2 * Copyright (c) 2010-2012 Citrix Inc.
3 * Copyright (c) 2009-2012,2016 Microsoft Corp.
4 * Copyright (c) 2012 NetApp Inc.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice unmodified, this list of conditions, and the following
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30 * Copyright (c) 2004-2006 Kip Macy
31 * All rights reserved.
33 * Redistribution and use in source and binary forms, with or without
34 * modification, are permitted provided that the following conditions
36 * 1. Redistributions of source code must retain the above copyright
37 * notice, this list of conditions and the following disclaimer.
38 * 2. Redistributions in binary form must reproduce the above copyright
39 * notice, this list of conditions and the following disclaimer in the
40 * documentation and/or other materials provided with the distribution.
42 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
43 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
44 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
45 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
46 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
47 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
48 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
49 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
50 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
51 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
55 #include <sys/cdefs.h>
56 __FBSDID("$FreeBSD$");
58 #include "opt_inet6.h"
61 #include <sys/param.h>
62 #include <sys/systm.h>
63 #include <sys/sockio.h>
65 #include <sys/malloc.h>
66 #include <sys/module.h>
67 #include <sys/kernel.h>
68 #include <sys/socket.h>
70 #include <sys/queue.h>
74 #include <sys/sysctl.h>
75 #include <sys/buf_ring.h>
78 #include <net/if_arp.h>
79 #include <net/ethernet.h>
80 #include <net/if_dl.h>
81 #include <net/if_media.h>
85 #include <net/if_types.h>
86 #include <net/if_vlan_var.h>
89 #include <netinet/in_systm.h>
90 #include <netinet/in.h>
91 #include <netinet/ip.h>
92 #include <netinet/if_ether.h>
93 #include <netinet/tcp.h>
94 #include <netinet/udp.h>
95 #include <netinet/ip6.h>
98 #include <vm/vm_param.h>
99 #include <vm/vm_kern.h>
102 #include <machine/bus.h>
103 #include <machine/resource.h>
104 #include <machine/frame.h>
105 #include <machine/vmparam.h>
108 #include <sys/rman.h>
109 #include <sys/mutex.h>
110 #include <sys/errno.h>
111 #include <sys/types.h>
112 #include <machine/atomic.h>
114 #include <machine/intr_machdep.h>
116 #include <machine/in_cksum.h>
118 #include <dev/hyperv/include/hyperv.h>
119 #include <dev/hyperv/include/hyperv_busdma.h>
121 #include "hv_net_vsc.h"
122 #include "hv_rndis.h"
123 #include "hv_rndis_filter.h"
124 #include "vmbus_if.h"
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 /* Limit TSO burst size */
245 static int hn_tso_maxlen = 0;
246 SYSCTL_INT(_hw_hn, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN,
247 &hn_tso_maxlen, 0, "TSO burst limit");
249 /* Limit chimney send size */
250 static int hn_tx_chimney_size = 0;
251 SYSCTL_INT(_hw_hn, OID_AUTO, tx_chimney_size, CTLFLAG_RDTUN,
252 &hn_tx_chimney_size, 0, "Chimney send packet size limit");
254 /* Limit the size of packet for direct transmission */
255 static int hn_direct_tx_size = HN_DIRECT_TX_SIZE_DEF;
256 SYSCTL_INT(_hw_hn, OID_AUTO, direct_tx_size, CTLFLAG_RDTUN,
257 &hn_direct_tx_size, 0, "Size of the packet for direct transmission");
259 #if defined(INET) || defined(INET6)
260 #if __FreeBSD_version >= 1100095
261 static int hn_lro_entry_count = HN_LROENT_CNT_DEF;
262 SYSCTL_INT(_hw_hn, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN,
263 &hn_lro_entry_count, 0, "LRO entry count");
267 static int hn_share_tx_taskq = 0;
268 SYSCTL_INT(_hw_hn, OID_AUTO, share_tx_taskq, CTLFLAG_RDTUN,
269 &hn_share_tx_taskq, 0, "Enable shared TX taskqueue");
271 static struct taskqueue *hn_tx_taskq;
273 #ifndef HN_USE_TXDESC_BUFRING
274 static int hn_use_txdesc_bufring = 0;
276 static int hn_use_txdesc_bufring = 1;
278 SYSCTL_INT(_hw_hn, OID_AUTO, use_txdesc_bufring, CTLFLAG_RD,
279 &hn_use_txdesc_bufring, 0, "Use buf_ring for TX descriptors");
281 static int hn_bind_tx_taskq = -1;
282 SYSCTL_INT(_hw_hn, OID_AUTO, bind_tx_taskq, CTLFLAG_RDTUN,
283 &hn_bind_tx_taskq, 0, "Bind TX taskqueue to the specified cpu");
285 static int hn_use_if_start = 0;
286 SYSCTL_INT(_hw_hn, OID_AUTO, use_if_start, CTLFLAG_RDTUN,
287 &hn_use_if_start, 0, "Use if_start TX method");
289 static int hn_chan_cnt = 0;
290 SYSCTL_INT(_hw_hn, OID_AUTO, chan_cnt, CTLFLAG_RDTUN,
292 "# of channels to use; each channel has one RX ring and one TX ring");
294 static int hn_tx_ring_cnt = 0;
295 SYSCTL_INT(_hw_hn, OID_AUTO, tx_ring_cnt, CTLFLAG_RDTUN,
296 &hn_tx_ring_cnt, 0, "# of TX rings to use");
298 static int hn_tx_swq_depth = 0;
299 SYSCTL_INT(_hw_hn, OID_AUTO, tx_swq_depth, CTLFLAG_RDTUN,
300 &hn_tx_swq_depth, 0, "Depth of IFQ or BUFRING");
302 #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 #if __FreeBSD_version < 1100095
329 static int hn_rx_stat_int_sysctl(SYSCTL_HANDLER_ARGS);
331 static int hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS);
333 static int hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS);
334 static int hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS);
335 static int hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS);
336 static int hn_check_iplen(const struct mbuf *, int);
337 static int hn_create_tx_ring(struct hn_softc *, int);
338 static void hn_destroy_tx_ring(struct hn_tx_ring *);
339 static int hn_create_tx_data(struct hn_softc *, int);
340 static void hn_destroy_tx_data(struct hn_softc *);
341 static void hn_start_taskfunc(void *, int);
342 static void hn_start_txeof_taskfunc(void *, int);
343 static void hn_stop_tx_tasks(struct hn_softc *);
344 static int hn_encap(struct hn_tx_ring *, struct hn_txdesc *, struct mbuf **);
345 static void hn_create_rx_data(struct hn_softc *sc, int);
346 static void hn_destroy_rx_data(struct hn_softc *sc);
347 static void hn_set_tx_chimney_size(struct hn_softc *, int);
348 static void hn_channel_attach(struct hn_softc *, struct vmbus_channel *);
349 static void hn_subchan_attach(struct hn_softc *, struct vmbus_channel *);
350 static void hn_subchan_setup(struct hn_softc *);
352 static int hn_transmit(struct ifnet *, struct mbuf *);
353 static void hn_xmit_qflush(struct ifnet *);
354 static int hn_xmit(struct hn_tx_ring *, int);
355 static void hn_xmit_txeof(struct hn_tx_ring *);
356 static void hn_xmit_taskfunc(void *, int);
357 static void hn_xmit_txeof_taskfunc(void *, int);
359 #if __FreeBSD_version >= 1100099
361 hn_set_lro_lenlim(struct hn_softc *sc, int lenlim)
365 for (i = 0; i < sc->hn_rx_ring_inuse; ++i)
366 sc->hn_rx_ring[i].hn_lro.lro_length_lim = lenlim;
371 hn_get_txswq_depth(const struct hn_tx_ring *txr)
374 KASSERT(txr->hn_txdesc_cnt > 0, ("tx ring is not setup yet"));
375 if (hn_tx_swq_depth < txr->hn_txdesc_cnt)
376 return txr->hn_txdesc_cnt;
377 return hn_tx_swq_depth;
381 hn_ifmedia_upd(struct ifnet *ifp __unused)
388 hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
390 struct hn_softc *sc = ifp->if_softc;
392 ifmr->ifm_status = IFM_AVALID;
393 ifmr->ifm_active = IFM_ETHER;
395 if (!sc->hn_carrier) {
396 ifmr->ifm_active |= IFM_NONE;
399 ifmr->ifm_status |= IFM_ACTIVE;
400 ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
403 /* {F8615163-DF3E-46c5-913F-F2D2F965ED0E} */
404 static const struct hyperv_guid g_net_vsc_device_type = {
405 .hv_guid = {0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
406 0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E}
410 * Standard probe entry point.
414 netvsc_probe(device_t dev)
416 if (VMBUS_PROBE_GUID(device_get_parent(dev), dev,
417 &g_net_vsc_device_type) == 0) {
418 device_set_desc(dev, "Hyper-V Network Interface");
419 return BUS_PROBE_DEFAULT;
425 hn_cpuset_setthread_task(void *xmask, int pending __unused)
427 cpuset_t *mask = xmask;
430 error = cpuset_setthread(curthread->td_tid, mask);
432 panic("curthread=%ju: can't pin; error=%d",
433 (uintmax_t)curthread->td_tid, error);
438 * Standard attach entry point.
440 * Called when the driver is loaded. It allocates needed resources,
441 * and initializes the "hardware" and software.
444 netvsc_attach(device_t dev)
446 netvsc_device_info device_info;
448 int unit = device_get_unit(dev);
449 struct ifnet *ifp = NULL;
450 int error, ring_cnt, tx_ring_cnt;
453 sc = device_get_softc(dev);
457 sc->hn_prichan = vmbus_get_channel(dev);
459 if (hn_tx_taskq == NULL) {
460 sc->hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
461 taskqueue_thread_enqueue, &sc->hn_tx_taskq);
462 taskqueue_start_threads(&sc->hn_tx_taskq, 1, PI_NET, "%s tx",
463 device_get_nameunit(dev));
464 if (hn_bind_tx_taskq >= 0) {
465 int cpu = hn_bind_tx_taskq;
466 struct task cpuset_task;
469 if (cpu > mp_ncpus - 1)
471 CPU_SETOF(cpu, &cpu_set);
472 TASK_INIT(&cpuset_task, 0, hn_cpuset_setthread_task,
474 taskqueue_enqueue(sc->hn_tx_taskq, &cpuset_task);
475 taskqueue_drain(sc->hn_tx_taskq, &cpuset_task);
478 sc->hn_tx_taskq = hn_tx_taskq;
480 NV_LOCK_INIT(sc, "NetVSCLock");
482 ifp = sc->hn_ifp = sc->arpcom.ac_ifp = if_alloc(IFT_ETHER);
484 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
487 * Figure out the # of RX rings (ring_cnt) and the # of TX rings
488 * to use (tx_ring_cnt).
491 * The # of RX rings to use is same as the # of channels to use.
