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_tx_chimney_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 void 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_tx_chimney_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 hn_create_rx_data(sc, ring_cnt);
526 * Associate the first TX/RX ring w/ the primary channel.
528 hn_channel_attach(sc, sc->hn_prichan);
530 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
531 ifp->if_ioctl = hn_ioctl;
532 ifp->if_init = hn_ifinit;
533 /* needed by hv_rf_on_device_add() code */
534 ifp->if_mtu = ETHERMTU;
535 if (hn_use_if_start) {
536 int qdepth = hn_get_txswq_depth(&sc->hn_tx_ring[0]);
538 ifp->if_start = hn_start;
539 IFQ_SET_MAXLEN(&ifp->if_snd, qdepth);
540 ifp->if_snd.ifq_drv_maxlen = qdepth - 1;
541 IFQ_SET_READY(&ifp->if_snd);
543 ifp->if_transmit = hn_transmit;
544 ifp->if_qflush = hn_xmit_qflush;
547 ifmedia_init(&sc->hn_media, 0, hn_ifmedia_upd, hn_ifmedia_sts);
548 ifmedia_add(&sc->hn_media, IFM_ETHER | IFM_AUTO, 0, NULL);
549 ifmedia_set(&sc->hn_media, IFM_ETHER | IFM_AUTO);
550 /* XXX ifmedia_set really should do this for us */
551 sc->hn_media.ifm_media = sc->hn_media.ifm_cur->ifm_media;
554 * Tell upper layers that we support full VLAN capability.
556 ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
557 ifp->if_capabilities |=
558 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
561 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
563 ifp->if_hwassist = sc->hn_tx_ring[0].hn_csum_assist | CSUM_TSO;
565 sc->hn_xact = vmbus_xact_ctx_create(bus_get_dma_tag(dev),
566 HN_XACT_REQ_SIZE, HN_XACT_RESP_SIZE, 0);
567 if (sc->hn_xact == NULL)
570 error = hv_rf_on_device_add(sc, &device_info, ring_cnt,
574 KASSERT(sc->net_dev->num_channel > 0 &&
575 sc->net_dev->num_channel <= sc->hn_rx_ring_inuse,
576 ("invalid channel count %u, should be less than %d",
577 sc->net_dev->num_channel, sc->hn_rx_ring_inuse));
580 * Set the # of TX/RX rings that could be used according to
581 * the # of channels that host offered.
583 if (sc->hn_tx_ring_inuse > sc->net_dev->num_channel)
584 sc->hn_tx_ring_inuse = sc->net_dev->num_channel;
585 sc->hn_rx_ring_inuse = sc->net_dev->num_channel;
586 device_printf(dev, "%d TX ring, %d RX ring\n",
587 sc->hn_tx_ring_inuse, sc->hn_rx_ring_inuse);
589 if (sc->net_dev->num_channel > 1)
590 hn_subchan_setup(sc);
592 #if __FreeBSD_version >= 1100099
593 if (sc->hn_rx_ring_inuse > 1) {
595 * Reduce TCP segment aggregation limit for multiple
596 * RX rings to increase ACK timeliness.
598 hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MULTIRX_DEF);
602 if (device_info.link_state == 0) {
606 tso_maxlen = hn_tso_maxlen;
607 if (tso_maxlen <= 0 || tso_maxlen > IP_MAXPACKET)
608 tso_maxlen = IP_MAXPACKET;
610 ifp->if_hw_tsomaxsegcount = HN_TX_DATA_SEGCNT_MAX;
611 ifp->if_hw_tsomaxsegsize = PAGE_SIZE;
612 ifp->if_hw_tsomax = tso_maxlen -
613 (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
615 ether_ifattach(ifp, device_info.mac_addr);
617 if_printf(ifp, "TSO: %u/%u/%u\n", ifp->if_hw_tsomax,
618 ifp->if_hw_tsomaxsegcount, ifp->if_hw_tsomaxsegsize);
620 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
621 hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max);
622 if (hn_tx_chimney_size > 0 &&
623 hn_tx_chimney_size < sc->hn_tx_chimney_max)
624 hn_set_tx_chimney_size(sc, hn_tx_chimney_size);
628 hn_destroy_tx_data(sc);
635 * Standard detach entry point
638 netvsc_detach(device_t dev)
640 struct hn_softc *sc = device_get_softc(dev);
643 printf("netvsc_detach\n");
646 * XXXKYS: Need to clean up all our
647 * driver state; this is the driver
652 * XXXKYS: Need to stop outgoing traffic and unregister
656 hv_rf_on_device_remove(sc, HV_RF_NV_DESTROY_CHANNEL);
658 hn_stop_tx_tasks(sc);
660 ifmedia_removeall(&sc->hn_media);
661 hn_destroy_rx_data(sc);
662 hn_destroy_tx_data(sc);
664 if (sc->hn_tx_taskq != hn_tx_taskq)
665 taskqueue_free(sc->hn_tx_taskq);
667 vmbus_xact_ctx_destroy(sc->hn_xact);
672 * Standard shutdown entry point
675 netvsc_shutdown(device_t dev)
681 hn_txdesc_dmamap_load(struct hn_tx_ring *txr, struct hn_txdesc *txd,
682 struct mbuf **m_head, bus_dma_segment_t *segs, int *nsegs)
684 struct mbuf *m = *m_head;
687 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag, txd->data_dmap,
688 m, segs, nsegs, BUS_DMA_NOWAIT);
689 if (error == EFBIG) {
692 m_new = m_collapse(m, M_NOWAIT, HN_TX_DATA_SEGCNT_MAX);
697 txr->hn_tx_collapsed++;
699 error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag,
700 txd->data_dmap, m, segs, nsegs, BUS_DMA_NOWAIT);
703 bus_dmamap_sync(txr->hn_tx_data_dtag, txd->data_dmap,
704 BUS_DMASYNC_PREWRITE);
705 txd->flags |= HN_TXD_FLAG_DMAMAP;
711 hn_txdesc_dmamap_unload(struct hn_tx_ring *txr, struct hn_txdesc *txd)
714 if (txd->flags & HN_TXD_FLAG_DMAMAP) {
715 bus_dmamap_sync(txr->hn_tx_data_dtag,
716 txd->data_dmap, BUS_DMASYNC_POSTWRITE);
717 bus_dmamap_unload(txr->hn_tx_data_dtag,
719 txd->flags &= ~HN_TXD_FLAG_DMAMAP;
724 hn_txdesc_put(struct hn_tx_ring *txr, struct hn_txdesc *txd)
727 KASSERT((txd->flags & HN_TXD_FLAG_ONLIST) == 0,
728 ("put an onlist txd %#x", txd->flags));
730 KASSERT(txd->refs > 0, ("invalid txd refs %d", txd->refs));
731 if (atomic_fetchadd_int(&txd->refs, -1) != 1)
734 hn_txdesc_dmamap_unload(txr, txd);
735 if (txd->m != NULL) {
740 txd->flags |= HN_TXD_FLAG_ONLIST;
742 #ifndef HN_USE_TXDESC_BUFRING
743 mtx_lock_spin(&txr->hn_txlist_spin);
744 KASSERT(txr->hn_txdesc_avail >= 0 &&
745 txr->hn_txdesc_avail < txr->hn_txdesc_cnt,
746 ("txdesc_put: invalid txd avail %d", txr->hn_txdesc_avail));
747 txr->hn_txdesc_avail++;
748 SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
749 mtx_unlock_spin(&txr->hn_txlist_spin);
751 atomic_add_int(&txr->hn_txdesc_avail, 1);
752 buf_ring_enqueue(txr->hn_txdesc_br, txd);
758 static __inline struct hn_txdesc *
759 hn_txdesc_get(struct hn_tx_ring *txr)
761 struct hn_txdesc *txd;
763 #ifndef HN_USE_TXDESC_BUFRING
764 mtx_lock_spin(&txr->hn_txlist_spin);
765 txd = SLIST_FIRST(&txr->hn_txlist);
767 KASSERT(txr->hn_txdesc_avail > 0,
768 ("txdesc_get: invalid txd avail %d", txr->hn_txdesc_avail));
769 txr->hn_txdesc_avail--;
770 SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
772 mtx_unlock_spin(&txr->hn_txlist_spin);
774 txd = buf_ring_dequeue_sc(txr->hn_txdesc_br);
778 #ifdef HN_USE_TXDESC_BUFRING
779 atomic_subtract_int(&txr->hn_txdesc_avail, 1);
781 KASSERT(txd->m == NULL && txd->refs == 0 &&
782 (txd->flags & HN_TXD_FLAG_ONLIST), ("invalid txd"));
783 txd->flags &= ~HN_TXD_FLAG_ONLIST;
790 hn_txdesc_hold(struct hn_txdesc *txd)
793 /* 0->1 transition will never work */
794 KASSERT(txd->refs > 0, ("invalid refs %d", txd->refs));
795 atomic_add_int(&txd->refs, 1);
799 hn_txeof(struct hn_tx_ring *txr)
801 txr->hn_has_txeof = 0;
806 hn_tx_done(struct hn_send_ctx *sndc, struct netvsc_dev_ *net_dev,
807 struct vmbus_channel *chan, const struct nvsp_msg_ *msg __unused,
810 struct hn_txdesc *txd = sndc->hn_cbarg;
811 struct hn_tx_ring *txr;
813 if (sndc->hn_chim_idx != NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX)
814 hn_chim_free(net_dev, sndc->hn_chim_idx);
817 KASSERT(txr->hn_chan == chan,
818 ("channel mismatch, on chan%u, should be chan%u",
819 vmbus_chan_subidx(chan), vmbus_chan_subidx(txr->hn_chan)));
821 txr->hn_has_txeof = 1;
822 hn_txdesc_put(txr, txd);
824 ++txr->hn_txdone_cnt;
825 if (txr->hn_txdone_cnt >= HN_EARLY_TXEOF_THRESH) {
826 txr->hn_txdone_cnt = 0;
833 netvsc_channel_rollup(struct hn_rx_ring *rxr, struct hn_tx_ring *txr)
835 #if defined(INET) || defined(INET6)
836 struct lro_ctrl *lro = &rxr->hn_lro;
837 struct lro_entry *queued;
839 while ((queued = SLIST_FIRST(&lro->lro_active)) != NULL) {
840 SLIST_REMOVE_HEAD(&lro->lro_active, next);
841 tcp_lro_flush(lro, queued);
847 * 'txr' could be NULL, if multiple channels and
848 * ifnet.if_start method are enabled.
