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"
122 #define hv_chan_rxr hv_chan_priv1
123 #define hv_chan_txr hv_chan_priv2
125 /* Short for Hyper-V network interface */
126 #define NETVSC_DEVNAME "hn"
129 * It looks like offset 0 of buf is reserved to hold the softc pointer.
130 * The sc pointer evidently not needed, and is not presently populated.
131 * The packet offset is where the netvsc_packet starts in the buffer.
133 #define HV_NV_SC_PTR_OFFSET_IN_BUF 0
134 #define HV_NV_PACKET_OFFSET_IN_BUF 16
136 /* YYY should get it from the underlying channel */
137 #define HN_TX_DESC_CNT 512
139 #define HN_LROENT_CNT_DEF 128
141 #define HN_RNDIS_MSG_LEN \
142 (sizeof(rndis_msg) + \
143 RNDIS_HASH_PPI_SIZE + \
144 RNDIS_VLAN_PPI_SIZE + \
145 RNDIS_TSO_PPI_SIZE + \
147 #define HN_RNDIS_MSG_BOUNDARY PAGE_SIZE
148 #define HN_RNDIS_MSG_ALIGN CACHE_LINE_SIZE
150 #define HN_TX_DATA_BOUNDARY PAGE_SIZE
151 #define HN_TX_DATA_MAXSIZE IP_MAXPACKET
152 #define HN_TX_DATA_SEGSIZE PAGE_SIZE
153 #define HN_TX_DATA_SEGCNT_MAX \
154 (NETVSC_PACKET_MAXPAGE - HV_RF_NUM_TX_RESERVED_PAGE_BUFS)
156 #define HN_DIRECT_TX_SIZE_DEF 128
159 #ifndef HN_USE_TXDESC_BUFRING
160 SLIST_ENTRY(hn_txdesc) link;
163 struct hn_tx_ring *txr;
165 uint32_t flags; /* HN_TXD_FLAG_ */
166 netvsc_packet netvsc_pkt; /* XXX to be removed */
168 bus_dmamap_t data_dmap;
170 bus_addr_t rndis_msg_paddr;
171 rndis_msg *rndis_msg;
172 bus_dmamap_t rndis_msg_dmap;
175 #define HN_TXD_FLAG_ONLIST 0x1
176 #define HN_TXD_FLAG_DMAMAP 0x2
179 * Only enable UDP checksum offloading when it is on 2012R2 or
180 * later. UDP checksum offloading doesn't work on earlier
183 #define HN_CSUM_ASSIST_WIN8 (CSUM_IP | CSUM_TCP)
184 #define HN_CSUM_ASSIST (CSUM_IP | CSUM_UDP | CSUM_TCP)
186 #define HN_LRO_LENLIM_DEF (25 * ETHERMTU)
187 /* YYY 2*MTU is a bit rough, but should be good enough. */
188 #define HN_LRO_LENLIM_MIN(ifp) (2 * (ifp)->if_mtu)
190 #define HN_LRO_ACKCNT_DEF 1
193 * Be aware that this sleepable mutex will exhibit WITNESS errors when
194 * certain TCP and ARP code paths are taken. This appears to be a
195 * well-known condition, as all other drivers checked use a sleeping
196 * mutex to protect their transmit paths.
197 * Also Be aware that mutexes do not play well with semaphores, and there
198 * is a conflicting semaphore in a certain channel code path.
200 #define NV_LOCK_INIT(_sc, _name) \
201 mtx_init(&(_sc)->hn_lock, _name, MTX_NETWORK_LOCK, MTX_DEF)
202 #define NV_LOCK(_sc) mtx_lock(&(_sc)->hn_lock)
203 #define NV_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->hn_lock, MA_OWNED)
204 #define NV_UNLOCK(_sc) mtx_unlock(&(_sc)->hn_lock)
205 #define NV_LOCK_DESTROY(_sc) mtx_destroy(&(_sc)->hn_lock)
212 int hv_promisc_mode = 0; /* normal mode by default */
214 SYSCTL_NODE(_hw, OID_AUTO, hn, CTLFLAG_RD, NULL, "Hyper-V network interface");
216 /* Trust tcp segements verification on host side. */
217 static int hn_trust_hosttcp = 1;
218 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hosttcp, CTLFLAG_RDTUN,
219 &hn_trust_hosttcp, 0,
220 "Trust tcp segement verification on host side, "
221 "when csum info is missing (global setting)");
223 /* Trust udp datagrams verification on host side. */
224 static int hn_trust_hostudp = 1;
225 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostudp, CTLFLAG_RDTUN,
226 &hn_trust_hostudp, 0,
227 "Trust udp datagram verification on host side, "
228 "when csum info is missing (global setting)");
230 /* Trust ip packets verification on host side. */
231 static int hn_trust_hostip = 1;
232 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostip, CTLFLAG_RDTUN,
234 "Trust ip packet verification on host side, "
235 "when csum info is missing (global setting)");
237 #if __FreeBSD_version >= 1100045
238 /* Limit TSO burst size */
239 static int hn_tso_maxlen = 0;
240 SYSCTL_INT(_hw_hn, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN,
241 &hn_tso_maxlen, 0, "TSO burst limit");
244 /* Limit chimney send size */
245 static int hn_tx_chimney_size = 0;
246 SYSCTL_INT(_hw_hn, OID_AUTO, tx_chimney_size, CTLFLAG_RDTUN,
247 &hn_tx_chimney_size, 0, "Chimney send packet size limit");
249 /* Limit the size of packet for direct transmission */
250 static int hn_direct_tx_size = HN_DIRECT_TX_SIZE_DEF;
251 SYSCTL_INT(_hw_hn, OID_AUTO, direct_tx_size, CTLFLAG_RDTUN,
252 &hn_direct_tx_size, 0, "Size of the packet for direct transmission");
254 #if defined(INET) || defined(INET6)
255 #if __FreeBSD_version >= 1100095
256 static int hn_lro_entry_count = HN_LROENT_CNT_DEF;
257 SYSCTL_INT(_hw_hn, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN,
258 &hn_lro_entry_count, 0, "LRO entry count");
262 static int hn_share_tx_taskq = 0;
263 SYSCTL_INT(_hw_hn, OID_AUTO, share_tx_taskq, CTLFLAG_RDTUN,
264 &hn_share_tx_taskq, 0, "Enable shared TX taskqueue");
266 static struct taskqueue *hn_tx_taskq;
268 #ifndef HN_USE_TXDESC_BUFRING
269 static int hn_use_txdesc_bufring = 0;
271 static int hn_use_txdesc_bufring = 1;
273 SYSCTL_INT(_hw_hn, OID_AUTO, use_txdesc_bufring, CTLFLAG_RD,
274 &hn_use_txdesc_bufring, 0, "Use buf_ring for TX descriptors");
276 static int hn_bind_tx_taskq = -1;
277 SYSCTL_INT(_hw_hn, OID_AUTO, bind_tx_taskq, CTLFLAG_RDTUN,
278 &hn_bind_tx_taskq, 0, "Bind TX taskqueue to the specified cpu");
280 static int hn_use_if_start = 0;
281 SYSCTL_INT(_hw_hn, OID_AUTO, use_if_start, CTLFLAG_RDTUN,
282 &hn_use_if_start, 0, "Use if_start TX method");
284 static int hn_ring_cnt = 1;
285 SYSCTL_INT(_hw_hn, OID_AUTO, ring_cnt, CTLFLAG_RDTUN,
286 &hn_ring_cnt, 0, "# of TX/RX rings to used");
288 static int hn_single_tx_ring = 1;
289 SYSCTL_INT(_hw_hn, OID_AUTO, single_tx_ring, CTLFLAG_RDTUN,
290 &hn_single_tx_ring, 0, "Use one TX ring");
293 * Forward declarations
295 static void hn_stop(hn_softc_t *sc);
296 static void hn_ifinit_locked(hn_softc_t *sc);
297 static void hn_ifinit(void *xsc);
298 static int hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data);
299 static int hn_start_locked(struct hn_tx_ring *txr, int len);
300 static void hn_start(struct ifnet *ifp);
301 static void hn_start_txeof(struct hn_tx_ring *);
302 static int hn_ifmedia_upd(struct ifnet *ifp);
303 static void hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr);
304 #if __FreeBSD_version >= 1100099
305 static int hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS);
306 static int hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS);
308 static int hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS);
309 static int hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS);
310 #if __FreeBSD_version < 1100095
311 static int hn_rx_stat_int_sysctl(SYSCTL_HANDLER_ARGS);
313 static int hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS);
315 static int hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS);
316 static int hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS);
317 static int hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS);
318 static int hn_check_iplen(const struct mbuf *, int);
319 static int hn_create_tx_ring(struct hn_softc *, int);
320 static void hn_destroy_tx_ring(struct hn_tx_ring *);
321 static int hn_create_tx_data(struct hn_softc *, int);
322 static void hn_destroy_tx_data(struct hn_softc *);
323 static void hn_start_taskfunc(void *, int);
324 static void hn_start_txeof_taskfunc(void *, int);
325 static void hn_stop_tx_tasks(struct hn_softc *);
326 static int hn_encap(struct hn_tx_ring *, struct hn_txdesc *, struct mbuf **);
327 static void hn_create_rx_data(struct hn_softc *sc, int);
328 static void hn_destroy_rx_data(struct hn_softc *sc);
329 static void hn_set_tx_chimney_size(struct hn_softc *, int);
331 static int hn_transmit(struct ifnet *, struct mbuf *);
332 static void hn_xmit_qflush(struct ifnet *);
333 static int hn_xmit(struct hn_tx_ring *, int);
334 static void hn_xmit_txeof(struct hn_tx_ring *);
335 static void hn_xmit_taskfunc(void *, int);
336 static void hn_xmit_txeof_taskfunc(void *, int);
339 hn_ifmedia_upd(struct ifnet *ifp __unused)
346 hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
348 struct hn_softc *sc = ifp->if_softc;
350 ifmr->ifm_status = IFM_AVALID;
351 ifmr->ifm_active = IFM_ETHER;
353 if (!sc->hn_carrier) {
354 ifmr->ifm_active |= IFM_NONE;
357 ifmr->ifm_status |= IFM_ACTIVE;
358 ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
361 /* {F8615163-DF3E-46c5-913F-F2D2F965ED0E} */
362 static const hv_guid g_net_vsc_device_type = {
363 .data = {0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
364 0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E}
368 * Standard probe entry point.
