projects / dragonfly.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | history | HEAD
igb: Main serializer should be enough to protect link status
[dragonfly.git] / sys / dev / netif / igb / if_igb.c
CommitLineData
1f7e3916
SZ		 1/*
2 * Copyright (c) 2001-2011, Intel Corporation
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions are met:
7 *
8 * 1. Redistributions of source code must retain the above copyright notice,
9 * this list of conditions and the following disclaimer.
10 *
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.
14 *
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.
18 *
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.
30 */
31
32#include "opt_polling.h"
8d6600da 33#include "opt_igb.h"
1f7e3916
SZ		 34
35#include <sys/param.h>
36#include <sys/bus.h>
37#include <sys/endian.h>
38#include <sys/interrupt.h>
39#include <sys/kernel.h>
40#include <sys/malloc.h>
41#include <sys/mbuf.h>
42#include <sys/proc.h>
43#include <sys/rman.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>
50
51#include <net/bpf.h>
52#include <net/ethernet.h>
53#include <net/if.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>
63
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>
69
70#include <bus/pci/pcivar.h>
71#include <bus/pci/pcireg.h>
72
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>
76
8d6600da
SZ		 77#ifdef IGB_RSS_DEBUG
78#define IGB_RSS_DPRINTF(sc, lvl, fmt, ...) \
79do { \
80 if (sc->rss_debug >= lvl) \
81 if_printf(&sc->arpcom.ac_if, fmt, __VA_ARGS__); \
82} while (0)
83#else /* !IGB_RSS_DEBUG */
84#define IGB_RSS_DPRINTF(sc, lvl, fmt, ...) ((void)0)
85#endif /* IGB_RSS_DEBUG */
86
1f7e3916
SZ		 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 }
91
92static struct igb_device {
93 uint16_t vid;
94 uint16_t did;
95 const char *desc;
96} igb_devices[] = {
97 IGB_DEVICE(82575EB_COPPER),
98 IGB_DEVICE(82575EB_FIBER_SERDES),
99 IGB_DEVICE(82575GB_QUAD_COPPER),
100 IGB_DEVICE(82576),
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),
123 IGB_DEVICE(I350_VF),
124
125 /* required last entry */
126 IGB_DEVICE_NULL
127};
128
129static int igb_probe(device_t);
130static int igb_attach(device_t);
131static int igb_detach(device_t);
132static int igb_shutdown(device_t);
133static int igb_suspend(device_t);
134static int igb_resume(device_t);
135
136static boolean_t igb_is_valid_ether_addr(const uint8_t *);
137static void igb_setup_ifp(struct igb_softc *);
138static int igb_txctx_pullup(struct igb_tx_ring *, struct mbuf **);
139static boolean_t igb_txctx(struct igb_tx_ring *, struct mbuf *);
140static void igb_add_sysctl(struct igb_softc *);
141static int igb_sysctl_intr_rate(SYSCTL_HANDLER_ARGS);
b6220144 142static int igb_sysctl_tx_intr_nsegs(SYSCTL_HANDLER_ARGS);
1f7e3916
SZ		 143
144static void igb_vf_init_stats(struct igb_softc *);
145static void igb_reset(struct igb_softc *);
146static void igb_update_stats_counters(struct igb_softc *);
147static void igb_update_vf_stats_counters(struct igb_softc *);
148static void igb_update_link_status(struct igb_softc *);
149static void igb_init_tx_unit(struct igb_softc *);
150static void igb_init_rx_unit(struct igb_softc *);
151
152static void igb_set_vlan(struct igb_softc *);
153static void igb_set_multi(struct igb_softc *);
154static void igb_set_promisc(struct igb_softc *);
155static void igb_disable_promisc(struct igb_softc *);
156
a619b256
SZ		 157static int igb_alloc_rings(struct igb_softc *);
158static void igb_free_rings(struct igb_softc *);
1f7e3916
SZ		 159static int igb_create_tx_ring(struct igb_tx_ring *);
160static int igb_create_rx_ring(struct igb_rx_ring *);
161static void igb_free_tx_ring(struct igb_tx_ring *);
162static void igb_free_rx_ring(struct igb_rx_ring *);
163static void igb_destroy_tx_ring(struct igb_tx_ring *, int);
164static void igb_destroy_rx_ring(struct igb_rx_ring *, int);
165static void igb_init_tx_ring(struct igb_tx_ring *);
166static int igb_init_rx_ring(struct igb_rx_ring *);
167static int igb_newbuf(struct igb_rx_ring *, int, boolean_t);
168static int igb_encap(struct igb_tx_ring *, struct mbuf **);
169
170static void igb_stop(struct igb_softc *);
171static void igb_init(void *);
172static int igb_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
173static void igb_media_status(struct ifnet *, struct ifmediareq *);
174static int igb_media_change(struct ifnet *);
175static void igb_timer(void *);
176static void igb_watchdog(struct ifnet *);
177static void igb_start(struct ifnet *);
178#ifdef DEVICE_POLLING
179static void igb_poll(struct ifnet *, enum poll_cmd, int);
180#endif
7d235eb5
SZ		 181static void igb_serialize(struct ifnet *, enum ifnet_serialize);
182static void igb_deserialize(struct ifnet *, enum ifnet_serialize);
183static int igb_tryserialize(struct ifnet *, enum ifnet_serialize);
184#ifdef INVARIANTS
185static void igb_serialize_assert(struct ifnet *, enum ifnet_serialize,
186 boolean_t);
187#endif
1f7e3916
SZ		 188
189static void igb_intr(void *);
f6167a56 190static void igb_shared_intr(void *);
1f7e3916
SZ		 191static void igb_rxeof(struct igb_rx_ring *, int);
192static void igb_txeof(struct igb_tx_ring *);
f6167a56 193static void igb_set_eitr(struct igb_softc *);
1f7e3916
SZ		 194static void igb_enable_intr(struct igb_softc *);
195static void igb_disable_intr(struct igb_softc *);
f6167a56
SZ		 196static void igb_init_unshared_intr(struct igb_softc *);
197static void igb_init_intr(struct igb_softc *);
198static int igb_setup_intr(struct igb_softc *);
8d6600da
SZ		 199static void igb_setup_tx_intr(struct igb_tx_ring *, int *, int);
200static void igb_setup_rx_intr(struct igb_rx_ring *, int *, int);
1f7e3916
SZ		 201
202/* Management and WOL Support */
203static void igb_get_mgmt(struct igb_softc *);
204static void igb_rel_mgmt(struct igb_softc *);
205static void igb_get_hw_control(struct igb_softc *);
206static void igb_rel_hw_control(struct igb_softc *);
207static void igb_enable_wol(device_t);
208
209static device_method_t igb_methods[] = {
210 /* Device interface */
211 DEVMETHOD(device_probe, igb_probe),
212 DEVMETHOD(device_attach, igb_attach),
213 DEVMETHOD(device_detach, igb_detach),
214 DEVMETHOD(device_shutdown, igb_shutdown),
215 DEVMETHOD(device_suspend, igb_suspend),
216 DEVMETHOD(device_resume, igb_resume),
217 { 0, 0 }
218};
219
220static driver_t igb_driver = {
221 "igb",
222 igb_methods,
223 sizeof(struct igb_softc),
224};
225
226static devclass_t igb_devclass;
227
228DECLARE_DUMMY_MODULE(if_igb);
229MODULE_DEPEND(igb, ig_hal, 1, 1, 1);
230DRIVER_MODULE(if_igb, pci, igb_driver, igb_devclass, NULL, NULL);
231
232static int igb_rxd = IGB_DEFAULT_RXD;
233static int igb_txd = IGB_DEFAULT_TXD;
8d6600da 234static int igb_rxr = 0;
1f7e3916
SZ		 235static int igb_msi_enable = 1;
236static int igb_msix_enable = 1;
237static int igb_eee_disabled = 1; /* Energy Efficient Ethernet */
238static int igb_fc_setting = e1000_fc_full;
239
240/*
241 * DMA Coalescing, only for i350 - default to off,
242 * this feature is for power savings
243 */
244static int igb_dma_coalesce = 0;
245
246TUNABLE_INT("hw.igb.rxd", &igb_rxd);
247TUNABLE_INT("hw.igb.txd", &igb_txd);
8d6600da 248TUNABLE_INT("hw.igb.rxr", &igb_rxr);
1f7e3916
SZ		 249TUNABLE_INT("hw.igb.msi.enable", &igb_msi_enable);
250TUNABLE_INT("hw.igb.msix.enable", &igb_msix_enable);
251TUNABLE_INT("hw.igb.fc_setting", &igb_fc_setting);
252
253/* i350 specific */
254TUNABLE_INT("hw.igb.eee_disabled", &igb_eee_disabled);
255TUNABLE_INT("hw.igb.dma_coalesce", &igb_dma_coalesce);
256
257static __inline void
258igb_rxcsum(uint32_t staterr, struct mbuf *mp)
259{
260 /* Ignore Checksum bit is set */
261 if (staterr & E1000_RXD_STAT_IXSM)
262 return;
263
264 if ((staterr & (E1000_RXD_STAT_IPCS | E1000_RXDEXT_STATERR_IPE)) ==
265 E1000_RXD_STAT_IPCS)
266 mp->m_pkthdr.csum_flags |= CSUM_IP_CHECKED | CSUM_IP_VALID;
267
268 if (staterr & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)) {
269 if ((staterr & E1000_RXDEXT_STATERR_TCPE) == 0) {
270 mp->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
271 CSUM_PSEUDO_HDR | CSUM_FRAG_NOT_CHECKED;
272 mp->m_pkthdr.csum_data = htons(0xffff);
273 }
274 }
275}
276
8d6600da
SZ		 277static __inline struct pktinfo *
278igb_rssinfo(struct mbuf *m, struct pktinfo *pi,
279 uint32_t hash, uint32_t hashtype, uint32_t staterr)
280{
281 switch (hashtype) {
282 case E1000_RXDADV_RSSTYPE_IPV4_TCP:
283 pi->pi_netisr = NETISR_IP;
284 pi->pi_flags = 0;
285 pi->pi_l3proto = IPPROTO_TCP;
286 break;
287
288 case E1000_RXDADV_RSSTYPE_IPV4:
289 if (staterr & E1000_RXD_STAT_IXSM)
290 return NULL;
291
292 if ((staterr &
293 (E1000_RXD_STAT_TCPCS | E1000_RXDEXT_STATERR_TCPE)) ==
294 E1000_RXD_STAT_TCPCS) {
295 pi->pi_netisr = NETISR_IP;
296 pi->pi_flags = 0;
297 pi->pi_l3proto = IPPROTO_UDP;
298 break;
299 }
300 /* FALL THROUGH */
301 default:
302 return NULL;
303 }
304
305 m->m_flags |= M_HASH;
306 m->m_pkthdr.hash = toeplitz_hash(hash);
307 return pi;
308}
309
1f7e3916
SZ		 310static int
311igb_probe(device_t dev)
312{
313 const struct igb_device *d;
314 uint16_t vid, did;
315
316 vid = pci_get_vendor(dev);
317 did = pci_get_device(dev);
318
319 for (d = igb_devices; d->desc != NULL; ++d) {
320 if (vid == d->vid && did == d->did) {
321 device_set_desc(dev, d->desc);
322 return 0;
323 }
324 }
325 return ENXIO;
326}
327
328static int
329igb_attach(device_t dev)
330{
331 struct igb_softc *sc = device_get_softc(dev);
1f7e3916
SZ		 332 uint16_t eeprom_data;
333 u_int intr_flags;
8d6600da 334 int error = 0, i, j, ring_max;
1f7e3916
SZ		 335
336#ifdef notyet
337 /* SYSCTL stuff */
338 SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev),
339 SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
340 OID_AUTO, "nvm", CTLTYPE_INT|CTLFLAG_RW, adapter, 0,
341 igb_sysctl_nvm_info, "I", "NVM Information");
342
343 SYSCTL_ADD_INT(device_get_sysctl_ctx(dev),
344 SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
345 OID_AUTO, "enable_aim", CTLTYPE_INT|CTLFLAG_RW,
346 &igb_enable_aim, 1, "Interrupt Moderation");
347
348 SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev),
349 SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
350 OID_AUTO, "flow_control", CTLTYPE_INT|CTLFLAG_RW,
351 adapter, 0, igb_set_flowcntl, "I", "Flow Control");
352#endif
353
354 callout_init_mp(&sc->timer);
355
356 sc->dev = sc->osdep.dev = dev;
357
358 /*
359 * Determine hardware and mac type
360 */
361 sc->hw.vendor_id = pci_get_vendor(dev);
362 sc->hw.device_id = pci_get_device(dev);
363 sc->hw.revision_id = pci_read_config(dev, PCIR_REVID, 1);
364 sc->hw.subsystem_vendor_id = pci_read_config(dev, PCIR_SUBVEND_0, 2);
365 sc->hw.subsystem_device_id = pci_read_config(dev, PCIR_SUBDEV_0, 2);
366
367 if (e1000_set_mac_type(&sc->hw))
368 return ENXIO;
369