493 ring_cnt = hn_chan_cnt;
497 if (ring_cnt > HN_RING_CNT_DEF_MAX)
498 ring_cnt = HN_RING_CNT_DEF_MAX;
499 } else if (ring_cnt > mp_ncpus) {
503 tx_ring_cnt = hn_tx_ring_cnt;
504 if (tx_ring_cnt <= 0 || tx_ring_cnt > ring_cnt)
505 tx_ring_cnt = ring_cnt;
506 if (hn_use_if_start) {
507 /* ifnet.if_start only needs one TX ring. */
512 * Set the leader CPU for channels.
514 sc->hn_cpu = atomic_fetchadd_int(&hn_cpu_index, ring_cnt) % mp_ncpus;
516 error = hn_create_tx_data(sc, tx_ring_cnt);
519 hn_create_rx_data(sc, ring_cnt);
522 * Associate the first TX/RX ring w/ the primary channel.
524 hn_channel_attach(sc, sc->hn_prichan);
526 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
527 ifp->if_ioctl = hn_ioctl;
528 ifp->if_init = hn_ifinit;
529 /* needed by hv_rf_on_device_add() code */
530 ifp->if_mtu = ETHERMTU;
531 if (hn_use_if_start) {
532 int qdepth = hn_get_txswq_depth(&sc->hn_tx_ring[0]);
534 ifp->if_start = hn_start;
535 IFQ_SET_MAXLEN(&ifp->if_snd, qdepth);
536 ifp->if_snd.ifq_drv_maxlen = qdepth - 1;
537 IFQ_SET_READY(&ifp->if_snd);
539 ifp->if_transmit = hn_transmit;
540 ifp->if_qflush = hn_xmit_qflush;
543 ifmedia_init(&sc->hn_media, 0, hn_ifmedia_upd, hn_ifmedia_sts);
544 ifmedia_add(&sc->hn_media, IFM_ETHER | IFM_AUTO, 0, NULL);
545 ifmedia_set(&sc->hn_media, IFM_ETHER | IFM_AUTO);
546 /* XXX ifmedia_set really should do this for us */
547 sc->hn_media.ifm_media = sc->hn_media.ifm_cur->ifm_media;
550 * Tell upper layers that we support full VLAN capability.
552 ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
553 ifp->if_capabilities |=
554 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
557 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
559 ifp->if_hwassist = sc->hn_tx_ring[0].hn_csum_assist | CSUM_TSO;
561 error = hv_rf_on_device_add(sc, &device_info, ring_cnt,
565 KASSERT(sc->net_dev->num_channel > 0 &&
566 sc->net_dev->num_channel <= sc->hn_rx_ring_inuse,
567 ("invalid channel count %u, should be less than %d",
568 sc->net_dev->num_channel, sc->hn_rx_ring_inuse));
571 * Set the # of TX/RX rings that could be used according to
572 * the # of channels that host offered.
574 if (sc->hn_tx_ring_inuse > sc->net_dev->num_channel)
575 sc->hn_tx_ring_inuse = sc->net_dev->num_channel;
576 sc->hn_rx_ring_inuse = sc->net_dev->num_channel;
577 device_printf(dev, "%d TX ring, %d RX ring\n",
578 sc->hn_tx_ring_inuse, sc->hn_rx_ring_inuse);
580 if (sc->net_dev->num_channel > 1)
581 hn_subchan_setup(sc);
583 #if __FreeBSD_version >= 1100099
584 if (sc->hn_rx_ring_inuse > 1) {
586 * Reduce TCP segment aggregation limit for multiple
587 * RX rings to increase ACK timeliness.
589 hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MULTIRX_DEF);
593 if (device_info.link_state == 0) {
597 tso_maxlen = hn_tso_maxlen;
598 if (tso_maxlen <= 0 || tso_maxlen > IP_MAXPACKET)
599 tso_maxlen = IP_MAXPACKET;
601 ifp->if_hw_tsomaxsegcount = HN_TX_DATA_SEGCNT_MAX;
602 ifp->if_hw_tsomaxsegsize = PAGE_SIZE;
603 ifp->if_hw_tsomax = tso_maxlen -
604 (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
606 ether_ifattach(ifp, device_info.mac_addr);
608 if_printf(ifp, "TSO: %u/%u/%u\n", ifp->if_hw_tsomax,
609 ifp->if_hw_tsomaxsegcount, ifp->if_hw_tsomaxsegsize);
611 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
612 hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max);
613 if (hn_tx_chimney_size > 0 &&
614 hn_tx_chimney_size < sc->hn_tx_chimney_max)
615 hn_set_tx_chimney_size(sc, hn_tx_chimney_size);
619 hn_destroy_tx_data(sc);
626 * Standard detach entry point
629 netvsc_detach(device_t dev)
631 struct hn_softc *sc = device_get_softc(dev);
634 printf("netvsc_detach\n");
637 * XXXKYS: Need to clean up all our
638 * driver state; this is the driver
643 * XXXKYS: Need to stop outgoing traffic and unregister
647 hv_rf_on_device_remove(sc, HV_RF_NV_DESTROY_CHANNEL);
649 hn_stop_tx_tasks(sc);
651 ifmedia_removeall(&sc->hn_media);
652 hn_destroy_rx_data(sc);
653 hn_destroy_tx_data(sc);
655 if (sc->hn_tx_taskq != hn_tx_taskq)
656 taskqueue_free(sc->hn_tx_taskq);
662 * Standard shutdown entry point
665 netvsc_shutdown(device_t dev)
671 hn_txdesc_dmamap_load(struct hn_tx_ring *txr, struct hn_txdesc *txd,
672 struct mbuf **m_head, bus_dma_segment_t *segs, int *nsegs)
674 struct mbuf *m = *m_head;
677 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag, txd->data_dmap,
678 m, segs, nsegs, BUS_DMA_NOWAIT);
679 if (error == EFBIG) {
682 m_new = m_collapse(m, M_NOWAIT, HN_TX_DATA_SEGCNT_MAX);
687 txr->hn_tx_collapsed++;
689 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag,
690 txd->data_dmap, m, segs, nsegs, BUS_DMA_NOWAIT);
693 bus_dmamap_sync(txr->hn_tx_data_dtag, txd->data_dmap,
694 BUS_DMASYNC_PREWRITE);
695 txd->flags |= HN_TXD_FLAG_DMAMAP;
701 hn_txdesc_dmamap_unload(struct hn_tx_ring *txr, struct hn_txdesc *txd)
704 if (txd->flags & HN_TXD_FLAG_DMAMAP) {
705 bus_dmamap_sync(txr->hn_tx_data_dtag,
706 txd->data_dmap, BUS_DMASYNC_POSTWRITE);
707 bus_dmamap_unload(txr->hn_tx_data_dtag,
709 txd->flags &= ~HN_TXD_FLAG_DMAMAP;
714 hn_txdesc_put(struct hn_tx_ring *txr, struct hn_txdesc *txd)
717 KASSERT((txd->flags & HN_TXD_FLAG_ONLIST) == 0,
718 ("put an onlist txd %#x", txd->flags));
720 KASSERT(txd->refs > 0, ("invalid txd refs %d", txd->refs));
721 if (atomic_fetchadd_int(&txd->refs, -1) != 1)
724 hn_txdesc_dmamap_unload(txr, txd);
725 if (txd->m != NULL) {
730 txd->flags |= HN_TXD_FLAG_ONLIST;
732 #ifndef HN_USE_TXDESC_BUFRING
733 mtx_lock_spin(&txr->hn_txlist_spin);
734 KASSERT(txr->hn_txdesc_avail >= 0 &&
735 txr->hn_txdesc_avail < txr->hn_txdesc_cnt,
736 ("txdesc_put: invalid txd avail %d", txr->hn_txdesc_avail));
737 txr->hn_txdesc_avail++;
738 SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
739 mtx_unlock_spin(&txr->hn_txlist_spin);
741 atomic_add_int(&txr->hn_txdesc_avail, 1);
742 buf_ring_enqueue(txr->hn_txdesc_br, txd);
748 static __inline struct hn_txdesc *
749 hn_txdesc_get(struct hn_tx_ring *txr)
751 struct hn_txdesc *txd;
753 #ifndef HN_USE_TXDESC_BUFRING
754 mtx_lock_spin(&txr->hn_txlist_spin);
755 txd = SLIST_FIRST(&txr->hn_txlist);
757 KASSERT(txr->hn_txdesc_avail > 0,
758 ("txdesc_get: invalid txd avail %d", txr->hn_txdesc_avail));
759 txr->hn_txdesc_avail--;
760 SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
762 mtx_unlock_spin(&txr->hn_txlist_spin);
764 txd = buf_ring_dequeue_sc(txr->hn_txdesc_br);
768 #ifdef HN_USE_TXDESC_BUFRING
769 atomic_subtract_int(&txr->hn_txdesc_avail, 1);
771 KASSERT(txd->m == NULL && txd->refs == 0 &&
772 (txd->flags & HN_TXD_FLAG_ONLIST), ("invalid txd"));
773 txd->flags &= ~HN_TXD_FLAG_ONLIST;
780 hn_txdesc_hold(struct hn_txdesc *txd)
783 /* 0->1 transition will never work */
784 KASSERT(txd->refs > 0, ("invalid refs %d", txd->refs));
785 atomic_add_int(&txd->refs, 1);
789 hn_txeof(struct hn_tx_ring *txr)
791 txr->hn_has_txeof = 0;
796 hn_tx_done(struct vmbus_channel *chan, void *xpkt)
798 netvsc_packet *packet = xpkt;
799 struct hn_txdesc *txd;
800 struct hn_tx_ring *txr;
802 txd = (struct hn_txdesc *)(uintptr_t)
803 packet->compl.send.send_completion_tid;
806 KASSERT(txr->hn_chan == chan,
807 ("channel mismatch, on chan%u, should be chan%u",
808 vmbus_chan_subidx(chan), vmbus_chan_subidx(txr->hn_chan)));
810 txr->hn_has_txeof = 1;
811 hn_txdesc_put(txr, txd);
813 ++txr->hn_txdone_cnt;
814 if (txr->hn_txdone_cnt >= HN_EARLY_TXEOF_THRESH) {
815 txr->hn_txdone_cnt = 0;
822 netvsc_channel_rollup(struct hn_rx_ring *rxr, struct hn_tx_ring *txr)
824 #if defined(INET) || defined(INET6)
825 struct lro_ctrl *lro = &rxr->hn_lro;
826 struct lro_entry *queued;
828 while ((queued = SLIST_FIRST(&lro->lro_active)) != NULL) {
829 SLIST_REMOVE_HEAD(&lro->lro_active, next);
830 tcp_lro_flush(lro, queued);
836 * 'txr' could be NULL, if multiple channels and
837 * ifnet.if_start method are enabled.
839 if (txr == NULL || !txr->hn_has_txeof)
842 txr->hn_txdone_cnt = 0;
848 * If this function fails, then both txd and m_head0 will be freed.
