2 * Copyright (c) 2010-2012 Citrix Inc.
3 * Copyright (c) 2009-2012,2016 Microsoft Corp.
4 * Copyright (c) 2012 NetApp Inc.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice unmodified, this list of conditions, and the following
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30 * Copyright (c) 2004-2006 Kip Macy
31 * All rights reserved.
33 * Redistribution and use in source and binary forms, with or without
34 * modification, are permitted provided that the following conditions
36 * 1. Redistributions of source code must retain the above copyright
37 * notice, this list of conditions and the following disclaimer.
38 * 2. Redistributions in binary form must reproduce the above copyright
39 * notice, this list of conditions and the following disclaimer in the
40 * documentation and/or other materials provided with the distribution.
42 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
43 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
44 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
45 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
46 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
47 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
48 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
49 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
50 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
51 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
55 #include <sys/cdefs.h>
56 __FBSDID("$FreeBSD$");
58 #include "opt_inet6.h"
61 #include <sys/param.h>
62 #include <sys/systm.h>
63 #include <sys/sockio.h>
65 #include <sys/malloc.h>
66 #include <sys/module.h>
67 #include <sys/kernel.h>
68 #include <sys/socket.h>
70 #include <sys/queue.h>
74 #include <sys/sysctl.h>
75 #include <sys/buf_ring.h>
78 #include <net/if_arp.h>
79 #include <net/ethernet.h>
80 #include <net/if_dl.h>
81 #include <net/if_media.h>
85 #include <net/if_types.h>
86 #include <net/if_vlan_var.h>
89 #include <netinet/in_systm.h>
90 #include <netinet/in.h>
91 #include <netinet/ip.h>
92 #include <netinet/if_ether.h>
93 #include <netinet/tcp.h>
94 #include <netinet/udp.h>
95 #include <netinet/ip6.h>
98 #include <vm/vm_param.h>
99 #include <vm/vm_kern.h>
102 #include <machine/bus.h>
103 #include <machine/resource.h>
104 #include <machine/frame.h>
105 #include <machine/vmparam.h>
108 #include <sys/rman.h>
109 #include <sys/mutex.h>
110 #include <sys/errno.h>
111 #include <sys/types.h>
112 #include <machine/atomic.h>
114 #include <machine/intr_machdep.h>
116 #include <machine/in_cksum.h>
118 #include <dev/hyperv/include/hyperv.h>
119 #include <dev/hyperv/include/hyperv_busdma.h>
120 #include <dev/hyperv/include/vmbus_xact.h>
122 #include "hv_net_vsc.h"
123 #include "hv_rndis.h"
124 #include "hv_rndis_filter.h"
125 #include "vmbus_if.h"
127 /* Short for Hyper-V network interface */
128 #define NETVSC_DEVNAME "hn"
130 #define HN_XACT_REQ_SIZE (2 * PAGE_SIZE)
131 #define HN_XACT_RESP_SIZE (2 * PAGE_SIZE)
134 * It looks like offset 0 of buf is reserved to hold the softc pointer.
135 * The sc pointer evidently not needed, and is not presently populated.
136 * The packet offset is where the netvsc_packet starts in the buffer.
138 #define HV_NV_SC_PTR_OFFSET_IN_BUF 0
139 #define HV_NV_PACKET_OFFSET_IN_BUF 16
141 /* YYY should get it from the underlying channel */
142 #define HN_TX_DESC_CNT 512
144 #define HN_LROENT_CNT_DEF 128
146 #define HN_RING_CNT_DEF_MAX 8
148 #define HN_RNDIS_MSG_LEN \
149 (sizeof(rndis_msg) + \
150 RNDIS_HASHVAL_PPI_SIZE + \
151 RNDIS_VLAN_PPI_SIZE + \
152 RNDIS_TSO_PPI_SIZE + \
154 #define HN_RNDIS_MSG_BOUNDARY PAGE_SIZE
155 #define HN_RNDIS_MSG_ALIGN CACHE_LINE_SIZE
157 #define HN_TX_DATA_BOUNDARY PAGE_SIZE
158 #define HN_TX_DATA_MAXSIZE IP_MAXPACKET
159 #define HN_TX_DATA_SEGSIZE PAGE_SIZE
160 #define HN_TX_DATA_SEGCNT_MAX \
161 (NETVSC_PACKET_MAXPAGE - HV_RF_NUM_TX_RESERVED_PAGE_BUFS)
163 #define HN_DIRECT_TX_SIZE_DEF 128
165 #define HN_EARLY_TXEOF_THRESH 8
168 #ifndef HN_USE_TXDESC_BUFRING
169 SLIST_ENTRY(hn_txdesc) link;
172 struct hn_tx_ring *txr;
174 uint32_t flags; /* HN_TXD_FLAG_ */
175 struct hn_send_ctx send_ctx;
177 bus_dmamap_t data_dmap;
179 bus_addr_t rndis_msg_paddr;
180 rndis_msg *rndis_msg;
181 bus_dmamap_t rndis_msg_dmap;
184 #define HN_TXD_FLAG_ONLIST 0x1
185 #define HN_TXD_FLAG_DMAMAP 0x2
188 * Only enable UDP checksum offloading when it is on 2012R2 or
189 * later. UDP checksum offloading doesn't work on earlier
192 #define HN_CSUM_ASSIST_WIN8 (CSUM_IP | CSUM_TCP)
193 #define HN_CSUM_ASSIST (CSUM_IP | CSUM_UDP | CSUM_TCP)
195 #define HN_LRO_LENLIM_MULTIRX_DEF (12 * ETHERMTU)
196 #define HN_LRO_LENLIM_DEF (25 * ETHERMTU)
197 /* YYY 2*MTU is a bit rough, but should be good enough. */
198 #define HN_LRO_LENLIM_MIN(ifp) (2 * (ifp)->if_mtu)
200 #define HN_LRO_ACKCNT_DEF 1
203 * Be aware that this sleepable mutex will exhibit WITNESS errors when
204 * certain TCP and ARP code paths are taken. This appears to be a
205 * well-known condition, as all other drivers checked use a sleeping
206 * mutex to protect their transmit paths.
207 * Also Be aware that mutexes do not play well with semaphores, and there
208 * is a conflicting semaphore in a certain channel code path.
210 #define NV_LOCK_INIT(_sc, _name) \
211 mtx_init(&(_sc)->hn_lock, _name, MTX_NETWORK_LOCK, MTX_DEF)
212 #define NV_LOCK(_sc) mtx_lock(&(_sc)->hn_lock)
213 #define NV_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->hn_lock, MA_OWNED)
214 #define NV_UNLOCK(_sc) mtx_unlock(&(_sc)->hn_lock)
215 #define NV_LOCK_DESTROY(_sc) mtx_destroy(&(_sc)->hn_lock)
222 int hv_promisc_mode = 0; /* normal mode by default */
224 SYSCTL_NODE(_hw, OID_AUTO, hn, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
225 "Hyper-V network interface");
227 /* Trust tcp segements verification on host side. */
228 static int hn_trust_hosttcp = 1;
229 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hosttcp, CTLFLAG_RDTUN,
230 &hn_trust_hosttcp, 0,
231 "Trust tcp segement verification on host side, "
232 "when csum info is missing (global setting)");
234 /* Trust udp datagrams verification on host side. */
235 static int hn_trust_hostudp = 1;
236 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostudp, CTLFLAG_RDTUN,
237 &hn_trust_hostudp, 0,
238 "Trust udp datagram verification on host side, "
239 "when csum info is missing (global setting)");
241 /* Trust ip packets verification on host side. */
242 static int hn_trust_hostip = 1;
243 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostip, CTLFLAG_RDTUN,
245 "Trust ip packet verification on host side, "
246 "when csum info is missing (global setting)");
248 /* Limit TSO burst size */
249 static int hn_tso_maxlen = 0;
250 SYSCTL_INT(_hw_hn, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN,
251 &hn_tso_maxlen, 0, "TSO burst limit");
253 /* Limit chimney send size */
254 static int hn_tx_chimney_size = 0;
255 SYSCTL_INT(_hw_hn, OID_AUTO, tx_chimney_size, CTLFLAG_RDTUN,
256 &hn_tx_chimney_size, 0, "Chimney send packet size limit");
258 /* Limit the size of packet for direct transmission */
259 static int hn_direct_tx_size = HN_DIRECT_TX_SIZE_DEF;
260 SYSCTL_INT(_hw_hn, OID_AUTO, direct_tx_size, CTLFLAG_RDTUN,
261 &hn_direct_tx_size, 0, "Size of the packet for direct transmission");
263 #if defined(INET) || defined(INET6)
264 #if __FreeBSD_version >= 1100095
265 static int hn_lro_entry_count = HN_LROENT_CNT_DEF;
266 SYSCTL_INT(_hw_hn, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN,
267 &hn_lro_entry_count, 0, "LRO entry count");
271 static int hn_share_tx_taskq = 0;
272 SYSCTL_INT(_hw_hn, OID_AUTO, share_tx_taskq, CTLFLAG_RDTUN,
273 &hn_share_tx_taskq, 0, "Enable shared TX taskqueue");
275 static struct taskqueue *hn_tx_taskq;
277 #ifndef HN_USE_TXDESC_BUFRING
278 static int hn_use_txdesc_bufring = 0;
280 static int hn_use_txdesc_bufring = 1;
282 SYSCTL_INT(_hw_hn, OID_AUTO, use_txdesc_bufring, CTLFLAG_RD,
283 &hn_use_txdesc_bufring, 0, "Use buf_ring for TX descriptors");
285 static int hn_bind_tx_taskq = -1;
286 SYSCTL_INT(_hw_hn, OID_AUTO, bind_tx_taskq, CTLFLAG_RDTUN,
287 &hn_bind_tx_taskq, 0, "Bind TX taskqueue to the specified cpu");
289 static int hn_use_if_start = 0;
290 SYSCTL_INT(_hw_hn, OID_AUTO, use_if_start, CTLFLAG_RDTUN,
291 &hn_use_if_start, 0, "Use if_start TX method");
293 static int hn_chan_cnt = 0;
294 SYSCTL_INT(_hw_hn, OID_AUTO, chan_cnt, CTLFLAG_RDTUN,
296 "# of channels to use; each channel has one RX ring and one TX ring");
298 static int hn_tx_ring_cnt = 0;
299 SYSCTL_INT(_hw_hn, OID_AUTO, tx_ring_cnt, CTLFLAG_RDTUN,
300 &hn_tx_ring_cnt, 0, "# of TX rings to use");
302 static int hn_tx_swq_depth = 0;
303 SYSCTL_INT(_hw_hn, OID_AUTO, tx_swq_depth, CTLFLAG_RDTUN,
304 &hn_tx_swq_depth, 0, "Depth of IFQ or BUFRING");
306 #if __FreeBSD_version >= 1100095
307 static u_int hn_lro_mbufq_depth = 0;
308 SYSCTL_UINT(_hw_hn, OID_AUTO, lro_mbufq_depth, CTLFLAG_RDTUN,
309 &hn_lro_mbufq_depth, 0, "Depth of LRO mbuf queue");
312 static u_int hn_cpu_index;
315 * Forward declarations
317 static void hn_stop(hn_softc_t *sc);
318 static void hn_ifinit_locked(hn_softc_t *sc);
319 static void hn_ifinit(void *xsc);
320 static int hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data);
321 static int hn_start_locked(struct hn_tx_ring *txr, int len);
322 static void hn_start(struct ifnet *ifp);
323 static void hn_start_txeof(struct hn_tx_ring *);
324 static int hn_ifmedia_upd(struct ifnet *ifp);
325 static void hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr);
326 #if __FreeBSD_version >= 1100099
327 static int hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS);
328 static int hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS);
330 static int hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS);
331 static int hn_chim_size_sysctl(SYSCTL_HANDLER_ARGS);
332 #if __FreeBSD_version < 1100095
333 static int hn_rx_stat_int_sysctl(SYSCTL_HANDLER_ARGS);
335 static int hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS);
337 static int hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS);
338 static int hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS);
339 static int hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS);
340 static int hn_check_iplen(const struct mbuf *, int);
341 static int hn_create_tx_ring(struct hn_softc *, int);
342 static void hn_destroy_tx_ring(struct hn_tx_ring *);
343 static int hn_create_tx_data(struct hn_softc *, int);
344 static void hn_destroy_tx_data(struct hn_softc *);
345 static void hn_start_taskfunc(void *, int);
346 static void hn_start_txeof_taskfunc(void *, int);
347 static void hn_stop_tx_tasks(struct hn_softc *);
348 static int hn_encap(struct hn_tx_ring *, struct hn_txdesc *, struct mbuf **);
349 static int hn_create_rx_data(struct hn_softc *sc, int);
350 static void hn_destroy_rx_data(struct hn_softc *sc);
351 static void hn_set_chim_size(struct hn_softc *, int);
352 static void hn_channel_attach(struct hn_softc *, struct vmbus_channel *);
353 static void hn_subchan_attach(struct hn_softc *, struct vmbus_channel *);
354 static void hn_subchan_setup(struct hn_softc *);
356 static int hn_transmit(struct ifnet *, struct mbuf *);
357 static void hn_xmit_qflush(struct ifnet *);
358 static int hn_xmit(struct hn_tx_ring *, int);
359 static void hn_xmit_txeof(struct hn_tx_ring *);
360 static void hn_xmit_taskfunc(void *, int);
361 static void hn_xmit_txeof_taskfunc(void *, int);
363 #if __FreeBSD_version >= 1100099
365 hn_set_lro_lenlim(struct hn_softc *sc, int lenlim)
369 for (i = 0; i < sc->hn_rx_ring_inuse; ++i)
370 sc->hn_rx_ring[i].hn_lro.lro_length_lim = lenlim;
375 hn_get_txswq_depth(const struct hn_tx_ring *txr)
378 KASSERT(txr->hn_txdesc_cnt > 0, ("tx ring is not setup yet"));
379 if (hn_tx_swq_depth < txr->hn_txdesc_cnt)
380 return txr->hn_txdesc_cnt;
381 return hn_tx_swq_depth;
385 hn_ifmedia_upd(struct ifnet *ifp __unused)
392 hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
394 struct hn_softc *sc = ifp->if_softc;
396 ifmr->ifm_status = IFM_AVALID;
397 ifmr->ifm_active = IFM_ETHER;
399 if (!sc->hn_carrier) {
400 ifmr->ifm_active |= IFM_NONE;
403 ifmr->ifm_status |= IFM_ACTIVE;
404 ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
407 /* {F8615163-DF3E-46c5-913F-F2D2F965ED0E} */
408 static const struct hyperv_guid g_net_vsc_device_type = {
409 .hv_guid = {0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
410 0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E}
414 * Standard probe entry point.
