2 * Copyright (c) 2001-2011, Intel Corporation
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions are met:
8 * 1. Redistributions of source code must retain the above copyright notice,
9 * this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the Intel Corporation nor the names of its
16 * contributors may be used to endorse or promote products derived from
17 * this software without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
20 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
23 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
32 #include "opt_ifpoll.h"
35 #include <sys/param.h>
37 #include <sys/endian.h>
38 #include <sys/interrupt.h>
39 #include <sys/kernel.h>
40 #include <sys/malloc.h>
44 #include <sys/serialize.h>
45 #include <sys/serialize2.h>
46 #include <sys/socket.h>
47 #include <sys/sockio.h>
48 #include <sys/sysctl.h>
49 #include <sys/systm.h>
52 #include <net/ethernet.h>
54 #include <net/if_arp.h>
55 #include <net/if_dl.h>
56 #include <net/if_media.h>
57 #include <net/ifq_var.h>
58 #include <net/toeplitz.h>
59 #include <net/toeplitz2.h>
60 #include <net/vlan/if_vlan_var.h>
61 #include <net/vlan/if_vlan_ether.h>
62 #include <net/if_poll.h>
64 #include <netinet/in_systm.h>
65 #include <netinet/in.h>
66 #include <netinet/ip.h>
67 #include <netinet/tcp.h>
68 #include <netinet/udp.h>
70 #include <bus/pci/pcivar.h>
71 #include <bus/pci/pcireg.h>
73 #include <dev/netif/ig_hal/e1000_api.h>
74 #include <dev/netif/ig_hal/e1000_82575.h>
75 #include <dev/netif/igb/if_igb.h>
78 #define IGB_RSS_DPRINTF(sc, lvl, fmt, ...) \
80 if (sc->rss_debug >= lvl) \
81 if_printf(&sc->arpcom.ac_if, fmt, __VA_ARGS__); \
83 #else /* !IGB_RSS_DEBUG */
84 #define IGB_RSS_DPRINTF(sc, lvl, fmt, ...) ((void)0)
85 #endif /* IGB_RSS_DEBUG */
87 #define IGB_NAME "Intel(R) PRO/1000 "
88 #define IGB_DEVICE(id) \
89 { IGB_VENDOR_ID, E1000_DEV_ID_##id, IGB_NAME #id }
90 #define IGB_DEVICE_NULL { 0, 0, NULL }
92 static struct igb_device {
97 IGB_DEVICE(82575EB_COPPER),
98 IGB_DEVICE(82575EB_FIBER_SERDES),
99 IGB_DEVICE(82575GB_QUAD_COPPER),
101 IGB_DEVICE(82576_NS),
102 IGB_DEVICE(82576_NS_SERDES),
103 IGB_DEVICE(82576_FIBER),
104 IGB_DEVICE(82576_SERDES),
105 IGB_DEVICE(82576_SERDES_QUAD),
106 IGB_DEVICE(82576_QUAD_COPPER),
107 IGB_DEVICE(82576_QUAD_COPPER_ET2),
108 IGB_DEVICE(82576_VF),
109 IGB_DEVICE(82580_COPPER),
110 IGB_DEVICE(82580_FIBER),
111 IGB_DEVICE(82580_SERDES),
112 IGB_DEVICE(82580_SGMII),
113 IGB_DEVICE(82580_COPPER_DUAL),
114 IGB_DEVICE(82580_QUAD_FIBER),
115 IGB_DEVICE(DH89XXCC_SERDES),
116 IGB_DEVICE(DH89XXCC_SGMII),
117 IGB_DEVICE(DH89XXCC_SFP),
118 IGB_DEVICE(DH89XXCC_BACKPLANE),
119 IGB_DEVICE(I350_COPPER),
120 IGB_DEVICE(I350_FIBER),
121 IGB_DEVICE(I350_SERDES),
122 IGB_DEVICE(I350_SGMII),
125 /* required last entry */
129 static int igb_probe(device_t);
130 static int igb_attach(device_t);
131 static int igb_detach(device_t);
132 static int igb_shutdown(device_t);
133 static int igb_suspend(device_t);
134 static int igb_resume(device_t);
136 static boolean_t igb_is_valid_ether_addr(const uint8_t *);
137 static void igb_setup_ifp(struct igb_softc *);
138 static boolean_t igb_txcsum_ctx(struct igb_tx_ring *, struct mbuf *);
139 static int igb_tso_pullup(struct igb_tx_ring *, struct mbuf **);
140 static void igb_tso_ctx(struct igb_tx_ring *, struct mbuf *, uint32_t *);
141 static void igb_add_sysctl(struct igb_softc *);
142 static int igb_sysctl_intr_rate(SYSCTL_HANDLER_ARGS);
143 static int igb_sysctl_msix_rate(SYSCTL_HANDLER_ARGS);
144 static int igb_sysctl_tx_intr_nsegs(SYSCTL_HANDLER_ARGS);
145 static int igb_sysctl_tx_wreg_nsegs(SYSCTL_HANDLER_ARGS);
146 static int igb_sysctl_rx_wreg_nsegs(SYSCTL_HANDLER_ARGS);
147 static void igb_set_ring_inuse(struct igb_softc *, boolean_t);
148 static int igb_get_rxring_inuse(const struct igb_softc *, boolean_t);
149 static int igb_get_txring_inuse(const struct igb_softc *, boolean_t);
151 static int igb_sysctl_npoll_rxoff(SYSCTL_HANDLER_ARGS);
152 static int igb_sysctl_npoll_txoff(SYSCTL_HANDLER_ARGS);
155 static void igb_vf_init_stats(struct igb_softc *);
156 static void igb_reset(struct igb_softc *);
157 static void igb_update_stats_counters(struct igb_softc *);
158 static void igb_update_vf_stats_counters(struct igb_softc *);
159 static void igb_update_link_status(struct igb_softc *);
160 static void igb_init_tx_unit(struct igb_softc *);
161 static void igb_init_rx_unit(struct igb_softc *);
163 static void igb_set_vlan(struct igb_softc *);
164 static void igb_set_multi(struct igb_softc *);
165 static void igb_set_promisc(struct igb_softc *);
166 static void igb_disable_promisc(struct igb_softc *);
168 static int igb_alloc_rings(struct igb_softc *);
169 static void igb_free_rings(struct igb_softc *);
170 static int igb_create_tx_ring(struct igb_tx_ring *);
171 static int igb_create_rx_ring(struct igb_rx_ring *);
172 static void igb_free_tx_ring(struct igb_tx_ring *);
173 static void igb_free_rx_ring(struct igb_rx_ring *);
174 static void igb_destroy_tx_ring(struct igb_tx_ring *, int);
175 static void igb_destroy_rx_ring(struct igb_rx_ring *, int);
176 static void igb_init_tx_ring(struct igb_tx_ring *);
177 static int igb_init_rx_ring(struct igb_rx_ring *);
178 static int igb_newbuf(struct igb_rx_ring *, int, boolean_t);
179 static int igb_encap(struct igb_tx_ring *, struct mbuf **, int *, int *);
180 static void igb_rx_refresh(struct igb_rx_ring *, int);
182 static void igb_stop(struct igb_softc *);
183 static void igb_init(void *);
184 static int igb_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
185 static void igb_media_status(struct ifnet *, struct ifmediareq *);
186 static int igb_media_change(struct ifnet *);
187 static void igb_timer(void *);
188 static void igb_watchdog(struct ifaltq_subque *);
189 static void igb_start(struct ifnet *, struct ifaltq_subque *);
191 static void igb_npoll(struct ifnet *, struct ifpoll_info *);
192 static void igb_npoll_rx(struct ifnet *, void *, int);
193 static void igb_npoll_tx(struct ifnet *, void *, int);
194 static void igb_npoll_status(struct ifnet *);
196 static void igb_serialize(struct ifnet *, enum ifnet_serialize);
197 static void igb_deserialize(struct ifnet *, enum ifnet_serialize);
198 static int igb_tryserialize(struct ifnet *, enum ifnet_serialize);
200 static void igb_serialize_assert(struct ifnet *, enum ifnet_serialize,
204 static void igb_intr(void *);
205 static void igb_intr_shared(void *);
206 static void igb_rxeof(struct igb_rx_ring *, int);
207 static void igb_txeof(struct igb_tx_ring *);
208 static void igb_set_eitr(struct igb_softc *, int, int);
209 static void igb_enable_intr(struct igb_softc *);
210 static void igb_disable_intr(struct igb_softc *);
211 static void igb_init_unshared_intr(struct igb_softc *);
212 static void igb_init_intr(struct igb_softc *);
213 static int igb_setup_intr(struct igb_softc *);
214 static void igb_set_txintr_mask(struct igb_tx_ring *, int *, int);
215 static void igb_set_rxintr_mask(struct igb_rx_ring *, int *, int);
216 static void igb_set_intr_mask(struct igb_softc *);
217 static int igb_alloc_intr(struct igb_softc *);
218 static void igb_free_intr(struct igb_softc *);
219 static void igb_teardown_intr(struct igb_softc *);
220 static void igb_msix_try_alloc(struct igb_softc *);
221 static void igb_msix_free(struct igb_softc *, boolean_t);
222 static int igb_msix_setup(struct igb_softc *);
223 static void igb_msix_teardown(struct igb_softc *, int);
224 static void igb_msix_rx(void *);
225 static void igb_msix_tx(void *);
226 static void igb_msix_status(void *);
228 /* Management and WOL Support */
229 static void igb_get_mgmt(struct igb_softc *);
230 static void igb_rel_mgmt(struct igb_softc *);
231 static void igb_get_hw_control(struct igb_softc *);
232 static void igb_rel_hw_control(struct igb_softc *);
233 static void igb_enable_wol(device_t);
235 static device_method_t igb_methods[] = {
236 /* Device interface */
237 DEVMETHOD(device_probe, igb_probe),
238 DEVMETHOD(device_attach, igb_attach),
239 DEVMETHOD(device_detach, igb_detach),
240 DEVMETHOD(device_shutdown, igb_shutdown),
241 DEVMETHOD(device_suspend, igb_suspend),
242 DEVMETHOD(device_resume, igb_resume),
246 static driver_t igb_driver = {
249 sizeof(struct igb_softc),
252 static devclass_t igb_devclass;
254 DECLARE_DUMMY_MODULE(if_igb);
255 MODULE_DEPEND(igb, ig_hal, 1, 1, 1);
256 DRIVER_MODULE(if_igb, pci, igb_driver, igb_devclass, NULL, NULL);
258 static int igb_rxd = IGB_DEFAULT_RXD;
259 static int igb_txd = IGB_DEFAULT_TXD;
260 static int igb_rxr = 0;
261 static int igb_txr = 0;
262 static int igb_msi_enable = 1;
263 static int igb_msix_enable = 1;
264 static int igb_eee_disabled = 1; /* Energy Efficient Ethernet */
265 static int igb_fc_setting = e1000_fc_full;
268 * DMA Coalescing, only for i350 - default to off,
269 * this feature is for power savings
271 static int igb_dma_coalesce = 0;
273 TUNABLE_INT("hw.igb.rxd", &igb_rxd);
274 TUNABLE_INT("hw.igb.txd", &igb_txd);
275 TUNABLE_INT("hw.igb.rxr", &igb_rxr);
276 TUNABLE_INT("hw.igb.txr", &igb_txr);
277 TUNABLE_INT("hw.igb.msi.enable", &igb_msi_enable);
278 TUNABLE_INT("hw.igb.msix.enable", &igb_msix_enable);
279 TUNABLE_INT("hw.igb.fc_setting", &igb_fc_setting);
282 TUNABLE_INT("hw.igb.eee_disabled", &igb_eee_disabled);
283 TUNABLE_INT("hw.igb.dma_coalesce", &igb_dma_coalesce);
286 igb_rxcsum(uint32_t staterr, struct mbuf *mp)
288 /* Ignore Checksum bit is set */
289 if (staterr & E1000_RXD_STAT_IXSM)
292 if ((staterr & (E1000_RXD_STAT_IPCS | E1000_RXDEXT_STATERR_IPE)) ==
294 mp->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID;
296 if (staterr & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)) {
297 if ((staterr & E1000_RXDEXT_STATERR_TCPE) == 0) {
298 mp->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
299 CSUM_PSEUDO_HDR | CSUM_FRAG_NOT_CHECKED;
300 mp->m_pkthdr.csum_data = htons(0xffff);
305 static __inline struct pktinfo *
306 igb_rssinfo(struct mbuf *m, struct pktinfo *pi,
307 uint32_t hash, uint32_t hashtype, uint32_t staterr)
310 case E1000_RXDADV_RSSTYPE_IPV4_TCP:
311 pi->pi_netisr = NETISR_IP;
313 pi->pi_l3proto = IPPROTO_TCP;
316 case E1000_RXDADV_RSSTYPE_IPV4:
317 if (staterr & E1000_RXD_STAT_IXSM)
321 (E1000_RXD_STAT_TCPCS | E1000_RXDEXT_STATERR_TCPE)) ==
322 E1000_RXD_STAT_TCPCS) {
323 pi->pi_netisr = NETISR_IP;
325 pi->pi_l3proto = IPPROTO_UDP;
333 m->m_flags |= M_HASH;
334 m->m_pkthdr.hash = toeplitz_hash(hash);
339 igb_probe(device_t dev)
341 const struct igb_device *d;
344 vid = pci_get_vendor(dev);
345 did = pci_get_device(dev);
347 for (d = igb_devices; d->desc != NULL; ++d) {
348 if (vid == d->vid && did == d->did) {
349 device_set_desc(dev, d->desc);
357 igb_attach(device_t dev)
359 struct igb_softc *sc = device_get_softc(dev);
360 uint16_t eeprom_data;
361 int error = 0, i, j, ring_max;
363 int offset, offset_def;
368 SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev),
369 SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
370 OID_AUTO, "nvm", CTLTYPE_INT|CTLFLAG_RW, adapter, 0,
371 igb_sysctl_nvm_info, "I", "NVM Information");
372 SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev),
373 SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
374 OID_AUTO, "flow_control", CTLTYPE_INT|CTLFLAG_RW,
375 adapter, 0, igb_set_flowcntl, "I", "Flow Control");
378 callout_init_mp(&sc->timer);
379 lwkt_serialize_init(&sc->main_serialize);
381 if_initname(&sc->arpcom.ac_if, device_get_name(dev),
382 device_get_unit(dev));
383 sc->dev = sc->osdep.dev = dev;
386 * Determine hardware and mac type
388 sc->hw.vendor_id = pci_get_vendor(dev);
389 sc->hw.device_id = pci_get_device(dev);
390 sc->hw.revision_id = pci_read_config(dev, PCIR_REVID, 1);
391 sc->hw.subsystem_vendor_id = pci_read_config(dev, PCIR_SUBVEND_0, 2);
392 sc->hw.subsystem_device_id = pci_read_config(dev, PCIR_SUBDEV_0, 2);
394 if (e1000_set_mac_type(&sc->hw))
397 /* Are we a VF device? */
398 if (sc->hw.mac.type == e1000_vfadapt ||
399 sc->hw.mac.type == e1000_vfadapt_i350)
405 * Configure total supported RX/TX ring count
407 switch (sc->hw.mac.type) {
409 ring_max = IGB_MAX_RING_82575;
412 ring_max = IGB_MAX_RING_82580;
415 ring_max = IGB_MAX_RING_I350;
418 ring_max = IGB_MAX_RING_82576;
421 ring_max = IGB_MIN_RING;
425 sc->rx_ring_cnt = device_getenv_int(dev, "rxr", igb_rxr);
426 sc->rx_ring_cnt = if_ring_count2(sc->rx_ring_cnt, ring_max);
428 sc->rx_ring_cnt = device_getenv_int(dev, "rxr_debug", sc->rx_ring_cnt);
430 sc->rx_ring_inuse = sc->rx_ring_cnt;
432 sc->tx_ring_cnt = device_getenv_int(dev, "txr", igb_txr);
433 sc->tx_ring_cnt = if_ring_count2(sc->tx_ring_cnt, /* XXX ring_max */1);
435 sc->tx_ring_cnt = device_getenv_int(dev, "txr_debug", sc->tx_ring_cnt);
437 sc->tx_ring_inuse = sc->tx_ring_cnt;
439 /* Enable bus mastering */
440 pci_enable_busmaster(dev);
445 sc->mem_rid = PCIR_BAR(0);
446 sc->mem_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &sc->mem_rid,
448 if (sc->mem_res == NULL) {
449 device_printf(dev, "Unable to allocate bus resource: memory\n");
453 sc->osdep.mem_bus_space_tag = rman_get_bustag(sc->mem_res);
454 sc->osdep.mem_bus_space_handle = rman_get_bushandle(sc->mem_res);
456 sc->hw.hw_addr = (uint8_t *)&sc->osdep.mem_bus_space_handle;
458 /* Save PCI command register for Shared Code */
459 sc->hw.bus.pci_cmd_word = pci_read_config(dev, PCIR_COMMAND, 2);
460 sc->hw.back = &sc->osdep;
462 /* Do Shared Code initialization */
463 if (e1000_setup_init_funcs(&sc->hw, TRUE)) {
464 device_printf(dev, "Setup of Shared code failed\n");
469 e1000_get_bus_info(&sc->hw);
471 sc->hw.mac.autoneg = DO_AUTO_NEG;
472 sc->hw.phy.autoneg_wait_to_complete = FALSE;
473 sc->hw.phy.autoneg_advertised = AUTONEG_ADV_DEFAULT;
476 if (sc->hw.phy.media_type == e1000_media_type_copper) {
477 sc->hw.phy.mdix = AUTO_ALL_MODES;
478 sc->hw.phy.disable_polarity_correction = FALSE;
479 sc->hw.phy.ms_type = IGB_MASTER_SLAVE;
482 /* Set the frame limits assuming standard ethernet sized frames. */
483 sc->max_frame_size = ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN;
485 /* Allocate RX/TX rings */
486 error = igb_alloc_rings(sc);
492 * NPOLLING RX CPU offset
494 if (sc->rx_ring_cnt == ncpus2) {
497 offset_def = (sc->rx_ring_cnt * device_get_unit(dev)) % ncpus2;
498 offset = device_getenv_int(dev, "npoll.rxoff", offset_def);
499 if (offset >= ncpus2 ||
500 offset % sc->rx_ring_cnt != 0) {
501 device_printf(dev, "invalid npoll.rxoff %d, use %d\n",
506 sc->rx_npoll_off = offset;
509 * NPOLLING TX CPU offset
511 if (sc->tx_ring_cnt == ncpus2) {
514 offset_def = (sc->tx_ring_cnt * device_get_unit(dev)) % ncpus2;
515 offset = device_getenv_int(dev, "npoll.txoff", offset_def);
516 if (offset >= ncpus2 ||
517 offset % sc->tx_ring_cnt != 0) {
518 device_printf(dev, "invalid npoll.txoff %d, use %d\n",
523 sc->tx_npoll_off = offset;
526 /* Allocate interrupt */
527 error = igb_alloc_intr(sc);
535 sc->serializes[i++] = &sc->main_serialize;
537 sc->tx_serialize = i;
538 for (j = 0; j < sc->tx_ring_cnt; ++j)
539 sc->serializes[i++] = &sc->tx_rings[j].tx_serialize;
541 sc->rx_serialize = i;
542 for (j = 0; j < sc->rx_ring_cnt; ++j)
543 sc->serializes[i++] = &sc->rx_rings[j].rx_serialize;
545 sc->serialize_cnt = i;
546 KKASSERT(sc->serialize_cnt <= IGB_NSERIALIZE);
548 /* Allocate the appropriate stats memory */
550 sc->stats = kmalloc(sizeof(struct e1000_vf_stats), M_DEVBUF,
552 igb_vf_init_stats(sc);
554 sc->stats = kmalloc(sizeof(struct e1000_hw_stats), M_DEVBUF,
558 /* Allocate multicast array memory. */
559 sc->mta = kmalloc(ETHER_ADDR_LEN * MAX_NUM_MULTICAST_ADDRESSES,
562 /* Some adapter-specific advanced features */
563 if (sc->hw.mac.type >= e1000_i350) {
565 igb_set_sysctl_value(adapter, "dma_coalesce",
566 "configure dma coalesce",
567 &adapter->dma_coalesce, igb_dma_coalesce);
568 igb_set_sysctl_value(adapter, "eee_disabled",
569 "enable Energy Efficient Ethernet",
570 &adapter->hw.dev_spec._82575.eee_disable,
573 sc->dma_coalesce = igb_dma_coalesce;
574 sc->hw.dev_spec._82575.eee_disable = igb_eee_disabled;
576 e1000_set_eee_i350(&sc->hw);
580 * Start from a known state, this is important in reading the nvm and
583 e1000_reset_hw(&sc->hw);
585 /* Make sure we have a good EEPROM before we read from it */
586 if (e1000_validate_nvm_checksum(&sc->hw) < 0) {
588 * Some PCI-E parts fail the first check due to
589 * the link being in sleep state, call it again,
590 * if it fails a second time its a real issue.
