2 * Copyright (c) 2010-2012 Citrix Inc.
3 * Copyright (c) 2009-2012 Microsoft Corp.
4 * Copyright (c) 2012 NetApp Inc.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice unmodified, this list of conditions, and the following
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30 * Copyright (c) 2004-2006 Kip Macy
31 * All rights reserved.
33 * Redistribution and use in source and binary forms, with or without
34 * modification, are permitted provided that the following conditions
36 * 1. Redistributions of source code must retain the above copyright
37 * notice, this list of conditions and the following disclaimer.
38 * 2. Redistributions in binary form must reproduce the above copyright
39 * notice, this list of conditions and the following disclaimer in the
40 * documentation and/or other materials provided with the distribution.
42 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
43 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
44 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
45 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
46 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
47 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
48 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
49 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
50 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
51 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
55 #include <sys/cdefs.h>
56 __FBSDID("$FreeBSD$");
58 #include "opt_inet6.h"
61 #include <sys/param.h>
62 #include <sys/systm.h>
63 #include <sys/sockio.h>
65 #include <sys/malloc.h>
66 #include <sys/module.h>
67 #include <sys/kernel.h>
68 #include <sys/socket.h>
70 #include <sys/queue.h>
73 #include <sys/sysctl.h>
74 #include <sys/buf_ring.h>
77 #include <net/if_arp.h>
78 #include <net/ethernet.h>
79 #include <net/if_dl.h>
80 #include <net/if_media.h>
84 #include <net/if_types.h>
85 #include <net/if_vlan_var.h>
88 #include <netinet/in_systm.h>
89 #include <netinet/in.h>
90 #include <netinet/ip.h>
91 #include <netinet/if_ether.h>
92 #include <netinet/tcp.h>
93 #include <netinet/udp.h>
94 #include <netinet/ip6.h>
97 #include <vm/vm_param.h>
98 #include <vm/vm_kern.h>
101 #include <machine/bus.h>
102 #include <machine/resource.h>
103 #include <machine/frame.h>
104 #include <machine/vmparam.h>
107 #include <sys/rman.h>
108 #include <sys/mutex.h>
109 #include <sys/errno.h>
110 #include <sys/types.h>
111 #include <machine/atomic.h>
113 #include <machine/intr_machdep.h>
115 #include <machine/in_cksum.h>
117 #include <dev/hyperv/include/hyperv.h>
118 #include "hv_net_vsc.h"
119 #include "hv_rndis.h"
120 #include "hv_rndis_filter.h"
123 /* Short for Hyper-V network interface */
124 #define NETVSC_DEVNAME "hn"
127 * It looks like offset 0 of buf is reserved to hold the softc pointer.
128 * The sc pointer evidently not needed, and is not presently populated.
129 * The packet offset is where the netvsc_packet starts in the buffer.
131 #define HV_NV_SC_PTR_OFFSET_IN_BUF 0
132 #define HV_NV_PACKET_OFFSET_IN_BUF 16
134 /* YYY should get it from the underlying channel */
135 #define HN_TX_DESC_CNT 512
137 #define HN_LROENT_CNT_DEF 128
139 #define HN_RNDIS_MSG_LEN \
140 (sizeof(rndis_msg) + \
141 RNDIS_VLAN_PPI_SIZE + \
142 RNDIS_TSO_PPI_SIZE + \
144 #define HN_RNDIS_MSG_BOUNDARY PAGE_SIZE
145 #define HN_RNDIS_MSG_ALIGN CACHE_LINE_SIZE
147 #define HN_TX_DATA_BOUNDARY PAGE_SIZE
148 #define HN_TX_DATA_MAXSIZE IP_MAXPACKET
149 #define HN_TX_DATA_SEGSIZE PAGE_SIZE
150 #define HN_TX_DATA_SEGCNT_MAX \
151 (NETVSC_PACKET_MAXPAGE - HV_RF_NUM_TX_RESERVED_PAGE_BUFS)
153 #define HN_DIRECT_TX_SIZE_DEF 128
156 #ifndef HN_USE_TXDESC_BUFRING
157 SLIST_ENTRY(hn_txdesc) link;
160 struct hn_tx_ring *txr;
162 uint32_t flags; /* HN_TXD_FLAG_ */
163 netvsc_packet netvsc_pkt; /* XXX to be removed */
165 bus_dmamap_t data_dmap;
167 bus_addr_t rndis_msg_paddr;
168 rndis_msg *rndis_msg;
169 bus_dmamap_t rndis_msg_dmap;
172 #define HN_TXD_FLAG_ONLIST 0x1
173 #define HN_TXD_FLAG_DMAMAP 0x2
176 * Only enable UDP checksum offloading when it is on 2012R2 or
177 * later. UDP checksum offloading doesn't work on earlier
180 #define HN_CSUM_ASSIST_WIN8 (CSUM_IP | CSUM_TCP)
181 #define HN_CSUM_ASSIST (CSUM_IP | CSUM_UDP | CSUM_TCP)
183 #define HN_LRO_LENLIM_DEF (25 * ETHERMTU)
184 /* YYY 2*MTU is a bit rough, but should be good enough. */
185 #define HN_LRO_LENLIM_MIN(ifp) (2 * (ifp)->if_mtu)
187 #define HN_LRO_ACKCNT_DEF 1
190 * Be aware that this sleepable mutex will exhibit WITNESS errors when
191 * certain TCP and ARP code paths are taken. This appears to be a
192 * well-known condition, as all other drivers checked use a sleeping
193 * mutex to protect their transmit paths.
194 * Also Be aware that mutexes do not play well with semaphores, and there
195 * is a conflicting semaphore in a certain channel code path.
197 #define NV_LOCK_INIT(_sc, _name) \
198 mtx_init(&(_sc)->hn_lock, _name, MTX_NETWORK_LOCK, MTX_DEF)
199 #define NV_LOCK(_sc) mtx_lock(&(_sc)->hn_lock)
200 #define NV_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->hn_lock, MA_OWNED)
201 #define NV_UNLOCK(_sc) mtx_unlock(&(_sc)->hn_lock)
202 #define NV_LOCK_DESTROY(_sc) mtx_destroy(&(_sc)->hn_lock)
209 int hv_promisc_mode = 0; /* normal mode by default */
211 SYSCTL_NODE(_hw, OID_AUTO, hn, CTLFLAG_RD, NULL, "Hyper-V network interface");
213 /* Trust tcp segements verification on host side. */
214 static int hn_trust_hosttcp = 1;
215 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hosttcp, CTLFLAG_RDTUN,
216 &hn_trust_hosttcp, 0,
217 "Trust tcp segement verification on host side, "
218 "when csum info is missing (global setting)");
220 /* Trust udp datagrams verification on host side. */
221 static int hn_trust_hostudp = 1;
222 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostudp, CTLFLAG_RDTUN,
223 &hn_trust_hostudp, 0,
224 "Trust udp datagram verification on host side, "
225 "when csum info is missing (global setting)");
227 /* Trust ip packets verification on host side. */
228 static int hn_trust_hostip = 1;
229 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostip, CTLFLAG_RDTUN,
231 "Trust ip packet verification on host side, "
232 "when csum info is missing (global setting)");
234 #if __FreeBSD_version >= 1100045
235 /* Limit TSO burst size */
236 static int hn_tso_maxlen = 0;
237 SYSCTL_INT(_hw_hn, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN,
238 &hn_tso_maxlen, 0, "TSO burst limit");
241 /* Limit chimney send size */
242 static int hn_tx_chimney_size = 0;
243 SYSCTL_INT(_hw_hn, OID_AUTO, tx_chimney_size, CTLFLAG_RDTUN,
244 &hn_tx_chimney_size, 0, "Chimney send packet size limit");
246 /* Limit the size of packet for direct transmission */
247 static int hn_direct_tx_size = HN_DIRECT_TX_SIZE_DEF;
248 SYSCTL_INT(_hw_hn, OID_AUTO, direct_tx_size, CTLFLAG_RDTUN,
249 &hn_direct_tx_size, 0, "Size of the packet for direct transmission");
251 #if defined(INET) || defined(INET6)
252 #if __FreeBSD_version >= 1100095
253 static int hn_lro_entry_count = HN_LROENT_CNT_DEF;
254 SYSCTL_INT(_hw_hn, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN,
255 &hn_lro_entry_count, 0, "LRO entry count");
259 static int hn_share_tx_taskq = 0;
260 SYSCTL_INT(_hw_hn, OID_AUTO, share_tx_taskq, CTLFLAG_RDTUN,
261 &hn_share_tx_taskq, 0, "Enable shared TX taskqueue");
263 static struct taskqueue *hn_tx_taskq;
265 #ifndef HN_USE_TXDESC_BUFRING
266 static int hn_use_txdesc_bufring = 0;
268 static int hn_use_txdesc_bufring = 1;
270 SYSCTL_INT(_hw_hn, OID_AUTO, use_txdesc_bufring, CTLFLAG_RD,
271 &hn_use_txdesc_bufring, 0, "Use buf_ring for TX descriptors");
273 static int hn_bind_tx_taskq = -1;
274 SYSCTL_INT(_hw_hn, OID_AUTO, bind_tx_taskq, CTLFLAG_RDTUN,
275 &hn_bind_tx_taskq, 0, "Bind TX taskqueue to the specified cpu");
278 * Forward declarations
280 static void hn_stop(hn_softc_t *sc);
281 static void hn_ifinit_locked(hn_softc_t *sc);
282 static void hn_ifinit(void *xsc);
283 static int hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data);
284 static int hn_start_locked(struct hn_tx_ring *txr, int len);
285 static void hn_start(struct ifnet *ifp);
286 static void hn_start_txeof(struct hn_tx_ring *);
287 static int hn_ifmedia_upd(struct ifnet *ifp);
288 static void hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr);
289 #if __FreeBSD_version >= 1100099
290 static int hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS);
291 static int hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS);
293 static int hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS);
294 static int hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS);
295 #if __FreeBSD_version < 1100095
296 static int hn_rx_stat_int_sysctl(SYSCTL_HANDLER_ARGS);
298 static int hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS);
300 static int hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS);
301 static int hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS);
302 static int hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS);
303 static int hn_check_iplen(const struct mbuf *, int);
304 static int hn_create_tx_ring(struct hn_softc *, int);
305 static void hn_destroy_tx_ring(struct hn_tx_ring *);
306 static int hn_create_tx_data(struct hn_softc *);
307 static void hn_destroy_tx_data(struct hn_softc *);
308 static void hn_start_taskfunc(void *, int);
309 static void hn_start_txeof_taskfunc(void *, int);
310 static void hn_stop_tx_tasks(struct hn_softc *);
311 static int hn_encap(struct hn_tx_ring *, struct hn_txdesc *, struct mbuf **);
312 static void hn_create_rx_data(struct hn_softc *sc);
313 static void hn_destroy_rx_data(struct hn_softc *sc);
314 static void hn_set_tx_chimney_size(struct hn_softc *, int);
317 hn_ifmedia_upd(struct ifnet *ifp __unused)
324 hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
326 struct hn_softc *sc = ifp->if_softc;
328 ifmr->ifm_status = IFM_AVALID;
329 ifmr->ifm_active = IFM_ETHER;
331 if (!sc->hn_carrier) {
332 ifmr->ifm_active |= IFM_NONE;
335 ifmr->ifm_status |= IFM_ACTIVE;
336 ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
339 /* {F8615163-DF3E-46c5-913F-F2D2F965ED0E} */
340 static const hv_guid g_net_vsc_device_type = {
341 .data = {0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
342 0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E}
346 * Standard probe entry point.