418 netvsc_probe(device_t dev)
420 if (VMBUS_PROBE_GUID(device_get_parent(dev), dev,
421 &g_net_vsc_device_type) == 0) {
422 device_set_desc(dev, "Hyper-V Network Interface");
423 return BUS_PROBE_DEFAULT;
429 hn_cpuset_setthread_task(void *xmask, int pending __unused)
431 cpuset_t *mask = xmask;
434 error = cpuset_setthread(curthread->td_tid, mask);
436 panic("curthread=%ju: can't pin; error=%d",
437 (uintmax_t)curthread->td_tid, error);
442 * Standard attach entry point.
444 * Called when the driver is loaded. It allocates needed resources,
445 * and initializes the "hardware" and software.
448 netvsc_attach(device_t dev)
450 netvsc_device_info device_info;
452 int unit = device_get_unit(dev);
453 struct ifnet *ifp = NULL;
454 int error, ring_cnt, tx_ring_cnt;
457 sc = device_get_softc(dev);
461 sc->hn_prichan = vmbus_get_channel(dev);
463 if (hn_tx_taskq == NULL) {
464 sc->hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
465 taskqueue_thread_enqueue, &sc->hn_tx_taskq);
466 taskqueue_start_threads(&sc->hn_tx_taskq, 1, PI_NET, "%s tx",
467 device_get_nameunit(dev));
468 if (hn_bind_tx_taskq >= 0) {
469 int cpu = hn_bind_tx_taskq;
470 struct task cpuset_task;
473 if (cpu > mp_ncpus - 1)
475 CPU_SETOF(cpu, &cpu_set);
476 TASK_INIT(&cpuset_task, 0, hn_cpuset_setthread_task,
478 taskqueue_enqueue(sc->hn_tx_taskq, &cpuset_task);
479 taskqueue_drain(sc->hn_tx_taskq, &cpuset_task);
482 sc->hn_tx_taskq = hn_tx_taskq;
484 NV_LOCK_INIT(sc, "NetVSCLock");
486 ifp = sc->hn_ifp = sc->arpcom.ac_ifp = if_alloc(IFT_ETHER);
488 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
491 * Figure out the # of RX rings (ring_cnt) and the # of TX rings
492 * to use (tx_ring_cnt).
495 * The # of RX rings to use is same as the # of channels to use.
497 ring_cnt = hn_chan_cnt;
501 if (ring_cnt > HN_RING_CNT_DEF_MAX)
502 ring_cnt = HN_RING_CNT_DEF_MAX;
503 } else if (ring_cnt > mp_ncpus) {
507 tx_ring_cnt = hn_tx_ring_cnt;
508 if (tx_ring_cnt <= 0 || tx_ring_cnt > ring_cnt)
509 tx_ring_cnt = ring_cnt;
510 if (hn_use_if_start) {
511 /* ifnet.if_start only needs one TX ring. */
516 * Set the leader CPU for channels.
518 sc->hn_cpu = atomic_fetchadd_int(&hn_cpu_index, ring_cnt) % mp_ncpus;
520 error = hn_create_tx_data(sc, tx_ring_cnt);
523 error = hn_create_rx_data(sc, ring_cnt);
528 * Associate the first TX/RX ring w/ the primary channel.
530 hn_channel_attach(sc, sc->hn_prichan);
532 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
533 ifp->if_ioctl = hn_ioctl;
534 ifp->if_init = hn_ifinit;
535 /* needed by hv_rf_on_device_add() code */
536 ifp->if_mtu = ETHERMTU;
537 if (hn_use_if_start) {
538 int qdepth = hn_get_txswq_depth(&sc->hn_tx_ring[0]);
540 ifp->if_start = hn_start;
541 IFQ_SET_MAXLEN(&ifp->if_snd, qdepth);
542 ifp->if_snd.ifq_drv_maxlen = qdepth - 1;
543 IFQ_SET_READY(&ifp->if_snd);
545 ifp->if_transmit = hn_transmit;
546 ifp->if_qflush = hn_xmit_qflush;
549 ifmedia_init(&sc->hn_media, 0, hn_ifmedia_upd, hn_ifmedia_sts);
550 ifmedia_add(&sc->hn_media, IFM_ETHER | IFM_AUTO, 0, NULL);
551 ifmedia_set(&sc->hn_media, IFM_ETHER | IFM_AUTO);
552 /* XXX ifmedia_set really should do this for us */
553 sc->hn_media.ifm_media = sc->hn_media.ifm_cur->ifm_media;
556 * Tell upper layers that we support full VLAN capability.
558 ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
559 ifp->if_capabilities |=
560 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
563 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
565 ifp->if_hwassist = sc->hn_tx_ring[0].hn_csum_assist | CSUM_TSO;
567 sc->hn_xact = vmbus_xact_ctx_create(bus_get_dma_tag(dev),
568 HN_XACT_REQ_SIZE, HN_XACT_RESP_SIZE, 0);
569 if (sc->hn_xact == NULL)
572 error = hv_rf_on_device_add(sc, &device_info, &ring_cnt,
576 KASSERT(ring_cnt > 0 && ring_cnt <= sc->hn_rx_ring_inuse,
577 ("invalid channel count %d, should be less than %d",
578 ring_cnt, sc->hn_rx_ring_inuse));
581 * Set the # of TX/RX rings that could be used according to
582 * the # of channels that host offered.
584 if (sc->hn_tx_ring_inuse > ring_cnt)
585 sc->hn_tx_ring_inuse = ring_cnt;
586 sc->hn_rx_ring_inuse = ring_cnt;
587 device_printf(dev, "%d TX ring, %d RX ring\n",
588 sc->hn_tx_ring_inuse, sc->hn_rx_ring_inuse);
590 if (sc->hn_rx_ring_inuse > 1)
591 hn_subchan_setup(sc);
593 #if __FreeBSD_version >= 1100099
594 if (sc->hn_rx_ring_inuse > 1) {
596 * Reduce TCP segment aggregation limit for multiple
597 * RX rings to increase ACK timeliness.