370 /* Are we a VF device? */
371 if (sc->hw.mac.type == e1000_vfadapt ||
372 sc->hw.mac.type == e1000_vfadapt_i350)
373 sc->vf_ifp = 1;
374 else
375 sc->vf_ifp = 0;
376
377 /* Enable bus mastering */
378 pci_enable_busmaster(dev);
379
380 /*
381 * Allocate IO memory
382 */
383 sc->mem_rid = PCIR_BAR(0);
384 sc->mem_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &sc->mem_rid,
385 RF_ACTIVE);
386 if (sc->mem_res == NULL) {
387 device_printf(dev, "Unable to allocate bus resource: memory\n");
388 error = ENXIO;
389 goto failed;
390 }
391 sc->osdep.mem_bus_space_tag = rman_get_bustag(sc->mem_res);
392 sc->osdep.mem_bus_space_handle = rman_get_bushandle(sc->mem_res);
393
394 sc->hw.hw_addr = (uint8_t *)&sc->osdep.mem_bus_space_handle;
395
396 /*
397 * Allocate interrupt
398 */
399 sc->intr_type = pci_alloc_1intr(dev, igb_msi_enable,
400 &sc->intr_rid, &intr_flags);
401
402 sc->intr_res = bus_alloc_resource_any(dev, SYS_RES_IRQ, &sc->intr_rid,
403 intr_flags);
404 if (sc->intr_res == NULL) {
405 device_printf(dev, "Unable to allocate bus resource: "
406 "interrupt\n");
407 error = ENXIO;
408 goto failed;
409 }
410
411 /* Save PCI command register for Shared Code */
412 sc->hw.bus.pci_cmd_word = pci_read_config(dev, PCIR_COMMAND, 2);
413 sc->hw.back = &sc->osdep;
414
8d6600da
SZ		 415 switch (sc->hw.mac.type) {
416 case e1000_82575:
417 ring_max = IGB_MAX_RING_82575;
418 break;
419 case e1000_82580:
420 ring_max = IGB_MAX_RING_82580;
421 break;
422 case e1000_i350:
423 ring_max = IGB_MAX_RING_I350;
424 break;
425 case e1000_82576:
426 ring_max = IGB_MAX_RING_82576;
427 break;
428 default:
429 ring_max = IGB_MIN_RING;
430 break;
431 }
432 sc->rx_ring_cnt = device_getenv_int(dev, "rxr", igb_rxr);
433 sc->rx_ring_cnt = if_ring_count2(sc->rx_ring_cnt, ring_max);
27866bf1 434 sc->tx_ring_cnt = 1; /* XXX */
8d6600da 435
1f7e3916
SZ		 436 sc->intr_rate = IGB_INTR_RATE;
437
438 /* Do Shared Code initialization */
439 if (e1000_setup_init_funcs(&sc->hw, TRUE)) {
440 device_printf(dev, "Setup of Shared code failed\n");
441 error = ENXIO;
442 goto failed;
443 }
444
445 e1000_get_bus_info(&sc->hw);
446
447 sc->hw.mac.autoneg = DO_AUTO_NEG;
448 sc->hw.phy.autoneg_wait_to_complete = FALSE;
449 sc->hw.phy.autoneg_advertised = AUTONEG_ADV_DEFAULT;
450
451 /* Copper options */
452 if (sc->hw.phy.media_type == e1000_media_type_copper) {
453 sc->hw.phy.mdix = AUTO_ALL_MODES;
454 sc->hw.phy.disable_polarity_correction = FALSE;
455 sc->hw.phy.ms_type = IGB_MASTER_SLAVE;
456 }
457
458 /* Set the frame limits assuming standard ethernet sized frames. */
459 sc->max_frame_size = ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN;
1f7e3916 460
a619b256
SZ		 461 /* Allocate RX/TX rings */
462 error = igb_alloc_rings(sc);
1f7e3916
SZ		 463 if (error)
464 goto failed;
465
7d235eb5
SZ		 466 /*
467 * Setup serializers
468 */
469 lwkt_serialize_init(&sc->main_serialize);
470 i = 0;
471 sc->serializes[i++] = &sc->main_serialize;
472
473 sc->tx_serialize = i;
474 for (j = 0; j < sc->tx_ring_cnt; ++j)
475 sc->serializes[i++] = &sc->tx_rings[j].tx_serialize;
476
477 sc->rx_serialize = i;
478 for (j = 0; j < sc->rx_ring_cnt; ++j)
479 sc->serializes[i++] = &sc->rx_rings[j].rx_serialize;
480
481 sc->serialize_cnt = i;
482 KKASSERT(sc->serialize_cnt <= IGB_NSERIALIZE);
483
1f7e3916
SZ		 484 /* Allocate the appropriate stats memory */
485 if (sc->vf_ifp) {
486 sc->stats = kmalloc(sizeof(struct e1000_vf_stats), M_DEVBUF,
487 M_WAITOK | M_ZERO);
488 igb_vf_init_stats(sc);
489 } else {
490 sc->stats = kmalloc(sizeof(struct e1000_hw_stats), M_DEVBUF,
491 M_WAITOK | M_ZERO);
492 }
493
494 /* Allocate multicast array memory. */
495 sc->mta = kmalloc(ETHER_ADDR_LEN * MAX_NUM_MULTICAST_ADDRESSES,
496 M_DEVBUF, M_WAITOK);
497
498 /* Some adapter-specific advanced features */
499 if (sc->hw.mac.type >= e1000_i350) {
500#ifdef notyet
501 igb_set_sysctl_value(adapter, "dma_coalesce",
502 "configure dma coalesce",
503 &adapter->dma_coalesce, igb_dma_coalesce);
504 igb_set_sysctl_value(adapter, "eee_disabled",
505 "enable Energy Efficient Ethernet",
506 &adapter->hw.dev_spec._82575.eee_disable,
507 igb_eee_disabled);
508#else
509 sc->dma_coalesce = igb_dma_coalesce;
510 sc->hw.dev_spec._82575.eee_disable = igb_eee_disabled;
511#endif
512 e1000_set_eee_i350(&sc->hw);
513 }
514
515 /*
516 * Start from a known state, this is important in reading the nvm and
517 * mac from that.
518 */
519 e1000_reset_hw(&sc->hw);
520
521 /* Make sure we have a good EEPROM before we read from it */
522 if (e1000_validate_nvm_checksum(&sc->hw) < 0) {
523 /*
524 * Some PCI-E parts fail the first check due to
525 * the link being in sleep state, call it again,
526 * if it fails a second time its a real issue.
527 */
528 if (e1000_validate_nvm_checksum(&sc->hw) < 0) {
529 device_printf(dev,
530 "The EEPROM Checksum Is Not Valid\n");
531 error = EIO;
532 goto failed;
533 }
534 }
535
536 /* Copy the permanent MAC address out of the EEPROM */
537 if (e1000_read_mac_addr(&sc->hw) < 0) {
538 device_printf(dev, "EEPROM read error while reading MAC"
539 " address\n");
540 error = EIO;
541 goto failed;
542 }
543 if (!igb_is_valid_ether_addr(sc->hw.mac.addr)) {
544 device_printf(dev, "Invalid MAC address\n");
545 error = EIO;
546 goto failed;
547 }
548
549#ifdef notyet
550 /*
551 ** Configure Interrupts
552 */
553 if ((adapter->msix > 1) && (igb_enable_msix))
554 error = igb_allocate_msix(adapter);
555 else /* MSI or Legacy */
556 error = igb_allocate_legacy(adapter);
557 if (error)
558 goto err_late;
559#endif
560
561 /* Setup OS specific network interface */
562 igb_setup_ifp(sc);
563
564 /* Add sysctl tree, must after igb_setup_ifp() */
565 igb_add_sysctl(sc);
566
567 /* Now get a good starting state */
568 igb_reset(sc);
569
570 /* Initialize statistics */
571 igb_update_stats_counters(sc);
572
573 sc->hw.mac.get_link_status = 1;
574 igb_update_link_status(sc);
575
576 /* Indicate SOL/IDER usage */
577 if (e1000_check_reset_block(&sc->hw)) {
578 device_printf(dev,
579 "PHY reset is blocked due to SOL/IDER session.\n");
580 }
581
582 /* Determine if we have to control management hardware */
396b7048
SZ		 583 if (e1000_enable_mng_pass_thru(&sc->hw))
584 sc->flags |= IGB_FLAG_HAS_MGMT;
1f7e3916
SZ		 585
586 /*
587 * Setup Wake-on-Lan
588 */
589 /* APME bit in EEPROM is mapped to WUC.APME */
590 eeprom_data = E1000_READ_REG(&sc->hw, E1000_WUC) & E1000_WUC_APME;
591 if (eeprom_data)
592 sc->wol = E1000_WUFC_MAG;
593 /* XXX disable WOL */
594 sc->wol = 0;
595
596#ifdef notyet
597 /* Register for VLAN events */
598 adapter->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
599 igb_register_vlan, adapter, EVENTHANDLER_PRI_FIRST);
600 adapter->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
601 igb_unregister_vlan, adapter, EVENTHANDLER_PRI_FIRST);
602#endif
603
604#ifdef notyet
605 igb_add_hw_stats(adapter);
606#endif
607
f6167a56 608 error = igb_setup_intr(sc);
1f7e3916 609 if (error) {
1f7e3916
SZ		 610 ether_ifdetach(&sc->arpcom.ac_if);
611 goto failed;
612 }
1f7e3916
SZ		 613 return 0;
614
615failed:
616 igb_detach(dev);
617 return error;
618}
619
620static int
621igb_detach(device_t dev)
622{
623 struct igb_softc *sc = device_get_softc(dev);
624
625 if (device_is_attached(dev)) {
626 struct ifnet *ifp = &sc->arpcom.ac_if;
627
628 ifnet_serialize_all(ifp);
629
630 igb_stop(sc);
631
632 e1000_phy_hw_reset(&sc->hw);
633
634 /* Give control back to firmware */
635 igb_rel_mgmt(sc);
636 igb_rel_hw_control(sc);
637
638 if (sc->wol) {
639 E1000_WRITE_REG(&sc->hw, E1000_WUC, E1000_WUC_PME_EN);
640 E1000_WRITE_REG(&sc->hw, E1000_WUFC, sc->wol);
641 igb_enable_wol(dev);
642 }
643
644 bus_teardown_intr(dev, sc->intr_res, sc->intr_tag);
645
646 ifnet_deserialize_all(ifp);
647
648 ether_ifdetach(ifp);
649 } else if (sc->mem_res != NULL) {
650 igb_rel_hw_control(sc);
651 }
652 bus_generic_detach(dev);
653
654 if (sc->intr_res != NULL) {
655 bus_release_resource(dev, SYS_RES_IRQ, sc->intr_rid,
656 sc->intr_res);
657 }
658 if (sc->intr_type == PCI_INTR_TYPE_MSI)
659 pci_release_msi(dev);
660
661 if (sc->mem_res != NULL) {
662 bus_release_resource(dev, SYS_RES_MEMORY, sc->mem_rid,
663 sc->mem_res);
664 }
665
a619b256 666 igb_free_rings(sc);
1f7e3916
SZ		 667
668 if (sc->mta != NULL)
669 kfree(sc->mta, M_DEVBUF);
670 if (sc->stats != NULL)
671 kfree(sc->stats, M_DEVBUF);
672
673 if (sc->sysctl_tree != NULL)
674 sysctl_ctx_free(&sc->sysctl_ctx);
675
676 return 0;
677}
678
679static int
680igb_shutdown(device_t dev)
681{
682 return igb_suspend(dev);
683}
684
685static int
686igb_suspend(device_t dev)
687{
688 struct igb_softc *sc = device_get_softc(dev);
689 struct ifnet *ifp = &sc->arpcom.ac_if;
690
691 ifnet_serialize_all(ifp);
692
693 igb_stop(sc);
694
695 igb_rel_mgmt(sc);
696 igb_rel_hw_control(sc);
697
698 if (sc->wol) {
699 E1000_WRITE_REG(&sc->hw, E1000_WUC, E1000_WUC_PME_EN);
700 E1000_WRITE_REG(&sc->hw, E1000_WUFC, sc->wol);
701 igb_enable_wol(dev);
702 }
703
704 ifnet_deserialize_all(ifp);
705
706 return bus_generic_suspend(dev);
707}
708
709static int
710igb_resume(device_t dev)
711{
712 struct igb_softc *sc = device_get_softc(dev);
713 struct ifnet *ifp = &sc->arpcom.ac_if;
714
715 ifnet_serialize_all(ifp);
716
717 igb_init(sc);
718 igb_get_mgmt(sc);
719
720 if_devstart(ifp);
721
722 ifnet_deserialize_all(ifp);
723
724 return bus_generic_resume(dev);
725}
726
727static int
728igb_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
729{
730 struct igb_softc *sc = ifp->if_softc;
731 struct ifreq *ifr = (struct ifreq *)data;
732 int max_frame_size, mask, reinit;
733 int error = 0;
734
735 ASSERT_IFNET_SERIALIZED_ALL(ifp);
736
737 switch (command) {
738 case SIOCSIFMTU:
739 max_frame_size = 9234;
740 if (ifr->ifr_mtu > max_frame_size - ETHER_HDR_LEN -
741 ETHER_CRC_LEN) {
742 error = EINVAL;
743 break;
744 }
745
746 ifp->if_mtu = ifr->ifr_mtu;
747 sc->max_frame_size = ifp->if_mtu + ETHER_HDR_LEN +
748 ETHER_CRC_LEN;
749
750 if (ifp->if_flags & IFF_RUNNING)
751 igb_init(sc);
752 break;
753
754 case SIOCSIFFLAGS:
755 if (ifp->if_flags & IFF_UP) {
756 if (ifp->if_flags & IFF_RUNNING) {
757 if ((ifp->if_flags ^ sc->if_flags) &
758 (IFF_PROMISC | IFF_ALLMULTI)) {
759 igb_disable_promisc(sc);
760 igb_set_promisc(sc);
761 }
762 } else {
763 igb_init(sc);
764 }
765 } else if (ifp->if_flags & IFF_RUNNING) {
766 igb_stop(sc);
767 }
768 sc->if_flags = ifp->if_flags;
769 break;
770
771 case SIOCADDMULTI:
772 case SIOCDELMULTI:
773 if (ifp->if_flags & IFF_RUNNING) {
774 igb_disable_intr(sc);
775 igb_set_multi(sc);
776#ifdef DEVICE_POLLING
777 if (!(ifp->if_flags & IFF_POLLING))
778#endif
779 igb_enable_intr(sc);
780 }
781 break;
782
783 case SIOCSIFMEDIA:
784 /*
785 * As the speed/duplex settings are being
786 * changed, we need toreset the PHY.