592 if (e1000_validate_nvm_checksum(&sc->hw) < 0) {
594 "The EEPROM Checksum Is Not Valid\n");
600 /* Copy the permanent MAC address out of the EEPROM */
601 if (e1000_read_mac_addr(&sc->hw) < 0) {
602 device_printf(dev, "EEPROM read error while reading MAC"
607 if (!igb_is_valid_ether_addr(sc->hw.mac.addr)) {
608 device_printf(dev, "Invalid MAC address\n");
613 /* Setup OS specific network interface */
616 /* Add sysctl tree, must after igb_setup_ifp() */
619 /* Now get a good starting state */
622 /* Initialize statistics */
623 igb_update_stats_counters(sc);
625 sc->hw.mac.get_link_status = 1;
626 igb_update_link_status(sc);
628 /* Indicate SOL/IDER usage */
629 if (e1000_check_reset_block(&sc->hw)) {
631 "PHY reset is blocked due to SOL/IDER session.\n");
634 /* Determine if we have to control management hardware */
635 if (e1000_enable_mng_pass_thru(&sc->hw))
636 sc->flags |= IGB_FLAG_HAS_MGMT;
641 /* APME bit in EEPROM is mapped to WUC.APME */
642 eeprom_data = E1000_READ_REG(&sc->hw, E1000_WUC) & E1000_WUC_APME;
644 sc->wol = E1000_WUFC_MAG;
645 /* XXX disable WOL */
649 /* Register for VLAN events */
650 adapter->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
651 igb_register_vlan, adapter, EVENTHANDLER_PRI_FIRST);
652 adapter->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
653 igb_unregister_vlan, adapter, EVENTHANDLER_PRI_FIRST);
657 igb_add_hw_stats(adapter);
660 error = igb_setup_intr(sc);
662 ether_ifdetach(&sc->arpcom.ac_if);
666 for (i = 0; i < sc->tx_ring_cnt; ++i) {
667 struct ifaltq_subque *ifsq =
668 ifq_get_subq(&sc->arpcom.ac_if.if_snd, i);
669 struct igb_tx_ring *txr = &sc->tx_rings[i];
671 ifsq_set_cpuid(ifsq, txr->tx_intr_cpuid);
672 ifsq_set_priv(ifsq, txr);
675 ifsq_watchdog_init(&txr->tx_watchdog, ifsq, igb_watchdog);
686 igb_detach(device_t dev)
688 struct igb_softc *sc = device_get_softc(dev);
690 if (device_is_attached(dev)) {
691 struct ifnet *ifp = &sc->arpcom.ac_if;
693 ifnet_serialize_all(ifp);
697 e1000_phy_hw_reset(&sc->hw);
699 /* Give control back to firmware */
701 igb_rel_hw_control(sc);
704 E1000_WRITE_REG(&sc->hw, E1000_WUC, E1000_WUC_PME_EN);
705 E1000_WRITE_REG(&sc->hw, E1000_WUFC, sc->wol);
709 igb_teardown_intr(sc);
711 ifnet_deserialize_all(ifp);
714 } else if (sc->mem_res != NULL) {
715 igb_rel_hw_control(sc);
717 bus_generic_detach(dev);
719 if (sc->sysctl_tree != NULL)
720 sysctl_ctx_free(&sc->sysctl_ctx);
724 if (sc->msix_mem_res != NULL) {
725 bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_mem_rid,
728 if (sc->mem_res != NULL) {
729 bus_release_resource(dev, SYS_RES_MEMORY, sc->mem_rid,
736 kfree(sc->mta, M_DEVBUF);
737 if (sc->stats != NULL)
738 kfree(sc->stats, M_DEVBUF);
744 igb_shutdown(device_t dev)
746 return igb_suspend(dev);
750 igb_suspend(device_t dev)
752 struct igb_softc *sc = device_get_softc(dev);
753 struct ifnet *ifp = &sc->arpcom.ac_if;
755 ifnet_serialize_all(ifp);
760 igb_rel_hw_control(sc);
763 E1000_WRITE_REG(&sc->hw, E1000_WUC, E1000_WUC_PME_EN);
764 E1000_WRITE_REG(&sc->hw, E1000_WUFC, sc->wol);
768 ifnet_deserialize_all(ifp);
770 return bus_generic_suspend(dev);
774 igb_resume(device_t dev)
776 struct igb_softc *sc = device_get_softc(dev);
777 struct ifnet *ifp = &sc->arpcom.ac_if;
780 ifnet_serialize_all(ifp);
785 for (i = 0; i < sc->tx_ring_inuse; ++i)
786 ifsq_devstart_sched(sc->tx_rings[i].ifsq);
788 ifnet_deserialize_all(ifp);
790 return bus_generic_resume(dev);
794 igb_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
796 struct igb_softc *sc = ifp->if_softc;
797 struct ifreq *ifr = (struct ifreq *)data;
798 int max_frame_size, mask, reinit;
801 ASSERT_IFNET_SERIALIZED_ALL(ifp);
805 max_frame_size = 9234;
806 if (ifr->ifr_mtu > max_frame_size - ETHER_HDR_LEN -
812 ifp->if_mtu = ifr->ifr_mtu;
813 sc->max_frame_size = ifp->if_mtu + ETHER_HDR_LEN +
816 if (ifp->if_flags & IFF_RUNNING)
821 if (ifp->if_flags & IFF_UP) {
822 if (ifp->if_flags & IFF_RUNNING) {
823 if ((ifp->if_flags ^ sc->if_flags) &
824 (IFF_PROMISC | IFF_ALLMULTI)) {
825 igb_disable_promisc(sc);
831 } else if (ifp->if_flags & IFF_RUNNING) {
834 sc->if_flags = ifp->if_flags;
839 if (ifp->if_flags & IFF_RUNNING) {
840 igb_disable_intr(sc);
843 if (!(ifp->if_flags & IFF_NPOLLING))
851 * As the speed/duplex settings are being
852 * changed, we need toreset the PHY.
854 sc->hw.phy.reset_disable = FALSE;
856 /* Check SOL/IDER usage */
857 if (e1000_check_reset_block(&sc->hw)) {
858 if_printf(ifp, "Media change is "
859 "blocked due to SOL/IDER session.\n");
865 error = ifmedia_ioctl(ifp, ifr, &sc->media, command);
870 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
871 if (mask & IFCAP_RXCSUM) {
872 ifp->if_capenable ^= IFCAP_RXCSUM;
875 if (mask & IFCAP_VLAN_HWTAGGING) {
876 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
879 if (mask & IFCAP_TXCSUM) {
880 ifp->if_capenable ^= IFCAP_TXCSUM;
881 if (ifp->if_capenable & IFCAP_TXCSUM)
882 ifp->if_hwassist |= IGB_CSUM_FEATURES;
884 ifp->if_hwassist &= ~IGB_CSUM_FEATURES;
886 if (mask & IFCAP_TSO) {
887 ifp->if_capenable ^= IFCAP_TSO;
888 if (ifp->if_capenable & IFCAP_TSO)
889 ifp->if_hwassist |= CSUM_TSO;
891 ifp->if_hwassist &= ~CSUM_TSO;
893 if (mask & IFCAP_RSS)
894 ifp->if_capenable ^= IFCAP_RSS;
895 if (reinit && (ifp->if_flags & IFF_RUNNING))
900 error = ether_ioctl(ifp, command, data);
909 struct igb_softc *sc = xsc;
910 struct ifnet *ifp = &sc->arpcom.ac_if;
914 ASSERT_IFNET_SERIALIZED_ALL(ifp);
918 /* Get the latest mac address, User can use a LAA */
919 bcopy(IF_LLADDR(ifp), sc->hw.mac.addr, ETHER_ADDR_LEN);
921 /* Put the address into the Receive Address Array */
922 e1000_rar_set(&sc->hw, sc->hw.mac.addr, 0);
925 igb_update_link_status(sc);
927 E1000_WRITE_REG(&sc->hw, E1000_VET, ETHERTYPE_VLAN);
929 /* Configure for OS presence */
934 if (ifp->if_flags & IFF_NPOLLING)
938 /* Configured used RX/TX rings */
939 igb_set_ring_inuse(sc, polling);
941 /* Initialize interrupt */
944 /* Prepare transmit descriptors and buffers */
945 for (i = 0; i < sc->tx_ring_inuse; ++i)
946 igb_init_tx_ring(&sc->tx_rings[i]);
947 igb_init_tx_unit(sc);
949 /* Setup Multicast table */
954 * Figure out the desired mbuf pool
955 * for doing jumbo/packetsplit
957 if (adapter->max_frame_size <= 2048)
958 adapter->rx_mbuf_sz = MCLBYTES;
959 else if (adapter->max_frame_size <= 4096)
960 adapter->rx_mbuf_sz = MJUMPAGESIZE;
962 adapter->rx_mbuf_sz = MJUM9BYTES;
965 /* Prepare receive descriptors and buffers */
966 for (i = 0; i < sc->rx_ring_inuse; ++i) {
969 error = igb_init_rx_ring(&sc->rx_rings[i]);
971 if_printf(ifp, "Could not setup receive structures\n");
976 igb_init_rx_unit(sc);
978 /* Enable VLAN support */
979 if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING)
982 /* Don't lose promiscuous settings */
985 ifp->if_flags |= IFF_RUNNING;
986 for (i = 0; i < sc->tx_ring_inuse; ++i) {
987 ifsq_clr_oactive(sc->tx_rings[i].ifsq);
988 ifsq_watchdog_start(&sc->tx_rings[i].tx_watchdog);
991 if (polling || sc->intr_type == PCI_INTR_TYPE_MSIX)
992 sc->timer_cpuid = 0; /* XXX fixed */
994 sc->timer_cpuid = rman_get_cpuid(sc->intr_res);
995 callout_reset_bycpu(&sc->timer, hz, igb_timer, sc, sc->timer_cpuid);
996 e1000_clear_hw_cntrs_base_generic(&sc->hw);
998 /* This clears any pending interrupts */
999 E1000_READ_REG(&sc->hw, E1000_ICR);
1002 * Only enable interrupts if we are not polling, make sure
1003 * they are off otherwise.
1006 igb_disable_intr(sc);
1008 igb_enable_intr(sc);
1009 E1000_WRITE_REG(&sc->hw, E1000_ICS, E1000_ICS_LSC);
1012 /* Set Energy Efficient Ethernet */
1013 e1000_set_eee_i350(&sc->hw);
1015 /* Don't reset the phy next time init gets called */
1016 sc->hw.phy.reset_disable = TRUE;
1020 igb_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
1022 struct igb_softc *sc = ifp->if_softc;
1023 u_char fiber_type = IFM_1000_SX;
1025 ASSERT_IFNET_SERIALIZED_ALL(ifp);
1027 igb_update_link_status(sc);
1029 ifmr->ifm_status = IFM_AVALID;
1030 ifmr->ifm_active = IFM_ETHER;
1032 if (!sc->link_active)
1035 ifmr->ifm_status |= IFM_ACTIVE;
1037 if (sc->hw.phy.media_type == e1000_media_type_fiber ||
1038 sc->hw.phy.media_type == e1000_media_type_internal_serdes) {
1039 ifmr->ifm_active |= fiber_type | IFM_FDX;
1041 switch (sc->link_speed) {
1043 ifmr->ifm_active |= IFM_10_T;
1047 ifmr->ifm_active |= IFM_100_TX;
1051 ifmr->ifm_active |= IFM_1000_T;
1054 if (sc->link_duplex == FULL_DUPLEX)
1055 ifmr->ifm_active |= IFM_FDX;
1057 ifmr->ifm_active |= IFM_HDX;
1062 igb_media_change(struct ifnet *ifp)
1064 struct igb_softc *sc = ifp->if_softc;
1065 struct ifmedia *ifm = &sc->media;
1067 ASSERT_IFNET_SERIALIZED_ALL(ifp);
1069 if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
1072 switch (IFM_SUBTYPE(ifm->ifm_media)) {
1074 sc->hw.mac.autoneg = DO_AUTO_NEG;
1075 sc->hw.phy.autoneg_advertised = AUTONEG_ADV_DEFAULT;
1081 sc->hw.mac.autoneg = DO_AUTO_NEG;
1082 sc->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
1086 sc->hw.mac.autoneg = FALSE;
1087 sc->hw.phy.autoneg_advertised = 0;
1088 if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
1089 sc->hw.mac.forced_speed_duplex = ADVERTISE_100_FULL;
1091 sc->hw.mac.forced_speed_duplex = ADVERTISE_100_HALF;
1095 sc->hw.mac.autoneg = FALSE;
1096 sc->hw.phy.autoneg_advertised = 0;
1097 if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
1098 sc->hw.mac.forced_speed_duplex = ADVERTISE_10_FULL;
1100 sc->hw.mac.forced_speed_duplex = ADVERTISE_10_HALF;
1104 if_printf(ifp, "Unsupported media type\n");
1114 igb_set_promisc(struct igb_softc *sc)
1116 struct ifnet *ifp = &sc->arpcom.ac_if;
1117 struct e1000_hw *hw = &sc->hw;
1121 e1000_promisc_set_vf(hw, e1000_promisc_enabled);
1125 reg = E1000_READ_REG(hw, E1000_RCTL);
1126 if (ifp->if_flags & IFF_PROMISC) {
1127 reg |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
1128 E1000_WRITE_REG(hw, E1000_RCTL, reg);
1129 } else if (ifp->if_flags & IFF_ALLMULTI) {
1130 reg |= E1000_RCTL_MPE;
1131 reg &= ~E1000_RCTL_UPE;
1132 E1000_WRITE_REG(hw, E1000_RCTL, reg);
1137 igb_disable_promisc(struct igb_softc *sc)
1139 struct e1000_hw *hw = &sc->hw;
1143 e1000_promisc_set_vf(hw, e1000_promisc_disabled);
1146 reg = E1000_READ_REG(hw, E1000_RCTL);
1147 reg &= ~E1000_RCTL_UPE;
1148 reg &= ~E1000_RCTL_MPE;
1149 E1000_WRITE_REG(hw, E1000_RCTL, reg);
1153 igb_set_multi(struct igb_softc *sc)
1155 struct ifnet *ifp = &sc->arpcom.ac_if;
1156 struct ifmultiaddr *ifma;
1157 uint32_t reg_rctl = 0;
1162 bzero(mta, ETH_ADDR_LEN * MAX_NUM_MULTICAST_ADDRESSES);
1164 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1165 if (ifma->ifma_addr->sa_family != AF_LINK)
1168 if (mcnt == MAX_NUM_MULTICAST_ADDRESSES)
1171 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1172 &mta[mcnt * ETH_ADDR_LEN], ETH_ADDR_LEN);
1176 if (mcnt >= MAX_NUM_MULTICAST_ADDRESSES) {
1177 reg_rctl = E1000_READ_REG(&sc->hw, E1000_RCTL);
1178 reg_rctl |= E1000_RCTL_MPE;
1179 E1000_WRITE_REG(&sc->hw, E1000_RCTL, reg_rctl);
1181 e1000_update_mc_addr_list(&sc->hw, mta, mcnt);
1186 igb_timer(void *xsc)
1188 struct igb_softc *sc = xsc;
1190 lwkt_serialize_enter(&sc->main_serialize);
1192 igb_update_link_status(sc);
1193 igb_update_stats_counters(sc);
1195 callout_reset_bycpu(&sc->timer, hz, igb_timer, sc, sc->timer_cpuid);
1197 lwkt_serialize_exit(&sc->main_serialize);
1201 igb_update_link_status(struct igb_softc *sc)
1203 struct ifnet *ifp = &sc->arpcom.ac_if;
1204 struct e1000_hw *hw = &sc->hw;
1205 uint32_t link_check, thstat, ctrl;
1207 link_check = thstat = ctrl = 0;
1209 /* Get the cached link value or read for real */
1210 switch (hw->phy.media_type) {
1211 case e1000_media_type_copper:
1212 if (hw->mac.get_link_status) {
1213 /* Do the work to read phy */
1214 e1000_check_for_link(hw);
1215 link_check = !hw->mac.get_link_status;
1221 case e1000_media_type_fiber:
1222 e1000_check_for_link(hw);
1223 link_check = E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU;
1226 case e1000_media_type_internal_serdes:
1227 e1000_check_for_link(hw);
1228 link_check = hw->mac.serdes_has_link;
1231 /* VF device is type_unknown */
1232 case e1000_media_type_unknown:
1233 e1000_check_for_link(hw);
1234 link_check = !hw->mac.get_link_status;
1240 /* Check for thermal downshift or shutdown */
1241 if (hw->mac.type == e1000_i350) {
1242 thstat = E1000_READ_REG(hw, E1000_THSTAT);
1243 ctrl = E1000_READ_REG(hw, E1000_CTRL_EXT);
1246 /* Now we check if a transition has happened */
1247 if (link_check && sc->link_active == 0) {
1248 e1000_get_speed_and_duplex(hw,
1249 &sc->link_speed, &sc->link_duplex);
1251 if_printf(ifp, "Link is up %d Mbps %s\n",
1253 sc->link_duplex == FULL_DUPLEX ?