372 netvsc_probe(device_t dev)
376 p = vmbus_get_type(dev);
377 if (!memcmp(p, &g_net_vsc_device_type.data, sizeof(hv_guid))) {
378 device_set_desc(dev, "Synthetic Network Interface");
380 printf("Netvsc probe... DONE \n");
382 return (BUS_PROBE_DEFAULT);
389 hn_cpuset_setthread_task(void *xmask, int pending __unused)
391 cpuset_t *mask = xmask;
394 error = cpuset_setthread(curthread->td_tid, mask);
396 panic("curthread=%ju: can't pin; error=%d",
397 (uintmax_t)curthread->td_tid, error);
402 * Standard attach entry point.
404 * Called when the driver is loaded. It allocates needed resources,
405 * and initializes the "hardware" and software.
408 netvsc_attach(device_t dev)
410 struct hv_device *device_ctx = vmbus_get_devctx(dev);
411 struct hv_vmbus_channel *chan;
412 netvsc_device_info device_info;
414 int unit = device_get_unit(dev);
415 struct ifnet *ifp = NULL;
416 int error, ring_cnt, tx_ring_cnt;
417 #if __FreeBSD_version >= 1100045
421 sc = device_get_softc(dev);
426 bzero(sc, sizeof(hn_softc_t));
430 if (hn_tx_taskq == NULL) {
431 sc->hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
432 taskqueue_thread_enqueue, &sc->hn_tx_taskq);
433 taskqueue_start_threads(&sc->hn_tx_taskq, 1, PI_NET, "%s tx",
434 device_get_nameunit(dev));
435 if (hn_bind_tx_taskq >= 0) {
436 int cpu = hn_bind_tx_taskq;
437 struct task cpuset_task;
440 if (cpu > mp_ncpus - 1)
442 CPU_SETOF(cpu, &cpu_set);
443 TASK_INIT(&cpuset_task, 0, hn_cpuset_setthread_task,
445 taskqueue_enqueue(sc->hn_tx_taskq, &cpuset_task);
446 taskqueue_drain(sc->hn_tx_taskq, &cpuset_task);
449 sc->hn_tx_taskq = hn_tx_taskq;
451 NV_LOCK_INIT(sc, "NetVSCLock");
453 sc->hn_dev_obj = device_ctx;
455 ifp = sc->hn_ifp = sc->arpcom.ac_ifp = if_alloc(IFT_ETHER);
458 ring_cnt = hn_ring_cnt;
459 if (ring_cnt <= 0 || ring_cnt >= mp_ncpus)
462 tx_ring_cnt = ring_cnt;
463 if (hn_single_tx_ring || hn_use_if_start) {
465 * - Explicitly asked to use single TX ring.
466 * - ifnet.if_start is used; ifnet.if_start only needs
471 error = hn_create_tx_data(sc, tx_ring_cnt);
475 hn_create_rx_data(sc, ring_cnt);
478 * Associate the first TX/RX ring w/ the primary channel.
480 chan = device_ctx->channel;
481 chan->hv_chan_rxr = &sc->hn_rx_ring[0];
482 chan->hv_chan_txr = &sc->hn_tx_ring[0];
483 sc->hn_tx_ring[0].hn_chan = chan;
485 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
486 ifp->if_dunit = unit;
487 ifp->if_dname = NETVSC_DEVNAME;
489 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
490 ifp->if_ioctl = hn_ioctl;
491 ifp->if_init = hn_ifinit;
492 /* needed by hv_rf_on_device_add() code */
493 ifp->if_mtu = ETHERMTU;
494 if (hn_use_if_start) {
495 ifp->if_start = hn_start;
496 IFQ_SET_MAXLEN(&ifp->if_snd, 512);
497 ifp->if_snd.ifq_drv_maxlen = 511;
498 IFQ_SET_READY(&ifp->if_snd);
500 ifp->if_transmit = hn_transmit;
501 ifp->if_qflush = hn_xmit_qflush;
504 ifmedia_init(&sc->hn_media, 0, hn_ifmedia_upd, hn_ifmedia_sts);
505 ifmedia_add(&sc->hn_media, IFM_ETHER | IFM_AUTO, 0, NULL);
506 ifmedia_set(&sc->hn_media, IFM_ETHER | IFM_AUTO);
507 /* XXX ifmedia_set really should do this for us */
508 sc->hn_media.ifm_media = sc->hn_media.ifm_cur->ifm_media;
511 * Tell upper layers that we support full VLAN capability.
513 ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
514 ifp->if_capabilities |=
515 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
518 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
520 ifp->if_hwassist = sc->hn_tx_ring[0].hn_csum_assist | CSUM_TSO;
522 error = hv_rf_on_device_add(device_ctx, &device_info, ring_cnt);
527 sc->hn_tx_ring_inuse = 1;
528 sc->hn_rx_ring_inuse = 1;
529 device_printf(dev, "%d TX ring, %d RX ring\n",
530 sc->hn_tx_ring_inuse, sc->hn_rx_ring_inuse);
532 if (device_info.link_state == 0) {
536 #if __FreeBSD_version >= 1100045
537 tso_maxlen = hn_tso_maxlen;
538 if (tso_maxlen <= 0 || tso_maxlen > IP_MAXPACKET)
539 tso_maxlen = IP_MAXPACKET;
541 ifp->if_hw_tsomaxsegcount = HN_TX_DATA_SEGCNT_MAX;
542 ifp->if_hw_tsomaxsegsize = PAGE_SIZE;
543 ifp->if_hw_tsomax = tso_maxlen -
544 (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
547 ether_ifattach(ifp, device_info.mac_addr);
549 #if __FreeBSD_version >= 1100045
550 if_printf(ifp, "TSO: %u/%u/%u\n", ifp->if_hw_tsomax,
551 ifp->if_hw_tsomaxsegcount, ifp->if_hw_tsomaxsegsize);
554 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
555 hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max);
556 if (hn_tx_chimney_size > 0 &&
557 hn_tx_chimney_size < sc->hn_tx_chimney_max)
558 hn_set_tx_chimney_size(sc, hn_tx_chimney_size);
562 hn_destroy_tx_data(sc);
569 * Standard detach entry point
572 netvsc_detach(device_t dev)
574 struct hn_softc *sc = device_get_softc(dev);
575 struct hv_device *hv_device = vmbus_get_devctx(dev);
578 printf("netvsc_detach\n");
581 * XXXKYS: Need to clean up all our
582 * driver state; this is the driver
587 * XXXKYS: Need to stop outgoing traffic and unregister
591 hv_rf_on_device_remove(hv_device, HV_RF_NV_DESTROY_CHANNEL);
593 hn_stop_tx_tasks(sc);
595 ifmedia_removeall(&sc->hn_media);
596 hn_destroy_rx_data(sc);
597 hn_destroy_tx_data(sc);
599 if (sc->hn_tx_taskq != hn_tx_taskq)
600 taskqueue_free(sc->hn_tx_taskq);
606 * Standard shutdown entry point
609 netvsc_shutdown(device_t dev)
615 hn_txdesc_dmamap_load(struct hn_tx_ring *txr, struct hn_txdesc *txd,
616 struct mbuf **m_head, bus_dma_segment_t *segs, int *nsegs)
618 struct mbuf *m = *m_head;
621 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag, txd->data_dmap,
622 m, segs, nsegs, BUS_DMA_NOWAIT);
623 if (error == EFBIG) {
626 m_new = m_collapse(m, M_NOWAIT, HN_TX_DATA_SEGCNT_MAX);
631 txr->hn_tx_collapsed++;
633 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag,
634 txd->data_dmap, m, segs, nsegs, BUS_DMA_NOWAIT);
637 bus_dmamap_sync(txr->hn_tx_data_dtag, txd->data_dmap,
638 BUS_DMASYNC_PREWRITE);
639 txd->flags |= HN_TXD_FLAG_DMAMAP;
645 hn_txdesc_dmamap_unload(struct hn_tx_ring *txr, struct hn_txdesc *txd)
648 if (txd->flags & HN_TXD_FLAG_DMAMAP) {
649 bus_dmamap_sync(txr->hn_tx_data_dtag,
650 txd->data_dmap, BUS_DMASYNC_POSTWRITE);
651 bus_dmamap_unload(txr->hn_tx_data_dtag,
653 txd->flags &= ~HN_TXD_FLAG_DMAMAP;
658 hn_txdesc_put(struct hn_tx_ring *txr, struct hn_txdesc *txd)
661 KASSERT((txd->flags & HN_TXD_FLAG_ONLIST) == 0,
662 ("put an onlist txd %#x", txd->flags));
664 KASSERT(txd->refs > 0, ("invalid txd refs %d", txd->refs));
665 if (atomic_fetchadd_int(&txd->refs, -1) != 1)
668 hn_txdesc_dmamap_unload(txr, txd);
669 if (txd->m != NULL) {
674 txd->flags |= HN_TXD_FLAG_ONLIST;
676 #ifndef HN_USE_TXDESC_BUFRING
677 mtx_lock_spin(&txr->hn_txlist_spin);
678 KASSERT(txr->hn_txdesc_avail >= 0 &&
679 txr->hn_txdesc_avail < txr->hn_txdesc_cnt,
680 ("txdesc_put: invalid txd avail %d", txr->hn_txdesc_avail));
681 txr->hn_txdesc_avail++;
682 SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
683 mtx_unlock_spin(&txr->hn_txlist_spin);
685 atomic_add_int(&txr->hn_txdesc_avail, 1);
686 buf_ring_enqueue(txr->hn_txdesc_br, txd);
692 static __inline struct hn_txdesc *
693 hn_txdesc_get(struct hn_tx_ring *txr)
695 struct hn_txdesc *txd;
697 #ifndef HN_USE_TXDESC_BUFRING
698 mtx_lock_spin(&txr->hn_txlist_spin);
699 txd = SLIST_FIRST(&txr->hn_txlist);
701 KASSERT(txr->hn_txdesc_avail > 0,
702 ("txdesc_get: invalid txd avail %d", txr->hn_txdesc_avail));
703 txr->hn_txdesc_avail--;
704 SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
706 mtx_unlock_spin(&txr->hn_txlist_spin);
708 txd = buf_ring_dequeue_sc(txr->hn_txdesc_br);
712 #ifdef HN_USE_TXDESC_BUFRING
713 atomic_subtract_int(&txr->hn_txdesc_avail, 1);
715 KASSERT(txd->m == NULL && txd->refs == 0 &&
716 (txd->flags & HN_TXD_FLAG_ONLIST), ("invalid txd"));
717 txd->flags &= ~HN_TXD_FLAG_ONLIST;
724 hn_txdesc_hold(struct hn_txdesc *txd)
727 /* 0->1 transition will never work */
728 KASSERT(txd->refs > 0, ("invalid refs %d", txd->refs));
729 atomic_add_int(&txd->refs, 1);
733 hn_tx_done(void *xpkt)
735 netvsc_packet *packet = xpkt;
736 struct hn_txdesc *txd;
737 struct hn_tx_ring *txr;
739 txd = (struct hn_txdesc *)(uintptr_t)
740 packet->compl.send.send_completion_tid;
743 txr->hn_has_txeof = 1;
744 hn_txdesc_put(txr, txd);
748 netvsc_channel_rollup(struct hv_vmbus_channel *chan)
750 struct hn_tx_ring *txr = chan->hv_chan_txr;
751 #if defined(INET) || defined(INET6)
752 struct hn_rx_ring *rxr = chan->hv_chan_rxr;
753 struct lro_ctrl *lro = &rxr->hn_lro;
754 struct lro_entry *queued;
756 while ((queued = SLIST_FIRST(&lro->lro_active)) != NULL) {
757 SLIST_REMOVE_HEAD(&lro->lro_active, next);
758 tcp_lro_flush(lro, queued);
764 * 'txr' could be NULL, if multiple channels and
765 * ifnet.if_start method are enabled.
