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 <dev/hyperv/include/hyperv_busdma.h>
118 #include "hv_net_vsc.h"
119 #include "hv_rndis.h"
120 #include "hv_rndis_filter.h"
121 #include "vmbus_if.h"
123 #define hv_chan_rxr hv_chan_priv1
124 #define hv_chan_txr hv_chan_priv2
126 /* Short for Hyper-V network interface */
127 #define NETVSC_DEVNAME "hn"
130 * It looks like offset 0 of buf is reserved to hold the softc pointer.
131 * The sc pointer evidently not needed, and is not presently populated.
132 * The packet offset is where the netvsc_packet starts in the buffer.
134 #define HV_NV_SC_PTR_OFFSET_IN_BUF 0
135 #define HV_NV_PACKET_OFFSET_IN_BUF 16
137 /* YYY should get it from the underlying channel */
138 #define HN_TX_DESC_CNT 512
140 #define HN_LROENT_CNT_DEF 128
142 #define HN_RING_CNT_DEF_MAX 8
144 #define HN_RNDIS_MSG_LEN \
145 (sizeof(rndis_msg) + \
146 RNDIS_HASHVAL_PPI_SIZE + \
147 RNDIS_VLAN_PPI_SIZE + \
148 RNDIS_TSO_PPI_SIZE + \
150 #define HN_RNDIS_MSG_BOUNDARY PAGE_SIZE
151 #define HN_RNDIS_MSG_ALIGN CACHE_LINE_SIZE
153 #define HN_TX_DATA_BOUNDARY PAGE_SIZE
154 #define HN_TX_DATA_MAXSIZE IP_MAXPACKET
155 #define HN_TX_DATA_SEGSIZE PAGE_SIZE
156 #define HN_TX_DATA_SEGCNT_MAX \
157 (VMBUS_CHAN_SGLIST_MAX - HV_RF_NUM_TX_RESERVED_PAGE_BUFS)
159 #define HN_DIRECT_TX_SIZE_DEF 128
161 #define HN_EARLY_TXEOF_THRESH 8
164 #ifndef HN_USE_TXDESC_BUFRING
165 SLIST_ENTRY(hn_txdesc) link;
168 struct hn_tx_ring *txr;
170 uint32_t flags; /* HN_TXD_FLAG_ */
171 netvsc_packet netvsc_pkt; /* XXX to be removed */
173 bus_dmamap_t data_dmap;
175 bus_addr_t rndis_msg_paddr;
176 rndis_msg *rndis_msg;
177 bus_dmamap_t rndis_msg_dmap;
180 #define HN_TXD_FLAG_ONLIST 0x1
181 #define HN_TXD_FLAG_DMAMAP 0x2
184 * Only enable UDP checksum offloading when it is on 2012R2 or
185 * later. UDP checksum offloading doesn't work on earlier
188 #define HN_CSUM_ASSIST_WIN8 (CSUM_IP | CSUM_TCP)
189 #define HN_CSUM_ASSIST (CSUM_IP | CSUM_UDP | CSUM_TCP)
191 #define HN_LRO_LENLIM_MULTIRX_DEF (12 * ETHERMTU)
192 #define HN_LRO_LENLIM_DEF (25 * ETHERMTU)
193 /* YYY 2*MTU is a bit rough, but should be good enough. */
194 #define HN_LRO_LENLIM_MIN(ifp) (2 * (ifp)->if_mtu)
196 #define HN_LRO_ACKCNT_DEF 1
199 * Be aware that this sleepable mutex will exhibit WITNESS errors when
200 * certain TCP and ARP code paths are taken. This appears to be a
201 * well-known condition, as all other drivers checked use a sleeping
202 * mutex to protect their transmit paths.
203 * Also Be aware that mutexes do not play well with semaphores, and there
204 * is a conflicting semaphore in a certain channel code path.
206 #define NV_LOCK_INIT(_sc, _name) \
207 mtx_init(&(_sc)->hn_lock, _name, MTX_NETWORK_LOCK, MTX_DEF)
208 #define NV_LOCK(_sc) mtx_lock(&(_sc)->hn_lock)
209 #define NV_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->hn_lock, MA_OWNED)
210 #define NV_UNLOCK(_sc) mtx_unlock(&(_sc)->hn_lock)
211 #define NV_LOCK_DESTROY(_sc) mtx_destroy(&(_sc)->hn_lock)
218 int hv_promisc_mode = 0; /* normal mode by default */
220 SYSCTL_NODE(_hw, OID_AUTO, hn, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
221 "Hyper-V network interface");
223 /* Trust tcp segements verification on host side. */
224 static int hn_trust_hosttcp = 1;
225 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hosttcp, CTLFLAG_RDTUN,
226 &hn_trust_hosttcp, 0,
227 "Trust tcp segement verification on host side, "
228 "when csum info is missing (global setting)");
230 /* Trust udp datagrams verification on host side. */
231 static int hn_trust_hostudp = 1;
232 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostudp, CTLFLAG_RDTUN,
233 &hn_trust_hostudp, 0,
234 "Trust udp datagram verification on host side, "
235 "when csum info is missing (global setting)");
237 /* Trust ip packets verification on host side. */
238 static int hn_trust_hostip = 1;
239 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostip, CTLFLAG_RDTUN,
241 "Trust ip packet verification on host side, "
242 "when csum info is missing (global setting)");
244 #if __FreeBSD_version >= 1100045
245 /* Limit TSO burst size */
246 static int hn_tso_maxlen = 0;
247 SYSCTL_INT(_hw_hn, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN,
248 &hn_tso_maxlen, 0, "TSO burst limit");
251 /* Limit chimney send size */
252 static int hn_tx_chimney_size = 0;
253 SYSCTL_INT(_hw_hn, OID_AUTO, tx_chimney_size, CTLFLAG_RDTUN,
254 &hn_tx_chimney_size, 0, "Chimney send packet size limit");
256 /* Limit the size of packet for direct transmission */
257 static int hn_direct_tx_size = HN_DIRECT_TX_SIZE_DEF;
258 SYSCTL_INT(_hw_hn, OID_AUTO, direct_tx_size, CTLFLAG_RDTUN,
259 &hn_direct_tx_size, 0, "Size of the packet for direct transmission");
261 #if defined(INET) || defined(INET6)
262 #if __FreeBSD_version >= 1100095
263 static int hn_lro_entry_count = HN_LROENT_CNT_DEF;
264 SYSCTL_INT(_hw_hn, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN,
265 &hn_lro_entry_count, 0, "LRO entry count");
269 static int hn_share_tx_taskq = 0;
270 SYSCTL_INT(_hw_hn, OID_AUTO, share_tx_taskq, CTLFLAG_RDTUN,
271 &hn_share_tx_taskq, 0, "Enable shared TX taskqueue");
273 static struct taskqueue *hn_tx_taskq;
275 #ifndef HN_USE_TXDESC_BUFRING
276 static int hn_use_txdesc_bufring = 0;
278 static int hn_use_txdesc_bufring = 1;
280 SYSCTL_INT(_hw_hn, OID_AUTO, use_txdesc_bufring, CTLFLAG_RD,
281 &hn_use_txdesc_bufring, 0, "Use buf_ring for TX descriptors");
283 static int hn_bind_tx_taskq = -1;
284 SYSCTL_INT(_hw_hn, OID_AUTO, bind_tx_taskq, CTLFLAG_RDTUN,
285 &hn_bind_tx_taskq, 0, "Bind TX taskqueue to the specified cpu");
287 static int hn_use_if_start = 0;
288 SYSCTL_INT(_hw_hn, OID_AUTO, use_if_start, CTLFLAG_RDTUN,
289 &hn_use_if_start, 0, "Use if_start TX method");
291 static int hn_chan_cnt = 0;
292 SYSCTL_INT(_hw_hn, OID_AUTO, chan_cnt, CTLFLAG_RDTUN,
294 "# of channels to use; each channel has one RX ring and one TX ring");
296 static int hn_tx_ring_cnt = 0;
297 SYSCTL_INT(_hw_hn, OID_AUTO, tx_ring_cnt, CTLFLAG_RDTUN,
298 &hn_tx_ring_cnt, 0, "# of TX rings to use");
300 static int hn_tx_swq_depth = 0;
301 SYSCTL_INT(_hw_hn, OID_AUTO, tx_swq_depth, CTLFLAG_RDTUN,
302 &hn_tx_swq_depth, 0, "Depth of IFQ or BUFRING");
304 #if __FreeBSD_version >= 1100095
305 static u_int hn_lro_mbufq_depth = 0;
306 SYSCTL_UINT(_hw_hn, OID_AUTO, lro_mbufq_depth, CTLFLAG_RDTUN,
307 &hn_lro_mbufq_depth, 0, "Depth of LRO mbuf queue");
310 static u_int hn_cpu_index;
313 * Forward declarations
315 static void hn_stop(hn_softc_t *sc);
316 static void hn_ifinit_locked(hn_softc_t *sc);
317 static void hn_ifinit(void *xsc);
318 static int hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data);
319 static int hn_start_locked(struct hn_tx_ring *txr, int len);
320 static void hn_start(struct ifnet *ifp);
321 static void hn_start_txeof(struct hn_tx_ring *);
322 static int hn_ifmedia_upd(struct ifnet *ifp);
323 static void hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr);
324 #if __FreeBSD_version >= 1100099