851 hn_encap(struct hn_tx_ring *txr, struct hn_txdesc *txd, struct mbuf **m_head0)
853 bus_dma_segment_t segs[HN_TX_DATA_SEGCNT_MAX];
855 struct mbuf *m_head = *m_head0;
856 netvsc_packet *packet;
857 rndis_msg *rndis_mesg;
858 rndis_packet *rndis_pkt;
859 rndis_per_packet_info *rppi;
860 struct rndis_hash_value *hash_value;
861 uint32_t rndis_msg_size;
863 packet = &txd->netvsc_pkt;
864 packet->is_data_pkt = TRUE;
865 packet->tot_data_buf_len = m_head->m_pkthdr.len;
868 * extension points to the area reserved for the
869 * rndis_filter_packet, which is placed just after
870 * the netvsc_packet (and rppi struct, if present;
871 * length is updated later).
873 rndis_mesg = txd->rndis_msg;
874 /* XXX not necessary */
875 memset(rndis_mesg, 0, HN_RNDIS_MSG_LEN);
876 rndis_mesg->ndis_msg_type = REMOTE_NDIS_PACKET_MSG;
878 rndis_pkt = &rndis_mesg->msg.packet;
879 rndis_pkt->data_offset = sizeof(rndis_packet);
880 rndis_pkt->data_length = packet->tot_data_buf_len;
881 rndis_pkt->per_pkt_info_offset = sizeof(rndis_packet);
883 rndis_msg_size = RNDIS_MESSAGE_SIZE(rndis_packet);
886 * Set the hash value for this packet, so that the host could
887 * dispatch the TX done event for this packet back to this TX
890 rndis_msg_size += RNDIS_HASHVAL_PPI_SIZE;
891 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_HASHVAL_PPI_SIZE,
893 hash_value = (struct rndis_hash_value *)((uint8_t *)rppi +
894 rppi->per_packet_info_offset);
895 hash_value->hash_value = txr->hn_tx_idx;
897 if (m_head->m_flags & M_VLANTAG) {
898 ndis_8021q_info *rppi_vlan_info;
900 rndis_msg_size += RNDIS_VLAN_PPI_SIZE;
901 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_VLAN_PPI_SIZE,
904 rppi_vlan_info = (ndis_8021q_info *)((uint8_t *)rppi +
905 rppi->per_packet_info_offset);
906 rppi_vlan_info->u1.s1.vlan_id =
907 m_head->m_pkthdr.ether_vtag & 0xfff;
910 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
911 rndis_tcp_tso_info *tso_info;
912 struct ether_vlan_header *eh;
916 * XXX need m_pullup and use mtodo
918 eh = mtod(m_head, struct ether_vlan_header*);
919 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN))
920 ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
922 ether_len = ETHER_HDR_LEN;
924 rndis_msg_size += RNDIS_TSO_PPI_SIZE;
925 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_TSO_PPI_SIZE,
926 tcp_large_send_info);
928 tso_info = (rndis_tcp_tso_info *)((uint8_t *)rppi +
929 rppi->per_packet_info_offset);
930 tso_info->lso_v2_xmit.type =
931 RNDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
934 if (m_head->m_pkthdr.csum_flags & CSUM_IP_TSO) {
936 (struct ip *)(m_head->m_data + ether_len);
937 unsigned long iph_len = ip->ip_hl << 2;
939 (struct tcphdr *)((caddr_t)ip + iph_len);
941 tso_info->lso_v2_xmit.ip_version =
942 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
946 th->th_sum = in_pseudo(ip->ip_src.s_addr,
947 ip->ip_dst.s_addr, htons(IPPROTO_TCP));
950 #if defined(INET6) && defined(INET)
955 struct ip6_hdr *ip6 = (struct ip6_hdr *)
956 (m_head->m_data + ether_len);
957 struct tcphdr *th = (struct tcphdr *)(ip6 + 1);
959 tso_info->lso_v2_xmit.ip_version =
960 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
962 th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
965 tso_info->lso_v2_xmit.tcp_header_offset = 0;
966 tso_info->lso_v2_xmit.mss = m_head->m_pkthdr.tso_segsz;
967 } else if (m_head->m_pkthdr.csum_flags & txr->hn_csum_assist) {
968 rndis_tcp_ip_csum_info *csum_info;
970 rndis_msg_size += RNDIS_CSUM_PPI_SIZE;
971 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_CSUM_PPI_SIZE,
973 csum_info = (rndis_tcp_ip_csum_info *)((uint8_t *)rppi +
974 rppi->per_packet_info_offset);
976 csum_info->xmit.is_ipv4 = 1;
977 if (m_head->m_pkthdr.csum_flags & CSUM_IP)
978 csum_info->xmit.ip_header_csum = 1;
980 if (m_head->m_pkthdr.csum_flags & CSUM_TCP) {
981 csum_info->xmit.tcp_csum = 1;
982 csum_info->xmit.tcp_header_offset = 0;
983 } else if (m_head->m_pkthdr.csum_flags & CSUM_UDP) {
984 csum_info->xmit.udp_csum = 1;
988 rndis_mesg->msg_len = packet->tot_data_buf_len + rndis_msg_size;
989 packet->tot_data_buf_len = rndis_mesg->msg_len;
992 * Chimney send, if the packet could fit into one chimney buffer.
994 if (packet->tot_data_buf_len < txr->hn_tx_chimney_size) {
995 netvsc_dev *net_dev = txr->hn_sc->net_dev;
996 uint32_t send_buf_section_idx;
998 txr->hn_tx_chimney_tried++;
999 send_buf_section_idx =
1000 hv_nv_get_next_send_section(net_dev);
1001 if (send_buf_section_idx !=
1002 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX) {
1003 uint8_t *dest = ((uint8_t *)net_dev->send_buf +
1004 (send_buf_section_idx *
1005 net_dev->send_section_size));
1007 memcpy(dest, rndis_mesg, rndis_msg_size);
1008 dest += rndis_msg_size;
1009 m_copydata(m_head, 0, m_head->m_pkthdr.len, dest);
1011 packet->send_buf_section_idx = send_buf_section_idx;
1012 packet->send_buf_section_size =
1013 packet->tot_data_buf_len;
1014 packet->gpa_cnt = 0;
1015 txr->hn_tx_chimney++;
1020 error = hn_txdesc_dmamap_load(txr, txd, &m_head, segs, &nsegs);
1025 * This mbuf is not linked w/ the txd yet, so free it now.
1030 freed = hn_txdesc_put(txr, txd);
1032 ("fail to free txd upon txdma error"));
1034 txr->hn_txdma_failed++;
1035 if_inc_counter(txr->hn_sc->hn_ifp, IFCOUNTER_OERRORS, 1);
1040 packet->gpa_cnt = nsegs + HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
1042 /* send packet with page buffer */
1043 packet->gpa[0].gpa_page = atop(txd->rndis_msg_paddr);
1044 packet->gpa[0].gpa_ofs = txd->rndis_msg_paddr & PAGE_MASK;
1045 packet->gpa[0].gpa_len = rndis_msg_size;
1048 * Fill the page buffers with mbuf info starting at index
1049 * HV_RF_NUM_TX_RESERVED_PAGE_BUFS.
1051 for (i = 0; i < nsegs; ++i) {
1052 struct vmbus_gpa *gpa = &packet->gpa[
1053 i + HV_RF_NUM_TX_RESERVED_PAGE_BUFS];
1055 gpa->gpa_page = atop(segs[i].ds_addr);
1056 gpa->gpa_ofs = segs[i].ds_addr & PAGE_MASK;
1057 gpa->gpa_len = segs[i].ds_len;
1060 packet->send_buf_section_idx =
1061 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX;
1062 packet->send_buf_section_size = 0;
1066 /* Set the completion routine */
1067 packet->compl.send.on_send_completion = hn_tx_done;
1068 packet->compl.send.send_completion_context = packet;
1069 packet->compl.send.send_completion_tid = (uint64_t)(uintptr_t)txd;
1076 * If this function fails, then txd will be freed, but the mbuf
1077 * associated w/ the txd will _not_ be freed.
1080 hn_send_pkt(struct ifnet *ifp, struct hn_tx_ring *txr, struct hn_txdesc *txd)
1082 int error, send_failed = 0;
1086 * Make sure that txd is not freed before ETHER_BPF_MTAP.
1088 hn_txdesc_hold(txd);
1089 error = hv_nv_on_send(txr->hn_chan, &txd->netvsc_pkt);
1091 ETHER_BPF_MTAP(ifp, txd->m);
1092 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1093 if (!hn_use_if_start) {
1094 if_inc_counter(ifp, IFCOUNTER_OBYTES,
1095 txd->m->m_pkthdr.len);
1096 if (txd->m->m_flags & M_MCAST)
1097 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
1101 hn_txdesc_put(txr, txd);
1103 if (__predict_false(error)) {
1107 * This should "really rarely" happen.
1109 * XXX Too many RX to be acked or too many sideband
1110 * commands to run? Ask netvsc_channel_rollup()
1111 * to kick start later.
1113 txr->hn_has_txeof = 1;
1115 txr->hn_send_failed++;
1118 * Try sending again after set hn_has_txeof;
1119 * in case that we missed the last
1120 * netvsc_channel_rollup().
1124 if_printf(ifp, "send failed\n");
1127 * Caller will perform further processing on the
1128 * associated mbuf, so don't free it in hn_txdesc_put();
1129 * only unload it from the DMA map in hn_txdesc_put(),
1133 freed = hn_txdesc_put(txr, txd);
1135 ("fail to free txd upon send error"));
1137 txr->hn_send_failed++;
1143 * Start a transmit of one or more packets
1146 hn_start_locked(struct hn_tx_ring *txr, int len)
1148 struct hn_softc *sc = txr->hn_sc;
1149 struct ifnet *ifp = sc->hn_ifp;
1151 KASSERT(hn_use_if_start,
1152 ("hn_start_locked is called, when if_start is disabled"));
1153 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1154 mtx_assert(&txr->hn_tx_lock, MA_OWNED);
1156 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1160 while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
1161 struct hn_txdesc *txd;
1162 struct mbuf *m_head;
1165 IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
1169 if (len > 0 && m_head->m_pkthdr.len > len) {
1171 * This sending could be time consuming; let callers
1172 * dispatch this packet sending (and sending of any
1173 * following up packets) to tx taskqueue.
1175 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1179 txd = hn_txdesc_get(txr);
1181 txr->hn_no_txdescs++;
1182 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1183 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1187 error = hn_encap(txr, txd, &m_head);
1189 /* Both txd and m_head are freed */
1193 error = hn_send_pkt(ifp, txr, txd);
1194 if (__predict_false(error)) {
1195 /* txd is freed, but m_head is not */
1196 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1197 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1205 * Link up/down notification
1208 netvsc_linkstatus_callback(struct hn_softc *sc, uint32_t status)
1218 * Append the specified data to the indicated mbuf chain,
1219 * Extend the mbuf chain if the new data does not fit in
1222 * This is a minor rewrite of m_append() from sys/kern/uipc_mbuf.c.