787 */
788 sc->hw.phy.reset_disable = FALSE;
789
790 /* Check SOL/IDER usage */
791 if (e1000_check_reset_block(&sc->hw)) {
792 if_printf(ifp, "Media change is "
793 "blocked due to SOL/IDER session.\n");
794 break;
795 }
796 /* FALL THROUGH */
797
798 case SIOCGIFMEDIA:
799 error = ifmedia_ioctl(ifp, ifr, &sc->media, command);
800 break;
801
802 case SIOCSIFCAP:
803 reinit = 0;
804 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
805 if (mask & IFCAP_HWCSUM) {
806 ifp->if_capenable ^= (mask & IFCAP_HWCSUM);
807 reinit = 1;
808 }
809 if (mask & IFCAP_VLAN_HWTAGGING) {
810 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
811 reinit = 1;
812 }
8d6600da
SZ		 813 if (mask & IFCAP_RSS)
814 ifp->if_capenable ^= IFCAP_RSS;
1f7e3916
SZ		 815 if (reinit && (ifp->if_flags & IFF_RUNNING))
816 igb_init(sc);
817 break;
818
819 default:
820 error = ether_ioctl(ifp, command, data);
821 break;
822 }
823 return error;
824}
825
826static void
827igb_init(void *xsc)
828{
829 struct igb_softc *sc = xsc;
830 struct ifnet *ifp = &sc->arpcom.ac_if;
831 int i;
832
833 ASSERT_IFNET_SERIALIZED_ALL(ifp);
834
835 igb_stop(sc);
836
837 /* Get the latest mac address, User can use a LAA */
838 bcopy(IF_LLADDR(ifp), sc->hw.mac.addr, ETHER_ADDR_LEN);
839
840 /* Put the address into the Receive Address Array */
841 e1000_rar_set(&sc->hw, sc->hw.mac.addr, 0);
842
843 igb_reset(sc);
844 igb_update_link_status(sc);
845
846 E1000_WRITE_REG(&sc->hw, E1000_VET, ETHERTYPE_VLAN);
847
848 /* Set hardware offload abilities */
849 if (ifp->if_capenable & IFCAP_TXCSUM)
850 ifp->if_hwassist = IGB_CSUM_FEATURES;
851 else
852 ifp->if_hwassist = 0;
853
854 /* Configure for OS presence */
855 igb_get_mgmt(sc);
856
857 /* Prepare transmit descriptors and buffers */
27866bf1 858 for (i = 0; i < sc->tx_ring_cnt; ++i)
1f7e3916
SZ		 859 igb_init_tx_ring(&sc->tx_rings[i]);
860 igb_init_tx_unit(sc);
861
862 /* Setup Multicast table */
863 igb_set_multi(sc);
864
865#if 0
866 /*
867 * Figure out the desired mbuf pool
868 * for doing jumbo/packetsplit
869 */
870 if (adapter->max_frame_size <= 2048)
871 adapter->rx_mbuf_sz = MCLBYTES;
872 else if (adapter->max_frame_size <= 4096)
873 adapter->rx_mbuf_sz = MJUMPAGESIZE;
874 else
875 adapter->rx_mbuf_sz = MJUM9BYTES;
1f7e3916
SZ		 876#endif
877
f6167a56
SZ		 878 /* Initialize interrupt */
879 igb_init_intr(sc);
880
1f7e3916 881 /* Prepare receive descriptors and buffers */
27866bf1 882 for (i = 0; i < sc->rx_ring_cnt; ++i) {
1f7e3916
SZ		 883 int error;
884
885 error = igb_init_rx_ring(&sc->rx_rings[i]);
886 if (error) {
887 if_printf(ifp, "Could not setup receive structures\n");
888 igb_stop(sc);
889 return;
890 }
891 }
892 igb_init_rx_unit(sc);
893
894 /* Enable VLAN support */
895 if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING)
896 igb_set_vlan(sc);
897
898 /* Don't lose promiscuous settings */
899 igb_set_promisc(sc);
900
1f7e3916
SZ		 901 ifp->if_flags |= IFF_RUNNING;
902 ifp->if_flags &= ~IFF_OACTIVE;
903
904 callout_reset(&sc->timer, hz, igb_timer, sc);
905 e1000_clear_hw_cntrs_base_generic(&sc->hw);
906
907#if 0
908 if (adapter->msix > 1) /* Set up queue routing */
909 igb_configure_queues(adapter);
910#endif
911
912 /* this clears any pending interrupts */
913 E1000_READ_REG(&sc->hw, E1000_ICR);
914#ifdef DEVICE_POLLING
915 /*
916 * Only enable interrupts if we are not polling, make sure
917 * they are off otherwise.
918 */
919 if (ifp->if_flags & IFF_POLLING)
920 igb_disable_intr(sc);
921 else
922#endif /* DEVICE_POLLING */
923 {
924 igb_enable_intr(sc);
925 E1000_WRITE_REG(&sc->hw, E1000_ICS, E1000_ICS_LSC);
926 }
927
928 /* Set Energy Efficient Ethernet */
929 e1000_set_eee_i350(&sc->hw);
930
931 /* Don't reset the phy next time init gets called */
932 sc->hw.phy.reset_disable = TRUE;
933}
934
935static void
936igb_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
937{
938 struct igb_softc *sc = ifp->if_softc;
939 u_char fiber_type = IFM_1000_SX;
940
941 ASSERT_IFNET_SERIALIZED_ALL(ifp);
942
943 igb_update_link_status(sc);
944
945 ifmr->ifm_status = IFM_AVALID;
946 ifmr->ifm_active = IFM_ETHER;
947
948 if (!sc->link_active)
949 return;
950
951 ifmr->ifm_status |= IFM_ACTIVE;
952
953 if (sc->hw.phy.media_type == e1000_media_type_fiber ||
954 sc->hw.phy.media_type == e1000_media_type_internal_serdes) {
955 ifmr->ifm_active |= fiber_type | IFM_FDX;
956 } else {
957 switch (sc->link_speed) {
958 case 10:
959 ifmr->ifm_active |= IFM_10_T;
960 break;
961
962 case 100:
963 ifmr->ifm_active |= IFM_100_TX;
964 break;
965
966 case 1000:
967 ifmr->ifm_active |= IFM_1000_T;
968 break;
969 }
970 if (sc->link_duplex == FULL_DUPLEX)
971 ifmr->ifm_active |= IFM_FDX;
972 else
973 ifmr->ifm_active |= IFM_HDX;
974 }
975}
976
977static int
978igb_media_change(struct ifnet *ifp)
979{
980 struct igb_softc *sc = ifp->if_softc;
981 struct ifmedia *ifm = &sc->media;
982
983 ASSERT_IFNET_SERIALIZED_ALL(ifp);
984
985 if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
986 return EINVAL;
987
988 switch (IFM_SUBTYPE(ifm->ifm_media)) {
989 case IFM_AUTO:
990 sc->hw.mac.autoneg = DO_AUTO_NEG;
991 sc->hw.phy.autoneg_advertised = AUTONEG_ADV_DEFAULT;
992 break;
993
994 case IFM_1000_LX:
995 case IFM_1000_SX:
996 case IFM_1000_T:
997 sc->hw.mac.autoneg = DO_AUTO_NEG;
998 sc->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
999 break;
1000
1001 case IFM_100_TX:
1002 sc->hw.mac.autoneg = FALSE;
1003 sc->hw.phy.autoneg_advertised = 0;
1004 if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
1005 sc->hw.mac.forced_speed_duplex = ADVERTISE_100_FULL;
1006 else
1007 sc->hw.mac.forced_speed_duplex = ADVERTISE_100_HALF;
1008 break;
1009
1010 case IFM_10_T:
1011 sc->hw.mac.autoneg = FALSE;
1012 sc->hw.phy.autoneg_advertised = 0;
1013 if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
1014 sc->hw.mac.forced_speed_duplex = ADVERTISE_10_FULL;
1015 else
1016 sc->hw.mac.forced_speed_duplex = ADVERTISE_10_HALF;
1017 break;
1018
1019 default:
1020 if_printf(ifp, "Unsupported media type\n");
1021 break;
1022 }
1023
1024 igb_init(sc);
1025
1026 return 0;
1027}
1028
1029static void
1030igb_set_promisc(struct igb_softc *sc)
1031{
1032 struct ifnet *ifp = &sc->arpcom.ac_if;
1033 struct e1000_hw *hw = &sc->hw;
1034 uint32_t reg;
1035
1036 if (sc->vf_ifp) {
1037 e1000_promisc_set_vf(hw, e1000_promisc_enabled);
1038 return;
1039 }
1040
1041 reg = E1000_READ_REG(hw, E1000_RCTL);
1042 if (ifp->if_flags & IFF_PROMISC) {
1043 reg |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
1044 E1000_WRITE_REG(hw, E1000_RCTL, reg);
1045 } else if (ifp->if_flags & IFF_ALLMULTI) {
1046 reg |= E1000_RCTL_MPE;
1047 reg &= ~E1000_RCTL_UPE;
1048 E1000_WRITE_REG(hw, E1000_RCTL, reg);
1049 }
1050}
1051
1052static void
1053igb_disable_promisc(struct igb_softc *sc)
1054{
1055 struct e1000_hw *hw = &sc->hw;
1056 uint32_t reg;
1057
1058 if (sc->vf_ifp) {
1059 e1000_promisc_set_vf(hw, e1000_promisc_disabled);
1060 return;
1061 }
1062 reg = E1000_READ_REG(hw, E1000_RCTL);
1063 reg &= ~E1000_RCTL_UPE;
1064 reg &= ~E1000_RCTL_MPE;
1065 E1000_WRITE_REG(hw, E1000_RCTL, reg);
1066}
1067
1068static void
1069igb_set_multi(struct igb_softc *sc)
1070{
1071 struct ifnet *ifp = &sc->arpcom.ac_if;
1072 struct ifmultiaddr *ifma;
1073 uint32_t reg_rctl = 0;
1074 uint8_t *mta;
1075 int mcnt = 0;
1076
1077 mta = sc->mta;
1078 bzero(mta, ETH_ADDR_LEN * MAX_NUM_MULTICAST_ADDRESSES);
1079
1080 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1081 if (ifma->ifma_addr->sa_family != AF_LINK)
1082 continue;
1083
1084 if (mcnt == MAX_NUM_MULTICAST_ADDRESSES)
1085 break;
1086
1087 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1088 &mta[mcnt * ETH_ADDR_LEN], ETH_ADDR_LEN);
1089 mcnt++;
1090 }
1091
1092 if (mcnt >= MAX_NUM_MULTICAST_ADDRESSES) {
1093 reg_rctl = E1000_READ_REG(&sc->hw, E1000_RCTL);
1094 reg_rctl |= E1000_RCTL_MPE;
1095 E1000_WRITE_REG(&sc->hw, E1000_RCTL, reg_rctl);
1096 } else {
1097 e1000_update_mc_addr_list(&sc->hw, mta, mcnt);
1098 }
1099}
1100
1101static void
1102igb_timer(void *xsc)
1103{
1104 struct igb_softc *sc = xsc;
1f7e3916 1105
27dd00d6 1106 lwkt_serialize_enter(&sc->main_serialize);
1f7e3916
SZ		 1107
1108 igb_update_link_status(sc);
1109 igb_update_stats_counters(sc);
1110
1111 callout_reset(&sc->timer, hz, igb_timer, sc);
1112
27dd00d6 1113 lwkt_serialize_exit(&sc->main_serialize);
1f7e3916
SZ		 1114}
1115
1116static void
1117igb_update_link_status(struct igb_softc *sc)
1118{
1119 struct ifnet *ifp = &sc->arpcom.ac_if;
1120 struct e1000_hw *hw = &sc->hw;
1121 uint32_t link_check, thstat, ctrl;
1122
1123 link_check = thstat = ctrl = 0;
1124
1125 /* Get the cached link value or read for real */
1126 switch (hw->phy.media_type) {
1127 case e1000_media_type_copper:
1128 if (hw->mac.get_link_status) {
1129 /* Do the work to read phy */
1130 e1000_check_for_link(hw);
1131 link_check = !hw->mac.get_link_status;
1132 } else {
1133 link_check = TRUE;
1134 }
1135 break;
1136
1137 case e1000_media_type_fiber:
1138 e1000_check_for_link(hw);
1139 link_check = E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU;
1140 break;
1141
1142 case e1000_media_type_internal_serdes:
1143 e1000_check_for_link(hw);
1144 link_check = hw->mac.serdes_has_link;
1145 break;
1146
1147 /* VF device is type_unknown */
1148 case e1000_media_type_unknown:
1149 e1000_check_for_link(hw);
1150 link_check = !hw->mac.get_link_status;
1151 /* Fall thru */
1152 default:
1153 break;
1154 }
1155
1156 /* Check for thermal downshift or shutdown */
1157 if (hw->mac.type == e1000_i350) {
1158 thstat = E1000_READ_REG(hw, E1000_THSTAT);
1159 ctrl = E1000_READ_REG(hw, E1000_CTRL_EXT);
1160 }
1161
1162 /* Now we check if a transition has happened */
1163 if (link_check && sc->link_active == 0) {
1164 e1000_get_speed_and_duplex(hw,
1165 &sc->link_speed, &sc->link_duplex);
1166 if (bootverbose) {
1167 if_printf(ifp, "Link is up %d Mbps %s\n",
1168 sc->link_speed,
1169 sc->link_duplex == FULL_DUPLEX ?
1170 "Full Duplex" : "Half Duplex");
1171 }
1172 sc->link_active = 1;
1173
1174 ifp->if_baudrate = sc->link_speed * 1000000;
1175 if ((ctrl & E1000_CTRL_EXT_LINK_MODE_GMII) &&
1176 (thstat & E1000_THSTAT_LINK_THROTTLE))
1177 if_printf(ifp, "Link: thermal downshift\n");
1178 /* This can sleep */
1179 ifp->if_link_state = LINK_STATE_UP;
1180 if_link_state_change(ifp);
1181 } else if (!link_check && sc->link_active == 1) {
1182 ifp->if_baudrate = sc->link_speed = 0;
1183 sc->link_duplex = 0;
1184 if (bootverbose)
1185 if_printf(ifp, "Link is Down\n");
1186 if ((ctrl & E1000_CTRL_EXT_LINK_MODE_GMII) &&
1187 (thstat & E1000_THSTAT_PWR_DOWN))
1188 if_printf(ifp, "Link: thermal shutdown\n");
1189 sc->link_active = 0;
1190 /* This can sleep */
1191 ifp->if_link_state = LINK_STATE_DOWN;
1192 if_link_state_change(ifp);
1193 }
1194}
1195
1196static void
1197igb_stop(struct igb_softc *sc)
1198{
1199 struct ifnet *ifp = &sc->arpcom.ac_if;
1200 int i;
1201
1202 ASSERT_IFNET_SERIALIZED_ALL(ifp);
1203
1204 igb_disable_intr(sc);
1205
1206 callout_stop(&sc->timer);
1207
1208 ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
1209 ifp->if_timer = 0;
1210
1211 e1000_reset_hw(&sc->hw);
1212 E1000_WRITE_REG(&sc->hw, E1000_WUC, 0);
1213
1214 e1000_led_off(&sc->hw);
1215 e1000_cleanup_led(&sc->hw);
1216
27866bf1 1217 for (i = 0; i < sc->tx_ring_cnt; ++i)
1f7e3916 1218 igb_free_tx_ring(&sc->tx_rings[i]);
27866bf1 1219 for (i = 0; i < sc->rx_ring_cnt; ++i)
1f7e3916
SZ		 1220 igb_free_rx_ring(&sc->rx_rings[i]);
1221}
1222
1223static void
1224igb_reset(struct igb_softc *sc)
1225{
1226 struct ifnet *ifp = &sc->arpcom.ac_if;
1227 struct e1000_hw *hw = &sc->hw;
1228 struct e1000_fc_info *fc = &hw->fc;
1229 uint32_t pba = 0;
1230 uint16_t hwm;
1231
1232 /* Let the firmware know the OS is in control */
1233 igb_get_hw_control(sc);
1234
1235 /*
1236 * Packet Buffer Allocation (PBA)
1237 * Writing PBA sets the receive portion of the buffer
1238 * the remainder is used for the transmit buffer.
