
2576 hn_txdesc_dmamap_destroy(txd);
2579 mtx_lock(&txr->hn_tx_lock);
2580 while ((txd = buf_ring_dequeue_sc(txr->hn_txdesc_br)) != NULL)
2581 hn_txdesc_dmamap_destroy(txd);
2582 mtx_unlock(&txr->hn_tx_lock);
2585 if (txr->hn_tx_data_dtag != NULL)
2586 bus_dma_tag_destroy(txr->hn_tx_data_dtag);
2587 if (txr->hn_tx_rndis_dtag != NULL)
2588 bus_dma_tag_destroy(txr->hn_tx_rndis_dtag);
2590 #ifdef HN_USE_TXDESC_BUFRING
2591 buf_ring_free(txr->hn_txdesc_br, M_NETVSC);
2594 free(txr->hn_txdesc, M_NETVSC);
2595 txr->hn_txdesc = NULL;
2597 if (txr->hn_mbuf_br != NULL)
2598 buf_ring_free(txr->hn_mbuf_br, M_NETVSC);
2600 #ifndef HN_USE_TXDESC_BUFRING
2601 mtx_destroy(&txr->hn_txlist_spin);
2603 mtx_destroy(&txr->hn_tx_lock);
2607 hn_create_tx_data(struct hn_softc *sc, int ring_cnt)
2609 struct sysctl_oid_list *child;
2610 struct sysctl_ctx_list *ctx;
2613 sc->hn_tx_ring_cnt = ring_cnt;
2614 sc->hn_tx_ring_inuse = sc->hn_tx_ring_cnt;
2616 sc->hn_tx_ring = malloc(sizeof(struct hn_tx_ring) * sc->hn_tx_ring_cnt,
2617 M_NETVSC, M_WAITOK | M_ZERO);
2619 ctx = device_get_sysctl_ctx(sc->hn_dev);
2620 child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->hn_dev));
2622 /* Create dev.hn.UNIT.tx sysctl tree */
2623 sc->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "tx",
2624 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2626 for (i = 0; i < sc->hn_tx_ring_cnt; ++i) {
2629 error = hn_create_tx_ring(sc, i);
2634 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "no_txdescs",
2635 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2636 __offsetof(struct hn_tx_ring, hn_no_txdescs),
2637 hn_tx_stat_ulong_sysctl, "LU", "# of times short of TX descs");
2638 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "send_failed",
2639 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2640 __offsetof(struct hn_tx_ring, hn_send_failed),
2641 hn_tx_stat_ulong_sysctl, "LU", "# of hyper-v sending failure");
2642 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "txdma_failed",
2643 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2644 __offsetof(struct hn_tx_ring, hn_txdma_failed),
2645 hn_tx_stat_ulong_sysctl, "LU", "# of TX DMA failure");
2646 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_collapsed",
2647 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2648 __offsetof(struct hn_tx_ring, hn_tx_collapsed),
2649 hn_tx_stat_ulong_sysctl, "LU", "# of TX mbuf collapsed");
2650 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney",
2651 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2652 __offsetof(struct hn_tx_ring, hn_tx_chimney),
2653 hn_tx_stat_ulong_sysctl, "LU", "# of chimney send");
2654 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_tried",
2655 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2656 __offsetof(struct hn_tx_ring, hn_tx_chimney_tried),
2657 hn_tx_stat_ulong_sysctl, "LU", "# of chimney send tries");
2658 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_cnt",
2659 CTLFLAG_RD, &sc->hn_tx_ring[0].hn_txdesc_cnt, 0,
2660 "# of total TX descs");
2661 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_chimney_max",
2662 CTLFLAG_RD, &sc->hn_tx_chimney_max, 0,
2663 "Chimney send packet size upper boundary");
2664 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_size",
2665 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2666 hn_tx_chimney_size_sysctl,
2667 "I", "Chimney send packet size limit");
2668 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "direct_tx_size",
2669 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2670 __offsetof(struct hn_tx_ring, hn_direct_tx_size),
2671 hn_tx_conf_int_sysctl, "I",
2672 "Size of the packet for direct transmission");
2673 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "sched_tx",
2674 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2675 __offsetof(struct hn_tx_ring, hn_sched_tx),
2676 hn_tx_conf_int_sysctl, "I",