599 hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MULTIRX_DEF);
603 if (device_info.link_state == 0) {
607 tso_maxlen = hn_tso_maxlen;
608 if (tso_maxlen <= 0 || tso_maxlen > IP_MAXPACKET)
609 tso_maxlen = IP_MAXPACKET;
611 ifp->if_hw_tsomaxsegcount = HN_TX_DATA_SEGCNT_MAX;
612 ifp->if_hw_tsomaxsegsize = PAGE_SIZE;
613 ifp->if_hw_tsomax = tso_maxlen -
614 (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
616 ether_ifattach(ifp, device_info.mac_addr);
618 if_printf(ifp, "TSO: %u/%u/%u\n", ifp->if_hw_tsomax,
619 ifp->if_hw_tsomaxsegcount, ifp->if_hw_tsomaxsegsize);
621 hn_set_chim_size(sc, sc->hn_chim_szmax);
622 if (hn_tx_chimney_size > 0 &&
623 hn_tx_chimney_size < sc->hn_chim_szmax)
624 hn_set_chim_size(sc, hn_tx_chimney_size);
626 SYSCTL_ADD_UINT(device_get_sysctl_ctx(dev),
627 SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), OID_AUTO,
628 "nvs_version", CTLFLAG_RD, &sc->hn_nvs_ver, 0, "NVS version");
632 hn_destroy_tx_data(sc);
639 * Standard detach entry point
642 netvsc_detach(device_t dev)
644 struct hn_softc *sc = device_get_softc(dev);
647 printf("netvsc_detach\n");
650 * XXXKYS: Need to clean up all our
651 * driver state; this is the driver
656 * XXXKYS: Need to stop outgoing traffic and unregister
660 hv_rf_on_device_remove(sc, HV_RF_NV_DESTROY_CHANNEL);
662 hn_stop_tx_tasks(sc);
664 ifmedia_removeall(&sc->hn_media);
665 hn_destroy_rx_data(sc);
666 hn_destroy_tx_data(sc);
668 if (sc->hn_tx_taskq != hn_tx_taskq)
669 taskqueue_free(sc->hn_tx_taskq);
671 vmbus_xact_ctx_destroy(sc->hn_xact);
676 * Standard shutdown entry point
679 netvsc_shutdown(device_t dev)
685 hn_txdesc_dmamap_load(struct hn_tx_ring *txr, struct hn_txdesc *txd,
686 struct mbuf **m_head, bus_dma_segment_t *segs, int *nsegs)
688 struct mbuf *m = *m_head;
691 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag, txd->data_dmap,
692 m, segs, nsegs, BUS_DMA_NOWAIT);
693 if (error == EFBIG) {
696 m_new = m_collapse(m, M_NOWAIT, HN_TX_DATA_SEGCNT_MAX);
701 txr->hn_tx_collapsed++;
703 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag,
704 txd->data_dmap, m, segs, nsegs, BUS_DMA_NOWAIT);
707 bus_dmamap_sync(txr->hn_tx_data_dtag, txd->data_dmap,
708 BUS_DMASYNC_PREWRITE);
709 txd->flags |= HN_TXD_FLAG_DMAMAP;
715 hn_txdesc_dmamap_unload(struct hn_tx_ring *txr, struct hn_txdesc *txd)
718 if (txd->flags & HN_TXD_FLAG_DMAMAP) {
719 bus_dmamap_sync(txr->hn_tx_data_dtag,
720 txd->data_dmap, BUS_DMASYNC_POSTWRITE);
721 bus_dmamap_unload(txr->hn_tx_data_dtag,
723 txd->flags &= ~HN_TXD_FLAG_DMAMAP;
728 hn_txdesc_put(struct hn_tx_ring *txr, struct hn_txdesc *txd)
731 KASSERT((txd->flags & HN_TXD_FLAG_ONLIST) == 0,
732 ("put an onlist txd %#x", txd->flags));
734 KASSERT(txd->refs > 0, ("invalid txd refs %d", txd->refs));
735 if (atomic_fetchadd_int(&txd->refs, -1) != 1)
738 hn_txdesc_dmamap_unload(txr, txd);
739 if (txd->m != NULL) {
744 txd->flags |= HN_TXD_FLAG_ONLIST;
746 #ifndef HN_USE_TXDESC_BUFRING
747 mtx_lock_spin(&txr->hn_txlist_spin);
748 KASSERT(txr->hn_txdesc_avail >= 0 &&
749 txr->hn_txdesc_avail < txr->hn_txdesc_cnt,
750 ("txdesc_put: invalid txd avail %d", txr->hn_txdesc_avail));
751 txr->hn_txdesc_avail++;
752 SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
753 mtx_unlock_spin(&txr->hn_txlist_spin);
755 atomic_add_int(&txr->hn_txdesc_avail, 1);
756 buf_ring_enqueue(txr->hn_txdesc_br, txd);
762 static __inline struct hn_txdesc *
763 hn_txdesc_get(struct hn_tx_ring *txr)
765 struct hn_txdesc *txd;
767 #ifndef HN_USE_TXDESC_BUFRING
768 mtx_lock_spin(&txr->hn_txlist_spin);
769 txd = SLIST_FIRST(&txr->hn_txlist);
771 KASSERT(txr->hn_txdesc_avail > 0,
772 ("txdesc_get: invalid txd avail %d", txr->hn_txdesc_avail));
773 txr->hn_txdesc_avail--;
774 SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
776 mtx_unlock_spin(&txr->hn_txlist_spin);
778 txd = buf_ring_dequeue_sc(txr->hn_txdesc_br);
782 #ifdef HN_USE_TXDESC_BUFRING
783 atomic_subtract_int(&txr->hn_txdesc_avail, 1);
785 KASSERT(txd->m == NULL && txd->refs == 0 &&
786 (txd->flags & HN_TXD_FLAG_ONLIST), ("invalid txd"));
787 txd->flags &= ~HN_TXD_FLAG_ONLIST;
794 hn_txdesc_hold(struct hn_txdesc *txd)
797 /* 0->1 transition will never work */
798 KASSERT(txd->refs > 0, ("invalid refs %d", txd->refs));
799 atomic_add_int(&txd->refs, 1);
803 hn_txeof(struct hn_tx_ring *txr)
805 txr->hn_has_txeof = 0;
810 hn_tx_done(struct hn_send_ctx *sndc, struct hn_softc *sc,
811 struct vmbus_channel *chan, const void *data __unused, int dlen __unused)
813 struct hn_txdesc *txd = sndc->hn_cbarg;
814 struct hn_tx_ring *txr;
816 if (sndc->hn_chim_idx != HN_NVS_CHIM_IDX_INVALID)
817 hn_chim_free(sc, sndc->hn_chim_idx);
820 KASSERT(txr->hn_chan == chan,
821 ("channel mismatch, on chan%u, should be chan%u",
822 vmbus_chan_subidx(chan), vmbus_chan_subidx(txr->hn_chan)));
824 txr->hn_has_txeof = 1;
825 hn_txdesc_put(txr, txd);
827 ++txr->hn_txdone_cnt;
828 if (txr->hn_txdone_cnt >= HN_EARLY_TXEOF_THRESH) {
829 txr->hn_txdone_cnt = 0;
836 netvsc_channel_rollup(struct hn_rx_ring *rxr, struct hn_tx_ring *txr)
838 #if defined(INET) || defined(INET6)
839 struct lro_ctrl *lro = &rxr->hn_lro;
840 struct lro_entry *queued;
842 while ((queued = SLIST_FIRST(&lro->lro_active)) != NULL) {
843 SLIST_REMOVE_HEAD(&lro->lro_active, next);
844 tcp_lro_flush(lro, queued);
850 * 'txr' could be NULL, if multiple channels and
851 * ifnet.if_start method are enabled.
853 if (txr == NULL || !txr->hn_has_txeof)
856 txr->hn_txdone_cnt = 0;
862 * If this function fails, then both txd and m_head0 will be freed.
865 hn_encap(struct hn_tx_ring *txr, struct hn_txdesc *txd, struct mbuf **m_head0)
867 bus_dma_segment_t segs[HN_TX_DATA_SEGCNT_MAX];
869 struct mbuf *m_head = *m_head0;
870 rndis_msg *rndis_mesg;
871 rndis_packet *rndis_pkt;
872 rndis_per_packet_info *rppi;
873 struct rndis_hash_value *hash_value;
874 uint32_t rndis_msg_size, tot_data_buf_len, send_buf_section_idx;
875 int send_buf_section_size;
877 tot_data_buf_len = m_head->m_pkthdr.len;
880 * extension points to the area reserved for the
881 * rndis_filter_packet, which is placed just after
882 * the netvsc_packet (and rppi struct, if present;
883 * length is updated later).
885 rndis_mesg = txd->rndis_msg;
886 /* XXX not necessary */
887 memset(rndis_mesg, 0, HN_RNDIS_MSG_LEN);
888 rndis_mesg->ndis_msg_type = REMOTE_NDIS_PACKET_MSG;
890 rndis_pkt = &rndis_mesg->msg.packet;
891 rndis_pkt->data_offset = sizeof(rndis_packet);
892 rndis_pkt->data_length = tot_data_buf_len;
893 rndis_pkt->per_pkt_info_offset = sizeof(rndis_packet);
895 rndis_msg_size = RNDIS_MESSAGE_SIZE(rndis_packet);
898 * Set the hash value for this packet, so that the host could
899 * dispatch the TX done event for this packet back to this TX
902 rndis_msg_size += RNDIS_HASHVAL_PPI_SIZE;
903 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_HASHVAL_PPI_SIZE,
905 hash_value = (struct rndis_hash_value *)((uint8_t *)rppi +
906 rppi->per_packet_info_offset);
907 hash_value->hash_value = txr->hn_tx_idx;
909 if (m_head->m_flags & M_VLANTAG) {
910 ndis_8021q_info *rppi_vlan_info;
912 rndis_msg_size += RNDIS_VLAN_PPI_SIZE;
913 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_VLAN_PPI_SIZE,
916 rppi_vlan_info = (ndis_8021q_info *)((uint8_t *)rppi +
917 rppi->per_packet_info_offset);
918 rppi_vlan_info->u1.s1.vlan_id =
919 m_head->m_pkthdr.ether_vtag & 0xfff;
922 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
923 rndis_tcp_tso_info *tso_info;
924 struct ether_vlan_header *eh;
928 * XXX need m_pullup and use mtodo
930 eh = mtod(m_head, struct ether_vlan_header*);
931 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN))
932 ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
934 ether_len = ETHER_HDR_LEN;
936 rndis_msg_size += RNDIS_TSO_PPI_SIZE;
937 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_TSO_PPI_SIZE,
938 tcp_large_send_info);
940 tso_info = (rndis_tcp_tso_info *)((uint8_t *)rppi +
941 rppi->per_packet_info_offset);
942 tso_info->lso_v2_xmit.type =
943 RNDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
946 if (m_head->m_pkthdr.csum_flags & CSUM_IP_TSO) {
948 (struct ip *)(m_head->m_data + ether_len);
949 unsigned long iph_len = ip->ip_hl << 2;
951 (struct tcphdr *)((caddr_t)ip + iph_len);
953 tso_info->lso_v2_xmit.ip_version =
954 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
958 th->th_sum = in_pseudo(ip->ip_src.s_addr,
959 ip->ip_dst.s_addr, htons(IPPROTO_TCP));
962 #if defined(INET6) && defined(INET)
967 struct ip6_hdr *ip6 = (struct ip6_hdr *)
968 (m_head->m_data + ether_len);
969 struct tcphdr *th = (struct tcphdr *)(ip6 + 1);
971 tso_info->lso_v2_xmit.ip_version =
972 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
974 th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
977 tso_info->lso_v2_xmit.tcp_header_offset = 0;
978 tso_info->lso_v2_xmit.mss = m_head->m_pkthdr.tso_segsz;
979 } else if (m_head->m_pkthdr.csum_flags & txr->hn_csum_assist) {
980 rndis_tcp_ip_csum_info *csum_info;
982 rndis_msg_size += RNDIS_CSUM_PPI_SIZE;
983 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_CSUM_PPI_SIZE,
985 csum_info = (rndis_tcp_ip_csum_info *)((uint8_t *)rppi +
986 rppi->per_packet_info_offset);
988 csum_info->xmit.is_ipv4 = 1;
989 if (m_head->m_pkthdr.csum_flags & CSUM_IP)
990 csum_info->xmit.ip_header_csum = 1;
992 if (m_head->m_pkthdr.csum_flags & CSUM_TCP) {
993 csum_info->xmit.tcp_csum = 1;
994 csum_info->xmit.tcp_header_offset = 0;
995 } else if (m_head->m_pkthdr.csum_flags & CSUM_UDP) {
996 csum_info->xmit.udp_csum = 1;
1000 rndis_mesg->msg_len = tot_data_buf_len + rndis_msg_size;
1001 tot_data_buf_len = rndis_mesg->msg_len;
1004 * Chimney send, if the packet could fit into one chimney buffer.
1006 if (tot_data_buf_len < txr->hn_chim_size) {
1007 txr->hn_tx_chimney_tried++;
1008 send_buf_section_idx = hn_chim_alloc(txr->hn_sc);
1009 if (send_buf_section_idx != HN_NVS_CHIM_IDX_INVALID) {
1010 uint8_t *dest = txr->hn_sc->hn_chim +
1011 (send_buf_section_idx * txr->hn_sc->hn_chim_szmax);
1013 memcpy(dest, rndis_mesg, rndis_msg_size);
1014 dest += rndis_msg_size;
1015 m_copydata(m_head, 0, m_head->m_pkthdr.len, dest);
1017 send_buf_section_size = tot_data_buf_len;
1018 txr->hn_gpa_cnt = 0;
1019 txr->hn_tx_chimney++;
1024 error = hn_txdesc_dmamap_load(txr, txd, &m_head, segs, &nsegs);
1029 * This mbuf is not linked w/ the txd yet, so free it now.
1034 freed = hn_txdesc_put(txr, txd);
1036 ("fail to free txd upon txdma error"));
1038 txr->hn_txdma_failed++;
1039 if_inc_counter(txr->hn_sc->hn_ifp, IFCOUNTER_OERRORS, 1);
1044 txr->hn_gpa_cnt = nsegs + HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
1046 /* send packet with page buffer */
1047 txr->hn_gpa[0].gpa_page = atop(txd->rndis_msg_paddr);
1048 txr->hn_gpa[0].gpa_ofs = txd->rndis_msg_paddr & PAGE_MASK;
1049 txr->hn_gpa[0].gpa_len = rndis_msg_size;
1052 * Fill the page buffers with mbuf info starting at index
1053 * HV_RF_NUM_TX_RESERVED_PAGE_BUFS.