850 if (txr == NULL || !txr->hn_has_txeof)
853 txr->hn_txdone_cnt = 0;
859 * If this function fails, then both txd and m_head0 will be freed.
862 hn_encap(struct hn_tx_ring *txr, struct hn_txdesc *txd, struct mbuf **m_head0)
864 bus_dma_segment_t segs[HN_TX_DATA_SEGCNT_MAX];
866 struct mbuf *m_head = *m_head0;
867 rndis_msg *rndis_mesg;
868 rndis_packet *rndis_pkt;
869 rndis_per_packet_info *rppi;
870 struct rndis_hash_value *hash_value;
871 uint32_t rndis_msg_size, tot_data_buf_len, send_buf_section_idx;
872 int send_buf_section_size;
874 tot_data_buf_len = m_head->m_pkthdr.len;
877 * extension points to the area reserved for the
878 * rndis_filter_packet, which is placed just after
879 * the netvsc_packet (and rppi struct, if present;
880 * length is updated later).
882 rndis_mesg = txd->rndis_msg;
883 /* XXX not necessary */
884 memset(rndis_mesg, 0, HN_RNDIS_MSG_LEN);
885 rndis_mesg->ndis_msg_type = REMOTE_NDIS_PACKET_MSG;
887 rndis_pkt = &rndis_mesg->msg.packet;
888 rndis_pkt->data_offset = sizeof(rndis_packet);
889 rndis_pkt->data_length = tot_data_buf_len;
890 rndis_pkt->per_pkt_info_offset = sizeof(rndis_packet);
892 rndis_msg_size = RNDIS_MESSAGE_SIZE(rndis_packet);
895 * Set the hash value for this packet, so that the host could
896 * dispatch the TX done event for this packet back to this TX
899 rndis_msg_size += RNDIS_HASHVAL_PPI_SIZE;
900 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_HASHVAL_PPI_SIZE,
902 hash_value = (struct rndis_hash_value *)((uint8_t *)rppi +
903 rppi->per_packet_info_offset);
904 hash_value->hash_value = txr->hn_tx_idx;
906 if (m_head->m_flags & M_VLANTAG) {
907 ndis_8021q_info *rppi_vlan_info;
909 rndis_msg_size += RNDIS_VLAN_PPI_SIZE;
910 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_VLAN_PPI_SIZE,
913 rppi_vlan_info = (ndis_8021q_info *)((uint8_t *)rppi +
914 rppi->per_packet_info_offset);
915 rppi_vlan_info->u1.s1.vlan_id =
916 m_head->m_pkthdr.ether_vtag & 0xfff;
919 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
920 rndis_tcp_tso_info *tso_info;
921 struct ether_vlan_header *eh;
925 * XXX need m_pullup and use mtodo
927 eh = mtod(m_head, struct ether_vlan_header*);
928 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN))
929 ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
931 ether_len = ETHER_HDR_LEN;
933 rndis_msg_size += RNDIS_TSO_PPI_SIZE;
934 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_TSO_PPI_SIZE,
935 tcp_large_send_info);
937 tso_info = (rndis_tcp_tso_info *)((uint8_t *)rppi +
938 rppi->per_packet_info_offset);
939 tso_info->lso_v2_xmit.type =
940 RNDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
943 if (m_head->m_pkthdr.csum_flags & CSUM_IP_TSO) {
945 (struct ip *)(m_head->m_data + ether_len);
946 unsigned long iph_len = ip->ip_hl << 2;
948 (struct tcphdr *)((caddr_t)ip + iph_len);
950 tso_info->lso_v2_xmit.ip_version =
951 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
955 th->th_sum = in_pseudo(ip->ip_src.s_addr,
956 ip->ip_dst.s_addr, htons(IPPROTO_TCP));
959 #if defined(INET6) && defined(INET)
964 struct ip6_hdr *ip6 = (struct ip6_hdr *)
965 (m_head->m_data + ether_len);
966 struct tcphdr *th = (struct tcphdr *)(ip6 + 1);
968 tso_info->lso_v2_xmit.ip_version =
969 RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
971 th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
974 tso_info->lso_v2_xmit.tcp_header_offset = 0;
975 tso_info->lso_v2_xmit.mss = m_head->m_pkthdr.tso_segsz;
976 } else if (m_head->m_pkthdr.csum_flags & txr->hn_csum_assist) {
977 rndis_tcp_ip_csum_info *csum_info;
979 rndis_msg_size += RNDIS_CSUM_PPI_SIZE;
980 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_CSUM_PPI_SIZE,
982 csum_info = (rndis_tcp_ip_csum_info *)((uint8_t *)rppi +
983 rppi->per_packet_info_offset);
985 csum_info->xmit.is_ipv4 = 1;
986 if (m_head->m_pkthdr.csum_flags & CSUM_IP)
987 csum_info->xmit.ip_header_csum = 1;
989 if (m_head->m_pkthdr.csum_flags & CSUM_TCP) {
990 csum_info->xmit.tcp_csum = 1;
991 csum_info->xmit.tcp_header_offset = 0;
992 } else if (m_head->m_pkthdr.csum_flags & CSUM_UDP) {
993 csum_info->xmit.udp_csum = 1;
997 rndis_mesg->msg_len = tot_data_buf_len + rndis_msg_size;
998 tot_data_buf_len = rndis_mesg->msg_len;
1001 * Chimney send, if the packet could fit into one chimney buffer.
1003 if (tot_data_buf_len < txr->hn_tx_chimney_size) {
1004 netvsc_dev *net_dev = txr->hn_sc->net_dev;
1006 txr->hn_tx_chimney_tried++;
1007 send_buf_section_idx =
1008 hv_nv_get_next_send_section(net_dev);
1009 if (send_buf_section_idx !=
1010 NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX) {
1011 uint8_t *dest = ((uint8_t *)net_dev->send_buf +
1012 (send_buf_section_idx *
1013 net_dev->send_section_size));
1015 memcpy(dest, rndis_mesg, rndis_msg_size);
1016 dest += rndis_msg_size;
1017 m_copydata(m_head, 0, m_head->m_pkthdr.len, dest);
1019 send_buf_section_size = tot_data_buf_len;
1020 txr->hn_gpa_cnt = 0;
1021 txr->hn_tx_chimney++;
1026 error = hn_txdesc_dmamap_load(txr, txd, &m_head, segs, &nsegs);
1031 * This mbuf is not linked w/ the txd yet, so free it now.
1036 freed = hn_txdesc_put(txr, txd);
1038 ("fail to free txd upon txdma error"));
1040 txr->hn_txdma_failed++;
1041 if_inc_counter(txr->hn_sc->hn_ifp, IFCOUNTER_OERRORS, 1);
1046 txr->hn_gpa_cnt = nsegs + HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
1048 /* send packet with page buffer */
1049 txr->hn_gpa[0].gpa_page = atop(txd->rndis_msg_paddr);
1050 txr->hn_gpa[0].gpa_ofs = txd->rndis_msg_paddr & PAGE_MASK;
1051 txr->hn_gpa[0].gpa_len = rndis_msg_size;
1054 * Fill the page buffers with mbuf info starting at index
1055 * HV_RF_NUM_TX_RESERVED_PAGE_BUFS.
1057 for (i = 0; i < nsegs; ++i) {
1058 struct vmbus_gpa *gpa = &txr->hn_gpa[
1059 i + HV_RF_NUM_TX_RESERVED_PAGE_BUFS];
1061 gpa->gpa_page = atop(segs[i].ds_addr);
1062 gpa->gpa_ofs = segs[i].ds_addr & PAGE_MASK;
1063 gpa->gpa_len = segs[i].ds_len;
1066 send_buf_section_idx = NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX;
1067 send_buf_section_size = 0;
1071 /* Set the completion routine */
1072 hn_send_ctx_init(&txd->send_ctx, hn_tx_done, txd,
1073 send_buf_section_idx, send_buf_section_size);
1080 * If this function fails, then txd will be freed, but the mbuf
1081 * associated w/ the txd will _not_ be freed.
1084 hn_send_pkt(struct ifnet *ifp, struct hn_tx_ring *txr, struct hn_txdesc *txd)
1086 int error, send_failed = 0;
1090 * Make sure that txd is not freed before ETHER_BPF_MTAP.
1092 hn_txdesc_hold(txd);
1093 error = hv_nv_on_send(txr->hn_chan, true, &txd->send_ctx,
1094 txr->hn_gpa, txr->hn_gpa_cnt);
1096 ETHER_BPF_MTAP(ifp, txd->m);
1097 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1098 if (!hn_use_if_start) {
1099 if_inc_counter(ifp, IFCOUNTER_OBYTES,
1100 txd->m->m_pkthdr.len);
1101 if (txd->m->m_flags & M_MCAST)
1102 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
1106 hn_txdesc_put(txr, txd);
1108 if (__predict_false(error)) {
1112 * This should "really rarely" happen.