767 if (txr == NULL || !txr->hn_has_txeof)
770 txr->hn_has_txeof = 0;
776 * If this function fails, then both txd and m_head0 will be freed.
779 hn_encap(struct hn_tx_ring *txr, struct hn_txdesc *txd, struct mbuf **m_head0)
781 bus_dma_segment_t segs[HN_TX_DATA_SEGCNT_MAX];
783 struct mbuf *m_head = *m_head0;
784 netvsc_packet *packet;
785 rndis_msg *rndis_mesg;
786 rndis_packet *rndis_pkt;
787 rndis_per_packet_info *rppi;
788 struct ndis_hash_info *hash_info;
789 uint32_t rndis_msg_size;
791 packet = &txd->netvsc_pkt;
792 packet->is_data_pkt = TRUE;
793 packet->tot_data_buf_len = m_head->m_pkthdr.len;
796 * extension points to the area reserved for the
797 * rndis_filter_packet, which is placed just after
798 * the netvsc_packet (and rppi struct, if present;
799 * length is updated later).
801 rndis_mesg = txd->rndis_msg;
802 /* XXX not necessary */
803 memset(rndis_mesg, 0, HN_RNDIS_MSG_LEN);
804 rndis_mesg->ndis_msg_type = REMOTE_NDIS_PACKET_MSG;
806 rndis_pkt = &rndis_mesg->msg.packet;
807 rndis_pkt->data_offset = sizeof(rndis_packet);
808 rndis_pkt->data_length = packet->tot_data_buf_len;
809 rndis_pkt->per_pkt_info_offset = sizeof(rndis_packet);
811 rndis_msg_size = RNDIS_MESSAGE_SIZE(rndis_packet);
814 * Set the hash info for this packet, so that the host could
815 * dispatch the TX done event for this packet back to this TX
818 rndis_msg_size += RNDIS_HASH_PPI_SIZE;
819 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_HASH_PPI_SIZE,
821 hash_info = (struct ndis_hash_info *)((uint8_t *)rppi +
822 rppi->per_packet_info_offset);
823 hash_info->hash = txr->hn_tx_idx;
825 if (m_head->m_flags & M_VLANTAG) {
826 ndis_8021q_info *rppi_vlan_info;
828 rndis_msg_size += RNDIS_VLAN_PPI_SIZE;
829 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_VLAN_PPI_SIZE,
832 rppi_vlan_info = (ndis_8021q_info *)((uint8_t *)rppi +
833 rppi->per_packet_info_offset);
834 rppi_vlan_info->u1.s1.vlan_id =
835 m_head->m_pkthdr.ether_vtag & 0xfff;
838 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
839 rndis_tcp_tso_info *tso_info;
840 struct ether_vlan_header *eh;
844 * XXX need m_pullup and use mtodo
846 eh = mtod(m_head, struct ether_vlan_header*);
847 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN))
848 ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
850 ether_len = ETHER_HDR_LEN;
852 rndis_msg_size += RNDIS_TSO_PPI_SIZE;
853 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_TSO_PPI_SIZE,
854 tcp_large_send_info);
856 tso_info = (rndis_tcp_tso_info *)((uint8_t *)rppi +
857 rppi->per_packet_info_offset);
858 tso_info->lso_v2_xmit.type =
859 RNDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
862 if (m_head->m_pkthdr.csum_flags & CSUM_IP_TSO) {
864 (struct ip *)(m_head->m_data + ether_len);
865 unsigned long iph_len = ip->ip_hl << 2;
867 (struct tcphdr *)((caddr_t)ip + iph_len);
869 tso_info->lso_v2_xmit.ip_version =
870 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
874 th->th_sum = in_pseudo(ip->ip_src.s_addr,
875 ip->ip_dst.s_addr, htons(IPPROTO_TCP));
878 #if defined(INET6) && defined(INET)
883 struct ip6_hdr *ip6 = (struct ip6_hdr *)
884 (m_head->m_data + ether_len);
885 struct tcphdr *th = (struct tcphdr *)(ip6 + 1);
887 tso_info->lso_v2_xmit.ip_version =
888 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
890 th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
893 tso_info->lso_v2_xmit.tcp_header_offset = 0;
894 tso_info->lso_v2_xmit.mss = m_head->m_pkthdr.tso_segsz;
895 } else if (m_head->m_pkthdr.csum_flags & txr->hn_csum_assist) {
896 rndis_tcp_ip_csum_info *csum_info;
898 rndis_msg_size += RNDIS_CSUM_PPI_SIZE;
899 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_CSUM_PPI_SIZE,
901 csum_info = (rndis_tcp_ip_csum_info *)((uint8_t *)rppi +
902 rppi->per_packet_info_offset);
904 csum_info->xmit.is_ipv4 = 1;
905 if (m_head->m_pkthdr.csum_flags & CSUM_IP)
906 csum_info->xmit.ip_header_csum = 1;
908 if (m_head->m_pkthdr.csum_flags & CSUM_TCP) {
909 csum_info->xmit.tcp_csum = 1;
910 csum_info->xmit.tcp_header_offset = 0;
911 } else if (m_head->m_pkthdr.csum_flags & CSUM_UDP) {
912 csum_info->xmit.udp_csum = 1;
916 rndis_mesg->msg_len = packet->tot_data_buf_len + rndis_msg_size;
917 packet->tot_data_buf_len = rndis_mesg->msg_len;
920 * Chimney send, if the packet could fit into one chimney buffer.
922 * TODO: vRSS, chimney buffer should be per-channel.
924 if (packet->tot_data_buf_len < txr->hn_tx_chimney_size) {
925 netvsc_dev *net_dev = txr->hn_sc->net_dev;
926 uint32_t send_buf_section_idx;
928 send_buf_section_idx =
929 hv_nv_get_next_send_section(net_dev);
930 if (send_buf_section_idx !=
931 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX) {
932 uint8_t *dest = ((uint8_t *)net_dev->send_buf +
933 (send_buf_section_idx *
934 net_dev->send_section_size));
936 memcpy(dest, rndis_mesg, rndis_msg_size);
937 dest += rndis_msg_size;
938 m_copydata(m_head, 0, m_head->m_pkthdr.len, dest);
940 packet->send_buf_section_idx = send_buf_section_idx;
941 packet->send_buf_section_size =
942 packet->tot_data_buf_len;
943 packet->page_buf_count = 0;
944 txr->hn_tx_chimney++;
949 error = hn_txdesc_dmamap_load(txr, txd, &m_head, segs, &nsegs);
954 * This mbuf is not linked w/ the txd yet, so free it now.
959 freed = hn_txdesc_put(txr, txd);
961 ("fail to free txd upon txdma error"));
963 txr->hn_txdma_failed++;
964 if_inc_counter(txr->hn_sc->hn_ifp, IFCOUNTER_OERRORS, 1);
969 packet->page_buf_count = nsegs + HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
971 /* send packet with page buffer */
972 packet->page_buffers[0].pfn = atop(txd->rndis_msg_paddr);
973 packet->page_buffers[0].offset = txd->rndis_msg_paddr & PAGE_MASK;
974 packet->page_buffers[0].length = rndis_msg_size;
977 * Fill the page buffers with mbuf info starting at index
978 * HV_RF_NUM_TX_RESERVED_PAGE_BUFS.
980 for (i = 0; i < nsegs; ++i) {
981 hv_vmbus_page_buffer *pb = &packet->page_buffers[
982 i + HV_RF_NUM_TX_RESERVED_PAGE_BUFS];
984 pb->pfn = atop(segs[i].ds_addr);
985 pb->offset = segs[i].ds_addr & PAGE_MASK;
986 pb->length = segs[i].ds_len;
989 packet->send_buf_section_idx =
990 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX;
991 packet->send_buf_section_size = 0;
995 /* Set the completion routine */
996 packet->compl.send.on_send_completion = hn_tx_done;
997 packet->compl.send.send_completion_context = packet;
998 packet->compl.send.send_completion_tid = (uint64_t)(uintptr_t)txd;
1005 * If this function fails, then txd will be freed, but the mbuf
1006 * associated w/ the txd will _not_ be freed.
1009 hn_send_pkt(struct ifnet *ifp, struct hn_tx_ring *txr, struct hn_txdesc *txd)
1011 int error, send_failed = 0;
1015 * Make sure that txd is not freed before ETHER_BPF_MTAP.