1055 for (i = 0; i < nsegs; ++i) {
1056 struct vmbus_gpa *gpa = &txr->hn_gpa[
1057 i + HV_RF_NUM_TX_RESERVED_PAGE_BUFS];
1059 gpa->gpa_page = atop(segs[i].ds_addr);
1060 gpa->gpa_ofs = segs[i].ds_addr & PAGE_MASK;
1061 gpa->gpa_len = segs[i].ds_len;
1064 send_buf_section_idx = HN_NVS_CHIM_IDX_INVALID;
1065 send_buf_section_size = 0;
1069 /* Set the completion routine */
1070 hn_send_ctx_init(&txd->send_ctx, hn_tx_done, txd,
1071 send_buf_section_idx, send_buf_section_size);
1078 * If this function fails, then txd will be freed, but the mbuf
1079 * associated w/ the txd will _not_ be freed.
1082 hn_send_pkt(struct ifnet *ifp, struct hn_tx_ring *txr, struct hn_txdesc *txd)
1084 int error, send_failed = 0;
1088 * Make sure that txd is not freed before ETHER_BPF_MTAP.
1090 hn_txdesc_hold(txd);
1091 error = hv_nv_on_send(txr->hn_chan, HN_NVS_RNDIS_MTYPE_DATA,
1092 &txd->send_ctx, txr->hn_gpa, txr->hn_gpa_cnt);
1094 ETHER_BPF_MTAP(ifp, txd->m);
1095 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1096 if (!hn_use_if_start) {
1097 if_inc_counter(ifp, IFCOUNTER_OBYTES,
1098 txd->m->m_pkthdr.len);
1099 if (txd->m->m_flags & M_MCAST)
1100 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
1104 hn_txdesc_put(txr, txd);
1106 if (__predict_false(error)) {
1110 * This should "really rarely" happen.
1112 * XXX Too many RX to be acked or too many sideband
1113 * commands to run? Ask netvsc_channel_rollup()
1114 * to kick start later.
1116 txr->hn_has_txeof = 1;
1118 txr->hn_send_failed++;
1121 * Try sending again after set hn_has_txeof;
1122 * in case that we missed the last
1123 * netvsc_channel_rollup().
1127 if_printf(ifp, "send failed\n");
1130 * Caller will perform further processing on the
1131 * associated mbuf, so don't free it in hn_txdesc_put();
1132 * only unload it from the DMA map in hn_txdesc_put(),
1136 freed = hn_txdesc_put(txr, txd);
1138 ("fail to free txd upon send error"));
1140 txr->hn_send_failed++;
1146 * Start a transmit of one or more packets
1149 hn_start_locked(struct hn_tx_ring *txr, int len)
1151 struct hn_softc *sc = txr->hn_sc;
1152 struct ifnet *ifp = sc->hn_ifp;
1154 KASSERT(hn_use_if_start,
1155 ("hn_start_locked is called, when if_start is disabled"));
1156 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1157 mtx_assert(&txr->hn_tx_lock, MA_OWNED);
1159 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1163 while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
1164 struct hn_txdesc *txd;
1165 struct mbuf *m_head;
1168 IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
1172 if (len > 0 && m_head->m_pkthdr.len > len) {
1174 * This sending could be time consuming; let callers
1175 * dispatch this packet sending (and sending of any
1176 * following up packets) to tx taskqueue.
1178 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1182 txd = hn_txdesc_get(txr);
1184 txr->hn_no_txdescs++;
1185 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1186 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1190 error = hn_encap(txr, txd, &m_head);
1192 /* Both txd and m_head are freed */
1196 error = hn_send_pkt(ifp, txr, txd);
1197 if (__predict_false(error)) {
1198 /* txd is freed, but m_head is not */
1199 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1200 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1208 * Link up/down notification
1211 netvsc_linkstatus_callback(struct hn_softc *sc, uint32_t status)
1221 * Append the specified data to the indicated mbuf chain,
1222 * Extend the mbuf chain if the new data does not fit in
1225 * This is a minor rewrite of m_append() from sys/kern/uipc_mbuf.c.
1226 * There should be an equivalent in the kernel mbuf code,
1227 * but there does not appear to be one yet.
1229 * Differs from m_append() in that additional mbufs are
1230 * allocated with cluster size MJUMPAGESIZE, and filled
1233 * Return 1 if able to complete the job; otherwise 0.
1236 hv_m_append(struct mbuf *m0, int len, c_caddr_t cp)
1239 int remainder, space;
1241 for (m = m0; m->m_next != NULL; m = m->m_next)
1244 space = M_TRAILINGSPACE(m);
1247 * Copy into available space.
1249 if (space > remainder)
1251 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1256 while (remainder > 0) {
1258 * Allocate a new mbuf; could check space
1259 * and allocate a cluster instead.
1261 n = m_getjcl(M_DONTWAIT, m->m_type, 0, MJUMPAGESIZE);
1264 n->m_len = min(MJUMPAGESIZE, remainder);
1265 bcopy(cp, mtod(n, caddr_t), n->m_len);
1267 remainder -= n->m_len;
1271 if (m0->m_flags & M_PKTHDR)
1272 m0->m_pkthdr.len += len - remainder;
1274 return (remainder == 0);
1277 #if defined(INET) || defined(INET6)
1279 hn_lro_rx(struct lro_ctrl *lc, struct mbuf *m)
1281 #if __FreeBSD_version >= 1100095
1282 if (hn_lro_mbufq_depth) {
1283 tcp_lro_queue_mbuf(lc, m);
1287 return tcp_lro_rx(lc, m, 0);
1292 * Called when we receive a data packet from the "wire" on the
1295 * Note: This is no longer used as a callback
1298 netvsc_recv(struct hn_rx_ring *rxr, const void *data, int dlen,
1299 const struct hn_recvinfo *info)
1301 struct ifnet *ifp = rxr->hn_ifp;
1303 int size, do_lro = 0, do_csum = 1;
1305 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1309 * Bail out if packet contains more data than configured MTU.
1311 if (dlen > (ifp->if_mtu + ETHER_HDR_LEN)) {
1313 } else if (dlen <= MHLEN) {
1314 m_new = m_gethdr(M_NOWAIT, MT_DATA);
1315 if (m_new == NULL) {
1316 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1319 memcpy(mtod(m_new, void *), data, dlen);
1320 m_new->m_pkthdr.len = m_new->m_len = dlen;
1321 rxr->hn_small_pkts++;
1324 * Get an mbuf with a cluster. For packets 2K or less,
1325 * get a standard 2K cluster. For anything larger, get a
1326 * 4K cluster. Any buffers larger than 4K can cause problems
1327 * if looped around to the Hyper-V TX channel, so avoid them.
1330 if (dlen > MCLBYTES) {
1332 size = MJUMPAGESIZE;
1335 m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, size);
1336 if (m_new == NULL) {
1337 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1341 hv_m_append(m_new, dlen, data);
1343 m_new->m_pkthdr.rcvif = ifp;
1345 if (__predict_false((ifp->if_capenable & IFCAP_RXCSUM) == 0))
1348 /* receive side checksum offload */
1349 if (info->csum_info != NULL) {
1350 /* IP csum offload */
1351 if (info->csum_info->receive.ip_csum_succeeded && do_csum) {
1352 m_new->m_pkthdr.csum_flags |=
1353 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1357 /* TCP/UDP csum offload */
1358 if ((info->csum_info->receive.tcp_csum_succeeded ||
1359 info->csum_info->receive.udp_csum_succeeded) && do_csum) {
1360 m_new->m_pkthdr.csum_flags |=
1361 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1362 m_new->m_pkthdr.csum_data = 0xffff;
1363 if (info->csum_info->receive.tcp_csum_succeeded)
1369 if (info->csum_info->receive.ip_csum_succeeded &&
1370 info->csum_info->receive.tcp_csum_succeeded)
1373 const struct ether_header *eh;
1378 if (m_new->m_len < hoff)
1380 eh = mtod(m_new, struct ether_header *);
1381 etype = ntohs(eh->ether_type);
1382 if (etype == ETHERTYPE_VLAN) {
1383 const struct ether_vlan_header *evl;
1385 hoff = sizeof(*evl);
1386 if (m_new->m_len < hoff)
1388 evl = mtod(m_new, struct ether_vlan_header *);
1389 etype = ntohs(evl->evl_proto);
1392 if (etype == ETHERTYPE_IP) {
1395 pr = hn_check_iplen(m_new, hoff);
1396 if (pr == IPPROTO_TCP) {
1398 (rxr->hn_trust_hcsum &
1399 HN_TRUST_HCSUM_TCP)) {
1400 rxr->hn_csum_trusted++;
1401 m_new->m_pkthdr.csum_flags |=
1402 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1403 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1404 m_new->m_pkthdr.csum_data = 0xffff;
1407 } else if (pr == IPPROTO_UDP) {
1409 (rxr->hn_trust_hcsum &
1410 HN_TRUST_HCSUM_UDP)) {
1411 rxr->hn_csum_trusted++;
1412 m_new->m_pkthdr.csum_flags |=
1413 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1414 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1415 m_new->m_pkthdr.csum_data = 0xffff;
1417 } else if (pr != IPPROTO_DONE && do_csum &&
1418 (rxr->hn_trust_hcsum & HN_TRUST_HCSUM_IP)) {
1419 rxr->hn_csum_trusted++;
1420 m_new->m_pkthdr.csum_flags |=
1421 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1426 if (info->vlan_info != NULL) {
1427 m_new->m_pkthdr.ether_vtag = info->vlan_info->u1.s1.vlan_id;
1428 m_new->m_flags |= M_VLANTAG;
1431 if (info->hash_info != NULL && info->hash_value != NULL) {
1432 int hash_type = M_HASHTYPE_OPAQUE;
1435 m_new->m_pkthdr.flowid = info->hash_value->hash_value;
1436 if ((info->hash_info->hash_info & NDIS_HASH_FUNCTION_MASK) ==
1437 NDIS_HASH_FUNCTION_TOEPLITZ) {
1439 (info->hash_info->hash_info & NDIS_HASH_TYPE_MASK);
1442 case NDIS_HASH_IPV4:
1443 hash_type = M_HASHTYPE_RSS_IPV4;
1446 case NDIS_HASH_TCP_IPV4:
1447 hash_type = M_HASHTYPE_RSS_TCP_IPV4;
1450 case NDIS_HASH_IPV6:
1451 hash_type = M_HASHTYPE_RSS_IPV6;
1454 case NDIS_HASH_IPV6_EX:
1455 hash_type = M_HASHTYPE_RSS_IPV6_EX;
1458 case NDIS_HASH_TCP_IPV6:
1459 hash_type = M_HASHTYPE_RSS_TCP_IPV6;
1462 case NDIS_HASH_TCP_IPV6_EX:
1463 hash_type = M_HASHTYPE_RSS_TCP_IPV6_EX;
1467 M_HASHTYPE_SET(m_new, hash_type);
1469 if (info->hash_value != NULL)
1470 m_new->m_pkthdr.flowid = info->hash_value->hash_value;
1472 m_new->m_pkthdr.flowid = rxr->hn_rx_idx;
1473 M_HASHTYPE_SET(m_new, M_HASHTYPE_OPAQUE);
1477 * Note: Moved RX completion back to hv_nv_on_receive() so all
1478 * messages (not just data messages) will trigger a response.