350 netvsc_probe(device_t dev)
354 p = vmbus_get_type(dev);
355 if (!memcmp(p, &g_net_vsc_device_type.data, sizeof(hv_guid))) {
356 device_set_desc(dev, "Synthetic Network Interface");
358 printf("Netvsc probe... DONE \n");
360 return (BUS_PROBE_DEFAULT);
367 hn_cpuset_setthread_task(void *xmask, int pending __unused)
369 cpuset_t *mask = xmask;
372 error = cpuset_setthread(curthread->td_tid, mask);
374 panic("curthread=%ju: can't pin; error=%d",
375 (uintmax_t)curthread->td_tid, error);
380 * Standard attach entry point.
382 * Called when the driver is loaded. It allocates needed resources,
383 * and initializes the "hardware" and software.
386 netvsc_attach(device_t dev)
388 struct hv_device *device_ctx = vmbus_get_devctx(dev);
389 netvsc_device_info device_info;
391 int unit = device_get_unit(dev);
392 struct ifnet *ifp = NULL;
394 #if __FreeBSD_version >= 1100045
398 sc = device_get_softc(dev);
403 bzero(sc, sizeof(hn_softc_t));
407 if (hn_tx_taskq == NULL) {
408 sc->hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
409 taskqueue_thread_enqueue, &sc->hn_tx_taskq);
410 taskqueue_start_threads(&sc->hn_tx_taskq, 1, PI_NET, "%s tx",
411 device_get_nameunit(dev));
412 if (hn_bind_tx_taskq >= 0) {
413 int cpu = hn_bind_tx_taskq;
414 struct task cpuset_task;
417 if (cpu > mp_ncpus - 1)
419 CPU_SETOF(cpu, &cpu_set);
420 TASK_INIT(&cpuset_task, 0, hn_cpuset_setthread_task,
422 taskqueue_enqueue(sc->hn_tx_taskq, &cpuset_task);
423 taskqueue_drain(sc->hn_tx_taskq, &cpuset_task);
426 sc->hn_tx_taskq = hn_tx_taskq;
428 NV_LOCK_INIT(sc, "NetVSCLock");
430 sc->hn_dev_obj = device_ctx;
432 ifp = sc->hn_ifp = sc->arpcom.ac_ifp = if_alloc(IFT_ETHER);
435 error = hn_create_tx_data(sc);
439 hn_create_rx_data(sc);
441 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
442 ifp->if_dunit = unit;
443 ifp->if_dname = NETVSC_DEVNAME;
445 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
446 ifp->if_ioctl = hn_ioctl;
447 ifp->if_start = hn_start;
448 ifp->if_init = hn_ifinit;
449 /* needed by hv_rf_on_device_add() code */
450 ifp->if_mtu = ETHERMTU;
451 IFQ_SET_MAXLEN(&ifp->if_snd, 512);
452 ifp->if_snd.ifq_drv_maxlen = 511;
453 IFQ_SET_READY(&ifp->if_snd);
455 ifmedia_init(&sc->hn_media, 0, hn_ifmedia_upd, hn_ifmedia_sts);
456 ifmedia_add(&sc->hn_media, IFM_ETHER | IFM_AUTO, 0, NULL);
457 ifmedia_set(&sc->hn_media, IFM_ETHER | IFM_AUTO);
458 /* XXX ifmedia_set really should do this for us */
459 sc->hn_media.ifm_media = sc->hn_media.ifm_cur->ifm_media;
462 * Tell upper layers that we support full VLAN capability.
464 ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
465 ifp->if_capabilities |=
466 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
469 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
471 ifp->if_hwassist = sc->hn_tx_ring[0].hn_csum_assist | CSUM_TSO;
473 error = hv_rf_on_device_add(device_ctx, &device_info);
477 if (device_info.link_state == 0) {
481 #if __FreeBSD_version >= 1100045
482 tso_maxlen = hn_tso_maxlen;
483 if (tso_maxlen <= 0 || tso_maxlen > IP_MAXPACKET)
484 tso_maxlen = IP_MAXPACKET;
486 ifp->if_hw_tsomaxsegcount = HN_TX_DATA_SEGCNT_MAX;
487 ifp->if_hw_tsomaxsegsize = PAGE_SIZE;
488 ifp->if_hw_tsomax = tso_maxlen -
489 (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
492 ether_ifattach(ifp, device_info.mac_addr);
494 #if __FreeBSD_version >= 1100045
495 if_printf(ifp, "TSO: %u/%u/%u\n", ifp->if_hw_tsomax,
496 ifp->if_hw_tsomaxsegcount, ifp->if_hw_tsomaxsegsize);
499 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
500 hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max);
501 if (hn_tx_chimney_size > 0 &&
502 hn_tx_chimney_size < sc->hn_tx_chimney_max)
503 hn_set_tx_chimney_size(sc, hn_tx_chimney_size);
507 hn_destroy_tx_data(sc);
514 * Standard detach entry point
517 netvsc_detach(device_t dev)
519 struct hn_softc *sc = device_get_softc(dev);
520 struct hv_device *hv_device = vmbus_get_devctx(dev);
523 printf("netvsc_detach\n");
526 * XXXKYS: Need to clean up all our
527 * driver state; this is the driver
532 * XXXKYS: Need to stop outgoing traffic and unregister
536 hv_rf_on_device_remove(hv_device, HV_RF_NV_DESTROY_CHANNEL);
538 hn_stop_tx_tasks(sc);
540 ifmedia_removeall(&sc->hn_media);
541 hn_destroy_rx_data(sc);
542 hn_destroy_tx_data(sc);
544 if (sc->hn_tx_taskq != hn_tx_taskq)
545 taskqueue_free(sc->hn_tx_taskq);
551 * Standard shutdown entry point
554 netvsc_shutdown(device_t dev)
560 hn_txdesc_dmamap_load(struct hn_tx_ring *txr, struct hn_txdesc *txd,
561 struct mbuf **m_head, bus_dma_segment_t *segs, int *nsegs)
563 struct mbuf *m = *m_head;
566 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag, txd->data_dmap,
567 m, segs, nsegs, BUS_DMA_NOWAIT);
568 if (error == EFBIG) {
571 m_new = m_collapse(m, M_NOWAIT, HN_TX_DATA_SEGCNT_MAX);
576 txr->hn_tx_collapsed++;
578 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag,
579 txd->data_dmap, m, segs, nsegs, BUS_DMA_NOWAIT);
582 bus_dmamap_sync(txr->hn_tx_data_dtag, txd->data_dmap,
583 BUS_DMASYNC_PREWRITE);
584 txd->flags |= HN_TXD_FLAG_DMAMAP;
590 hn_txdesc_dmamap_unload(struct hn_tx_ring *txr, struct hn_txdesc *txd)
593 if (txd->flags & HN_TXD_FLAG_DMAMAP) {
594 bus_dmamap_sync(txr->hn_tx_data_dtag,
595 txd->data_dmap, BUS_DMASYNC_POSTWRITE);
596 bus_dmamap_unload(txr->hn_tx_data_dtag,
598 txd->flags &= ~HN_TXD_FLAG_DMAMAP;
603 hn_txdesc_put(struct hn_tx_ring *txr, struct hn_txdesc *txd)
606 KASSERT((txd->flags & HN_TXD_FLAG_ONLIST) == 0,
607 ("put an onlist txd %#x", txd->flags));
609 KASSERT(txd->refs > 0, ("invalid txd refs %d", txd->refs));
610 if (atomic_fetchadd_int(&txd->refs, -1) != 1)
613 hn_txdesc_dmamap_unload(txr, txd);
614 if (txd->m != NULL) {
619 txd->flags |= HN_TXD_FLAG_ONLIST;
621 #ifndef HN_USE_TXDESC_BUFRING
622 mtx_lock_spin(&txr->hn_txlist_spin);
623 KASSERT(txr->hn_txdesc_avail >= 0 &&
624 txr->hn_txdesc_avail < txr->hn_txdesc_cnt,
625 ("txdesc_put: invalid txd avail %d", txr->hn_txdesc_avail));
626 txr->hn_txdesc_avail++;
627 SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
628 mtx_unlock_spin(&txr->hn_txlist_spin);
630 atomic_add_int(&txr->hn_txdesc_avail, 1);
631 buf_ring_enqueue(txr->hn_txdesc_br, txd);
637 static __inline struct hn_txdesc *
638 hn_txdesc_get(struct hn_tx_ring *txr)
640 struct hn_txdesc *txd;
642 #ifndef HN_USE_TXDESC_BUFRING