1017 hn_txdesc_hold(txd);
1018 error = hv_nv_on_send(txr->hn_chan, &txd->netvsc_pkt);
1020 ETHER_BPF_MTAP(ifp, txd->m);
1021 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1022 if (!hn_use_if_start) {
1023 if_inc_counter(ifp, IFCOUNTER_OBYTES,
1024 txd->m->m_pkthdr.len);
1025 if (txd->m->m_flags & M_MCAST)
1026 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
1029 hn_txdesc_put(txr, txd);
1031 if (__predict_false(error)) {
1035 * This should "really rarely" happen.
1037 * XXX Too many RX to be acked or too many sideband
1038 * commands to run? Ask netvsc_channel_rollup()
1039 * to kick start later.
1041 txr->hn_has_txeof = 1;
1043 txr->hn_send_failed++;
1046 * Try sending again after set hn_has_txeof;
1047 * in case that we missed the last
1048 * netvsc_channel_rollup().
1052 if_printf(ifp, "send failed\n");
1055 * Caller will perform further processing on the
1056 * associated mbuf, so don't free it in hn_txdesc_put();
1057 * only unload it from the DMA map in hn_txdesc_put(),
1061 freed = hn_txdesc_put(txr, txd);
1063 ("fail to free txd upon send error"));
1065 txr->hn_send_failed++;
1071 * Start a transmit of one or more packets
1074 hn_start_locked(struct hn_tx_ring *txr, int len)
1076 struct hn_softc *sc = txr->hn_sc;
1077 struct ifnet *ifp = sc->hn_ifp;
1079 KASSERT(hn_use_if_start,
1080 ("hn_start_locked is called, when if_start is disabled"));
1081 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1082 mtx_assert(&txr->hn_tx_lock, MA_OWNED);
1084 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1088 while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
1089 struct hn_txdesc *txd;
1090 struct mbuf *m_head;
1093 IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
1097 if (len > 0 && m_head->m_pkthdr.len > len) {
1099 * This sending could be time consuming; let callers
1100 * dispatch this packet sending (and sending of any
1101 * following up packets) to tx taskqueue.
1103 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1107 txd = hn_txdesc_get(txr);
1109 txr->hn_no_txdescs++;
1110 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1111 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1115 error = hn_encap(txr, txd, &m_head);
1117 /* Both txd and m_head are freed */
1121 error = hn_send_pkt(ifp, txr, txd);
1122 if (__predict_false(error)) {
1123 /* txd is freed, but m_head is not */
1124 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1125 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1133 * Link up/down notification
1136 netvsc_linkstatus_callback(struct hv_device *device_obj, uint32_t status)
1138 hn_softc_t *sc = device_get_softc(device_obj->device);
1152 * Append the specified data to the indicated mbuf chain,
1153 * Extend the mbuf chain if the new data does not fit in
1156 * This is a minor rewrite of m_append() from sys/kern/uipc_mbuf.c.
1157 * There should be an equivalent in the kernel mbuf code,
1158 * but there does not appear to be one yet.
1160 * Differs from m_append() in that additional mbufs are
1161 * allocated with cluster size MJUMPAGESIZE, and filled
1164 * Return 1 if able to complete the job; otherwise 0.
1167 hv_m_append(struct mbuf *m0, int len, c_caddr_t cp)
1170 int remainder, space;
1172 for (m = m0; m->m_next != NULL; m = m->m_next)
1175 space = M_TRAILINGSPACE(m);
1178 * Copy into available space.
1180 if (space > remainder)
1182 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1187 while (remainder > 0) {
1189 * Allocate a new mbuf; could check space
1190 * and allocate a cluster instead.
1192 n = m_getjcl(M_DONTWAIT, m->m_type, 0, MJUMPAGESIZE);
1195 n->m_len = min(MJUMPAGESIZE, remainder);
1196 bcopy(cp, mtod(n, caddr_t), n->m_len);
1198 remainder -= n->m_len;
1202 if (m0->m_flags & M_PKTHDR)
1203 m0->m_pkthdr.len += len - remainder;
1205 return (remainder == 0);
1210 * Called when we receive a data packet from the "wire" on the
1213 * Note: This is no longer used as a callback
1216 netvsc_recv(struct hv_vmbus_channel *chan, netvsc_packet *packet,
1217 rndis_tcp_ip_csum_info *csum_info)
1219 struct hn_rx_ring *rxr = chan->hv_chan_rxr;
1220 struct ifnet *ifp = rxr->hn_ifp;
1222 int size, do_lro = 0, do_csum = 1;
1224 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1228 * Bail out if packet contains more data than configured MTU.
1230 if (packet->tot_data_buf_len > (ifp->if_mtu + ETHER_HDR_LEN)) {
1232 } else if (packet->tot_data_buf_len <= MHLEN) {
1233 m_new = m_gethdr(M_NOWAIT, MT_DATA);
1236 memcpy(mtod(m_new, void *), packet->data,
1237 packet->tot_data_buf_len);
1238 m_new->m_pkthdr.len = m_new->m_len = packet->tot_data_buf_len;
1239 rxr->hn_small_pkts++;
1242 * Get an mbuf with a cluster. For packets 2K or less,
1243 * get a standard 2K cluster. For anything larger, get a
1244 * 4K cluster. Any buffers larger than 4K can cause problems
1245 * if looped around to the Hyper-V TX channel, so avoid them.
1248 if (packet->tot_data_buf_len > MCLBYTES) {
1250 size = MJUMPAGESIZE;
1253 m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, size);
1254 if (m_new == NULL) {
1255 if_printf(ifp, "alloc mbuf failed.\n");
1259 hv_m_append(m_new, packet->tot_data_buf_len, packet->data);
1261 m_new->m_pkthdr.rcvif = ifp;
1263 if (__predict_false((ifp->if_capenable & IFCAP_RXCSUM) == 0))
1266 /* receive side checksum offload */
1267 if (csum_info != NULL) {
1268 /* IP csum offload */
1269 if (csum_info->receive.ip_csum_succeeded && do_csum) {
1270 m_new->m_pkthdr.csum_flags |=
1271 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1275 /* TCP/UDP csum offload */
1276 if ((csum_info->receive.tcp_csum_succeeded ||
1277 csum_info->receive.udp_csum_succeeded) && do_csum) {
1278 m_new->m_pkthdr.csum_flags |=
1279 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1280 m_new->m_pkthdr.csum_data = 0xffff;
1281 if (csum_info->receive.tcp_csum_succeeded)
1287 if (csum_info->receive.ip_csum_succeeded &&
1288 csum_info->receive.tcp_csum_succeeded)
1291 const struct ether_header *eh;
1296 if (m_new->m_len < hoff)
1298 eh = mtod(m_new, struct ether_header *);
1299 etype = ntohs(eh->ether_type);
1300 if (etype == ETHERTYPE_VLAN) {
1301 const struct ether_vlan_header *evl;
1303 hoff = sizeof(*evl);
1304 if (m_new->m_len < hoff)
1306 evl = mtod(m_new, struct ether_vlan_header *);
1307 etype = ntohs(evl->evl_proto);
1310 if (etype == ETHERTYPE_IP) {
1313 pr = hn_check_iplen(m_new, hoff);
1314 if (pr == IPPROTO_TCP) {
1316 (rxr->hn_trust_hcsum &
1317 HN_TRUST_HCSUM_TCP)) {
1318 rxr->hn_csum_trusted++;
1319 m_new->m_pkthdr.csum_flags |=
1320 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1321 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1322 m_new->m_pkthdr.csum_data = 0xffff;
1324 /* Rely on SW csum verification though... */
1326 } else if (pr == IPPROTO_UDP) {
1328 (rxr->hn_trust_hcsum &
1329 HN_TRUST_HCSUM_UDP)) {
1330 rxr->hn_csum_trusted++;
1331 m_new->m_pkthdr.csum_flags |=
1332 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1333 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1334 m_new->m_pkthdr.csum_data = 0xffff;
1336 } else if (pr != IPPROTO_DONE && do_csum &&
1337 (rxr->hn_trust_hcsum & HN_TRUST_HCSUM_IP)) {
1338 rxr->hn_csum_trusted++;
1339 m_new->m_pkthdr.csum_flags |=
1340 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1345 if ((packet->vlan_tci != 0) &&
1346 (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0) {
1347 m_new->m_pkthdr.ether_vtag = packet->vlan_tci;
1348 m_new->m_flags |= M_VLANTAG;
1351 m_new->m_pkthdr.flowid = rxr->hn_rx_idx;
1352 M_HASHTYPE_SET(m_new, M_HASHTYPE_OPAQUE);
1355 * Note: Moved RX completion back to hv_nv_on_receive() so all
1356 * messages (not just data messages) will trigger a response.
1361 if ((ifp->if_capenable & IFCAP_LRO) && do_lro) {
1362 #if defined(INET) || defined(INET6)
1363 struct lro_ctrl *lro = &rxr->hn_lro;
1366 rxr->hn_lro_tried++;
1367 if (tcp_lro_rx(lro, m_new, 0) == 0) {
1375 /* We're not holding the lock here, so don't release it */
1376 (*ifp->if_input)(ifp, m_new);
1382 * Rules for using sc->temp_unusable:
1383 * 1. sc->temp_unusable can only be read or written while holding NV_LOCK()
1384 * 2. code reading sc->temp_unusable under NV_LOCK(), and finding
1385 * sc->temp_unusable set, must release NV_LOCK() and exit
1386 * 3. to retain exclusive control of the interface,
1387 * sc->temp_unusable must be set by code before releasing NV_LOCK()
1388 * 4. only code setting sc->temp_unusable can clear sc->temp_unusable
1389 * 5. code setting sc->temp_unusable must eventually clear sc->temp_unusable
1393 * Standard ioctl entry point. Called when the user wants to configure
1397 hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1399 hn_softc_t *sc = ifp->if_softc;
1400 struct ifreq *ifr = (struct ifreq *)data;
1402 struct ifaddr *ifa = (struct ifaddr *)data;
1404 netvsc_device_info device_info;
1405 struct hv_device *hn_dev;
1406 int mask, error = 0;
1407 int retry_cnt = 500;
1413 if (ifa->ifa_addr->sa_family == AF_INET) {
1414 ifp->if_flags |= IFF_UP;
1415 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1417 arp_ifinit(ifp, ifa);
1420 error = ether_ioctl(ifp, cmd, data);
1423 hn_dev = vmbus_get_devctx(sc->hn_dev);
1425 /* Check MTU value change */
1426 if (ifp->if_mtu == ifr->ifr_mtu)
1429 if (ifr->ifr_mtu > NETVSC_MAX_CONFIGURABLE_MTU) {
1434 /* Obtain and record requested MTU */
1435 ifp->if_mtu = ifr->ifr_mtu;
1437 #if __FreeBSD_version >= 1100099
1439 * Make sure that LRO aggregation length limit is still
1440 * valid, after the MTU change.