1114 * XXX Too many RX to be acked or too many sideband
1115 * commands to run? Ask netvsc_channel_rollup()
1116 * to kick start later.
1118 txr->hn_has_txeof = 1;
1120 txr->hn_send_failed++;
1123 * Try sending again after set hn_has_txeof;
1124 * in case that we missed the last
1125 * netvsc_channel_rollup().
1129 if_printf(ifp, "send failed\n");
1132 * Caller will perform further processing on the
1133 * associated mbuf, so don't free it in hn_txdesc_put();
1134 * only unload it from the DMA map in hn_txdesc_put(),
1138 freed = hn_txdesc_put(txr, txd);
1140 ("fail to free txd upon send error"));
1142 txr->hn_send_failed++;
1148 * Start a transmit of one or more packets
1151 hn_start_locked(struct hn_tx_ring *txr, int len)
1153 struct hn_softc *sc = txr->hn_sc;
1154 struct ifnet *ifp = sc->hn_ifp;
1156 KASSERT(hn_use_if_start,
1157 ("hn_start_locked is called, when if_start is disabled"));
1158 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1159 mtx_assert(&txr->hn_tx_lock, MA_OWNED);
1161 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1165 while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
1166 struct hn_txdesc *txd;
1167 struct mbuf *m_head;
1170 IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
1174 if (len > 0 && m_head->m_pkthdr.len > len) {
1176 * This sending could be time consuming; let callers
1177 * dispatch this packet sending (and sending of any
1178 * following up packets) to tx taskqueue.
1180 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1184 txd = hn_txdesc_get(txr);
1186 txr->hn_no_txdescs++;
1187 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1188 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1192 error = hn_encap(txr, txd, &m_head);
1194 /* Both txd and m_head are freed */
1198 error = hn_send_pkt(ifp, txr, txd);
1199 if (__predict_false(error)) {
1200 /* txd is freed, but m_head is not */
1201 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1202 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1210 * Link up/down notification
1213 netvsc_linkstatus_callback(struct hn_softc *sc, uint32_t status)
1223 * Append the specified data to the indicated mbuf chain,
1224 * Extend the mbuf chain if the new data does not fit in
1227 * This is a minor rewrite of m_append() from sys/kern/uipc_mbuf.c.
1228 * There should be an equivalent in the kernel mbuf code,
1229 * but there does not appear to be one yet.
1231 * Differs from m_append() in that additional mbufs are
1232 * allocated with cluster size MJUMPAGESIZE, and filled
1235 * Return 1 if able to complete the job; otherwise 0.
1238 hv_m_append(struct mbuf *m0, int len, c_caddr_t cp)
1241 int remainder, space;
1243 for (m = m0; m->m_next != NULL; m = m->m_next)
1246 space = M_TRAILINGSPACE(m);
1249 * Copy into available space.
1251 if (space > remainder)
1253 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
1258 while (remainder > 0) {
1260 * Allocate a new mbuf; could check space
1261 * and allocate a cluster instead.
1263 n = m_getjcl(M_DONTWAIT, m->m_type, 0, MJUMPAGESIZE);
1266 n->m_len = min(MJUMPAGESIZE, remainder);
1267 bcopy(cp, mtod(n, caddr_t), n->m_len);
1269 remainder -= n->m_len;
1273 if (m0->m_flags & M_PKTHDR)
1274 m0->m_pkthdr.len += len - remainder;
1276 return (remainder == 0);
1279 #if defined(INET) || defined(INET6)
1281 hn_lro_rx(struct lro_ctrl *lc, struct mbuf *m)
1283 #if __FreeBSD_version >= 1100095
1284 if (hn_lro_mbufq_depth) {
1285 tcp_lro_queue_mbuf(lc, m);
1289 return tcp_lro_rx(lc, m, 0);
1294 * Called when we receive a data packet from the "wire" on the
1297 * Note: This is no longer used as a callback
1300 netvsc_recv(struct hn_rx_ring *rxr, netvsc_packet *packet,
1301 const rndis_tcp_ip_csum_info *csum_info,
1302 const struct rndis_hash_info *hash_info,
1303 const struct rndis_hash_value *hash_value)
1305 struct ifnet *ifp = rxr->hn_ifp;
1307 int size, do_lro = 0, do_csum = 1;
1309 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1313 * Bail out if packet contains more data than configured MTU.
1315 if (packet->tot_data_buf_len > (ifp->if_mtu + ETHER_HDR_LEN)) {
1317 } else if (packet->tot_data_buf_len <= MHLEN) {
1318 m_new = m_gethdr(M_NOWAIT, MT_DATA);
1319 if (m_new == NULL) {
1320 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1323 memcpy(mtod(m_new, void *), packet->data,
1324 packet->tot_data_buf_len);
1325 m_new->m_pkthdr.len = m_new->m_len = packet->tot_data_buf_len;
1326 rxr->hn_small_pkts++;
1329 * Get an mbuf with a cluster. For packets 2K or less,
1330 * get a standard 2K cluster. For anything larger, get a
1331 * 4K cluster. Any buffers larger than 4K can cause problems
1332 * if looped around to the Hyper-V TX channel, so avoid them.
1335 if (packet->tot_data_buf_len > MCLBYTES) {
1337 size = MJUMPAGESIZE;
1340 m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, size);
1341 if (m_new == NULL) {
1342 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1346 hv_m_append(m_new, packet->tot_data_buf_len, packet->data);
1348 m_new->m_pkthdr.rcvif = ifp;
1350 if (__predict_false((ifp->if_capenable & IFCAP_RXCSUM) == 0))
1353 /* receive side checksum offload */
1354 if (csum_info != NULL) {
1355 /* IP csum offload */
1356 if (csum_info->receive.ip_csum_succeeded && do_csum) {
1357 m_new->m_pkthdr.csum_flags |=
1358 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1362 /* TCP/UDP csum offload */
1363 if ((csum_info->receive.tcp_csum_succeeded ||
1364 csum_info->receive.udp_csum_succeeded) && do_csum) {
1365 m_new->m_pkthdr.csum_flags |=
1366 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1367 m_new->m_pkthdr.csum_data = 0xffff;
1368 if (csum_info->receive.tcp_csum_succeeded)
1374 if (csum_info->receive.ip_csum_succeeded &&
1375 csum_info->receive.tcp_csum_succeeded)
1378 const struct ether_header *eh;
1383 if (m_new->m_len < hoff)
1385 eh = mtod(m_new, struct ether_header *);
1386 etype = ntohs(eh->ether_type);
1387 if (etype == ETHERTYPE_VLAN) {
1388 const struct ether_vlan_header *evl;
1390 hoff = sizeof(*evl);
1391 if (m_new->m_len < hoff)
1393 evl = mtod(m_new, struct ether_vlan_header *);
1394 etype = ntohs(evl->evl_proto);
1397 if (etype == ETHERTYPE_IP) {
1400 pr = hn_check_iplen(m_new, hoff);
1401 if (pr == IPPROTO_TCP) {
1403 (rxr->hn_trust_hcsum &
1404 HN_TRUST_HCSUM_TCP)) {
1405 rxr->hn_csum_trusted++;
1406 m_new->m_pkthdr.csum_flags |=
1407 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1408 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1409 m_new->m_pkthdr.csum_data = 0xffff;
1412 } else if (pr == IPPROTO_UDP) {
1414 (rxr->hn_trust_hcsum &
1415 HN_TRUST_HCSUM_UDP)) {
1416 rxr->hn_csum_trusted++;
1417 m_new->m_pkthdr.csum_flags |=
1418 (CSUM_IP_CHECKED | CSUM_IP_VALID |
1419 CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1420 m_new->m_pkthdr.csum_data = 0xffff;
1422 } else if (pr != IPPROTO_DONE && do_csum &&
1423 (rxr->hn_trust_hcsum & HN_TRUST_HCSUM_IP)) {
1424 rxr->hn_csum_trusted++;
1425 m_new->m_pkthdr.csum_flags |=
1426 (CSUM_IP_CHECKED | CSUM_IP_VALID);
1431 if ((packet->vlan_tci != 0) &&
1432 (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0) {
1433 m_new->m_pkthdr.ether_vtag = packet->vlan_tci;
1434 m_new->m_flags |= M_VLANTAG;
1437 if (hash_info != NULL && hash_value != NULL) {
1438 int hash_type = M_HASHTYPE_OPAQUE;
1441 m_new->m_pkthdr.flowid = hash_value->hash_value;
1442 if ((hash_info->hash_info & NDIS_HASH_FUNCTION_MASK) ==
1443 NDIS_HASH_FUNCTION_TOEPLITZ) {
1445 (hash_info->hash_info & NDIS_HASH_TYPE_MASK);
1448 case NDIS_HASH_IPV4:
1449 hash_type = M_HASHTYPE_RSS_IPV4;
1452 case NDIS_HASH_TCP_IPV4:
1453 hash_type = M_HASHTYPE_RSS_TCP_IPV4;
1456 case NDIS_HASH_IPV6:
1457 hash_type = M_HASHTYPE_RSS_IPV6;
1460 case NDIS_HASH_IPV6_EX:
1461 hash_type = M_HASHTYPE_RSS_IPV6_EX;
1464 case NDIS_HASH_TCP_IPV6:
1465 hash_type = M_HASHTYPE_RSS_TCP_IPV6;
1468 case NDIS_HASH_TCP_IPV6_EX:
1469 hash_type = M_HASHTYPE_RSS_TCP_IPV6_EX;
1473 M_HASHTYPE_SET(m_new, hash_type);
1475 if (hash_value != NULL)
1476 m_new->m_pkthdr.flowid = hash_value->hash_value;
1478 m_new->m_pkthdr.flowid = rxr->hn_rx_idx;
1479 M_HASHTYPE_SET(m_new, M_HASHTYPE_OPAQUE);
1483 * Note: Moved RX completion back to hv_nv_on_receive() so all
1484 * messages (not just data messages) will trigger a response.