1239 */
1240 switch (hw->mac.type) {
1241 case e1000_82575:
1242 pba = E1000_PBA_32K;
1243 break;
1244
1245 case e1000_82576:
1246 case e1000_vfadapt:
1247 pba = E1000_READ_REG(hw, E1000_RXPBS);
1248 pba &= E1000_RXPBS_SIZE_MASK_82576;
1249 break;
1250
1251 case e1000_82580:
1252 case e1000_i350:
1253 case e1000_vfadapt_i350:
1254 pba = E1000_READ_REG(hw, E1000_RXPBS);
1255 pba = e1000_rxpbs_adjust_82580(pba);
1256 break;
1257 /* XXX pba = E1000_PBA_35K; */
1258
1259 default:
1260 break;
1261 }
1262
1263 /* Special needs in case of Jumbo frames */
1264 if (hw->mac.type == e1000_82575 && ifp->if_mtu > ETHERMTU) {
1265 uint32_t tx_space, min_tx, min_rx;
1266
1267 pba = E1000_READ_REG(hw, E1000_PBA);
1268 tx_space = pba >> 16;
1269 pba &= 0xffff;
1270
1271 min_tx = (sc->max_frame_size +
1272 sizeof(struct e1000_tx_desc) - ETHER_CRC_LEN) * 2;
1273 min_tx = roundup2(min_tx, 1024);
1274 min_tx >>= 10;
1275 min_rx = sc->max_frame_size;
1276 min_rx = roundup2(min_rx, 1024);
1277 min_rx >>= 10;
1278 if (tx_space < min_tx && (min_tx - tx_space) < pba) {
1279 pba = pba - (min_tx - tx_space);
1280 /*
1281 * if short on rx space, rx wins
1282 * and must trump tx adjustment
1283 */
1284 if (pba < min_rx)
1285 pba = min_rx;
1286 }
1287 E1000_WRITE_REG(hw, E1000_PBA, pba);
1288 }
1289
1290 /*
1291 * These parameters control the automatic generation (Tx) and
1292 * response (Rx) to Ethernet PAUSE frames.
1293 * - High water mark should allow for at least two frames to be
1294 * received after sending an XOFF.
1295 * - Low water mark works best when it is very near the high water mark.
1296 * This allows the receiver to restart by sending XON when it has
1297 * drained a bit.
1298 */
1299 hwm = min(((pba << 10) * 9 / 10),
1300 ((pba << 10) - 2 * sc->max_frame_size));
1301
1302 if (hw->mac.type < e1000_82576) {
1303 fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
1304 fc->low_water = fc->high_water - 8;
1305 } else {
1306 fc->high_water = hwm & 0xFFF0; /* 16-byte granularity */
1307 fc->low_water = fc->high_water - 16;
1308 }
1309 fc->pause_time = IGB_FC_PAUSE_TIME;
1310 fc->send_xon = TRUE;
1311
1312 /* Issue a global reset */
1313 e1000_reset_hw(hw);
1314 E1000_WRITE_REG(hw, E1000_WUC, 0);
1315
1316 if (e1000_init_hw(hw) < 0)
1317 if_printf(ifp, "Hardware Initialization Failed\n");
1318
1319 /* Setup DMA Coalescing */
1320 if (hw->mac.type == e1000_i350 && sc->dma_coalesce) {
1321 uint32_t reg;
1322
1323 hwm = (pba - 4) << 10;
1324 reg = ((pba - 6) << E1000_DMACR_DMACTHR_SHIFT)
1325 & E1000_DMACR_DMACTHR_MASK;
1326
1327 /* transition to L0x or L1 if available..*/
1328 reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
1329
1330 /* timer = +-1000 usec in 32usec intervals */
1331 reg |= (1000 >> 5);
1332 E1000_WRITE_REG(hw, E1000_DMACR, reg);
1333
1334 /* No lower threshold */
1335 E1000_WRITE_REG(hw, E1000_DMCRTRH, 0);
1336
1337 /* set hwm to PBA - 2 * max frame size */
1338 E1000_WRITE_REG(hw, E1000_FCRTC, hwm);
1339
1340 /* Set the interval before transition */
1341 reg = E1000_READ_REG(hw, E1000_DMCTLX);
1342 reg |= 0x800000FF; /* 255 usec */
1343 E1000_WRITE_REG(hw, E1000_DMCTLX, reg);
1344
1345 /* free space in tx packet buffer to wake from DMA coal */
1346 E1000_WRITE_REG(hw, E1000_DMCTXTH,
1347 (20480 - (2 * sc->max_frame_size)) >> 6);
1348
1349 /* make low power state decision controlled by DMA coal */
1350 reg = E1000_READ_REG(hw, E1000_PCIEMISC);
1351 E1000_WRITE_REG(hw, E1000_PCIEMISC,
1352 reg | E1000_PCIEMISC_LX_DECISION);
1353 if_printf(ifp, "DMA Coalescing enabled\n");
1354 }
1355
1356 E1000_WRITE_REG(&sc->hw, E1000_VET, ETHERTYPE_VLAN);
1357 e1000_get_phy_info(hw);
1358 e1000_check_for_link(hw);
1359}
1360
1361static void
1362igb_setup_ifp(struct igb_softc *sc)
1363{
1364 struct ifnet *ifp = &sc->arpcom.ac_if;
1365
1366 if_initname(ifp, device_get_name(sc->dev), device_get_unit(sc->dev));
1367 ifp->if_softc = sc;
1368 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1369 ifp->if_init = igb_init;
1370 ifp->if_ioctl = igb_ioctl;
1371 ifp->if_start = igb_start;
7d235eb5
SZ		 1372 ifp->if_serialize = igb_serialize;
1373 ifp->if_deserialize = igb_deserialize;
1374 ifp->if_tryserialize = igb_tryserialize;
1375#ifdef INVARIANTS
1376 ifp->if_serialize_assert = igb_serialize_assert;
1377#endif
1f7e3916
SZ		 1378#ifdef DEVICE_POLLING
1379 ifp->if_poll = igb_poll;
1380#endif
1381 ifp->if_watchdog = igb_watchdog;
1382
91b8700a 1383 ifq_set_maxlen(&ifp->if_snd, sc->tx_rings[0].num_tx_desc - 1);
1f7e3916
SZ		 1384 ifq_set_ready(&ifp->if_snd);
1385
1386 ether_ifattach(ifp, sc->hw.mac.addr, NULL);
1387
1388 ifp->if_capabilities =
1389 IFCAP_HWCSUM | IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU;
8d6600da
SZ		 1390 if (IGB_ENABLE_HWRSS(sc))
1391 ifp->if_capabilities |= IFCAP_RSS;
1f7e3916
SZ		 1392 ifp->if_capenable = ifp->if_capabilities;
1393 ifp->if_hwassist = IGB_CSUM_FEATURES;
1394
1395 /*
1396 * Tell the upper layer(s) we support long frames
1397 */
1398 ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
1399
1400 /*
1401 * Specify the media types supported by this adapter and register
1402 * callbacks to update media and link information
1403 */
1404 ifmedia_init(&sc->media, IFM_IMASK, igb_media_change, igb_media_status);
1405 if (sc->hw.phy.media_type == e1000_media_type_fiber ||
1406 sc->hw.phy.media_type == e1000_media_type_internal_serdes) {
1407 ifmedia_add(&sc->media, IFM_ETHER | IFM_1000_SX | IFM_FDX,
1408 0, NULL);
1409 ifmedia_add(&sc->media, IFM_ETHER | IFM_1000_SX, 0, NULL);
1410 } else {
1411 ifmedia_add(&sc->media, IFM_ETHER | IFM_10_T, 0, NULL);
1412 ifmedia_add(&sc->media, IFM_ETHER | IFM_10_T | IFM_FDX,
1413 0, NULL);
1414 ifmedia_add(&sc->media, IFM_ETHER | IFM_100_TX, 0, NULL);
1415 ifmedia_add(&sc->media, IFM_ETHER | IFM_100_TX | IFM_FDX,
1416 0, NULL);
1417 if (sc->hw.phy.type != e1000_phy_ife) {
1418 ifmedia_add(&sc->media,
1419 IFM_ETHER | IFM_1000_T | IFM_FDX, 0, NULL);
1420 ifmedia_add(&sc->media,
1421 IFM_ETHER | IFM_1000_T, 0, NULL);
1422 }
1423 }
1424 ifmedia_add(&sc->media, IFM_ETHER | IFM_AUTO, 0, NULL);
1425 ifmedia_set(&sc->media, IFM_ETHER | IFM_AUTO);
1426}
1427
1428static void
1429igb_add_sysctl(struct igb_softc *sc)
1430{
8d6600da
SZ		 1431#ifdef IGB_RSS_DEBUG
1432 char rx_pkt[32];
1433 int i;
1434#endif
1435
1f7e3916
SZ		 1436 sysctl_ctx_init(&sc->sysctl_ctx);
1437 sc->sysctl_tree = SYSCTL_ADD_NODE(&sc->sysctl_ctx,
1438 SYSCTL_STATIC_CHILDREN(_hw), OID_AUTO,
1439 device_get_nameunit(sc->dev), CTLFLAG_RD, 0, "");
1440 if (sc->sysctl_tree == NULL) {
1441 device_printf(sc->dev, "can't add sysctl node\n");
1442 return;
1443 }
1444
1445 SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
8d6600da
SZ		 1446 OID_AUTO, "rxr", CTLFLAG_RD, &sc->rx_ring_cnt, 0, "# of RX rings");
1447 SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1448 OID_AUTO, "rxd", CTLFLAG_RD, &sc->rx_rings[0].num_rx_desc, 0,
1449 "# of RX descs");
1f7e3916 1450 SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
8d6600da
SZ		 1451 OID_AUTO, "txd", CTLFLAG_RD, &sc->tx_rings[0].num_tx_desc, 0,
1452 "# of TX descs");
1f7e3916
SZ		 1453
1454 SYSCTL_ADD_PROC(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1455 OID_AUTO, "intr_rate", CTLTYPE_INT | CTLFLAG_RW,
1456 sc, 0, igb_sysctl_intr_rate, "I", "interrupt rate");
b6220144
SZ		 1457
1458 SYSCTL_ADD_PROC(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1459 OID_AUTO, "tx_intr_nsegs", CTLTYPE_INT | CTLFLAG_RW,
1460 sc, 0, igb_sysctl_tx_intr_nsegs, "I",
8d6600da
SZ		 1461 "# of segments per TX interrupt");
1462
1463#ifdef IGB_RSS_DEBUG
1464 SYSCTL_ADD_INT(&sc->sysctl_ctx, SYSCTL_CHILDREN(sc->sysctl_tree),
1465 OID_AUTO, "rss_debug", CTLFLAG_RW, &sc->rss_debug, 0,
1466 "RSS debug level");
1467 for (i = 0; i < sc->rx_ring_cnt; ++i) {
1468 ksnprintf(rx_pkt, sizeof(rx_pkt), "rx%d_pkt", i);
1469 SYSCTL_ADD_ULONG(&sc->sysctl_ctx,
1470 SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO, rx_pkt,
1471 CTLFLAG_RW, &sc->rx_rings[i].rx_packets, "RXed packets");
1472 }
1473#endif
1f7e3916
SZ		 1474}
1475
1476static int
a619b256 1477igb_alloc_rings(struct igb_softc *sc)
1f7e3916
SZ		 1478{
1479 int error, i;
1480
1f7e3916
SZ		 1481 /*
1482 * Create top level busdma tag
1483 */
1484 error = bus_dma_tag_create(NULL, 1, 0,
1485 BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
1486 BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
1487 &sc->parent_tag);
1488 if (error) {
1489 device_printf(sc->dev, "could not create top level DMA tag\n");
1490 return error;
1491 }
1492
1493 /*
1494 * Allocate TX descriptor rings and buffers
1495 */
27866bf1 1496 sc->tx_rings = kmalloc(sizeof(struct igb_tx_ring) * sc->tx_ring_cnt,
1f7e3916 1497 M_DEVBUF, M_WAITOK | M_ZERO);
27866bf1 1498 for (i = 0; i < sc->tx_ring_cnt; ++i) {
1f7e3916
SZ		 1499 struct igb_tx_ring *txr = &sc->tx_rings[i];
1500
1501 /* Set up some basics */
1502 txr->sc = sc;
1503 txr->me = i;
7d235eb5 1504 lwkt_serialize_init(&txr->tx_serialize);
1f7e3916
SZ		 1505
1506 error = igb_create_tx_ring(txr);
1507 if (error)
1508 return error;
1509 }
1510
1511 /*
1512 * Allocate RX descriptor rings and buffers
1513 */
27866bf1 1514 sc->rx_rings = kmalloc(sizeof(struct igb_rx_ring) * sc->rx_ring_cnt,
1f7e3916 1515 M_DEVBUF, M_WAITOK | M_ZERO);
27866bf1 1516 for (i = 0; i < sc->rx_ring_cnt; ++i) {
1f7e3916
SZ		 1517 struct igb_rx_ring *rxr = &sc->rx_rings[i];
1518
1519 /* Set up some basics */
1520 rxr->sc = sc;
1521 rxr->me = i;
7d235eb5 1522 lwkt_serialize_init(&rxr->rx_serialize);
1f7e3916
SZ		 1523
1524 error = igb_create_rx_ring(rxr);
1525 if (error)
1526 return error;
1527 }
1528
1f7e3916
SZ		 1529 return 0;
1530}
1531
1532static void
a619b256 1533igb_free_rings(struct igb_softc *sc)
1f7e3916
SZ		 1534{
1535 int i;
1536
1f7e3916 1537 if (sc->tx_rings != NULL) {
91b8700a
SZ		 1538 for (i = 0; i < sc->tx_ring_cnt; ++i) {
1539 struct igb_tx_ring *txr = &sc->tx_rings[i];
1540
1541 igb_destroy_tx_ring(txr, txr->num_tx_desc);
1542 }
1f7e3916
SZ		 1543 kfree(sc->tx_rings, M_DEVBUF);
1544 }
1545
1546 if (sc->rx_rings != NULL) {
91b8700a
SZ		 1547 for (i = 0; i < sc->rx_ring_cnt; ++i) {
1548 struct igb_rx_ring *rxr = &sc->rx_rings[i];
1549
1550 igb_destroy_rx_ring(rxr, rxr->num_rx_desc);
1551 }
1f7e3916
SZ		 1552 kfree(sc->rx_rings, M_DEVBUF);
1553 }
1554}
1555
1556static int
1557igb_create_tx_ring(struct igb_tx_ring *txr)
1558{
1559 int tsize, error, i;
1560
1561 /*
1562 * Validate number of transmit descriptors. It must not exceed
1563 * hardware maximum, and must be multiple of IGB_DBA_ALIGN.