1254 "Full Duplex" : "Half Duplex");
1256 sc->link_active = 1;
1258 ifp->if_baudrate = sc->link_speed * 1000000;
1259 if ((ctrl & E1000_CTRL_EXT_LINK_MODE_GMII) &&
1260 (thstat & E1000_THSTAT_LINK_THROTTLE))
1261 if_printf(ifp, "Link: thermal downshift\n");
1262 /* This can sleep */
1263 ifp->if_link_state = LINK_STATE_UP;
1264 if_link_state_change(ifp);
1265 } else if (!link_check && sc->link_active == 1) {
1266 ifp->if_baudrate = sc->link_speed = 0;
1267 sc->link_duplex = 0;
1269 if_printf(ifp, "Link is Down\n");
1270 if ((ctrl & E1000_CTRL_EXT_LINK_MODE_GMII) &&
1271 (thstat & E1000_THSTAT_PWR_DOWN))
1272 if_printf(ifp, "Link: thermal shutdown\n");
1273 sc->link_active = 0;
1274 /* This can sleep */
1275 ifp->if_link_state = LINK_STATE_DOWN;
1276 if_link_state_change(ifp);
1281 igb_stop(struct igb_softc *sc)
1283 struct ifnet *ifp = &sc->arpcom.ac_if;
1286 ASSERT_IFNET_SERIALIZED_ALL(ifp);
1288 igb_disable_intr(sc);
1290 callout_stop(&sc->timer);
1292 ifp->if_flags &= ~IFF_RUNNING;
1293 for (i = 0; i < sc->tx_ring_cnt; ++i) {
1294 ifsq_clr_oactive(sc->tx_rings[i].ifsq);
1295 ifsq_watchdog_stop(&sc->tx_rings[i].tx_watchdog);
1298 e1000_reset_hw(&sc->hw);
1299 E1000_WRITE_REG(&sc->hw, E1000_WUC, 0);
1301 e1000_led_off(&sc->hw);
1302 e1000_cleanup_led(&sc->hw);
1304 for (i = 0; i < sc->tx_ring_cnt; ++i)
1305 igb_free_tx_ring(&sc->tx_rings[i]);
1306 for (i = 0; i < sc->rx_ring_cnt; ++i)
1307 igb_free_rx_ring(&sc->rx_rings[i]);
1311 igb_reset(struct igb_softc *sc)
1313 struct ifnet *ifp = &sc->arpcom.ac_if;
1314 struct e1000_hw *hw = &sc->hw;
1315 struct e1000_fc_info *fc = &hw->fc;
1319 /* Let the firmware know the OS is in control */
1320 igb_get_hw_control(sc);
1323 * Packet Buffer Allocation (PBA)
1324 * Writing PBA sets the receive portion of the buffer
1325 * the remainder is used for the transmit buffer.
1327 switch (hw->mac.type) {
1329 pba = E1000_PBA_32K;
1334 pba = E1000_READ_REG(hw, E1000_RXPBS);
1335 pba &= E1000_RXPBS_SIZE_MASK_82576;
1340 case e1000_vfadapt_i350:
1341 pba = E1000_READ_REG(hw, E1000_RXPBS);
1342 pba = e1000_rxpbs_adjust_82580(pba);
1344 /* XXX pba = E1000_PBA_35K; */
1350 /* Special needs in case of Jumbo frames */
1351 if (hw->mac.type == e1000_82575 && ifp->if_mtu > ETHERMTU) {
1352 uint32_t tx_space, min_tx, min_rx;
1354 pba = E1000_READ_REG(hw, E1000_PBA);
1355 tx_space = pba >> 16;
1358 min_tx = (sc->max_frame_size +
1359 sizeof(struct e1000_tx_desc) - ETHER_CRC_LEN) * 2;
1360 min_tx = roundup2(min_tx, 1024);
1362 min_rx = sc->max_frame_size;
1363 min_rx = roundup2(min_rx, 1024);
1365 if (tx_space < min_tx && (min_tx - tx_space) < pba) {
1366 pba = pba - (min_tx - tx_space);
1368 * if short on rx space, rx wins
1369 * and must trump tx adjustment
1374 E1000_WRITE_REG(hw, E1000_PBA, pba);
1378 * These parameters control the automatic generation (Tx) and
1379 * response (Rx) to Ethernet PAUSE frames.
1380 * - High water mark should allow for at least two frames to be
1381 * received after sending an XOFF.
1382 * - Low water mark works best when it is very near the high water mark.
1383 * This allows the receiver to restart by sending XON when it has
1386 hwm = min(((pba << 10) * 9 / 10),
1387 ((pba << 10) - 2 * sc->max_frame_size));
1389 if (hw->mac.type < e1000_82576) {
1390 fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
1391 fc->low_water = fc->high_water - 8;
1393 fc->high_water = hwm & 0xFFF0; /* 16-byte granularity */
1394 fc->low_water = fc->high_water - 16;
1396 fc->pause_time = IGB_FC_PAUSE_TIME;
1397 fc->send_xon = TRUE;
1399 /* Issue a global reset */
1401 E1000_WRITE_REG(hw, E1000_WUC, 0);
1403 if (e1000_init_hw(hw) < 0)
1404 if_printf(ifp, "Hardware Initialization Failed\n");
1406 /* Setup DMA Coalescing */
1407 if (hw->mac.type == e1000_i350 && sc->dma_coalesce) {
1410 hwm = (pba - 4) << 10;
1411 reg = ((pba - 6) << E1000_DMACR_DMACTHR_SHIFT)
1412 & E1000_DMACR_DMACTHR_MASK;
1414 /* transition to L0x or L1 if available..*/
1415 reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
1417 /* timer = +-1000 usec in 32usec intervals */
1419 E1000_WRITE_REG(hw, E1000_DMACR, reg);
1421 /* No lower threshold */
1422 E1000_WRITE_REG(hw, E1000_DMCRTRH, 0);
1424 /* set hwm to PBA - 2 * max frame size */
1425 E1000_WRITE_REG(hw, E1000_FCRTC, hwm);
1427 /* Set the interval before transition */
1428 reg = E1000_READ_REG(hw, E1000_DMCTLX);
1429 reg |= 0x800000FF; /* 255 usec */
1430 E1000_WRITE_REG(hw, E1000_DMCTLX, reg);
1432 /* free space in tx packet buffer to wake from DMA coal */
1433 E1000_WRITE_REG(hw, E1000_DMCTXTH,
1434 (20480 - (2 * sc->max_frame_size)) >> 6);
1436 /* make low power state decision controlled by DMA coal */
1437 reg = E1000_READ_REG(hw, E1000_PCIEMISC);
1438 E1000_WRITE_REG(hw, E1000_PCIEMISC,
1439 reg | E1000_PCIEMISC_LX_DECISION);
1440 if_printf(ifp, "DMA Coalescing enabled\n");
1443 E1000_WRITE_REG(&sc->hw, E1000_VET, ETHERTYPE_VLAN);
1444 e1000_get_phy_info(hw);
1445 e1000_check_for_link(hw);
1449 igb_setup_ifp(struct igb_softc *sc)
1451 struct ifnet *ifp = &sc->arpcom.ac_if;
1454 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1455 ifp->if_init = igb_init;
1456 ifp->if_ioctl = igb_ioctl;
1457 ifp->if_start = igb_start;
1458 ifp->if_serialize = igb_serialize;
1459 ifp->if_deserialize = igb_deserialize;
1460 ifp->if_tryserialize = igb_tryserialize;
1462 ifp->if_serialize_assert = igb_serialize_assert;
1464 #ifdef IFPOLL_ENABLE
1465 ifp->if_npoll = igb_npoll;
1468 ifq_set_maxlen(&ifp->if_snd, sc->tx_rings[0].num_tx_desc - 1);
1469 ifq_set_ready(&ifp->if_snd);
1471 ether_ifattach(ifp, sc->hw.mac.addr, NULL);
1473 ifp->if_capabilities =
1474 IFCAP_HWCSUM | IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_TSO;
1475 if (IGB_ENABLE_HWRSS(sc))
1476 ifp->if_capabilities |= IFCAP_RSS;
1477 ifp->if_capenable = ifp->if_capabilities;
1478 ifp->if_hwassist = IGB_CSUM_FEATURES | CSUM_TSO;
1481 * Tell the upper layer(s) we support long frames
1483 ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1486 * Specify the media types supported by this adapter and register
1487 * callbacks to update media and link information
1489 ifmedia_init(&sc->media, IFM_IMASK, igb_media_change, igb_media_status);
1490 if (sc->hw.phy.media_type == e1000_media_type_fiber ||
1491 sc->hw.phy.media_type == e1000_media_type_internal_serdes) {
1492 ifmedia_add(&sc->media, IFM_ETHER | IFM_1000_SX | IFM_FDX,
1494 ifmedia_add(&sc->media, IFM_ETHER | IFM_1000_SX, 0, NULL);
1496 ifmedia_add(&sc->media, IFM_ETHER | IFM_10_T, 0, NULL);
1497 ifmedia_add(&sc->media, IFM_ETHER | IFM_10_T | IFM_FDX,
1499 ifmedia_add(&sc->media, IFM_ETHER | IFM_100_TX, 0, NULL);
1500 ifmedia_add(&sc->media, IFM_ETHER | IFM_100_TX | IFM_FDX,
1502 if (sc->hw.phy.type != e1000_phy_ife) {
1503 ifmedia_add(&sc->media,
1504 IFM_ETHER | IFM_1000_T | IFM_FDX, 0, NULL);
1505 ifmedia_add(&sc->media,
1506 IFM_ETHER | IFM_1000_T, 0, NULL);
1509 ifmedia_add(&sc->media, IFM_ETHER | IFM_AUTO, 0, NULL);
1510 ifmedia_set(&sc->media, IFM_ETHER | IFM_AUTO);
1514 igb_add_sysctl(struct igb_softc *sc)
1519 sysctl_ctx_init(&sc->sysctl_ctx);
1520 sc->sysctl_tree = SYSCTL_ADD_NODE(&sc->sysctl_ctx,
1521 SYSCTL_STATIC_CHILDREN(_hw), OID_AUTO,
1522 device_get_nameunit(sc->dev), CTLFLAG_RD, 0, "");
1523 if (sc->sysctl_tree == NULL) {
1524 device_printf(sc->dev, "can't add sysctl node\n");
1528 SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1529 OID_AUTO, "rxr", CTLFLAG_RD, &sc->rx_ring_cnt, 0, "# of RX rings");
1530 SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1531 OID_AUTO, "rxr_inuse", CTLFLAG_RD, &sc->rx_ring_inuse, 0,
1532 "# of RX rings used");
1533 SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1534 OID_AUTO, "txr", CTLFLAG_RD, &sc->tx_ring_cnt, 0, "# of TX rings");
1535 SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1536 OID_AUTO, "txr_inuse", CTLFLAG_RD, &sc->tx_ring_inuse, 0,
1537 "# of TX rings used");
1538 SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1539 OID_AUTO, "rxd", CTLFLAG_RD, &sc->rx_rings[0].num_rx_desc, 0,
1541 SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1542 OID_AUTO, "txd", CTLFLAG_RD, &sc->tx_rings[0].num_tx_desc, 0,
1545 if (sc->intr_type != PCI_INTR_TYPE_MSIX) {
1546 SYSCTL_ADD_PROC(&sc->sysctl_ctx,
1547 SYSCTL_CHILDREN(sc->sysctl_tree),
1548 OID_AUTO, "intr_rate", CTLTYPE_INT | CTLFLAG_RW,
1549 sc, 0, igb_sysctl_intr_rate, "I", "interrupt rate");
1551 for (i = 0; i < sc->msix_cnt; ++i) {
1552 struct igb_msix_data *msix = &sc->msix_data[i];
1554 ksnprintf(node, sizeof(node), "msix%d_rate", i);
1555 SYSCTL_ADD_PROC(&sc->sysctl_ctx,
1556 SYSCTL_CHILDREN(sc->sysctl_tree),
1557 OID_AUTO, node, CTLTYPE_INT | CTLFLAG_RW,
1558 msix, 0, igb_sysctl_msix_rate, "I",
1559 msix->msix_rate_desc);
1563 SYSCTL_ADD_PROC(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1564 OID_AUTO, "tx_intr_nsegs", CTLTYPE_INT | CTLFLAG_RW,
1565 sc, 0, igb_sysctl_tx_intr_nsegs, "I",
1566 "# of segments per TX interrupt");
1568 SYSCTL_ADD_PROC(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1569 OID_AUTO, "tx_wreg_nsegs", CTLTYPE_INT | CTLFLAG_RW,
1570 sc, 0, igb_sysctl_tx_wreg_nsegs, "I",
1571 "# of segments sent before write to hardware register");
1573 SYSCTL_ADD_PROC(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1574 OID_AUTO, "rx_wreg_nsegs", CTLTYPE_INT | CTLFLAG_RW,
1575 sc, 0, igb_sysctl_rx_wreg_nsegs, "I",
1576 "# of segments received before write to hardware register");
1578 #ifdef IFPOLL_ENABLE
1579 SYSCTL_ADD_PROC(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1580 OID_AUTO, "npoll_rxoff", CTLTYPE_INT|CTLFLAG_RW,
1581 sc, 0, igb_sysctl_npoll_rxoff, "I", "NPOLLING RX cpu offset");
1582 SYSCTL_ADD_PROC(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1583 OID_AUTO, "npoll_txoff", CTLTYPE_INT|CTLFLAG_RW,
1584 sc, 0, igb_sysctl_npoll_txoff, "I", "NPOLLING TX cpu offset");
1587 #ifdef IGB_RSS_DEBUG
1588 SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1589 OID_AUTO, "rss_debug", CTLFLAG_RW, &sc->rss_debug, 0,
1591 for (i = 0; i < sc->rx_ring_cnt; ++i) {
1592 ksnprintf(node, sizeof(node), "rx%d_pkt", i);
1593 SYSCTL_ADD_ULONG(&sc->sysctl_ctx,
1594 SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO, node,
1595 CTLFLAG_RW, &sc->rx_rings[i].rx_packets, "RXed packets");
1601 igb_alloc_rings(struct igb_softc *sc)
1606 * Create top level busdma tag
1608 error = bus_dma_tag_create(NULL, 1, 0,
1609 BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1610 BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
1613 device_printf(sc->dev, "could not create top level DMA tag\n");
1618 * Allocate TX descriptor rings and buffers
1620 sc->tx_rings = kmalloc_cachealign(
1621 sizeof(struct igb_tx_ring) * sc->tx_ring_cnt,
1622 M_DEVBUF, M_WAITOK | M_ZERO);
1623 for (i = 0; i < sc->tx_ring_cnt; ++i) {
1624 struct igb_tx_ring *txr = &sc->tx_rings[i];
1626 /* Set up some basics */
1629 lwkt_serialize_init(&txr->tx_serialize);
1631 error = igb_create_tx_ring(txr);
1637 * Allocate RX descriptor rings and buffers
1639 sc->rx_rings = kmalloc_cachealign(
1640 sizeof(struct igb_rx_ring) * sc->rx_ring_cnt,
1641 M_DEVBUF, M_WAITOK | M_ZERO);
1642 for (i = 0; i < sc->rx_ring_cnt; ++i) {
1643 struct igb_rx_ring *rxr = &sc->rx_rings[i];
1645 /* Set up some basics */
1648 lwkt_serialize_init(&rxr->rx_serialize);
1650 error = igb_create_rx_ring(rxr);
1659 igb_free_rings(struct igb_softc *sc)
1663 if (sc->tx_rings != NULL) {
1664 for (i = 0; i < sc->tx_ring_cnt; ++i) {
1665 struct igb_tx_ring *txr = &sc->tx_rings[i];
1667 igb_destroy_tx_ring(txr, txr->num_tx_desc);
1669 kfree(sc->tx_rings, M_DEVBUF);
1672 if (sc->rx_rings != NULL) {
1673 for (i = 0; i < sc->rx_ring_cnt; ++i) {
1674 struct igb_rx_ring *rxr = &sc->rx_rings[i];
1676 igb_destroy_rx_ring(rxr, rxr->num_rx_desc);
1678 kfree(sc->rx_rings, M_DEVBUF);
1683 igb_create_tx_ring(struct igb_tx_ring *txr)
1685 int tsize, error, i, ntxd;
1688 * Validate number of transmit descriptors. It must not exceed
1689 * hardware maximum, and must be multiple of IGB_DBA_ALIGN.