1490 if ((ifp->if_capenable & IFCAP_LRO) && do_lro) {
1491 #if defined(INET) || defined(INET6)
1492 struct lro_ctrl *lro = &rxr->hn_lro;
1495 rxr->hn_lro_tried++;
1496 if (hn_lro_rx(lro, m_new) == 0) {
1504 /* We're not holding the lock here, so don't release it */
1505 (*ifp->if_input)(ifp, m_new);
1511 * Rules for using sc->temp_unusable:
1512 * 1. sc->temp_unusable can only be read or written while holding NV_LOCK()
1513 * 2. code reading sc->temp_unusable under NV_LOCK(), and finding
1514 * sc->temp_unusable set, must release NV_LOCK() and exit
1515 * 3. to retain exclusive control of the interface,
1516 * sc->temp_unusable must be set by code before releasing NV_LOCK()
1517 * 4. only code setting sc->temp_unusable can clear sc->temp_unusable
1518 * 5. code setting sc->temp_unusable must eventually clear sc->temp_unusable
1522 * Standard ioctl entry point. Called when the user wants to configure
1526 hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1528 hn_softc_t *sc = ifp->if_softc;
1529 struct ifreq *ifr = (struct ifreq *)data;
1531 struct ifaddr *ifa = (struct ifaddr *)data;
1533 netvsc_device_info device_info;
1534 int mask, error = 0;
1535 int retry_cnt = 500;
1541 if (ifa->ifa_addr->sa_family == AF_INET) {
1542 ifp->if_flags |= IFF_UP;
1543 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1545 arp_ifinit(ifp, ifa);
1548 error = ether_ioctl(ifp, cmd, data);
1551 /* Check MTU value change */
1552 if (ifp->if_mtu == ifr->ifr_mtu)
1555 if (ifr->ifr_mtu > NETVSC_MAX_CONFIGURABLE_MTU) {
1560 /* Obtain and record requested MTU */
1561 ifp->if_mtu = ifr->ifr_mtu;
1563 #if __FreeBSD_version >= 1100099
1565 * Make sure that LRO aggregation length limit is still
1566 * valid, after the MTU change.
1569 if (sc->hn_rx_ring[0].hn_lro.lro_length_lim <
1570 HN_LRO_LENLIM_MIN(ifp))
1571 hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MIN(ifp));
1577 if (!sc->temp_unusable) {
1578 sc->temp_unusable = TRUE;
1582 if (retry_cnt > 0) {
1586 } while (retry_cnt > 0);
1588 if (retry_cnt == 0) {
1593 /* We must remove and add back the device to cause the new
1594 * MTU to take effect. This includes tearing down, but not
1595 * deleting the channel, then bringing it back up.
1597 error = hv_rf_on_device_remove(sc, HV_RF_NV_RETAIN_CHANNEL);
1600 sc->temp_unusable = FALSE;
1605 /* Wait for subchannels to be destroyed */
1606 vmbus_subchan_drain(sc->hn_prichan);
1608 error = hv_rf_on_device_add(sc, &device_info,
1609 sc->hn_rx_ring_inuse, &sc->hn_rx_ring[0]);
1612 sc->temp_unusable = FALSE;
1616 KASSERT(sc->hn_rx_ring_cnt == sc->net_dev->num_channel,
1617 ("RX ring count %d and channel count %u mismatch",
1618 sc->hn_rx_ring_cnt, sc->net_dev->num_channel));
1619 if (sc->net_dev->num_channel > 1) {
1623 * Skip the rings on primary channel; they are
1624 * handled by the hv_rf_on_device_add() above.
1626 for (r = 1; r < sc->hn_rx_ring_cnt; ++r) {
1627 sc->hn_rx_ring[r].hn_rx_flags &=
1628 ~HN_RX_FLAG_ATTACHED;
1630 for (r = 1; r < sc->hn_tx_ring_cnt; ++r) {
1631 sc->hn_tx_ring[r].hn_tx_flags &=
1632 ~HN_TX_FLAG_ATTACHED;
1634 hn_subchan_setup(sc);
1637 sc->hn_tx_chimney_max = sc->net_dev->send_section_size;
1638 if (sc->hn_tx_ring[0].hn_tx_chimney_size >
1639 sc->hn_tx_chimney_max)
1640 hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max);
1642 hn_ifinit_locked(sc);
1645 sc->temp_unusable = FALSE;
1651 if (!sc->temp_unusable) {
1652 sc->temp_unusable = TRUE;
1656 if (retry_cnt > 0) {
1660 } while (retry_cnt > 0);
1662 if (retry_cnt == 0) {
1667 if (ifp->if_flags & IFF_UP) {
1669 * If only the state of the PROMISC flag changed,
1670 * then just use the 'set promisc mode' command
1671 * instead of reinitializing the entire NIC. Doing
1672 * a full re-init means reloading the firmware and
1673 * waiting for it to start up, which may take a
1677 /* Fixme: Promiscuous mode? */
1678 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1679 ifp->if_flags & IFF_PROMISC &&
1680 !(sc->hn_if_flags & IFF_PROMISC)) {
1681 /* do something here for Hyper-V */
1682 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1683 !(ifp->if_flags & IFF_PROMISC) &&
1684 sc->hn_if_flags & IFF_PROMISC) {
1685 /* do something here for Hyper-V */
1688 hn_ifinit_locked(sc);
1690 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1695 sc->temp_unusable = FALSE;
1697 sc->hn_if_flags = ifp->if_flags;
1703 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1704 if (mask & IFCAP_TXCSUM) {
1705 ifp->if_capenable ^= IFCAP_TXCSUM;
1706 if (ifp->if_capenable & IFCAP_TXCSUM) {
1708 sc->hn_tx_ring[0].hn_csum_assist;
1711 ~sc->hn_tx_ring[0].hn_csum_assist;
1715 if (mask & IFCAP_RXCSUM)
1716 ifp->if_capenable ^= IFCAP_RXCSUM;
1718 if (mask & IFCAP_LRO)
1719 ifp->if_capenable ^= IFCAP_LRO;
1721 if (mask & IFCAP_TSO4) {
1722 ifp->if_capenable ^= IFCAP_TSO4;
1723 if (ifp->if_capenable & IFCAP_TSO4)
1724 ifp->if_hwassist |= CSUM_IP_TSO;
1726 ifp->if_hwassist &= ~CSUM_IP_TSO;
1729 if (mask & IFCAP_TSO6) {
1730 ifp->if_capenable ^= IFCAP_TSO6;
1731 if (ifp->if_capenable & IFCAP_TSO6)
1732 ifp->if_hwassist |= CSUM_IP6_TSO;
1734 ifp->if_hwassist &= ~CSUM_IP6_TSO;
1743 /* Fixme: Multicast mode? */
1744 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1746 netvsc_setmulti(sc);
1755 error = ifmedia_ioctl(ifp, ifr, &sc->hn_media, cmd);
1758 error = ether_ioctl(ifp, cmd, data);
1769 hn_stop(hn_softc_t *sc)
1777 printf(" Closing Device ...\n");
1779 atomic_clear_int(&ifp->if_drv_flags,
1780 (IFF_DRV_RUNNING | IFF_DRV_OACTIVE));
1781 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
1782 sc->hn_tx_ring[i].hn_oactive = 0;
1784 if_link_state_change(ifp, LINK_STATE_DOWN);
1785 sc->hn_initdone = 0;
1787 ret = hv_rf_on_close(sc);
1791 * FreeBSD transmit entry point
1794 hn_start(struct ifnet *ifp)
1796 struct hn_softc *sc = ifp->if_softc;
1797 struct hn_tx_ring *txr = &sc->hn_tx_ring[0];
1799 if (txr->hn_sched_tx)
1802 if (mtx_trylock(&txr->hn_tx_lock)) {
1805 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1806 mtx_unlock(&txr->hn_tx_lock);
1811 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
1815 hn_start_txeof(struct hn_tx_ring *txr)
1817 struct hn_softc *sc = txr->hn_sc;
1818 struct ifnet *ifp = sc->hn_ifp;
1820 KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
1822 if (txr->hn_sched_tx)
1825 if (mtx_trylock(&txr->hn_tx_lock)) {
1828 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1829 sched = hn_start_locked(txr, txr->hn_direct_tx_size);
1830 mtx_unlock(&txr->hn_tx_lock);
1832 taskqueue_enqueue(txr->hn_tx_taskq,
1838 * Release the OACTIVE earlier, with the hope, that
1839 * others could catch up. The task will clear the
1840 * flag again with the hn_tx_lock to avoid possible
1843 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1844 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
1852 hn_ifinit_locked(hn_softc_t *sc)
1859 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1863 hv_promisc_mode = 1;
1865 ret = hv_rf_on_open(sc);
1869 sc->hn_initdone = 1;
1872 atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
1873 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
1874 sc->hn_tx_ring[i].hn_oactive = 0;
1876 atomic_set_int(&ifp->if_drv_flags, IFF_DRV_RUNNING);
1877 if_link_state_change(ifp, LINK_STATE_UP);
1884 hn_ifinit(void *xsc)
1886 hn_softc_t *sc = xsc;
1889 if (sc->temp_unusable) {
1893 sc->temp_unusable = TRUE;
1896 hn_ifinit_locked(sc);
1899 sc->temp_unusable = FALSE;
1908 hn_watchdog(struct ifnet *ifp)
1913 printf("hn%d: watchdog timeout -- resetting\n", sc->hn_unit);
1914 hn_ifinit(sc); /*???*/
1919 #if __FreeBSD_version >= 1100099
1922 hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS)
1924 struct hn_softc *sc = arg1;
1925 unsigned int lenlim;
1928 lenlim = sc->hn_rx_ring[0].hn_lro.lro_length_lim;
1929 error = sysctl_handle_int(oidp, &lenlim, 0, req);
1930 if (error || req->newptr == NULL)
1933 if (lenlim < HN_LRO_LENLIM_MIN(sc->hn_ifp) ||
1934 lenlim > TCP_LRO_LENGTH_MAX)
1938 hn_set_lro_lenlim(sc, lenlim);
1944 hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS)
1946 struct hn_softc *sc = arg1;
1947 int ackcnt, error, i;
1950 * lro_ackcnt_lim is append count limit,
1951 * +1 to turn it into aggregation limit.