325 static int hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS);
326 static int hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS);
328 static int hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS);
329 static int hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS);
330 static int hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS);
331 static int hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS);
332 static int hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS);
333 static int hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS);
334 static int hn_check_iplen(const struct mbuf *, int);
335 static int hn_create_tx_ring(struct hn_softc *, int);
336 static void hn_destroy_tx_ring(struct hn_tx_ring *);
337 static int hn_create_tx_data(struct hn_softc *, int);
338 static void hn_destroy_tx_data(struct hn_softc *);
339 static void hn_start_taskfunc(void *, int);
340 static void hn_start_txeof_taskfunc(void *, int);
341 static void hn_stop_tx_tasks(struct hn_softc *);
342 static int hn_encap(struct hn_tx_ring *, struct hn_txdesc *, struct mbuf **);
343 static void hn_create_rx_data(struct hn_softc *sc, int);
344 static void hn_destroy_rx_data(struct hn_softc *sc);
345 static void hn_set_tx_chimney_size(struct hn_softc *, int);
346 static void hn_channel_attach(struct hn_softc *, struct hv_vmbus_channel *);
347 static void hn_subchan_attach(struct hn_softc *, struct hv_vmbus_channel *);
348 static void hn_subchan_setup(struct hn_softc *);
350 static int hn_transmit(struct ifnet *, struct mbuf *);
351 static void hn_xmit_qflush(struct ifnet *);
352 static int hn_xmit(struct hn_tx_ring *, int);
353 static void hn_xmit_txeof(struct hn_tx_ring *);
354 static void hn_xmit_taskfunc(void *, int);
355 static void hn_xmit_txeof_taskfunc(void *, int);
357 #if __FreeBSD_version >= 1100099
359 hn_set_lro_lenlim(struct hn_softc *sc, int lenlim)
363 for (i = 0; i < sc->hn_rx_ring_inuse; ++i)
364 sc->hn_rx_ring[i].hn_lro.lro_length_lim = lenlim;
369 hn_get_txswq_depth(const struct hn_tx_ring *txr)
372 KASSERT(txr->hn_txdesc_cnt > 0, ("tx ring is not setup yet"));
373 if (hn_tx_swq_depth < txr->hn_txdesc_cnt)
374 return txr->hn_txdesc_cnt;
375 return hn_tx_swq_depth;
379 hn_ifmedia_upd(struct ifnet *ifp __unused)
386 hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
388 struct hn_softc *sc = ifp->if_softc;
390 ifmr->ifm_status = IFM_AVALID;
391 ifmr->ifm_active = IFM_ETHER;
393 if (!sc->hn_carrier) {
394 ifmr->ifm_active |= IFM_NONE;
397 ifmr->ifm_status |= IFM_ACTIVE;
398 ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
401 /* {F8615163-DF3E-46c5-913F-F2D2F965ED0E} */
402 static const struct hyperv_guid g_net_vsc_device_type = {
403 .hv_guid = {0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
404 0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E}
408 * Standard probe entry point.
412 netvsc_probe(device_t dev)
414 if (VMBUS_PROBE_GUID(device_get_parent(dev), dev,
415 &g_net_vsc_device_type) == 0) {
416 device_set_desc(dev, "Hyper-V Network Interface");
417 return BUS_PROBE_DEFAULT;
423 * Standard attach entry point.
425 * Called when the driver is loaded. It allocates needed resources,
426 * and initializes the "hardware" and software.
429 netvsc_attach(device_t dev)
431 netvsc_device_info device_info;
433 int unit = device_get_unit(dev);
434 struct ifnet *ifp = NULL;
435 int error, ring_cnt, tx_ring_cnt;
436 #if __FreeBSD_version >= 1100045
440 sc = device_get_softc(dev);
444 sc->hn_prichan = vmbus_get_channel(dev);
446 if (hn_tx_taskq == NULL) {
447 sc->hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
448 taskqueue_thread_enqueue, &sc->hn_tx_taskq);
449 if (hn_bind_tx_taskq >= 0) {
450 int cpu = hn_bind_tx_taskq;
453 if (cpu > mp_ncpus - 1)
455 CPU_SETOF(cpu, &cpu_set);
456 taskqueue_start_threads_cpuset(&sc->hn_tx_taskq, 1,
457 PI_NET, &cpu_set, "%s tx",
458 device_get_nameunit(dev));
460 taskqueue_start_threads(&sc->hn_tx_taskq, 1, PI_NET,
461 "%s tx", device_get_nameunit(dev));
464 sc->hn_tx_taskq = hn_tx_taskq;
466 NV_LOCK_INIT(sc, "NetVSCLock");
468 ifp = sc->hn_ifp = if_alloc(IFT_ETHER);
470 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
473 * Figure out the # of RX rings (ring_cnt) and the # of TX rings
474 * to use (tx_ring_cnt).
477 * The # of RX rings to use is same as the # of channels to use.
479 ring_cnt = hn_chan_cnt;
483 if (ring_cnt > HN_RING_CNT_DEF_MAX)
484 ring_cnt = HN_RING_CNT_DEF_MAX;
485 } else if (ring_cnt > mp_ncpus) {
489 tx_ring_cnt = hn_tx_ring_cnt;
490 if (tx_ring_cnt <= 0 || tx_ring_cnt > ring_cnt)
491 tx_ring_cnt = ring_cnt;
492 if (hn_use_if_start) {
493 /* ifnet.if_start only needs one TX ring. */
498 * Set the leader CPU for channels.
500 sc->hn_cpu = atomic_fetchadd_int(&hn_cpu_index, ring_cnt) % mp_ncpus;
502 error = hn_create_tx_data(sc, tx_ring_cnt);
505 hn_create_rx_data(sc, ring_cnt);
508 * Associate the first TX/RX ring w/ the primary channel.
510 hn_channel_attach(sc, sc->hn_prichan);
512 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
513 ifp->if_ioctl = hn_ioctl;
514 ifp->if_init = hn_ifinit;
515 /* needed by hv_rf_on_device_add() code */
516 ifp->if_mtu = ETHERMTU;
517 if (hn_use_if_start) {
518 int qdepth = hn_get_txswq_depth(&sc->hn_tx_ring[0]);
520 ifp->if_start = hn_start;
521 IFQ_SET_MAXLEN(&ifp->if_snd, qdepth);
522 ifp->if_snd.ifq_drv_maxlen = qdepth - 1;
523 IFQ_SET_READY(&ifp->if_snd);
525 ifp->if_transmit = hn_transmit;
526 ifp->if_qflush = hn_xmit_qflush;
529 ifmedia_init(&sc->hn_media, 0, hn_ifmedia_upd, hn_ifmedia_sts);
530 ifmedia_add(&sc->hn_media, IFM_ETHER | IFM_AUTO, 0, NULL);
531 ifmedia_set(&sc->hn_media, IFM_ETHER | IFM_AUTO);
532 /* XXX ifmedia_set really should do this for us */
533 sc->hn_media.ifm_media = sc->hn_media.ifm_cur->ifm_media;
536 * Tell upper layers that we support full VLAN capability.
538 ifp->if_hdrlen = sizeof(struct ether_vlan_header);
539 ifp->if_capabilities |=
540 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
543 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
545 ifp->if_hwassist = sc->hn_tx_ring[0].hn_csum_assist | CSUM_TSO;
547 error = hv_rf_on_device_add(sc, &device_info, ring_cnt);
550 KASSERT(sc->net_dev->num_channel > 0 &&
551 sc->net_dev->num_channel <= sc->hn_rx_ring_inuse,
552 ("invalid channel count %u, should be less than %d",
553 sc->net_dev->num_channel, sc->hn_rx_ring_inuse));
556 * Set the # of TX/RX rings that could be used according to
557 * the # of channels that host offered.
559 if (sc->hn_tx_ring_inuse > sc->net_dev->num_channel)
560 sc->hn_tx_ring_inuse = sc->net_dev->num_channel;
561 sc->hn_rx_ring_inuse = sc->net_dev->num_channel;
562 device_printf(dev, "%d TX ring, %d RX ring\n",
563 sc->hn_tx_ring_inuse, sc->hn_rx_ring_inuse);
565 if (sc->net_dev->num_channel > 1)
566 hn_subchan_setup(sc);
568 #if __FreeBSD_version >= 1100099
569 if (sc->hn_rx_ring_inuse > 1) {
571 * Reduce TCP segment aggregation limit for multiple
572 * RX rings to increase ACK timeliness.