1484 if ((ifp->if_capenable & IFCAP_LRO) && do_lro) {
1485 #if defined(INET) || defined(INET6)
1486 struct lro_ctrl *lro = &rxr->hn_lro;
1489 rxr->hn_lro_tried++;
1490 if (hn_lro_rx(lro, m_new) == 0) {
1498 /* We're not holding the lock here, so don't release it */
1499 (*ifp->if_input)(ifp, m_new);
1505 * Rules for using sc->temp_unusable:
1506 * 1. sc->temp_unusable can only be read or written while holding NV_LOCK()
1507 * 2. code reading sc->temp_unusable under NV_LOCK(), and finding
1508 * sc->temp_unusable set, must release NV_LOCK() and exit
1509 * 3. to retain exclusive control of the interface,
1510 * sc->temp_unusable must be set by code before releasing NV_LOCK()
1511 * 4. only code setting sc->temp_unusable can clear sc->temp_unusable
1512 * 5. code setting sc->temp_unusable must eventually clear sc->temp_unusable
1516 * Standard ioctl entry point. Called when the user wants to configure
1520 hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1522 hn_softc_t *sc = ifp->if_softc;
1523 struct ifreq *ifr = (struct ifreq *)data;
1525 struct ifaddr *ifa = (struct ifaddr *)data;
1527 netvsc_device_info device_info;
1528 int mask, error = 0, ring_cnt;
1529 int retry_cnt = 500;
1535 if (ifa->ifa_addr->sa_family == AF_INET) {
1536 ifp->if_flags |= IFF_UP;
1537 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1539 arp_ifinit(ifp, ifa);
1542 error = ether_ioctl(ifp, cmd, data);
1545 /* Check MTU value change */
1546 if (ifp->if_mtu == ifr->ifr_mtu)
1549 if (ifr->ifr_mtu > NETVSC_MAX_CONFIGURABLE_MTU) {
1554 /* Obtain and record requested MTU */
1555 ifp->if_mtu = ifr->ifr_mtu;
1557 #if __FreeBSD_version >= 1100099
1559 * Make sure that LRO aggregation length limit is still
1560 * valid, after the MTU change.
1563 if (sc->hn_rx_ring[0].hn_lro.lro_length_lim <
1564 HN_LRO_LENLIM_MIN(ifp))
1565 hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MIN(ifp));
1571 if (!sc->temp_unusable) {
1572 sc->temp_unusable = TRUE;
1576 if (retry_cnt > 0) {
1580 } while (retry_cnt > 0);
1582 if (retry_cnt == 0) {
1587 /* We must remove and add back the device to cause the new
1588 * MTU to take effect. This includes tearing down, but not
1589 * deleting the channel, then bringing it back up.
1591 error = hv_rf_on_device_remove(sc, HV_RF_NV_RETAIN_CHANNEL);
1594 sc->temp_unusable = FALSE;
1599 /* Wait for subchannels to be destroyed */
1600 vmbus_subchan_drain(sc->hn_prichan);
1602 ring_cnt = sc->hn_rx_ring_inuse;
1603 error = hv_rf_on_device_add(sc, &device_info, &ring_cnt,
1604 &sc->hn_rx_ring[0]);
1607 sc->temp_unusable = FALSE;
1611 /* # of channels can _not_ be changed */
1612 KASSERT(sc->hn_rx_ring_inuse == ring_cnt,
1613 ("RX ring count %d and channel count %u mismatch",
1614 sc->hn_rx_ring_cnt, ring_cnt));
1615 if (sc->hn_rx_ring_inuse > 1) {
1619 * Skip the rings on primary channel; they are
1620 * handled by the hv_rf_on_device_add() above.
1622 for (r = 1; r < sc->hn_rx_ring_cnt; ++r) {
1623 sc->hn_rx_ring[r].hn_rx_flags &=
1624 ~HN_RX_FLAG_ATTACHED;
1626 for (r = 1; r < sc->hn_tx_ring_cnt; ++r) {
1627 sc->hn_tx_ring[r].hn_tx_flags &=
1628 ~HN_TX_FLAG_ATTACHED;
1630 hn_subchan_setup(sc);
1633 if (sc->hn_tx_ring[0].hn_chim_size > sc->hn_chim_szmax)
1634 hn_set_chim_size(sc, sc->hn_chim_szmax);
1636 hn_ifinit_locked(sc);
1639 sc->temp_unusable = FALSE;
1645 if (!sc->temp_unusable) {
1646 sc->temp_unusable = TRUE;
1650 if (retry_cnt > 0) {
1654 } while (retry_cnt > 0);
1656 if (retry_cnt == 0) {
1661 if (ifp->if_flags & IFF_UP) {
1663 * If only the state of the PROMISC flag changed,
1664 * then just use the 'set promisc mode' command
1665 * instead of reinitializing the entire NIC. Doing
1666 * a full re-init means reloading the firmware and
1667 * waiting for it to start up, which may take a
1671 /* Fixme: Promiscuous mode? */
1672 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1673 ifp->if_flags & IFF_PROMISC &&
1674 !(sc->hn_if_flags & IFF_PROMISC)) {
1675 /* do something here for Hyper-V */
1676 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1677 !(ifp->if_flags & IFF_PROMISC) &&
1678 sc->hn_if_flags & IFF_PROMISC) {
1679 /* do something here for Hyper-V */
1682 hn_ifinit_locked(sc);
1684 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1689 sc->temp_unusable = FALSE;
1691 sc->hn_if_flags = ifp->if_flags;
1697 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1698 if (mask & IFCAP_TXCSUM) {
1699 ifp->if_capenable ^= IFCAP_TXCSUM;
1700 if (ifp->if_capenable & IFCAP_TXCSUM) {
1702 sc->hn_tx_ring[0].hn_csum_assist;
1705 ~sc->hn_tx_ring[0].hn_csum_assist;
1709 if (mask & IFCAP_RXCSUM)
1710 ifp->if_capenable ^= IFCAP_RXCSUM;
1712 if (mask & IFCAP_LRO)
1713 ifp->if_capenable ^= IFCAP_LRO;
1715 if (mask & IFCAP_TSO4) {
1716 ifp->if_capenable ^= IFCAP_TSO4;
1717 if (ifp->if_capenable & IFCAP_TSO4)
1718 ifp->if_hwassist |= CSUM_IP_TSO;
1720 ifp->if_hwassist &= ~CSUM_IP_TSO;
1723 if (mask & IFCAP_TSO6) {
1724 ifp->if_capenable ^= IFCAP_TSO6;
1725 if (ifp->if_capenable & IFCAP_TSO6)
1726 ifp->if_hwassist |= CSUM_IP6_TSO;
1728 ifp->if_hwassist &= ~CSUM_IP6_TSO;
1737 /* Fixme: Multicast mode? */
1738 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1740 netvsc_setmulti(sc);
1749 error = ifmedia_ioctl(ifp, ifr, &sc->hn_media, cmd);
1752 error = ether_ioctl(ifp, cmd, data);
1763 hn_stop(hn_softc_t *sc)
1771 printf(" Closing Device ...\n");
1773 atomic_clear_int(&ifp->if_drv_flags,
1774 (IFF_DRV_RUNNING | IFF_DRV_OACTIVE));
1775 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
1776 sc->hn_tx_ring[i].hn_oactive = 0;
1778 if_link_state_change(ifp, LINK_STATE_DOWN);
1779 sc->hn_initdone = 0;
1781 ret = hv_rf_on_close(sc);
1785 * FreeBSD transmit entry point
1788 hn_start(struct ifnet *ifp)
1790 struct hn_softc *sc = ifp->if_softc;
1791 struct hn_tx_ring *txr = &sc->hn_tx_ring[0];
1793 if (txr->hn_sched_tx)
1796 if (mtx_trylock(&txr->hn_tx_lock)) {
1799 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1800 mtx_unlock(&txr->hn_tx_lock);
1805 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
1809 hn_start_txeof(struct hn_tx_ring *txr)
1811 struct hn_softc *sc = txr->hn_sc;
1812 struct ifnet *ifp = sc->hn_ifp;
1814 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1816 if (txr->hn_sched_tx)
1819 if (mtx_trylock(&txr->hn_tx_lock)) {
1822 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1823 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1824 mtx_unlock(&txr->hn_tx_lock);
1826 taskqueue_enqueue(txr->hn_tx_taskq,
1832 * Release the OACTIVE earlier, with the hope, that
1833 * others could catch up. The task will clear the
1834 * flag again with the hn_tx_lock to avoid possible
1837 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1838 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
1846 hn_ifinit_locked(hn_softc_t *sc)
1853 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1857 hv_promisc_mode = 1;
1859 ret = hv_rf_on_open(sc);
1863 sc->hn_initdone = 1;
1866 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1867 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
1868 sc->hn_tx_ring[i].hn_oactive = 0;
1870 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_RUNNING);
1871 if_link_state_change(ifp, LINK_STATE_UP);
1878 hn_ifinit(void *xsc)
1880 hn_softc_t *sc = xsc;
1883 if (sc->temp_unusable) {
1887 sc->temp_unusable = TRUE;
1890 hn_ifinit_locked(sc);
1893 sc->temp_unusable = FALSE;
1902 hn_watchdog(struct ifnet *ifp)
1907 printf("hn%d: watchdog timeout -- resetting\n", sc->hn_unit);
1908 hn_ifinit(sc); /*???*/
1913 #if __FreeBSD_version >= 1100099
1916 hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS)
1918 struct hn_softc *sc = arg1;
1919 unsigned int lenlim;
1922 lenlim = sc->hn_rx_ring[0].hn_lro.lro_length_lim;
1923 error = sysctl_handle_int(oidp, &lenlim, 0, req);
1924 if (error || req->newptr == NULL)
1927 if (lenlim < HN_LRO_LENLIM_MIN(sc->hn_ifp) ||
1928 lenlim > TCP_LRO_LENGTH_MAX)
1932 hn_set_lro_lenlim(sc, lenlim);
1938 hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS)
1940 struct hn_softc *sc = arg1;
1941 int ackcnt, error, i;
1944 * lro_ackcnt_lim is append count limit,
1945 * +1 to turn it into aggregation limit.