1953 ackcnt = sc->hn_rx_ring[0].hn_lro.lro_ackcnt_lim + 1;
1954 error = sysctl_handle_int(oidp, &ackcnt, 0, req);
1955 if (error || req->newptr == NULL)
1958 if (ackcnt < 2 || ackcnt > (TCP_LRO_ACKCNT_MAX + 1))
1962 * Convert aggregation limit back to append
1967 for (i = 0; i < sc->hn_rx_ring_inuse; ++i)
1968 sc->hn_rx_ring[i].hn_lro.lro_ackcnt_lim = ackcnt;
1976 hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS)
1978 struct hn_softc *sc = arg1;
1983 if (sc->hn_rx_ring[0].hn_trust_hcsum & hcsum)
1986 error = sysctl_handle_int(oidp, &on, 0, req);
1987 if (error || req->newptr == NULL)
1991 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
1992 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
1995 rxr->hn_trust_hcsum |= hcsum;
1997 rxr->hn_trust_hcsum &= ~hcsum;
2004 hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS)
2006 struct hn_softc *sc = arg1;
2007 int chimney_size, error;
2009 chimney_size = sc->hn_tx_ring[0].hn_tx_chimney_size;
2010 error = sysctl_handle_int(oidp, &chimney_size, 0, req);
2011 if (error || req->newptr == NULL)
2014 if (chimney_size > sc->hn_tx_chimney_max || chimney_size <= 0)
2017 hn_set_tx_chimney_size(sc, chimney_size);
2021 #if __FreeBSD_version < 1100095
2023 hn_rx_stat_int_sysctl(SYSCTL_HANDLER_ARGS)
2025 struct hn_softc *sc = arg1;
2026 int ofs = arg2, i, error;
2027 struct hn_rx_ring *rxr;
2031 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2032 rxr = &sc->hn_rx_ring[i];
2033 stat += *((int *)((uint8_t *)rxr + ofs));
2036 error = sysctl_handle_64(oidp, &stat, 0, req);
2037 if (error || req->newptr == NULL)
2040 /* Zero out this stat. */
2041 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2042 rxr = &sc->hn_rx_ring[i];
2043 *((int *)((uint8_t *)rxr + ofs)) = 0;
2049 hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS)
2051 struct hn_softc *sc = arg1;
2052 int ofs = arg2, i, error;
2053 struct hn_rx_ring *rxr;
2057 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2058 rxr = &sc->hn_rx_ring[i];
2059 stat += *((uint64_t *)((uint8_t *)rxr + ofs));
2062 error = sysctl_handle_64(oidp, &stat, 0, req);
2063 if (error || req->newptr == NULL)
2066 /* Zero out this stat. */
2067 for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
2068 rxr = &sc->hn_rx_ring[i];
2069 *((uint64_t *)((uint8_t *)rxr + ofs)) = 0;
2077 hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
2079 struct hn_softc *sc = arg1;
2080 int ofs = arg2, i, error;
2081 struct hn_rx_ring *rxr;
2085 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2086 rxr = &sc->hn_rx_ring[i];
2087 stat += *((u_long *)((uint8_t *)rxr + ofs));
2090 error = sysctl_handle_long(oidp, &stat, 0, req);
2091 if (error || req->newptr == NULL)
2094 /* Zero out this stat. */
2095 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2096 rxr = &sc->hn_rx_ring[i];
2097 *((u_long *)((uint8_t *)rxr + ofs)) = 0;
2103 hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
2105 struct hn_softc *sc = arg1;
2106 int ofs = arg2, i, error;
2107 struct hn_tx_ring *txr;
2111 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2112 txr = &sc->hn_tx_ring[i];
2113 stat += *((u_long *)((uint8_t *)txr + ofs));
2116 error = sysctl_handle_long(oidp, &stat, 0, req);
2117 if (error || req->newptr == NULL)
2120 /* Zero out this stat. */
2121 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2122 txr = &sc->hn_tx_ring[i];
2123 *((u_long *)((uint8_t *)txr + ofs)) = 0;
2129 hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS)
2131 struct hn_softc *sc = arg1;
2132 int ofs = arg2, i, error, conf;
2133 struct hn_tx_ring *txr;
2135 txr = &sc->hn_tx_ring[0];
2136 conf = *((int *)((uint8_t *)txr + ofs));
2138 error = sysctl_handle_int(oidp, &conf, 0, req);
2139 if (error || req->newptr == NULL)
2143 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2144 txr = &sc->hn_tx_ring[i];
2145 *((int *)((uint8_t *)txr + ofs)) = conf;
2153 hn_check_iplen(const struct mbuf *m, int hoff)
2155 const struct ip *ip;
2156 int len, iphlen, iplen;
2157 const struct tcphdr *th;
2158 int thoff; /* TCP data offset */
2160 len = hoff + sizeof(struct ip);
2162 /* The packet must be at least the size of an IP header. */
2163 if (m->m_pkthdr.len < len)
2164 return IPPROTO_DONE;
2166 /* The fixed IP header must reside completely in the first mbuf. */
2168 return IPPROTO_DONE;
2170 ip = mtodo(m, hoff);
2172 /* Bound check the packet's stated IP header length. */
2173 iphlen = ip->ip_hl << 2;
2174 if (iphlen < sizeof(struct ip)) /* minimum header length */
2175 return IPPROTO_DONE;
2177 /* The full IP header must reside completely in the one mbuf. */
2178 if (m->m_len < hoff + iphlen)
2179 return IPPROTO_DONE;
2181 iplen = ntohs(ip->ip_len);
2184 * Check that the amount of data in the buffers is as
2185 * at least much as the IP header would have us expect.
2187 if (m->m_pkthdr.len < hoff + iplen)
2188 return IPPROTO_DONE;
2191 * Ignore IP fragments.
2193 if (ntohs(ip->ip_off) & (IP_OFFMASK | IP_MF))
2194 return IPPROTO_DONE;
2197 * The TCP/IP or UDP/IP header must be entirely contained within
2198 * the first fragment of a packet.