1691 ntxd = device_getenv_int(txr->sc->dev, "txd", igb_txd);
1692 if ((ntxd * sizeof(struct e1000_tx_desc)) % IGB_DBA_ALIGN != 0 ||
1693 ntxd > IGB_MAX_TXD || ntxd < IGB_MIN_TXD) {
1694 device_printf(txr->sc->dev,
1695 "Using %d TX descriptors instead of %d!\n",
1696 IGB_DEFAULT_TXD, ntxd);
1697 txr->num_tx_desc = IGB_DEFAULT_TXD;
1699 txr->num_tx_desc = ntxd;
1703 * Allocate TX descriptor ring
1705 tsize = roundup2(txr->num_tx_desc * sizeof(union e1000_adv_tx_desc),
1707 txr->txdma.dma_vaddr = bus_dmamem_coherent_any(txr->sc->parent_tag,
1708 IGB_DBA_ALIGN, tsize, BUS_DMA_WAITOK,
1709 &txr->txdma.dma_tag, &txr->txdma.dma_map, &txr->txdma.dma_paddr);
1710 if (txr->txdma.dma_vaddr == NULL) {
1711 device_printf(txr->sc->dev,
1712 "Unable to allocate TX Descriptor memory\n");
1715 txr->tx_base = txr->txdma.dma_vaddr;
1716 bzero(txr->tx_base, tsize);
1718 tsize = __VM_CACHELINE_ALIGN(
1719 sizeof(struct igb_tx_buf) * txr->num_tx_desc);
1720 txr->tx_buf = kmalloc_cachealign(tsize, M_DEVBUF, M_WAITOK | M_ZERO);
1723 * Allocate TX head write-back buffer
1725 txr->tx_hdr = bus_dmamem_coherent_any(txr->sc->parent_tag,
1726 __VM_CACHELINE_SIZE, __VM_CACHELINE_SIZE, BUS_DMA_WAITOK,
1727 &txr->tx_hdr_dtag, &txr->tx_hdr_dmap, &txr->tx_hdr_paddr);
1728 if (txr->tx_hdr == NULL) {
1729 device_printf(txr->sc->dev,
1730 "Unable to allocate TX head write-back buffer\n");
1735 * Create DMA tag for TX buffers
1737 error = bus_dma_tag_create(txr->sc->parent_tag,
1738 1, 0, /* alignment, bounds */
1739 BUS_SPACE_MAXADDR, /* lowaddr */
1740 BUS_SPACE_MAXADDR, /* highaddr */
1741 NULL, NULL, /* filter, filterarg */
1742 IGB_TSO_SIZE, /* maxsize */
1743 IGB_MAX_SCATTER, /* nsegments */
1744 PAGE_SIZE, /* maxsegsize */
1745 BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW |
1746 BUS_DMA_ONEBPAGE, /* flags */
1749 device_printf(txr->sc->dev, "Unable to allocate TX DMA tag\n");
1750 kfree(txr->tx_buf, M_DEVBUF);
1756 * Create DMA maps for TX buffers
1758 for (i = 0; i < txr->num_tx_desc; ++i) {
1759 struct igb_tx_buf *txbuf = &txr->tx_buf[i];
1761 error = bus_dmamap_create(txr->tx_tag,
1762 BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE, &txbuf->map);
1764 device_printf(txr->sc->dev,
1765 "Unable to create TX DMA map\n");
1766 igb_destroy_tx_ring(txr, i);
1771 if (txr->sc->hw.mac.type == e1000_82575)
1772 txr->tx_flags |= IGB_TXFLAG_TSO_IPLEN0;
1775 * Initialize various watermark
1777 txr->spare_desc = IGB_TX_SPARE;
1778 txr->intr_nsegs = txr->num_tx_desc / 16;
1779 txr->wreg_nsegs = IGB_DEF_TXWREG_NSEGS;
1780 txr->oact_hi_desc = txr->num_tx_desc / 2;
1781 txr->oact_lo_desc = txr->num_tx_desc / 8;
1782 if (txr->oact_lo_desc > IGB_TX_OACTIVE_MAX)
1783 txr->oact_lo_desc = IGB_TX_OACTIVE_MAX;
1784 if (txr->oact_lo_desc < txr->spare_desc + IGB_TX_RESERVED)
1785 txr->oact_lo_desc = txr->spare_desc + IGB_TX_RESERVED;
1791 igb_free_tx_ring(struct igb_tx_ring *txr)
1795 for (i = 0; i < txr->num_tx_desc; ++i) {
1796 struct igb_tx_buf *txbuf = &txr->tx_buf[i];
1798 if (txbuf->m_head != NULL) {
1799 bus_dmamap_unload(txr->tx_tag, txbuf->map);
1800 m_freem(txbuf->m_head);
1801 txbuf->m_head = NULL;
1807 igb_destroy_tx_ring(struct igb_tx_ring *txr, int ndesc)
1811 if (txr->txdma.dma_vaddr != NULL) {
1812 bus_dmamap_unload(txr->txdma.dma_tag, txr->txdma.dma_map);
1813 bus_dmamem_free(txr->txdma.dma_tag, txr->txdma.dma_vaddr,
1814 txr->txdma.dma_map);
1815 bus_dma_tag_destroy(txr->txdma.dma_tag);
1816 txr->txdma.dma_vaddr = NULL;
1819 if (txr->tx_hdr != NULL) {
1820 bus_dmamap_unload(txr->tx_hdr_dtag, txr->tx_hdr_dmap);
1821 bus_dmamem_free(txr->tx_hdr_dtag, txr->tx_hdr,
1823 bus_dma_tag_destroy(txr->tx_hdr_dtag);
1827 if (txr->tx_buf == NULL)
1830 for (i = 0; i < ndesc; ++i) {
1831 struct igb_tx_buf *txbuf = &txr->tx_buf[i];
1833 KKASSERT(txbuf->m_head == NULL);
1834 bus_dmamap_destroy(txr->tx_tag, txbuf->map);
1836 bus_dma_tag_destroy(txr->tx_tag);
1838 kfree(txr->tx_buf, M_DEVBUF);
1843 igb_init_tx_ring(struct igb_tx_ring *txr)
1845 /* Clear the old descriptor contents */
1847 sizeof(union e1000_adv_tx_desc) * txr->num_tx_desc);
1849 /* Clear TX head write-back buffer */
1853 txr->next_avail_desc = 0;
1854 txr->next_to_clean = 0;
1857 /* Set number of descriptors available */
1858 txr->tx_avail = txr->num_tx_desc;
1862 igb_init_tx_unit(struct igb_softc *sc)
1864 struct e1000_hw *hw = &sc->hw;
1868 /* Setup the Tx Descriptor Rings */
1869 for (i = 0; i < sc->tx_ring_inuse; ++i) {
1870 struct igb_tx_ring *txr = &sc->tx_rings[i];
1871 uint64_t bus_addr = txr->txdma.dma_paddr;
1872 uint64_t hdr_paddr = txr->tx_hdr_paddr;
1873 uint32_t txdctl = 0;
1874 uint32_t dca_txctrl;
1876 E1000_WRITE_REG(hw, E1000_TDLEN(i),
1877 txr->num_tx_desc * sizeof(struct e1000_tx_desc));
1878 E1000_WRITE_REG(hw, E1000_TDBAH(i),
1879 (uint32_t)(bus_addr >> 32));
1880 E1000_WRITE_REG(hw, E1000_TDBAL(i),
1881 (uint32_t)bus_addr);
1883 /* Setup the HW Tx Head and Tail descriptor pointers */
1884 E1000_WRITE_REG(hw, E1000_TDT(i), 0);
1885 E1000_WRITE_REG(hw, E1000_TDH(i), 0);
1887 dca_txctrl = E1000_READ_REG(hw, E1000_DCA_TXCTRL(i));
1888 dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1889 E1000_WRITE_REG(hw, E1000_DCA_TXCTRL(i), dca_txctrl);
1892 * Don't set WB_on_EITR:
1893 * - 82575 does not have it
1894 * - It almost has no effect on 82576, see:
1895 * 82576 specification update errata #26
1896 * - It causes unnecessary bus traffic
1898 E1000_WRITE_REG(hw, E1000_TDWBAH(i),
1899 (uint32_t)(hdr_paddr >> 32));
1900 E1000_WRITE_REG(hw, E1000_TDWBAL(i),
1901 ((uint32_t)hdr_paddr) | E1000_TX_HEAD_WB_ENABLE);
1904 * WTHRESH is ignored by the hardware, since header
1905 * write back mode is used.
1907 txdctl |= IGB_TX_PTHRESH;
1908 txdctl |= IGB_TX_HTHRESH << 8;
1909 txdctl |= IGB_TX_WTHRESH << 16;
1910 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1911 E1000_WRITE_REG(hw, E1000_TXDCTL(i), txdctl);
1917 e1000_config_collision_dist(hw);
1919 /* Program the Transmit Control Register */
1920 tctl = E1000_READ_REG(hw, E1000_TCTL);
1921 tctl &= ~E1000_TCTL_CT;
1922 tctl |= (E1000_TCTL_PSP | E1000_TCTL_RTLC | E1000_TCTL_EN |
1923 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT));
1925 /* This write will effectively turn on the transmit unit. */
1926 E1000_WRITE_REG(hw, E1000_TCTL, tctl);
1930 igb_txcsum_ctx(struct igb_tx_ring *txr, struct mbuf *mp)
1932 struct e1000_adv_tx_context_desc *TXD;
1933 uint32_t vlan_macip_lens, type_tucmd_mlhl, mss_l4len_idx;
1934 int ehdrlen, ctxd, ip_hlen = 0;
1935 boolean_t offload = TRUE;
1937 if ((mp->m_pkthdr.csum_flags & IGB_CSUM_FEATURES) == 0)
1940 vlan_macip_lens = type_tucmd_mlhl = mss_l4len_idx = 0;
1942 ctxd = txr->next_avail_desc;
1943 TXD = (struct e1000_adv_tx_context_desc *)&txr->tx_base[ctxd];
1946 * In advanced descriptors the vlan tag must
1947 * be placed into the context descriptor, thus
1948 * we need to be here just for that setup.
1950 if (mp->m_flags & M_VLANTAG) {
1953 vlantag = htole16(mp->m_pkthdr.ether_vlantag);
1954 vlan_macip_lens |= (vlantag << E1000_ADVTXD_VLAN_SHIFT);
1955 } else if (!offload) {
1959 ehdrlen = mp->m_pkthdr.csum_lhlen;
1960 KASSERT(ehdrlen > 0, ("invalid ether hlen"));
1962 /* Set the ether header length */
1963 vlan_macip_lens |= ehdrlen << E1000_ADVTXD_MACLEN_SHIFT;
1964 if (mp->m_pkthdr.csum_flags & CSUM_IP) {
1965 type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_IPV4;
1966 ip_hlen = mp->m_pkthdr.csum_iphlen;
1967 KASSERT(ip_hlen > 0, ("invalid ip hlen"));
1969 vlan_macip_lens |= ip_hlen;
1971 type_tucmd_mlhl |= E1000_ADVTXD_DCMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
1972 if (mp->m_pkthdr.csum_flags & CSUM_TCP)
1973 type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_L4T_TCP;
1974 else if (mp->m_pkthdr.csum_flags & CSUM_UDP)
1975 type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_L4T_UDP;
1977 /* 82575 needs the queue index added */
1978 if (txr->sc->hw.mac.type == e1000_82575)
1979 mss_l4len_idx = txr->me << 4;
1981 /* Now copy bits into descriptor */
1982 TXD->vlan_macip_lens = htole32(vlan_macip_lens);
1983 TXD->type_tucmd_mlhl = htole32(type_tucmd_mlhl);
1984 TXD->seqnum_seed = htole32(0);
1985 TXD->mss_l4len_idx = htole32(mss_l4len_idx);
1987 /* We've consumed the first desc, adjust counters */
1988 if (++ctxd == txr->num_tx_desc)
1990 txr->next_avail_desc = ctxd;
1997 igb_txeof(struct igb_tx_ring *txr)
1999 struct ifnet *ifp = &txr->sc->arpcom.ac_if;
2000 int first, hdr, avail;
2002 if (txr->tx_avail == txr->num_tx_desc)
2005 first = txr->next_to_clean;
2006 hdr = *(txr->tx_hdr);
2011 avail = txr->tx_avail;
2012 while (first != hdr) {
2013 struct igb_tx_buf *txbuf = &txr->tx_buf[first];
2016 if (txbuf->m_head) {
2017 bus_dmamap_unload(txr->tx_tag, txbuf->map);
2018 m_freem(txbuf->m_head);
2019 txbuf->m_head = NULL;
2022 if (++first == txr->num_tx_desc)
2025 txr->next_to_clean = first;
2026 txr->tx_avail = avail;
2029 * If we have a minimum free, clear OACTIVE
2030 * to tell the stack that it is OK to send packets.
2032 if (IGB_IS_NOT_OACTIVE(txr)) {
2033 ifsq_clr_oactive(txr->ifsq);
2036 * We have enough TX descriptors, turn off
2037 * the watchdog. We allow small amount of
2038 * packets (roughly intr_nsegs) pending on
2039 * the transmit ring.
2041 txr->tx_watchdog.wd_timer = 0;
2046 igb_create_rx_ring(struct igb_rx_ring *rxr)
2048 int rsize, i, error, nrxd;
2051 * Validate number of receive descriptors. It must not exceed
2052 * hardware maximum, and must be multiple of IGB_DBA_ALIGN.
2054 nrxd = device_getenv_int(rxr->sc->dev, "rxd", igb_rxd);
2055 if ((nrxd * sizeof(struct e1000_rx_desc)) % IGB_DBA_ALIGN != 0 ||
2056 nrxd > IGB_MAX_RXD || nrxd < IGB_MIN_RXD) {
2057 device_printf(rxr->sc->dev,
2058 "Using %d RX descriptors instead of %d!\n",
2059 IGB_DEFAULT_RXD, nrxd);
2060 rxr->num_rx_desc = IGB_DEFAULT_RXD;
2062 rxr->num_rx_desc = nrxd;
2066 * Allocate RX descriptor ring
2068 rsize = roundup2(rxr->num_rx_desc * sizeof(union e1000_adv_rx_desc),
2070 rxr->rxdma.dma_vaddr = bus_dmamem_coherent_any(rxr->sc->parent_tag,
2071 IGB_DBA_ALIGN, rsize, BUS_DMA_WAITOK,
2072 &rxr->rxdma.dma_tag, &rxr->rxdma.dma_map,
2073 &rxr->rxdma.dma_paddr);
2074 if (rxr->rxdma.dma_vaddr == NULL) {
2075 device_printf(rxr->sc->dev,
2076 "Unable to allocate RxDescriptor memory\n");
2079 rxr->rx_base = rxr->rxdma.dma_vaddr;
2080 bzero(rxr->rx_base, rsize);
2082 rsize = __VM_CACHELINE_ALIGN(
2083 sizeof(struct igb_rx_buf) * rxr->num_rx_desc);
2084 rxr->rx_buf = kmalloc_cachealign(rsize, M_DEVBUF, M_WAITOK | M_ZERO);
2087 * Create DMA tag for RX buffers
2089 error = bus_dma_tag_create(rxr->sc->parent_tag,
2090 1, 0, /* alignment, bounds */
2091 BUS_SPACE_MAXADDR, /* lowaddr */
2092 BUS_SPACE_MAXADDR, /* highaddr */
2093 NULL, NULL, /* filter, filterarg */
2094 MCLBYTES, /* maxsize */
2096 MCLBYTES, /* maxsegsize */
2097 BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW, /* flags */
2100 device_printf(rxr->sc->dev,
2101 "Unable to create RX payload DMA tag\n");
2102 kfree(rxr->rx_buf, M_DEVBUF);
2108 * Create spare DMA map for RX buffers
2110 error = bus_dmamap_create(rxr->rx_tag, BUS_DMA_WAITOK,
2113 device_printf(rxr->sc->dev,
2114 "Unable to create spare RX DMA maps\n");
2115 bus_dma_tag_destroy(rxr->rx_tag);
2116 kfree(rxr->rx_buf, M_DEVBUF);
2122 * Create DMA maps for RX buffers
2124 for (i = 0; i < rxr->num_rx_desc; i++) {
2125 struct igb_rx_buf *rxbuf = &rxr->rx_buf[i];
2127 error = bus_dmamap_create(rxr->rx_tag,
2128 BUS_DMA_WAITOK, &rxbuf->map);
2130 device_printf(rxr->sc->dev,
2131 "Unable to create RX DMA maps\n");
2132 igb_destroy_rx_ring(rxr, i);
2138 * Initialize various watermark
2140 rxr->wreg_nsegs = IGB_DEF_RXWREG_NSEGS;
2146 igb_free_rx_ring(struct igb_rx_ring *rxr)
2150 for (i = 0; i < rxr->num_rx_desc; ++i) {
2151 struct igb_rx_buf *rxbuf = &rxr->rx_buf[i];
2153 if (rxbuf->m_head != NULL) {
2154 bus_dmamap_unload(rxr->rx_tag, rxbuf->map);
2155 m_freem(rxbuf->m_head);
2156 rxbuf->m_head = NULL;
2160 if (rxr->fmp != NULL)
2167 igb_destroy_rx_ring(struct igb_rx_ring *rxr, int ndesc)
2171 if (rxr->rxdma.dma_vaddr != NULL) {
2172 bus_dmamap_unload(rxr->rxdma.dma_tag, rxr->rxdma.dma_map);
2173 bus_dmamem_free(rxr->rxdma.dma_tag, rxr->rxdma.dma_vaddr,
2174 rxr->rxdma.dma_map);
2175 bus_dma_tag_destroy(rxr->rxdma.dma_tag);
2176 rxr->rxdma.dma_vaddr = NULL;
2179 if (rxr->rx_buf == NULL)
2182 for (i = 0; i < ndesc; ++i) {
2183 struct igb_rx_buf *rxbuf = &rxr->rx_buf[i];
2185 KKASSERT(rxbuf->m_head == NULL);
2186 bus_dmamap_destroy(rxr->rx_tag, rxbuf->map);
2188 bus_dmamap_destroy(rxr->rx_tag, rxr->rx_sparemap);
2189 bus_dma_tag_destroy(rxr->rx_tag);
2191 kfree(rxr->rx_buf, M_DEVBUF);
2196 igb_setup_rxdesc(union e1000_adv_rx_desc *rxd, const struct igb_rx_buf *rxbuf)
2198 rxd->read.pkt_addr = htole64(rxbuf->paddr);
2199 rxd->wb.upper.status_error = 0;
2203 igb_newbuf(struct igb_rx_ring *rxr, int i, boolean_t wait)
2206 bus_dma_segment_t seg;
2208 struct igb_rx_buf *rxbuf;
2211 m = m_getcl(wait ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
2214 if_printf(&rxr->sc->arpcom.ac_if,
2215 "Unable to allocate RX mbuf\n");
2219 m->m_len = m->m_pkthdr.len = MCLBYTES;
2221 if (rxr->sc->max_frame_size <= MCLBYTES - ETHER_ALIGN)
2222 m_adj(m, ETHER_ALIGN);
2224 error = bus_dmamap_load_mbuf_segment(rxr->rx_tag,
2225 rxr->rx_sparemap, m, &seg, 1, &nseg, BUS_DMA_NOWAIT);
2229 if_printf(&rxr->sc->arpcom.ac_if,
2230 "Unable to load RX mbuf\n");
2235 rxbuf = &rxr->rx_buf[i];
2236 if (rxbuf->m_head != NULL)
2237 bus_dmamap_unload(rxr->rx_tag, rxbuf->map);
2240 rxbuf->map = rxr->rx_sparemap;
2241 rxr->rx_sparemap = map;
2244 rxbuf->paddr = seg.ds_addr;
2246 igb_setup_rxdesc(&rxr->rx_base[i], rxbuf);
2251 igb_init_rx_ring(struct igb_rx_ring *rxr)
2255 /* Clear the ring contents */
2257 rxr->num_rx_desc * sizeof(union e1000_adv_rx_desc));
2259 /* Now replenish the ring mbufs */
2260 for (i = 0; i < rxr->num_rx_desc; ++i) {
2263 error = igb_newbuf(rxr, i, TRUE);
2268 /* Setup our descriptor indices */
2269 rxr->next_to_check = 0;
2273 rxr->discard = FALSE;
2279 igb_init_rx_unit(struct igb_softc *sc)
2281 struct ifnet *ifp = &sc->arpcom.ac_if;
2282 struct e1000_hw *hw = &sc->hw;
2283 uint32_t rctl, rxcsum, srrctl = 0;
2287 * Make sure receives are disabled while setting
2288 * up the descriptor ring
2290 rctl = E1000_READ_REG(hw, E1000_RCTL);
2291 E1000_WRITE_REG(hw, E1000_RCTL, rctl & ~E1000_RCTL_EN);
2295 ** Set up for header split
2297 if (igb_header_split) {
2298 /* Use a standard mbuf for the header */
2299 srrctl |= IGB_HDR_BUF << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
2300 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2303 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
2306 ** Set up for jumbo frames
2308 if (ifp->if_mtu > ETHERMTU) {
2309 rctl |= E1000_RCTL_LPE;
2311 if (adapter->rx_mbuf_sz == MJUMPAGESIZE) {
2312 srrctl |= 4096 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2313 rctl |= E1000_RCTL_SZ_4096 | E1000_RCTL_BSEX;
2314 } else if (adapter->rx_mbuf_sz > MJUMPAGESIZE) {
2315 srrctl |= 8192 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2316 rctl |= E1000_RCTL_SZ_8192 | E1000_RCTL_BSEX;
2318 /* Set maximum packet len */
2319 psize = adapter->max_frame_size;
2320 /* are we on a vlan? */
2321 if (adapter->ifp->if_vlantrunk != NULL)
2322 psize += VLAN_TAG_SIZE;
2323 E1000_WRITE_REG(&adapter->hw, E1000_RLPML, psize);
2325 srrctl |= 2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2326 rctl |= E1000_RCTL_SZ_2048;
2329 rctl &= ~E1000_RCTL_LPE;
2330 srrctl |= 2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2331 rctl |= E1000_RCTL_SZ_2048;
2334 /* Setup the Base and Length of the Rx Descriptor Rings */
2335 for (i = 0; i < sc->rx_ring_inuse; ++i) {
2336 struct igb_rx_ring *rxr = &sc->rx_rings[i];
2337 uint64_t bus_addr = rxr->rxdma.dma_paddr;
2340 E1000_WRITE_REG(hw, E1000_RDLEN(i),
2341 rxr->num_rx_desc * sizeof(struct e1000_rx_desc));
2342 E1000_WRITE_REG(hw, E1000_RDBAH(i),
2343 (uint32_t)(bus_addr >> 32));
2344 E1000_WRITE_REG(hw, E1000_RDBAL(i),
2345 (uint32_t)bus_addr);
2346 E1000_WRITE_REG(hw, E1000_SRRCTL(i), srrctl);
2347 /* Enable this Queue */
2348 rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(i));
2349 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
2350 rxdctl &= 0xFFF00000;
2351 rxdctl |= IGB_RX_PTHRESH;
2352 rxdctl |= IGB_RX_HTHRESH << 8;
2354 * Don't set WTHRESH to a value above 1 on 82576, see:
2355 * 82576 specification update errata #26
2357 rxdctl |= IGB_RX_WTHRESH << 16;
2358 E1000_WRITE_REG(hw, E1000_RXDCTL(i), rxdctl);
2361 rxcsum = E1000_READ_REG(&sc->hw, E1000_RXCSUM);
2362 rxcsum &= ~(E1000_RXCSUM_PCSS_MASK | E1000_RXCSUM_IPPCSE);
2365 * Receive Checksum Offload for TCP and UDP
2367 * Checksum offloading is also enabled if multiple receive
2368 * queue is to be supported, since we need it to figure out
2371 if ((ifp->if_capenable & IFCAP_RXCSUM) || IGB_ENABLE_HWRSS(sc)) {
2374 * PCSD must be enabled to enable multiple
2377 rxcsum |= E1000_RXCSUM_IPOFL | E1000_RXCSUM_TUOFL |
2380 rxcsum &= ~(E1000_RXCSUM_IPOFL | E1000_RXCSUM_TUOFL |
2383 E1000_WRITE_REG(&sc->hw, E1000_RXCSUM, rxcsum);
2385 if (IGB_ENABLE_HWRSS(sc)) {
2386 uint8_t key[IGB_NRSSRK * IGB_RSSRK_SIZE];
2387 uint32_t reta_shift;
2392 * When we reach here, RSS has already been disabled
2393 * in igb_stop(), so we could safely configure RSS key
2394 * and redirect table.