1443 if (sc->hn_rx_ring[0].hn_lro.lro_length_lim <
1444 HN_LRO_LENLIM_MIN(ifp)) {
1446 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1447 sc->hn_rx_ring[i].hn_lro.lro_length_lim =
1448 HN_LRO_LENLIM_MIN(ifp);
1456 if (!sc->temp_unusable) {
1457 sc->temp_unusable = TRUE;
1461 if (retry_cnt > 0) {
1465 } while (retry_cnt > 0);
1467 if (retry_cnt == 0) {
1472 /* We must remove and add back the device to cause the new
1473 * MTU to take effect. This includes tearing down, but not
1474 * deleting the channel, then bringing it back up.
1476 error = hv_rf_on_device_remove(hn_dev, HV_RF_NV_RETAIN_CHANNEL);
1479 sc->temp_unusable = FALSE;
1483 error = hv_rf_on_device_add(hn_dev, &device_info,
1484 sc->hn_rx_ring_inuse);
1487 sc->temp_unusable = FALSE;
1492 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
1493 if (sc->hn_tx_ring[0].hn_tx_chimney_size >
1494 sc->hn_tx_chimney_max)
1495 hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max);
1497 hn_ifinit_locked(sc);
1500 sc->temp_unusable = FALSE;
1506 if (!sc->temp_unusable) {
1507 sc->temp_unusable = TRUE;
1511 if (retry_cnt > 0) {
1515 } while (retry_cnt > 0);
1517 if (retry_cnt == 0) {
1522 if (ifp->if_flags & IFF_UP) {
1524 * If only the state of the PROMISC flag changed,
1525 * then just use the 'set promisc mode' command
1526 * instead of reinitializing the entire NIC. Doing
1527 * a full re-init means reloading the firmware and
1528 * waiting for it to start up, which may take a
1532 /* Fixme: Promiscuous mode? */
1533 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1534 ifp->if_flags & IFF_PROMISC &&
1535 !(sc->hn_if_flags & IFF_PROMISC)) {
1536 /* do something here for Hyper-V */
1537 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1538 !(ifp->if_flags & IFF_PROMISC) &&
1539 sc->hn_if_flags & IFF_PROMISC) {
1540 /* do something here for Hyper-V */
1543 hn_ifinit_locked(sc);
1545 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1550 sc->temp_unusable = FALSE;
1552 sc->hn_if_flags = ifp->if_flags;
1558 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1559 if (mask & IFCAP_TXCSUM) {
1560 ifp->if_capenable ^= IFCAP_TXCSUM;
1561 if (ifp->if_capenable & IFCAP_TXCSUM) {
1563 sc->hn_tx_ring[0].hn_csum_assist;
1566 ~sc->hn_tx_ring[0].hn_csum_assist;
1570 if (mask & IFCAP_RXCSUM)
1571 ifp->if_capenable ^= IFCAP_RXCSUM;
1573 if (mask & IFCAP_LRO)
1574 ifp->if_capenable ^= IFCAP_LRO;
1576 if (mask & IFCAP_TSO4) {
1577 ifp->if_capenable ^= IFCAP_TSO4;
1578 if (ifp->if_capenable & IFCAP_TSO4)
1579 ifp->if_hwassist |= CSUM_IP_TSO;
1581 ifp->if_hwassist &= ~CSUM_IP_TSO;
1584 if (mask & IFCAP_TSO6) {
1585 ifp->if_capenable ^= IFCAP_TSO6;
1586 if (ifp->if_capenable & IFCAP_TSO6)
1587 ifp->if_hwassist |= CSUM_IP6_TSO;
1589 ifp->if_hwassist &= ~CSUM_IP6_TSO;
1598 /* Fixme: Multicast mode? */
1599 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1601 netvsc_setmulti(sc);
1610 error = ifmedia_ioctl(ifp, ifr, &sc->hn_media, cmd);
1613 error = ether_ioctl(ifp, cmd, data);
1624 hn_stop(hn_softc_t *sc)
1628 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1633 printf(" Closing Device ...\n");
1635 atomic_clear_int(&ifp->if_drv_flags,
1636 (IFF_DRV_RUNNING | IFF_DRV_OACTIVE));
1637 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
1638 sc->hn_tx_ring[i].hn_oactive = 0;
1640 if_link_state_change(ifp, LINK_STATE_DOWN);
1641 sc->hn_initdone = 0;
1643 ret = hv_rf_on_close(device_ctx);
1647 * FreeBSD transmit entry point
1650 hn_start(struct ifnet *ifp)
1652 struct hn_softc *sc = ifp->if_softc;
1653 struct hn_tx_ring *txr = &sc->hn_tx_ring[0];
1655 if (txr->hn_sched_tx)
1658 if (mtx_trylock(&txr->hn_tx_lock)) {
1661 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1662 mtx_unlock(&txr->hn_tx_lock);
1667 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
1671 hn_start_txeof(struct hn_tx_ring *txr)
1673 struct hn_softc *sc = txr->hn_sc;
1674 struct ifnet *ifp = sc->hn_ifp;
1676 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1678 if (txr->hn_sched_tx)
1681 if (mtx_trylock(&txr->hn_tx_lock)) {
1684 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1685 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1686 mtx_unlock(&txr->hn_tx_lock);
1688 taskqueue_enqueue(txr->hn_tx_taskq,
1694 * Release the OACTIVE earlier, with the hope, that
1695 * others could catch up. The task will clear the
1696 * flag again with the hn_tx_lock to avoid possible
1699 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1700 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
1708 hn_ifinit_locked(hn_softc_t *sc)
1711 struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1716 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1720 hv_promisc_mode = 1;
1722 ret = hv_rf_on_open(device_ctx);
1726 sc->hn_initdone = 1;
1729 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1730 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
1731 sc->hn_tx_ring[i].hn_oactive = 0;
1733 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_RUNNING);
1734 if_link_state_change(ifp, LINK_STATE_UP);
1741 hn_ifinit(void *xsc)
1743 hn_softc_t *sc = xsc;
1746 if (sc->temp_unusable) {
1750 sc->temp_unusable = TRUE;
1753 hn_ifinit_locked(sc);
1756 sc->temp_unusable = FALSE;
1765 hn_watchdog(struct ifnet *ifp)
1770 printf("hn%d: watchdog timeout -- resetting\n", sc->hn_unit);
1771 hn_ifinit(sc); /*???*/
1776 #if __FreeBSD_version >= 1100099
1779 hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS)
1781 struct hn_softc *sc = arg1;
1782 unsigned int lenlim;
1785 lenlim = sc->hn_rx_ring[0].hn_lro.lro_length_lim;
1786 error = sysctl_handle_int(oidp, &lenlim, 0, req);
1787 if (error || req->newptr == NULL)
1790 if (lenlim < HN_LRO_LENLIM_MIN(sc->hn_ifp) ||
1791 lenlim > TCP_LRO_LENGTH_MAX)
1795 for (i = 0; i < sc->hn_rx_ring_inuse; ++i)
1796 sc->hn_rx_ring[i].hn_lro.lro_length_lim = lenlim;
1802 hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS)
1804 struct hn_softc *sc = arg1;
1805 int ackcnt, error, i;
1808 * lro_ackcnt_lim is append count limit,
1809 * +1 to turn it into aggregation limit.