574 hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MULTIRX_DEF);
578 if (device_info.link_state == 0) {
582 #if __FreeBSD_version >= 1100045
583 tso_maxlen = hn_tso_maxlen;
584 if (tso_maxlen <= 0 || tso_maxlen > IP_MAXPACKET)
585 tso_maxlen = IP_MAXPACKET;
587 ifp->if_hw_tsomaxsegcount = HN_TX_DATA_SEGCNT_MAX;
588 ifp->if_hw_tsomaxsegsize = PAGE_SIZE;
589 ifp->if_hw_tsomax = tso_maxlen -
590 (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
593 ether_ifattach(ifp, device_info.mac_addr);
595 #if __FreeBSD_version >= 1100045
596 if_printf(ifp, "TSO: %u/%u/%u\n", ifp->if_hw_tsomax,
597 ifp->if_hw_tsomaxsegcount, ifp->if_hw_tsomaxsegsize);
600 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
601 hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max);
602 if (hn_tx_chimney_size > 0 &&
603 hn_tx_chimney_size < sc->hn_tx_chimney_max)
604 hn_set_tx_chimney_size(sc, hn_tx_chimney_size);
608 hn_destroy_tx_data(sc);
615 * Standard detach entry point
618 netvsc_detach(device_t dev)
620 struct hn_softc *sc = device_get_softc(dev);
623 printf("netvsc_detach\n");
626 * XXXKYS: Need to clean up all our
627 * driver state; this is the driver
632 * XXXKYS: Need to stop outgoing traffic and unregister
636 hv_rf_on_device_remove(sc, HV_RF_NV_DESTROY_CHANNEL);
638 hn_stop_tx_tasks(sc);
640 ifmedia_removeall(&sc->hn_media);
641 hn_destroy_rx_data(sc);
642 hn_destroy_tx_data(sc);
644 if (sc->hn_tx_taskq != hn_tx_taskq)
645 taskqueue_free(sc->hn_tx_taskq);
651 * Standard shutdown entry point
654 netvsc_shutdown(device_t dev)
660 hn_txdesc_dmamap_load(struct hn_tx_ring *txr, struct hn_txdesc *txd,
661 struct mbuf **m_head, bus_dma_segment_t *segs, int *nsegs)
663 struct mbuf *m = *m_head;
666 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag, txd->data_dmap,
667 m, segs, nsegs, BUS_DMA_NOWAIT);
668 if (error == EFBIG) {
671 m_new = m_collapse(m, M_NOWAIT, HN_TX_DATA_SEGCNT_MAX);
676 txr->hn_tx_collapsed++;
678 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag,
679 txd->data_dmap, m, segs, nsegs, BUS_DMA_NOWAIT);
682 bus_dmamap_sync(txr->hn_tx_data_dtag, txd->data_dmap,
683 BUS_DMASYNC_PREWRITE);
684 txd->flags |= HN_TXD_FLAG_DMAMAP;
690 hn_txdesc_dmamap_unload(struct hn_tx_ring *txr, struct hn_txdesc *txd)
693 if (txd->flags & HN_TXD_FLAG_DMAMAP) {
694 bus_dmamap_sync(txr->hn_tx_data_dtag,
695 txd->data_dmap, BUS_DMASYNC_POSTWRITE);
696 bus_dmamap_unload(txr->hn_tx_data_dtag,
698 txd->flags &= ~HN_TXD_FLAG_DMAMAP;
703 hn_txdesc_put(struct hn_tx_ring *txr, struct hn_txdesc *txd)
706 KASSERT((txd->flags & HN_TXD_FLAG_ONLIST) == 0,
707 ("put an onlist txd %#x", txd->flags));
709 KASSERT(txd->refs > 0, ("invalid txd refs %d", txd->refs));
710 if (atomic_fetchadd_int(&txd->refs, -1) != 1)
713 hn_txdesc_dmamap_unload(txr, txd);
714 if (txd->m != NULL) {
719 txd->flags |= HN_TXD_FLAG_ONLIST;
721 #ifndef HN_USE_TXDESC_BUFRING
722 mtx_lock_spin(&txr->hn_txlist_spin);
723 KASSERT(txr->hn_txdesc_avail >= 0 &&
724 txr->hn_txdesc_avail < txr->hn_txdesc_cnt,
725 ("txdesc_put: invalid txd avail %d", txr->hn_txdesc_avail));
726 txr->hn_txdesc_avail++;
727 SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
728 mtx_unlock_spin(&txr->hn_txlist_spin);
730 atomic_add_int(&txr->hn_txdesc_avail, 1);
731 buf_ring_enqueue(txr->hn_txdesc_br, txd);
737 static __inline struct hn_txdesc *
738 hn_txdesc_get(struct hn_tx_ring *txr)
740 struct hn_txdesc *txd;
742 #ifndef HN_USE_TXDESC_BUFRING
743 mtx_lock_spin(&txr->hn_txlist_spin);
744 txd = SLIST_FIRST(&txr->hn_txlist);
746 KASSERT(txr->hn_txdesc_avail > 0,
747 ("txdesc_get: invalid txd avail %d", txr->hn_txdesc_avail));
748 txr->hn_txdesc_avail--;
749 SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
751 mtx_unlock_spin(&txr->hn_txlist_spin);
753 txd = buf_ring_dequeue_sc(txr->hn_txdesc_br);
757 #ifdef HN_USE_TXDESC_BUFRING
758 atomic_subtract_int(&txr->hn_txdesc_avail, 1);
760 KASSERT(txd->m == NULL && txd->refs == 0 &&
761 (txd->flags & HN_TXD_FLAG_ONLIST), ("invalid txd"));
762 txd->flags &= ~HN_TXD_FLAG_ONLIST;
769 hn_txdesc_hold(struct hn_txdesc *txd)
772 /* 0->1 transition will never work */
773 KASSERT(txd->refs > 0, ("invalid refs %d", txd->refs));
774 atomic_add_int(&txd->refs, 1);
778 hn_txeof(struct hn_tx_ring *txr)
780 txr->hn_has_txeof = 0;
785 hn_tx_done(struct hv_vmbus_channel *chan, void *xpkt)
787 netvsc_packet *packet = xpkt;
788 struct hn_txdesc *txd;
789 struct hn_tx_ring *txr;
791 txd = (struct hn_txdesc *)(uintptr_t)
792 packet->compl.send.send_completion_tid;
795 KASSERT(txr->hn_chan == chan,
796 ("channel mismatch, on channel%u, should be channel%u",
798 txr->hn_chan->ch_subidx));
800 txr->hn_has_txeof = 1;
801 hn_txdesc_put(txr, txd);
803 ++txr->hn_txdone_cnt;
804 if (txr->hn_txdone_cnt >= HN_EARLY_TXEOF_THRESH) {
805 txr->hn_txdone_cnt = 0;
812 netvsc_channel_rollup(struct hv_vmbus_channel *chan)
814 struct hn_tx_ring *txr = chan->hv_chan_txr;
815 #if defined(INET) || defined(INET6)
816 struct hn_rx_ring *rxr = chan->hv_chan_rxr;
818 tcp_lro_flush_all(&rxr->hn_lro);
823 * 'txr' could be NULL, if multiple channels and
824 * ifnet.if_start method are enabled.
826 if (txr == NULL || !txr->hn_has_txeof)
829 txr->hn_txdone_cnt = 0;
835 * If this function fails, then both txd and m_head0 will be freed.
838 hn_encap(struct hn_tx_ring *txr, struct hn_txdesc *txd, struct mbuf **m_head0)
840 bus_dma_segment_t segs[HN_TX_DATA_SEGCNT_MAX];
842 struct mbuf *m_head = *m_head0;
843 netvsc_packet *packet;
844 rndis_msg *rndis_mesg;
845 rndis_packet *rndis_pkt;
846 rndis_per_packet_info *rppi;
847 struct rndis_hash_value *hash_value;
848 uint32_t rndis_msg_size;
850 packet = &txd->netvsc_pkt;
851 packet->is_data_pkt = TRUE;
852 packet->tot_data_buf_len = m_head->m_pkthdr.len;
855 * extension points to the area reserved for the
856 * rndis_filter_packet, which is placed just after
857 * the netvsc_packet (and rppi struct, if present;
858 * length is updated later).