1564 */
1565 if (((igb_txd * sizeof(struct e1000_tx_desc)) % IGB_DBA_ALIGN) != 0 ||
1566 (igb_txd > IGB_MAX_TXD) || (igb_txd < IGB_MIN_TXD)) {
1567 device_printf(txr->sc->dev,
1568 "Using %d TX descriptors instead of %d!\n",
1569 IGB_DEFAULT_TXD, igb_txd);
91b8700a 1570 txr->num_tx_desc = IGB_DEFAULT_TXD;
1f7e3916 1571 } else {
91b8700a 1572 txr->num_tx_desc = igb_txd;
1f7e3916
SZ		 1573 }
1574
1575 /*
1576 * Allocate TX descriptor ring
1577 */
91b8700a 1578 tsize = roundup2(txr->num_tx_desc * sizeof(union e1000_adv_tx_desc),
1f7e3916
SZ		 1579 IGB_DBA_ALIGN);
1580 txr->txdma.dma_vaddr = bus_dmamem_coherent_any(txr->sc->parent_tag,
1581 IGB_DBA_ALIGN, tsize, BUS_DMA_WAITOK,
1582 &txr->txdma.dma_tag, &txr->txdma.dma_map, &txr->txdma.dma_paddr);
1583 if (txr->txdma.dma_vaddr == NULL) {
1584 device_printf(txr->sc->dev,
1585 "Unable to allocate TX Descriptor memory\n");
1586 return ENOMEM;
1587 }
1588 txr->tx_base = txr->txdma.dma_vaddr;
1589 bzero(txr->tx_base, tsize);
1590
91b8700a 1591 txr->tx_buf = kmalloc(sizeof(struct igb_tx_buf) * txr->num_tx_desc,
1f7e3916
SZ		 1592 M_DEVBUF, M_WAITOK | M_ZERO);
1593
1594 /*
b6220144
SZ		 1595 * Allocate TX head write-back buffer
1596 */
1597 txr->tx_hdr = bus_dmamem_coherent_any(txr->sc->parent_tag,
1598 __VM_CACHELINE_SIZE, __VM_CACHELINE_SIZE, BUS_DMA_WAITOK,
1599 &txr->tx_hdr_dtag, &txr->tx_hdr_dmap, &txr->tx_hdr_paddr);
1600 if (txr->tx_hdr == NULL) {
1601 device_printf(txr->sc->dev,
1602 "Unable to allocate TX head write-back buffer\n");
1603 return ENOMEM;
1604 }
1605
1606 /*
1f7e3916
SZ		 1607 * Create DMA tag for TX buffers
1608 */
1609 error = bus_dma_tag_create(txr->sc->parent_tag,
1610 1, 0, /* alignment, bounds */
1611 BUS_SPACE_MAXADDR, /* lowaddr */
1612 BUS_SPACE_MAXADDR, /* highaddr */
1613 NULL, NULL, /* filter, filterarg */
1614 IGB_TSO_SIZE, /* maxsize */
1615 IGB_MAX_SCATTER, /* nsegments */
1616 PAGE_SIZE, /* maxsegsize */
1617 BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW |
1618 BUS_DMA_ONEBPAGE, /* flags */
1619 &txr->tx_tag);
1620 if (error) {
1621 device_printf(txr->sc->dev, "Unable to allocate TX DMA tag\n");
1622 kfree(txr->tx_buf, M_DEVBUF);
1623 txr->tx_buf = NULL;
1624 return error;
1625 }
1626
1627 /*
1628 * Create DMA maps for TX buffers
1629 */
91b8700a 1630 for (i = 0; i < txr->num_tx_desc; ++i) {
1f7e3916
SZ		 1631 struct igb_tx_buf *txbuf = &txr->tx_buf[i];
1632
1633 error = bus_dmamap_create(txr->tx_tag,
1634 BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE, &txbuf->map);
1635 if (error) {
1636 device_printf(txr->sc->dev,
1637 "Unable to create TX DMA map\n");
1638 igb_destroy_tx_ring(txr, i);
1639 return error;
1640 }
1641 }
b6220144
SZ		 1642
1643 /*
1644 * Initialize various watermark
1645 */
1646 txr->spare_desc = IGB_TX_SPARE;
91b8700a
SZ		 1647 txr->intr_nsegs = txr->num_tx_desc / 16;
1648 txr->oact_hi_desc = txr->num_tx_desc / 2;
1649 txr->oact_lo_desc = txr->num_tx_desc / 8;
b6220144
SZ		 1650 if (txr->oact_lo_desc > IGB_TX_OACTIVE_MAX)
1651 txr->oact_lo_desc = IGB_TX_OACTIVE_MAX;
1652 if (txr->oact_lo_desc < txr->spare_desc + IGB_TX_RESERVED)
1653 txr->oact_lo_desc = txr->spare_desc + IGB_TX_RESERVED;
1654
1f7e3916
SZ		 1655 return 0;
1656}
1657
1658static void
1659igb_free_tx_ring(struct igb_tx_ring *txr)
1660{
1661 int i;
1662
91b8700a 1663 for (i = 0; i < txr->num_tx_desc; ++i) {
1f7e3916
SZ		 1664 struct igb_tx_buf *txbuf = &txr->tx_buf[i];
1665
1666 if (txbuf->m_head != NULL) {
1667 bus_dmamap_unload(txr->tx_tag, txbuf->map);
1668 m_freem(txbuf->m_head);
1669 txbuf->m_head = NULL;
1670 }
1671 }
1672}
1673
1674static void
1675igb_destroy_tx_ring(struct igb_tx_ring *txr, int ndesc)
1676{
1677 int i;
1678
1679 if (txr->txdma.dma_vaddr != NULL) {
1680 bus_dmamap_unload(txr->txdma.dma_tag, txr->txdma.dma_map);
1681 bus_dmamem_free(txr->txdma.dma_tag, txr->txdma.dma_vaddr,
1682 txr->txdma.dma_map);
1683 bus_dma_tag_destroy(txr->txdma.dma_tag);
1684 txr->txdma.dma_vaddr = NULL;
1685 }
1686
b6220144
SZ		 1687 if (txr->tx_hdr != NULL) {
1688 bus_dmamap_unload(txr->tx_hdr_dtag, txr->tx_hdr_dmap);
1689 bus_dmamem_free(txr->tx_hdr_dtag, txr->tx_hdr,
1690 txr->tx_hdr_dmap);
1691 bus_dma_tag_destroy(txr->tx_hdr_dtag);
1692 txr->tx_hdr = NULL;
1693 }
1694
1f7e3916
SZ		 1695 if (txr->tx_buf == NULL)
1696 return;
1697
1698 for (i = 0; i < ndesc; ++i) {
1699 struct igb_tx_buf *txbuf = &txr->tx_buf[i];
1700
1701 KKASSERT(txbuf->m_head == NULL);
1702 bus_dmamap_destroy(txr->tx_tag, txbuf->map);
1703 }
1704 bus_dma_tag_destroy(txr->tx_tag);
1705
1706 kfree(txr->tx_buf, M_DEVBUF);
1707 txr->tx_buf = NULL;
1708}
1709
1710static void
1711igb_init_tx_ring(struct igb_tx_ring *txr)
1712{
1f7e3916
SZ		 1713 /* Clear the old descriptor contents */
1714 bzero(txr->tx_base,
91b8700a 1715 sizeof(union e1000_adv_tx_desc) * txr->num_tx_desc);
1f7e3916 1716
b6220144
SZ		 1717 /* Clear TX head write-back buffer */
1718 *(txr->tx_hdr) = 0;
1719
1f7e3916
SZ		 1720 /* Reset indices */
1721 txr->next_avail_desc = 0;
1722 txr->next_to_clean = 0;
b6220144 1723 txr->tx_nsegs = 0;
1f7e3916
SZ		 1724
1725 /* Set number of descriptors available */
91b8700a 1726 txr->tx_avail = txr->num_tx_desc;
1f7e3916
SZ		 1727}
1728
1729static void
1730igb_init_tx_unit(struct igb_softc *sc)
1731{
1732 struct e1000_hw *hw = &sc->hw;
1733 uint32_t tctl;
1734 int i;
1735
1736 /* Setup the Tx Descriptor Rings */
27866bf1 1737 for (i = 0; i < sc->tx_ring_cnt; ++i) {
1f7e3916
SZ		 1738 struct igb_tx_ring *txr = &sc->tx_rings[i];
1739 uint64_t bus_addr = txr->txdma.dma_paddr;
c3162c4e 1740 uint64_t hdr_paddr = txr->tx_hdr_paddr;
1f7e3916 1741 uint32_t txdctl = 0;
b6220144 1742 uint32_t dca_txctrl;
1f7e3916
SZ		 1743
1744 E1000_WRITE_REG(hw, E1000_TDLEN(i),
91b8700a 1745 txr->num_tx_desc * sizeof(struct e1000_tx_desc));
1f7e3916
SZ		 1746 E1000_WRITE_REG(hw, E1000_TDBAH(i),
1747 (uint32_t)(bus_addr >> 32));
1748 E1000_WRITE_REG(hw, E1000_TDBAL(i),
1749 (uint32_t)bus_addr);
1750
1751 /* Setup the HW Tx Head and Tail descriptor pointers */
1752 E1000_WRITE_REG(hw, E1000_TDT(i), 0);
1753 E1000_WRITE_REG(hw, E1000_TDH(i), 0);
1754
1755 txdctl |= IGB_TX_PTHRESH;
1756 txdctl |= IGB_TX_HTHRESH << 8;
1757 txdctl |= IGB_TX_WTHRESH << 16;
1758 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1759 E1000_WRITE_REG(hw, E1000_TXDCTL(i), txdctl);
b6220144
SZ		 1760
1761 dca_txctrl = E1000_READ_REG(hw, E1000_DCA_TXCTRL(i));
1762 dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1763 E1000_WRITE_REG(hw, E1000_DCA_TXCTRL(i), dca_txctrl);
1764
1765 E1000_WRITE_REG(hw, E1000_TDWBAH(i),
c3162c4e 1766 (uint32_t)(hdr_paddr >> 32));
b6220144 1767 E1000_WRITE_REG(hw, E1000_TDWBAL(i),
c3162c4e 1768 ((uint32_t)hdr_paddr) | E1000_TX_HEAD_WB_ENABLE);
1f7e3916
SZ		 1769 }
1770
1771 if (sc->vf_ifp)
1772 return;
1773
1774 e1000_config_collision_dist(hw);
1775
1776 /* Program the Transmit Control Register */
1777 tctl = E1000_READ_REG(hw, E1000_TCTL);
1778 tctl &= ~E1000_TCTL_CT;
1779 tctl |= (E1000_TCTL_PSP | E1000_TCTL_RTLC | E1000_TCTL_EN |
1780 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT));
1781
1782 /* This write will effectively turn on the transmit unit. */
1783 E1000_WRITE_REG(hw, E1000_TCTL, tctl);
1784}
1785
1786static boolean_t
1787igb_txctx(struct igb_tx_ring *txr, struct mbuf *mp)
1788{
1789 struct e1000_adv_tx_context_desc *TXD;
1790 struct igb_tx_buf *txbuf;
1791 uint32_t vlan_macip_lens, type_tucmd_mlhl, mss_l4len_idx;
1792 struct ether_vlan_header *eh;
1793 struct ip *ip = NULL;
1794 int ehdrlen, ctxd, ip_hlen = 0;
1795 uint16_t etype, vlantag = 0;
1796 boolean_t offload = TRUE;
1797
1798 if ((mp->m_pkthdr.csum_flags & IGB_CSUM_FEATURES) == 0)
1799 offload = FALSE;
1800
1801 vlan_macip_lens = type_tucmd_mlhl = mss_l4len_idx = 0;
1802 ctxd = txr->next_avail_desc;
1803 txbuf = &txr->tx_buf[ctxd];
1804 TXD = (struct e1000_adv_tx_context_desc *)&txr->tx_base[ctxd];
1805
1806 /*
1807 * In advanced descriptors the vlan tag must
1808 * be placed into the context descriptor, thus
1809 * we need to be here just for that setup.
1810 */
1811 if (mp->m_flags & M_VLANTAG) {
1812 vlantag = htole16(mp->m_pkthdr.ether_vlantag);
1813 vlan_macip_lens |= (vlantag << E1000_ADVTXD_VLAN_SHIFT);
1814 } else if (!offload) {
1815 return FALSE;
1816 }
1817
1818 /*
1819 * Determine where frame payload starts.
1820 * Jump over vlan headers if already present,
1821 * helpful for QinQ too.