2202 if (iplen < iphlen + sizeof(struct tcphdr))
2203 return IPPROTO_DONE;
2204 if (m->m_len < hoff + iphlen + sizeof(struct tcphdr))
2205 return IPPROTO_DONE;
2206 th = (const struct tcphdr *)((const uint8_t *)ip + iphlen);
2207 thoff = th->th_off << 2;
2208 if (thoff < sizeof(struct tcphdr) || thoff + iphlen > iplen)
2209 return IPPROTO_DONE;
2210 if (m->m_len < hoff + iphlen + thoff)
2211 return IPPROTO_DONE;
2214 if (iplen < iphlen + sizeof(struct udphdr))
2215 return IPPROTO_DONE;
2216 if (m->m_len < hoff + iphlen + sizeof(struct udphdr))
2217 return IPPROTO_DONE;
2221 return IPPROTO_DONE;
2228 hn_create_rx_data(struct hn_softc *sc, int ring_cnt)
2230 struct sysctl_oid_list *child;
2231 struct sysctl_ctx_list *ctx;
2232 device_t dev = sc->hn_dev;
2233 #if defined(INET) || defined(INET6)
2234 #if __FreeBSD_version >= 1100095
2240 sc->hn_rx_ring_cnt = ring_cnt;
2241 sc->hn_rx_ring_inuse = sc->hn_rx_ring_cnt;
2243 sc->hn_rx_ring = malloc(sizeof(struct hn_rx_ring) * sc->hn_rx_ring_cnt,
2244 M_NETVSC, M_WAITOK | M_ZERO);
2246 #if defined(INET) || defined(INET6)
2247 #if __FreeBSD_version >= 1100095
2248 lroent_cnt = hn_lro_entry_count;
2249 if (lroent_cnt < TCP_LRO_ENTRIES)
2250 lroent_cnt = TCP_LRO_ENTRIES;
2251 device_printf(dev, "LRO: entry count %d\n", lroent_cnt);
2253 #endif /* INET || INET6 */
2255 ctx = device_get_sysctl_ctx(dev);
2256 child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
2258 /* Create dev.hn.UNIT.rx sysctl tree */
2259 sc->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "rx",
2260 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2262 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2263 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
2265 if (hn_trust_hosttcp)
2266 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_TCP;
2267 if (hn_trust_hostudp)
2268 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_UDP;
2269 if (hn_trust_hostip)
2270 rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_IP;
2271 rxr->hn_ifp = sc->hn_ifp;
2272 if (i < sc->hn_tx_ring_cnt)
2273 rxr->hn_txr = &sc->hn_tx_ring[i];
2274 rxr->hn_rdbuf = malloc(NETVSC_PACKET_SIZE, M_NETVSC, M_WAITOK);
2280 #if defined(INET) || defined(INET6)
2281 #if __FreeBSD_version >= 1100095
2282 tcp_lro_init_args(&rxr->hn_lro, sc->hn_ifp, lroent_cnt,
2283 hn_lro_mbufq_depth);
2285 tcp_lro_init(&rxr->hn_lro);
2286 rxr->hn_lro.ifp = sc->hn_ifp;
2288 #if __FreeBSD_version >= 1100099
2289 rxr->hn_lro.lro_length_lim = HN_LRO_LENLIM_DEF;
2290 rxr->hn_lro.lro_ackcnt_lim = HN_LRO_ACKCNT_DEF;
2292 #endif /* INET || INET6 */
2294 if (sc->hn_rx_sysctl_tree != NULL) {
2298 * Create per RX ring sysctl tree:
2299 * dev.hn.UNIT.rx.RINGID
2301 snprintf(name, sizeof(name), "%d", i);
2302 rxr->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx,
2303 SYSCTL_CHILDREN(sc->hn_rx_sysctl_tree),
2304 OID_AUTO, name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2306 if (rxr->hn_rx_sysctl_tree != NULL) {
2307 SYSCTL_ADD_ULONG(ctx,
2308 SYSCTL_CHILDREN(rxr->hn_rx_sysctl_tree),
2309 OID_AUTO, "packets", CTLFLAG_RW,
2310 &rxr->hn_pkts, "# of packets received");
2311 SYSCTL_ADD_ULONG(ctx,
2312 SYSCTL_CHILDREN(rxr->hn_rx_sysctl_tree),
2313 OID_AUTO, "rss_pkts", CTLFLAG_RW,
2315 "# of packets w/ RSS info received");
2320 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_queued",
2321 CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2322 __offsetof(struct hn_rx_ring, hn_lro.lro_queued),
2323 #if __FreeBSD_version < 1100095
2324 hn_rx_stat_int_sysctl,
2326 hn_rx_stat_u64_sysctl,
2328 "LU", "LRO queued");
2329 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_flushed",
2330 CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2331 __offsetof(struct hn_rx_ring, hn_lro.lro_flushed),
2332 #if __FreeBSD_version < 1100095
2333 hn_rx_stat_int_sysctl,
2335 hn_rx_stat_u64_sysctl,
2337 "LU", "LRO flushed");
2338 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_tried",
2339 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2340 __offsetof(struct hn_rx_ring, hn_lro_tried),
2341 hn_rx_stat_ulong_sysctl, "LU", "# of LRO tries");
2342 #if __FreeBSD_version >= 1100099
2343 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_length_lim",
2344 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2345 hn_lro_lenlim_sysctl, "IU",
2346 "Max # of data bytes to be aggregated by LRO");
2347 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_ackcnt_lim",
2348 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2349 hn_lro_ackcnt_sysctl, "I",
2350 "Max # of ACKs to be aggregated by LRO");
2352 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hosttcp",
2353 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_TCP,
2354 hn_trust_hcsum_sysctl, "I",
2355 "Trust tcp segement verification on host side, "
2356 "when csum info is missing");
2357 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostudp",
2358 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_UDP,
2359 hn_trust_hcsum_sysctl, "I",
2360 "Trust udp datagram verification on host side, "
2361 "when csum info is missing");
2362 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostip",
2363 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_IP,
2364 hn_trust_hcsum_sysctl, "I",
2365 "Trust ip packet verification on host side, "
2366 "when csum info is missing");
2367 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_ip",
2368 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2369 __offsetof(struct hn_rx_ring, hn_csum_ip),
2370 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM IP");
2371 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_tcp",
2372 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2373 __offsetof(struct hn_rx_ring, hn_csum_tcp),
2374 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM TCP");
2375 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_udp",
2376 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2377 __offsetof(struct hn_rx_ring, hn_csum_udp),
2378 hn_rx_stat_ulong_sysctl, "LU", "RXCSUM UDP");
2379 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_trusted",
2380 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2381 __offsetof(struct hn_rx_ring, hn_csum_trusted),
2382 hn_rx_stat_ulong_sysctl, "LU",
2383 "# of packets that we trust host's csum verification");
2384 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "small_pkts",
2385 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2386 __offsetof(struct hn_rx_ring, hn_small_pkts),
2387 hn_rx_stat_ulong_sysctl, "LU", "# of small packets received");
2388 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_cnt",
2389 CTLFLAG_RD, &sc->hn_rx_ring_cnt, 0, "# created RX rings");
2390 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_inuse",
2391 CTLFLAG_RD, &sc->hn_rx_ring_inuse, 0, "# used RX rings");
2395 hn_destroy_rx_data(struct hn_softc *sc)
2399 if (sc->hn_rx_ring_cnt == 0)
2402 for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
2403 struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
2405 #if defined(INET) || defined(INET6)
2406 tcp_lro_free(&rxr->hn_lro);
2408 free(rxr->hn_rdbuf, M_NETVSC);
2410 free(sc->hn_rx_ring, M_NETVSC);
2411 sc->hn_rx_ring = NULL;
2413 sc->hn_rx_ring_cnt = 0;
2414 sc->hn_rx_ring_inuse = 0;
2418 hn_create_tx_ring(struct hn_softc *sc, int id)
2420 struct hn_tx_ring *txr = &sc->hn_tx_ring[id];
2421 device_t dev = sc->hn_dev;
2422 bus_dma_tag_t parent_dtag;
2427 txr->hn_tx_idx = id;
2429 #ifndef HN_USE_TXDESC_BUFRING
2430 mtx_init(&txr->hn_txlist_spin, "hn txlist", NULL, MTX_SPIN);
2432 mtx_init(&txr->hn_tx_lock, "hn tx", NULL, MTX_DEF);
2434 txr->hn_txdesc_cnt = HN_TX_DESC_CNT;
2435 txr->hn_txdesc = malloc(sizeof(struct hn_txdesc) * txr->hn_txdesc_cnt,
2436 M_NETVSC, M_WAITOK | M_ZERO);
2437 #ifndef HN_USE_TXDESC_BUFRING
2438 SLIST_INIT(&txr->hn_txlist);
2440 txr->hn_txdesc_br = buf_ring_alloc(txr->hn_txdesc_cnt, M_NETVSC,
2441 M_WAITOK, &txr->hn_tx_lock);
2444 txr->hn_tx_taskq = sc->hn_tx_taskq;
2446 if (hn_use_if_start) {
2447 txr->hn_txeof = hn_start_txeof;
2448 TASK_INIT(&txr->hn_tx_task, 0, hn_start_taskfunc, txr);
2449 TASK_INIT(&txr->hn_txeof_task, 0, hn_start_txeof_taskfunc, txr);
2453 txr->hn_txeof = hn_xmit_txeof;
2454 TASK_INIT(&txr->hn_tx_task, 0, hn_xmit_taskfunc, txr);
2455 TASK_INIT(&txr->hn_txeof_task, 0, hn_xmit_txeof_taskfunc, txr);
2457 br_depth = hn_get_txswq_depth(txr);
2458 txr->hn_mbuf_br = buf_ring_alloc(br_depth, M_NETVSC,
2459 M_WAITOK, &txr->hn_tx_lock);
2462 txr->hn_direct_tx_size = hn_direct_tx_size;
2463 version = VMBUS_GET_VERSION(device_get_parent(dev), dev);
2464 if (version >= VMBUS_VERSION_WIN8_1) {
2465 txr->hn_csum_assist = HN_CSUM_ASSIST;
2467 txr->hn_csum_assist = HN_CSUM_ASSIST_WIN8;
2469 device_printf(dev, "bus version %u.%u, "
2470 "no UDP checksum offloading\n",
2471 VMBUS_VERSION_MAJOR(version),
2472 VMBUS_VERSION_MINOR(version));
2477 * Always schedule transmission instead of trying to do direct
2478 * transmission. This one gives the best performance so far.
2480 txr->hn_sched_tx = 1;
2482 parent_dtag = bus_get_dma_tag(dev);
2484 /* DMA tag for RNDIS messages. */
2485 error = bus_dma_tag_create(parent_dtag, /* parent */
2486 HN_RNDIS_MSG_ALIGN, /* alignment */
2487 HN_RNDIS_MSG_BOUNDARY, /* boundary */
2488 BUS_SPACE_MAXADDR, /* lowaddr */
2489 BUS_SPACE_MAXADDR, /* highaddr */
2490 NULL, NULL, /* filter, filterarg */
2491 HN_RNDIS_MSG_LEN, /* maxsize */
2493 HN_RNDIS_MSG_LEN, /* maxsegsize */
2495 NULL, /* lockfunc */
2496 NULL, /* lockfuncarg */
2497 &txr->hn_tx_rndis_dtag);
2499 device_printf(dev, "failed to create rndis dmatag\n");
2503 /* DMA tag for data. */
2504 error = bus_dma_tag_create(parent_dtag, /* parent */
2506 HN_TX_DATA_BOUNDARY, /* boundary */
2507 BUS_SPACE_MAXADDR, /* lowaddr */
2508 BUS_SPACE_MAXADDR, /* highaddr */
2509 NULL, NULL, /* filter, filterarg */
2510 HN_TX_DATA_MAXSIZE, /* maxsize */
2511 HN_TX_DATA_SEGCNT_MAX, /* nsegments */
2512 HN_TX_DATA_SEGSIZE, /* maxsegsize */
2514 NULL, /* lockfunc */
2515 NULL, /* lockfuncarg */
2516 &txr->hn_tx_data_dtag);
2518 device_printf(dev, "failed to create data dmatag\n");
2522 for (i = 0; i < txr->hn_txdesc_cnt; ++i) {
2523 struct hn_txdesc *txd = &txr->hn_txdesc[i];
2528 * Allocate and load RNDIS messages.