860 rndis_mesg = txd->rndis_msg;
861 /* XXX not necessary */
862 memset(rndis_mesg, 0, HN_RNDIS_MSG_LEN);
863 rndis_mesg->ndis_msg_type = REMOTE_NDIS_PACKET_MSG;
865 rndis_pkt = &rndis_mesg->msg.packet;
866 rndis_pkt->data_offset = sizeof(rndis_packet);
867 rndis_pkt->data_length = packet->tot_data_buf_len;
868 rndis_pkt->per_pkt_info_offset = sizeof(rndis_packet);
870 rndis_msg_size = RNDIS_MESSAGE_SIZE(rndis_packet);
873 * Set the hash value for this packet, so that the host could
874 * dispatch the TX done event for this packet back to this TX
877 rndis_msg_size += RNDIS_HASHVAL_PPI_SIZE;
878 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_HASHVAL_PPI_SIZE,
880 hash_value = (struct rndis_hash_value *)((uint8_t *)rppi +
881 rppi->per_packet_info_offset);
882 hash_value->hash_value = txr->hn_tx_idx;
884 if (m_head->m_flags & M_VLANTAG) {
885 ndis_8021q_info *rppi_vlan_info;
887 rndis_msg_size += RNDIS_VLAN_PPI_SIZE;
888 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_VLAN_PPI_SIZE,
891 rppi_vlan_info = (ndis_8021q_info *)((uint8_t *)rppi +
892 rppi->per_packet_info_offset);
893 rppi_vlan_info->u1.s1.vlan_id =
894 m_head->m_pkthdr.ether_vtag & 0xfff;
897 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
898 rndis_tcp_tso_info *tso_info;
899 struct ether_vlan_header *eh;
903 * XXX need m_pullup and use mtodo
905 eh = mtod(m_head, struct ether_vlan_header*);
906 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN))
907 ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
909 ether_len = ETHER_HDR_LEN;
911 rndis_msg_size += RNDIS_TSO_PPI_SIZE;
912 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_TSO_PPI_SIZE,
913 tcp_large_send_info);
915 tso_info = (rndis_tcp_tso_info *)((uint8_t *)rppi +
916 rppi->per_packet_info_offset);
917 tso_info->lso_v2_xmit.type =
918 RNDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
921 if (m_head->m_pkthdr.csum_flags & CSUM_IP_TSO) {
923 (struct ip *)(m_head->m_data + ether_len);
924 unsigned long iph_len = ip->ip_hl << 2;
926 (struct tcphdr *)((caddr_t)ip + iph_len);
928 tso_info->lso_v2_xmit.ip_version =
929 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
933 th->th_sum = in_pseudo(ip->ip_src.s_addr,
934 ip->ip_dst.s_addr, htons(IPPROTO_TCP));
937 #if defined(INET6) && defined(INET)
942 struct ip6_hdr *ip6 = (struct ip6_hdr *)
943 (m_head->m_data + ether_len);
944 struct tcphdr *th = (struct tcphdr *)(ip6 + 1);
946 tso_info->lso_v2_xmit.ip_version =
947 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
949 th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
952 tso_info->lso_v2_xmit.tcp_header_offset = 0;
953 tso_info->lso_v2_xmit.mss = m_head->m_pkthdr.tso_segsz;
954 } else if (m_head->m_pkthdr.csum_flags & txr->hn_csum_assist) {
955 rndis_tcp_ip_csum_info *csum_info;
957 rndis_msg_size += RNDIS_CSUM_PPI_SIZE;
958 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_CSUM_PPI_SIZE,
960 csum_info = (rndis_tcp_ip_csum_info *)((uint8_t *)rppi +
961 rppi->per_packet_info_offset);
963 csum_info->xmit.is_ipv4 = 1;
964 if (m_head->m_pkthdr.csum_flags & CSUM_IP)
965 csum_info->xmit.ip_header_csum = 1;
967 if (m_head->m_pkthdr.csum_flags & CSUM_TCP) {
968 csum_info->xmit.tcp_csum = 1;
969 csum_info->xmit.tcp_header_offset = 0;
970 } else if (m_head->m_pkthdr.csum_flags & CSUM_UDP) {
971 csum_info->xmit.udp_csum = 1;
975 rndis_mesg->msg_len = packet->tot_data_buf_len + rndis_msg_size;
976 packet->tot_data_buf_len = rndis_mesg->msg_len;
979 * Chimney send, if the packet could fit into one chimney buffer.
981 if (packet->tot_data_buf_len < txr->hn_tx_chimney_size) {
982 netvsc_dev *net_dev = txr->hn_sc->net_dev;
983 uint32_t send_buf_section_idx;
985 txr->hn_tx_chimney_tried++;
986 send_buf_section_idx =
987 hv_nv_get_next_send_section(net_dev);
988 if (send_buf_section_idx !=
989 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX) {
990 uint8_t *dest = ((uint8_t *)net_dev->send_buf +
991 (send_buf_section_idx *
992 net_dev->send_section_size));
994 memcpy(dest, rndis_mesg, rndis_msg_size);
995 dest += rndis_msg_size;
996 m_copydata(m_head, 0, m_head->m_pkthdr.len, dest);
998 packet->send_buf_section_idx = send_buf_section_idx;
999 packet->send_buf_section_size =
1000 packet->tot_data_buf_len;
1001 packet->gpa_cnt = 0;
1002 txr->hn_tx_chimney++;
1007 error = hn_txdesc_dmamap_load(txr, txd, &m_head, segs, &nsegs);
1012 * This mbuf is not linked w/ the txd yet, so free it now.
1017 freed = hn_txdesc_put(txr, txd);
1019 ("fail to free txd upon txdma error"));
1021 txr->hn_txdma_failed++;
1022 if_inc_counter(txr->hn_sc->hn_ifp, IFCOUNTER_OERRORS, 1);
1027 packet->gpa_cnt = nsegs + HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
1029 /* send packet with page buffer */
1030 packet->gpa[0].gpa_page = atop(txd->rndis_msg_paddr);
1031 packet->gpa[0].gpa_ofs = txd->rndis_msg_paddr & PAGE_MASK;
1032 packet->gpa[0].gpa_len = rndis_msg_size;
1035 * Fill the page buffers with mbuf info starting at index
1036 * HV_RF_NUM_TX_RESERVED_PAGE_BUFS.
1038 for (i = 0; i < nsegs; ++i) {
1039 struct vmbus_gpa *gpa = &packet->gpa[
1040 i + HV_RF_NUM_TX_RESERVED_PAGE_BUFS];
1042 gpa->gpa_page = atop(segs[i].ds_addr);
1043 gpa->gpa_ofs = segs[i].ds_addr & PAGE_MASK;
1044 gpa->gpa_len = segs[i].ds_len;
1047 packet->send_buf_section_idx =
1048 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX;
1049 packet->send_buf_section_size = 0;
1053 /* Set the completion routine */
1054 packet->compl.send.on_send_completion = hn_tx_done;
1055 packet->compl.send.send_completion_context = packet;
1056 packet->compl.send.send_completion_tid = (uint64_t)(uintptr_t)txd;
1063 * If this function fails, then txd will be freed, but the mbuf
1064 * associated w/ the txd will _not_ be freed.
1067 hn_send_pkt(struct ifnet *ifp, struct hn_tx_ring *txr, struct hn_txdesc *txd)
1069 int error, send_failed = 0;
1073 * Make sure that txd is not freed before ETHER_BPF_MTAP.
1075 hn_txdesc_hold(txd);
1076 error = hv_nv_on_send(txr->hn_chan, &txd->netvsc_pkt);
1078 ETHER_BPF_MTAP(ifp, txd->m);
1079 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1080 if (!hn_use_if_start) {
1081 if_inc_counter(ifp, IFCOUNTER_OBYTES,
1082 txd->m->m_pkthdr.len);
1083 if (txd->m->m_flags & M_MCAST)
1084 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
1088 hn_txdesc_put(txr, txd);
1090 if (__predict_false(error)) {
1094 * This should "really rarely" happen.
1096 * XXX Too many RX to be acked or too many sideband
1097 * commands to run? Ask netvsc_channel_rollup()
1098 * to kick start later.
1100 txr->hn_has_txeof = 1;
1102 txr->hn_send_failed++;
1105 * Try sending again after set hn_has_txeof;
1106 * in case that we missed the last
1107 * netvsc_channel_rollup().
1111 if_printf(ifp, "send failed\n");
1114 * Caller will perform further processing on the
1115 * associated mbuf, so don't free it in hn_txdesc_put();
1116 * only unload it from the DMA map in hn_txdesc_put(),
1120 freed = hn_txdesc_put(txr, txd);
1122 ("fail to free txd upon send error"));
1124 txr->hn_send_failed++;
1130 * Start a transmit of one or more packets
1133 hn_start_locked(struct hn_tx_ring *txr, int len)
1135 struct hn_softc *sc = txr->hn_sc;
1136 struct ifnet *ifp = sc->hn_ifp;
1138 KASSERT(hn_use_if_start,
1139 ("hn_start_locked is called, when if_start is disabled"));
1140 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1141 mtx_assert(&txr->hn_tx_lock, MA_OWNED);
1143 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1147 while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
1148 struct hn_txdesc *txd;
1149 struct mbuf *m_head;
1152 IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
1156 if (len > 0 && m_head->m_pkthdr.len > len) {
1158 * This sending could be time consuming; let callers
1159 * dispatch this packet sending (and sending of any
1160 * following up packets) to tx taskqueue.