643 mtx_lock_spin(&txr->hn_txlist_spin);
644 txd = SLIST_FIRST(&txr->hn_txlist);
646 KASSERT(txr->hn_txdesc_avail > 0,
647 ("txdesc_get: invalid txd avail %d", txr->hn_txdesc_avail));
648 txr->hn_txdesc_avail--;
649 SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
651 mtx_unlock_spin(&txr->hn_txlist_spin);
653 txd = buf_ring_dequeue_sc(txr->hn_txdesc_br);
657 #ifdef HN_USE_TXDESC_BUFRING
658 atomic_subtract_int(&txr->hn_txdesc_avail, 1);
660 KASSERT(txd->m == NULL && txd->refs == 0 &&
661 (txd->flags & HN_TXD_FLAG_ONLIST), ("invalid txd"));
662 txd->flags &= ~HN_TXD_FLAG_ONLIST;
669 hn_txdesc_hold(struct hn_txdesc *txd)
672 /* 0->1 transition will never work */
673 KASSERT(txd->refs > 0, ("invalid refs %d", txd->refs));
674 atomic_add_int(&txd->refs, 1);
678 hn_tx_done(void *xpkt)
680 netvsc_packet *packet = xpkt;
681 struct hn_txdesc *txd;
682 struct hn_tx_ring *txr;
684 txd = (struct hn_txdesc *)(uintptr_t)
685 packet->compl.send.send_completion_tid;
688 txr->hn_has_txeof = 1;
689 hn_txdesc_put(txr, txd);
693 netvsc_channel_rollup(struct hv_device *device_ctx)
695 struct hn_softc *sc = device_get_softc(device_ctx->device);
696 struct hn_tx_ring *txr = &sc->hn_tx_ring[0]; /* TODO: vRSS */
697 #if defined(INET) || defined(INET6)
698 struct hn_rx_ring *rxr = &sc->hn_rx_ring[0]; /* TODO: vRSS */
699 struct lro_ctrl *lro = &rxr->hn_lro;
700 struct lro_entry *queued;
702 while ((queued = SLIST_FIRST(&lro->lro_active)) != NULL) {
703 SLIST_REMOVE_HEAD(&lro->lro_active, next);
704 tcp_lro_flush(lro, queued);
708 if (!txr->hn_has_txeof)
711 txr->hn_has_txeof = 0;
717 * If this function fails, then both txd and m_head0 will be freed.
720 hn_encap(struct hn_tx_ring *txr, struct hn_txdesc *txd, struct mbuf **m_head0)
722 bus_dma_segment_t segs[HN_TX_DATA_SEGCNT_MAX];
724 struct mbuf *m_head = *m_head0;
725 netvsc_packet *packet;
726 rndis_msg *rndis_mesg;
727 rndis_packet *rndis_pkt;
728 rndis_per_packet_info *rppi;
729 uint32_t rndis_msg_size;
731 packet = &txd->netvsc_pkt;
732 packet->is_data_pkt = TRUE;
733 packet->tot_data_buf_len = m_head->m_pkthdr.len;
736 * extension points to the area reserved for the
737 * rndis_filter_packet, which is placed just after
738 * the netvsc_packet (and rppi struct, if present;
739 * length is updated later).
741 rndis_mesg = txd->rndis_msg;
742 /* XXX not necessary */
743 memset(rndis_mesg, 0, HN_RNDIS_MSG_LEN);
744 rndis_mesg->ndis_msg_type = REMOTE_NDIS_PACKET_MSG;
746 rndis_pkt = &rndis_mesg->msg.packet;
747 rndis_pkt->data_offset = sizeof(rndis_packet);
748 rndis_pkt->data_length = packet->tot_data_buf_len;
749 rndis_pkt->per_pkt_info_offset = sizeof(rndis_packet);
751 rndis_msg_size = RNDIS_MESSAGE_SIZE(rndis_packet);
753 if (m_head->m_flags & M_VLANTAG) {
754 ndis_8021q_info *rppi_vlan_info;
756 rndis_msg_size += RNDIS_VLAN_PPI_SIZE;
757 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_VLAN_PPI_SIZE,
760 rppi_vlan_info = (ndis_8021q_info *)((uint8_t *)rppi +
761 rppi->per_packet_info_offset);
762 rppi_vlan_info->u1.s1.vlan_id =
763 m_head->m_pkthdr.ether_vtag & 0xfff;
766 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
767 rndis_tcp_tso_info *tso_info;
768 struct ether_vlan_header *eh;
772 * XXX need m_pullup and use mtodo
774 eh = mtod(m_head, struct ether_vlan_header*);
775 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN))
776 ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
778 ether_len = ETHER_HDR_LEN;
780 rndis_msg_size += RNDIS_TSO_PPI_SIZE;
781 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_TSO_PPI_SIZE,
782 tcp_large_send_info);
784 tso_info = (rndis_tcp_tso_info *)((uint8_t *)rppi +
785 rppi->per_packet_info_offset);
786 tso_info->lso_v2_xmit.type =
787 RNDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
790 if (m_head->m_pkthdr.csum_flags & CSUM_IP_TSO) {
792 (struct ip *)(m_head->m_data + ether_len);
793 unsigned long iph_len = ip->ip_hl << 2;
795 (struct tcphdr *)((caddr_t)ip + iph_len);
797 tso_info->lso_v2_xmit.ip_version =
798 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
802 th->th_sum = in_pseudo(ip->ip_src.s_addr,
803 ip->ip_dst.s_addr, htons(IPPROTO_TCP));
806 #if defined(INET6) && defined(INET)
811 struct ip6_hdr *ip6 = (struct ip6_hdr *)
812 (m_head->m_data + ether_len);
813 struct tcphdr *th = (struct tcphdr *)(ip6 + 1);
815 tso_info->lso_v2_xmit.ip_version =
816 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
818 th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
821 tso_info->lso_v2_xmit.tcp_header_offset = 0;
822 tso_info->lso_v2_xmit.mss = m_head->m_pkthdr.tso_segsz;
823 } else if (m_head->m_pkthdr.csum_flags & txr->hn_csum_assist) {
824 rndis_tcp_ip_csum_info *csum_info;
826 rndis_msg_size += RNDIS_CSUM_PPI_SIZE;
827 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_CSUM_PPI_SIZE,
829 csum_info = (rndis_tcp_ip_csum_info *)((uint8_t *)rppi +
830 rppi->per_packet_info_offset);
832 csum_info->xmit.is_ipv4 = 1;
833 if (m_head->m_pkthdr.csum_flags & CSUM_IP)
834 csum_info->xmit.ip_header_csum = 1;
836 if (m_head->m_pkthdr.csum_flags & CSUM_TCP) {
837 csum_info->xmit.tcp_csum = 1;
838 csum_info->xmit.tcp_header_offset = 0;
839 } else if (m_head->m_pkthdr.csum_flags & CSUM_UDP) {
840 csum_info->xmit.udp_csum = 1;
844 rndis_mesg->msg_len = packet->tot_data_buf_len + rndis_msg_size;
845 packet->tot_data_buf_len = rndis_mesg->msg_len;
848 * Chimney send, if the packet could fit into one chimney buffer.
850 if (packet->tot_data_buf_len < txr->hn_tx_chimney_size) {
851 netvsc_dev *net_dev = txr->hn_sc->net_dev;
852 uint32_t send_buf_section_idx;
854 send_buf_section_idx =
855 hv_nv_get_next_send_section(net_dev);
856 if (send_buf_section_idx !=
857 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX) {
858 uint8_t *dest = ((uint8_t *)net_dev->send_buf +
859 (send_buf_section_idx *
860 net_dev->send_section_size));
862 memcpy(dest, rndis_mesg, rndis_msg_size);
863 dest += rndis_msg_size;
864 m_copydata(m_head, 0, m_head->m_pkthdr.len, dest);
866 packet->send_buf_section_idx = send_buf_section_idx;
867 packet->send_buf_section_size =
868 packet->tot_data_buf_len;
869 packet->page_buf_count = 0;
870 txr->hn_tx_chimney++;
875 error = hn_txdesc_dmamap_load(txr, txd, &m_head, segs, &nsegs);
880 * This mbuf is not linked w/ the txd yet, so free it now.