2530 error = bus_dmamem_alloc(txr->hn_tx_rndis_dtag,
2531 (void **)&txd->rndis_msg,
2532 BUS_DMA_WAITOK | BUS_DMA_COHERENT,
2533 &txd->rndis_msg_dmap);
2536 "failed to allocate rndis_msg, %d\n", i);
2540 error = bus_dmamap_load(txr->hn_tx_rndis_dtag,
2541 txd->rndis_msg_dmap,
2542 txd->rndis_msg, HN_RNDIS_MSG_LEN,
2543 hyperv_dma_map_paddr, &txd->rndis_msg_paddr,
2547 "failed to load rndis_msg, %d\n", i);
2548 bus_dmamem_free(txr->hn_tx_rndis_dtag,
2549 txd->rndis_msg, txd->rndis_msg_dmap);
2553 /* DMA map for TX data. */
2554 error = bus_dmamap_create(txr->hn_tx_data_dtag, 0,
2558 "failed to allocate tx data dmamap\n");
2559 bus_dmamap_unload(txr->hn_tx_rndis_dtag,
2560 txd->rndis_msg_dmap);
2561 bus_dmamem_free(txr->hn_tx_rndis_dtag,
2562 txd->rndis_msg, txd->rndis_msg_dmap);
2566 /* All set, put it to list */
2567 txd->flags |= HN_TXD_FLAG_ONLIST;
2568 #ifndef HN_USE_TXDESC_BUFRING
2569 SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
2571 buf_ring_enqueue(txr->hn_txdesc_br, txd);
2574 txr->hn_txdesc_avail = txr->hn_txdesc_cnt;
2576 if (sc->hn_tx_sysctl_tree != NULL) {
2577 struct sysctl_oid_list *child;
2578 struct sysctl_ctx_list *ctx;
2582 * Create per TX ring sysctl tree:
2583 * dev.hn.UNIT.tx.RINGID
2585 ctx = device_get_sysctl_ctx(dev);
2586 child = SYSCTL_CHILDREN(sc->hn_tx_sysctl_tree);
2588 snprintf(name, sizeof(name), "%d", id);
2589 txr->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO,
2590 name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2592 if (txr->hn_tx_sysctl_tree != NULL) {
2593 child = SYSCTL_CHILDREN(txr->hn_tx_sysctl_tree);
2595 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_avail",
2596 CTLFLAG_RD, &txr->hn_txdesc_avail, 0,
2597 "# of available TX descs");
2598 if (!hn_use_if_start) {
2599 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "oactive",
2600 CTLFLAG_RD, &txr->hn_oactive, 0,
2603 SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "packets",
2604 CTLFLAG_RW, &txr->hn_pkts,
2605 "# of packets transmitted");
2613 hn_txdesc_dmamap_destroy(struct hn_txdesc *txd)
2615 struct hn_tx_ring *txr = txd->txr;
2617 KASSERT(txd->m == NULL, ("still has mbuf installed"));
2618 KASSERT((txd->flags & HN_TXD_FLAG_DMAMAP) == 0, ("still dma mapped"));
2620 bus_dmamap_unload(txr->hn_tx_rndis_dtag, txd->rndis_msg_dmap);
2621 bus_dmamem_free(txr->hn_tx_rndis_dtag, txd->rndis_msg,
2622 txd->rndis_msg_dmap);
2623 bus_dmamap_destroy(txr->hn_tx_data_dtag, txd->data_dmap);
2627 hn_destroy_tx_ring(struct hn_tx_ring *txr)
2629 struct hn_txdesc *txd;
2631 if (txr->hn_txdesc == NULL)
2634 #ifndef HN_USE_TXDESC_BUFRING
2635 while ((txd = SLIST_FIRST(&txr->hn_txlist)) != NULL) {
2636 SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
2637 hn_txdesc_dmamap_destroy(txd);
2640 mtx_lock(&txr->hn_tx_lock);
2641 while ((txd = buf_ring_dequeue_sc(txr->hn_txdesc_br)) != NULL)
2642 hn_txdesc_dmamap_destroy(txd);
2643 mtx_unlock(&txr->hn_tx_lock);
2646 if (txr->hn_tx_data_dtag != NULL)
2647 bus_dma_tag_destroy(txr->hn_tx_data_dtag);
2648 if (txr->hn_tx_rndis_dtag != NULL)
2649 bus_dma_tag_destroy(txr->hn_tx_rndis_dtag);
2651 #ifdef HN_USE_TXDESC_BUFRING
2652 buf_ring_free(txr->hn_txdesc_br, M_NETVSC);
2655 free(txr->hn_txdesc, M_NETVSC);
2656 txr->hn_txdesc = NULL;
2658 if (txr->hn_mbuf_br != NULL)
2659 buf_ring_free(txr->hn_mbuf_br, M_NETVSC);
2661 #ifndef HN_USE_TXDESC_BUFRING
2662 mtx_destroy(&txr->hn_txlist_spin);
2664 mtx_destroy(&txr->hn_tx_lock);
2668 hn_create_tx_data(struct hn_softc *sc, int ring_cnt)
2670 struct sysctl_oid_list *child;
2671 struct sysctl_ctx_list *ctx;
2674 sc->hn_tx_ring_cnt = ring_cnt;
2675 sc->hn_tx_ring_inuse = sc->hn_tx_ring_cnt;
2677 sc->hn_tx_ring = malloc(sizeof(struct hn_tx_ring) * sc->hn_tx_ring_cnt,
2678 M_NETVSC, M_WAITOK | M_ZERO);
2680 ctx = device_get_sysctl_ctx(sc->hn_dev);
2681 child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->hn_dev));
2683 /* Create dev.hn.UNIT.tx sysctl tree */
2684 sc->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "tx",
2685 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
2687 for (i = 0; i < sc->hn_tx_ring_cnt; ++i) {
2690 error = hn_create_tx_ring(sc, i);
2695 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "no_txdescs",
2696 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2697 __offsetof(struct hn_tx_ring, hn_no_txdescs),
2698 hn_tx_stat_ulong_sysctl, "LU", "# of times short of TX descs");
2699 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "send_failed",
2700 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2701 __offsetof(struct hn_tx_ring, hn_send_failed),
2702 hn_tx_stat_ulong_sysctl, "LU", "# of hyper-v sending failure");
2703 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "txdma_failed",
2704 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2705 __offsetof(struct hn_tx_ring, hn_txdma_failed),
2706 hn_tx_stat_ulong_sysctl, "LU", "# of TX DMA failure");
2707 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_collapsed",
2708 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2709 __offsetof(struct hn_tx_ring, hn_tx_collapsed),
2710 hn_tx_stat_ulong_sysctl, "LU", "# of TX mbuf collapsed");
2711 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney",
2712 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2713 __offsetof(struct hn_tx_ring, hn_tx_chimney),
2714 hn_tx_stat_ulong_sysctl, "LU", "# of chimney send");
2715 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_tried",
2716 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2717 __offsetof(struct hn_tx_ring, hn_tx_chimney_tried),
2718 hn_tx_stat_ulong_sysctl, "LU", "# of chimney send tries");
2719 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_cnt",
2720 CTLFLAG_RD, &sc->hn_tx_ring[0].hn_txdesc_cnt, 0,
2721 "# of total TX descs");
2722 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_chimney_max",
2723 CTLFLAG_RD, &sc->hn_tx_chimney_max, 0,
2724 "Chimney send packet size upper boundary");
2725 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_size",
2726 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
2727 hn_tx_chimney_size_sysctl,
2728 "I", "Chimney send packet size limit");
2729 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "direct_tx_size",
2730 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2731 __offsetof(struct hn_tx_ring, hn_direct_tx_size),
2732 hn_tx_conf_int_sysctl, "I",
2733 "Size of the packet for direct transmission");
2734 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "sched_tx",
2735 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
2736 __offsetof(struct hn_tx_ring, hn_sched_tx),
2737 hn_tx_conf_int_sysctl, "I",
2738 "Always schedule transmission "
2739 "instead of doing direct transmission");
2740 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_cnt",
2741 CTLFLAG_RD, &sc->hn_tx_ring_cnt, 0, "# created TX rings");
2742 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_inuse",
2743 CTLFLAG_RD, &sc->hn_tx_ring_inuse, 0, "# used TX rings");
2749 hn_set_tx_chimney_size(struct hn_softc *sc, int chimney_size)
2754 for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
2755 sc->hn_tx_ring[i].hn_tx_chimney_size = chimney_size;
2760 hn_destroy_tx_data(struct hn_softc *sc)
2764 if (sc->hn_tx_ring_cnt == 0)
2767 for (i = 0; i < sc->hn_tx_ring_cnt; ++i)
2768 hn_destroy_tx_ring(&sc->hn_tx_ring[i]);
2770 free(sc->hn_tx_ring, M_NETVSC);
2771 sc->hn_tx_ring = NULL;
2773 sc->hn_tx_ring_cnt = 0;
2774 sc->hn_tx_ring_inuse = 0;
2778 hn_start_taskfunc(void *xtxr, int pending __unused)
2780 struct hn_tx_ring *txr = xtxr;
2782 mtx_lock(&txr->hn_tx_lock);
2783 hn_start_locked(txr, 0);
2784 mtx_unlock(&txr->hn_tx_lock);
2788 hn_start_txeof_taskfunc(void *xtxr, int pending __unused)
2790 struct hn_tx_ring *txr = xtxr;
2792 mtx_lock(&txr->hn_tx_lock);
2793 atomic_clear_int(&txr->hn_sc->hn_ifp->if_drv_flags, IFF_DRV_OACTIVE);
2794 hn_start_locked(txr, 0);
2795 mtx_unlock(&txr->hn_tx_lock);
2799 hn_stop_tx_tasks(struct hn_softc *sc)
2803 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2804 struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
2806 taskqueue_drain(txr->hn_tx_taskq, &txr->hn_tx_task);
2807 taskqueue_drain(txr->hn_tx_taskq, &txr->hn_txeof_task);
2812 hn_xmit(struct hn_tx_ring *txr, int len)
2814 struct hn_softc *sc = txr->hn_sc;
2815 struct ifnet *ifp = sc->hn_ifp;
2816 struct mbuf *m_head;
2818 mtx_assert(&txr->hn_tx_lock, MA_OWNED);
2819 KASSERT(hn_use_if_start == 0,
2820 ("hn_xmit is called, when if_start is enabled"));
2822 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || txr->hn_oactive)
2825 while ((m_head = drbr_peek(ifp, txr->hn_mbuf_br)) != NULL) {
2826 struct hn_txdesc *txd;
2829 if (len > 0 && m_head->m_pkthdr.len > len) {
2831 * This sending could be time consuming; let callers
2832 * dispatch this packet sending (and sending of any
2833 * following up packets) to tx taskqueue.