1162 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1166 txd = hn_txdesc_get(txr);
1168 txr->hn_no_txdescs++;
1169 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1170 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1174 error = hn_encap(txr, txd, &m_head);
1176 /* Both txd and m_head are freed */
1180 error = hn_send_pkt(ifp, txr, txd);
1181 if (__predict_false(error)) {
1182 /* txd is freed, but m_head is not */
1183 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1184 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1192 * Link up/down notification
1195 netvsc_linkstatus_callback(struct hn_softc *sc, uint32_t status)
1205 * Append the specified data to the indicated mbuf chain,
1206 * Extend the mbuf chain if the new data does not fit in
1209 * This is a minor rewrite of m_append() from sys/kern/uipc_mbuf.c.
1210 * There should be an equivalent in the kernel mbuf code,
1211 * but there does not appear to be one yet.
1213 * Differs from m_append() in that additional mbufs are
1214 * allocated with cluster size MJUMPAGESIZE, and filled
1217 * Return 1 if able to complete the job; otherwise 0.
1220 hv_m_append(struct mbuf *m0, int len, c_caddr_t cp)
1223 int remainder, space;
1225 for (m = m0; m->m_next != NULL; m = m->m_next)
1228 space = M_TRAILINGSPACE(m);
1231 * Copy into available space.
1233 if (space > remainder)
1235 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1240 while (remainder > 0) {
1242 * Allocate a new mbuf; could check space
1243 * and allocate a cluster instead.
1245 n = m_getjcl(M_NOWAIT, m->m_type, 0, MJUMPAGESIZE);
1248 n->m_len = min(MJUMPAGESIZE, remainder);
1249 bcopy(cp, mtod(n, caddr_t), n->m_len);
1251 remainder -= n->m_len;
1255 if (m0->m_flags & M_PKTHDR)
1256 m0->m_pkthdr.len += len - remainder;
1258 return (remainder == 0);
1261 #if defined(INET) || defined(INET6)
1263 hn_lro_rx(struct lro_ctrl *lc, struct mbuf *m)
1265 #if __FreeBSD_version >= 1100095
1266 if (hn_lro_mbufq_depth) {
1267 tcp_lro_queue_mbuf(lc, m);
1271 return tcp_lro_rx(lc, m, 0);
1276 * Called when we receive a data packet from the "wire" on the
1279 * Note: This is no longer used as a callback
1282 netvsc_recv(struct hv_vmbus_channel *chan, netvsc_packet *packet,
1283 const rndis_tcp_ip_csum_info *csum_info,
1284 const struct rndis_hash_info *hash_info,
1285 const struct rndis_hash_value *hash_value)
1287 struct hn_rx_ring *rxr = chan->hv_chan_rxr;
1288 struct ifnet *ifp = rxr->hn_ifp;
1290 int size, do_lro = 0, do_csum = 1;
1291 int hash_type = M_HASHTYPE_OPAQUE_HASH;
1293 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1297 * Bail out if packet contains more data than configured MTU.
1299 if (packet->tot_data_buf_len > (ifp->if_mtu + ETHER_HDR_LEN)) {
1301 } else if (packet->tot_data_buf_len <= MHLEN) {
1302 m_new = m_gethdr(M_NOWAIT, MT_DATA);
1303 if (m_new == NULL) {
1304 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1307 memcpy(mtod(m_new, void *), packet->data,
1308 packet->tot_data_buf_len);
1309 m_new->m_pkthdr.len = m_new->m_len = packet->tot_data_buf_len;
1310 rxr->hn_small_pkts++;
1313 * Get an mbuf with a cluster. For packets 2K or less,
1314 * get a standard 2K cluster. For anything larger, get a
1315 * 4K cluster. Any buffers larger than 4K can cause problems
1316 * if looped around to the Hyper-V TX channel, so avoid them.
1319 if (packet->tot_data_buf_len > MCLBYTES) {
1321 size = MJUMPAGESIZE;
1324 m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, size);
1325 if (m_new == NULL) {
1326 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1330 hv_m_append(m_new, packet->tot_data_buf_len, packet->data);
1332 m_new->m_pkthdr.rcvif = ifp;
1334 if (__predict_false((ifp->if_capenable & IFCAP_RXCSUM) == 0))
1337 /* receive side checksum offload */
1338 if (csum_info != NULL) {
1339 /* IP csum offload */
1340 if (csum_info->receive.ip_csum_succeeded && do_csum) {
1341 m_new->m_pkthdr.csum_flags |=
1342 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1346 /* TCP/UDP csum offload */
1347 if ((csum_info->receive.tcp_csum_succeeded ||
1348 csum_info->receive.udp_csum_succeeded) && do_csum) {
1349 m_new->m_pkthdr.csum_flags |=
1350 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1351 m_new->m_pkthdr.csum_data = 0xffff;
1352 if (csum_info->receive.tcp_csum_succeeded)
1358 if (csum_info->receive.ip_csum_succeeded &&
1359 csum_info->receive.tcp_csum_succeeded)
1362 const struct ether_header *eh;
1367 if (m_new->m_len < hoff)
1369 eh = mtod(m_new, struct ether_header *);
1370 etype = ntohs(eh->ether_type);
1371 if (etype == ETHERTYPE_VLAN) {
1372 const struct ether_vlan_header *evl;
1374 hoff = sizeof(*evl);
1375 if (m_new->m_len < hoff)
1377 evl = mtod(m_new, struct ether_vlan_header *);
1378 etype = ntohs(evl->evl_proto);
1381 if (etype == ETHERTYPE_IP) {
1384 pr = hn_check_iplen(m_new, hoff);
1385 if (pr == IPPROTO_TCP) {
1387 (rxr->hn_trust_hcsum &
1388 HN_TRUST_HCSUM_TCP)) {
1389 rxr->hn_csum_trusted++;
1390 m_new->m_pkthdr.csum_flags |=
1391 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1392 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1393 m_new->m_pkthdr.csum_data = 0xffff;
1396 } else if (pr == IPPROTO_UDP) {
1398 (rxr->hn_trust_hcsum &
1399 HN_TRUST_HCSUM_UDP)) {
1400 rxr->hn_csum_trusted++;
1401 m_new->m_pkthdr.csum_flags |=
1402 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1403 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1404 m_new->m_pkthdr.csum_data = 0xffff;
1406 } else if (pr != IPPROTO_DONE && do_csum &&
1407 (rxr->hn_trust_hcsum & HN_TRUST_HCSUM_IP)) {
1408 rxr->hn_csum_trusted++;
1409 m_new->m_pkthdr.csum_flags |=
1410 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1415 if ((packet->vlan_tci != 0) &&
1416 (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0) {
1417 m_new->m_pkthdr.ether_vtag = packet->vlan_tci;
1418 m_new->m_flags |= M_VLANTAG;
1421 if (hash_info != NULL && hash_value != NULL) {
1423 m_new->m_pkthdr.flowid = hash_value->hash_value;
1424 if ((hash_info->hash_info & NDIS_HASH_FUNCTION_MASK) ==
1425 NDIS_HASH_FUNCTION_TOEPLITZ) {
1427 (hash_info->hash_info & NDIS_HASH_TYPE_MASK);
1430 case NDIS_HASH_IPV4:
1431 hash_type = M_HASHTYPE_RSS_IPV4;
1434 case NDIS_HASH_TCP_IPV4:
1435 hash_type = M_HASHTYPE_RSS_TCP_IPV4;
1438 case NDIS_HASH_IPV6:
1439 hash_type = M_HASHTYPE_RSS_IPV6;
1442 case NDIS_HASH_IPV6_EX:
1443 hash_type = M_HASHTYPE_RSS_IPV6_EX;
1446 case NDIS_HASH_TCP_IPV6:
1447 hash_type = M_HASHTYPE_RSS_TCP_IPV6;
1450 case NDIS_HASH_TCP_IPV6_EX:
1451 hash_type = M_HASHTYPE_RSS_TCP_IPV6_EX;
1456 if (hash_value != NULL) {
1457 m_new->m_pkthdr.flowid = hash_value->hash_value;
1459 m_new->m_pkthdr.flowid = rxr->hn_rx_idx;
1460 hash_type = M_HASHTYPE_OPAQUE;
1463 M_HASHTYPE_SET(m_new, hash_type);
1466 * Note: Moved RX completion back to hv_nv_on_receive() so all
1467 * messages (not just data messages) will trigger a response.