1947 ackcnt = sc->hn_rx_ring[0].hn_lro.lro_ackcnt_lim + 1;
1948 error = sysctl_handle_int(oidp, &ackcnt, 0, req);
1949 if (error || req->newptr == NULL)
1952 if (ackcnt < 2 || ackcnt > (TCP_LRO_ACKCNT_MAX + 1))
1956 * Convert aggregation limit back to append
1961 for (i = 0; i < sc->hn_rx_ring_inuse; ++i)
1962 sc->hn_rx_ring[i].hn_lro.lro_ackcnt_lim = ackcnt;
1970 hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS)
1972 struct hn_softc *sc = arg1;
1977 if (sc->hn_rx_ring[0].hn_trust_hcsum & hcsum)
1980 error = sysctl_handle_int(oidp, &on, 0, req);
1981 if (error || req->newptr == NULL)
1985 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1986 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
1989 rxr->hn_trust_hcsum |= hcsum;
1991 rxr->hn_trust_hcsum &= ~hcsum;
1998 hn_chim_size_sysctl(SYSCTL_HANDLER_ARGS)
2000 struct hn_softc *sc = arg1;
2001 int chim_size, error;
2003 chim_size = sc->hn_tx_ring[0].hn_chim_size;
2004 error = sysctl_handle_int(oidp, &chim_size, 0, req);
2005 if (error || req->newptr == NULL)
2008 if (chim_size > sc->hn_chim_szmax || chim_size <= 0)
2011 hn_set_chim_size(sc, chim_size);
2015 #if __FreeBSD_version < 1100095
2017 hn_rx_stat_int_sysctl(SYSCTL_HANDLER_ARGS)
2019 struct hn_softc *sc = arg1;
2020 int ofs = arg2, i, error;
2021 struct hn_rx_ring *rxr;
2025 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2026 rxr = &sc->hn_rx_ring[i];
2027 stat += *((int *)((uint8_t *)rxr + ofs));
2030 error = sysctl_handle_64(oidp, &stat, 0, req);
2031 if (error || req->newptr == NULL)
2034 /* Zero out this stat. */
2035 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2036 rxr = &sc->hn_rx_ring[i];
2037 *((int *)((uint8_t *)rxr + ofs)) = 0;
2043 hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS)
2045 struct hn_softc *sc = arg1;
2046 int ofs = arg2, i, error;
2047 struct hn_rx_ring *rxr;
2051 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2052 rxr = &sc->hn_rx_ring[i];
2053 stat += *((uint64_t *)((uint8_t *)rxr + ofs));
2056 error = sysctl_handle_64(oidp, &stat, 0, req);
2057 if (error || req->newptr == NULL)
2060 /* Zero out this stat. */
2061 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2062 rxr = &sc->hn_rx_ring[i];
2063 *((uint64_t *)((uint8_t *)rxr + ofs)) = 0;
2071 hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
2073 struct hn_softc *sc = arg1;
2074 int ofs = arg2, i, error;
2075 struct hn_rx_ring *rxr;
2079 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2080 rxr = &sc->hn_rx_ring[i];
2081 stat += *((u_long *)((uint8_t *)rxr + ofs));
2084 error = sysctl_handle_long(oidp, &stat, 0, req);
2085 if (error || req->newptr == NULL)
2088 /* Zero out this stat. */
2089 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2090 rxr = &sc->hn_rx_ring[i];
2091 *((u_long *)((uint8_t *)rxr + ofs)) = 0;
2097 hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
2099 struct hn_softc *sc = arg1;
2100 int ofs = arg2, i, error;
2101 struct hn_tx_ring *txr;
2105 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2106 txr = &sc->hn_tx_ring[i];
2107 stat += *((u_long *)((uint8_t *)txr + ofs));
2110 error = sysctl_handle_long(oidp, &stat, 0, req);
2111 if (error || req->newptr == NULL)
2114 /* Zero out this stat. */
2115 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2116 txr = &sc->hn_tx_ring[i];
2117 *((u_long *)((uint8_t *)txr + ofs)) = 0;
2123 hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS)
2125 struct hn_softc *sc = arg1;
2126 int ofs = arg2, i, error, conf;
2127 struct hn_tx_ring *txr;
2129 txr = &sc->hn_tx_ring[0];
2130 conf = *((int *)((uint8_t *)txr + ofs));
2132 error = sysctl_handle_int(oidp, &conf, 0, req);
2133 if (error || req->newptr == NULL)
2137 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2138 txr = &sc->hn_tx_ring[i];
2139 *((int *)((uint8_t *)txr + ofs)) = conf;
2147 hn_check_iplen(const struct mbuf *m, int hoff)
2149 const struct ip *ip;
2150 int len, iphlen, iplen;
2151 const struct tcphdr *th;
2152 int thoff; /* TCP data offset */
2154 len = hoff + sizeof(struct ip);
2156 /* The packet must be at least the size of an IP header. */
2157 if (m->m_pkthdr.len < len)
2158 return IPPROTO_DONE;
2160 /* The fixed IP header must reside completely in the first mbuf. */
2162 return IPPROTO_DONE;
2164 ip = mtodo(m, hoff);
2166 /* Bound check the packet's stated IP header length. */
2167 iphlen = ip->ip_hl << 2;
2168 if (iphlen < sizeof(struct ip)) /* minimum header length */
2169 return IPPROTO_DONE;
2171 /* The full IP header must reside completely in the one mbuf. */
2172 if (m->m_len < hoff + iphlen)
2173 return IPPROTO_DONE;
2175 iplen = ntohs(ip->ip_len);
2178 * Check that the amount of data in the buffers is as
2179 * at least much as the IP header would have us expect.
2181 if (m->m_pkthdr.len < hoff + iplen)
2182 return IPPROTO_DONE;
2185 * Ignore IP fragments.
2187 if (ntohs(ip->ip_off) & (IP_OFFMASK | IP_MF))
2188 return IPPROTO_DONE;
2191 * The TCP/IP or UDP/IP header must be entirely contained within
2192 * the first fragment of a packet.
2196 if (iplen < iphlen + sizeof(struct tcphdr))
2197 return IPPROTO_DONE;
2198 if (m->m_len < hoff + iphlen + sizeof(struct tcphdr))
2199 return IPPROTO_DONE;
2200 th = (const struct tcphdr *)((const uint8_t *)ip + iphlen);
2201 thoff = th->th_off << 2;
2202 if (thoff < sizeof(struct tcphdr) || thoff + iphlen > iplen)
2203 return IPPROTO_DONE;
2204 if (m->m_len < hoff + iphlen + thoff)
2205 return IPPROTO_DONE;
2208 if (iplen < iphlen + sizeof(struct udphdr))
2209 return IPPROTO_DONE;
2210 if (m->m_len < hoff + iphlen + sizeof(struct udphdr))
2211 return IPPROTO_DONE;
2215 return IPPROTO_DONE;
2222 hn_create_rx_data(struct hn_softc *sc, int ring_cnt)
2224 struct sysctl_oid_list *child;
2225 struct sysctl_ctx_list *ctx;
2226 device_t dev = sc->hn_dev;
2227 #if defined(INET) || defined(INET6)
2228 #if __FreeBSD_version >= 1100095
2235 * Create RXBUF for reception.
2238 * - It is shared by all channels.
2239 * - A large enough buffer is allocated, certain version of NVSes
2240 * may further limit the usable space.
2242 sc->hn_rxbuf = hyperv_dmamem_alloc(bus_get_dma_tag(dev),
2243 PAGE_SIZE, 0, NETVSC_RECEIVE_BUFFER_SIZE, &sc->hn_rxbuf_dma,
2244 BUS_DMA_WAITOK | BUS_DMA_ZERO);
2245 if (sc->hn_rxbuf == NULL) {
2246 device_printf(sc->hn_dev, "allocate rxbuf failed\n");
2250 sc->hn_rx_ring_cnt = ring_cnt;
2251 sc->hn_rx_ring_inuse = sc->hn_rx_ring_cnt;
2253 sc->hn_rx_ring = malloc(sizeof(struct hn_rx_ring) * sc->hn_rx_ring_cnt,
2254 M_NETVSC, M_WAITOK | M_ZERO);
2256 #if defined(INET) || defined(INET6)
2257 #if __FreeBSD_version >= 1100095
2258 lroent_cnt = hn_lro_entry_count;
2259 if (lroent_cnt < TCP_LRO_ENTRIES)
2260 lroent_cnt = TCP_LRO_ENTRIES;
2261 device_printf(dev, "LRO: entry count %d\n", lroent_cnt);
2263 #endif /* INET || INET6 */
2265 ctx = device_get_sysctl_ctx(dev);
2266 child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
2268 /* Create dev.hn.UNIT.rx sysctl tree */
2269 sc->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "rx",
2270 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2272 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2273 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
2275 if (hn_trust_hosttcp)
2276 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_TCP;
2277 if (hn_trust_hostudp)
2278 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_UDP;
2279 if (hn_trust_hostip)
2280 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_IP;
2281 rxr->hn_ifp = sc->hn_ifp;
2282 if (i < sc->hn_tx_ring_cnt)
2283 rxr->hn_txr = &sc->hn_tx_ring[i];
2284 rxr->hn_rdbuf = malloc(NETVSC_PACKET_SIZE, M_NETVSC, M_WAITOK);
2286 rxr->hn_rxbuf = sc->hn_rxbuf;
2291 #if defined(INET) || defined(INET6)
2292 #if __FreeBSD_version >= 1100095
2293 tcp_lro_init_args(&rxr->hn_lro, sc->hn_ifp, lroent_cnt,
2294 hn_lro_mbufq_depth);
2296 tcp_lro_init(&rxr->hn_lro);
2297 rxr->hn_lro.ifp = sc->hn_ifp;
2299 #if __FreeBSD_version >= 1100099
2300 rxr->hn_lro.lro_length_lim = HN_LRO_LENLIM_DEF;
2301 rxr->hn_lro.lro_ackcnt_lim = HN_LRO_ACKCNT_DEF;
2303 #endif /* INET || INET6 */
2305 if (sc->hn_rx_sysctl_tree != NULL) {
2309 * Create per RX ring sysctl tree:
2310 * dev.hn.UNIT.rx.RINGID
2312 snprintf(name, sizeof(name), "%d", i);
2313 rxr->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx,
2314 SYSCTL_CHILDREN(sc->hn_rx_sysctl_tree),
2315 OID_AUTO, name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2317 if (rxr->hn_rx_sysctl_tree != NULL) {
2318 SYSCTL_ADD_ULONG(ctx,
2319 SYSCTL_CHILDREN(rxr->hn_rx_sysctl_tree),
2320 OID_AUTO, "packets", CTLFLAG_RW,
2321 &rxr->hn_pkts, "# of packets received");
2322 SYSCTL_ADD_ULONG(ctx,
2323 SYSCTL_CHILDREN(rxr->hn_rx_sysctl_tree),
2324 OID_AUTO, "rss_pkts", CTLFLAG_RW,
2326 "# of packets w/ RSS info received");
2331 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_queued",
2332 CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2333 __offsetof(struct hn_rx_ring, hn_lro.lro_queued),
2334 #if __FreeBSD_version < 1100095
2335 hn_rx_stat_int_sysctl,
2337 hn_rx_stat_u64_sysctl,
2339 "LU", "LRO queued");
2340 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_flushed",
2341 CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2342 __offsetof(struct hn_rx_ring, hn_lro.lro_flushed),
2343 #if __FreeBSD_version < 1100095
2344 hn_rx_stat_int_sysctl,
2346 hn_rx_stat_u64_sysctl,
2348 "LU", "LRO flushed");
2349 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_tried",
2350 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2351 __offsetof(struct hn_rx_ring, hn_lro_tried),
2352 hn_rx_stat_ulong_sysctl, "LU", "# of LRO tries");
2353 #if __FreeBSD_version >= 1100099
2354 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_length_lim",
2355 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2356 hn_lro_lenlim_sysctl, "IU",
2357 "Max # of data bytes to be aggregated by LRO");
2358 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_ackcnt_lim",
2359 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2360 hn_lro_ackcnt_sysctl, "I",
2361 "Max # of ACKs to be aggregated by LRO");
2363 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hosttcp",
2364 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_TCP,
2365 hn_trust_hcsum_sysctl, "I",
2366 "Trust tcp segement verification on host side, "
2367 "when csum info is missing");
2368 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostudp",
2369 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_UDP,
2370 hn_trust_hcsum_sysctl, "I",
2371 "Trust udp datagram verification on host side, "
2372 "when csum info is missing");
2373 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostip",
2374 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_IP,
2375 hn_trust_hcsum_sysctl, "I",
2376 "Trust ip packet verification on host side, "
2377 "when csum info is missing");
2378 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_ip",
2379 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2380 __offsetof(struct hn_rx_ring, hn_csum_ip),
2381 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM IP");
2382 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_tcp",
2383 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2384 __offsetof(struct hn_rx_ring, hn_csum_tcp),
2385 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM TCP");
2386 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_udp",
2387 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2388 __offsetof(struct hn_rx_ring, hn_csum_udp),
2389 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM UDP");
2390 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_trusted",
2391 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2392 __offsetof(struct hn_rx_ring, hn_csum_trusted),
2393 hn_rx_stat_ulong_sysctl, "LU",
2394 "# of packets that we trust host's csum verification");
2395 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "small_pkts",
2396 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2397 __offsetof(struct hn_rx_ring, hn_small_pkts),
2398 hn_rx_stat_ulong_sysctl, "LU", "# of small packets received");
2399 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_cnt",
2400 CTLFLAG_RD, &sc->hn_rx_ring_cnt, 0, "# created RX rings");
2401 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_inuse",
2402 CTLFLAG_RD, &sc->hn_rx_ring_inuse, 0, "# used RX rings");
2408 hn_destroy_rx_data(struct hn_softc *sc)
2412 if (sc->hn_rxbuf != NULL) {
2413 hyperv_dmamem_free(&sc->hn_rxbuf_dma, sc->hn_rxbuf);
2414 sc->hn_rxbuf = NULL;
2417 if (sc->hn_rx_ring_cnt == 0)
2420 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2421 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
2423 #if defined(INET) || defined(INET6)
2424 tcp_lro_free(&rxr->hn_lro);
2426 free(rxr->hn_rdbuf, M_NETVSC);
2428 free(sc->hn_rx_ring, M_NETVSC);
2429 sc->hn_rx_ring = NULL;
2431 sc->hn_rx_ring_cnt = 0;
2432 sc->hn_rx_ring_inuse = 0;
2436 hn_create_tx_ring(struct hn_softc *sc, int id)
2438 struct hn_tx_ring *txr = &sc->hn_tx_ring[id];
2439 device_t dev = sc->hn_dev;
2440 bus_dma_tag_t parent_dtag;
2445 txr->hn_tx_idx = id;
2447 #ifndef HN_USE_TXDESC_BUFRING
2448 mtx_init(&txr->hn_txlist_spin, "hn txlist", NULL, MTX_SPIN);
2450 mtx_init(&txr->hn_tx_lock, "hn tx", NULL, MTX_DEF);
2452 txr->hn_txdesc_cnt = HN_TX_DESC_CNT;
2453 txr->hn_txdesc = malloc(sizeof(struct hn_txdesc) * txr->hn_txdesc_cnt,
2454 M_NETVSC, M_WAITOK | M_ZERO);
2455 #ifndef HN_USE_TXDESC_BUFRING
2456 SLIST_INIT(&txr->hn_txlist);
2458 txr->hn_txdesc_br = buf_ring_alloc(txr->hn_txdesc_cnt, M_NETVSC,
2459 M_WAITOK, &txr->hn_tx_lock);
2462 txr->hn_tx_taskq = sc->hn_tx_taskq;
2464 if (hn_use_if_start) {
2465 txr->hn_txeof = hn_start_txeof;
2466 TASK_INIT(&txr->hn_tx_task, 0, hn_start_taskfunc, txr);
2467 TASK_INIT(&txr->hn_txeof_task, 0, hn_start_txeof_taskfunc, txr);
2471 txr->hn_txeof = hn_xmit_txeof;
2472 TASK_INIT(&txr->hn_tx_task, 0, hn_xmit_taskfunc, txr);
2473 TASK_INIT(&txr->hn_txeof_task, 0, hn_xmit_txeof_taskfunc, txr);
2475 br_depth = hn_get_txswq_depth(txr);
2476 txr->hn_mbuf_br = buf_ring_alloc(br_depth, M_NETVSC,
2477 M_WAITOK, &txr->hn_tx_lock);
2480 txr->hn_direct_tx_size = hn_direct_tx_size;
2481 version = VMBUS_GET_VERSION(device_get_parent(dev), dev);
2482 if (version >= VMBUS_VERSION_WIN8_1) {
2483 txr->hn_csum_assist = HN_CSUM_ASSIST;
2485 txr->hn_csum_assist = HN_CSUM_ASSIST_WIN8;
2487 device_printf(dev, "bus version %u.%u, "
2488 "no UDP checksum offloading\n",
2489 VMBUS_VERSION_MAJOR(version),
2490 VMBUS_VERSION_MINOR(version));
2495 * Always schedule transmission instead of trying to do direct
2496 * transmission. This one gives the best performance so far.