2400 toeplitz_get_key(key, sizeof(key));
2401 for (i = 0; i < IGB_NRSSRK; ++i) {
2404 rssrk = IGB_RSSRK_VAL(key, i);
2405 IGB_RSS_DPRINTF(sc, 1, "rssrk%d 0x%08x\n", i, rssrk);
2407 E1000_WRITE_REG(hw, E1000_RSSRK(i), rssrk);
2411 * Configure RSS redirect table in following fashion:
2412 * (hash & ring_cnt_mask) == rdr_table[(hash & rdr_table_mask)]
2414 reta_shift = IGB_RETA_SHIFT;
2415 if (hw->mac.type == e1000_82575)
2416 reta_shift = IGB_RETA_SHIFT_82575;
2419 for (j = 0; j < IGB_NRETA; ++j) {
2422 for (i = 0; i < IGB_RETA_SIZE; ++i) {
2425 q = (r % sc->rx_ring_inuse) << reta_shift;
2426 reta |= q << (8 * i);
2429 IGB_RSS_DPRINTF(sc, 1, "reta 0x%08x\n", reta);
2430 E1000_WRITE_REG(hw, E1000_RETA(j), reta);
2434 * Enable multiple receive queues.
2435 * Enable IPv4 RSS standard hash functions.
2436 * Disable RSS interrupt on 82575
2438 E1000_WRITE_REG(&sc->hw, E1000_MRQC,
2439 E1000_MRQC_ENABLE_RSS_4Q |
2440 E1000_MRQC_RSS_FIELD_IPV4_TCP |
2441 E1000_MRQC_RSS_FIELD_IPV4);
2444 /* Setup the Receive Control Register */
2445 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2446 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_LBM_NO |
2447 E1000_RCTL_RDMTS_HALF |
2448 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2449 /* Strip CRC bytes. */
2450 rctl |= E1000_RCTL_SECRC;
2451 /* Make sure VLAN Filters are off */
2452 rctl &= ~E1000_RCTL_VFE;
2453 /* Don't store bad packets */
2454 rctl &= ~E1000_RCTL_SBP;
2456 /* Enable Receives */
2457 E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2460 * Setup the HW Rx Head and Tail Descriptor Pointers
2461 * - needs to be after enable
2463 for (i = 0; i < sc->rx_ring_inuse; ++i) {
2464 struct igb_rx_ring *rxr = &sc->rx_rings[i];
2466 E1000_WRITE_REG(hw, E1000_RDH(i), rxr->next_to_check);
2467 E1000_WRITE_REG(hw, E1000_RDT(i), rxr->num_rx_desc - 1);
2472 igb_rx_refresh(struct igb_rx_ring *rxr, int i)
2475 i = rxr->num_rx_desc - 1;
2476 E1000_WRITE_REG(&rxr->sc->hw, E1000_RDT(rxr->me), i);
2480 igb_rxeof(struct igb_rx_ring *rxr, int count)
2482 struct ifnet *ifp = &rxr->sc->arpcom.ac_if;
2483 union e1000_adv_rx_desc *cur;
2487 i = rxr->next_to_check;
2488 cur = &rxr->rx_base[i];
2489 staterr = le32toh(cur->wb.upper.status_error);
2491 if ((staterr & E1000_RXD_STAT_DD) == 0)
2494 while ((staterr & E1000_RXD_STAT_DD) && count != 0) {
2495 struct pktinfo *pi = NULL, pi0;
2496 struct igb_rx_buf *rxbuf = &rxr->rx_buf[i];
2497 struct mbuf *m = NULL;
2500 eop = (staterr & E1000_RXD_STAT_EOP) ? TRUE : FALSE;
2505 if ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) == 0 &&
2507 struct mbuf *mp = rxbuf->m_head;
2508 uint32_t hash, hashtype;
2512 len = le16toh(cur->wb.upper.length);
2513 if (rxr->sc->hw.mac.type == e1000_i350 &&
2514 (staterr & E1000_RXDEXT_STATERR_LB))
2515 vlan = be16toh(cur->wb.upper.vlan);
2517 vlan = le16toh(cur->wb.upper.vlan);
2519 hash = le32toh(cur->wb.lower.hi_dword.rss);
2520 hashtype = le32toh(cur->wb.lower.lo_dword.data) &
2521 E1000_RXDADV_RSSTYPE_MASK;
2523 IGB_RSS_DPRINTF(rxr->sc, 10,
2524 "ring%d, hash 0x%08x, hashtype %u\n",
2525 rxr->me, hash, hashtype);
2527 bus_dmamap_sync(rxr->rx_tag, rxbuf->map,
2528 BUS_DMASYNC_POSTREAD);
2530 if (igb_newbuf(rxr, i, FALSE) != 0) {
2536 if (rxr->fmp == NULL) {
2537 mp->m_pkthdr.len = len;
2541 rxr->lmp->m_next = mp;
2542 rxr->lmp = rxr->lmp->m_next;
2543 rxr->fmp->m_pkthdr.len += len;
2551 m->m_pkthdr.rcvif = ifp;
2554 if (ifp->if_capenable & IFCAP_RXCSUM)
2555 igb_rxcsum(staterr, m);
2557 if (staterr & E1000_RXD_STAT_VP) {
2558 m->m_pkthdr.ether_vlantag = vlan;
2559 m->m_flags |= M_VLANTAG;
2562 if (ifp->if_capenable & IFCAP_RSS) {
2563 pi = igb_rssinfo(m, &pi0,
2564 hash, hashtype, staterr);
2566 #ifdef IGB_RSS_DEBUG
2573 igb_setup_rxdesc(cur, rxbuf);
2575 rxr->discard = TRUE;
2577 rxr->discard = FALSE;
2578 if (rxr->fmp != NULL) {
2587 ether_input_pkt(ifp, m, pi);
2589 /* Advance our pointers to the next descriptor. */
2590 if (++i == rxr->num_rx_desc)
2593 if (ncoll >= rxr->wreg_nsegs) {
2594 igb_rx_refresh(rxr, i);
2598 cur = &rxr->rx_base[i];
2599 staterr = le32toh(cur->wb.upper.status_error);
2601 rxr->next_to_check = i;
2604 igb_rx_refresh(rxr, i);
2609 igb_set_vlan(struct igb_softc *sc)
2611 struct e1000_hw *hw = &sc->hw;
2614 struct ifnet *ifp = sc->arpcom.ac_if;
2618 e1000_rlpml_set_vf(hw, sc->max_frame_size + VLAN_TAG_SIZE);
2622 reg = E1000_READ_REG(hw, E1000_CTRL);
2623 reg |= E1000_CTRL_VME;
2624 E1000_WRITE_REG(hw, E1000_CTRL, reg);
2627 /* Enable the Filter Table */
2628 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) {
2629 reg = E1000_READ_REG(hw, E1000_RCTL);
2630 reg &= ~E1000_RCTL_CFIEN;
2631 reg |= E1000_RCTL_VFE;
2632 E1000_WRITE_REG(hw, E1000_RCTL, reg);
2636 /* Update the frame size */
2637 E1000_WRITE_REG(&sc->hw, E1000_RLPML,
2638 sc->max_frame_size + VLAN_TAG_SIZE);
2641 /* Don't bother with table if no vlans */
2642 if ((adapter->num_vlans == 0) ||
2643 ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0))
2646 ** A soft reset zero's out the VFTA, so
2647 ** we need to repopulate it now.
2649 for (int i = 0; i < IGB_VFTA_SIZE; i++)
2650 if (adapter->shadow_vfta[i] != 0) {
2651 if (adapter->vf_ifp)
2652 e1000_vfta_set_vf(hw,
2653 adapter->shadow_vfta[i], TRUE);
2655 E1000_WRITE_REG_ARRAY(hw, E1000_VFTA,
2656 i, adapter->shadow_vfta[i]);
2662 igb_enable_intr(struct igb_softc *sc)
2664 if (sc->intr_type != PCI_INTR_TYPE_MSIX) {
2665 lwkt_serialize_handler_enable(&sc->main_serialize);
2669 for (i = 0; i < sc->msix_cnt; ++i) {
2670 lwkt_serialize_handler_enable(
2671 sc->msix_data[i].msix_serialize);
2675 if ((sc->flags & IGB_FLAG_SHARED_INTR) == 0) {
2676 if (sc->intr_type == PCI_INTR_TYPE_MSIX)
2677 E1000_WRITE_REG(&sc->hw, E1000_EIAC, sc->intr_mask);
2679 E1000_WRITE_REG(&sc->hw, E1000_EIAC, 0);
2680 E1000_WRITE_REG(&sc->hw, E1000_EIAM, sc->intr_mask);
2681 E1000_WRITE_REG(&sc->hw, E1000_EIMS, sc->intr_mask);
2682 E1000_WRITE_REG(&sc->hw, E1000_IMS, E1000_IMS_LSC);
2684 E1000_WRITE_REG(&sc->hw, E1000_IMS, IMS_ENABLE_MASK);
2686 E1000_WRITE_FLUSH(&sc->hw);
2690 igb_disable_intr(struct igb_softc *sc)
2692 if ((sc->flags & IGB_FLAG_SHARED_INTR) == 0) {
2693 E1000_WRITE_REG(&sc->hw, E1000_EIMC, 0xffffffff);
2694 E1000_WRITE_REG(&sc->hw, E1000_EIAC, 0);
2696 E1000_WRITE_REG(&sc->hw, E1000_IMC, 0xffffffff);
2697 E1000_WRITE_FLUSH(&sc->hw);
2699 if (sc->intr_type != PCI_INTR_TYPE_MSIX) {
2700 lwkt_serialize_handler_disable(&sc->main_serialize);
2704 for (i = 0; i < sc->msix_cnt; ++i) {
2705 lwkt_serialize_handler_disable(
2706 sc->msix_data[i].msix_serialize);
2712 * Bit of a misnomer, what this really means is
2713 * to enable OS management of the system... aka
2714 * to disable special hardware management features
2717 igb_get_mgmt(struct igb_softc *sc)
2719 if (sc->flags & IGB_FLAG_HAS_MGMT) {
2720 int manc2h = E1000_READ_REG(&sc->hw, E1000_MANC2H);
2721 int manc = E1000_READ_REG(&sc->hw, E1000_MANC);
2723 /* disable hardware interception of ARP */
2724 manc &= ~E1000_MANC_ARP_EN;
2726 /* enable receiving management packets to the host */
2727 manc |= E1000_MANC_EN_MNG2HOST;
2728 manc2h |= 1 << 5; /* Mng Port 623 */
2729 manc2h |= 1 << 6; /* Mng Port 664 */
2730 E1000_WRITE_REG(&sc->hw, E1000_MANC2H, manc2h);
2731 E1000_WRITE_REG(&sc->hw, E1000_MANC, manc);
2736 * Give control back to hardware management controller
2740 igb_rel_mgmt(struct igb_softc *sc)
2742 if (sc->flags & IGB_FLAG_HAS_MGMT) {
2743 int manc = E1000_READ_REG(&sc->hw, E1000_MANC);
2745 /* Re-enable hardware interception of ARP */
2746 manc |= E1000_MANC_ARP_EN;
2747 manc &= ~E1000_MANC_EN_MNG2HOST;
2749 E1000_WRITE_REG(&sc->hw, E1000_MANC, manc);
2754 * Sets CTRL_EXT:DRV_LOAD bit.
2756 * For ASF and Pass Through versions of f/w this means that
2757 * the driver is loaded.
2760 igb_get_hw_control(struct igb_softc *sc)
2767 /* Let firmware know the driver has taken over */
2768 ctrl_ext = E1000_READ_REG(&sc->hw, E1000_CTRL_EXT);
2769 E1000_WRITE_REG(&sc->hw, E1000_CTRL_EXT,
2770 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2774 * Resets CTRL_EXT:DRV_LOAD bit.
2776 * For ASF and Pass Through versions of f/w this means that the
2777 * driver is no longer loaded.
2780 igb_rel_hw_control(struct igb_softc *sc)
2787 /* Let firmware taken over control of h/w */
2788 ctrl_ext = E1000_READ_REG(&sc->hw, E1000_CTRL_EXT);
2789 E1000_WRITE_REG(&sc->hw, E1000_CTRL_EXT,
2790 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2794 igb_is_valid_ether_addr(const uint8_t *addr)
2796 uint8_t zero_addr[ETHER_ADDR_LEN] = { 0, 0, 0, 0, 0, 0 };
2798 if ((addr[0] & 1) || !bcmp(addr, zero_addr, ETHER_ADDR_LEN))
2804 * Enable PCI Wake On Lan capability
2807 igb_enable_wol(device_t dev)
2809 uint16_t cap, status;
2812 /* First find the capabilities pointer*/
2813 cap = pci_read_config(dev, PCIR_CAP_PTR, 2);
2815 /* Read the PM Capabilities */
2816 id = pci_read_config(dev, cap, 1);
2817 if (id != PCIY_PMG) /* Something wrong */
2821 * OK, we have the power capabilities,
2822 * so now get the status register
2824 cap += PCIR_POWER_STATUS;
2825 status = pci_read_config(dev, cap, 2);
2826 status |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
2827 pci_write_config(dev, cap, status, 2);
2831 igb_update_stats_counters(struct igb_softc *sc)
2833 struct e1000_hw *hw = &sc->hw;
2834 struct e1000_hw_stats *stats;
2835 struct ifnet *ifp = &sc->arpcom.ac_if;
2838 * The virtual function adapter has only a
2839 * small controlled set of stats, do only
2843 igb_update_vf_stats_counters(sc);
2848 if (sc->hw.phy.media_type == e1000_media_type_copper ||
2849 (E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU)) {
2851 E1000_READ_REG(hw,E1000_SYMERRS);
2852 stats->sec += E1000_READ_REG(hw, E1000_SEC);
2855 stats->crcerrs += E1000_READ_REG(hw, E1000_CRCERRS);
2856 stats->mpc += E1000_READ_REG(hw, E1000_MPC);
2857 stats->scc += E1000_READ_REG(hw, E1000_SCC);
2858 stats->ecol += E1000_READ_REG(hw, E1000_ECOL);
2860 stats->mcc += E1000_READ_REG(hw, E1000_MCC);
2861 stats->latecol += E1000_READ_REG(hw, E1000_LATECOL);
2862 stats->colc += E1000_READ_REG(hw, E1000_COLC);
2863 stats->dc += E1000_READ_REG(hw, E1000_DC);
2864 stats->rlec += E1000_READ_REG(hw, E1000_RLEC);
2865 stats->xonrxc += E1000_READ_REG(hw, E1000_XONRXC);
2866 stats->xontxc += E1000_READ_REG(hw, E1000_XONTXC);
2869 * For watchdog management we need to know if we have been
2870 * paused during the last interval, so capture that here.