1470 if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1473 if ((ifp->if_capenable & IFCAP_LRO) && do_lro) {
1474 #if defined(INET) || defined(INET6)
1475 struct lro_ctrl *lro = &rxr->hn_lro;
1478 rxr->hn_lro_tried++;
1479 if (hn_lro_rx(lro, m_new) == 0) {
1487 /* We're not holding the lock here, so don't release it */
1488 (*ifp->if_input)(ifp, m_new);
1494 * Rules for using sc->temp_unusable:
1495 * 1. sc->temp_unusable can only be read or written while holding NV_LOCK()
1496 * 2. code reading sc->temp_unusable under NV_LOCK(), and finding
1497 * sc->temp_unusable set, must release NV_LOCK() and exit
1498 * 3. to retain exclusive control of the interface,
1499 * sc->temp_unusable must be set by code before releasing NV_LOCK()
1500 * 4. only code setting sc->temp_unusable can clear sc->temp_unusable
1501 * 5. code setting sc->temp_unusable must eventually clear sc->temp_unusable
1505 * Standard ioctl entry point. Called when the user wants to configure
1509 hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1511 hn_softc_t *sc = ifp->if_softc;
1512 struct ifreq *ifr = (struct ifreq *)data;
1514 struct ifaddr *ifa = (struct ifaddr *)data;
1516 netvsc_device_info device_info;
1517 int mask, error = 0;
1518 int retry_cnt = 500;
1524 if (ifa->ifa_addr->sa_family == AF_INET) {
1525 ifp->if_flags |= IFF_UP;
1526 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1528 arp_ifinit(ifp, ifa);
1531 error = ether_ioctl(ifp, cmd, data);
1534 /* Check MTU value change */
1535 if (ifp->if_mtu == ifr->ifr_mtu)
1538 if (ifr->ifr_mtu > NETVSC_MAX_CONFIGURABLE_MTU) {
1543 /* Obtain and record requested MTU */
1544 ifp->if_mtu = ifr->ifr_mtu;
1546 #if __FreeBSD_version >= 1100099
1548 * Make sure that LRO aggregation length limit is still
1549 * valid, after the MTU change.
1552 if (sc->hn_rx_ring[0].hn_lro.lro_length_lim <
1553 HN_LRO_LENLIM_MIN(ifp))
1554 hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MIN(ifp));
1560 if (!sc->temp_unusable) {
1561 sc->temp_unusable = TRUE;
1565 if (retry_cnt > 0) {
1569 } while (retry_cnt > 0);
1571 if (retry_cnt == 0) {
1576 /* We must remove and add back the device to cause the new
1577 * MTU to take effect. This includes tearing down, but not
1578 * deleting the channel, then bringing it back up.
1580 error = hv_rf_on_device_remove(sc, HV_RF_NV_RETAIN_CHANNEL);
1583 sc->temp_unusable = FALSE;
1588 /* Wait for subchannels to be destroyed */
1589 vmbus_subchan_drain(sc->hn_prichan);
1591 error = hv_rf_on_device_add(sc, &device_info,
1592 sc->hn_rx_ring_inuse);
1595 sc->temp_unusable = FALSE;
1599 KASSERT(sc->hn_rx_ring_cnt == sc->net_dev->num_channel,
1600 ("RX ring count %d and channel count %u mismatch",
1601 sc->hn_rx_ring_cnt, sc->net_dev->num_channel));
1602 if (sc->net_dev->num_channel > 1) {
1606 * Skip the rings on primary channel; they are
1607 * handled by the hv_rf_on_device_add() above.
1609 for (r = 1; r < sc->hn_rx_ring_cnt; ++r) {
1610 sc->hn_rx_ring[r].hn_rx_flags &=
1611 ~HN_RX_FLAG_ATTACHED;
1613 for (r = 1; r < sc->hn_tx_ring_cnt; ++r) {
1614 sc->hn_tx_ring[r].hn_tx_flags &=
1615 ~HN_TX_FLAG_ATTACHED;
1617 hn_subchan_setup(sc);
1620 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
1621 if (sc->hn_tx_ring[0].hn_tx_chimney_size >
1622 sc->hn_tx_chimney_max)
1623 hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max);
1625 hn_ifinit_locked(sc);
1628 sc->temp_unusable = FALSE;
1634 if (!sc->temp_unusable) {
1635 sc->temp_unusable = TRUE;
1639 if (retry_cnt > 0) {
1643 } while (retry_cnt > 0);
1645 if (retry_cnt == 0) {
1650 if (ifp->if_flags & IFF_UP) {
1652 * If only the state of the PROMISC flag changed,
1653 * then just use the 'set promisc mode' command
1654 * instead of reinitializing the entire NIC. Doing
1655 * a full re-init means reloading the firmware and
1656 * waiting for it to start up, which may take a
1660 /* Fixme: Promiscuous mode? */
1661 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1662 ifp->if_flags & IFF_PROMISC &&
1663 !(sc->hn_if_flags & IFF_PROMISC)) {
1664 /* do something here for Hyper-V */
1665 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1666 !(ifp->if_flags & IFF_PROMISC) &&
1667 sc->hn_if_flags & IFF_PROMISC) {
1668 /* do something here for Hyper-V */
1671 hn_ifinit_locked(sc);
1673 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1678 sc->temp_unusable = FALSE;
1680 sc->hn_if_flags = ifp->if_flags;
1686 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1687 if (mask & IFCAP_TXCSUM) {
1688 ifp->if_capenable ^= IFCAP_TXCSUM;
1689 if (ifp->if_capenable & IFCAP_TXCSUM) {
1691 sc->hn_tx_ring[0].hn_csum_assist;
1694 ~sc->hn_tx_ring[0].hn_csum_assist;
1698 if (mask & IFCAP_RXCSUM)
1699 ifp->if_capenable ^= IFCAP_RXCSUM;
1701 if (mask & IFCAP_LRO)
1702 ifp->if_capenable ^= IFCAP_LRO;
1704 if (mask & IFCAP_TSO4) {
1705 ifp->if_capenable ^= IFCAP_TSO4;
1706 if (ifp->if_capenable & IFCAP_TSO4)
1707 ifp->if_hwassist |= CSUM_IP_TSO;
1709 ifp->if_hwassist &= ~CSUM_IP_TSO;
1712 if (mask & IFCAP_TSO6) {
1713 ifp->if_capenable ^= IFCAP_TSO6;
1714 if (ifp->if_capenable & IFCAP_TSO6)
1715 ifp->if_hwassist |= CSUM_IP6_TSO;
1717 ifp->if_hwassist &= ~CSUM_IP6_TSO;
1726 /* Fixme: Multicast mode? */
1727 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1729 netvsc_setmulti(sc);
1738 error = ifmedia_ioctl(ifp, ifr, &sc->hn_media, cmd);
1741 error = ether_ioctl(ifp, cmd, data);
1752 hn_stop(hn_softc_t *sc)
1760 printf(" Closing Device ...\n");
1762 atomic_clear_int(&ifp->if_drv_flags,
1763 (IFF_DRV_RUNNING | IFF_DRV_OACTIVE));
1764 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
1765 sc->hn_tx_ring[i].hn_oactive = 0;
1767 if_link_state_change(ifp, LINK_STATE_DOWN);
1768 sc->hn_initdone = 0;
1770 ret = hv_rf_on_close(sc);
1774 * FreeBSD transmit entry point
1777 hn_start(struct ifnet *ifp)
1779 struct hn_softc *sc = ifp->if_softc;
1780 struct hn_tx_ring *txr = &sc->hn_tx_ring[0];
1782 if (txr->hn_sched_tx)
1785 if (mtx_trylock(&txr->hn_tx_lock)) {
1788 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1789 mtx_unlock(&txr->hn_tx_lock);
1794 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
1798 hn_start_txeof(struct hn_tx_ring *txr)
1800 struct hn_softc *sc = txr->hn_sc;
1801 struct ifnet *ifp = sc->hn_ifp;
1803 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1805 if (txr->hn_sched_tx)
1808 if (mtx_trylock(&txr->hn_tx_lock)) {
1811 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1812 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1813 mtx_unlock(&txr->hn_tx_lock);
1815 taskqueue_enqueue(txr->hn_tx_taskq,
1821 * Release the OACTIVE earlier, with the hope, that
1822 * others could catch up. The task will clear the
1823 * flag again with the hn_tx_lock to avoid possible
1826 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1827 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
1835 hn_ifinit_locked(hn_softc_t *sc)
1842 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1846 hv_promisc_mode = 1;
1848 ret = hv_rf_on_open(sc);
1852 sc->hn_initdone = 1;
1855 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1856 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
1857 sc->hn_tx_ring[i].hn_oactive = 0;
1859 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_RUNNING);
1860 if_link_state_change(ifp, LINK_STATE_UP);
1867 hn_ifinit(void *xsc)
1869 hn_softc_t *sc = xsc;
1872 if (sc->temp_unusable) {
1876 sc->temp_unusable = TRUE;
1879 hn_ifinit_locked(sc);
1882 sc->temp_unusable = FALSE;
1891 hn_watchdog(struct ifnet *ifp)
1896 printf("hn%d: watchdog timeout -- resetting\n", sc->hn_unit);
1897 hn_ifinit(sc); /*???*/
1898 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1902 #if __FreeBSD_version >= 1100099
1905 hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS)
1907 struct hn_softc *sc = arg1;
1908 unsigned int lenlim;
1911 lenlim = sc->hn_rx_ring[0].hn_lro.lro_length_lim;
1912 error = sysctl_handle_int(oidp, &lenlim, 0, req);
1913 if (error || req->newptr == NULL)
1916 if (lenlim < HN_LRO_LENLIM_MIN(sc->hn_ifp) ||
1917 lenlim > TCP_LRO_LENGTH_MAX)
1921 hn_set_lro_lenlim(sc, lenlim);
1927 hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS)
1929 struct hn_softc *sc = arg1;
1930 int ackcnt, error, i;
1933 * lro_ackcnt_lim is append count limit,
1934 * +1 to turn it into aggregation limit.