2835 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2839 txd = hn_txdesc_get(txr);
2841 txr->hn_no_txdescs++;
2842 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2843 txr->hn_oactive = 1;
2847 error = hn_encap(txr, txd, &m_head);
2849 /* Both txd and m_head are freed; discard */
2850 drbr_advance(ifp, txr->hn_mbuf_br);
2854 error = hn_send_pkt(ifp, txr, txd);
2855 if (__predict_false(error)) {
2856 /* txd is freed, but m_head is not */
2857 drbr_putback(ifp, txr->hn_mbuf_br, m_head);
2858 txr->hn_oactive = 1;
2863 drbr_advance(ifp, txr->hn_mbuf_br);
2869 hn_transmit(struct ifnet *ifp, struct mbuf *m)
2871 struct hn_softc *sc = ifp->if_softc;
2872 struct hn_tx_ring *txr;
2876 * Select the TX ring based on flowid
2878 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2879 idx = m->m_pkthdr.flowid % sc->hn_tx_ring_inuse;
2880 txr = &sc->hn_tx_ring[idx];
2882 error = drbr_enqueue(ifp, txr->hn_mbuf_br, m);
2884 if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1);
2888 if (txr->hn_oactive)
2891 if (txr->hn_sched_tx)
2894 if (mtx_trylock(&txr->hn_tx_lock)) {
2897 sched = hn_xmit(txr, txr->hn_direct_tx_size);
2898 mtx_unlock(&txr->hn_tx_lock);
2903 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
2908 hn_xmit_qflush(struct ifnet *ifp)
2910 struct hn_softc *sc = ifp->if_softc;
2913 for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
2914 struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
2917 mtx_lock(&txr->hn_tx_lock);
2918 while ((m = buf_ring_dequeue_sc(txr->hn_mbuf_br)) != NULL)
2920 mtx_unlock(&txr->hn_tx_lock);
2926 hn_xmit_txeof(struct hn_tx_ring *txr)
2929 if (txr->hn_sched_tx)
2932 if (mtx_trylock(&txr->hn_tx_lock)) {
2935 txr->hn_oactive = 0;
2936 sched = hn_xmit(txr, txr->hn_direct_tx_size);
2937 mtx_unlock(&txr->hn_tx_lock);
2939 taskqueue_enqueue(txr->hn_tx_taskq,
2945 * Release the oactive earlier, with the hope, that
2946 * others could catch up. The task will clear the
2947 * oactive again with the hn_tx_lock to avoid possible
2950 txr->hn_oactive = 0;
2951 taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
2956 hn_xmit_taskfunc(void *xtxr, int pending __unused)
2958 struct hn_tx_ring *txr = xtxr;
2960 mtx_lock(&txr->hn_tx_lock);
2962 mtx_unlock(&txr->hn_tx_lock);
2966 hn_xmit_txeof_taskfunc(void *xtxr, int pending __unused)
2968 struct hn_tx_ring *txr = xtxr;
2970 mtx_lock(&txr->hn_tx_lock);
2971 txr->hn_oactive = 0;
2973 mtx_unlock(&txr->hn_tx_lock);
2977 hn_channel_attach(struct hn_softc *sc, struct vmbus_channel *chan)
2979 struct hn_rx_ring *rxr;
2982 idx = vmbus_chan_subidx(chan);
2984 KASSERT(idx >= 0 && idx < sc->hn_rx_ring_inuse,
2985 ("invalid channel index %d, should > 0 && < %d",
2986 idx, sc->hn_rx_ring_inuse));
2987 rxr = &sc->hn_rx_ring[idx];
2988 KASSERT((rxr->hn_rx_flags & HN_RX_FLAG_ATTACHED) == 0,
2989 ("RX ring %d already attached", idx));
2990 rxr->hn_rx_flags |= HN_RX_FLAG_ATTACHED;
2993 if_printf(sc->hn_ifp, "link RX ring %d to channel%u\n",
2994 idx, vmbus_chan_id(chan));
2997 if (idx < sc->hn_tx_ring_inuse) {
2998 struct hn_tx_ring *txr = &sc->hn_tx_ring[idx];
3000 KASSERT((txr->hn_tx_flags & HN_TX_FLAG_ATTACHED) == 0,
3001 ("TX ring %d already attached", idx));
3002 txr->hn_tx_flags |= HN_TX_FLAG_ATTACHED;
3004 txr->hn_chan = chan;
3006 if_printf(sc->hn_ifp, "link TX ring %d to channel%u\n",
3007 idx, vmbus_chan_id(chan));
3011 /* Bind channel to a proper CPU */
3012 vmbus_chan_cpu_set(chan, (sc->hn_cpu + idx) % mp_ncpus);
3016 hn_subchan_attach(struct hn_softc *sc, struct vmbus_channel *chan)
3019 KASSERT(!vmbus_chan_is_primary(chan),
3020 ("subchannel callback on primary channel"));
3021 hn_channel_attach(sc, chan);
3025 hn_subchan_setup(struct hn_softc *sc)
3027 struct vmbus_channel **subchans;
3028 int subchan_cnt = sc->net_dev->num_channel - 1;
3031 /* Wait for sub-channels setup to complete. */
3032 subchans = vmbus_subchan_get(sc->hn_prichan, subchan_cnt);
3034 /* Attach the sub-channels. */
3035 for (i = 0; i < subchan_cnt; ++i) {
3036 struct vmbus_channel *subchan = subchans[i];
3038 /* NOTE: Calling order is critical. */
3039 hn_subchan_attach(sc, subchan);
3040 hv_nv_subchan_attach(subchan,
3041 &sc->hn_rx_ring[vmbus_chan_subidx(subchan)]);
3044 /* Release the sub-channels */
3045 vmbus_subchan_rel(subchans, subchan_cnt);
3046 if_printf(sc->hn_ifp, "%d sub-channels setup done\n", subchan_cnt);
3050 hn_tx_taskq_create(void *arg __unused)
3052 if (!hn_share_tx_taskq)
3055 hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
3056 taskqueue_thread_enqueue, &hn_tx_taskq);
3057 taskqueue_start_threads(&hn_tx_taskq, 1, PI_NET, "hn tx");
3058 if (hn_bind_tx_taskq >= 0) {
3059 int cpu = hn_bind_tx_taskq;
3060 struct task cpuset_task;
3063 if (cpu > mp_ncpus - 1)
3065 CPU_SETOF(cpu, &cpu_set);
3066 TASK_INIT(&cpuset_task, 0, hn_cpuset_setthread_task, &cpu_set);
3067 taskqueue_enqueue(hn_tx_taskq, &cpuset_task);
3068 taskqueue_drain(hn_tx_taskq, &cpuset_task);
3071 SYSINIT(hn_txtq_create, SI_SUB_DRIVERS, SI_ORDER_FIRST,
3072 hn_tx_taskq_create, NULL);
3075 hn_tx_taskq_destroy(void *arg __unused)
3077 if (hn_tx_taskq != NULL)
3078 taskqueue_free(hn_tx_taskq);
3080 SYSUNINIT(hn_txtq_destroy, SI_SUB_DRIVERS, SI_ORDER_FIRST,
3081 hn_tx_taskq_destroy, NULL);
3083 static device_method_t netvsc_methods[] = {
3084 /* Device interface */
3085 DEVMETHOD(device_probe, netvsc_probe),
3086 DEVMETHOD(device_attach, netvsc_attach),
3087 DEVMETHOD(device_detach, netvsc_detach),
3088 DEVMETHOD(device_shutdown, netvsc_shutdown),
3093 static driver_t netvsc_driver = {
3099 static devclass_t netvsc_devclass;
3101 DRIVER_MODULE(hn, vmbus, netvsc_driver, netvsc_devclass, 0, 0);
3102 MODULE_VERSION(hn, 1);
3103 MODULE_DEPEND(hn, vmbus, 1, 1, 1);