1223 * There should be an equivalent in the kernel mbuf code,
1224 * but there does not appear to be one yet.
1226 * Differs from m_append() in that additional mbufs are
1227 * allocated with cluster size MJUMPAGESIZE, and filled
1230 * Return 1 if able to complete the job; otherwise 0.
1233 hv_m_append(struct mbuf *m0, int len, c_caddr_t cp)
1236 int remainder, space;
1238 for (m = m0; m->m_next != NULL; m = m->m_next)
1241 space = M_TRAILINGSPACE(m);
1244 * Copy into available space.
1246 if (space > remainder)
1248 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1253 while (remainder > 0) {
1255 * Allocate a new mbuf; could check space
1256 * and allocate a cluster instead.
1258 n = m_getjcl(M_DONTWAIT, m->m_type, 0, MJUMPAGESIZE);
1261 n->m_len = min(MJUMPAGESIZE, remainder);
1262 bcopy(cp, mtod(n, caddr_t), n->m_len);
1264 remainder -= n->m_len;
1268 if (m0->m_flags & M_PKTHDR)
1269 m0->m_pkthdr.len += len - remainder;
1271 return (remainder == 0);
1274 #if defined(INET) || defined(INET6)
1276 hn_lro_rx(struct lro_ctrl *lc, struct mbuf *m)
1278 #if __FreeBSD_version >= 1100095
1279 if (hn_lro_mbufq_depth) {
1280 tcp_lro_queue_mbuf(lc, m);
1284 return tcp_lro_rx(lc, m, 0);
1289 * Called when we receive a data packet from the "wire" on the
1292 * Note: This is no longer used as a callback
1295 netvsc_recv(struct hn_rx_ring *rxr, netvsc_packet *packet,
1296 const rndis_tcp_ip_csum_info *csum_info,
1297 const struct rndis_hash_info *hash_info,
1298 const struct rndis_hash_value *hash_value)
1300 struct ifnet *ifp = rxr->hn_ifp;
1302 int size, do_lro = 0, do_csum = 1;
1304 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1308 * Bail out if packet contains more data than configured MTU.
1310 if (packet->tot_data_buf_len > (ifp->if_mtu + ETHER_HDR_LEN)) {
1312 } else if (packet->tot_data_buf_len <= MHLEN) {
1313 m_new = m_gethdr(M_NOWAIT, MT_DATA);
1314 if (m_new == NULL) {
1315 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1318 memcpy(mtod(m_new, void *), packet->data,
1319 packet->tot_data_buf_len);
1320 m_new->m_pkthdr.len = m_new->m_len = packet->tot_data_buf_len;
1321 rxr->hn_small_pkts++;
1324 * Get an mbuf with a cluster. For packets 2K or less,
1325 * get a standard 2K cluster. For anything larger, get a
1326 * 4K cluster. Any buffers larger than 4K can cause problems
1327 * if looped around to the Hyper-V TX channel, so avoid them.
1330 if (packet->tot_data_buf_len > MCLBYTES) {
1332 size = MJUMPAGESIZE;
1335 m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, size);
1336 if (m_new == NULL) {
1337 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1341 hv_m_append(m_new, packet->tot_data_buf_len, packet->data);
1343 m_new->m_pkthdr.rcvif = ifp;
1345 if (__predict_false((ifp->if_capenable & IFCAP_RXCSUM) == 0))
1348 /* receive side checksum offload */
1349 if (csum_info != NULL) {
1350 /* IP csum offload */
1351 if (csum_info->receive.ip_csum_succeeded && do_csum) {
1352 m_new->m_pkthdr.csum_flags |=
1353 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1357 /* TCP/UDP csum offload */
1358 if ((csum_info->receive.tcp_csum_succeeded ||
1359 csum_info->receive.udp_csum_succeeded) && do_csum) {
1360 m_new->m_pkthdr.csum_flags |=
1361 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1362 m_new->m_pkthdr.csum_data = 0xffff;
1363 if (csum_info->receive.tcp_csum_succeeded)
1369 if (csum_info->receive.ip_csum_succeeded &&
1370 csum_info->receive.tcp_csum_succeeded)
1373 const struct ether_header *eh;
1378 if (m_new->m_len < hoff)
1380 eh = mtod(m_new, struct ether_header *);
1381 etype = ntohs(eh->ether_type);
1382 if (etype == ETHERTYPE_VLAN) {
1383 const struct ether_vlan_header *evl;
1385 hoff = sizeof(*evl);
1386 if (m_new->m_len < hoff)
1388 evl = mtod(m_new, struct ether_vlan_header *);
1389 etype = ntohs(evl->evl_proto);
1392 if (etype == ETHERTYPE_IP) {
1395 pr = hn_check_iplen(m_new, hoff);
1396 if (pr == IPPROTO_TCP) {
1398 (rxr->hn_trust_hcsum &
1399 HN_TRUST_HCSUM_TCP)) {
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;
1407 } else if (pr == IPPROTO_UDP) {
1409 (rxr->hn_trust_hcsum &
1410 HN_TRUST_HCSUM_UDP)) {
1411 rxr->hn_csum_trusted++;
1412 m_new->m_pkthdr.csum_flags |=
1413 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1414 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1415 m_new->m_pkthdr.csum_data = 0xffff;
1417 } else if (pr != IPPROTO_DONE && do_csum &&
1418 (rxr->hn_trust_hcsum & HN_TRUST_HCSUM_IP)) {
1419 rxr->hn_csum_trusted++;
1420 m_new->m_pkthdr.csum_flags |=
1421 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1426 if ((packet->vlan_tci != 0) &&
1427 (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0) {
1428 m_new->m_pkthdr.ether_vtag = packet->vlan_tci;
1429 m_new->m_flags |= M_VLANTAG;
1432 if (hash_info != NULL && hash_value != NULL) {
1433 int hash_type = M_HASHTYPE_OPAQUE;
1436 m_new->m_pkthdr.flowid = hash_value->hash_value;
1437 if ((hash_info->hash_info & NDIS_HASH_FUNCTION_MASK) ==
1438 NDIS_HASH_FUNCTION_TOEPLITZ) {
1440 (hash_info->hash_info & NDIS_HASH_TYPE_MASK);
1443 case NDIS_HASH_IPV4:
1444 hash_type = M_HASHTYPE_RSS_IPV4;
1447 case NDIS_HASH_TCP_IPV4:
1448 hash_type = M_HASHTYPE_RSS_TCP_IPV4;
1451 case NDIS_HASH_IPV6:
1452 hash_type = M_HASHTYPE_RSS_IPV6;
1455 case NDIS_HASH_IPV6_EX:
1456 hash_type = M_HASHTYPE_RSS_IPV6_EX;
1459 case NDIS_HASH_TCP_IPV6:
1460 hash_type = M_HASHTYPE_RSS_TCP_IPV6;
1463 case NDIS_HASH_TCP_IPV6_EX:
1464 hash_type = M_HASHTYPE_RSS_TCP_IPV6_EX;
1468 M_HASHTYPE_SET(m_new, hash_type);
1470 if (hash_value != NULL)
1471 m_new->m_pkthdr.flowid = hash_value->hash_value;
1473 m_new->m_pkthdr.flowid = rxr->hn_rx_idx;
1474 M_HASHTYPE_SET(m_new, M_HASHTYPE_OPAQUE);
1478 * Note: Moved RX completion back to hv_nv_on_receive() so all
1479 * messages (not just data messages) will trigger a response.
1485 if ((ifp->if_capenable & IFCAP_LRO) && do_lro) {
1486 #if defined(INET) || defined(INET6)
1487 struct lro_ctrl *lro = &rxr->hn_lro;
1490 rxr->hn_lro_tried++;
1491 if (hn_lro_rx(lro, m_new) == 0) {
1499 /* We're not holding the lock here, so don't release it */
1500 (*ifp->if_input)(ifp, m_new);
1506 * Rules for using sc->temp_unusable:
1507 * 1. sc->temp_unusable can only be read or written while holding NV_LOCK()
1508 * 2. code reading sc->temp_unusable under NV_LOCK(), and finding
1509 * sc->temp_unusable set, must release NV_LOCK() and exit
1510 * 3. to retain exclusive control of the interface,
1511 * sc->temp_unusable must be set by code before releasing NV_LOCK()
1512 * 4. only code setting sc->temp_unusable can clear sc->temp_unusable
1513 * 5. code setting sc->temp_unusable must eventually clear sc->temp_unusable
1517 * Standard ioctl entry point. Called when the user wants to configure
1521 hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1523 hn_softc_t *sc = ifp->if_softc;
1524 struct ifreq *ifr = (struct ifreq *)data;
1526 struct ifaddr *ifa = (struct ifaddr *)data;
1528 netvsc_device_info device_info;
1529 int mask, error = 0;
1530 int retry_cnt = 500;
1536 if (ifa->ifa_addr->sa_family == AF_INET) {
1537 ifp->if_flags |= IFF_UP;
1538 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1540 arp_ifinit(ifp, ifa);
1543 error = ether_ioctl(ifp, cmd, data);
1546 /* Check MTU value change */
1547 if (ifp->if_mtu == ifr->ifr_mtu)
1550 if (ifr->ifr_mtu > NETVSC_MAX_CONFIGURABLE_MTU) {
1555 /* Obtain and record requested MTU */
1556 ifp->if_mtu = ifr->ifr_mtu;
1558 #if __FreeBSD_version >= 1100099
1560 * Make sure that LRO aggregation length limit is still
1561 * valid, after the MTU change.
1564 if (sc->hn_rx_ring[0].hn_lro.lro_length_lim <
1565 HN_LRO_LENLIM_MIN(ifp))
1566 hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MIN(ifp));
1572 if (!sc->temp_unusable) {
1573 sc->temp_unusable = TRUE;
1577 if (retry_cnt > 0) {
1581 } while (retry_cnt > 0);
1583 if (retry_cnt == 0) {
1588 /* We must remove and add back the device to cause the new
1589 * MTU to take effect. This includes tearing down, but not
1590 * deleting the channel, then bringing it back up.
1592 error = hv_rf_on_device_remove(sc, HV_RF_NV_RETAIN_CHANNEL);
1595 sc->temp_unusable = FALSE;
1600 /* Wait for subchannels to be destroyed */
1601 vmbus_subchan_drain(sc->hn_prichan);
1603 error = hv_rf_on_device_add(sc, &device_info,
1604 sc->hn_rx_ring_inuse, &sc->hn_rx_ring[0]);
1607 sc->temp_unusable = FALSE;
1611 KASSERT(sc->hn_rx_ring_cnt == sc->net_dev->num_channel,
1612 ("RX ring count %d and channel count %u mismatch",
1613 sc->hn_rx_ring_cnt, sc->net_dev->num_channel));
1614 if (sc->net_dev->num_channel > 1) {
1618 * Skip the rings on primary channel; they are
1619 * handled by the hv_rf_on_device_add() above.