2677 "Always schedule transmission "
2678 "instead of doing direct transmission");
2679 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_cnt",
2680 CTLFLAG_RD, &sc->hn_tx_ring_cnt, 0, "# created TX rings");
2681 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_inuse",
2682 CTLFLAG_RD, &sc->hn_tx_ring_inuse, 0, "# used TX rings");
2688 hn_set_tx_chimney_size(struct hn_softc *sc, int chimney_size)
2693 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
2694 sc->hn_tx_ring[i].hn_tx_chimney_size = chimney_size;
2699 hn_destroy_tx_data(struct hn_softc *sc)
2703 if (sc->hn_tx_ring_cnt == 0)
2706 for (i = 0; i < sc->hn_tx_ring_cnt; ++i)
2707 hn_destroy_tx_ring(&sc->hn_tx_ring[i]);
2709 free(sc->hn_tx_ring, M_NETVSC);
2710 sc->hn_tx_ring = NULL;
2712 sc->hn_tx_ring_cnt = 0;
2713 sc->hn_tx_ring_inuse = 0;
2717 hn_start_taskfunc(void *xtxr, int pending __unused)
2719 struct hn_tx_ring *txr = xtxr;
2721 mtx_lock(&txr->hn_tx_lock);
2722 hn_start_locked(txr, 0);
2723 mtx_unlock(&txr->hn_tx_lock);
2727 hn_start_txeof_taskfunc(void *xtxr, int pending __unused)
2729 struct hn_tx_ring *txr = xtxr;
2731 mtx_lock(&txr->hn_tx_lock);
2732 atomic_clear_int(&txr->hn_sc->hn_ifp->if_drv_flags, IFF_DRV_OACTIVE);
2733 hn_start_locked(txr, 0);
2734 mtx_unlock(&txr->hn_tx_lock);
2738 hn_stop_tx_tasks(struct hn_softc *sc)
2742 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2743 struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
2745 taskqueue_drain(txr->hn_tx_taskq, &txr->hn_tx_task);
2746 taskqueue_drain(txr->hn_tx_taskq, &txr->hn_txeof_task);
2751 hn_xmit(struct hn_tx_ring *txr, int len)
2753 struct hn_softc *sc = txr->hn_sc;
2754 struct ifnet *ifp = sc->hn_ifp;
2755 struct mbuf *m_head;
2757 mtx_assert(&txr->hn_tx_lock, MA_OWNED);
2758 KASSERT(hn_use_if_start == 0,
2759 ("hn_xmit is called, when if_start is enabled"));
2761 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || txr->hn_oactive)
2764 while ((m_head = drbr_peek(ifp, txr->hn_mbuf_br)) != NULL) {
2765 struct hn_txdesc *txd;
2768 if (len > 0 && m_head->m_pkthdr.len > len) {
2770 * This sending could be time consuming; let callers
2771 * dispatch this packet sending (and sending of any
2772 * following up packets) to tx taskqueue.
2774 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2778 txd = hn_txdesc_get(txr);
2780 txr->hn_no_txdescs++;
2781 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2782 txr->hn_oactive = 1;
2786 error = hn_encap(txr, txd, &m_head);
2788 /* Both txd and m_head are freed; discard */
2789 drbr_advance(ifp, txr->hn_mbuf_br);
2793 error = hn_send_pkt(ifp, txr, txd);
2794 if (__predict_false(error)) {
2795 /* txd is freed, but m_head is not */
2796 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2797 txr->hn_oactive = 1;
2802 drbr_advance(ifp, txr->hn_mbuf_br);
2808 hn_transmit(struct ifnet *ifp, struct mbuf *m)
2810 struct hn_softc *sc = ifp->if_softc;
2811 struct hn_tx_ring *txr;
2815 * Select the TX ring based on flowid
2817 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2818 idx = m->m_pkthdr.flowid % sc->hn_tx_ring_inuse;
2819 txr = &sc->hn_tx_ring[idx];
2821 error = drbr_enqueue(ifp, txr->hn_mbuf_br, m);
2823 if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1);
2827 if (txr->hn_oactive)
2830 if (txr->hn_sched_tx)
2833 if (mtx_trylock(&txr->hn_tx_lock)) {
2836 sched = hn_xmit(txr, txr->hn_direct_tx_size);
2837 mtx_unlock(&txr->hn_tx_lock);
2842 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
2847 hn_xmit_qflush(struct ifnet *ifp)
2849 struct hn_softc *sc = ifp->if_softc;
2852 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2853 struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
2856 mtx_lock(&txr->hn_tx_lock);
2857 while ((m = buf_ring_dequeue_sc(txr->hn_mbuf_br)) != NULL)
2859 mtx_unlock(&txr->hn_tx_lock);
2865 hn_xmit_txeof(struct hn_tx_ring *txr)
2868 if (txr->hn_sched_tx)
2871 if (mtx_trylock(&txr->hn_tx_lock)) {