885 freed = hn_txdesc_put(txr, txd);
887 ("fail to free txd upon txdma error"));
889 txr->hn_txdma_failed++;
890 if_inc_counter(txr->hn_sc->hn_ifp, IFCOUNTER_OERRORS, 1);
895 packet->page_buf_count = nsegs + HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
897 /* send packet with page buffer */
898 packet->page_buffers[0].pfn = atop(txd->rndis_msg_paddr);
899 packet->page_buffers[0].offset = txd->rndis_msg_paddr & PAGE_MASK;
900 packet->page_buffers[0].length = rndis_msg_size;
903 * Fill the page buffers with mbuf info starting at index
904 * HV_RF_NUM_TX_RESERVED_PAGE_BUFS.
906 for (i = 0; i < nsegs; ++i) {
907 hv_vmbus_page_buffer *pb = &packet->page_buffers[
908 i + HV_RF_NUM_TX_RESERVED_PAGE_BUFS];
910 pb->pfn = atop(segs[i].ds_addr);
911 pb->offset = segs[i].ds_addr & PAGE_MASK;
912 pb->length = segs[i].ds_len;
915 packet->send_buf_section_idx =
916 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX;
917 packet->send_buf_section_size = 0;
921 /* Set the completion routine */
922 packet->compl.send.on_send_completion = hn_tx_done;
923 packet->compl.send.send_completion_context = packet;
924 packet->compl.send.send_completion_tid = (uint64_t)(uintptr_t)txd;
931 * If this function fails, then txd will be freed, but the mbuf
932 * associated w/ the txd will _not_ be freed.
935 hn_send_pkt(struct ifnet *ifp, struct hv_device *device_ctx,
936 struct hn_tx_ring *txr, struct hn_txdesc *txd)
938 int error, send_failed = 0;
942 * Make sure that txd is not freed before ETHER_BPF_MTAP.
945 error = hv_nv_on_send(device_ctx, &txd->netvsc_pkt);
947 ETHER_BPF_MTAP(ifp, txd->m);
948 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
950 hn_txdesc_put(txr, txd);
952 if (__predict_false(error)) {
956 * This should "really rarely" happen.
958 * XXX Too many RX to be acked or too many sideband
959 * commands to run? Ask netvsc_channel_rollup()
960 * to kick start later.
962 txr->hn_has_txeof = 1;
964 txr->hn_send_failed++;
967 * Try sending again after set hn_has_txeof;
968 * in case that we missed the last
969 * netvsc_channel_rollup().
973 if_printf(ifp, "send failed\n");
976 * Caller will perform further processing on the
977 * associated mbuf, so don't free it in hn_txdesc_put();
978 * only unload it from the DMA map in hn_txdesc_put(),
982 freed = hn_txdesc_put(txr, txd);
984 ("fail to free txd upon send error"));
986 txr->hn_send_failed++;
992 * Start a transmit of one or more packets
995 hn_start_locked(struct hn_tx_ring *txr, int len)
997 struct hn_softc *sc = txr->hn_sc;
998 struct ifnet *ifp = sc->hn_ifp;
999 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1001 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1002 mtx_assert(&txr->hn_tx_lock, MA_OWNED);
1004 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1008 while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
1009 struct hn_txdesc *txd;
1010 struct mbuf *m_head;
1013 IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
1017 if (len > 0 && m_head->m_pkthdr.len > len) {
1019 * This sending could be time consuming; let callers
1020 * dispatch this packet sending (and sending of any
1021 * following up packets) to tx taskqueue.
1023 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1027 txd = hn_txdesc_get(txr);
1029 txr->hn_no_txdescs++;
1030 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1031 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1035 error = hn_encap(txr, txd, &m_head);
1037 /* Both txd and m_head are freed */
1041 error = hn_send_pkt(ifp, device_ctx, txr, txd);
1042 if (__predict_false(error)) {
1043 /* txd is freed, but m_head is not */
1044 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1045 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1053 * Link up/down notification
1056 netvsc_linkstatus_callback(struct hv_device *device_obj, uint32_t status)
1058 hn_softc_t *sc = device_get_softc(device_obj->device);
1072 * Append the specified data to the indicated mbuf chain,
1073 * Extend the mbuf chain if the new data does not fit in
1076 * This is a minor rewrite of m_append() from sys/kern/uipc_mbuf.c.
1077 * There should be an equivalent in the kernel mbuf code,
1078 * but there does not appear to be one yet.
1080 * Differs from m_append() in that additional mbufs are
1081 * allocated with cluster size MJUMPAGESIZE, and filled
1084 * Return 1 if able to complete the job; otherwise 0.
1087 hv_m_append(struct mbuf *m0, int len, c_caddr_t cp)
1090 int remainder, space;
1092 for (m = m0; m->m_next != NULL; m = m->m_next)
1095 space = M_TRAILINGSPACE(m);
1098 * Copy into available space.
1100 if (space > remainder)
1102 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1107 while (remainder > 0) {
1109 * Allocate a new mbuf; could check space
1110 * and allocate a cluster instead.
1112 n = m_getjcl(M_DONTWAIT, m->m_type, 0, MJUMPAGESIZE);
1115 n->m_len = min(MJUMPAGESIZE, remainder);
1116 bcopy(cp, mtod(n, caddr_t), n->m_len);
1118 remainder -= n->m_len;
1122 if (m0->m_flags & M_PKTHDR)
1123 m0->m_pkthdr.len += len - remainder;
1125 return (remainder == 0);
1130 * Called when we receive a data packet from the "wire" on the
1133 * Note: This is no longer used as a callback
1136 netvsc_recv(struct hv_device *device_ctx, netvsc_packet *packet,
1137 rndis_tcp_ip_csum_info *csum_info)
1139 struct hn_softc *sc = device_get_softc(device_ctx->device);
1140 struct hn_rx_ring *rxr = &sc->hn_rx_ring[0]; /* TODO: vRSS */
1143 int size, do_lro = 0, do_csum = 1;
1146 return (0); /* TODO: KYS how can this be! */
1151 ifp = sc->arpcom.ac_ifp;
1153 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1158 * Bail out if packet contains more data than configured MTU.
1160 if (packet->tot_data_buf_len > (ifp->if_mtu + ETHER_HDR_LEN)) {
1162 } else if (packet->tot_data_buf_len <= MHLEN) {
1163 m_new = m_gethdr(M_NOWAIT, MT_DATA);
1166 memcpy(mtod(m_new, void *), packet->data,
1167 packet->tot_data_buf_len);
1168 m_new->m_pkthdr.len = m_new->m_len = packet->tot_data_buf_len;
1169 rxr->hn_small_pkts++;
1172 * Get an mbuf with a cluster. For packets 2K or less,
1173 * get a standard 2K cluster. For anything larger, get a
1174 * 4K cluster. Any buffers larger than 4K can cause problems
1175 * if looped around to the Hyper-V TX channel, so avoid them.
1178 if (packet->tot_data_buf_len > MCLBYTES) {
1180 size = MJUMPAGESIZE;
1183 m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, size);
1184 if (m_new == NULL) {
1185 if_printf(ifp, "alloc mbuf failed.\n");
1189 hv_m_append(m_new, packet->tot_data_buf_len, packet->data);
1191 m_new->m_pkthdr.rcvif = ifp;
1193 if (__predict_false((ifp->if_capenable & IFCAP_RXCSUM) == 0))
1196 /* receive side checksum offload */
1197 if (csum_info != NULL) {
1198 /* IP csum offload */
1199 if (csum_info->receive.ip_csum_succeeded && do_csum) {
1200 m_new->m_pkthdr.csum_flags |=
1201 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1205 /* TCP/UDP csum offload */
1206 if ((csum_info->receive.tcp_csum_succeeded ||
1207 csum_info->receive.udp_csum_succeeded) && do_csum) {
1208 m_new->m_pkthdr.csum_flags |=
1209 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1210 m_new->m_pkthdr.csum_data = 0xffff;
1211 if (csum_info->receive.tcp_csum_succeeded)
1217 if (csum_info->receive.ip_csum_succeeded &&
1218 csum_info->receive.tcp_csum_succeeded)
1221 const struct ether_header *eh;
1226 if (m_new->m_len < hoff)
1228 eh = mtod(m_new, struct ether_header *);
1229 etype = ntohs(eh->ether_type);
1230 if (etype == ETHERTYPE_VLAN) {
1231 const struct ether_vlan_header *evl;
1233 hoff = sizeof(*evl);
1234 if (m_new->m_len < hoff)
1236 evl = mtod(m_new, struct ether_vlan_header *);
1237 etype = ntohs(evl->evl_proto);
1240 if (etype == ETHERTYPE_IP) {
1243 pr = hn_check_iplen(m_new, hoff);
1244 if (pr == IPPROTO_TCP) {
1246 (rxr->hn_trust_hcsum &
1247 HN_TRUST_HCSUM_TCP)) {
1248 rxr->hn_csum_trusted++;
1249 m_new->m_pkthdr.csum_flags |=
1250 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1251 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1252 m_new->m_pkthdr.csum_data = 0xffff;
1254 /* Rely on SW csum verification though... */
1256 } else if (pr == IPPROTO_UDP) {
1258 (rxr->hn_trust_hcsum &
1259 HN_TRUST_HCSUM_UDP)) {
1260 rxr->hn_csum_trusted++;
1261 m_new->m_pkthdr.csum_flags |=
1262 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1263 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1264 m_new->m_pkthdr.csum_data = 0xffff;
1266 } else if (pr != IPPROTO_DONE && do_csum &&
1267 (rxr->hn_trust_hcsum & HN_TRUST_HCSUM_IP)) {
1268 rxr->hn_csum_trusted++;
1269 m_new->m_pkthdr.csum_flags |=
1270 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1275 if ((packet->vlan_tci != 0) &&
1276 (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0) {
1277 m_new->m_pkthdr.ether_vtag = packet->vlan_tci;
1278 m_new->m_flags |= M_VLANTAG;
1282 * Note: Moved RX completion back to hv_nv_on_receive() so all
1283 * messages (not just data messages) will trigger a response.