1822 */
1823 KASSERT(mp->m_len >= ETHER_HDR_LEN,
1824 ("igb_txctx_pullup is not called (eh)?\n"));
1825 eh = mtod(mp, struct ether_vlan_header *);
1826 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
1827 KASSERT(mp->m_len >= ETHER_HDR_LEN + EVL_ENCAPLEN,
1828 ("igb_txctx_pullup is not called (evh)?\n"));
1829 etype = ntohs(eh->evl_proto);
1830 ehdrlen = ETHER_HDR_LEN + EVL_ENCAPLEN;
1831 } else {
1832 etype = ntohs(eh->evl_encap_proto);
1833 ehdrlen = ETHER_HDR_LEN;
1834 }
1835
1836 /* Set the ether header length */
1837 vlan_macip_lens |= ehdrlen << E1000_ADVTXD_MACLEN_SHIFT;
1838
1839 switch (etype) {
1840 case ETHERTYPE_IP:
1841 KASSERT(mp->m_len >= ehdrlen + IGB_IPVHL_SIZE,
1842 ("igb_txctx_pullup is not called (eh+ip_vhl)?\n"));
1843
1844 /* NOTE: We could only safely access ip.ip_vhl part */
1845 ip = (struct ip *)(mp->m_data + ehdrlen);
1846 ip_hlen = ip->ip_hl << 2;
1847
1848 if (mp->m_pkthdr.csum_flags & CSUM_IP)
1849 type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_IPV4;
1850 break;
1851
1852#ifdef notyet
1853 case ETHERTYPE_IPV6:
1854 ip6 = (struct ip6_hdr *)(mp->m_data + ehdrlen);
1855 ip_hlen = sizeof(struct ip6_hdr);
1856 type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_IPV6;
1857 break;
1858#endif
1859
1860 default:
1861 offload = FALSE;
1862 break;
1863 }
1864
1865 vlan_macip_lens |= ip_hlen;
1866 type_tucmd_mlhl |= E1000_ADVTXD_DCMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
1867
1868 if (mp->m_pkthdr.csum_flags & CSUM_TCP)
1869 type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_L4T_TCP;
1870 else if (mp->m_pkthdr.csum_flags & CSUM_UDP)
1871 type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_L4T_UDP;
1872
1873 /* 82575 needs the queue index added */
1874 if (txr->sc->hw.mac.type == e1000_82575)
1875 mss_l4len_idx = txr->me << 4;
1876
1877 /* Now copy bits into descriptor */
1878 TXD->vlan_macip_lens = htole32(vlan_macip_lens);
1879 TXD->type_tucmd_mlhl = htole32(type_tucmd_mlhl);
1880 TXD->seqnum_seed = htole32(0);
1881 TXD->mss_l4len_idx = htole32(mss_l4len_idx);
1882
1883 txbuf->m_head = NULL;
1f7e3916
SZ		 1884
1885 /* We've consumed the first desc, adjust counters */
91b8700a 1886 if (++ctxd == txr->num_tx_desc)
1f7e3916
SZ		 1887 ctxd = 0;
1888 txr->next_avail_desc = ctxd;
1889 --txr->tx_avail;
1890
1891 return offload;
1892}
1893
1894static void
1895igb_txeof(struct igb_tx_ring *txr)
1896{
1897 struct ifnet *ifp = &txr->sc->arpcom.ac_if;
b6220144 1898 int first, hdr, avail;
1f7e3916 1899
91b8700a 1900 if (txr->tx_avail == txr->num_tx_desc)
1f7e3916
SZ		 1901 return;
1902
1903 first = txr->next_to_clean;
b6220144 1904 hdr = *(txr->tx_hdr);
1f7e3916 1905
b6220144
SZ		 1906 if (first == hdr)
1907 return;
1f7e3916 1908
b6220144
SZ		 1909 avail = txr->tx_avail;
1910 while (first != hdr) {
1911 struct igb_tx_buf *txbuf = &txr->tx_buf[first];
1f7e3916 1912
b6220144
SZ		 1913 ++avail;
1914 if (txbuf->m_head) {
1915 bus_dmamap_unload(txr->tx_tag, txbuf->map);
1916 m_freem(txbuf->m_head);
1917 txbuf->m_head = NULL;
1918 ++ifp->if_opackets;
1f7e3916 1919 }
91b8700a 1920 if (++first == txr->num_tx_desc)
b6220144 1921 first = 0;
1f7e3916
SZ		 1922 }
1923 txr->next_to_clean = first;
b6220144 1924 txr->tx_avail = avail;
1f7e3916
SZ		 1925
1926 /*
1927 * If we have a minimum free, clear IFF_OACTIVE
1928 * to tell the stack that it is OK to send packets.
1929 */
b6220144 1930 if (IGB_IS_NOT_OACTIVE(txr)) {
1f7e3916
SZ		 1931 ifp->if_flags &= ~IFF_OACTIVE;
1932
1f7e3916
SZ		 1933 /*
1934 * We have enough TX descriptors, turn off
b6220144
SZ		 1935 * the watchdog. We allow small amount of
1936 * packets (roughly intr_nsegs) pending on
1937 * the transmit ring.
1f7e3916
SZ		 1938 */
1939 ifp->if_timer = 0;
1f7e3916
SZ		 1940 }
1941}
1942
1943static int
1944igb_create_rx_ring(struct igb_rx_ring *rxr)
1945{
1946 int rsize, i, error;
1947
1948 /*
1949 * Validate number of receive descriptors. It must not exceed
1950 * hardware maximum, and must be multiple of IGB_DBA_ALIGN.
1951 */
1952 if (((igb_rxd * sizeof(struct e1000_rx_desc)) % IGB_DBA_ALIGN) != 0 ||
1953 (igb_rxd > IGB_MAX_RXD) || (igb_rxd < IGB_MIN_RXD)) {
1954 device_printf(rxr->sc->dev,
1955 "Using %d RX descriptors instead of %d!\n",
1956 IGB_DEFAULT_RXD, igb_rxd);
91b8700a 1957 rxr->num_rx_desc = IGB_DEFAULT_RXD;
1f7e3916 1958 } else {
91b8700a 1959 rxr->num_rx_desc = igb_rxd;
1f7e3916
SZ		 1960 }
1961
1962 /*
1963 * Allocate RX descriptor ring
1964 */
91b8700a 1965 rsize = roundup2(rxr->num_rx_desc * sizeof(union e1000_adv_rx_desc),
1f7e3916
SZ		 1966 IGB_DBA_ALIGN);
1967 rxr->rxdma.dma_vaddr = bus_dmamem_coherent_any(rxr->sc->parent_tag,
1968 IGB_DBA_ALIGN, rsize, BUS_DMA_WAITOK,
1969 &rxr->rxdma.dma_tag, &rxr->rxdma.dma_map,
1970 &rxr->rxdma.dma_paddr);
1971 if (rxr->rxdma.dma_vaddr == NULL) {
1972 device_printf(rxr->sc->dev,
1973 "Unable to allocate RxDescriptor memory\n");
1974 return ENOMEM;
1975 }
1976 rxr->rx_base = rxr->rxdma.dma_vaddr;
1977 bzero(rxr->rx_base, rsize);
1978
91b8700a 1979 rxr->rx_buf = kmalloc(sizeof(struct igb_rx_buf) * rxr->num_rx_desc,
1f7e3916
SZ		 1980 M_DEVBUF, M_WAITOK | M_ZERO);
1981
1982 /*
1983 * Create DMA tag for RX buffers
1984 */
1985 error = bus_dma_tag_create(rxr->sc->parent_tag,
1986 1, 0, /* alignment, bounds */
1987 BUS_SPACE_MAXADDR, /* lowaddr */
1988 BUS_SPACE_MAXADDR, /* highaddr */
1989 NULL, NULL, /* filter, filterarg */
1990 MCLBYTES, /* maxsize */
1991 1, /* nsegments */
1992 MCLBYTES, /* maxsegsize */
1993 BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW, /* flags */
1994 &rxr->rx_tag);
1995 if (error) {
1996 device_printf(rxr->sc->dev,
1997 "Unable to create RX payload DMA tag\n");
1998 kfree(rxr->rx_buf, M_DEVBUF);
1999 rxr->rx_buf = NULL;
2000 return error;
2001 }
2002
2003 /*
2004 * Create spare DMA map for RX buffers
2005 */
2006 error = bus_dmamap_create(rxr->rx_tag, BUS_DMA_WAITOK,
2007 &rxr->rx_sparemap);
2008 if (error) {
2009 device_printf(rxr->sc->dev,
2010 "Unable to create spare RX DMA maps\n");
2011 bus_dma_tag_destroy(rxr->rx_tag);
2012 kfree(rxr->rx_buf, M_DEVBUF);
2013 rxr->rx_buf = NULL;
2014 return error;
2015 }
2016
2017 /*
2018 * Create DMA maps for RX buffers
2019 */
91b8700a 2020 for (i = 0; i < rxr->num_rx_desc; i++) {
1f7e3916
SZ		 2021 struct igb_rx_buf *rxbuf = &rxr->rx_buf[i];
2022
2023 error = bus_dmamap_create(rxr->rx_tag,
2024 BUS_DMA_WAITOK, &rxbuf->map);
2025 if (error) {
2026 device_printf(rxr->sc->dev,
2027 "Unable to create RX DMA maps\n");
2028 igb_destroy_rx_ring(rxr, i);
2029 return error;
2030 }
2031 }
2032 return 0;
2033}
2034
2035static void
2036igb_free_rx_ring(struct igb_rx_ring *rxr)
2037{
2038 int i;
2039
91b8700a 2040 for (i = 0; i < rxr->num_rx_desc; ++i) {
1f7e3916
SZ		 2041 struct igb_rx_buf *rxbuf = &rxr->rx_buf[i];
2042
2043 if (rxbuf->m_head != NULL) {
2044 bus_dmamap_unload(rxr->rx_tag, rxbuf->map);
2045 m_freem(rxbuf->m_head);
2046 rxbuf->m_head = NULL;
2047 }
2048 }
2049
2050 if (rxr->fmp != NULL)
2051 m_freem(rxr->fmp);
2052 rxr->fmp = NULL;
2053 rxr->lmp = NULL;
2054}
2055
2056static void
2057igb_destroy_rx_ring(struct igb_rx_ring *rxr, int ndesc)
2058{
2059 int i;
2060
2061 if (rxr->rxdma.dma_vaddr != NULL) {
2062 bus_dmamap_unload(rxr->rxdma.dma_tag, rxr->rxdma.dma_map);
2063 bus_dmamem_free(rxr->rxdma.dma_tag, rxr->rxdma.dma_vaddr,
2064 rxr->rxdma.dma_map);
2065 bus_dma_tag_destroy(rxr->rxdma.dma_tag);
2066 rxr->rxdma.dma_vaddr = NULL;
2067 }
2068
2069 if (rxr->rx_buf == NULL)
2070 return;
2071
2072 for (i = 0; i < ndesc; ++i) {
2073 struct igb_rx_buf *rxbuf = &rxr->rx_buf[i];
2074
2075 KKASSERT(rxbuf->m_head == NULL);
2076 bus_dmamap_destroy(rxr->rx_tag, rxbuf->map);
2077 }
2078 bus_dmamap_destroy(rxr->rx_tag, rxr->rx_sparemap);
2079 bus_dma_tag_destroy(rxr->rx_tag);
2080
2081 kfree(rxr->rx_buf, M_DEVBUF);
2082 rxr->rx_buf = NULL;
2083}
2084
2085static void
2086igb_setup_rxdesc(union e1000_adv_rx_desc *rxd, const struct igb_rx_buf *rxbuf)
2087{
2088 rxd->read.pkt_addr = htole64(rxbuf->paddr);
2089 rxd->wb.upper.status_error = 0;
2090}
2091
2092static int
2093igb_newbuf(struct igb_rx_ring *rxr, int i, boolean_t wait)
2094{
2095 struct mbuf *m;
2096 bus_dma_segment_t seg;
2097 bus_dmamap_t map;
2098 struct igb_rx_buf *rxbuf;
2099 int error, nseg;
2100
2101 m = m_getcl(wait ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
2102 if (m == NULL) {
2103 if (wait) {
2104 if_printf(&rxr->sc->arpcom.ac_if,
2105 "Unable to allocate RX mbuf\n");
2106 }
2107 return ENOBUFS;
2108 }
2109 m->m_len = m->m_pkthdr.len = MCLBYTES;
2110
2111 if (rxr->sc->max_frame_size <= MCLBYTES - ETHER_ALIGN)
2112 m_adj(m, ETHER_ALIGN);
2113
2114 error = bus_dmamap_load_mbuf_segment(rxr->rx_tag,
2115 rxr->rx_sparemap, m, &seg, 1, &nseg, BUS_DMA_NOWAIT);
2116 if (error) {
2117 m_freem(m);
2118 if (wait) {
2119 if_printf(&rxr->sc->arpcom.ac_if,
2120 "Unable to load RX mbuf\n");
2121 }
2122 return error;
2123 }
2124
2125 rxbuf = &rxr->rx_buf[i];
2126 if (rxbuf->m_head != NULL)
2127 bus_dmamap_unload(rxr->rx_tag, rxbuf->map);
2128
2129 map = rxbuf->map;
2130 rxbuf->map = rxr->rx_sparemap;
2131 rxr->rx_sparemap = map;
2132
2133 rxbuf->m_head = m;
2134 rxbuf->paddr = seg.ds_addr;
2135
2136 igb_setup_rxdesc(&rxr->rx_base[i], rxbuf);
2137 return 0;
2138}
2139
2140static int
2141igb_init_rx_ring(struct igb_rx_ring *rxr)
2142{
2143 int i;
2144
2145 /* Clear the ring contents */
2146 bzero(rxr->rx_base,
91b8700a 2147 rxr->num_rx_desc * sizeof(union e1000_adv_rx_desc));
1f7e3916
SZ		 2148
2149 /* Now replenish the ring mbufs */
91b8700a 2150 for (i = 0; i < rxr->num_rx_desc; ++i) {
1f7e3916
SZ		 2151 int error;
2152
2153 error = igb_newbuf(rxr, i, TRUE);
2154 if (error)
2155 return error;
2156 }
2157
2158 /* Setup our descriptor indices */
2159 rxr->next_to_check = 0;
2160
2161 rxr->fmp = NULL;
2162 rxr->lmp = NULL;
2163 rxr->discard = FALSE;
2164
2165 return 0;
2166}
2167
2168static void
2169igb_init_rx_unit(struct igb_softc *sc)
2170{
2171 struct ifnet *ifp = &sc->arpcom.ac_if;
2172 struct e1000_hw *hw = &sc->hw;
2173 uint32_t rctl, rxcsum, srrctl = 0;
2174 int i;
2175
2176 /*
2177 * Make sure receives are disabled while setting
2178 * up the descriptor ring
2179 */
2180 rctl = E1000_READ_REG(hw, E1000_RCTL);
2181 E1000_WRITE_REG(hw, E1000_RCTL, rctl & ~E1000_RCTL_EN);
2182
2183#if 0
2184 /*
2185 ** Set up for header split
2186 */
2187 if (igb_header_split) {
2188 /* Use a standard mbuf for the header */
2189 srrctl |= IGB_HDR_BUF << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
2190 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2191 } else
2192#endif
2193 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
2194
2195 /*
2196 ** Set up for jumbo frames
2197 */
2198 if (ifp->if_mtu > ETHERMTU) {
2199 rctl |= E1000_RCTL_LPE;
2200#if 0
2201 if (adapter->rx_mbuf_sz == MJUMPAGESIZE) {
2202 srrctl |= 4096 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2203 rctl |= E1000_RCTL_SZ_4096 | E1000_RCTL_BSEX;
2204 } else if (adapter->rx_mbuf_sz > MJUMPAGESIZE) {
2205 srrctl |= 8192 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2206 rctl |= E1000_RCTL_SZ_8192 | E1000_RCTL_BSEX;
2207 }
2208 /* Set maximum packet len */
2209 psize = adapter->max_frame_size;
2210 /* are we on a vlan? */
2211 if (adapter->ifp->if_vlantrunk != NULL)
2212 psize += VLAN_TAG_SIZE;
2213 E1000_WRITE_REG(&adapter->hw, E1000_RLPML, psize);
2214#else
2215 srrctl |= 2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2216 rctl |= E1000_RCTL_SZ_2048;
2217#endif
2218 } else {
2219 rctl &= ~E1000_RCTL_LPE;
2220 srrctl |= 2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2221 rctl |= E1000_RCTL_SZ_2048;
2222 }
2223
2224 /* Setup the Base and Length of the Rx Descriptor Rings */
27866bf1 2225 for (i = 0; i < sc->rx_ring_cnt; ++i) {
1f7e3916
SZ		 2226 struct igb_rx_ring *rxr = &sc->rx_rings[i];
2227 uint64_t bus_addr = rxr->rxdma.dma_paddr;
2228 uint32_t rxdctl;
2229
2230 E1000_WRITE_REG(hw, E1000_RDLEN(i),
91b8700a 2231 rxr->num_rx_desc * sizeof(struct e1000_rx_desc));
1f7e3916
SZ		 2232 E1000_WRITE_REG(hw, E1000_RDBAH(i),
2233 (uint32_t)(bus_addr >> 32));
2234 E1000_WRITE_REG(hw, E1000_RDBAL(i),
2235 (uint32_t)bus_addr);
2236 E1000_WRITE_REG(hw, E1000_SRRCTL(i), srrctl);
2237 /* Enable this Queue */
2238 rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(i));
2239 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
2240 rxdctl &= 0xFFF00000;
2241 rxdctl |= IGB_RX_PTHRESH;
2242 rxdctl |= IGB_RX_HTHRESH << 8;
2243 rxdctl |= IGB_RX_WTHRESH << 16;
2244 E1000_WRITE_REG(hw, E1000_RXDCTL(i), rxdctl);
2245 }
2246
8d6600da
SZ		 2247 rxcsum = E1000_READ_REG(&sc->hw, E1000_RXCSUM);
2248 rxcsum &= ~(E1000_RXCSUM_PCSS_MASK | E1000_RXCSUM_IPPCSE);
2249
1f7e3916 2250 /*
8d6600da
SZ		 2251 * Receive Checksum Offload for TCP and UDP
2252 *
2253 * Checksum offloading is also enabled if multiple receive
2254 * queue is to be supported, since we need it to figure out
2255 * fragments.