2498 txr->hn_sched_tx = 1;
2500 parent_dtag = bus_get_dma_tag(dev);
2502 /* DMA tag for RNDIS messages. */
2503 error = bus_dma_tag_create(parent_dtag, /* parent */
2504 HN_RNDIS_MSG_ALIGN, /* alignment */
2505 HN_RNDIS_MSG_BOUNDARY, /* boundary */
2506 BUS_SPACE_MAXADDR, /* lowaddr */
2507 BUS_SPACE_MAXADDR, /* highaddr */
2508 NULL, NULL, /* filter, filterarg */
2509 HN_RNDIS_MSG_LEN, /* maxsize */
2511 HN_RNDIS_MSG_LEN, /* maxsegsize */
2513 NULL, /* lockfunc */
2514 NULL, /* lockfuncarg */
2515 &txr->hn_tx_rndis_dtag);
2517 device_printf(dev, "failed to create rndis dmatag\n");
2521 /* DMA tag for data. */
2522 error = bus_dma_tag_create(parent_dtag, /* parent */
2524 HN_TX_DATA_BOUNDARY, /* boundary */
2525 BUS_SPACE_MAXADDR, /* lowaddr */
2526 BUS_SPACE_MAXADDR, /* highaddr */
2527 NULL, NULL, /* filter, filterarg */
2528 HN_TX_DATA_MAXSIZE, /* maxsize */
2529 HN_TX_DATA_SEGCNT_MAX, /* nsegments */
2530 HN_TX_DATA_SEGSIZE, /* maxsegsize */
2532 NULL, /* lockfunc */
2533 NULL, /* lockfuncarg */
2534 &txr->hn_tx_data_dtag);
2536 device_printf(dev, "failed to create data dmatag\n");
2540 for (i = 0; i < txr->hn_txdesc_cnt; ++i) {
2541 struct hn_txdesc *txd = &txr->hn_txdesc[i];
2546 * Allocate and load RNDIS messages.
2548 error = bus_dmamem_alloc(txr->hn_tx_rndis_dtag,
2549 (void **)&txd->rndis_msg,
2550 BUS_DMA_WAITOK | BUS_DMA_COHERENT,
2551 &txd->rndis_msg_dmap);
2554 "failed to allocate rndis_msg, %d\n", i);
2558 error = bus_dmamap_load(txr->hn_tx_rndis_dtag,
2559 txd->rndis_msg_dmap,
2560 txd->rndis_msg, HN_RNDIS_MSG_LEN,
2561 hyperv_dma_map_paddr, &txd->rndis_msg_paddr,
2565 "failed to load rndis_msg, %d\n", i);
2566 bus_dmamem_free(txr->hn_tx_rndis_dtag,
2567 txd->rndis_msg, txd->rndis_msg_dmap);
2571 /* DMA map for TX data. */
2572 error = bus_dmamap_create(txr->hn_tx_data_dtag, 0,
2576 "failed to allocate tx data dmamap\n");
2577 bus_dmamap_unload(txr->hn_tx_rndis_dtag,
2578 txd->rndis_msg_dmap);
2579 bus_dmamem_free(txr->hn_tx_rndis_dtag,
2580 txd->rndis_msg, txd->rndis_msg_dmap);
2584 /* All set, put it to list */
2585 txd->flags |= HN_TXD_FLAG_ONLIST;
2586 #ifndef HN_USE_TXDESC_BUFRING
2587 SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
2589 buf_ring_enqueue(txr->hn_txdesc_br, txd);
2592 txr->hn_txdesc_avail = txr->hn_txdesc_cnt;
2594 if (sc->hn_tx_sysctl_tree != NULL) {
2595 struct sysctl_oid_list *child;
2596 struct sysctl_ctx_list *ctx;
2600 * Create per TX ring sysctl tree:
2601 * dev.hn.UNIT.tx.RINGID
2603 ctx = device_get_sysctl_ctx(dev);
2604 child = SYSCTL_CHILDREN(sc->hn_tx_sysctl_tree);
2606 snprintf(name, sizeof(name), "%d", id);
2607 txr->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO,
2608 name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2610 if (txr->hn_tx_sysctl_tree != NULL) {
2611 child = SYSCTL_CHILDREN(txr->hn_tx_sysctl_tree);
2613 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_avail",
2614 CTLFLAG_RD, &txr->hn_txdesc_avail, 0,
2615 "# of available TX descs");
2616 if (!hn_use_if_start) {
2617 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "oactive",
2618 CTLFLAG_RD, &txr->hn_oactive, 0,
2621 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "packets",
2622 CTLFLAG_RW, &txr->hn_pkts,
2623 "# of packets transmitted");
2631 hn_txdesc_dmamap_destroy(struct hn_txdesc *txd)
2633 struct hn_tx_ring *txr = txd->txr;
2635 KASSERT(txd->m == NULL, ("still has mbuf installed"));
2636 KASSERT((txd->flags & HN_TXD_FLAG_DMAMAP) == 0, ("still dma mapped"));
2638 bus_dmamap_unload(txr->hn_tx_rndis_dtag, txd->rndis_msg_dmap);
2639 bus_dmamem_free(txr->hn_tx_rndis_dtag, txd->rndis_msg,
2640 txd->rndis_msg_dmap);
2641 bus_dmamap_destroy(txr->hn_tx_data_dtag, txd->data_dmap);
2645 hn_destroy_tx_ring(struct hn_tx_ring *txr)
2647 struct hn_txdesc *txd;
2649 if (txr->hn_txdesc == NULL)
2652 #ifndef HN_USE_TXDESC_BUFRING
2653 while ((txd = SLIST_FIRST(&txr->hn_txlist)) != NULL) {
2654 SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
2655 hn_txdesc_dmamap_destroy(txd);
2658 mtx_lock(&txr->hn_tx_lock);
2659 while ((txd = buf_ring_dequeue_sc(txr->hn_txdesc_br)) != NULL)
2660 hn_txdesc_dmamap_destroy(txd);
2661 mtx_unlock(&txr->hn_tx_lock);
2664 if (txr->hn_tx_data_dtag != NULL)
2665 bus_dma_tag_destroy(txr->hn_tx_data_dtag);
2666 if (txr->hn_tx_rndis_dtag != NULL)
2667 bus_dma_tag_destroy(txr->hn_tx_rndis_dtag);
2669 #ifdef HN_USE_TXDESC_BUFRING
2670 buf_ring_free(txr->hn_txdesc_br, M_NETVSC);
2673 free(txr->hn_txdesc, M_NETVSC);
2674 txr->hn_txdesc = NULL;
2676 if (txr->hn_mbuf_br != NULL)
2677 buf_ring_free(txr->hn_mbuf_br, M_NETVSC);
2679 #ifndef HN_USE_TXDESC_BUFRING
2680 mtx_destroy(&txr->hn_txlist_spin);
2682 mtx_destroy(&txr->hn_tx_lock);
2686 hn_create_tx_data(struct hn_softc *sc, int ring_cnt)
2688 struct sysctl_oid_list *child;
2689 struct sysctl_ctx_list *ctx;
2693 * Create TXBUF for chimney sending.
2695 * NOTE: It is shared by all channels.