2872 sc->pause_frames = E1000_READ_REG(hw, E1000_XOFFRXC);
2873 stats->xoffrxc += sc->pause_frames;
2874 stats->xofftxc += E1000_READ_REG(hw, E1000_XOFFTXC);
2875 stats->fcruc += E1000_READ_REG(hw, E1000_FCRUC);
2876 stats->prc64 += E1000_READ_REG(hw, E1000_PRC64);
2877 stats->prc127 += E1000_READ_REG(hw, E1000_PRC127);
2878 stats->prc255 += E1000_READ_REG(hw, E1000_PRC255);
2879 stats->prc511 += E1000_READ_REG(hw, E1000_PRC511);
2880 stats->prc1023 += E1000_READ_REG(hw, E1000_PRC1023);
2881 stats->prc1522 += E1000_READ_REG(hw, E1000_PRC1522);
2882 stats->gprc += E1000_READ_REG(hw, E1000_GPRC);
2883 stats->bprc += E1000_READ_REG(hw, E1000_BPRC);
2884 stats->mprc += E1000_READ_REG(hw, E1000_MPRC);
2885 stats->gptc += E1000_READ_REG(hw, E1000_GPTC);
2887 /* For the 64-bit byte counters the low dword must be read first. */
2888 /* Both registers clear on the read of the high dword */
2890 stats->gorc += E1000_READ_REG(hw, E1000_GORCL) +
2891 ((uint64_t)E1000_READ_REG(hw, E1000_GORCH) << 32);
2892 stats->gotc += E1000_READ_REG(hw, E1000_GOTCL) +
2893 ((uint64_t)E1000_READ_REG(hw, E1000_GOTCH) << 32);
2895 stats->rnbc += E1000_READ_REG(hw, E1000_RNBC);
2896 stats->ruc += E1000_READ_REG(hw, E1000_RUC);
2897 stats->rfc += E1000_READ_REG(hw, E1000_RFC);
2898 stats->roc += E1000_READ_REG(hw, E1000_ROC);
2899 stats->rjc += E1000_READ_REG(hw, E1000_RJC);
2901 stats->tor += E1000_READ_REG(hw, E1000_TORH);
2902 stats->tot += E1000_READ_REG(hw, E1000_TOTH);
2904 stats->tpr += E1000_READ_REG(hw, E1000_TPR);
2905 stats->tpt += E1000_READ_REG(hw, E1000_TPT);
2906 stats->ptc64 += E1000_READ_REG(hw, E1000_PTC64);
2907 stats->ptc127 += E1000_READ_REG(hw, E1000_PTC127);
2908 stats->ptc255 += E1000_READ_REG(hw, E1000_PTC255);
2909 stats->ptc511 += E1000_READ_REG(hw, E1000_PTC511);
2910 stats->ptc1023 += E1000_READ_REG(hw, E1000_PTC1023);
2911 stats->ptc1522 += E1000_READ_REG(hw, E1000_PTC1522);
2912 stats->mptc += E1000_READ_REG(hw, E1000_MPTC);
2913 stats->bptc += E1000_READ_REG(hw, E1000_BPTC);
2915 /* Interrupt Counts */
2917 stats->iac += E1000_READ_REG(hw, E1000_IAC);
2918 stats->icrxptc += E1000_READ_REG(hw, E1000_ICRXPTC);
2919 stats->icrxatc += E1000_READ_REG(hw, E1000_ICRXATC);
2920 stats->ictxptc += E1000_READ_REG(hw, E1000_ICTXPTC);
2921 stats->ictxatc += E1000_READ_REG(hw, E1000_ICTXATC);
2922 stats->ictxqec += E1000_READ_REG(hw, E1000_ICTXQEC);
2923 stats->ictxqmtc += E1000_READ_REG(hw, E1000_ICTXQMTC);
2924 stats->icrxdmtc += E1000_READ_REG(hw, E1000_ICRXDMTC);
2925 stats->icrxoc += E1000_READ_REG(hw, E1000_ICRXOC);
2927 /* Host to Card Statistics */
2929 stats->cbtmpc += E1000_READ_REG(hw, E1000_CBTMPC);
2930 stats->htdpmc += E1000_READ_REG(hw, E1000_HTDPMC);
2931 stats->cbrdpc += E1000_READ_REG(hw, E1000_CBRDPC);
2932 stats->cbrmpc += E1000_READ_REG(hw, E1000_CBRMPC);
2933 stats->rpthc += E1000_READ_REG(hw, E1000_RPTHC);
2934 stats->hgptc += E1000_READ_REG(hw, E1000_HGPTC);
2935 stats->htcbdpc += E1000_READ_REG(hw, E1000_HTCBDPC);
2936 stats->hgorc += (E1000_READ_REG(hw, E1000_HGORCL) +
2937 ((uint64_t)E1000_READ_REG(hw, E1000_HGORCH) << 32));
2938 stats->hgotc += (E1000_READ_REG(hw, E1000_HGOTCL) +
2939 ((uint64_t)E1000_READ_REG(hw, E1000_HGOTCH) << 32));
2940 stats->lenerrs += E1000_READ_REG(hw, E1000_LENERRS);
2941 stats->scvpc += E1000_READ_REG(hw, E1000_SCVPC);
2942 stats->hrmpc += E1000_READ_REG(hw, E1000_HRMPC);
2944 stats->algnerrc += E1000_READ_REG(hw, E1000_ALGNERRC);
2945 stats->rxerrc += E1000_READ_REG(hw, E1000_RXERRC);
2946 stats->tncrs += E1000_READ_REG(hw, E1000_TNCRS);
2947 stats->cexterr += E1000_READ_REG(hw, E1000_CEXTERR);
2948 stats->tsctc += E1000_READ_REG(hw, E1000_TSCTC);
2949 stats->tsctfc += E1000_READ_REG(hw, E1000_TSCTFC);
2951 ifp->if_collisions = stats->colc;
2954 ifp->if_ierrors = stats->rxerrc + stats->crcerrs + stats->algnerrc +
2955 stats->ruc + stats->roc + stats->mpc + stats->cexterr;
2958 ifp->if_oerrors = stats->ecol + stats->latecol + sc->watchdog_events;
2960 /* Driver specific counters */
2961 sc->device_control = E1000_READ_REG(hw, E1000_CTRL);
2962 sc->rx_control = E1000_READ_REG(hw, E1000_RCTL);
2963 sc->int_mask = E1000_READ_REG(hw, E1000_IMS);
2964 sc->eint_mask = E1000_READ_REG(hw, E1000_EIMS);
2965 sc->packet_buf_alloc_tx =
2966 ((E1000_READ_REG(hw, E1000_PBA) & 0xffff0000) >> 16);
2967 sc->packet_buf_alloc_rx =
2968 (E1000_READ_REG(hw, E1000_PBA) & 0xffff);
2972 igb_vf_init_stats(struct igb_softc *sc)
2974 struct e1000_hw *hw = &sc->hw;
2975 struct e1000_vf_stats *stats;
2978 stats->last_gprc = E1000_READ_REG(hw, E1000_VFGPRC);
2979 stats->last_gorc = E1000_READ_REG(hw, E1000_VFGORC);
2980 stats->last_gptc = E1000_READ_REG(hw, E1000_VFGPTC);
2981 stats->last_gotc = E1000_READ_REG(hw, E1000_VFGOTC);
2982 stats->last_mprc = E1000_READ_REG(hw, E1000_VFMPRC);
2986 igb_update_vf_stats_counters(struct igb_softc *sc)
2988 struct e1000_hw *hw = &sc->hw;
2989 struct e1000_vf_stats *stats;
2991 if (sc->link_speed == 0)
2995 UPDATE_VF_REG(E1000_VFGPRC, stats->last_gprc, stats->gprc);
2996 UPDATE_VF_REG(E1000_VFGORC, stats->last_gorc, stats->gorc);
2997 UPDATE_VF_REG(E1000_VFGPTC, stats->last_gptc, stats->gptc);
2998 UPDATE_VF_REG(E1000_VFGOTC, stats->last_gotc, stats->gotc);
2999 UPDATE_VF_REG(E1000_VFMPRC, stats->last_mprc, stats->mprc);
3002 #ifdef IFPOLL_ENABLE
3005 igb_npoll_status(struct ifnet *ifp)
3007 struct igb_softc *sc = ifp->if_softc;
3010 ASSERT_SERIALIZED(&sc->main_serialize);
3012 reg_icr = E1000_READ_REG(&sc->hw, E1000_ICR);
3013 if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3014 sc->hw.mac.get_link_status = 1;
3015 igb_update_link_status(sc);
3020 igb_npoll_tx(struct ifnet *ifp, void *arg, int cycle __unused)
3022 struct igb_tx_ring *txr = arg;
3024 ASSERT_SERIALIZED(&txr->tx_serialize);
3027 if (!ifsq_is_empty(txr->ifsq))
3028 ifsq_devstart(txr->ifsq);
3032 igb_npoll_rx(struct ifnet *ifp __unused, void *arg, int cycle)
3034 struct igb_rx_ring *rxr = arg;
3036 ASSERT_SERIALIZED(&rxr->rx_serialize);
3038 igb_rxeof(rxr, cycle);
3042 igb_npoll(struct ifnet *ifp, struct ifpoll_info *info)
3044 struct igb_softc *sc = ifp->if_softc;
3045 int i, txr_cnt, rxr_cnt;
3047 ASSERT_IFNET_SERIALIZED_ALL(ifp);
3052 info->ifpi_status.status_func = igb_npoll_status;
3053 info->ifpi_status.serializer = &sc->main_serialize;
3055 txr_cnt = igb_get_txring_inuse(sc, TRUE);
3056 off = sc->tx_npoll_off;
3057 for (i = 0; i < txr_cnt; ++i) {
3058 struct igb_tx_ring *txr = &sc->tx_rings[i];
3061 KKASSERT(idx < ncpus2);
3062 info->ifpi_tx[idx].poll_func = igb_npoll_tx;
3063 info->ifpi_tx[idx].arg = txr;
3064 info->ifpi_tx[idx].serializer = &txr->tx_serialize;
3065 ifsq_set_cpuid(txr->ifsq, idx);
3068 rxr_cnt = igb_get_rxring_inuse(sc, TRUE);
3069 off = sc->rx_npoll_off;
3070 for (i = 0; i < rxr_cnt; ++i) {
3071 struct igb_rx_ring *rxr = &sc->rx_rings[i];
3074 KKASSERT(idx < ncpus2);
3075 info->ifpi_rx[idx].poll_func = igb_npoll_rx;
3076 info->ifpi_rx[idx].arg = rxr;
3077 info->ifpi_rx[idx].serializer = &rxr->rx_serialize;
3080 if (ifp->if_flags & IFF_RUNNING) {
3081 if (rxr_cnt == sc->rx_ring_inuse &&
3082 txr_cnt == sc->tx_ring_inuse)
3083 igb_disable_intr(sc);
3088 for (i = 0; i < sc->tx_ring_cnt; ++i) {
3089 struct igb_tx_ring *txr = &sc->tx_rings[i];
3091 ifsq_set_cpuid(txr->ifsq, txr->tx_intr_cpuid);
3094 if (ifp->if_flags & IFF_RUNNING) {
3095 txr_cnt = igb_get_txring_inuse(sc, FALSE);
3096 rxr_cnt = igb_get_rxring_inuse(sc, FALSE);
3098 if (rxr_cnt == sc->rx_ring_inuse &&
3099 txr_cnt == sc->tx_ring_inuse)
3100 igb_enable_intr(sc);
3107 #endif /* IFPOLL_ENABLE */
3112 struct igb_softc *sc = xsc;
3113 struct ifnet *ifp = &sc->arpcom.ac_if;
3116 ASSERT_SERIALIZED(&sc->main_serialize);
3118 eicr = E1000_READ_REG(&sc->hw, E1000_EICR);
3123 if (ifp->if_flags & IFF_RUNNING) {
3124 struct igb_tx_ring *txr = &sc->tx_rings[0];
3127 for (i = 0; i < sc->rx_ring_inuse; ++i) {
3128 struct igb_rx_ring *rxr = &sc->rx_rings[i];
3130 if (eicr & rxr->rx_intr_mask) {
3131 lwkt_serialize_enter(&rxr->rx_serialize);
3133 lwkt_serialize_exit(&rxr->rx_serialize);
3137 if (eicr & txr->tx_intr_mask) {
3138 lwkt_serialize_enter(&txr->tx_serialize);
3140 if (!ifsq_is_empty(txr->ifsq))
3141 ifsq_devstart(txr->ifsq);
3142 lwkt_serialize_exit(&txr->tx_serialize);
3146 if (eicr & E1000_EICR_OTHER) {
3147 uint32_t icr = E1000_READ_REG(&sc->hw, E1000_ICR);
3149 /* Link status change */
3150 if (icr & E1000_ICR_LSC) {
3151 sc->hw.mac.get_link_status = 1;
3152 igb_update_link_status(sc);
3157 * Reading EICR has the side effect to clear interrupt mask,
3158 * so all interrupts need to be enabled here.
3160 E1000_WRITE_REG(&sc->hw, E1000_EIMS, sc->intr_mask);
3164 igb_intr_shared(void *xsc)
3166 struct igb_softc *sc = xsc;
3167 struct ifnet *ifp = &sc->arpcom.ac_if;
3170 ASSERT_SERIALIZED(&sc->main_serialize);
3172 reg_icr = E1000_READ_REG(&sc->hw, E1000_ICR);
3175 if (reg_icr == 0xffffffff)
3178 /* Definitely not our interrupt. */
3182 if ((reg_icr & E1000_ICR_INT_ASSERTED) == 0)
3185 if (ifp->if_flags & IFF_RUNNING) {
3187 (E1000_ICR_RXT0 | E1000_ICR_RXDMT0 | E1000_ICR_RXO)) {
3190 for (i = 0; i < sc->rx_ring_inuse; ++i) {
3191 struct igb_rx_ring *rxr = &sc->rx_rings[i];
3193 lwkt_serialize_enter(&rxr->rx_serialize);
3195 lwkt_serialize_exit(&rxr->rx_serialize);
3199 if (reg_icr & E1000_ICR_TXDW) {
3200 struct igb_tx_ring *txr = &sc->tx_rings[0];
3202 lwkt_serialize_enter(&txr->tx_serialize);
3204 if (!ifsq_is_empty(txr->ifsq))
3205 ifsq_devstart(txr->ifsq);
3206 lwkt_serialize_exit(&txr->tx_serialize);
3210 /* Link status change */
3211 if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3212 sc->hw.mac.get_link_status = 1;
3213 igb_update_link_status(sc);
3216 if (reg_icr & E1000_ICR_RXO)
3221 igb_encap(struct igb_tx_ring *txr, struct mbuf **m_headp,
3222 int *segs_used, int *idx)
3224 bus_dma_segment_t segs[IGB_MAX_SCATTER];
3226 struct igb_tx_buf *tx_buf, *tx_buf_mapped;
3227 union e1000_adv_tx_desc *txd = NULL;
3228 struct mbuf *m_head = *m_headp;
3229 uint32_t olinfo_status = 0, cmd_type_len = 0, cmd_rs = 0;
3230 int maxsegs, nsegs, i, j, error;
3231 uint32_t hdrlen = 0;
3233 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
3234 error = igb_tso_pullup(txr, m_headp);
3240 /* Set basic descriptor constants */
3241 cmd_type_len |= E1000_ADVTXD_DTYP_DATA;
3242 cmd_type_len |= E1000_ADVTXD_DCMD_IFCS | E1000_ADVTXD_DCMD_DEXT;
3243 if (m_head->m_flags & M_VLANTAG)
3244 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
3247 * Map the packet for DMA.
3249 tx_buf = &txr->tx_buf[txr->next_avail_desc];
3250 tx_buf_mapped = tx_buf;
3253 maxsegs = txr->tx_avail - IGB_TX_RESERVED;
3254 KASSERT(maxsegs >= txr->spare_desc, ("not enough spare TX desc\n"));
3255 if (maxsegs > IGB_MAX_SCATTER)
3256 maxsegs = IGB_MAX_SCATTER;
3258 error = bus_dmamap_load_mbuf_defrag(txr->tx_tag, map, m_headp,
3259 segs, maxsegs, &nsegs, BUS_DMA_NOWAIT);
3261 if (error == ENOBUFS)
3262 txr->sc->mbuf_defrag_failed++;
3264 txr->sc->no_tx_dma_setup++;
3270 bus_dmamap_sync(txr->tx_tag, map, BUS_DMASYNC_PREWRITE);
3275 * Set up the TX context descriptor, if any hardware offloading is
3276 * needed. This includes CSUM, VLAN, and TSO. It will consume one
3279 * Unlike these chips' predecessors (em/emx), TX context descriptor
3280 * will _not_ interfere TX data fetching pipelining.
3282 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
3283 igb_tso_ctx(txr, m_head, &hdrlen);
3284 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
3285 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
3286 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3289 } else if (igb_txcsum_ctx(txr, m_head)) {
3290 if (m_head->m_pkthdr.csum_flags & CSUM_IP)
3291 olinfo_status |= (E1000_TXD_POPTS_IXSM << 8);
3292 if (m_head->m_pkthdr.csum_flags & (CSUM_UDP | CSUM_TCP))
3293 olinfo_status |= (E1000_TXD_POPTS_TXSM << 8);
3298 *segs_used += nsegs;
3299 txr->tx_nsegs += nsegs;
3300 if (txr->tx_nsegs >= txr->intr_nsegs) {
3302 * Report Status (RS) is turned on every intr_nsegs
3303 * descriptors (roughly).