1f7e3916 2256 */
8d6600da
SZ		 2257 if ((ifp->if_capenable & IFCAP_RXCSUM) || IGB_ENABLE_HWRSS(sc)) {
2258 /*
2259 * NOTE:
2260 * PCSD must be enabled to enable multiple
2261 * receive queues.
2262 */
2263 rxcsum |= E1000_RXCSUM_IPOFL | E1000_RXCSUM_TUOFL |
2264 E1000_RXCSUM_PCSD;
2265 } else {
2266 rxcsum &= ~(E1000_RXCSUM_IPOFL | E1000_RXCSUM_TUOFL |
2267 E1000_RXCSUM_PCSD);
2268 }
2269 E1000_WRITE_REG(&sc->hw, E1000_RXCSUM, rxcsum);
2270
2271 if (IGB_ENABLE_HWRSS(sc)) {
2272 uint8_t key[IGB_NRSSRK * IGB_RSSRK_SIZE];
2273 uint32_t reta, reta_shift;
2274
2275 /*
2276 * NOTE:
2277 * When we reach here, RSS has already been disabled
2278 * in igb_stop(), so we could safely configure RSS key
2279 * and redirect table.
2280 */
2281
2282 /*
2283 * Configure RSS key
2284 */
2285 toeplitz_get_key(key, sizeof(key));
2286 for (i = 0; i < IGB_NRSSRK; ++i) {
2287 uint32_t rssrk;
2288
2289 rssrk = IGB_RSSRK_VAL(key, i);
2290 IGB_RSS_DPRINTF(sc, 1, "rssrk%d 0x%08x\n", i, rssrk);
2291
2292 E1000_WRITE_REG(hw, E1000_RSSRK(i), rssrk);
1f7e3916 2293 }
1f7e3916
SZ		 2294
2295 /*
8d6600da
SZ		 2296 * Configure RSS redirect table in following fashion:
2297 * (hash & ring_cnt_mask) == rdr_table[(hash & rdr_table_mask)]
2298 */
2299 reta_shift = IGB_RETA_SHIFT;
2300 if (hw->mac.type == e1000_82575)
2301 reta_shift = IGB_RETA_SHIFT_82575;
2302 reta = 0;
2303 for (i = 0; i < IGB_RETA_SIZE; ++i) {
2304 uint32_t q;
2305
2306 q = (i % sc->rx_ring_cnt) << reta_shift;
2307 reta |= q << (8 * i);
2308 }
2309 IGB_RSS_DPRINTF(sc, 1, "reta 0x%08x\n", reta);
2310
2311 for (i = 0; i < IGB_NRETA; ++i)
2312 E1000_WRITE_REG(hw, E1000_RETA(i), reta);
2313
2314 /*
2315 * Enable multiple receive queues.
2316 * Enable IPv4 RSS standard hash functions.
2317 * Disable RSS interrupt on 82575
2318 */
2319 E1000_WRITE_REG(&sc->hw, E1000_MRQC,
2320 E1000_MRQC_ENABLE_RSS_4Q |
2321 E1000_MRQC_RSS_FIELD_IPV4_TCP |
2322 E1000_MRQC_RSS_FIELD_IPV4);
1f7e3916 2323 }
1f7e3916
SZ		 2324
2325 /* Setup the Receive Control Register */
2326 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2327 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_LBM_NO |
2328 E1000_RCTL_RDMTS_HALF |
2329 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2330 /* Strip CRC bytes. */
2331 rctl |= E1000_RCTL_SECRC;
2332 /* Make sure VLAN Filters are off */
2333 rctl &= ~E1000_RCTL_VFE;
2334 /* Don't store bad packets */
2335 rctl &= ~E1000_RCTL_SBP;
2336
2337 /* Enable Receives */
2338 E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2339
2340 /*
2341 * Setup the HW Rx Head and Tail Descriptor Pointers
2342 * - needs to be after enable
2343 */
27866bf1 2344 for (i = 0; i < sc->rx_ring_cnt; ++i) {
1f7e3916
SZ		 2345 struct igb_rx_ring *rxr = &sc->rx_rings[i];
2346
2347 E1000_WRITE_REG(hw, E1000_RDH(i), rxr->next_to_check);
91b8700a 2348 E1000_WRITE_REG(hw, E1000_RDT(i), rxr->num_rx_desc - 1);
1f7e3916
SZ		 2349 }
2350}
2351
2352static void
2353igb_rxeof(struct igb_rx_ring *rxr, int count)
2354{
2355 struct ifnet *ifp = &rxr->sc->arpcom.ac_if;
2356 union e1000_adv_rx_desc *cur;
2357 uint32_t staterr;
2358 int i;
2359
2360 i = rxr->next_to_check;
2361 cur = &rxr->rx_base[i];
2362 staterr = le32toh(cur->wb.upper.status_error);
2363
2364 if ((staterr & E1000_RXD_STAT_DD) == 0)
2365 return;
2366
2367 while ((staterr & E1000_RXD_STAT_DD) && count != 0) {
8d6600da 2368 struct pktinfo *pi = NULL, pi0;
1f7e3916
SZ		 2369 struct igb_rx_buf *rxbuf = &rxr->rx_buf[i];
2370 struct mbuf *m = NULL;
2371 boolean_t eop;
2372
2373 eop = (staterr & E1000_RXD_STAT_EOP) ? TRUE : FALSE;
2374 if (eop)
2375 --count;
2376
2377 if ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) == 0 &&
2378 !rxr->discard) {
2379 struct mbuf *mp = rxbuf->m_head;
8d6600da 2380 uint32_t hash, hashtype;
1f7e3916
SZ		 2381 uint16_t vlan;
2382 int len;
2383
2384 len = le16toh(cur->wb.upper.length);
2385 if (rxr->sc->hw.mac.type == e1000_i350 &&
2386 (staterr & E1000_RXDEXT_STATERR_LB))
2387 vlan = be16toh(cur->wb.upper.vlan);
2388 else
2389 vlan = le16toh(cur->wb.upper.vlan);
2390
8d6600da
SZ		 2391 hash = le32toh(cur->wb.lower.hi_dword.rss);
2392 hashtype = le32toh(cur->wb.lower.lo_dword.data) &
2393 E1000_RXDADV_RSSTYPE_MASK;
2394
2395 IGB_RSS_DPRINTF(rxr->sc, 10,
2396 "ring%d, hash 0x%08x, hashtype %u\n",
2397 rxr->me, hash, hashtype);
2398
1f7e3916
SZ		 2399 bus_dmamap_sync(rxr->rx_tag, rxbuf->map,
2400 BUS_DMASYNC_POSTREAD);
2401
2402 if (igb_newbuf(rxr, i, FALSE) != 0) {
2403 ifp->if_iqdrops++;
2404 goto discard;
2405 }
2406
2407 mp->m_len = len;
2408 if (rxr->fmp == NULL) {
2409 mp->m_pkthdr.len = len;
2410 rxr->fmp = mp;
2411 rxr->lmp = mp;
2412 } else {
2413 rxr->lmp->m_next = mp;
2414 rxr->lmp = rxr->lmp->m_next;
2415 rxr->fmp->m_pkthdr.len += len;
2416 }
2417
2418 if (eop) {
2419 m = rxr->fmp;
2420 rxr->fmp = NULL;
2421 rxr->lmp = NULL;
2422
2423 m->m_pkthdr.rcvif = ifp;
2424 ifp->if_ipackets++;
2425
2426 if (ifp->if_capenable & IFCAP_RXCSUM)
2427 igb_rxcsum(staterr, m);
2428
2429 if (staterr & E1000_RXD_STAT_VP) {
2430 m->m_pkthdr.ether_vlantag = vlan;
2431 m->m_flags |= M_VLANTAG;
2432 }
2433
1f7e3916 2434 if (ifp->if_capenable & IFCAP_RSS) {
8d6600da
SZ		 2435 pi = igb_rssinfo(m, &pi0,
2436 hash, hashtype, staterr);
1f7e3916 2437 }
8d6600da
SZ		 2438#ifdef IGB_RSS_DEBUG
2439 rxr->rx_packets++;
1f7e3916
SZ		 2440#endif
2441 }
2442 } else {
2443 ifp->if_ierrors++;
2444discard:
2445 igb_setup_rxdesc(cur, rxbuf);
2446 if (!eop)
2447 rxr->discard = TRUE;
2448 else
2449 rxr->discard = FALSE;
2450 if (rxr->fmp != NULL) {
2451 m_freem(rxr->fmp);
2452 rxr->fmp = NULL;
2453 rxr->lmp = NULL;
2454 }
2455 m = NULL;
2456 }
2457
2458 if (m != NULL)
8d6600da 2459 ether_input_pkt(ifp, m, pi);
1f7e3916
SZ		 2460
2461 /* Advance our pointers to the next descriptor. */
91b8700a 2462 if (++i == rxr->num_rx_desc)
1f7e3916
SZ		 2463 i = 0;
2464
2465 cur = &rxr->rx_base[i];
2466 staterr = le32toh(cur->wb.upper.status_error);
2467 }
2468 rxr->next_to_check = i;
2469
2470 if (--i < 0)
91b8700a 2471 i = rxr->num_rx_desc - 1;
1f7e3916
SZ		 2472 E1000_WRITE_REG(&rxr->sc->hw, E1000_RDT(rxr->me), i);
2473}
2474
2475
2476static void
2477igb_set_vlan(struct igb_softc *sc)
2478{
2479 struct e1000_hw *hw = &sc->hw;
2480 uint32_t reg;
2481#if 0
2482 struct ifnet *ifp = sc->arpcom.ac_if;
2483#endif
2484
2485 if (sc->vf_ifp) {
2486 e1000_rlpml_set_vf(hw, sc->max_frame_size + VLAN_TAG_SIZE);
2487 return;
2488 }
2489
2490 reg = E1000_READ_REG(hw, E1000_CTRL);
2491 reg |= E1000_CTRL_VME;
2492 E1000_WRITE_REG(hw, E1000_CTRL, reg);
2493
2494#if 0
2495 /* Enable the Filter Table */
2496 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) {
2497 reg = E1000_READ_REG(hw, E1000_RCTL);
2498 reg &= ~E1000_RCTL_CFIEN;
2499 reg |= E1000_RCTL_VFE;
2500 E1000_WRITE_REG(hw, E1000_RCTL, reg);
2501 }
2502#endif
2503
2504 /* Update the frame size */
2505 E1000_WRITE_REG(&sc->hw, E1000_RLPML,
2506 sc->max_frame_size + VLAN_TAG_SIZE);
2507
2508#if 0
2509 /* Don't bother with table if no vlans */
2510 if ((adapter->num_vlans == 0) ||
2511 ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0))
2512 return;
2513 /*
2514 ** A soft reset zero's out the VFTA, so
2515 ** we need to repopulate it now.