1936 ackcnt = sc->hn_rx_ring[0].hn_lro.lro_ackcnt_lim + 1;
1937 error = sysctl_handle_int(oidp, &ackcnt, 0, req);
1938 if (error || req->newptr == NULL)
1941 if (ackcnt < 2 || ackcnt > (TCP_LRO_ACKCNT_MAX + 1))
1945 * Convert aggregation limit back to append
1950 for (i = 0; i < sc->hn_rx_ring_inuse; ++i)
1951 sc->hn_rx_ring[i].hn_lro.lro_ackcnt_lim = ackcnt;
1959 hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS)
1961 struct hn_softc *sc = arg1;
1966 if (sc->hn_rx_ring[0].hn_trust_hcsum & hcsum)
1969 error = sysctl_handle_int(oidp, &on, 0, req);
1970 if (error || req->newptr == NULL)
1974 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1975 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
1978 rxr->hn_trust_hcsum |= hcsum;
1980 rxr->hn_trust_hcsum &= ~hcsum;
1987 hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS)
1989 struct hn_softc *sc = arg1;
1990 int chimney_size, error;
1992 chimney_size = sc->hn_tx_ring[0].hn_tx_chimney_size;
1993 error = sysctl_handle_int(oidp, &chimney_size, 0, req);
1994 if (error || req->newptr == NULL)
1997 if (chimney_size > sc->hn_tx_chimney_max || chimney_size <= 0)
2000 hn_set_tx_chimney_size(sc, chimney_size);
2005 hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
2007 struct hn_softc *sc = arg1;
2008 int ofs = arg2, i, error;
2009 struct hn_rx_ring *rxr;
2013 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2014 rxr = &sc->hn_rx_ring[i];
2015 stat += *((u_long *)((uint8_t *)rxr + ofs));
2018 error = sysctl_handle_long(oidp, &stat, 0, req);
2019 if (error || req->newptr == NULL)
2022 /* Zero out this stat. */
2023 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2024 rxr = &sc->hn_rx_ring[i];
2025 *((u_long *)((uint8_t *)rxr + ofs)) = 0;
2031 hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS)
2033 struct hn_softc *sc = arg1;
2034 int ofs = arg2, i, error;
2035 struct hn_rx_ring *rxr;
2039 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2040 rxr = &sc->hn_rx_ring[i];
2041 stat += *((uint64_t *)((uint8_t *)rxr + ofs));
2044 error = sysctl_handle_64(oidp, &stat, 0, req);
2045 if (error || req->newptr == NULL)
2048 /* Zero out this stat. */
2049 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2050 rxr = &sc->hn_rx_ring[i];
2051 *((uint64_t *)((uint8_t *)rxr + ofs)) = 0;
2057 hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
2059 struct hn_softc *sc = arg1;
2060 int ofs = arg2, i, error;
2061 struct hn_tx_ring *txr;
2065 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2066 txr = &sc->hn_tx_ring[i];
2067 stat += *((u_long *)((uint8_t *)txr + ofs));
2070 error = sysctl_handle_long(oidp, &stat, 0, req);
2071 if (error || req->newptr == NULL)
2074 /* Zero out this stat. */
2075 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2076 txr = &sc->hn_tx_ring[i];
2077 *((u_long *)((uint8_t *)txr + ofs)) = 0;
2083 hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS)
2085 struct hn_softc *sc = arg1;
2086 int ofs = arg2, i, error, conf;
2087 struct hn_tx_ring *txr;
2089 txr = &sc->hn_tx_ring[0];
2090 conf = *((int *)((uint8_t *)txr + ofs));
2092 error = sysctl_handle_int(oidp, &conf, 0, req);
2093 if (error || req->newptr == NULL)
2097 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2098 txr = &sc->hn_tx_ring[i];
2099 *((int *)((uint8_t *)txr + ofs)) = conf;
2107 hn_check_iplen(const struct mbuf *m, int hoff)
2109 const struct ip *ip;
2110 int len, iphlen, iplen;
2111 const struct tcphdr *th;
2112 int thoff; /* TCP data offset */
2114 len = hoff + sizeof(struct ip);
2116 /* The packet must be at least the size of an IP header. */
2117 if (m->m_pkthdr.len < len)
2118 return IPPROTO_DONE;
2120 /* The fixed IP header must reside completely in the first mbuf. */
2122 return IPPROTO_DONE;
2124 ip = mtodo(m, hoff);
2126 /* Bound check the packet's stated IP header length. */
2127 iphlen = ip->ip_hl << 2;
2128 if (iphlen < sizeof(struct ip)) /* minimum header length */
2129 return IPPROTO_DONE;
2131 /* The full IP header must reside completely in the one mbuf. */
2132 if (m->m_len < hoff + iphlen)
2133 return IPPROTO_DONE;
2135 iplen = ntohs(ip->ip_len);
2138 * Check that the amount of data in the buffers is as
2139 * at least much as the IP header would have us expect.
2141 if (m->m_pkthdr.len < hoff + iplen)
2142 return IPPROTO_DONE;
2145 * Ignore IP fragments.
2147 if (ntohs(ip->ip_off) & (IP_OFFMASK | IP_MF))
2148 return IPPROTO_DONE;
2151 * The TCP/IP or UDP/IP header must be entirely contained within
2152 * the first fragment of a packet.