1621 for (r = 1; r < sc->hn_rx_ring_cnt; ++r) {
1622 sc->hn_rx_ring[r].hn_rx_flags &=
1623 ~HN_RX_FLAG_ATTACHED;
1625 for (r = 1; r < sc->hn_tx_ring_cnt; ++r) {
1626 sc->hn_tx_ring[r].hn_tx_flags &=
1627 ~HN_TX_FLAG_ATTACHED;
1629 hn_subchan_setup(sc);
1632 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
1633 if (sc->hn_tx_ring[0].hn_tx_chimney_size >
1634 sc->hn_tx_chimney_max)
1635 hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max);
1637 hn_ifinit_locked(sc);
1640 sc->temp_unusable = FALSE;
1646 if (!sc->temp_unusable) {
1647 sc->temp_unusable = TRUE;
1651 if (retry_cnt > 0) {
1655 } while (retry_cnt > 0);
1657 if (retry_cnt == 0) {
1662 if (ifp->if_flags & IFF_UP) {
1664 * If only the state of the PROMISC flag changed,
1665 * then just use the 'set promisc mode' command
1666 * instead of reinitializing the entire NIC. Doing
1667 * a full re-init means reloading the firmware and
1668 * waiting for it to start up, which may take a
1672 /* Fixme: Promiscuous mode? */
1673 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1674 ifp->if_flags & IFF_PROMISC &&
1675 !(sc->hn_if_flags & IFF_PROMISC)) {
1676 /* do something here for Hyper-V */
1677 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1678 !(ifp->if_flags & IFF_PROMISC) &&
1679 sc->hn_if_flags & IFF_PROMISC) {
1680 /* do something here for Hyper-V */
1683 hn_ifinit_locked(sc);
1685 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1690 sc->temp_unusable = FALSE;
1692 sc->hn_if_flags = ifp->if_flags;
1698 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1699 if (mask & IFCAP_TXCSUM) {
1700 ifp->if_capenable ^= IFCAP_TXCSUM;
1701 if (ifp->if_capenable & IFCAP_TXCSUM) {
1703 sc->hn_tx_ring[0].hn_csum_assist;
1706 ~sc->hn_tx_ring[0].hn_csum_assist;
1710 if (mask & IFCAP_RXCSUM)
1711 ifp->if_capenable ^= IFCAP_RXCSUM;
1713 if (mask & IFCAP_LRO)
1714 ifp->if_capenable ^= IFCAP_LRO;
1716 if (mask & IFCAP_TSO4) {
1717 ifp->if_capenable ^= IFCAP_TSO4;
1718 if (ifp->if_capenable & IFCAP_TSO4)
1719 ifp->if_hwassist |= CSUM_IP_TSO;
1721 ifp->if_hwassist &= ~CSUM_IP_TSO;
1724 if (mask & IFCAP_TSO6) {
1725 ifp->if_capenable ^= IFCAP_TSO6;
1726 if (ifp->if_capenable & IFCAP_TSO6)
1727 ifp->if_hwassist |= CSUM_IP6_TSO;
1729 ifp->if_hwassist &= ~CSUM_IP6_TSO;
1738 /* Fixme: Multicast mode? */
1739 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1741 netvsc_setmulti(sc);
1750 error = ifmedia_ioctl(ifp, ifr, &sc->hn_media, cmd);
1753 error = ether_ioctl(ifp, cmd, data);
1764 hn_stop(hn_softc_t *sc)
1772 printf(" Closing Device ...\n");
1774 atomic_clear_int(&ifp->if_drv_flags,
1775 (IFF_DRV_RUNNING | IFF_DRV_OACTIVE));
1776 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
1777 sc->hn_tx_ring[i].hn_oactive = 0;
1779 if_link_state_change(ifp, LINK_STATE_DOWN);
1780 sc->hn_initdone = 0;
1782 ret = hv_rf_on_close(sc);
1786 * FreeBSD transmit entry point
1789 hn_start(struct ifnet *ifp)
1791 struct hn_softc *sc = ifp->if_softc;
1792 struct hn_tx_ring *txr = &sc->hn_tx_ring[0];
1794 if (txr->hn_sched_tx)
1797 if (mtx_trylock(&txr->hn_tx_lock)) {
1800 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1801 mtx_unlock(&txr->hn_tx_lock);
1806 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
1810 hn_start_txeof(struct hn_tx_ring *txr)
1812 struct hn_softc *sc = txr->hn_sc;
1813 struct ifnet *ifp = sc->hn_ifp;
1815 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1817 if (txr->hn_sched_tx)
1820 if (mtx_trylock(&txr->hn_tx_lock)) {
1823 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1824 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1825 mtx_unlock(&txr->hn_tx_lock);
1827 taskqueue_enqueue(txr->hn_tx_taskq,
1833 * Release the OACTIVE earlier, with the hope, that
1834 * others could catch up. The task will clear the
1835 * flag again with the hn_tx_lock to avoid possible
1838 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1839 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
1847 hn_ifinit_locked(hn_softc_t *sc)
1854 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1858 hv_promisc_mode = 1;
1860 ret = hv_rf_on_open(sc);
1864 sc->hn_initdone = 1;
1867 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1868 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
1869 sc->hn_tx_ring[i].hn_oactive = 0;
1871 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_RUNNING);
1872 if_link_state_change(ifp, LINK_STATE_UP);
1879 hn_ifinit(void *xsc)
1881 hn_softc_t *sc = xsc;
1884 if (sc->temp_unusable) {
1888 sc->temp_unusable = TRUE;
1891 hn_ifinit_locked(sc);
1894 sc->temp_unusable = FALSE;
1903 hn_watchdog(struct ifnet *ifp)
1908 printf("hn%d: watchdog timeout -- resetting\n", sc->hn_unit);
1909 hn_ifinit(sc); /*???*/
1914 #if __FreeBSD_version >= 1100099
1917 hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS)
1919 struct hn_softc *sc = arg1;
1920 unsigned int lenlim;
1923 lenlim = sc->hn_rx_ring[0].hn_lro.lro_length_lim;
1924 error = sysctl_handle_int(oidp, &lenlim, 0, req);
1925 if (error || req->newptr == NULL)
1928 if (lenlim < HN_LRO_LENLIM_MIN(sc->hn_ifp) ||
1929 lenlim > TCP_LRO_LENGTH_MAX)
1933 hn_set_lro_lenlim(sc, lenlim);
1939 hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS)
1941 struct hn_softc *sc = arg1;
1942 int ackcnt, error, i;
1945 * lro_ackcnt_lim is append count limit,
1946 * +1 to turn it into aggregation limit.
1948 ackcnt = sc->hn_rx_ring[0].hn_lro.lro_ackcnt_lim + 1;
1949 error = sysctl_handle_int(oidp, &ackcnt, 0, req);
1950 if (error || req->newptr == NULL)
1953 if (ackcnt < 2 || ackcnt > (TCP_LRO_ACKCNT_MAX + 1))
1957 * Convert aggregation limit back to append
1962 for (i = 0; i < sc->hn_rx_ring_inuse; ++i)
1963 sc->hn_rx_ring[i].hn_lro.lro_ackcnt_lim = ackcnt;
1971 hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS)
1973 struct hn_softc *sc = arg1;
1978 if (sc->hn_rx_ring[0].hn_trust_hcsum & hcsum)
1981 error = sysctl_handle_int(oidp, &on, 0, req);
1982 if (error || req->newptr == NULL)
1986 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1987 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
1990 rxr->hn_trust_hcsum |= hcsum;
1992 rxr->hn_trust_hcsum &= ~hcsum;
1999 hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS)
2001 struct hn_softc *sc = arg1;
2002 int chimney_size, error;
2004 chimney_size = sc->hn_tx_ring[0].hn_tx_chimney_size;
2005 error = sysctl_handle_int(oidp, &chimney_size, 0, req);
2006 if (error || req->newptr == NULL)
2009 if (chimney_size > sc->hn_tx_chimney_max || chimney_size <= 0)
2012 hn_set_tx_chimney_size(sc, chimney_size);
2016 #if __FreeBSD_version < 1100095
2018 hn_rx_stat_int_sysctl(SYSCTL_HANDLER_ARGS)
2020 struct hn_softc *sc = arg1;
2021 int ofs = arg2, i, error;
2022 struct hn_rx_ring *rxr;
2026 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2027 rxr = &sc->hn_rx_ring[i];
2028 stat += *((int *)((uint8_t *)rxr + ofs));
2031 error = sysctl_handle_64(oidp, &stat, 0, req);
2032 if (error || req->newptr == NULL)
2035 /* Zero out this stat. */
2036 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2037 rxr = &sc->hn_rx_ring[i];
2038 *((int *)((uint8_t *)rxr + ofs)) = 0;
2044 hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS)
2046 struct hn_softc *sc = arg1;
2047 int ofs = arg2, i, error;
2048 struct hn_rx_ring *rxr;
2052 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2053 rxr = &sc->hn_rx_ring[i];
2054 stat += *((uint64_t *)((uint8_t *)rxr + ofs));
2057 error = sysctl_handle_64(oidp, &stat, 0, req);
2058 if (error || req->newptr == NULL)
2061 /* Zero out this stat. */
2062 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2063 rxr = &sc->hn_rx_ring[i];
2064 *((uint64_t *)((uint8_t *)rxr + ofs)) = 0;
2072 hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
2074 struct hn_softc *sc = arg1;
2075 int ofs = arg2, i, error;
2076 struct hn_rx_ring *rxr;
2080 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2081 rxr = &sc->hn_rx_ring[i];
2082 stat += *((u_long *)((uint8_t *)rxr + ofs));
2085 error = sysctl_handle_long(oidp, &stat, 0, req);
2086 if (error || req->newptr == NULL)
2089 /* Zero out this stat. */
2090 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2091 rxr = &sc->hn_rx_ring[i];
2092 *((u_long *)((uint8_t *)rxr + ofs)) = 0;
2098 hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
2100 struct hn_softc *sc = arg1;
2101 int ofs = arg2, i, error;
2102 struct hn_tx_ring *txr;
2106 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2107 txr = &sc->hn_tx_ring[i];
2108 stat += *((u_long *)((uint8_t *)txr + ofs));
2111 error = sysctl_handle_long(oidp, &stat, 0, req);
2112 if (error || req->newptr == NULL)
2115 /* Zero out this stat. */
2116 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2117 txr = &sc->hn_tx_ring[i];
2118 *((u_long *)((uint8_t *)txr + ofs)) = 0;
2124 hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS)
2126 struct hn_softc *sc = arg1;
2127 int ofs = arg2, i, error, conf;
2128 struct hn_tx_ring *txr;
2130 txr = &sc->hn_tx_ring[0];
2131 conf = *((int *)((uint8_t *)txr + ofs));
2133 error = sysctl_handle_int(oidp, &conf, 0, req);
2134 if (error || req->newptr == NULL)
2138 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2139 txr = &sc->hn_tx_ring[i];
2140 *((int *)((uint8_t *)txr + ofs)) = conf;
2148 hn_check_iplen(const struct mbuf *m, int hoff)
2150 const struct ip *ip;
2151 int len, iphlen, iplen;
2152 const struct tcphdr *th;
2153 int thoff; /* TCP data offset */
2155 len = hoff + sizeof(struct ip);
2157 /* The packet must be at least the size of an IP header. */
2158 if (m->m_pkthdr.len < len)
2159 return IPPROTO_DONE;
2161 /* The fixed IP header must reside completely in the first mbuf. */
2163 return IPPROTO_DONE;
2165 ip = mtodo(m, hoff);
2167 /* Bound check the packet's stated IP header length. */
2168 iphlen = ip->ip_hl << 2;
2169 if (iphlen < sizeof(struct ip)) /* minimum header length */
2170 return IPPROTO_DONE;
2172 /* The full IP header must reside completely in the one mbuf. */
2173 if (m->m_len < hoff + iphlen)
2174 return IPPROTO_DONE;
2176 iplen = ntohs(ip->ip_len);
2179 * Check that the amount of data in the buffers is as
2180 * at least much as the IP header would have us expect.