1288 if ((ifp->if_capenable & IFCAP_LRO) && do_lro) {
1289 #if defined(INET) || defined(INET6)
1290 struct lro_ctrl *lro = &rxr->hn_lro;
1293 rxr->hn_lro_tried++;
1294 if (tcp_lro_rx(lro, m_new, 0) == 0) {
1302 /* We're not holding the lock here, so don't release it */
1303 (*ifp->if_input)(ifp, m_new);
1309 netvsc_recv_rollup(struct hv_device *device_ctx __unused)
1314 * Rules for using sc->temp_unusable:
1315 * 1. sc->temp_unusable can only be read or written while holding NV_LOCK()
1316 * 2. code reading sc->temp_unusable under NV_LOCK(), and finding
1317 * sc->temp_unusable set, must release NV_LOCK() and exit
1318 * 3. to retain exclusive control of the interface,
1319 * sc->temp_unusable must be set by code before releasing NV_LOCK()
1320 * 4. only code setting sc->temp_unusable can clear sc->temp_unusable
1321 * 5. code setting sc->temp_unusable must eventually clear sc->temp_unusable
1325 * Standard ioctl entry point. Called when the user wants to configure
1329 hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1331 hn_softc_t *sc = ifp->if_softc;
1332 struct ifreq *ifr = (struct ifreq *)data;
1334 struct ifaddr *ifa = (struct ifaddr *)data;
1336 netvsc_device_info device_info;
1337 struct hv_device *hn_dev;
1338 int mask, error = 0;
1339 int retry_cnt = 500;
1345 if (ifa->ifa_addr->sa_family == AF_INET) {
1346 ifp->if_flags |= IFF_UP;
1347 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1349 arp_ifinit(ifp, ifa);
1352 error = ether_ioctl(ifp, cmd, data);
1355 hn_dev = vmbus_get_devctx(sc->hn_dev);
1357 /* Check MTU value change */
1358 if (ifp->if_mtu == ifr->ifr_mtu)
1361 if (ifr->ifr_mtu > NETVSC_MAX_CONFIGURABLE_MTU) {
1366 /* Obtain and record requested MTU */
1367 ifp->if_mtu = ifr->ifr_mtu;
1369 #if __FreeBSD_version >= 1100099
1371 * Make sure that LRO aggregation length limit is still
1372 * valid, after the MTU change.
1375 if (sc->hn_rx_ring[0].hn_lro.lro_length_lim <
1376 HN_LRO_LENLIM_MIN(ifp)) {
1378 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
1379 sc->hn_rx_ring[i].hn_lro.lro_length_lim =
1380 HN_LRO_LENLIM_MIN(ifp);
1388 if (!sc->temp_unusable) {
1389 sc->temp_unusable = TRUE;
1393 if (retry_cnt > 0) {
1397 } while (retry_cnt > 0);
1399 if (retry_cnt == 0) {
1404 /* We must remove and add back the device to cause the new
1405 * MTU to take effect. This includes tearing down, but not
1406 * deleting the channel, then bringing it back up.
1408 error = hv_rf_on_device_remove(hn_dev, HV_RF_NV_RETAIN_CHANNEL);
1411 sc->temp_unusable = FALSE;
1415 error = hv_rf_on_device_add(hn_dev, &device_info);
1418 sc->temp_unusable = FALSE;
1423 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
1424 if (sc->hn_tx_ring[0].hn_tx_chimney_size >
1425 sc->hn_tx_chimney_max)
1426 hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max);
1428 hn_ifinit_locked(sc);
1431 sc->temp_unusable = FALSE;
1437 if (!sc->temp_unusable) {
1438 sc->temp_unusable = TRUE;
1442 if (retry_cnt > 0) {
1446 } while (retry_cnt > 0);
1448 if (retry_cnt == 0) {
1453 if (ifp->if_flags & IFF_UP) {
1455 * If only the state of the PROMISC flag changed,
1456 * then just use the 'set promisc mode' command
1457 * instead of reinitializing the entire NIC. Doing
1458 * a full re-init means reloading the firmware and
1459 * waiting for it to start up, which may take a
1463 /* Fixme: Promiscuous mode? */
1464 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1465 ifp->if_flags & IFF_PROMISC &&
1466 !(sc->hn_if_flags & IFF_PROMISC)) {
1467 /* do something here for Hyper-V */
1468 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1469 !(ifp->if_flags & IFF_PROMISC) &&
1470 sc->hn_if_flags & IFF_PROMISC) {
1471 /* do something here for Hyper-V */
1474 hn_ifinit_locked(sc);
1476 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1481 sc->temp_unusable = FALSE;
1483 sc->hn_if_flags = ifp->if_flags;
1489 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1490 if (mask & IFCAP_TXCSUM) {
1491 ifp->if_capenable ^= IFCAP_TXCSUM;
1492 if (ifp->if_capenable & IFCAP_TXCSUM) {
1494 sc->hn_tx_ring[0].hn_csum_assist;
1497 ~sc->hn_tx_ring[0].hn_csum_assist;
1501 if (mask & IFCAP_RXCSUM)
1502 ifp->if_capenable ^= IFCAP_RXCSUM;
1504 if (mask & IFCAP_LRO)
1505 ifp->if_capenable ^= IFCAP_LRO;
1507 if (mask & IFCAP_TSO4) {
1508 ifp->if_capenable ^= IFCAP_TSO4;
1509 if (ifp->if_capenable & IFCAP_TSO4)
1510 ifp->if_hwassist |= CSUM_IP_TSO;
1512 ifp->if_hwassist &= ~CSUM_IP_TSO;
1515 if (mask & IFCAP_TSO6) {
1516 ifp->if_capenable ^= IFCAP_TSO6;
1517 if (ifp->if_capenable & IFCAP_TSO6)
1518 ifp->if_hwassist |= CSUM_IP6_TSO;
1520 ifp->if_hwassist &= ~CSUM_IP6_TSO;
1529 /* Fixme: Multicast mode? */
1530 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1532 netvsc_setmulti(sc);
1541 error = ifmedia_ioctl(ifp, ifr, &sc->hn_media, cmd);
1544 error = ether_ioctl(ifp, cmd, data);
1555 hn_stop(hn_softc_t *sc)
1559 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1564 printf(" Closing Device ...\n");
1566 atomic_clear_int(&ifp->if_drv_flags,
1567 (IFF_DRV_RUNNING | IFF_DRV_OACTIVE));
1568 if_link_state_change(ifp, LINK_STATE_DOWN);
1569 sc->hn_initdone = 0;
1571 ret = hv_rf_on_close(device_ctx);
1575 * FreeBSD transmit entry point
1578 hn_start(struct ifnet *ifp)
1580 struct hn_softc *sc = ifp->if_softc;
1581 struct hn_tx_ring *txr = &sc->hn_tx_ring[0];
1583 if (txr->hn_sched_tx)
1586 if (mtx_trylock(&txr->hn_tx_lock)) {
1589 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1590 mtx_unlock(&txr->hn_tx_lock);
1595 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
1599 hn_start_txeof(struct hn_tx_ring *txr)
1601 struct hn_softc *sc = txr->hn_sc;
1602 struct ifnet *ifp = sc->hn_ifp;
1604 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1606 if (txr->hn_sched_tx)
1609 if (mtx_trylock(&txr->hn_tx_lock)) {
1612 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1613 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1614 mtx_unlock(&txr->hn_tx_lock);
1616 taskqueue_enqueue(txr->hn_tx_taskq,
1622 * Release the OACTIVE earlier, with the hope, that
1623 * others could catch up. The task will clear the
1624 * flag again with the hn_tx_lock to avoid possible
1627 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1628 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
1636 hn_ifinit_locked(hn_softc_t *sc)
1639 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1644 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1648 hv_promisc_mode = 1;
1650 ret = hv_rf_on_open(device_ctx);
1654 sc->hn_initdone = 1;
1656 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1657 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_RUNNING);
1658 if_link_state_change(ifp, LINK_STATE_UP);
1665 hn_ifinit(void *xsc)
1667 hn_softc_t *sc = xsc;
1670 if (sc->temp_unusable) {
1674 sc->temp_unusable = TRUE;
1677 hn_ifinit_locked(sc);
1680 sc->temp_unusable = FALSE;
1689 hn_watchdog(struct ifnet *ifp)
1694 printf("hn%d: watchdog timeout -- resetting\n", sc->hn_unit);
1695 hn_ifinit(sc); /*???*/
1700 #if __FreeBSD_version >= 1100099
1703 hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS)
1705 struct hn_softc *sc = arg1;
1706 unsigned int lenlim;
1709 lenlim = sc->hn_rx_ring[0].hn_lro.lro_length_lim;
1710 error = sysctl_handle_int(oidp, &lenlim, 0, req);
1711 if (error || req->newptr == NULL)
1714 if (lenlim < HN_LRO_LENLIM_MIN(sc->hn_ifp) ||
1715 lenlim > TCP_LRO_LENGTH_MAX)
1719 for (i = 0; i < sc->hn_rx_ring_cnt; ++i)
1720 sc->hn_rx_ring[i].hn_lro.lro_length_lim = lenlim;
1726 hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS)
1728 struct hn_softc *sc = arg1;
1729 int ackcnt, error, i;
1732 * lro_ackcnt_lim is append count limit,
1733 * +1 to turn it into aggregation limit.