2156 if (iplen < iphlen + sizeof(struct tcphdr))
2157 return IPPROTO_DONE;
2158 if (m->m_len < hoff + iphlen + sizeof(struct tcphdr))
2159 return IPPROTO_DONE;
2160 th = (const struct tcphdr *)((const uint8_t *)ip + iphlen);
2161 thoff = th->th_off << 2;
2162 if (thoff < sizeof(struct tcphdr) || thoff + iphlen > iplen)
2163 return IPPROTO_DONE;
2164 if (m->m_len < hoff + iphlen + thoff)
2165 return IPPROTO_DONE;
2168 if (iplen < iphlen + sizeof(struct udphdr))
2169 return IPPROTO_DONE;
2170 if (m->m_len < hoff + iphlen + sizeof(struct udphdr))
2171 return IPPROTO_DONE;
2175 return IPPROTO_DONE;
2182 hn_create_rx_data(struct hn_softc *sc, int ring_cnt)
2184 struct sysctl_oid_list *child;
2185 struct sysctl_ctx_list *ctx;
2186 device_t dev = sc->hn_dev;
2187 #if defined(INET) || defined(INET6)
2188 #if __FreeBSD_version >= 1100095
2194 sc->hn_rx_ring_cnt = ring_cnt;
2195 sc->hn_rx_ring_inuse = sc->hn_rx_ring_cnt;
2197 sc->hn_rx_ring = malloc(sizeof(struct hn_rx_ring) * sc->hn_rx_ring_cnt,
2198 M_NETVSC, M_WAITOK | M_ZERO);
2200 #if defined(INET) || defined(INET6)
2201 #if __FreeBSD_version >= 1100095
2202 lroent_cnt = hn_lro_entry_count;
2203 if (lroent_cnt < TCP_LRO_ENTRIES)
2204 lroent_cnt = TCP_LRO_ENTRIES;
2205 device_printf(dev, "LRO: entry count %d\n", lroent_cnt);
2207 #endif /* INET || INET6 */
2209 ctx = device_get_sysctl_ctx(dev);
2210 child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
2212 /* Create dev.hn.UNIT.rx sysctl tree */
2213 sc->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "rx",
2214 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2216 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2217 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
2219 if (hn_trust_hosttcp)
2220 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_TCP;
2221 if (hn_trust_hostudp)
2222 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_UDP;
2223 if (hn_trust_hostip)
2224 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_IP;
2225 rxr->hn_ifp = sc->hn_ifp;
2231 #if defined(INET) || defined(INET6)
2232 #if __FreeBSD_version >= 1100095
2233 tcp_lro_init_args(&rxr->hn_lro, sc->hn_ifp, lroent_cnt,
2234 hn_lro_mbufq_depth);
2236 tcp_lro_init(&rxr->hn_lro);
2237 rxr->hn_lro.ifp = sc->hn_ifp;
2239 #if __FreeBSD_version >= 1100099
2240 rxr->hn_lro.lro_length_lim = HN_LRO_LENLIM_DEF;
2241 rxr->hn_lro.lro_ackcnt_lim = HN_LRO_ACKCNT_DEF;
2243 #endif /* INET || INET6 */
2245 if (sc->hn_rx_sysctl_tree != NULL) {
2249 * Create per RX ring sysctl tree:
2250 * dev.hn.UNIT.rx.RINGID
2252 snprintf(name, sizeof(name), "%d", i);
2253 rxr->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx,
2254 SYSCTL_CHILDREN(sc->hn_rx_sysctl_tree),
2255 OID_AUTO, name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2257 if (rxr->hn_rx_sysctl_tree != NULL) {
2258 SYSCTL_ADD_ULONG(ctx,
2259 SYSCTL_CHILDREN(rxr->hn_rx_sysctl_tree),
2260 OID_AUTO, "packets", CTLFLAG_RW,
2261 &rxr->hn_pkts, "# of packets received");
2262 SYSCTL_ADD_ULONG(ctx,
2263 SYSCTL_CHILDREN(rxr->hn_rx_sysctl_tree),
2264 OID_AUTO, "rss_pkts", CTLFLAG_RW,
2266 "# of packets w/ RSS info received");
2271 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_queued",
2272 CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2273 __offsetof(struct hn_rx_ring, hn_lro.lro_queued),
2274 hn_rx_stat_u64_sysctl, "LU", "LRO queued");
2275 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_flushed",
2276 CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2277 __offsetof(struct hn_rx_ring, hn_lro.lro_flushed),
2278 hn_rx_stat_u64_sysctl, "LU", "LRO flushed");
2279 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_tried",
2280 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2281 __offsetof(struct hn_rx_ring, hn_lro_tried),
2282 hn_rx_stat_ulong_sysctl, "LU", "# of LRO tries");
2283 #if __FreeBSD_version >= 1100099
2284 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_length_lim",
2285 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2286 hn_lro_lenlim_sysctl, "IU",
2287 "Max # of data bytes to be aggregated by LRO");
2288 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_ackcnt_lim",
2289 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2290 hn_lro_ackcnt_sysctl, "I",
2291 "Max # of ACKs to be aggregated by LRO");
2293 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hosttcp",
2294 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_TCP,
2295 hn_trust_hcsum_sysctl, "I",
2296 "Trust tcp segement verification on host side, "
2297 "when csum info is missing");
2298 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostudp",
2299 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_UDP,
2300 hn_trust_hcsum_sysctl, "I",
2301 "Trust udp datagram verification on host side, "
2302 "when csum info is missing");
2303 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostip",
2304 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_IP,
2305 hn_trust_hcsum_sysctl, "I",
2306 "Trust ip packet verification on host side, "
2307 "when csum info is missing");
2308 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_ip",
2309 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2310 __offsetof(struct hn_rx_ring, hn_csum_ip),
2311 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM IP");
2312 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_tcp",
2313 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2314 __offsetof(struct hn_rx_ring, hn_csum_tcp),
2315 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM TCP");
2316 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_udp",
2317 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2318 __offsetof(struct hn_rx_ring, hn_csum_udp),
2319 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM UDP");
2320 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_trusted",
2321 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2322 __offsetof(struct hn_rx_ring, hn_csum_trusted),
2323 hn_rx_stat_ulong_sysctl, "LU",
2324 "# of packets that we trust host's csum verification");
2325 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "small_pkts",
2326 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2327 __offsetof(struct hn_rx_ring, hn_small_pkts),
2328 hn_rx_stat_ulong_sysctl, "LU", "# of small packets received");
2329 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_cnt",
2330 CTLFLAG_RD, &sc->hn_rx_ring_cnt, 0, "# created RX rings");
2331 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_inuse",
2332 CTLFLAG_RD, &sc->hn_rx_ring_inuse, 0, "# used RX rings");
2336 hn_destroy_rx_data(struct hn_softc *sc)
2338 #if defined(INET) || defined(INET6)
2342 if (sc->hn_rx_ring_cnt == 0)
2345 #if defined(INET) || defined(INET6)
2346 for (i = 0; i < sc->hn_rx_ring_cnt; ++i)
2347 tcp_lro_free(&sc->hn_rx_ring[i].hn_lro);
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 hv_vmbus_channel *chan)
2918 struct hn_rx_ring *rxr;
2921 idx = chan->ch_subidx;
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;
2931 chan->hv_chan_rxr = rxr;
2933 if_printf(sc->hn_ifp, "link RX ring %d to channel%u\n",
2937 if (idx < sc->hn_tx_ring_inuse) {
2938 struct hn_tx_ring *txr = &sc->hn_tx_ring[idx];
2940 KASSERT((txr->hn_tx_flags & HN_TX_FLAG_ATTACHED) == 0,
2941 ("TX ring %d already attached", idx));
2942 txr->hn_tx_flags |= HN_TX_FLAG_ATTACHED;
2944 chan->hv_chan_txr = txr;
2945 txr->hn_chan = chan;
2947 if_printf(sc->hn_ifp, "link TX ring %d to channel%u\n",
2952 /* Bind channel to a proper CPU */
2953 vmbus_chan_cpu_set(chan, (sc->hn_cpu + idx) % mp_ncpus);
2957 hn_subchan_attach(struct hn_softc *sc, struct hv_vmbus_channel *chan)
2960 KASSERT(!VMBUS_CHAN_ISPRIMARY(chan),
2961 ("subchannel callback on primary channel"));
2962 KASSERT(chan->ch_subidx > 0,
2963 ("invalid channel subidx %u",
2965 hn_channel_attach(sc, chan);
2969 hn_subchan_setup(struct hn_softc *sc)
2971 struct hv_vmbus_channel **subchan;
2972 int subchan_cnt = sc->net_dev->num_channel - 1;
2975 /* Wait for sub-channels setup to complete. */
2976 subchan = vmbus_subchan_get(sc->hn_prichan, subchan_cnt);
2978 /* Attach the sub-channels. */
2979 for (i = 0; i < subchan_cnt; ++i) {
2980 /* NOTE: Calling order is critical. */
2981 hn_subchan_attach(sc, subchan[i]);
2982 hv_nv_subchan_attach(subchan[i]);
2985 /* Release the sub-channels */
2986 vmbus_subchan_rel(subchan, subchan_cnt);
2987 if_printf(sc->hn_ifp, "%d sub-channels setup done\n", subchan_cnt);
2991 hn_tx_taskq_create(void *arg __unused)
2993 if (!hn_share_tx_taskq)
2996 hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
2997 taskqueue_thread_enqueue, &hn_tx_taskq);
2998 if (hn_bind_tx_taskq >= 0) {
2999 int cpu = hn_bind_tx_taskq;
3002 if (cpu > mp_ncpus - 1)
3004 CPU_SETOF(cpu, &cpu_set);
3005 taskqueue_start_threads_cpuset(&hn_tx_taskq, 1, PI_NET,
3008 taskqueue_start_threads(&hn_tx_taskq, 1, PI_NET, "hn tx");
3011 SYSINIT(hn_txtq_create, SI_SUB_DRIVERS, SI_ORDER_FIRST,
3012 hn_tx_taskq_create, NULL);
3015 hn_tx_taskq_destroy(void *arg __unused)
3017 if (hn_tx_taskq != NULL)
3018 taskqueue_free(hn_tx_taskq);
3020 SYSUNINIT(hn_txtq_destroy, SI_SUB_DRIVERS, SI_ORDER_FIRST,
3021 hn_tx_taskq_destroy, NULL);
3023 static device_method_t netvsc_methods[] = {
3024 /* Device interface */
3025 DEVMETHOD(device_probe, netvsc_probe),
3026 DEVMETHOD(device_attach, netvsc_attach),
3027 DEVMETHOD(device_detach, netvsc_detach),
3028 DEVMETHOD(device_shutdown, netvsc_shutdown),
3033 static driver_t netvsc_driver = {
3039 static devclass_t netvsc_devclass;
3041 DRIVER_MODULE(hn, vmbus, netvsc_driver, netvsc_devclass, 0, 0);
3042 MODULE_VERSION(hn, 1);
3043 MODULE_DEPEND(hn, vmbus, 1, 1, 1);