2182 if (m->m_pkthdr.len < hoff + iplen)
2183 return IPPROTO_DONE;
2186 * Ignore IP fragments.
2188 if (ntohs(ip->ip_off) & (IP_OFFMASK | IP_MF))
2189 return IPPROTO_DONE;
2192 * The TCP/IP or UDP/IP header must be entirely contained within
2193 * the first fragment of a packet.
2197 if (iplen < iphlen + sizeof(struct tcphdr))
2198 return IPPROTO_DONE;
2199 if (m->m_len < hoff + iphlen + sizeof(struct tcphdr))
2200 return IPPROTO_DONE;
2201 th = (const struct tcphdr *)((const uint8_t *)ip + iphlen);
2202 thoff = th->th_off << 2;
2203 if (thoff < sizeof(struct tcphdr) || thoff + iphlen > iplen)
2204 return IPPROTO_DONE;
2205 if (m->m_len < hoff + iphlen + thoff)
2206 return IPPROTO_DONE;
2209 if (iplen < iphlen + sizeof(struct udphdr))
2210 return IPPROTO_DONE;
2211 if (m->m_len < hoff + iphlen + sizeof(struct udphdr))
2212 return IPPROTO_DONE;
2216 return IPPROTO_DONE;
2223 hn_create_rx_data(struct hn_softc *sc, int ring_cnt)
2225 struct sysctl_oid_list *child;
2226 struct sysctl_ctx_list *ctx;
2227 device_t dev = sc->hn_dev;
2228 #if defined(INET) || defined(INET6)
2229 #if __FreeBSD_version >= 1100095
2235 sc->hn_rx_ring_cnt = ring_cnt;
2236 sc->hn_rx_ring_inuse = sc->hn_rx_ring_cnt;
2238 sc->hn_rx_ring = malloc(sizeof(struct hn_rx_ring) * sc->hn_rx_ring_cnt,
2239 M_NETVSC, M_WAITOK | M_ZERO);
2241 #if defined(INET) || defined(INET6)
2242 #if __FreeBSD_version >= 1100095
2243 lroent_cnt = hn_lro_entry_count;
2244 if (lroent_cnt < TCP_LRO_ENTRIES)
2245 lroent_cnt = TCP_LRO_ENTRIES;
2246 device_printf(dev, "LRO: entry count %d\n", lroent_cnt);
2248 #endif /* INET || INET6 */
2250 ctx = device_get_sysctl_ctx(dev);
2251 child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
2253 /* Create dev.hn.UNIT.rx sysctl tree */
2254 sc->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "rx",
2255 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2257 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2258 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
2260 if (hn_trust_hosttcp)
2261 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_TCP;
2262 if (hn_trust_hostudp)
2263 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_UDP;
2264 if (hn_trust_hostip)
2265 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_IP;
2266 rxr->hn_ifp = sc->hn_ifp;
2267 if (i < sc->hn_tx_ring_cnt)
2268 rxr->hn_txr = &sc->hn_tx_ring[i];
2269 rxr->hn_rdbuf = malloc(NETVSC_PACKET_SIZE, M_NETVSC, M_WAITOK);
2275 #if defined(INET) || defined(INET6)
2276 #if __FreeBSD_version >= 1100095
2277 tcp_lro_init_args(&rxr->hn_lro, sc->hn_ifp, lroent_cnt,
2278 hn_lro_mbufq_depth);
2280 tcp_lro_init(&rxr->hn_lro);
2281 rxr->hn_lro.ifp = sc->hn_ifp;
2283 #if __FreeBSD_version >= 1100099
2284 rxr->hn_lro.lro_length_lim = HN_LRO_LENLIM_DEF;
2285 rxr->hn_lro.lro_ackcnt_lim = HN_LRO_ACKCNT_DEF;
2287 #endif /* INET || INET6 */
2289 if (sc->hn_rx_sysctl_tree != NULL) {
2293 * Create per RX ring sysctl tree:
2294 * dev.hn.UNIT.rx.RINGID
2296 snprintf(name, sizeof(name), "%d", i);
2297 rxr->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx,
2298 SYSCTL_CHILDREN(sc->hn_rx_sysctl_tree),
2299 OID_AUTO, name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2301 if (rxr->hn_rx_sysctl_tree != NULL) {
2302 SYSCTL_ADD_ULONG(ctx,
2303 SYSCTL_CHILDREN(rxr->hn_rx_sysctl_tree),
2304 OID_AUTO, "packets", CTLFLAG_RW,
2305 &rxr->hn_pkts, "# of packets received");
2306 SYSCTL_ADD_ULONG(ctx,
2307 SYSCTL_CHILDREN(rxr->hn_rx_sysctl_tree),
2308 OID_AUTO, "rss_pkts", CTLFLAG_RW,
2310 "# of packets w/ RSS info received");
2315 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_queued",
2316 CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2317 __offsetof(struct hn_rx_ring, hn_lro.lro_queued),
2318 #if __FreeBSD_version < 1100095
2319 hn_rx_stat_int_sysctl,
2321 hn_rx_stat_u64_sysctl,
2323 "LU", "LRO queued");
2324 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_flushed",
2325 CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2326 __offsetof(struct hn_rx_ring, hn_lro.lro_flushed),
2327 #if __FreeBSD_version < 1100095
2328 hn_rx_stat_int_sysctl,
2330 hn_rx_stat_u64_sysctl,
2332 "LU", "LRO flushed");
2333 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_tried",
2334 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2335 __offsetof(struct hn_rx_ring, hn_lro_tried),
2336 hn_rx_stat_ulong_sysctl, "LU", "# of LRO tries");
2337 #if __FreeBSD_version >= 1100099
2338 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_length_lim",
2339 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2340 hn_lro_lenlim_sysctl, "IU",
2341 "Max # of data bytes to be aggregated by LRO");
2342 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_ackcnt_lim",
2343 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2344 hn_lro_ackcnt_sysctl, "I",
2345 "Max # of ACKs to be aggregated by LRO");
2347 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hosttcp",
2348 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_TCP,
2349 hn_trust_hcsum_sysctl, "I",
2350 "Trust tcp segement verification on host side, "
2351 "when csum info is missing");
2352 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostudp",
2353 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_UDP,
2354 hn_trust_hcsum_sysctl, "I",
2355 "Trust udp datagram verification on host side, "
2356 "when csum info is missing");
2357 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostip",
2358 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_IP,
2359 hn_trust_hcsum_sysctl, "I",
2360 "Trust ip packet verification on host side, "
2361 "when csum info is missing");
2362 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_ip",
2363 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2364 __offsetof(struct hn_rx_ring, hn_csum_ip),
2365 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM IP");
2366 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_tcp",
2367 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2368 __offsetof(struct hn_rx_ring, hn_csum_tcp),
2369 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM TCP");
2370 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_udp",
2371 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2372 __offsetof(struct hn_rx_ring, hn_csum_udp),
2373 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM UDP");
2374 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_trusted",
2375 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2376 __offsetof(struct hn_rx_ring, hn_csum_trusted),
2377 hn_rx_stat_ulong_sysctl, "LU",
2378 "# of packets that we trust host's csum verification");
2379 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "small_pkts",
2380 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2381 __offsetof(struct hn_rx_ring, hn_small_pkts),
2382 hn_rx_stat_ulong_sysctl, "LU", "# of small packets received");
2383 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_cnt",
2384 CTLFLAG_RD, &sc->hn_rx_ring_cnt, 0, "# created RX rings");
2385 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_inuse",
2386 CTLFLAG_RD, &sc->hn_rx_ring_inuse, 0, "# used RX rings");
2390 hn_destroy_rx_data(struct hn_softc *sc)
2394 if (sc->hn_rx_ring_cnt == 0)
2397 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2398 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
2400 #if defined(INET) || defined(INET6)
2401 tcp_lro_free(&rxr->hn_lro);
2403 free(rxr->hn_rdbuf, M_NETVSC);
2405 free(sc->hn_rx_ring, M_NETVSC);
2406 sc->hn_rx_ring = NULL;
2408 sc->hn_rx_ring_cnt = 0;
2409 sc->hn_rx_ring_inuse = 0;
2413 hn_create_tx_ring(struct hn_softc *sc, int id)
2415 struct hn_tx_ring *txr = &sc->hn_tx_ring[id];
2416 device_t dev = sc->hn_dev;
2417 bus_dma_tag_t parent_dtag;
2422 txr->hn_tx_idx = id;
2424 #ifndef HN_USE_TXDESC_BUFRING
2425 mtx_init(&txr->hn_txlist_spin, "hn txlist", NULL, MTX_SPIN);
2427 mtx_init(&txr->hn_tx_lock, "hn tx", NULL, MTX_DEF);
2429 txr->hn_txdesc_cnt = HN_TX_DESC_CNT;
2430 txr->hn_txdesc = malloc(sizeof(struct hn_txdesc) * txr->hn_txdesc_cnt,
2431 M_NETVSC, M_WAITOK | M_ZERO);
2432 #ifndef HN_USE_TXDESC_BUFRING
2433 SLIST_INIT(&txr->hn_txlist);
2435 txr->hn_txdesc_br = buf_ring_alloc(txr->hn_txdesc_cnt, M_NETVSC,
2436 M_WAITOK, &txr->hn_tx_lock);
2439 txr->hn_tx_taskq = sc->hn_tx_taskq;
2441 if (hn_use_if_start) {
2442 txr->hn_txeof = hn_start_txeof;
2443 TASK_INIT(&txr->hn_tx_task, 0, hn_start_taskfunc, txr);
2444 TASK_INIT(&txr->hn_txeof_task, 0, hn_start_txeof_taskfunc, txr);
2448 txr->hn_txeof = hn_xmit_txeof;
2449 TASK_INIT(&txr->hn_tx_task, 0, hn_xmit_taskfunc, txr);
2450 TASK_INIT(&txr->hn_txeof_task, 0, hn_xmit_txeof_taskfunc, txr);
2452 br_depth = hn_get_txswq_depth(txr);
2453 txr->hn_mbuf_br = buf_ring_alloc(br_depth, M_NETVSC,
2454 M_WAITOK, &txr->hn_tx_lock);
2457 txr->hn_direct_tx_size = hn_direct_tx_size;
2458 version = VMBUS_GET_VERSION(device_get_parent(dev), dev);
2459 if (version >= VMBUS_VERSION_WIN8_1) {
2460 txr->hn_csum_assist = HN_CSUM_ASSIST;
2462 txr->hn_csum_assist = HN_CSUM_ASSIST_WIN8;
2464 device_printf(dev, "bus version %u.%u, "
2465 "no UDP checksum offloading\n",
2466 VMBUS_VERSION_MAJOR(version),
2467 VMBUS_VERSION_MINOR(version));
2472 * Always schedule transmission instead of trying to do direct
2473 * transmission. This one gives the best performance so far.