3306 cmd_rs = E1000_ADVTXD_DCMD_RS;
3309 /* Calculate payload length */
3310 olinfo_status |= ((m_head->m_pkthdr.len - hdrlen)
3311 << E1000_ADVTXD_PAYLEN_SHIFT);
3313 /* 82575 needs the queue index added */
3314 if (txr->sc->hw.mac.type == e1000_82575)
3315 olinfo_status |= txr->me << 4;
3317 /* Set up our transmit descriptors */
3318 i = txr->next_avail_desc;
3319 for (j = 0; j < nsegs; j++) {
3321 bus_addr_t seg_addr;
3323 tx_buf = &txr->tx_buf[i];
3324 txd = (union e1000_adv_tx_desc *)&txr->tx_base[i];
3325 seg_addr = segs[j].ds_addr;
3326 seg_len = segs[j].ds_len;
3328 txd->read.buffer_addr = htole64(seg_addr);
3329 txd->read.cmd_type_len = htole32(cmd_type_len | seg_len);
3330 txd->read.olinfo_status = htole32(olinfo_status);
3331 if (++i == txr->num_tx_desc)
3333 tx_buf->m_head = NULL;
3336 KASSERT(txr->tx_avail > nsegs, ("invalid avail TX desc\n"));
3337 txr->next_avail_desc = i;
3338 txr->tx_avail -= nsegs;
3340 tx_buf->m_head = m_head;
3341 tx_buf_mapped->map = tx_buf->map;
3345 * Last Descriptor of Packet needs End Of Packet (EOP)
3347 txd->read.cmd_type_len |= htole32(E1000_ADVTXD_DCMD_EOP | cmd_rs);
3350 * Defer TDT updating, until enough descrptors are setup
3359 igb_start(struct ifnet *ifp, struct ifaltq_subque *ifsq)
3361 struct igb_softc *sc = ifp->if_softc;
3362 struct igb_tx_ring *txr = ifsq_get_priv(ifsq);
3363 struct mbuf *m_head;
3364 int idx = -1, nsegs = 0;
3366 KKASSERT(txr->ifsq == ifsq);
3367 ASSERT_SERIALIZED(&txr->tx_serialize);
3369 if ((ifp->if_flags & IFF_RUNNING) == 0 || ifsq_is_oactive(ifsq))
3372 if (!sc->link_active) {
3377 if (!IGB_IS_NOT_OACTIVE(txr))
3380 while (!ifsq_is_empty(ifsq)) {
3381 if (IGB_IS_OACTIVE(txr)) {
3382 ifsq_set_oactive(ifsq);
3383 /* Set watchdog on */
3384 txr->tx_watchdog.wd_timer = 5;
3388 m_head = ifsq_dequeue(ifsq, NULL);
3392 if (igb_encap(txr, &m_head, &nsegs, &idx)) {
3397 if (nsegs >= txr->wreg_nsegs) {
3398 E1000_WRITE_REG(&txr->sc->hw, E1000_TDT(txr->me), idx);
3403 /* Send a copy of the frame to the BPF listener */
3404 ETHER_BPF_MTAP(ifp, m_head);
3407 E1000_WRITE_REG(&txr->sc->hw, E1000_TDT(txr->me), idx);
3411 igb_watchdog(struct ifaltq_subque *ifsq)
3413 struct igb_tx_ring *txr = ifsq_get_priv(ifsq);
3414 struct ifnet *ifp = ifsq_get_ifp(ifsq);
3415 struct igb_softc *sc = ifp->if_softc;
3418 KKASSERT(txr->ifsq == ifsq);
3419 ASSERT_IFNET_SERIALIZED_ALL(ifp);
3422 * If flow control has paused us since last checking
3423 * it invalidates the watchdog timing, so dont run it.
3425 if (sc->pause_frames) {
3426 sc->pause_frames = 0;
3427 txr->tx_watchdog.wd_timer = 5;
3431 if_printf(ifp, "Watchdog timeout -- resetting\n");
3432 if_printf(ifp, "Queue(%d) tdh = %d, hw tdt = %d\n", txr->me,
3433 E1000_READ_REG(&sc->hw, E1000_TDH(txr->me)),
3434 E1000_READ_REG(&sc->hw, E1000_TDT(txr->me)));
3435 if_printf(ifp, "TX(%d) desc avail = %d, "
3436 "Next TX to Clean = %d\n",
3437 txr->me, txr->tx_avail, txr->next_to_clean);
3440 sc->watchdog_events++;
3443 for (i = 0; i < sc->tx_ring_inuse; ++i)
3444 ifsq_devstart_sched(sc->tx_rings[i].ifsq);
3448 igb_set_eitr(struct igb_softc *sc, int idx, int rate)
3453 if (sc->hw.mac.type == e1000_82575) {
3454 eitr = 1000000000 / 256 / rate;
3457 * Document is wrong on the 2 bits left shift
3460 eitr = 1000000 / rate;
3461 eitr <<= IGB_EITR_INTVL_SHIFT;
3465 /* Don't disable it */
3466 eitr = 1 << IGB_EITR_INTVL_SHIFT;
3467 } else if (eitr > IGB_EITR_INTVL_MASK) {
3468 /* Don't allow it to be too large */
3469 eitr = IGB_EITR_INTVL_MASK;
3472 if (sc->hw.mac.type == e1000_82575)
3475 eitr |= E1000_EITR_CNT_IGNR;
3476 E1000_WRITE_REG(&sc->hw, E1000_EITR(idx), eitr);
3480 igb_sysctl_intr_rate(SYSCTL_HANDLER_ARGS)
3482 struct igb_softc *sc = (void *)arg1;
3483 struct ifnet *ifp = &sc->arpcom.ac_if;
3484 int error, intr_rate;
3486 intr_rate = sc->intr_rate;
3487 error = sysctl_handle_int(oidp, &intr_rate, 0, req);
3488 if (error || req->newptr == NULL)
3493 ifnet_serialize_all(ifp);
3495 sc->intr_rate = intr_rate;
3496 if (ifp->if_flags & IFF_RUNNING)
3497 igb_set_eitr(sc, 0, sc->intr_rate);
3500 if_printf(ifp, "interrupt rate set to %d/sec\n", sc->intr_rate);
3502 ifnet_deserialize_all(ifp);
3508 igb_sysctl_msix_rate(SYSCTL_HANDLER_ARGS)
3510 struct igb_msix_data *msix = (void *)arg1;
3511 struct igb_softc *sc = msix->msix_sc;
3512 struct ifnet *ifp = &sc->arpcom.ac_if;
3513 int error, msix_rate;
3515 msix_rate = msix->msix_rate;
3516 error = sysctl_handle_int(oidp, &msix_rate, 0, req);
3517 if (error || req->newptr == NULL)
3522 lwkt_serialize_enter(msix->msix_serialize);
3524 msix->msix_rate = msix_rate;
3525 if (ifp->if_flags & IFF_RUNNING)
3526 igb_set_eitr(sc, msix->msix_vector, msix->msix_rate);
3529 if_printf(ifp, "%s set to %d/sec\n", msix->msix_rate_desc,
3533 lwkt_serialize_exit(msix->msix_serialize);
3539 igb_sysctl_tx_intr_nsegs(SYSCTL_HANDLER_ARGS)
3541 struct igb_softc *sc = (void *)arg1;
3542 struct ifnet *ifp = &sc->arpcom.ac_if;
3543 struct igb_tx_ring *txr = &sc->tx_rings[0];
3546 nsegs = txr->intr_nsegs;
3547 error = sysctl_handle_int(oidp, &nsegs, 0, req);
3548 if (error || req->newptr == NULL)
3553 ifnet_serialize_all(ifp);
3555 if (nsegs >= txr->num_tx_desc - txr->oact_lo_desc ||
3556 nsegs >= txr->oact_hi_desc - IGB_MAX_SCATTER) {
3562 for (i = 0; i < sc->tx_ring_cnt; ++i)
3563 sc->tx_rings[i].intr_nsegs = nsegs;
3566 ifnet_deserialize_all(ifp);
3572 igb_sysctl_rx_wreg_nsegs(SYSCTL_HANDLER_ARGS)
3574 struct igb_softc *sc = (void *)arg1;
3575 struct ifnet *ifp = &sc->arpcom.ac_if;
3576 int error, nsegs, i;
3578 nsegs = sc->rx_rings[0].wreg_nsegs;
3579 error = sysctl_handle_int(oidp, &nsegs, 0, req);
3580 if (error || req->newptr == NULL)
3583 ifnet_serialize_all(ifp);
3584 for (i = 0; i < sc->rx_ring_cnt; ++i)
3585 sc->rx_rings[i].wreg_nsegs =nsegs;
3586 ifnet_deserialize_all(ifp);
3592 igb_sysctl_tx_wreg_nsegs(SYSCTL_HANDLER_ARGS)
3594 struct igb_softc *sc = (void *)arg1;
3595 struct ifnet *ifp = &sc->arpcom.ac_if;
3596 int error, nsegs, i;
3598 nsegs = sc->tx_rings[0].wreg_nsegs;
3599 error = sysctl_handle_int(oidp, &nsegs, 0, req);
3600 if (error || req->newptr == NULL)
3603 ifnet_serialize_all(ifp);
3604 for (i = 0; i < sc->tx_ring_cnt; ++i)
3605 sc->tx_rings[i].wreg_nsegs =nsegs;
3606 ifnet_deserialize_all(ifp);
3611 #ifdef IFPOLL_ENABLE
3614 igb_sysctl_npoll_rxoff(SYSCTL_HANDLER_ARGS)
3616 struct igb_softc *sc = (void *)arg1;
3617 struct ifnet *ifp = &sc->arpcom.ac_if;
3620 off = sc->rx_npoll_off;
3621 error = sysctl_handle_int(oidp, &off, 0, req);
3622 if (error || req->newptr == NULL)
3627 ifnet_serialize_all(ifp);
3628 if (off >= ncpus2 || off % sc->rx_ring_cnt != 0) {
3632 sc->rx_npoll_off = off;
3634 ifnet_deserialize_all(ifp);
3640 igb_sysctl_npoll_txoff(SYSCTL_HANDLER_ARGS)
3642 struct igb_softc *sc = (void *)arg1;
3643 struct ifnet *ifp = &sc->arpcom.ac_if;
3646 off = sc->tx_npoll_off;
3647 error = sysctl_handle_int(oidp, &off, 0, req);
3648 if (error || req->newptr == NULL)
3653 ifnet_serialize_all(ifp);
3654 if (off >= ncpus2 || off % sc->tx_ring_cnt != 0) {
3658 sc->tx_npoll_off = off;
3660 ifnet_deserialize_all(ifp);
3665 #endif /* IFPOLL_ENABLE */
3668 igb_init_intr(struct igb_softc *sc)
3670 igb_set_intr_mask(sc);
3672 if ((sc->flags & IGB_FLAG_SHARED_INTR) == 0)
3673 igb_init_unshared_intr(sc);
3675 if (sc->intr_type != PCI_INTR_TYPE_MSIX) {
3676 igb_set_eitr(sc, 0, sc->intr_rate);
3680 for (i = 0; i < sc->msix_cnt; ++i)
3681 igb_set_eitr(sc, i, sc->msix_data[i].msix_rate);
3686 igb_init_unshared_intr(struct igb_softc *sc)
3688 struct e1000_hw *hw = &sc->hw;
3689 const struct igb_rx_ring *rxr;
3690 const struct igb_tx_ring *txr;
3691 uint32_t ivar, index;
3695 * Enable extended mode
3697 if (sc->hw.mac.type != e1000_82575) {
3701 gpie = E1000_GPIE_NSICR;
3702 if (sc->intr_type == PCI_INTR_TYPE_MSIX) {
3703 gpie |= E1000_GPIE_MSIX_MODE |
3707 E1000_WRITE_REG(hw, E1000_GPIE, gpie);
3712 switch (sc->hw.mac.type) {
3714 ivar_max = IGB_MAX_IVAR_82580;
3718 ivar_max = IGB_MAX_IVAR_I350;
3722 case e1000_vfadapt_i350:
3723 ivar_max = IGB_MAX_IVAR_VF;
3727 ivar_max = IGB_MAX_IVAR_82576;
3731 panic("unknown mac type %d\n", sc->hw.mac.type);
3733 for (i = 0; i < ivar_max; ++i)
3734 E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, i, 0);
3735 E1000_WRITE_REG(hw, E1000_IVAR_MISC, 0);
3739 KASSERT(sc->intr_type != PCI_INTR_TYPE_MSIX,
3740 ("82575 w/ MSI-X"));
3741 tmp = E1000_READ_REG(hw, E1000_CTRL_EXT);
3742 tmp |= E1000_CTRL_EXT_IRCA;
3743 E1000_WRITE_REG(hw, E1000_CTRL_EXT, tmp);
3747 * Map TX/RX interrupts to EICR
3749 switch (sc->hw.mac.type) {
3753 case e1000_vfadapt_i350:
3755 for (i = 0; i < sc->rx_ring_inuse; ++i) {
3756 rxr = &sc->rx_rings[i];
3759 ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
3764 (rxr->rx_intr_bit | E1000_IVAR_VALID) << 16;
3768 (rxr->rx_intr_bit | E1000_IVAR_VALID);
3770 E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
3773 for (i = 0; i < sc->tx_ring_inuse; ++i) {
3774 txr = &sc->tx_rings[i];
3777 ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
3782 (txr->tx_intr_bit | E1000_IVAR_VALID) << 24;
3786 (txr->tx_intr_bit | E1000_IVAR_VALID) << 8;
3788 E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
3790 if (sc->intr_type == PCI_INTR_TYPE_MSIX) {
3791 ivar = (sc->sts_intr_bit | E1000_IVAR_VALID) << 8;
3792 E1000_WRITE_REG(hw, E1000_IVAR_MISC, ivar);
3798 for (i = 0; i < sc->rx_ring_inuse; ++i) {
3799 rxr = &sc->rx_rings[i];
3801 index = i & 0x7; /* Each IVAR has two entries */
3802 ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
3807 (rxr->rx_intr_bit | E1000_IVAR_VALID);
3811 (rxr->rx_intr_bit | E1000_IVAR_VALID) << 16;
3813 E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
3816 for (i = 0; i < sc->tx_ring_inuse; ++i) {
3817 txr = &sc->tx_rings[i];
3819 index = i & 0x7; /* Each IVAR has two entries */
3820 ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
3825 (txr->tx_intr_bit | E1000_IVAR_VALID) << 8;
3829 (txr->tx_intr_bit | E1000_IVAR_VALID) << 24;
3831 E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
3833 if (sc->intr_type == PCI_INTR_TYPE_MSIX) {
3834 ivar = (sc->sts_intr_bit | E1000_IVAR_VALID) << 8;
3835 E1000_WRITE_REG(hw, E1000_IVAR_MISC, ivar);
3841 * Enable necessary interrupt bits.
3843 * The name of the register is confusing; in addition to
3844 * configuring the first vector of MSI-X, it also configures
3845 * which bits of EICR could be set by the hardware even when
3846 * MSI or line interrupt is used; it thus controls interrupt
3847 * generation. It MUST be configured explicitly; the default
3848 * value mentioned in the datasheet is wrong: RX queue0 and
3849 * TX queue0 are NOT enabled by default.