1735 ackcnt = sc->hn_rx_ring[0].hn_lro.lro_ackcnt_lim + 1;
1736 error = sysctl_handle_int(oidp, &ackcnt, 0, req);
1737 if (error || req->newptr == NULL)
1740 if (ackcnt < 2 || ackcnt > (TCP_LRO_ACKCNT_MAX + 1))
1744 * Convert aggregation limit back to append
1749 for (i = 0; i < sc->hn_rx_ring_cnt; ++i)
1750 sc->hn_rx_ring[i].hn_lro.lro_ackcnt_lim = ackcnt;
1758 hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS)
1760 struct hn_softc *sc = arg1;
1765 if (sc->hn_rx_ring[0].hn_trust_hcsum & hcsum)
1768 error = sysctl_handle_int(oidp, &on, 0, req);
1769 if (error || req->newptr == NULL)
1773 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
1774 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
1777 rxr->hn_trust_hcsum |= hcsum;
1779 rxr->hn_trust_hcsum &= ~hcsum;
1786 hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS)
1788 struct hn_softc *sc = arg1;
1789 int chimney_size, error;
1791 chimney_size = sc->hn_tx_ring[0].hn_tx_chimney_size;
1792 error = sysctl_handle_int(oidp, &chimney_size, 0, req);
1793 if (error || req->newptr == NULL)
1796 if (chimney_size > sc->hn_tx_chimney_max || chimney_size <= 0)
1799 hn_set_tx_chimney_size(sc, chimney_size);
1803 #if __FreeBSD_version < 1100095
1805 hn_rx_stat_int_sysctl(SYSCTL_HANDLER_ARGS)
1807 struct hn_softc *sc = arg1;
1808 int ofs = arg2, i, error;
1809 struct hn_rx_ring *rxr;
1813 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
1814 rxr = &sc->hn_rx_ring[i];
1815 stat += *((int *)((uint8_t *)rxr + ofs));
1818 error = sysctl_handle_64(oidp, &stat, 0, req);
1819 if (error || req->newptr == NULL)
1822 /* Zero out this stat. */
1823 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
1824 rxr = &sc->hn_rx_ring[i];
1825 *((int *)((uint8_t *)rxr + ofs)) = 0;
1831 hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS)
1833 struct hn_softc *sc = arg1;
1834 int ofs = arg2, i, error;
1835 struct hn_rx_ring *rxr;
1839 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
1840 rxr = &sc->hn_rx_ring[i];
1841 stat += *((uint64_t *)((uint8_t *)rxr + ofs));
1844 error = sysctl_handle_64(oidp, &stat, 0, req);
1845 if (error || req->newptr == NULL)
1848 /* Zero out this stat. */
1849 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
1850 rxr = &sc->hn_rx_ring[i];
1851 *((uint64_t *)((uint8_t *)rxr + ofs)) = 0;
1859 hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
1861 struct hn_softc *sc = arg1;
1862 int ofs = arg2, i, error;
1863 struct hn_rx_ring *rxr;
1867 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
1868 rxr = &sc->hn_rx_ring[i];
1869 stat += *((u_long *)((uint8_t *)rxr + ofs));
1872 error = sysctl_handle_long(oidp, &stat, 0, req);
1873 if (error || req->newptr == NULL)
1876 /* Zero out this stat. */
1877 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
1878 rxr = &sc->hn_rx_ring[i];
1879 *((u_long *)((uint8_t *)rxr + ofs)) = 0;
1885 hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
1887 struct hn_softc *sc = arg1;
1888 int ofs = arg2, i, error;
1889 struct hn_tx_ring *txr;
1893 for (i = 0; i < sc->hn_tx_ring_cnt; ++i) {
1894 txr = &sc->hn_tx_ring[i];
1895 stat += *((u_long *)((uint8_t *)txr + ofs));
1898 error = sysctl_handle_long(oidp, &stat, 0, req);
1899 if (error || req->newptr == NULL)
1902 /* Zero out this stat. */
1903 for (i = 0; i < sc->hn_tx_ring_cnt; ++i) {
1904 txr = &sc->hn_tx_ring[i];
1905 *((u_long *)((uint8_t *)txr + ofs)) = 0;
1911 hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS)
1913 struct hn_softc *sc = arg1;
1914 int ofs = arg2, i, error, conf;
1915 struct hn_tx_ring *txr;
1917 txr = &sc->hn_tx_ring[0];
1918 conf = *((int *)((uint8_t *)txr + ofs));
1920 error = sysctl_handle_int(oidp, &conf, 0, req);
1921 if (error || req->newptr == NULL)
1925 for (i = 0; i < sc->hn_tx_ring_cnt; ++i) {
1926 txr = &sc->hn_tx_ring[i];
1927 *((int *)((uint8_t *)txr + ofs)) = conf;
1935 hn_check_iplen(const struct mbuf *m, int hoff)
1937 const struct ip *ip;
1938 int len, iphlen, iplen;
1939 const struct tcphdr *th;
1940 int thoff; /* TCP data offset */
1942 len = hoff + sizeof(struct ip);
1944 /* The packet must be at least the size of an IP header. */
1945 if (m->m_pkthdr.len < len)
1946 return IPPROTO_DONE;
1948 /* The fixed IP header must reside completely in the first mbuf. */
1950 return IPPROTO_DONE;
1952 ip = mtodo(m, hoff);
1954 /* Bound check the packet's stated IP header length. */
1955 iphlen = ip->ip_hl << 2;
1956 if (iphlen < sizeof(struct ip)) /* minimum header length */
1957 return IPPROTO_DONE;
1959 /* The full IP header must reside completely in the one mbuf. */
1960 if (m->m_len < hoff + iphlen)
1961 return IPPROTO_DONE;
1963 iplen = ntohs(ip->ip_len);
1966 * Check that the amount of data in the buffers is as
1967 * at least much as the IP header would have us expect.
1969 if (m->m_pkthdr.len < hoff + iplen)
1970 return IPPROTO_DONE;
1973 * Ignore IP fragments.
1975 if (ntohs(ip->ip_off) & (IP_OFFMASK | IP_MF))
1976 return IPPROTO_DONE;
1979 * The TCP/IP or UDP/IP header must be entirely contained within
1980 * the first fragment of a packet.