2475 txr->hn_sched_tx = 1;
2477 parent_dtag = bus_get_dma_tag(dev);
2479 /* DMA tag for RNDIS messages. */
2480 error = bus_dma_tag_create(parent_dtag, /* parent */
2481 HN_RNDIS_MSG_ALIGN, /* alignment */
2482 HN_RNDIS_MSG_BOUNDARY, /* boundary */
2483 BUS_SPACE_MAXADDR, /* lowaddr */
2484 BUS_SPACE_MAXADDR, /* highaddr */
2485 NULL, NULL, /* filter, filterarg */
2486 HN_RNDIS_MSG_LEN, /* maxsize */
2488 HN_RNDIS_MSG_LEN, /* maxsegsize */
2490 NULL, /* lockfunc */
2491 NULL, /* lockfuncarg */
2492 &txr->hn_tx_rndis_dtag);
2494 device_printf(dev, "failed to create rndis dmatag\n");
2498 /* DMA tag for data. */
2499 error = bus_dma_tag_create(parent_dtag, /* parent */
2501 HN_TX_DATA_BOUNDARY, /* boundary */
2502 BUS_SPACE_MAXADDR, /* lowaddr */
2503 BUS_SPACE_MAXADDR, /* highaddr */
2504 NULL, NULL, /* filter, filterarg */
2505 HN_TX_DATA_MAXSIZE, /* maxsize */
2506 HN_TX_DATA_SEGCNT_MAX, /* nsegments */
2507 HN_TX_DATA_SEGSIZE, /* maxsegsize */
2509 NULL, /* lockfunc */
2510 NULL, /* lockfuncarg */
2511 &txr->hn_tx_data_dtag);
2513 device_printf(dev, "failed to create data dmatag\n");
2517 for (i = 0; i < txr->hn_txdesc_cnt; ++i) {
2518 struct hn_txdesc *txd = &txr->hn_txdesc[i];
2523 * Allocate and load RNDIS messages.
2525 error = bus_dmamem_alloc(txr->hn_tx_rndis_dtag,
2526 (void **)&txd->rndis_msg,
2527 BUS_DMA_WAITOK | BUS_DMA_COHERENT,
2528 &txd->rndis_msg_dmap);
2531 "failed to allocate rndis_msg, %d\n", i);
2535 error = bus_dmamap_load(txr->hn_tx_rndis_dtag,
2536 txd->rndis_msg_dmap,
2537 txd->rndis_msg, HN_RNDIS_MSG_LEN,
2538 hyperv_dma_map_paddr, &txd->rndis_msg_paddr,
2542 "failed to load rndis_msg, %d\n", i);
2543 bus_dmamem_free(txr->hn_tx_rndis_dtag,
2544 txd->rndis_msg, txd->rndis_msg_dmap);
2548 /* DMA map for TX data. */
2549 error = bus_dmamap_create(txr->hn_tx_data_dtag, 0,
2553 "failed to allocate tx data dmamap\n");
2554 bus_dmamap_unload(txr->hn_tx_rndis_dtag,
2555 txd->rndis_msg_dmap);
2556 bus_dmamem_free(txr->hn_tx_rndis_dtag,
2557 txd->rndis_msg, txd->rndis_msg_dmap);
2561 /* All set, put it to list */
2562 txd->flags |= HN_TXD_FLAG_ONLIST;
2563 #ifndef HN_USE_TXDESC_BUFRING
2564 SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
2566 buf_ring_enqueue(txr->hn_txdesc_br, txd);
2569 txr->hn_txdesc_avail = txr->hn_txdesc_cnt;
2571 if (sc->hn_tx_sysctl_tree != NULL) {
2572 struct sysctl_oid_list *child;
2573 struct sysctl_ctx_list *ctx;
2577 * Create per TX ring sysctl tree:
2578 * dev.hn.UNIT.tx.RINGID
2580 ctx = device_get_sysctl_ctx(dev);
2581 child = SYSCTL_CHILDREN(sc->hn_tx_sysctl_tree);
2583 snprintf(name, sizeof(name), "%d", id);
2584 txr->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO,
2585 name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2587 if (txr->hn_tx_sysctl_tree != NULL) {
2588 child = SYSCTL_CHILDREN(txr->hn_tx_sysctl_tree);
2590 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_avail",
2591 CTLFLAG_RD, &txr->hn_txdesc_avail, 0,
2592 "# of available TX descs");
2593 if (!hn_use_if_start) {
2594 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "oactive",
2595 CTLFLAG_RD, &txr->hn_oactive, 0,
2598 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "packets",
2599 CTLFLAG_RW, &txr->hn_pkts,
2600 "# of packets transmitted");
2608 hn_txdesc_dmamap_destroy(struct hn_txdesc *txd)
2610 struct hn_tx_ring *txr = txd->txr;
2612 KASSERT(txd->m == NULL, ("still has mbuf installed"));
2613 KASSERT((txd->flags & HN_TXD_FLAG_DMAMAP) == 0, ("still dma mapped"));
2615 bus_dmamap_unload(txr->hn_tx_rndis_dtag, txd->rndis_msg_dmap);
2616 bus_dmamem_free(txr->hn_tx_rndis_dtag, txd->rndis_msg,
2617 txd->rndis_msg_dmap);
2618 bus_dmamap_destroy(txr->hn_tx_data_dtag, txd->data_dmap);
2622 hn_destroy_tx_ring(struct hn_tx_ring *txr)
2624 struct hn_txdesc *txd;
2626 if (txr->hn_txdesc == NULL)
2629 #ifndef HN_USE_TXDESC_BUFRING
2630 while ((txd = SLIST_FIRST(&txr->hn_txlist)) != NULL) {
2631 SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
2632 hn_txdesc_dmamap_destroy(txd);
2635 mtx_lock(&txr->hn_tx_lock);
2636 while ((txd = buf_ring_dequeue_sc(txr->hn_txdesc_br)) != NULL)
2637 hn_txdesc_dmamap_destroy(txd);
2638 mtx_unlock(&txr->hn_tx_lock);
2641 if (txr->hn_tx_data_dtag != NULL)
2642 bus_dma_tag_destroy(txr->hn_tx_data_dtag);
2643 if (txr->hn_tx_rndis_dtag != NULL)
2644 bus_dma_tag_destroy(txr->hn_tx_rndis_dtag);
2646 #ifdef HN_USE_TXDESC_BUFRING
2647 buf_ring_free(txr->hn_txdesc_br, M_NETVSC);
2650 free(txr->hn_txdesc, M_NETVSC);
2651 txr->hn_txdesc = NULL;
2653 if (txr->hn_mbuf_br != NULL)
2654 buf_ring_free(txr->hn_mbuf_br, M_NETVSC);
2656 #ifndef HN_USE_TXDESC_BUFRING
2657 mtx_destroy(&txr->hn_txlist_spin);
2659 mtx_destroy(&txr->hn_tx_lock);
2663 hn_create_tx_data(struct hn_softc *sc, int ring_cnt)
2665 struct sysctl_oid_list *child;
2666 struct sysctl_ctx_list *ctx;
2669 sc->hn_tx_ring_cnt = ring_cnt;
2670 sc->hn_tx_ring_inuse = sc->hn_tx_ring_cnt;
2672 sc->hn_tx_ring = malloc(sizeof(struct hn_tx_ring) * sc->hn_tx_ring_cnt,
2673 M_NETVSC, M_WAITOK | M_ZERO);
2675 ctx = device_get_sysctl_ctx(sc->hn_dev);
2676 child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->hn_dev));
2678 /* Create dev.hn.UNIT.tx sysctl tree */
2679 sc->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "tx",
2680 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2682 for (i = 0; i < sc->hn_tx_ring_cnt; ++i) {
2685 error = hn_create_tx_ring(sc, i);
2690 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "no_txdescs",
2691 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2692 __offsetof(struct hn_tx_ring, hn_no_txdescs),
2693 hn_tx_stat_ulong_sysctl, "LU", "# of times short of TX descs");
2694 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "send_failed",
2695 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2696 __offsetof(struct hn_tx_ring, hn_send_failed),
2697 hn_tx_stat_ulong_sysctl, "LU", "# of hyper-v sending failure");
2698 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "txdma_failed",
2699 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2700 __offsetof(struct hn_tx_ring, hn_txdma_failed),
2701 hn_tx_stat_ulong_sysctl, "LU", "# of TX DMA failure");
2702 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_collapsed",
2703 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2704 __offsetof(struct hn_tx_ring, hn_tx_collapsed),
2705 hn_tx_stat_ulong_sysctl, "LU", "# of TX mbuf collapsed");
2706 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney",
2707 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2708 __offsetof(struct hn_tx_ring, hn_tx_chimney),
2709 hn_tx_stat_ulong_sysctl, "LU", "# of chimney send");
2710 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_tried",
2711 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2712 __offsetof(struct hn_tx_ring, hn_tx_chimney_tried),
2713 hn_tx_stat_ulong_sysctl, "LU", "# of chimney send tries");
2714 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_cnt",
2715 CTLFLAG_RD, &sc->hn_tx_ring[0].hn_txdesc_cnt, 0,
2716 "# of total TX descs");
2717 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_chimney_max",
2718 CTLFLAG_RD, &sc->hn_tx_chimney_max, 0,
2719 "Chimney send packet size upper boundary");
2720 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_size",
2721 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2722 hn_tx_chimney_size_sysctl,
2723 "I", "Chimney send packet size limit");
2724 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "direct_tx_size",
2725 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2726 __offsetof(struct hn_tx_ring, hn_direct_tx_size),
2727 hn_tx_conf_int_sysctl, "I",
2728 "Size of the packet for direct transmission");
2729 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "sched_tx",
2730 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2731 __offsetof(struct hn_tx_ring, hn_sched_tx),
2732 hn_tx_conf_int_sysctl, "I",
2733 "Always schedule transmission "
2734 "instead of doing direct transmission");
2735 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_cnt",
2736 CTLFLAG_RD, &sc->hn_tx_ring_cnt, 0, "# created TX rings");
2737 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_inuse",
2738 CTLFLAG_RD, &sc->hn_tx_ring_inuse, 0, "# used TX rings");
2744 hn_set_tx_chimney_size(struct hn_softc *sc, int chimney_size)
2749 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
2750 sc->hn_tx_ring[i].hn_tx_chimney_size = chimney_size;
2755 hn_destroy_tx_data(struct hn_softc *sc)
2759 if (sc->hn_tx_ring_cnt == 0)
2762 for (i = 0; i < sc->hn_tx_ring_cnt; ++i)
2763 hn_destroy_tx_ring(&sc->hn_tx_ring[i]);
2765 free(sc->hn_tx_ring, M_NETVSC);
2766 sc->hn_tx_ring = NULL;
2768 sc->hn_tx_ring_cnt = 0;
2769 sc->hn_tx_ring_inuse = 0;
2773 hn_start_taskfunc(void *xtxr, int pending __unused)
2775 struct hn_tx_ring *txr = xtxr;
2777 mtx_lock(&txr->hn_tx_lock);
2778 hn_start_locked(txr, 0);
2779 mtx_unlock(&txr->hn_tx_lock);
2783 hn_start_txeof_taskfunc(void *xtxr, int pending __unused)
2785 struct hn_tx_ring *txr = xtxr;
2787 mtx_lock(&txr->hn_tx_lock);
2788 atomic_clear_int(&txr->hn_sc->hn_ifp->if_drv_flags, IFF_DRV_OACTIVE);
2789 hn_start_locked(txr, 0);
2790 mtx_unlock(&txr->hn_tx_lock);
2794 hn_stop_tx_tasks(struct hn_softc *sc)
2798 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2799 struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
2801 taskqueue_drain(txr->hn_tx_taskq, &txr->hn_tx_task);
2802 taskqueue_drain(txr->hn_tx_taskq, &txr->hn_txeof_task);
2807 hn_xmit(struct hn_tx_ring *txr, int len)
2809 struct hn_softc *sc = txr->hn_sc;
2810 struct ifnet *ifp = sc->hn_ifp;
2811 struct mbuf *m_head;
2813 mtx_assert(&txr->hn_tx_lock, MA_OWNED);
2814 KASSERT(hn_use_if_start == 0,
2815 ("hn_xmit is called, when if_start is enabled"));
2817 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || txr->hn_oactive)
2820 while ((m_head = drbr_peek(ifp, txr->hn_mbuf_br)) != NULL) {
2821 struct hn_txdesc *txd;
2824 if (len > 0 && m_head->m_pkthdr.len > len) {
2826 * This sending could be time consuming; let callers
2827 * dispatch this packet sending (and sending of any
2828 * following up packets) to tx taskqueue.