1811 ackcnt = sc->hn_rx_ring[0].hn_lro.lro_ackcnt_lim + 1;
1812 error = sysctl_handle_int(oidp, &ackcnt, 0, req);
1813 if (error || req->newptr == NULL)
1816 if (ackcnt < 2 || ackcnt > (TCP_LRO_ACKCNT_MAX + 1))
1820 * Convert aggregation limit back to append
1825 for (i = 0; i < sc->hn_rx_ring_inuse; ++i)
1826 sc->hn_rx_ring[i].hn_lro.lro_ackcnt_lim = ackcnt;
1834 hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS)
1836 struct hn_softc *sc = arg1;
1841 if (sc->hn_rx_ring[0].hn_trust_hcsum & hcsum)
1844 error = sysctl_handle_int(oidp, &on, 0, req);
1845 if (error || req->newptr == NULL)
1849 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1850 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
1853 rxr->hn_trust_hcsum |= hcsum;
1855 rxr->hn_trust_hcsum &= ~hcsum;
1862 hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS)
1864 struct hn_softc *sc = arg1;
1865 int chimney_size, error;
1867 chimney_size = sc->hn_tx_ring[0].hn_tx_chimney_size;
1868 error = sysctl_handle_int(oidp, &chimney_size, 0, req);
1869 if (error || req->newptr == NULL)
1872 if (chimney_size > sc->hn_tx_chimney_max || chimney_size <= 0)
1875 hn_set_tx_chimney_size(sc, chimney_size);
1879 #if __FreeBSD_version < 1100095
1881 hn_rx_stat_int_sysctl(SYSCTL_HANDLER_ARGS)
1883 struct hn_softc *sc = arg1;
1884 int ofs = arg2, i, error;
1885 struct hn_rx_ring *rxr;
1889 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1890 rxr = &sc->hn_rx_ring[i];
1891 stat += *((int *)((uint8_t *)rxr + ofs));
1894 error = sysctl_handle_64(oidp, &stat, 0, req);
1895 if (error || req->newptr == NULL)
1898 /* Zero out this stat. */
1899 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1900 rxr = &sc->hn_rx_ring[i];
1901 *((int *)((uint8_t *)rxr + ofs)) = 0;
1907 hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS)
1909 struct hn_softc *sc = arg1;
1910 int ofs = arg2, i, error;
1911 struct hn_rx_ring *rxr;
1915 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1916 rxr = &sc->hn_rx_ring[i];
1917 stat += *((uint64_t *)((uint8_t *)rxr + ofs));
1920 error = sysctl_handle_64(oidp, &stat, 0, req);
1921 if (error || req->newptr == NULL)
1924 /* Zero out this stat. */
1925 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1926 rxr = &sc->hn_rx_ring[i];
1927 *((uint64_t *)((uint8_t *)rxr + ofs)) = 0;
1935 hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
1937 struct hn_softc *sc = arg1;
1938 int ofs = arg2, i, error;
1939 struct hn_rx_ring *rxr;
1943 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
1944 rxr = &sc->hn_rx_ring[i];
1945 stat += *((u_long *)((uint8_t *)rxr + ofs));
1948 error = sysctl_handle_long(oidp, &stat, 0, req);
1949 if (error || req->newptr == NULL)
1952 /* Zero out this stat. */
1953 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
1954 rxr = &sc->hn_rx_ring[i];
1955 *((u_long *)((uint8_t *)rxr + ofs)) = 0;
1961 hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
1963 struct hn_softc *sc = arg1;
1964 int ofs = arg2, i, error;
1965 struct hn_tx_ring *txr;
1969 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
1970 txr = &sc->hn_tx_ring[i];
1971 stat += *((u_long *)((uint8_t *)txr + ofs));
1974 error = sysctl_handle_long(oidp, &stat, 0, req);
1975 if (error || req->newptr == NULL)
1978 /* Zero out this stat. */
1979 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
1980 txr = &sc->hn_tx_ring[i];
1981 *((u_long *)((uint8_t *)txr + ofs)) = 0;
1987 hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS)
1989 struct hn_softc *sc = arg1;
1990 int ofs = arg2, i, error, conf;
1991 struct hn_tx_ring *txr;
1993 txr = &sc->hn_tx_ring[0];
1994 conf = *((int *)((uint8_t *)txr + ofs));
1996 error = sysctl_handle_int(oidp, &conf, 0, req);
1997 if (error || req->newptr == NULL)
2001 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2002 txr = &sc->hn_tx_ring[i];
2003 *((int *)((uint8_t *)txr + ofs)) = conf;
2011 hn_check_iplen(const struct mbuf *m, int hoff)
2013 const struct ip *ip;
2014 int len, iphlen, iplen;
2015 const struct tcphdr *th;
2016 int thoff; /* TCP data offset */
2018 len = hoff + sizeof(struct ip);
2020 /* The packet must be at least the size of an IP header. */
2021 if (m->m_pkthdr.len < len)
2022 return IPPROTO_DONE;
2024 /* The fixed IP header must reside completely in the first mbuf. */
2026 return IPPROTO_DONE;
2028 ip = mtodo(m, hoff);
2030 /* Bound check the packet's stated IP header length. */
2031 iphlen = ip->ip_hl << 2;
2032 if (iphlen < sizeof(struct ip)) /* minimum header length */
2033 return IPPROTO_DONE;
2035 /* The full IP header must reside completely in the one mbuf. */
2036 if (m->m_len < hoff + iphlen)
2037 return IPPROTO_DONE;
2039 iplen = ntohs(ip->ip_len);
2042 * Check that the amount of data in the buffers is as
2043 * at least much as the IP header would have us expect.
2045 if (m->m_pkthdr.len < hoff + iplen)
2046 return IPPROTO_DONE;
2049 * Ignore IP fragments.
2051 if (ntohs(ip->ip_off) & (IP_OFFMASK | IP_MF))
2052 return IPPROTO_DONE;
2055 * The TCP/IP or UDP/IP header must be entirely contained within
2056 * the first fragment of a packet.
2060 if (iplen < iphlen + sizeof(struct tcphdr))
2061 return IPPROTO_DONE;
2062 if (m->m_len < hoff + iphlen + sizeof(struct tcphdr))
2063 return IPPROTO_DONE;
2064 th = (const struct tcphdr *)((const uint8_t *)ip + iphlen);
2065 thoff = th->th_off << 2;
2066 if (thoff < sizeof(struct tcphdr) || thoff + iphlen > iplen)
2067 return IPPROTO_DONE;
2068 if (m->m_len < hoff + iphlen + thoff)
2069 return IPPROTO_DONE;
2072 if (iplen < iphlen + sizeof(struct udphdr))
2073 return IPPROTO_DONE;
2074 if (m->m_len < hoff + iphlen + sizeof(struct udphdr))
2075 return IPPROTO_DONE;
2079 return IPPROTO_DONE;
2086 hn_dma_map_paddr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
2088 bus_addr_t *paddr = arg;
2093 KASSERT(nseg == 1, ("too many segments %d!", nseg));
2094 *paddr = segs->ds_addr;
2098 hn_create_rx_data(struct hn_softc *sc, int ring_cnt)
2100 struct sysctl_oid_list *child;
2101 struct sysctl_ctx_list *ctx;
2102 device_t dev = sc->hn_dev;
2103 #if defined(INET) || defined(INET6)
2104 #if __FreeBSD_version >= 1100095
2110 sc->hn_rx_ring_cnt = ring_cnt;
2111 sc->hn_rx_ring_inuse = sc->hn_rx_ring_cnt;
2113 sc->hn_rx_ring = malloc(sizeof(struct hn_rx_ring) * sc->hn_rx_ring_cnt,
2114 M_NETVSC, M_WAITOK | M_ZERO);
2116 #if defined(INET) || defined(INET6)
2117 #if __FreeBSD_version >= 1100095
2118 lroent_cnt = hn_lro_entry_count;
2119 if (lroent_cnt < TCP_LRO_ENTRIES)
2120 lroent_cnt = TCP_LRO_ENTRIES;
2121 device_printf(dev, "LRO: entry count %d\n", lroent_cnt);
2123 #endif /* INET || INET6 */
2125 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2126 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
2128 if (hn_trust_hosttcp)
2129 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_TCP;
2130 if (hn_trust_hostudp)
2131 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_UDP;
2132 if (hn_trust_hostip)
2133 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_IP;
2134 rxr->hn_ifp = sc->hn_ifp;
2140 #if defined(INET) || defined(INET6)
2141 #if __FreeBSD_version >= 1100095
2142 tcp_lro_init_args(&rxr->hn_lro, sc->hn_ifp, lroent_cnt, 0);
2144 tcp_lro_init(&rxr->hn_lro);
2145 rxr->hn_lro.ifp = sc->hn_ifp;
2147 #if __FreeBSD_version >= 1100099
2148 rxr->hn_lro.lro_length_lim = HN_LRO_LENLIM_DEF;
2149 rxr->hn_lro.lro_ackcnt_lim = HN_LRO_ACKCNT_DEF;
2151 #endif /* INET || INET6 */
2154 ctx = device_get_sysctl_ctx(dev);
2155 child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
2157 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_queued",
2158 CTLTYPE_U64 | CTLFLAG_RW, sc,
2159 __offsetof(struct hn_rx_ring, hn_lro.lro_queued),
2160 #if __FreeBSD_version < 1100095
2161 hn_rx_stat_int_sysctl,
2163 hn_rx_stat_u64_sysctl,
2165 "LU", "LRO queued");
2166 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_flushed",
2167 CTLTYPE_U64 | CTLFLAG_RW, sc,
2168 __offsetof(struct hn_rx_ring, hn_lro.lro_flushed),
2169 #if __FreeBSD_version < 1100095
2170 hn_rx_stat_int_sysctl,
2172 hn_rx_stat_u64_sysctl,
2174 "LU", "LRO flushed");
2175 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_tried",
2176 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2177 __offsetof(struct hn_rx_ring, hn_lro_tried),
2178 hn_rx_stat_ulong_sysctl, "LU", "# of LRO tries");
2179 #if __FreeBSD_version >= 1100099
2180 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_length_lim",
2181 CTLTYPE_UINT | CTLFLAG_RW, sc, 0, hn_lro_lenlim_sysctl, "IU",
2182 "Max # of data bytes to be aggregated by LRO");
2183 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_ackcnt_lim",
2184 CTLTYPE_INT | CTLFLAG_RW, sc, 0, hn_lro_ackcnt_sysctl, "I",
2185 "Max # of ACKs to be aggregated by LRO");
2187 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hosttcp",
2188 CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_TCP,
2189 hn_trust_hcsum_sysctl, "I",
2190 "Trust tcp segement verification on host side, "
2191 "when csum info is missing");
2192 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostudp",
2193 CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_UDP,
2194 hn_trust_hcsum_sysctl, "I",
2195 "Trust udp datagram verification on host side, "
2196 "when csum info is missing");
2197 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostip",
2198 CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_IP,
2199 hn_trust_hcsum_sysctl, "I",
2200 "Trust ip packet verification on host side, "
2201 "when csum info is missing");
2202 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_ip",
2203 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2204 __offsetof(struct hn_rx_ring, hn_csum_ip),
2205 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM IP");
2206 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_tcp",
2207 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2208 __offsetof(struct hn_rx_ring, hn_csum_tcp),
2209 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM TCP");
2210 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_udp",
2211 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2212 __offsetof(struct hn_rx_ring, hn_csum_udp),
2213 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM UDP");
2214 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_trusted",
2215 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2216 __offsetof(struct hn_rx_ring, hn_csum_trusted),
2217 hn_rx_stat_ulong_sysctl, "LU",
2218 "# of packets that we trust host's csum verification");
2219 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "small_pkts",
2220 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2221 __offsetof(struct hn_rx_ring, hn_small_pkts),
2222 hn_rx_stat_ulong_sysctl, "LU", "# of small packets received");
2223 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_cnt",
2224 CTLFLAG_RD, &sc->hn_rx_ring_cnt, 0, "# created RX rings");
2225 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_inuse",
2226 CTLFLAG_RD, &sc->hn_rx_ring_inuse, 0, "# used RX rings");
2230 hn_destroy_rx_data(struct hn_softc *sc)
2232 #if defined(INET) || defined(INET6)
2236 if (sc->hn_rx_ring_cnt == 0)
2239 #if defined(INET) || defined(INET6)
2240 for (i = 0; i < sc->hn_rx_ring_cnt; ++i)
2241 tcp_lro_free(&sc->hn_rx_ring[i].hn_lro);
2243 free(sc->hn_rx_ring, M_NETVSC);
2244 sc->hn_rx_ring = NULL;
2246 sc->hn_rx_ring_cnt = 0;
2247 sc->hn_rx_ring_inuse = 0;
2251 hn_create_tx_ring(struct hn_softc *sc, int id)
2253 struct hn_tx_ring *txr = &sc->hn_tx_ring[id];
2254 bus_dma_tag_t parent_dtag;
2258 txr->hn_tx_idx = id;
2260 #ifndef HN_USE_TXDESC_BUFRING
2261 mtx_init(&txr->hn_txlist_spin, "hn txlist", NULL, MTX_SPIN);
2263 mtx_init(&txr->hn_tx_lock, "hn tx", NULL, MTX_DEF);
2265 txr->hn_txdesc_cnt = HN_TX_DESC_CNT;
2266 txr->hn_txdesc = malloc(sizeof(struct hn_txdesc) * txr->hn_txdesc_cnt,
2267 M_NETVSC, M_WAITOK | M_ZERO);
2268 #ifndef HN_USE_TXDESC_BUFRING
2269 SLIST_INIT(&txr->hn_txlist);
2271 txr->hn_txdesc_br = buf_ring_alloc(txr->hn_txdesc_cnt, M_NETVSC,
2272 M_WAITOK, &txr->hn_tx_lock);
2275 txr->hn_tx_taskq = sc->hn_tx_taskq;
2277 if (hn_use_if_start) {
2278 txr->hn_txeof = hn_start_txeof;
2279 TASK_INIT(&txr->hn_tx_task, 0, hn_start_taskfunc, txr);
2280 TASK_INIT(&txr->hn_txeof_task, 0, hn_start_txeof_taskfunc, txr);
2282 txr->hn_txeof = hn_xmit_txeof;
2283 TASK_INIT(&txr->hn_tx_task, 0, hn_xmit_taskfunc, txr);
2284 TASK_INIT(&txr->hn_txeof_task, 0, hn_xmit_txeof_taskfunc, txr);
2285 txr->hn_mbuf_br = buf_ring_alloc(txr->hn_txdesc_cnt, M_NETVSC,
2286 M_WAITOK, &txr->hn_tx_lock);
2289 txr->hn_direct_tx_size = hn_direct_tx_size;
2290 if (hv_vmbus_protocal_version >= HV_VMBUS_VERSION_WIN8_1)
2291 txr->hn_csum_assist = HN_CSUM_ASSIST;
2293 txr->hn_csum_assist = HN_CSUM_ASSIST_WIN8;
2296 * Always schedule transmission instead of trying to do direct
2297 * transmission. This one gives the best performance so far.