1984 if (iplen < iphlen + sizeof(struct tcphdr))
1985 return IPPROTO_DONE;
1986 if (m->m_len < hoff + iphlen + sizeof(struct tcphdr))
1987 return IPPROTO_DONE;
1988 th = (const struct tcphdr *)((const uint8_t *)ip + iphlen);
1989 thoff = th->th_off << 2;
1990 if (thoff < sizeof(struct tcphdr) || thoff + iphlen > iplen)
1991 return IPPROTO_DONE;
1992 if (m->m_len < hoff + iphlen + thoff)
1993 return IPPROTO_DONE;
1996 if (iplen < iphlen + sizeof(struct udphdr))
1997 return IPPROTO_DONE;
1998 if (m->m_len < hoff + iphlen + sizeof(struct udphdr))
1999 return IPPROTO_DONE;
2003 return IPPROTO_DONE;
2010 hn_dma_map_paddr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
2012 bus_addr_t *paddr = arg;
2017 KASSERT(nseg == 1, ("too many segments %d!", nseg));
2018 *paddr = segs->ds_addr;
2022 hn_create_rx_data(struct hn_softc *sc)
2024 struct sysctl_oid_list *child;
2025 struct sysctl_ctx_list *ctx;
2026 device_t dev = sc->hn_dev;
2027 #if defined(INET) || defined(INET6)
2028 #if __FreeBSD_version >= 1100095
2034 sc->hn_rx_ring_cnt = 1; /* TODO: vRSS */
2035 sc->hn_rx_ring = malloc(sizeof(struct hn_rx_ring) * sc->hn_rx_ring_cnt,
2036 M_NETVSC, M_WAITOK | M_ZERO);
2038 #if defined(INET) || defined(INET6)
2039 #if __FreeBSD_version >= 1100095
2040 lroent_cnt = hn_lro_entry_count;
2041 if (lroent_cnt < TCP_LRO_ENTRIES)
2042 lroent_cnt = TCP_LRO_ENTRIES;
2043 device_printf(dev, "LRO: entry count %d\n", lroent_cnt);
2045 #endif /* INET || INET6 */
2047 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2048 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
2050 if (hn_trust_hosttcp)
2051 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_TCP;
2052 if (hn_trust_hostudp)
2053 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_UDP;
2054 if (hn_trust_hostip)
2055 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_IP;
2060 #if defined(INET) || defined(INET6)
2061 #if __FreeBSD_version >= 1100095
2062 tcp_lro_init_args(&rxr->hn_lro, sc->hn_ifp, lroent_cnt, 0);
2064 tcp_lro_init(&rxr->hn_lro);
2065 rxr->hn_lro.ifp = sc->hn_ifp;
2067 #if __FreeBSD_version >= 1100099
2068 rxr->hn_lro.lro_length_lim = HN_LRO_LENLIM_DEF;
2069 rxr->hn_lro.lro_ackcnt_lim = HN_LRO_ACKCNT_DEF;
2071 #endif /* INET || INET6 */
2074 ctx = device_get_sysctl_ctx(dev);
2075 child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
2077 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_queued",
2078 CTLTYPE_U64 | CTLFLAG_RW, sc,
2079 __offsetof(struct hn_rx_ring, hn_lro.lro_queued),
2080 #if __FreeBSD_version < 1100095
2081 hn_rx_stat_int_sysctl,
2083 hn_rx_stat_u64_sysctl,
2085 "LU", "LRO queued");
2086 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_flushed",
2087 CTLTYPE_U64 | CTLFLAG_RW, sc,
2088 __offsetof(struct hn_rx_ring, hn_lro.lro_flushed),
2089 #if __FreeBSD_version < 1100095
2090 hn_rx_stat_int_sysctl,
2092 hn_rx_stat_u64_sysctl,
2094 "LU", "LRO flushed");
2095 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_tried",
2096 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2097 __offsetof(struct hn_rx_ring, hn_lro_tried),
2098 hn_rx_stat_ulong_sysctl, "LU", "# of LRO tries");
2099 #if __FreeBSD_version >= 1100099
2100 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_length_lim",
2101 CTLTYPE_UINT | CTLFLAG_RW, sc, 0, hn_lro_lenlim_sysctl, "IU",
2102 "Max # of data bytes to be aggregated by LRO");
2103 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_ackcnt_lim",
2104 CTLTYPE_INT | CTLFLAG_RW, sc, 0, hn_lro_ackcnt_sysctl, "I",
2105 "Max # of ACKs to be aggregated by LRO");
2107 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hosttcp",
2108 CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_TCP,
2109 hn_trust_hcsum_sysctl, "I",
2110 "Trust tcp segement verification on host side, "
2111 "when csum info is missing");
2112 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostudp",
2113 CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_UDP,
2114 hn_trust_hcsum_sysctl, "I",
2115 "Trust udp datagram verification on host side, "
2116 "when csum info is missing");
2117 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostip",
2118 CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_IP,
2119 hn_trust_hcsum_sysctl, "I",
2120 "Trust ip packet verification on host side, "
2121 "when csum info is missing");
2122 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_ip",
2123 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2124 __offsetof(struct hn_rx_ring, hn_csum_ip),
2125 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM IP");
2126 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_tcp",
2127 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2128 __offsetof(struct hn_rx_ring, hn_csum_tcp),
2129 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM TCP");
2130 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_udp",
2131 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2132 __offsetof(struct hn_rx_ring, hn_csum_udp),
2133 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM UDP");
2134 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_trusted",
2135 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2136 __offsetof(struct hn_rx_ring, hn_csum_trusted),
2137 hn_rx_stat_ulong_sysctl, "LU",
2138 "# of packets that we trust host's csum verification");
2139 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "small_pkts",
2140 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2141 __offsetof(struct hn_rx_ring, hn_small_pkts),
2142 hn_rx_stat_ulong_sysctl, "LU", "# of small packets received");
2146 hn_destroy_rx_data(struct hn_softc *sc)
2148 #if defined(INET) || defined(INET6)
2152 if (sc->hn_rx_ring_cnt == 0)
2155 #if defined(INET) || defined(INET6)
2156 for (i = 0; i < sc->hn_rx_ring_cnt; ++i)
2157 tcp_lro_free(&sc->hn_rx_ring[i].hn_lro);
2159 free(sc->hn_rx_ring, M_NETVSC);
2160 sc->hn_rx_ring = NULL;
2162 sc->hn_rx_ring_cnt = 0;
2166 hn_create_tx_ring(struct hn_softc *sc, int id)
2168 struct hn_tx_ring *txr = &sc->hn_tx_ring[id];
2169 bus_dma_tag_t parent_dtag;
2174 #ifndef HN_USE_TXDESC_BUFRING
2175 mtx_init(&txr->hn_txlist_spin, "hn txlist", NULL, MTX_SPIN);
2177 mtx_init(&txr->hn_tx_lock, "hn tx", NULL, MTX_DEF);
2179 txr->hn_txdesc_cnt = HN_TX_DESC_CNT;
2180 txr->hn_txdesc = malloc(sizeof(struct hn_txdesc) * txr->hn_txdesc_cnt,
2181 M_NETVSC, M_WAITOK | M_ZERO);
2182 #ifndef HN_USE_TXDESC_BUFRING
2183 SLIST_INIT(&txr->hn_txlist);
2185 txr->hn_txdesc_br = buf_ring_alloc(txr->hn_txdesc_cnt, M_NETVSC,
2186 M_WAITOK, &txr->hn_tx_lock);
2189 txr->hn_tx_taskq = sc->hn_tx_taskq;
2190 TASK_INIT(&txr->hn_tx_task, 0, hn_start_taskfunc, txr);
2191 TASK_INIT(&txr->hn_txeof_task, 0, hn_start_txeof_taskfunc, txr);
2193 txr->hn_direct_tx_size = hn_direct_tx_size;
2194 if (hv_vmbus_protocal_version >= HV_VMBUS_VERSION_WIN8_1)
2195 txr->hn_csum_assist = HN_CSUM_ASSIST;
2197 txr->hn_csum_assist = HN_CSUM_ASSIST_WIN8;
2200 * Always schedule transmission instead of trying to do direct
2201 * transmission. This one gives the best performance so far.
2203 txr->hn_sched_tx = 1;
2205 txr->hn_txeof = hn_start_txeof; /* TODO: if_transmit */
2207 parent_dtag = bus_get_dma_tag(sc->hn_dev);
2209 /* DMA tag for RNDIS messages. */
2210 error = bus_dma_tag_create(parent_dtag, /* parent */
2211 HN_RNDIS_MSG_ALIGN, /* alignment */
2212 HN_RNDIS_MSG_BOUNDARY, /* boundary */
2213 BUS_SPACE_MAXADDR, /* lowaddr */
2214 BUS_SPACE_MAXADDR, /* highaddr */
2215 NULL, NULL, /* filter, filterarg */
2216 HN_RNDIS_MSG_LEN, /* maxsize */
2218 HN_RNDIS_MSG_LEN, /* maxsegsize */
2220 NULL, /* lockfunc */
2221 NULL, /* lockfuncarg */
2222 &txr->hn_tx_rndis_dtag);
2224 device_printf(sc->hn_dev, "failed to create rndis dmatag\n");
2228 /* DMA tag for data. */
2229 error = bus_dma_tag_create(parent_dtag, /* parent */
2231 HN_TX_DATA_BOUNDARY, /* boundary */
2232 BUS_SPACE_MAXADDR, /* lowaddr */
2233 BUS_SPACE_MAXADDR, /* highaddr */
2234 NULL, NULL, /* filter, filterarg */
2235 HN_TX_DATA_MAXSIZE, /* maxsize */
2236 HN_TX_DATA_SEGCNT_MAX, /* nsegments */
2237 HN_TX_DATA_SEGSIZE, /* maxsegsize */
2239 NULL, /* lockfunc */
2240 NULL, /* lockfuncarg */
2241 &txr->hn_tx_data_dtag);
2243 device_printf(sc->hn_dev, "failed to create data dmatag\n");
2247 for (i = 0; i < txr->hn_txdesc_cnt; ++i) {
2248 struct hn_txdesc *txd = &txr->hn_txdesc[i];
2253 * Allocate and load RNDIS messages.