2830 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2834 txd = hn_txdesc_get(txr);
2836 txr->hn_no_txdescs++;
2837 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2838 txr->hn_oactive = 1;
2842 error = hn_encap(txr, txd, &m_head);
2844 /* Both txd and m_head are freed; discard */
2845 drbr_advance(ifp, txr->hn_mbuf_br);
2849 error = hn_send_pkt(ifp, txr, txd);
2850 if (__predict_false(error)) {
2851 /* txd is freed, but m_head is not */
2852 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2853 txr->hn_oactive = 1;
2858 drbr_advance(ifp, txr->hn_mbuf_br);
2864 hn_transmit(struct ifnet *ifp, struct mbuf *m)
2866 struct hn_softc *sc = ifp->if_softc;
2867 struct hn_tx_ring *txr;
2871 * Select the TX ring based on flowid
2873 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2874 idx = m->m_pkthdr.flowid % sc->hn_tx_ring_inuse;
2875 txr = &sc->hn_tx_ring[idx];
2877 error = drbr_enqueue(ifp, txr->hn_mbuf_br, m);
2879 if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1);
2883 if (txr->hn_oactive)
2886 if (txr->hn_sched_tx)
2889 if (mtx_trylock(&txr->hn_tx_lock)) {
2892 sched = hn_xmit(txr, txr->hn_direct_tx_size);
2893 mtx_unlock(&txr->hn_tx_lock);
2898 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
2903 hn_xmit_qflush(struct ifnet *ifp)
2905 struct hn_softc *sc = ifp->if_softc;
2908 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2909 struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
2912 mtx_lock(&txr->hn_tx_lock);
2913 while ((m = buf_ring_dequeue_sc(txr->hn_mbuf_br)) != NULL)
2915 mtx_unlock(&txr->hn_tx_lock);
2921 hn_xmit_txeof(struct hn_tx_ring *txr)
2924 if (txr->hn_sched_tx)
2927 if (mtx_trylock(&txr->hn_tx_lock)) {
2930 txr->hn_oactive = 0;
2931 sched = hn_xmit(txr, txr->hn_direct_tx_size);
2932 mtx_unlock(&txr->hn_tx_lock);
2934 taskqueue_enqueue(txr->hn_tx_taskq,
2940 * Release the oactive earlier, with the hope, that
2941 * others could catch up. The task will clear the
2942 * oactive again with the hn_tx_lock to avoid possible
2945 txr->hn_oactive = 0;
2946 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
2951 hn_xmit_taskfunc(void *xtxr, int pending __unused)
2953 struct hn_tx_ring *txr = xtxr;
2955 mtx_lock(&txr->hn_tx_lock);
2957 mtx_unlock(&txr->hn_tx_lock);
2961 hn_xmit_txeof_taskfunc(void *xtxr, int pending __unused)
2963 struct hn_tx_ring *txr = xtxr;
2965 mtx_lock(&txr->hn_tx_lock);
2966 txr->hn_oactive = 0;
2968 mtx_unlock(&txr->hn_tx_lock);
2972 hn_channel_attach(struct hn_softc *sc, struct vmbus_channel *chan)
2974 struct hn_rx_ring *rxr;
2977 idx = vmbus_chan_subidx(chan);
2979 KASSERT(idx >= 0 && idx < sc->hn_rx_ring_inuse,
2980 ("invalid channel index %d, should > 0 && < %d",
2981 idx, sc->hn_rx_ring_inuse));
2982 rxr = &sc->hn_rx_ring[idx];
2983 KASSERT((rxr->hn_rx_flags & HN_RX_FLAG_ATTACHED) == 0,
2984 ("RX ring %d already attached", idx));
2985 rxr->hn_rx_flags |= HN_RX_FLAG_ATTACHED;
2988 if_printf(sc->hn_ifp, "link RX ring %d to channel%u\n",
2989 idx, vmbus_chan_id(chan));
2992 if (idx < sc->hn_tx_ring_inuse) {
2993 struct hn_tx_ring *txr = &sc->hn_tx_ring[idx];
2995 KASSERT((txr->hn_tx_flags & HN_TX_FLAG_ATTACHED) == 0,
2996 ("TX ring %d already attached", idx));
2997 txr->hn_tx_flags |= HN_TX_FLAG_ATTACHED;
2999 txr->hn_chan = chan;
3001 if_printf(sc->hn_ifp, "link TX ring %d to channel%u\n",
3002 idx, vmbus_chan_id(chan));
3006 /* Bind channel to a proper CPU */
3007 vmbus_chan_cpu_set(chan, (sc->hn_cpu + idx) % mp_ncpus);
3011 hn_subchan_attach(struct hn_softc *sc, struct vmbus_channel *chan)
3014 KASSERT(!vmbus_chan_is_primary(chan),
3015 ("subchannel callback on primary channel"));
3016 hn_channel_attach(sc, chan);
3020 hn_subchan_setup(struct hn_softc *sc)
3022 struct vmbus_channel **subchans;
3023 int subchan_cnt = sc->net_dev->num_channel - 1;
3026 /* Wait for sub-channels setup to complete. */
3027 subchans = vmbus_subchan_get(sc->hn_prichan, subchan_cnt);
3029 /* Attach the sub-channels. */
3030 for (i = 0; i < subchan_cnt; ++i) {
3031 struct vmbus_channel *subchan = subchans[i];
3033 /* NOTE: Calling order is critical. */
3034 hn_subchan_attach(sc, subchan);
3035 hv_nv_subchan_attach(subchan,
3036 &sc->hn_rx_ring[vmbus_chan_subidx(subchan)]);
3039 /* Release the sub-channels */
3040 vmbus_subchan_rel(subchans, subchan_cnt);
3041 if_printf(sc->hn_ifp, "%d sub-channels setup done\n", subchan_cnt);
3045 hn_tx_taskq_create(void *arg __unused)
3047 if (!hn_share_tx_taskq)
3050 hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
3051 taskqueue_thread_enqueue, &hn_tx_taskq);
3052 taskqueue_start_threads(&hn_tx_taskq, 1, PI_NET, "hn tx");
3053 if (hn_bind_tx_taskq >= 0) {
3054 int cpu = hn_bind_tx_taskq;
3055 struct task cpuset_task;
3058 if (cpu > mp_ncpus - 1)
3060 CPU_SETOF(cpu, &cpu_set);
3061 TASK_INIT(&cpuset_task, 0, hn_cpuset_setthread_task, &cpu_set);
3062 taskqueue_enqueue(hn_tx_taskq, &cpuset_task);
3063 taskqueue_drain(hn_tx_taskq, &cpuset_task);
3066 SYSINIT(hn_txtq_create, SI_SUB_DRIVERS, SI_ORDER_FIRST,
3067 hn_tx_taskq_create, NULL);
3070 hn_tx_taskq_destroy(void *arg __unused)
3072 if (hn_tx_taskq != NULL)
3073 taskqueue_free(hn_tx_taskq);
3075 SYSUNINIT(hn_txtq_destroy, SI_SUB_DRIVERS, SI_ORDER_FIRST,
3076 hn_tx_taskq_destroy, NULL);
3078 static device_method_t netvsc_methods[] = {
3079 /* Device interface */
3080 DEVMETHOD(device_probe, netvsc_probe),
3081 DEVMETHOD(device_attach, netvsc_attach),
3082 DEVMETHOD(device_detach, netvsc_detach),
3083 DEVMETHOD(device_shutdown, netvsc_shutdown),
3088 static driver_t netvsc_driver = {
3094 static devclass_t netvsc_devclass;
3096 DRIVER_MODULE(hn, vmbus, netvsc_driver, netvsc_devclass, 0, 0);
3097 MODULE_VERSION(hn, 1);
3098 MODULE_DEPEND(hn, vmbus, 1, 1, 1);