2299 txr->hn_sched_tx = 1;
2301 parent_dtag = bus_get_dma_tag(sc->hn_dev);
2303 /* DMA tag for RNDIS messages. */
2304 error = bus_dma_tag_create(parent_dtag, /* parent */
2305 HN_RNDIS_MSG_ALIGN, /* alignment */
2306 HN_RNDIS_MSG_BOUNDARY, /* boundary */
2307 BUS_SPACE_MAXADDR, /* lowaddr */
2308 BUS_SPACE_MAXADDR, /* highaddr */
2309 NULL, NULL, /* filter, filterarg */
2310 HN_RNDIS_MSG_LEN, /* maxsize */
2312 HN_RNDIS_MSG_LEN, /* maxsegsize */
2314 NULL, /* lockfunc */
2315 NULL, /* lockfuncarg */
2316 &txr->hn_tx_rndis_dtag);
2318 device_printf(sc->hn_dev, "failed to create rndis dmatag\n");
2322 /* DMA tag for data. */
2323 error = bus_dma_tag_create(parent_dtag, /* parent */
2325 HN_TX_DATA_BOUNDARY, /* boundary */
2326 BUS_SPACE_MAXADDR, /* lowaddr */
2327 BUS_SPACE_MAXADDR, /* highaddr */
2328 NULL, NULL, /* filter, filterarg */
2329 HN_TX_DATA_MAXSIZE, /* maxsize */
2330 HN_TX_DATA_SEGCNT_MAX, /* nsegments */
2331 HN_TX_DATA_SEGSIZE, /* maxsegsize */
2333 NULL, /* lockfunc */
2334 NULL, /* lockfuncarg */
2335 &txr->hn_tx_data_dtag);
2337 device_printf(sc->hn_dev, "failed to create data dmatag\n");
2341 for (i = 0; i < txr->hn_txdesc_cnt; ++i) {
2342 struct hn_txdesc *txd = &txr->hn_txdesc[i];
2347 * Allocate and load RNDIS messages.
2349 error = bus_dmamem_alloc(txr->hn_tx_rndis_dtag,
2350 (void **)&txd->rndis_msg,
2351 BUS_DMA_WAITOK | BUS_DMA_COHERENT,
2352 &txd->rndis_msg_dmap);
2354 device_printf(sc->hn_dev,
2355 "failed to allocate rndis_msg, %d\n", i);
2359 error = bus_dmamap_load(txr->hn_tx_rndis_dtag,
2360 txd->rndis_msg_dmap,
2361 txd->rndis_msg, HN_RNDIS_MSG_LEN,
2362 hn_dma_map_paddr, &txd->rndis_msg_paddr,
2365 device_printf(sc->hn_dev,
2366 "failed to load rndis_msg, %d\n", i);
2367 bus_dmamem_free(txr->hn_tx_rndis_dtag,
2368 txd->rndis_msg, txd->rndis_msg_dmap);
2372 /* DMA map for TX data. */
2373 error = bus_dmamap_create(txr->hn_tx_data_dtag, 0,
2376 device_printf(sc->hn_dev,
2377 "failed to allocate tx data dmamap\n");
2378 bus_dmamap_unload(txr->hn_tx_rndis_dtag,
2379 txd->rndis_msg_dmap);
2380 bus_dmamem_free(txr->hn_tx_rndis_dtag,
2381 txd->rndis_msg, txd->rndis_msg_dmap);
2385 /* All set, put it to list */
2386 txd->flags |= HN_TXD_FLAG_ONLIST;
2387 #ifndef HN_USE_TXDESC_BUFRING
2388 SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
2390 buf_ring_enqueue(txr->hn_txdesc_br, txd);
2393 txr->hn_txdesc_avail = txr->hn_txdesc_cnt;
2395 if (sc->hn_tx_sysctl_tree != NULL) {
2396 struct sysctl_oid_list *child;
2397 struct sysctl_ctx_list *ctx;
2401 * Create per TX ring sysctl tree:
2402 * dev.hn.UNIT.tx.RINGID
2404 ctx = device_get_sysctl_ctx(sc->hn_dev);
2405 child = SYSCTL_CHILDREN(sc->hn_tx_sysctl_tree);
2407 snprintf(name, sizeof(name), "%d", id);
2408 txr->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO,
2409 name, CTLFLAG_RD, 0, "");
2411 if (txr->hn_tx_sysctl_tree != NULL) {
2412 child = SYSCTL_CHILDREN(txr->hn_tx_sysctl_tree);
2414 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_avail",
2415 CTLFLAG_RD, &txr->hn_txdesc_avail, 0,
2416 "# of available TX descs");
2417 if (!hn_use_if_start) {
2418 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "oactive",
2419 CTLFLAG_RD, &txr->hn_oactive, 0,
2429 hn_txdesc_dmamap_destroy(struct hn_txdesc *txd)
2431 struct hn_tx_ring *txr = txd->txr;
2433 KASSERT(txd->m == NULL, ("still has mbuf installed"));
2434 KASSERT((txd->flags & HN_TXD_FLAG_DMAMAP) == 0, ("still dma mapped"));
2436 bus_dmamap_unload(txr->hn_tx_rndis_dtag, txd->rndis_msg_dmap);
2437 bus_dmamem_free(txr->hn_tx_rndis_dtag, txd->rndis_msg,
2438 txd->rndis_msg_dmap);
2439 bus_dmamap_destroy(txr->hn_tx_data_dtag, txd->data_dmap);
2443 hn_destroy_tx_ring(struct hn_tx_ring *txr)
2445 struct hn_txdesc *txd;
2447 if (txr->hn_txdesc == NULL)
2450 #ifndef HN_USE_TXDESC_BUFRING
2451 while ((txd = SLIST_FIRST(&txr->hn_txlist)) != NULL) {
2452 SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
2453 hn_txdesc_dmamap_destroy(txd);
2456 mtx_lock(&txr->hn_tx_lock);
2457 while ((txd = buf_ring_dequeue_sc(txr->hn_txdesc_br)) != NULL)
2458 hn_txdesc_dmamap_destroy(txd);
2459 mtx_unlock(&txr->hn_tx_lock);
2462 if (txr->hn_tx_data_dtag != NULL)
2463 bus_dma_tag_destroy(txr->hn_tx_data_dtag);
2464 if (txr->hn_tx_rndis_dtag != NULL)
2465 bus_dma_tag_destroy(txr->hn_tx_rndis_dtag);
2467 #ifdef HN_USE_TXDESC_BUFRING
2468 buf_ring_free(txr->hn_txdesc_br, M_NETVSC);
2471 free(txr->hn_txdesc, M_NETVSC);
2472 txr->hn_txdesc = NULL;
2474 if (txr->hn_mbuf_br != NULL)
2475 buf_ring_free(txr->hn_mbuf_br, M_NETVSC);
2477 #ifndef HN_USE_TXDESC_BUFRING
2478 mtx_destroy(&txr->hn_txlist_spin);
2480 mtx_destroy(&txr->hn_tx_lock);
2484 hn_create_tx_data(struct hn_softc *sc, int ring_cnt)
2486 struct sysctl_oid_list *child;
2487 struct sysctl_ctx_list *ctx;
2490 sc->hn_tx_ring_cnt = ring_cnt;
2491 sc->hn_tx_ring_inuse = sc->hn_tx_ring_cnt;
2493 sc->hn_tx_ring = malloc(sizeof(struct hn_tx_ring) * sc->hn_tx_ring_cnt,
2494 M_NETVSC, M_WAITOK | M_ZERO);
2496 ctx = device_get_sysctl_ctx(sc->hn_dev);
2497 child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->hn_dev));
2499 /* Create dev.hn.UNIT.tx sysctl tree */
2500 sc->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "tx",
2503 for (i = 0; i < sc->hn_tx_ring_cnt; ++i) {
2506 error = hn_create_tx_ring(sc, i);
2511 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "no_txdescs",
2512 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2513 __offsetof(struct hn_tx_ring, hn_no_txdescs),
2514 hn_tx_stat_ulong_sysctl, "LU", "# of times short of TX descs");
2515 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "send_failed",
2516 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2517 __offsetof(struct hn_tx_ring, hn_send_failed),
2518 hn_tx_stat_ulong_sysctl, "LU", "# of hyper-v sending failure");
2519 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "txdma_failed",
2520 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2521 __offsetof(struct hn_tx_ring, hn_txdma_failed),
2522 hn_tx_stat_ulong_sysctl, "LU", "# of TX DMA failure");
2523 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_collapsed",
2524 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2525 __offsetof(struct hn_tx_ring, hn_tx_collapsed),
2526 hn_tx_stat_ulong_sysctl, "LU", "# of TX mbuf collapsed");
2527 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney",
2528 CTLTYPE_ULONG | CTLFLAG_RW, sc,
2529 __offsetof(struct hn_tx_ring, hn_tx_chimney),
2530 hn_tx_stat_ulong_sysctl, "LU", "# of chimney send");
2531 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_cnt",
2532 CTLFLAG_RD, &sc->hn_tx_ring[0].hn_txdesc_cnt, 0,
2533 "# of total TX descs");
2534 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_chimney_max",