3851 E1000_WRITE_REG(&sc->hw, E1000_MSIXBM(0), sc->intr_mask);
3855 panic("unknown mac type %d\n", sc->hw.mac.type);
3860 igb_setup_intr(struct igb_softc *sc)
3864 if (sc->intr_type == PCI_INTR_TYPE_MSIX)
3865 return igb_msix_setup(sc);
3867 error = bus_setup_intr(sc->dev, sc->intr_res, INTR_MPSAFE,
3868 (sc->flags & IGB_FLAG_SHARED_INTR) ? igb_intr_shared : igb_intr,
3869 sc, &sc->intr_tag, &sc->main_serialize);
3871 device_printf(sc->dev, "Failed to register interrupt handler");
3874 sc->tx_rings[0].tx_intr_cpuid = rman_get_cpuid(sc->intr_res);
3880 igb_set_txintr_mask(struct igb_tx_ring *txr, int *intr_bit0, int intr_bitmax)
3882 if (txr->sc->hw.mac.type == e1000_82575) {
3883 txr->tx_intr_bit = 0; /* unused */
3886 txr->tx_intr_mask = E1000_EICR_TX_QUEUE0;
3889 txr->tx_intr_mask = E1000_EICR_TX_QUEUE1;
3892 txr->tx_intr_mask = E1000_EICR_TX_QUEUE2;
3895 txr->tx_intr_mask = E1000_EICR_TX_QUEUE3;
3898 panic("unsupported # of TX ring, %d\n", txr->me);
3901 int intr_bit = *intr_bit0;
3903 txr->tx_intr_bit = intr_bit % intr_bitmax;
3904 txr->tx_intr_mask = 1 << txr->tx_intr_bit;
3906 *intr_bit0 = intr_bit + 1;
3911 igb_set_rxintr_mask(struct igb_rx_ring *rxr, int *intr_bit0, int intr_bitmax)
3913 if (rxr->sc->hw.mac.type == e1000_82575) {
3914 rxr->rx_intr_bit = 0; /* unused */
3917 rxr->rx_intr_mask = E1000_EICR_RX_QUEUE0;
3920 rxr->rx_intr_mask = E1000_EICR_RX_QUEUE1;
3923 rxr->rx_intr_mask = E1000_EICR_RX_QUEUE2;
3926 rxr->rx_intr_mask = E1000_EICR_RX_QUEUE3;
3929 panic("unsupported # of RX ring, %d\n", rxr->me);
3932 int intr_bit = *intr_bit0;
3934 rxr->rx_intr_bit = intr_bit % intr_bitmax;
3935 rxr->rx_intr_mask = 1 << rxr->rx_intr_bit;
3937 *intr_bit0 = intr_bit + 1;
3942 igb_serialize(struct ifnet *ifp, enum ifnet_serialize slz)
3944 struct igb_softc *sc = ifp->if_softc;
3946 ifnet_serialize_array_enter(sc->serializes, sc->serialize_cnt,
3947 sc->tx_serialize, sc->rx_serialize, slz);
3951 igb_deserialize(struct ifnet *ifp, enum ifnet_serialize slz)
3953 struct igb_softc *sc = ifp->if_softc;
3955 ifnet_serialize_array_exit(sc->serializes, sc->serialize_cnt,
3956 sc->tx_serialize, sc->rx_serialize, slz);
3960 igb_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz)
3962 struct igb_softc *sc = ifp->if_softc;
3964 return ifnet_serialize_array_try(sc->serializes, sc->serialize_cnt,
3965 sc->tx_serialize, sc->rx_serialize, slz);
3971 igb_serialize_assert(struct ifnet *ifp, enum ifnet_serialize slz,
3972 boolean_t serialized)
3974 struct igb_softc *sc = ifp->if_softc;
3976 ifnet_serialize_array_assert(sc->serializes, sc->serialize_cnt,
3977 sc->tx_serialize, sc->rx_serialize, slz, serialized);
3980 #endif /* INVARIANTS */
3983 igb_set_intr_mask(struct igb_softc *sc)
3987 sc->intr_mask = sc->sts_intr_mask;
3988 for (i = 0; i < sc->rx_ring_inuse; ++i)
3989 sc->intr_mask |= sc->rx_rings[i].rx_intr_mask;
3990 for (i = 0; i < sc->tx_ring_inuse; ++i)
3991 sc->intr_mask |= sc->tx_rings[i].tx_intr_mask;
3993 if_printf(&sc->arpcom.ac_if, "intr mask 0x%08x\n",
3999 igb_alloc_intr(struct igb_softc *sc)
4001 int i, intr_bit, intr_bitmax;
4004 igb_msix_try_alloc(sc);
4005 if (sc->intr_type == PCI_INTR_TYPE_MSIX)
4009 * Allocate MSI/legacy interrupt resource
4011 sc->intr_type = pci_alloc_1intr(sc->dev, igb_msi_enable,
4012 &sc->intr_rid, &intr_flags);
4014 if (sc->intr_type == PCI_INTR_TYPE_LEGACY) {
4017 unshared = device_getenv_int(sc->dev, "irq.unshared", 0);
4019 sc->flags |= IGB_FLAG_SHARED_INTR;
4021 device_printf(sc->dev, "IRQ shared\n");
4023 intr_flags &= ~RF_SHAREABLE;
4025 device_printf(sc->dev, "IRQ unshared\n");
4029 sc->intr_res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ,
4030 &sc->intr_rid, intr_flags);
4031 if (sc->intr_res == NULL) {
4032 device_printf(sc->dev, "Unable to allocate bus resource: "
4038 * Setup MSI/legacy interrupt mask
4040 switch (sc->hw.mac.type) {
4042 intr_bitmax = IGB_MAX_TXRXINT_82575;
4045 intr_bitmax = IGB_MAX_TXRXINT_82580;
4048 intr_bitmax = IGB_MAX_TXRXINT_I350;
4051 intr_bitmax = IGB_MAX_TXRXINT_82576;
4054 intr_bitmax = IGB_MIN_TXRXINT;
4058 for (i = 0; i < sc->tx_ring_cnt; ++i)
4059 igb_set_txintr_mask(&sc->tx_rings[i], &intr_bit, intr_bitmax);
4060 for (i = 0; i < sc->rx_ring_cnt; ++i)
4061 igb_set_rxintr_mask(&sc->rx_rings[i], &intr_bit, intr_bitmax);
4062 sc->sts_intr_bit = 0;
4063 sc->sts_intr_mask = E1000_EICR_OTHER;
4065 /* Initialize interrupt rate */
4066 sc->intr_rate = IGB_INTR_RATE;
4068 igb_set_ring_inuse(sc, FALSE);
4069 igb_set_intr_mask(sc);
4074 igb_free_intr(struct igb_softc *sc)
4076 if (sc->intr_type != PCI_INTR_TYPE_MSIX) {
4077 if (sc->intr_res != NULL) {
4078 bus_release_resource(sc->dev, SYS_RES_IRQ, sc->intr_rid,
4081 if (sc->intr_type == PCI_INTR_TYPE_MSI)
4082 pci_release_msi(sc->dev);
4084 igb_msix_free(sc, TRUE);
4089 igb_teardown_intr(struct igb_softc *sc)
4091 if (sc->intr_type != PCI_INTR_TYPE_MSIX)
4092 bus_teardown_intr(sc->dev, sc->intr_res, sc->intr_tag);
4094 igb_msix_teardown(sc, sc->msix_cnt);
4098 igb_msix_try_alloc(struct igb_softc *sc)
4100 int msix_enable, msix_cnt, msix_cnt2, alloc_cnt;
4102 struct igb_msix_data *msix;
4103 boolean_t aggregate, setup = FALSE;
4106 * Don't enable MSI-X on 82575, see:
4107 * 82575 specification update errata #25
4109 if (sc->hw.mac.type == e1000_82575)
4112 /* Don't enable MSI-X on VF */
4116 msix_enable = device_getenv_int(sc->dev, "msix.enable",
4121 msix_cnt = pci_msix_count(sc->dev);
4122 #ifdef IGB_MSIX_DEBUG
4123 msix_cnt = device_getenv_int(sc->dev, "msix.count", msix_cnt);
4125 if (msix_cnt <= 1) {
4126 /* One MSI-X model does not make sense */
4131 while ((1 << (i + 1)) <= msix_cnt)
4136 device_printf(sc->dev, "MSI-X count %d/%d\n",
4137 msix_cnt2, msix_cnt);
4140 KKASSERT(msix_cnt2 <= msix_cnt);
4141 if (msix_cnt == msix_cnt2) {
4142 /* We need at least one MSI-X for link status */
4144 if (msix_cnt2 <= 1) {
4145 /* One MSI-X for RX/TX does not make sense */
4146 device_printf(sc->dev, "not enough MSI-X for TX/RX, "
4147 "MSI-X count %d/%d\n", msix_cnt2, msix_cnt);
4150 KKASSERT(msix_cnt > msix_cnt2);
4153 device_printf(sc->dev, "MSI-X count fixup %d/%d\n",
4154 msix_cnt2, msix_cnt);
4158 sc->rx_ring_msix = sc->rx_ring_cnt;
4159 if (sc->rx_ring_msix > msix_cnt2)
4160 sc->rx_ring_msix = msix_cnt2;
4162 sc->tx_ring_msix = sc->tx_ring_cnt;
4163 if (sc->tx_ring_msix > msix_cnt2)
4164 sc->tx_ring_msix = msix_cnt2;
4166 if (msix_cnt >= sc->tx_ring_msix + sc->rx_ring_msix + 1) {
4168 * Independent TX/RX MSI-X
4172 device_printf(sc->dev, "independent TX/RX MSI-X\n");
4173 alloc_cnt = sc->tx_ring_msix + sc->rx_ring_msix;
4176 * Aggregate TX/RX MSI-X
4180 device_printf(sc->dev, "aggregate TX/RX MSI-X\n");
4181 alloc_cnt = msix_cnt2;
4182 if (alloc_cnt > ncpus2)
4184 if (sc->rx_ring_msix > alloc_cnt)
4185 sc->rx_ring_msix = alloc_cnt;
4187 ++alloc_cnt; /* For link status */
4190 device_printf(sc->dev, "MSI-X alloc %d, RX ring %d\n",
4191 alloc_cnt, sc->rx_ring_msix);
4194 sc->msix_mem_rid = PCIR_BAR(IGB_MSIX_BAR);
4195 sc->msix_mem_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
4196 &sc->msix_mem_rid, RF_ACTIVE);
4197 if (sc->msix_mem_res == NULL) {
4198 device_printf(sc->dev, "Unable to map MSI-X table\n");
4202 sc->msix_cnt = alloc_cnt;
4203 sc->msix_data = kmalloc_cachealign(
4204 sizeof(struct igb_msix_data) * sc->msix_cnt,
4205 M_DEVBUF, M_WAITOK | M_ZERO);
4206 for (x = 0; x < sc->msix_cnt; ++x) {
4207 msix = &sc->msix_data[x];
4209 lwkt_serialize_init(&msix->msix_serialize0);
4211 msix->msix_rid = -1;
4212 msix->msix_vector = x;
4213 msix->msix_mask = 1 << msix->msix_vector;
4214 msix->msix_rate = IGB_INTR_RATE;
4219 int offset, offset_def;
4221 if (sc->rx_ring_msix == ncpus2) {
4224 offset_def = (sc->rx_ring_msix *
4225 device_get_unit(sc->dev)) % ncpus2;
4227 offset = device_getenv_int(sc->dev,
4228 "msix.rxoff", offset_def);
4229 if (offset >= ncpus2 ||
4230 offset % sc->rx_ring_msix != 0) {
4231 device_printf(sc->dev,
4232 "invalid msix.rxoff %d, use %d\n",
4233 offset, offset_def);
4234 offset = offset_def;
4239 for (i = 0; i < sc->rx_ring_msix; ++i) {
4240 struct igb_rx_ring *rxr = &sc->rx_rings[i];
4242 KKASSERT(x < sc->msix_cnt);
4243 msix = &sc->msix_data[x++];
4244 rxr->rx_intr_bit = msix->msix_vector;
4245 rxr->rx_intr_mask = msix->msix_mask;
4247 msix->msix_serialize = &rxr->rx_serialize;
4248 msix->msix_func = igb_msix_rx;
4249 msix->msix_arg = rxr;
4250 msix->msix_cpuid = i + offset;
4251 KKASSERT(msix->msix_cpuid < ncpus2);
4252 ksnprintf(msix->msix_desc, sizeof(msix->msix_desc),
4253 "%s rx%d", device_get_nameunit(sc->dev), i);
4254 msix->msix_rate = IGB_MSIX_RX_RATE;
4255 ksnprintf(msix->msix_rate_desc,
4256 sizeof(msix->msix_rate_desc),
4257 "RX%d interrupt rate", i);
4260 offset_def = device_get_unit(sc->dev) % ncpus2;
4261 offset = device_getenv_int(sc->dev, "msix.txoff", offset_def);
4262 if (offset >= ncpus2) {
4263 device_printf(sc->dev, "invalid msix.txoff %d, "
4264 "use %d\n", offset, offset_def);
4265 offset = offset_def;
4269 for (i = 0; i < sc->tx_ring_msix; ++i) {
4270 struct igb_tx_ring *txr = &sc->tx_rings[i];
4272 KKASSERT(x < sc->msix_cnt);
4273 msix = &sc->msix_data[x++];
4274 txr->tx_intr_bit = msix->msix_vector;
4275 txr->tx_intr_mask = msix->msix_mask;
4277 msix->msix_serialize = &txr->tx_serialize;
4278 msix->msix_func = igb_msix_tx;
4279 msix->msix_arg = txr;
4280 msix->msix_cpuid = i + offset;
4281 txr->tx_intr_cpuid = msix->msix_cpuid;
4282 KKASSERT(msix->msix_cpuid < ncpus2);
4283 ksnprintf(msix->msix_desc, sizeof(msix->msix_desc),
4284 "%s tx%d", device_get_nameunit(sc->dev), i);
4285 msix->msix_rate = IGB_MSIX_TX_RATE;
4286 ksnprintf(msix->msix_rate_desc,
4287 sizeof(msix->msix_rate_desc),
4288 "TX%d interrupt rate", i);
4299 KKASSERT(x < sc->msix_cnt);
4300 msix = &sc->msix_data[x++];
4301 sc->sts_intr_bit = msix->msix_vector;
4302 sc->sts_intr_mask = msix->msix_mask;
4304 msix->msix_serialize = &sc->main_serialize;
4305 msix->msix_func = igb_msix_status;
4306 msix->msix_arg = sc;
4307 msix->msix_cpuid = 0; /* TODO tunable */
4308 ksnprintf(msix->msix_desc, sizeof(msix->msix_desc), "%s sts",
4309 device_get_nameunit(sc->dev));
4310 ksnprintf(msix->msix_rate_desc, sizeof(msix->msix_rate_desc),
4311 "status interrupt rate");
4313 KKASSERT(x == sc->msix_cnt);
4315 error = pci_setup_msix(sc->dev);
4317 device_printf(sc->dev, "Setup MSI-X failed\n");
4322 for (i = 0; i < sc->msix_cnt; ++i) {
4323 msix = &sc->msix_data[i];
4325 error = pci_alloc_msix_vector(sc->dev, msix->msix_vector,
4326 &msix->msix_rid, msix->msix_cpuid);
4328 device_printf(sc->dev,
4329 "Unable to allocate MSI-X %d on cpu%d\n",
4330 msix->msix_vector, msix->msix_cpuid);
4334 msix->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ,
4335 &msix->msix_rid, RF_ACTIVE);
4336 if (msix->msix_res == NULL) {
4337 device_printf(sc->dev,
4338 "Unable to allocate MSI-X %d resource\n",
4345 pci_enable_msix(sc->dev);
4346 sc->intr_type = PCI_INTR_TYPE_MSIX;
4349 igb_msix_free(sc, setup);
4353 igb_msix_free(struct igb_softc *sc, boolean_t setup)
4357 KKASSERT(sc->msix_cnt > 1);
4359 for (i = 0; i < sc->msix_cnt; ++i) {
4360 struct igb_msix_data *msix = &sc->msix_data[i];
4362 if (msix->msix_res != NULL) {
4363 bus_release_resource(sc->dev, SYS_RES_IRQ,
4364 msix->msix_rid, msix->msix_res);
4366 if (msix->msix_rid >= 0)
4367 pci_release_msix_vector(sc->dev, msix->msix_rid);
4370 pci_teardown_msix(sc->dev);
4373 kfree(sc->msix_data, M_DEVBUF);
4374 sc->msix_data = NULL;
4378 igb_msix_setup(struct igb_softc *sc)
4382 for (i = 0; i < sc->msix_cnt; ++i) {
4383 struct igb_msix_data *msix = &sc->msix_data[i];
4386 error = bus_setup_intr_descr(sc->dev, msix->msix_res,
4387 INTR_MPSAFE, msix->msix_func, msix->msix_arg,
4388 &msix->msix_handle, msix->msix_serialize, msix->msix_desc);
4390 device_printf(sc->dev, "could not set up %s "
4391 "interrupt handler.\n", msix->msix_desc);
4392 igb_msix_teardown(sc, i);
4400 igb_msix_teardown(struct igb_softc *sc, int msix_cnt)
4404 for (i = 0; i < msix_cnt; ++i) {
4405 struct igb_msix_data *msix = &sc->msix_data[i];
4407 bus_teardown_intr(sc->dev, msix->msix_res, msix->msix_handle);
4412 igb_msix_rx(void *arg)
4414 struct igb_rx_ring *rxr = arg;
4416 ASSERT_SERIALIZED(&rxr->rx_serialize);
4419 E1000_WRITE_REG(&rxr->sc->hw, E1000_EIMS, rxr->rx_intr_mask);
4423 igb_msix_tx(void *arg)
4425 struct igb_tx_ring *txr = arg;
4427 ASSERT_SERIALIZED(&txr->tx_serialize);
4430 if (!ifsq_is_empty(txr->ifsq))
4431 ifsq_devstart(txr->ifsq);
4433 E1000_WRITE_REG(&txr->sc->hw, E1000_EIMS, txr->tx_intr_mask);
4437 igb_msix_status(void *arg)
4439 struct igb_softc *sc = arg;
4442 ASSERT_SERIALIZED(&sc->main_serialize);
4444 icr = E1000_READ_REG(&sc->hw, E1000_ICR);
4445 if (icr & E1000_ICR_LSC) {
4446 sc->hw.mac.get_link_status = 1;
4447 igb_update_link_status(sc);
4450 E1000_WRITE_REG(&sc->hw, E1000_EIMS, sc->sts_intr_mask);
4454 igb_set_ring_inuse(struct igb_softc *sc, boolean_t polling)
4456 sc->rx_ring_inuse = igb_get_rxring_inuse(sc, polling);
4457 sc->tx_ring_inuse = igb_get_txring_inuse(sc, polling);
4459 if_printf(&sc->arpcom.ac_if, "RX rings %d/%d, TX rings %d/%d\n",
4460 sc->rx_ring_inuse, sc->rx_ring_cnt,
4461 sc->tx_ring_inuse, sc->tx_ring_cnt);
4466 igb_get_rxring_inuse(const struct igb_softc *sc, boolean_t polling)
4468 if (!IGB_ENABLE_HWRSS(sc))
4472 return sc->rx_ring_cnt;
4473 else if (sc->intr_type != PCI_INTR_TYPE_MSIX)
4474 return IGB_MIN_RING_RSS;
4476 return sc->rx_ring_msix;
4480 igb_get_txring_inuse(const struct igb_softc *sc, boolean_t polling)
4482 if (!IGB_ENABLE_HWTSS(sc))
4486 return sc->tx_ring_cnt;
4487 else if (sc->intr_type != PCI_INTR_TYPE_MSIX)
4488 return IGB_MIN_RING;
4490 return sc->tx_ring_msix;
4494 igb_tso_pullup(struct igb_tx_ring *txr, struct mbuf **mp)
4496 int hoff, iphlen, thoff;
4500 KASSERT(M_WRITABLE(m), ("TSO mbuf not writable"));
4502 iphlen = m->m_pkthdr.csum_iphlen;
4503 thoff = m->m_pkthdr.csum_thlen;
4504 hoff = m->m_pkthdr.csum_lhlen;
4506 KASSERT(iphlen > 0, ("invalid ip hlen"));
4507 KASSERT(thoff > 0, ("invalid tcp hlen"));
4508 KASSERT(hoff > 0, ("invalid ether hlen"));
4510 if (__predict_false(m->m_len < hoff + iphlen + thoff)) {
4511 m = m_pullup(m, hoff + iphlen + thoff);
4518 if (txr->tx_flags & IGB_TXFLAG_TSO_IPLEN0) {
4521 ip = mtodoff(m, struct ip *, hoff);
4529 igb_tso_ctx(struct igb_tx_ring *txr, struct mbuf *m, uint32_t *hlen)
4531 struct e1000_adv_tx_context_desc *TXD;
4532 uint32_t vlan_macip_lens, type_tucmd_mlhl, mss_l4len_idx;
4533 int hoff, ctxd, iphlen, thoff;
4535 iphlen = m->m_pkthdr.csum_iphlen;
4536 thoff = m->m_pkthdr.csum_thlen;
4537 hoff = m->m_pkthdr.csum_lhlen;
4539 vlan_macip_lens = type_tucmd_mlhl = mss_l4len_idx = 0;
4541 ctxd = txr->next_avail_desc;
4542 TXD = (struct e1000_adv_tx_context_desc *)&txr->tx_base[ctxd];
4544 if (m->m_flags & M_VLANTAG) {
4547 vlantag = htole16(m->m_pkthdr.ether_vlantag);
4548 vlan_macip_lens |= (vlantag << E1000_ADVTXD_VLAN_SHIFT);
4551 vlan_macip_lens |= (hoff << E1000_ADVTXD_MACLEN_SHIFT);
4552 vlan_macip_lens |= iphlen;
4554 type_tucmd_mlhl |= E1000_ADVTXD_DCMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
4555 type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_L4T_TCP;
4556 type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_IPV4;
4558 mss_l4len_idx |= (m->m_pkthdr.tso_segsz << E1000_ADVTXD_MSS_SHIFT);
4559 mss_l4len_idx |= (thoff << E1000_ADVTXD_L4LEN_SHIFT);
4560 /* 82575 needs the queue index added */
4561 if (txr->sc->hw.mac.type == e1000_82575)
4562 mss_l4len_idx |= txr->me << 4;
4564 TXD->vlan_macip_lens = htole32(vlan_macip_lens);
4565 TXD->type_tucmd_mlhl = htole32(type_tucmd_mlhl);
4566 TXD->seqnum_seed = htole32(0);
4567 TXD->mss_l4len_idx = htole32(mss_l4len_idx);
4569 /* We've consumed the first desc, adjust counters */
4570 if (++ctxd == txr->num_tx_desc)
4572 txr->next_avail_desc = ctxd;
4575 *hlen = hoff + iphlen + thoff;