2874 txr->hn_oactive = 0;
2875 sched = hn_xmit(txr, txr->hn_direct_tx_size);
2876 mtx_unlock(&txr->hn_tx_lock);
2878 taskqueue_enqueue(txr->hn_tx_taskq,
2884 * Release the oactive earlier, with the hope, that
2885 * others could catch up. The task will clear the
2886 * oactive again with the hn_tx_lock to avoid possible
2889 txr->hn_oactive = 0;
2890 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
2895 hn_xmit_taskfunc(void *xtxr, int pending __unused)
2897 struct hn_tx_ring *txr = xtxr;
2899 mtx_lock(&txr->hn_tx_lock);
2901 mtx_unlock(&txr->hn_tx_lock);
2905 hn_xmit_txeof_taskfunc(void *xtxr, int pending __unused)
2907 struct hn_tx_ring *txr = xtxr;
2909 mtx_lock(&txr->hn_tx_lock);
2910 txr->hn_oactive = 0;
2912 mtx_unlock(&txr->hn_tx_lock);
2916 hn_channel_attach(struct hn_softc *sc, struct vmbus_channel *chan)
2918 struct hn_rx_ring *rxr;
2921 idx = vmbus_chan_subidx(chan);
2923 KASSERT(idx >= 0 && idx < sc->hn_rx_ring_inuse,
2924 ("invalid channel index %d, should > 0 && < %d",
2925 idx, sc->hn_rx_ring_inuse));
2926 rxr = &sc->hn_rx_ring[idx];
2927 KASSERT((rxr->hn_rx_flags & HN_RX_FLAG_ATTACHED) == 0,
2928 ("RX ring %d already attached", idx));
2929 rxr->hn_rx_flags |= HN_RX_FLAG_ATTACHED;
2932 if_printf(sc->hn_ifp, "link RX ring %d to channel%u\n",
2933 idx, vmbus_chan_id(chan));
2936 if (idx < sc->hn_tx_ring_inuse) {
2937 struct hn_tx_ring *txr = &sc->hn_tx_ring[idx];
2939 KASSERT((txr->hn_tx_flags & HN_TX_FLAG_ATTACHED) == 0,
2940 ("TX ring %d already attached", idx));
2941 txr->hn_tx_flags |= HN_TX_FLAG_ATTACHED;
2943 txr->hn_chan = chan;
2945 if_printf(sc->hn_ifp, "link TX ring %d to channel%u\n",
2946 idx, vmbus_chan_id(chan));
2950 /* Bind channel to a proper CPU */
2951 vmbus_chan_cpu_set(chan, (sc->hn_cpu + idx) % mp_ncpus);
2955 hn_subchan_attach(struct hn_softc *sc, struct vmbus_channel *chan)
2958 KASSERT(!vmbus_chan_is_primary(chan),
2959 ("subchannel callback on primary channel"));
2960 hn_channel_attach(sc, chan);
2964 hn_subchan_setup(struct hn_softc *sc)
2966 struct vmbus_channel **subchans;
2967 int subchan_cnt = sc->net_dev->num_channel - 1;
2970 /* Wait for sub-channels setup to complete. */
2971 subchans = vmbus_subchan_get(sc->hn_prichan, subchan_cnt);
2973 /* Attach the sub-channels. */
2974 for (i = 0; i < subchan_cnt; ++i) {
2975 struct vmbus_channel *subchan = subchans[i];
2977 /* NOTE: Calling order is critical. */
2978 hn_subchan_attach(sc, subchan);
2979 hv_nv_subchan_attach(subchan,
2980 &sc->hn_rx_ring[vmbus_chan_subidx(subchan)]);
2983 /* Release the sub-channels */
2984 vmbus_subchan_rel(subchans, subchan_cnt);
2985 if_printf(sc->hn_ifp, "%d sub-channels setup done\n", subchan_cnt);
2989 hn_tx_taskq_create(void *arg __unused)
2991 if (!hn_share_tx_taskq)
2994 hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
2995 taskqueue_thread_enqueue, &hn_tx_taskq);
2996 if (hn_bind_tx_taskq >= 0) {
2997 int cpu = hn_bind_tx_taskq;
3000 if (cpu > mp_ncpus - 1)
3002 CPU_SETOF(cpu, &cpu_set);
3003 taskqueue_start_threads_cpuset(&hn_tx_taskq, 1, PI_NET,
3006 taskqueue_start_threads(&hn_tx_taskq, 1, PI_NET, "hn tx");
3009 SYSINIT(hn_txtq_create, SI_SUB_DRIVERS, SI_ORDER_FIRST,
3010 hn_tx_taskq_create, NULL);
3013 hn_tx_taskq_destroy(void *arg __unused)
3015 if (hn_tx_taskq != NULL)
3016 taskqueue_free(hn_tx_taskq);
3018 SYSUNINIT(hn_txtq_destroy, SI_SUB_DRIVERS, SI_ORDER_FIRST,
3019 hn_tx_taskq_destroy, NULL);
3021 static device_method_t netvsc_methods[] = {
3022 /* Device interface */
3023 DEVMETHOD(device_probe, netvsc_probe),
3024 DEVMETHOD(device_attach, netvsc_attach),
3025 DEVMETHOD(device_detach, netvsc_detach),
3026 DEVMETHOD(device_shutdown, netvsc_shutdown),
3031 static driver_t netvsc_driver = {
3037 static devclass_t netvsc_devclass;
3039 DRIVER_MODULE(hn, vmbus, netvsc_driver, netvsc_devclass, 0, 0);
3040 MODULE_VERSION(hn, 1);
3041 MODULE_DEPEND(hn, vmbus, 1, 1, 1);