2535 CTLFLAG_RD, &sc->hn_tx_chimney_max, 0,
2536 "Chimney send packet size upper boundary");
2537 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_size",
2538 CTLTYPE_INT | CTLFLAG_RW, sc, 0, hn_tx_chimney_size_sysctl,
2539 "I", "Chimney send packet size limit");
2540 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "direct_tx_size",
2541 CTLTYPE_INT | CTLFLAG_RW, sc,
2542 __offsetof(struct hn_tx_ring, hn_direct_tx_size),
2543 hn_tx_conf_int_sysctl, "I",
2544 "Size of the packet for direct transmission");
2545 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "sched_tx",
2546 CTLTYPE_INT | CTLFLAG_RW, sc,
2547 __offsetof(struct hn_tx_ring, hn_sched_tx),
2548 hn_tx_conf_int_sysctl, "I",
2549 "Always schedule transmission "
2550 "instead of doing direct transmission");
2551 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_cnt",
2552 CTLFLAG_RD, &sc->hn_tx_ring_cnt, 0, "# created TX rings");
2553 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_inuse",
2554 CTLFLAG_RD, &sc->hn_tx_ring_inuse, 0, "# used TX rings");
2560 hn_set_tx_chimney_size(struct hn_softc *sc, int chimney_size)
2565 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
2566 sc->hn_tx_ring[i].hn_tx_chimney_size = chimney_size;
2571 hn_destroy_tx_data(struct hn_softc *sc)
2575 if (sc->hn_tx_ring_cnt == 0)
2578 for (i = 0; i < sc->hn_tx_ring_cnt; ++i)
2579 hn_destroy_tx_ring(&sc->hn_tx_ring[i]);
2581 free(sc->hn_tx_ring, M_NETVSC);
2582 sc->hn_tx_ring = NULL;
2584 sc->hn_tx_ring_cnt = 0;
2585 sc->hn_tx_ring_inuse = 0;
2589 hn_start_taskfunc(void *xtxr, int pending __unused)
2591 struct hn_tx_ring *txr = xtxr;
2593 mtx_lock(&txr->hn_tx_lock);
2594 hn_start_locked(txr, 0);
2595 mtx_unlock(&txr->hn_tx_lock);
2599 hn_start_txeof_taskfunc(void *xtxr, int pending __unused)
2601 struct hn_tx_ring *txr = xtxr;
2603 mtx_lock(&txr->hn_tx_lock);
2604 atomic_clear_int(&txr->hn_sc->hn_ifp->if_drv_flags, IFF_DRV_OACTIVE);
2605 hn_start_locked(txr, 0);
2606 mtx_unlock(&txr->hn_tx_lock);
2610 hn_stop_tx_tasks(struct hn_softc *sc)
2614 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2615 struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
2617 taskqueue_drain(txr->hn_tx_taskq, &txr->hn_tx_task);
2618 taskqueue_drain(txr->hn_tx_taskq, &txr->hn_txeof_task);
2623 hn_xmit(struct hn_tx_ring *txr, int len)
2625 struct hn_softc *sc = txr->hn_sc;
2626 struct ifnet *ifp = sc->hn_ifp;
2627 struct mbuf *m_head;
2629 mtx_assert(&txr->hn_tx_lock, MA_OWNED);
2630 KASSERT(hn_use_if_start == 0,
2631 ("hn_xmit is called, when if_start is enabled"));
2633 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || txr->hn_oactive)
2636 while ((m_head = drbr_peek(ifp, txr->hn_mbuf_br)) != NULL) {
2637 struct hn_txdesc *txd;
2640 if (len > 0 && m_head->m_pkthdr.len > len) {
2642 * This sending could be time consuming; let callers
2643 * dispatch this packet sending (and sending of any
2644 * following up packets) to tx taskqueue.
2646 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2650 txd = hn_txdesc_get(txr);
2652 txr->hn_no_txdescs++;
2653 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2654 txr->hn_oactive = 1;
2658 error = hn_encap(txr, txd, &m_head);
2660 /* Both txd and m_head are freed; discard */
2661 drbr_advance(ifp, txr->hn_mbuf_br);
2665 error = hn_send_pkt(ifp, txr, txd);
2666 if (__predict_false(error)) {
2667 /* txd is freed, but m_head is not */
2668 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2669 txr->hn_oactive = 1;
2674 drbr_advance(ifp, txr->hn_mbuf_br);
2680 hn_transmit(struct ifnet *ifp, struct mbuf *m)
2682 struct hn_softc *sc = ifp->if_softc;
2683 struct hn_tx_ring *txr;
2687 * Select the TX ring based on flowid
2689 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2690 idx = m->m_pkthdr.flowid % sc->hn_tx_ring_inuse;
2691 txr = &sc->hn_tx_ring[idx];
2693 error = drbr_enqueue(ifp, txr->hn_mbuf_br, m);
2697 if (txr->hn_oactive)
2700 if (txr->hn_sched_tx)
2703 if (mtx_trylock(&txr->hn_tx_lock)) {
2706 sched = hn_xmit(txr, txr->hn_direct_tx_size);
2707 mtx_unlock(&txr->hn_tx_lock);
2712 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
2717 hn_xmit_qflush(struct ifnet *ifp)
2719 struct hn_softc *sc = ifp->if_softc;
2722 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2723 struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
2726 mtx_lock(&txr->hn_tx_lock);
2727 while ((m = buf_ring_dequeue_sc(txr->hn_mbuf_br)) != NULL)
2729 mtx_unlock(&txr->hn_tx_lock);
2735 hn_xmit_txeof(struct hn_tx_ring *txr)
2738 if (txr->hn_sched_tx)
2741 if (mtx_trylock(&txr->hn_tx_lock)) {
2744 txr->hn_oactive = 0;
2745 sched = hn_xmit(txr, txr->hn_direct_tx_size);
2746 mtx_unlock(&txr->hn_tx_lock);
2748 taskqueue_enqueue(txr->hn_tx_taskq,
2754 * Release the oactive earlier, with the hope, that
2755 * others could catch up. The task will clear the
2756 * oactive again with the hn_tx_lock to avoid possible
2759 txr->hn_oactive = 0;
2760 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
2765 hn_xmit_taskfunc(void *xtxr, int pending __unused)
2767 struct hn_tx_ring *txr = xtxr;
2769 mtx_lock(&txr->hn_tx_lock);
2771 mtx_unlock(&txr->hn_tx_lock);
2775 hn_xmit_txeof_taskfunc(void *xtxr, int pending __unused)
2777 struct hn_tx_ring *txr = xtxr;
2779 mtx_lock(&txr->hn_tx_lock);
2780 txr->hn_oactive = 0;
2782 mtx_unlock(&txr->hn_tx_lock);
2786 hn_tx_taskq_create(void *arg __unused)
2788 if (!hn_share_tx_taskq)
2791 hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
2792 taskqueue_thread_enqueue, &hn_tx_taskq);
2793 taskqueue_start_threads(&hn_tx_taskq, 1, PI_NET, "hn tx");
2794 if (hn_bind_tx_taskq >= 0) {
2795 int cpu = hn_bind_tx_taskq;
2796 struct task cpuset_task;
2799 if (cpu > mp_ncpus - 1)
2801 CPU_SETOF(cpu, &cpu_set);
2802 TASK_INIT(&cpuset_task, 0, hn_cpuset_setthread_task, &cpu_set);
2803 taskqueue_enqueue(hn_tx_taskq, &cpuset_task);
2804 taskqueue_drain(hn_tx_taskq, &cpuset_task);
2807 SYSINIT(hn_txtq_create, SI_SUB_DRIVERS, SI_ORDER_FIRST,
2808 hn_tx_taskq_create, NULL);
2811 hn_tx_taskq_destroy(void *arg __unused)
2813 if (hn_tx_taskq != NULL)
2814 taskqueue_free(hn_tx_taskq);
2816 SYSUNINIT(hn_txtq_destroy, SI_SUB_DRIVERS, SI_ORDER_FIRST,
2817 hn_tx_taskq_destroy, NULL);
2819 static device_method_t netvsc_methods[] = {
2820 /* Device interface */
2821 DEVMETHOD(device_probe, netvsc_probe),
2822 DEVMETHOD(device_attach, netvsc_attach),
2823 DEVMETHOD(device_detach, netvsc_detach),
2824 DEVMETHOD(device_shutdown, netvsc_shutdown),
2829 static driver_t netvsc_driver = {
2835 static devclass_t netvsc_devclass;
2837 DRIVER_MODULE(hn, vmbus, netvsc_driver, netvsc_devclass, 0, 0);
2838 MODULE_VERSION(hn, 1);
2839 MODULE_DEPEND(hn, vmbus, 1, 1, 1);