2255 error = bus_dmamem_alloc(txr->hn_tx_rndis_dtag,
2256 (void **)&txd->rndis_msg,
2257 BUS_DMA_WAITOK | BUS_DMA_COHERENT,
2258 &txd->rndis_msg_dmap);
2260 device_printf(sc->hn_dev,
2261 "failed to allocate rndis_msg, %d\n", i);
2265 error = bus_dmamap_load(txr->hn_tx_rndis_dtag,
2266 txd->rndis_msg_dmap,
2267 txd->rndis_msg, HN_RNDIS_MSG_LEN,
2268 hn_dma_map_paddr, &txd->rndis_msg_paddr,
2271 device_printf(sc->hn_dev,
2272 "failed to load rndis_msg, %d\n", i);
2273 bus_dmamem_free(txr->hn_tx_rndis_dtag,
2274 txd->rndis_msg, txd->rndis_msg_dmap);
2278 /* DMA map for TX data. */
2279 error = bus_dmamap_create(txr->hn_tx_data_dtag, 0,
2282 device_printf(sc->hn_dev,
2283 "failed to allocate tx data dmamap\n");
2284 bus_dmamap_unload(txr->hn_tx_rndis_dtag,
2285 txd->rndis_msg_dmap);
2286 bus_dmamem_free(txr->hn_tx_rndis_dtag,
2287 txd->rndis_msg, txd->rndis_msg_dmap);
2291 /* All set, put it to list */
2292 txd->flags |= HN_TXD_FLAG_ONLIST;
2293 #ifndef HN_USE_TXDESC_BUFRING
2294 SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
2296 buf_ring_enqueue(txr->hn_txdesc_br, txd);
2299 txr->hn_txdesc_avail = txr->hn_txdesc_cnt;
2301 if (sc->hn_tx_sysctl_tree != NULL) {
2302 struct sysctl_oid_list *child;
2303 struct sysctl_ctx_list *ctx;
2307 * Create per TX ring sysctl tree:
2308 * dev.hn.UNIT.tx.RINGID
2310 ctx = device_get_sysctl_ctx(sc->hn_dev);
2311 child = SYSCTL_CHILDREN(sc->hn_tx_sysctl_tree);
2313 snprintf(name, sizeof(name), "%d", id);
2314 txr->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO,
2315 name, CTLFLAG_RD, 0, "");
2317 if (txr->hn_tx_sysctl_tree != NULL) {
2318 child = SYSCTL_CHILDREN(txr->hn_tx_sysctl_tree);
2320 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_avail",
2321 CTLFLAG_RD, &txr->hn_txdesc_avail, 0,
2322 "# of available TX descs");
2330 hn_txdesc_dmamap_destroy(struct hn_txdesc *txd)
2332 struct hn_tx_ring *txr = txd->txr;
2334 KASSERT(txd->m == NULL, ("still has mbuf installed"));
2335 KASSERT((txd->flags & HN_TXD_FLAG_DMAMAP) == 0, ("still dma mapped"));
2337 bus_dmamap_unload(txr->hn_tx_rndis_dtag, txd->rndis_msg_dmap);
2338 bus_dmamem_free(txr->hn_tx_rndis_dtag, txd->rndis_msg,
2339 txd->rndis_msg_dmap);
2340 bus_dmamap_destroy(txr->hn_tx_data_dtag, txd->data_dmap);
2344 hn_destroy_tx_ring(struct hn_tx_ring *txr)
2346 struct hn_txdesc *txd;
2348 if (txr->hn_txdesc == NULL)
2351 #ifndef HN_USE_TXDESC_BUFRING
2352 while ((txd = SLIST_FIRST(&txr->hn_txlist)) != NULL) {
2353 SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
2354 hn_txdesc_dmamap_destroy(txd);
2357 while ((txd = buf_ring_dequeue_sc(txr->hn_txdesc_br)) != NULL)
2358 hn_txdesc_dmamap_destroy(txd);
2361 if (txr->hn_tx_data_dtag != NULL)
2362 bus_dma_tag_destroy(txr->hn_tx_data_dtag);
2363 if (txr->hn_tx_rndis_dtag != NULL)
2364 bus_dma_tag_destroy(txr->hn_tx_rndis_dtag);
2366 #ifdef HN_USE_TXDESC_BUFRING
2367 buf_ring_free(txr->hn_txdesc_br, M_NETVSC);
2370 free(txr->hn_txdesc, M_NETVSC);
2371 txr->hn_txdesc = NULL;
2373 #ifndef HN_USE_TXDESC_BUFRING
2374 mtx_destroy(&txr->hn_txlist_spin);
2376 mtx_destroy(&txr->hn_tx_lock);
2380 hn_create_tx_data(struct hn_softc *sc)
2382 struct sysctl_oid_list *child;
2383 struct sysctl_ctx_list *ctx;
2386 sc->hn_tx_ring_cnt = 1; /* TODO: vRSS */
2387 sc->hn_tx_ring = malloc(sizeof(struct hn_tx_ring) * sc->hn_tx_ring_cnt,
2388 M_NETVSC, M_WAITOK | M_ZERO);
2390 ctx = device_get_sysctl_ctx(sc->hn_dev);
2391 child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->hn_dev));
2393 /* Create dev.hn.UNIT.tx sysctl tree */
2394 sc->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "tx",
2397 for (i = 0; i < sc->hn_tx_ring_cnt; ++i) {
2400 error = hn_create_tx_ring(sc, i);
2405 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "no_txdescs",
2406 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2407 __offsetof(struct hn_tx_ring, hn_no_txdescs),
2408 hn_tx_stat_ulong_sysctl, "LU", "# of times short of TX descs");
2409 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "send_failed",
2410 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2411 __offsetof(struct hn_tx_ring, hn_send_failed),
2412 hn_tx_stat_ulong_sysctl, "LU", "# of hyper-v sending failure");
2413 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "txdma_failed",
2414 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2415 __offsetof(struct hn_tx_ring, hn_txdma_failed),
2416 hn_tx_stat_ulong_sysctl, "LU", "# of TX DMA failure");
2417 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_collapsed",
2418 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2419 __offsetof(struct hn_tx_ring, hn_tx_collapsed),
2420 hn_tx_stat_ulong_sysctl, "LU", "# of TX mbuf collapsed");
2421 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney",
2422 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2423 __offsetof(struct hn_tx_ring, hn_tx_chimney),
2424 hn_tx_stat_ulong_sysctl, "LU", "# of chimney send");
2425 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_cnt",
2426 CTLFLAG_RD, &sc->hn_tx_ring[0].hn_txdesc_cnt, 0,
2427 "# of total TX descs");
2428 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_chimney_max",
2429 CTLFLAG_RD, &sc->hn_tx_chimney_max, 0,
2430 "Chimney send packet size upper boundary");
2431 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_size",
2432 CTLTYPE_INT | CTLFLAG_RW, sc, 0, hn_tx_chimney_size_sysctl,
2433 "I", "Chimney send packet size limit");
2434 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "direct_tx_size",
2435 CTLTYPE_INT | CTLFLAG_RW, sc,
2436 __offsetof(struct hn_tx_ring, hn_direct_tx_size),
2437 hn_tx_conf_int_sysctl, "I",
2438 "Size of the packet for direct transmission");
2439 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "sched_tx",
2440 CTLTYPE_INT | CTLFLAG_RW, sc,
2441 __offsetof(struct hn_tx_ring, hn_sched_tx),
2442 hn_tx_conf_int_sysctl, "I",
2443 "Always schedule transmission "
2444 "instead of doing direct transmission");
2450 hn_set_tx_chimney_size(struct hn_softc *sc, int chimney_size)
2455 for (i = 0; i < sc->hn_tx_ring_cnt; ++i)
2456 sc->hn_tx_ring[i].hn_tx_chimney_size = chimney_size;
2461 hn_destroy_tx_data(struct hn_softc *sc)
2465 if (sc->hn_tx_ring_cnt == 0)
2468 for (i = 0; i < sc->hn_tx_ring_cnt; ++i)
2469 hn_destroy_tx_ring(&sc->hn_tx_ring[i]);
2471 free(sc->hn_tx_ring, M_NETVSC);
2472 sc->hn_tx_ring = NULL;
2474 sc->hn_tx_ring_cnt = 0;
2478 hn_start_taskfunc(void *xtxr, int pending __unused)
2480 struct hn_tx_ring *txr = xtxr;
2482 mtx_lock(&txr->hn_tx_lock);
2483 hn_start_locked(txr, 0);
2484 mtx_unlock(&txr->hn_tx_lock);
2488 hn_start_txeof_taskfunc(void *xtxr, int pending __unused)
2490 struct hn_tx_ring *txr = xtxr;
2492 mtx_lock(&txr->hn_tx_lock);
2493 atomic_clear_int(&txr->hn_sc->hn_ifp->if_drv_flags, IFF_DRV_OACTIVE);
2494 hn_start_locked(txr, 0);
2495 mtx_unlock(&txr->hn_tx_lock);
2499 hn_stop_tx_tasks(struct hn_softc *sc)
2503 for (i = 0; i < sc->hn_tx_ring_cnt; ++i) {
2504 struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
2506 taskqueue_drain(txr->hn_tx_taskq, &txr->hn_tx_task);
2507 taskqueue_drain(txr->hn_tx_taskq, &txr->hn_txeof_task);
2512 hn_tx_taskq_create(void *arg __unused)
2514 if (!hn_share_tx_taskq)
2517 hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
2518 taskqueue_thread_enqueue, &hn_tx_taskq);
2519 taskqueue_start_threads(&hn_tx_taskq, 1, PI_NET, "hn tx");
2520 if (hn_bind_tx_taskq >= 0) {
2521 int cpu = hn_bind_tx_taskq;
2522 struct task cpuset_task;
2525 if (cpu > mp_ncpus - 1)
2527 CPU_SETOF(cpu, &cpu_set);
2528 TASK_INIT(&cpuset_task, 0, hn_cpuset_setthread_task, &cpu_set);
2529 taskqueue_enqueue(hn_tx_taskq, &cpuset_task);
2530 taskqueue_drain(hn_tx_taskq, &cpuset_task);
2533 SYSINIT(hn_txtq_create, SI_SUB_DRIVERS, SI_ORDER_FIRST,
2534 hn_tx_taskq_create, NULL);
2537 hn_tx_taskq_destroy(void *arg __unused)
2539 if (hn_tx_taskq != NULL)
2540 taskqueue_free(hn_tx_taskq);
2542 SYSUNINIT(hn_txtq_destroy, SI_SUB_DRIVERS, SI_ORDER_FIRST,
2543 hn_tx_taskq_destroy, NULL);
2545 static device_method_t netvsc_methods[] = {
2546 /* Device interface */
2547 DEVMETHOD(device_probe, netvsc_probe),
2548 DEVMETHOD(device_attach, netvsc_attach),
2549 DEVMETHOD(device_detach, netvsc_detach),
2550 DEVMETHOD(device_shutdown, netvsc_shutdown),
2555 static driver_t netvsc_driver = {
2561 static devclass_t netvsc_devclass;
2563 DRIVER_MODULE(hn, vmbus, netvsc_driver, netvsc_devclass, 0, 0);
2564 MODULE_VERSION(hn, 1);
2565 MODULE_DEPEND(hn, vmbus, 1, 1, 1);