2697 sc->hn_chim = hyperv_dmamem_alloc(bus_get_dma_tag(sc->hn_dev),
2698 PAGE_SIZE, 0, NETVSC_SEND_BUFFER_SIZE, &sc->hn_chim_dma,
2699 BUS_DMA_WAITOK | BUS_DMA_ZERO);
2700 if (sc->hn_chim == NULL) {
2701 device_printf(sc->hn_dev, "allocate txbuf failed\n");
2705 sc->hn_tx_ring_cnt = ring_cnt;
2706 sc->hn_tx_ring_inuse = sc->hn_tx_ring_cnt;
2708 sc->hn_tx_ring = malloc(sizeof(struct hn_tx_ring) * sc->hn_tx_ring_cnt,
2709 M_NETVSC, M_WAITOK | M_ZERO);
2711 ctx = device_get_sysctl_ctx(sc->hn_dev);
2712 child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->hn_dev));
2714 /* Create dev.hn.UNIT.tx sysctl tree */
2715 sc->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "tx",
2716 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2718 for (i = 0; i < sc->hn_tx_ring_cnt; ++i) {
2721 error = hn_create_tx_ring(sc, i);
2726 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "no_txdescs",
2727 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2728 __offsetof(struct hn_tx_ring, hn_no_txdescs),
2729 hn_tx_stat_ulong_sysctl, "LU", "# of times short of TX descs");
2730 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "send_failed",
2731 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2732 __offsetof(struct hn_tx_ring, hn_send_failed),
2733 hn_tx_stat_ulong_sysctl, "LU", "# of hyper-v sending failure");
2734 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "txdma_failed",
2735 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2736 __offsetof(struct hn_tx_ring, hn_txdma_failed),
2737 hn_tx_stat_ulong_sysctl, "LU", "# of TX DMA failure");
2738 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_collapsed",
2739 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2740 __offsetof(struct hn_tx_ring, hn_tx_collapsed),
2741 hn_tx_stat_ulong_sysctl, "LU", "# of TX mbuf collapsed");
2742 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney",
2743 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2744 __offsetof(struct hn_tx_ring, hn_tx_chimney),
2745 hn_tx_stat_ulong_sysctl, "LU", "# of chimney send");
2746 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_tried",
2747 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2748 __offsetof(struct hn_tx_ring, hn_tx_chimney_tried),
2749 hn_tx_stat_ulong_sysctl, "LU", "# of chimney send tries");
2750 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_cnt",
2751 CTLFLAG_RD, &sc->hn_tx_ring[0].hn_txdesc_cnt, 0,
2752 "# of total TX descs");
2753 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_chimney_max",
2754 CTLFLAG_RD, &sc->hn_chim_szmax, 0,
2755 "Chimney send packet size upper boundary");
2756 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_size",
2757 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2758 hn_chim_size_sysctl, "I", "Chimney send packet size limit");
2759 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "direct_tx_size",
2760 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2761 __offsetof(struct hn_tx_ring, hn_direct_tx_size),
2762 hn_tx_conf_int_sysctl, "I",
2763 "Size of the packet for direct transmission");
2764 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "sched_tx",
2765 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2766 __offsetof(struct hn_tx_ring, hn_sched_tx),
2767 hn_tx_conf_int_sysctl, "I",
2768 "Always schedule transmission "
2769 "instead of doing direct transmission");
2770 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_cnt",
2771 CTLFLAG_RD, &sc->hn_tx_ring_cnt, 0, "# created TX rings");
2772 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_inuse",
2773 CTLFLAG_RD, &sc->hn_tx_ring_inuse, 0, "# used TX rings");
2779 hn_set_chim_size(struct hn_softc *sc, int chim_size)
2784 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
2785 sc->hn_tx_ring[i].hn_chim_size = chim_size;
2790 hn_destroy_tx_data(struct hn_softc *sc)
2794 if (sc->hn_chim != NULL) {
2795 hyperv_dmamem_free(&sc->hn_chim_dma, sc->hn_chim);
2799 if (sc->hn_tx_ring_cnt == 0)
2802 for (i = 0; i < sc->hn_tx_ring_cnt; ++i)
2803 hn_destroy_tx_ring(&sc->hn_tx_ring[i]);
2805 free(sc->hn_tx_ring, M_NETVSC);
2806 sc->hn_tx_ring = NULL;
2808 sc->hn_tx_ring_cnt = 0;
2809 sc->hn_tx_ring_inuse = 0;
2813 hn_start_taskfunc(void *xtxr, int pending __unused)
2815 struct hn_tx_ring *txr = xtxr;
2817 mtx_lock(&txr->hn_tx_lock);
2818 hn_start_locked(txr, 0);
2819 mtx_unlock(&txr->hn_tx_lock);
2823 hn_start_txeof_taskfunc(void *xtxr, int pending __unused)
2825 struct hn_tx_ring *txr = xtxr;
2827 mtx_lock(&txr->hn_tx_lock);
2828 atomic_clear_int(&txr->hn_sc->hn_ifp->if_drv_flags, IFF_DRV_OACTIVE);
2829 hn_start_locked(txr, 0);
2830 mtx_unlock(&txr->hn_tx_lock);
2834 hn_stop_tx_tasks(struct hn_softc *sc)
2838 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2839 struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
2841 taskqueue_drain(txr->hn_tx_taskq, &txr->hn_tx_task);
2842 taskqueue_drain(txr->hn_tx_taskq, &txr->hn_txeof_task);
2847 hn_xmit(struct hn_tx_ring *txr, int len)
2849 struct hn_softc *sc = txr->hn_sc;
2850 struct ifnet *ifp = sc->hn_ifp;
2851 struct mbuf *m_head;
2853 mtx_assert(&txr->hn_tx_lock, MA_OWNED);
2854 KASSERT(hn_use_if_start == 0,
2855 ("hn_xmit is called, when if_start is enabled"));
2857 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || txr->hn_oactive)
2860 while ((m_head = drbr_peek(ifp, txr->hn_mbuf_br)) != NULL) {
2861 struct hn_txdesc *txd;
2864 if (len > 0 && m_head->m_pkthdr.len > len) {
2866 * This sending could be time consuming; let callers
2867 * dispatch this packet sending (and sending of any
2868 * following up packets) to tx taskqueue.
2870 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2874 txd = hn_txdesc_get(txr);
2876 txr->hn_no_txdescs++;
2877 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2878 txr->hn_oactive = 1;
2882 error = hn_encap(txr, txd, &m_head);
2884 /* Both txd and m_head are freed; discard */
2885 drbr_advance(ifp, txr->hn_mbuf_br);
2889 error = hn_send_pkt(ifp, txr, txd);
2890 if (__predict_false(error)) {
2891 /* txd is freed, but m_head is not */
2892 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2893 txr->hn_oactive = 1;
2898 drbr_advance(ifp, txr->hn_mbuf_br);
2904 hn_transmit(struct ifnet *ifp, struct mbuf *m)
2906 struct hn_softc *sc = ifp->if_softc;
2907 struct hn_tx_ring *txr;
2911 * Select the TX ring based on flowid
2913 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2914 idx = m->m_pkthdr.flowid % sc->hn_tx_ring_inuse;
2915 txr = &sc->hn_tx_ring[idx];
2917 error = drbr_enqueue(ifp, txr->hn_mbuf_br, m);
2919 if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1);
2923 if (txr->hn_oactive)
2926 if (txr->hn_sched_tx)
2929 if (mtx_trylock(&txr->hn_tx_lock)) {
2932 sched = hn_xmit(txr, txr->hn_direct_tx_size);
2933 mtx_unlock(&txr->hn_tx_lock);
2938 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
2943 hn_xmit_qflush(struct ifnet *ifp)
2945 struct hn_softc *sc = ifp->if_softc;
2948 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2949 struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
2952 mtx_lock(&txr->hn_tx_lock);
2953 while ((m = buf_ring_dequeue_sc(txr->hn_mbuf_br)) != NULL)
2955 mtx_unlock(&txr->hn_tx_lock);
2961 hn_xmit_txeof(struct hn_tx_ring *txr)
2964 if (txr->hn_sched_tx)
2967 if (mtx_trylock(&txr->hn_tx_lock)) {
2970 txr->hn_oactive = 0;
2971 sched = hn_xmit(txr, txr->hn_direct_tx_size);
2972 mtx_unlock(&txr->hn_tx_lock);
2974 taskqueue_enqueue(txr->hn_tx_taskq,
2980 * Release the oactive earlier, with the hope, that
2981 * others could catch up. The task will clear the
2982 * oactive again with the hn_tx_lock to avoid possible
2985 txr->hn_oactive = 0;
2986 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
2991 hn_xmit_taskfunc(void *xtxr, int pending __unused)
2993 struct hn_tx_ring *txr = xtxr;
2995 mtx_lock(&txr->hn_tx_lock);
2997 mtx_unlock(&txr->hn_tx_lock);
3001 hn_xmit_txeof_taskfunc(void *xtxr, int pending __unused)
3003 struct hn_tx_ring *txr = xtxr;
3005 mtx_lock(&txr->hn_tx_lock);
3006 txr->hn_oactive = 0;
3008 mtx_unlock(&txr->hn_tx_lock);
3012 hn_channel_attach(struct hn_softc *sc, struct vmbus_channel *chan)
3014 struct hn_rx_ring *rxr;
3017 idx = vmbus_chan_subidx(chan);
3019 KASSERT(idx >= 0 && idx < sc->hn_rx_ring_inuse,
3020 ("invalid channel index %d, should > 0 && < %d",
3021 idx, sc->hn_rx_ring_inuse));
3022 rxr = &sc->hn_rx_ring[idx];
3023 KASSERT((rxr->hn_rx_flags & HN_RX_FLAG_ATTACHED) == 0,
3024 ("RX ring %d already attached", idx));
3025 rxr->hn_rx_flags |= HN_RX_FLAG_ATTACHED;
3028 if_printf(sc->hn_ifp, "link RX ring %d to channel%u\n",
3029 idx, vmbus_chan_id(chan));
3032 if (idx < sc->hn_tx_ring_inuse) {
3033 struct hn_tx_ring *txr = &sc->hn_tx_ring[idx];
3035 KASSERT((txr->hn_tx_flags & HN_TX_FLAG_ATTACHED) == 0,
3036 ("TX ring %d already attached", idx));
3037 txr->hn_tx_flags |= HN_TX_FLAG_ATTACHED;
3039 txr->hn_chan = chan;
3041 if_printf(sc->hn_ifp, "link TX ring %d to channel%u\n",
3042 idx, vmbus_chan_id(chan));
3046 /* Bind channel to a proper CPU */
3047 vmbus_chan_cpu_set(chan, (sc->hn_cpu + idx) % mp_ncpus);
3051 hn_subchan_attach(struct hn_softc *sc, struct vmbus_channel *chan)
3054 KASSERT(!vmbus_chan_is_primary(chan),
3055 ("subchannel callback on primary channel"));
3056 hn_channel_attach(sc, chan);
3060 hn_subchan_setup(struct hn_softc *sc)
3062 struct vmbus_channel **subchans;
3063 int subchan_cnt = sc->hn_rx_ring_inuse - 1;
3066 /* Wait for sub-channels setup to complete. */
3067 subchans = vmbus_subchan_get(sc->hn_prichan, subchan_cnt);
3069 /* Attach the sub-channels. */
3070 for (i = 0; i < subchan_cnt; ++i) {
3071 struct vmbus_channel *subchan = subchans[i];
3073 /* NOTE: Calling order is critical. */
3074 hn_subchan_attach(sc, subchan);
3075 hv_nv_subchan_attach(subchan,
3076 &sc->hn_rx_ring[vmbus_chan_subidx(subchan)]);
3079 /* Release the sub-channels */
3080 vmbus_subchan_rel(subchans, subchan_cnt);
3081 if_printf(sc->hn_ifp, "%d sub-channels setup done\n", subchan_cnt);
3085 hn_tx_taskq_create(void *arg __unused)
3087 if (!hn_share_tx_taskq)
3090 hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
3091 taskqueue_thread_enqueue, &hn_tx_taskq);
3092 taskqueue_start_threads(&hn_tx_taskq, 1, PI_NET, "hn tx");
3093 if (hn_bind_tx_taskq >= 0) {
3094 int cpu = hn_bind_tx_taskq;
3095 struct task cpuset_task;
3098 if (cpu > mp_ncpus - 1)
3100 CPU_SETOF(cpu, &cpu_set);
3101 TASK_INIT(&cpuset_task, 0, hn_cpuset_setthread_task, &cpu_set);
3102 taskqueue_enqueue(hn_tx_taskq, &cpuset_task);
3103 taskqueue_drain(hn_tx_taskq, &cpuset_task);
3106 SYSINIT(hn_txtq_create, SI_SUB_DRIVERS, SI_ORDER_FIRST,
3107 hn_tx_taskq_create, NULL);
3110 hn_tx_taskq_destroy(void *arg __unused)
3112 if (hn_tx_taskq != NULL)
3113 taskqueue_free(hn_tx_taskq);
3115 SYSUNINIT(hn_txtq_destroy, SI_SUB_DRIVERS, SI_ORDER_FIRST,
3116 hn_tx_taskq_destroy, NULL);
3118 static device_method_t netvsc_methods[] = {
3119 /* Device interface */
3120 DEVMETHOD(device_probe, netvsc_probe),
3121 DEVMETHOD(device_attach, netvsc_attach),
3122 DEVMETHOD(device_detach, netvsc_detach),
3123 DEVMETHOD(device_shutdown, netvsc_shutdown),
3128 static driver_t netvsc_driver = {
3134 static devclass_t netvsc_devclass;
3136 DRIVER_MODULE(hn, vmbus, netvsc_driver, netvsc_devclass, 0, 0);
3137 MODULE_VERSION(hn, 1);
3138 MODULE_DEPEND(hn, vmbus, 1, 1, 1);