2516 */
2517 for (int i = 0; i < IGB_VFTA_SIZE; i++)
2518 if (adapter->shadow_vfta[i] != 0) {
2519 if (adapter->vf_ifp)
2520 e1000_vfta_set_vf(hw,
2521 adapter->shadow_vfta[i], TRUE);
2522 else
2523 E1000_WRITE_REG_ARRAY(hw, E1000_VFTA,
2524 i, adapter->shadow_vfta[i]);
2525 }
2526#endif
2527}
2528
2529static void
2530igb_enable_intr(struct igb_softc *sc)
2531{
7d235eb5 2532 lwkt_serialize_handler_enable(&sc->main_serialize);
1f7e3916 2533
f6167a56
SZ		 2534 if ((sc->flags & IGB_FLAG_SHARED_INTR) == 0) {
2535 /* XXX MSI-X should use sc->intr_mask */
2536 E1000_WRITE_REG(&sc->hw, E1000_EIAC, 0);
2537 E1000_WRITE_REG(&sc->hw, E1000_EIAM, sc->intr_mask);
2538 E1000_WRITE_REG(&sc->hw, E1000_EIMS, sc->intr_mask);
1f7e3916
SZ		 2539 E1000_WRITE_REG(&sc->hw, E1000_IMS, E1000_IMS_LSC);
2540 } else {
2541 E1000_WRITE_REG(&sc->hw, E1000_IMS, IMS_ENABLE_MASK);
2542 }
2543 E1000_WRITE_FLUSH(&sc->hw);
2544}
2545
2546static void
2547igb_disable_intr(struct igb_softc *sc)
2548{
f6167a56 2549 if ((sc->flags & IGB_FLAG_SHARED_INTR) == 0) {
1f7e3916
SZ		 2550 E1000_WRITE_REG(&sc->hw, E1000_EIMC, 0xffffffff);
2551 E1000_WRITE_REG(&sc->hw, E1000_EIAC, 0);
f6167a56 2552 }
1f7e3916
SZ		 2553 E1000_WRITE_REG(&sc->hw, E1000_IMC, 0xffffffff);
2554 E1000_WRITE_FLUSH(&sc->hw);
2555
7d235eb5 2556 lwkt_serialize_handler_disable(&sc->main_serialize);
1f7e3916
SZ		 2557}
2558
2559/*
2560 * Bit of a misnomer, what this really means is
2561 * to enable OS management of the system... aka
2562 * to disable special hardware management features
2563 */
2564static void
2565igb_get_mgmt(struct igb_softc *sc)
2566{
396b7048 2567 if (sc->flags & IGB_FLAG_HAS_MGMT) {
1f7e3916
SZ		 2568 int manc2h = E1000_READ_REG(&sc->hw, E1000_MANC2H);
2569 int manc = E1000_READ_REG(&sc->hw, E1000_MANC);
2570
2571 /* disable hardware interception of ARP */
2572 manc &= ~E1000_MANC_ARP_EN;
2573
2574 /* enable receiving management packets to the host */
2575 manc |= E1000_MANC_EN_MNG2HOST;
2576 manc2h |= 1 << 5; /* Mng Port 623 */
2577 manc2h |= 1 << 6; /* Mng Port 664 */
2578 E1000_WRITE_REG(&sc->hw, E1000_MANC2H, manc2h);
2579 E1000_WRITE_REG(&sc->hw, E1000_MANC, manc);
2580 }
2581}
2582
2583/*
2584 * Give control back to hardware management controller
2585 * if there is one.
2586 */
2587static void
2588igb_rel_mgmt(struct igb_softc *sc)
2589{
396b7048 2590 if (sc->flags & IGB_FLAG_HAS_MGMT) {
1f7e3916
SZ		 2591 int manc = E1000_READ_REG(&sc->hw, E1000_MANC);
2592
2593 /* Re-enable hardware interception of ARP */
2594 manc |= E1000_MANC_ARP_EN;
2595 manc &= ~E1000_MANC_EN_MNG2HOST;
2596
2597 E1000_WRITE_REG(&sc->hw, E1000_MANC, manc);
2598 }
2599}
2600
2601/*
2602 * Sets CTRL_EXT:DRV_LOAD bit.
2603 *
2604 * For ASF and Pass Through versions of f/w this means that
2605 * the driver is loaded.
2606 */
2607static void
2608igb_get_hw_control(struct igb_softc *sc)
2609{
2610 uint32_t ctrl_ext;
2611
2612 if (sc->vf_ifp)
2613 return;
2614
2615 /* Let firmware know the driver has taken over */
2616 ctrl_ext = E1000_READ_REG(&sc->hw, E1000_CTRL_EXT);
2617 E1000_WRITE_REG(&sc->hw, E1000_CTRL_EXT,
2618 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2619}
2620
2621/*
2622 * Resets CTRL_EXT:DRV_LOAD bit.
2623 *
2624 * For ASF and Pass Through versions of f/w this means that the
2625 * driver is no longer loaded.
2626 */
2627static void
2628igb_rel_hw_control(struct igb_softc *sc)
2629{
2630 uint32_t ctrl_ext;
2631
2632 if (sc->vf_ifp)
2633 return;
2634
2635 /* Let firmware taken over control of h/w */
2636 ctrl_ext = E1000_READ_REG(&sc->hw, E1000_CTRL_EXT);
2637 E1000_WRITE_REG(&sc->hw, E1000_CTRL_EXT,
2638 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2639}
2640
2641static int
2642igb_is_valid_ether_addr(const uint8_t *addr)
2643{
2644 uint8_t zero_addr[ETHER_ADDR_LEN] = { 0, 0, 0, 0, 0, 0 };
2645
2646 if ((addr[0] & 1) || !bcmp(addr, zero_addr, ETHER_ADDR_LEN))
2647 return FALSE;
2648 return TRUE;
2649}
2650
2651/*
2652 * Enable PCI Wake On Lan capability
2653 */
2654static void
2655igb_enable_wol(device_t dev)
2656{
2657 uint16_t cap, status;
2658 uint8_t id;
2659
2660 /* First find the capabilities pointer*/
2661 cap = pci_read_config(dev, PCIR_CAP_PTR, 2);
2662
2663 /* Read the PM Capabilities */
2664 id = pci_read_config(dev, cap, 1);
2665 if (id != PCIY_PMG) /* Something wrong */
2666 return;
2667
2668 /*
2669 * OK, we have the power capabilities,
2670 * so now get the status register
2671 */
2672 cap += PCIR_POWER_STATUS;
2673 status = pci_read_config(dev, cap, 2);
2674 status |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
2675 pci_write_config(dev, cap, status, 2);
2676}
2677
2678static void
2679igb_update_stats_counters(struct igb_softc *sc)
2680{
2681 struct e1000_hw *hw = &sc->hw;
2682 struct e1000_hw_stats *stats;
2683 struct ifnet *ifp = &sc->arpcom.ac_if;
2684
2685 /*
2686 * The virtual function adapter has only a
2687 * small controlled set of stats, do only
2688 * those and return.
2689 */
2690 if (sc->vf_ifp) {
2691 igb_update_vf_stats_counters(sc);
2692 return;
2693 }
2694 stats = sc->stats;
2695
2696 if (sc->hw.phy.media_type == e1000_media_type_copper ||
2697 (E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU)) {
2698 stats->symerrs +=
2699 E1000_READ_REG(hw,E1000_SYMERRS);
2700 stats->sec += E1000_READ_REG(hw, E1000_SEC);
2701 }
2702
2703 stats->crcerrs += E1000_READ_REG(hw, E1000_CRCERRS);
2704 stats->mpc += E1000_READ_REG(hw, E1000_MPC);
2705 stats->scc += E1000_READ_REG(hw, E1000_SCC);
2706 stats->ecol += E1000_READ_REG(hw, E1000_ECOL);
2707
2708 stats->mcc += E1000_READ_REG(hw, E1000_MCC);
2709 stats->latecol += E1000_READ_REG(hw, E1000_LATECOL);
2710 stats->colc += E1000_READ_REG(hw, E1000_COLC);
2711 stats->dc += E1000_READ_REG(hw, E1000_DC);
2712 stats->rlec += E1000_READ_REG(hw, E1000_RLEC);
2713 stats->xonrxc += E1000_READ_REG(hw, E1000_XONRXC);
2714 stats->xontxc += E1000_READ_REG(hw, E1000_XONTXC);
2715
2716 /*
2717 * For watchdog management we need to know if we have been
2718 * paused during the last interval, so capture that here.
2719 */
2720 sc->pause_frames = E1000_READ_REG(hw, E1000_XOFFRXC);
2721 stats->xoffrxc += sc->pause_frames;
2722 stats->xofftxc += E1000_READ_REG(hw, E1000_XOFFTXC);
2723 stats->fcruc += E1000_READ_REG(hw, E1000_FCRUC);
2724 stats->prc64 += E1000_READ_REG(hw, E1000_PRC64);
2725 stats->prc127 += E1000_READ_REG(hw, E1000_PRC127);
2726 stats->prc255 += E1000_READ_REG(hw, E1000_PRC255);
2727 stats->prc511 += E1000_READ_REG(hw, E1000_PRC511);
2728 stats->prc1023 += E1000_READ_REG(hw, E1000_PRC1023);
2729 stats->prc1522 += E1000_READ_REG(hw, E1000_PRC1522);
2730 stats->gprc += E1000_READ_REG(hw, E1000_GPRC);
2731 stats->bprc += E1000_READ_REG(hw, E1000_BPRC);
2732 stats->mprc += E1000_READ_REG(hw, E1000_MPRC);
2733 stats->gptc += E1000_READ_REG(hw, E1000_GPTC);
2734
2735 /* For the 64-bit byte counters the low dword must be read first. */
2736 /* Both registers clear on the read of the high dword */
2737
2738 stats->gorc += E1000_READ_REG(hw, E1000_GORCL) +
2739 ((uint64_t)E1000_READ_REG(hw, E1000_GORCH) << 32);
2740 stats->gotc += E1000_READ_REG(hw, E1000_GOTCL) +
2741 ((uint64_t)E1000_READ_REG(hw, E1000_GOTCH) << 32);
2742
2743 stats->rnbc += E1000_READ_REG(hw, E1000_RNBC);
2744 stats->ruc += E1000_READ_REG(hw, E1000_RUC);
2745 stats->rfc += E1000_READ_REG(hw, E1000_RFC);
2746 stats->roc += E1000_READ_REG(hw, E1000_ROC);
2747 stats->rjc += E1000_READ_REG(hw, E1000_RJC);
2748
2749 stats->tor += E1000_READ_REG(hw, E1000_TORH);
2750 stats->tot += E1000_READ_REG(hw, E1000_TOTH);
2751
2752 stats->tpr += E1000_READ_REG(hw, E1000_TPR);
2753 stats->tpt += E1000_READ_REG(hw, E1000_TPT);
2754 stats->ptc64 += E1000_READ_REG(hw, E1000_PTC64);
2755 stats->ptc127 += E1000_READ_REG(hw, E1000_PTC127);
2756 stats->ptc255 += E1000_READ_REG(hw, E1000_PTC255);
2757 stats->ptc511 += E1000_READ_REG(hw, E1000_PTC511);
2758 stats->ptc1023 += E1000_READ_REG(hw, E1000_PTC1023);
2759 stats->ptc1522 += E1000_READ_REG(hw, E1000_PTC1522);
2760 stats->mptc += E1000_READ_REG(hw, E1000_MPTC);
2761 stats->bptc += E1000_READ_REG(hw, E1000_BPTC);
2762
2763 /* Interrupt Counts */
2764
2765 stats->iac += E1000_READ_REG(hw, E1000_IAC);
2766 stats->icrxptc += E1000_READ_REG(hw, E1000_ICRXPTC);
2767 stats->icrxatc += E1000_READ_REG(hw, E1000_ICRXATC);
2768 stats->ictxptc += E1000_READ_REG(hw, E1000_ICTXPTC);
2769 stats->ictxatc += E1000_READ_REG(hw, E1000_ICTXATC);
2770 stats->ictxqec += E1000_READ_REG(hw, E1000_ICTXQEC);
2771 stats->ictxqmtc += E1000_READ_REG(hw, E1000_ICTXQMTC);
2772 stats->icrxdmtc += E1000_READ_REG(hw, E1000_ICRXDMTC);
2773 stats->icrxoc += E1000_READ_REG(hw, E1000_ICRXOC);
2774
2775 /* Host to Card Statistics */
2776
2777 stats->cbtmpc += E1000_READ_REG(hw, E1000_CBTMPC);
2778 stats->htdpmc += E1000_READ_REG(hw, E1000_HTDPMC);
2779 stats->cbrdpc += E1000_READ_REG(hw, E1000_CBRDPC);
2780 stats->cbrmpc += E1000_READ_REG(hw, E1000_CBRMPC);
2781 stats->rpthc += E1000_READ_REG(hw, E1000_RPTHC);
2782 stats->hgptc += E1000_READ_REG(hw, E1000_HGPTC);
2783 stats->htcbdpc += E1000_READ_REG(hw, E1000_HTCBDPC);
2784 stats->hgorc += (E1000_READ_REG(hw, E1000_HGORCL) +
2785 ((uint64_t)E1000_READ_REG(hw, E1000_HGORCH) << 32));
2786 stats->hgotc += (E1000_READ_REG(hw, E1000_HGOTCL) +
2787 ((uint64_t)E1000_READ_REG(hw, E1000_HGOTCH) << 32));
2788 stats->lenerrs += E1000_READ_REG(hw, E1000_LENERRS);
2789 stats->scvpc += E1000_READ_REG(hw, E1000_SCVPC);
2790 stats->hrmpc += E1000_READ_REG(hw, E1000_HRMPC);
2791
2792 stats->algnerrc += E1000_READ_REG(hw, E1000_ALGNERRC);
2793 stats->rxerrc += E1000_READ_REG(hw, E1000_RXERRC);
2794 stats->tncrs += E1000_READ_REG(hw, E1000_TNCRS);
2795 stats->cexterr += E1000_READ_REG(hw, E1000_CEXTERR);
2796 stats->tsctc += E1000_READ_REG(hw, E1000_TSCTC);
2797 stats->tsctfc += E1000_READ_REG(hw, E1000_TSCTFC);
2798
2799 ifp->if_collisions = stats->colc;
2800
2801 /* Rx Errors */
2802 ifp->if_ierrors = stats->rxerrc + stats->crcerrs + stats->algnerrc +
2803 stats->ruc + stats->roc + stats->mpc + stats->cexterr;
2804
2805 /* Tx Errors */
2806 ifp->if_oerrors = stats->ecol + stats->latecol + sc->watchdog_events;
2807
2808 /* Driver specific counters */
2809 sc->device_control = E1000_READ_REG(hw, E1000_CTRL);
2810 sc->rx_control = E1000_READ_REG(hw, E1000_RCTL);
2811 sc->int_mask = E1000_READ_REG(hw, E1000_IMS);
2812 sc->eint_mask = E1000_READ_REG(hw, E1000_EIMS);
2813 sc->packet_buf_alloc_tx =
2814 ((E1000_READ_REG(hw, E1000_PBA) & 0xffff0000) >> 16);
2815 sc->packet_buf_alloc_rx =
2816 (E1000_READ_REG(hw, E1000_PBA) & 0xffff);
2817}
2818
2819static void
2820igb_vf_init_stats(struct igb_softc *sc)
2821{
2822 struct e1000_hw *hw = &sc->hw;
2823 struct e1000_vf_stats *stats;
2824
2825 stats = sc->stats;
2826 stats->last_gprc = E1000_READ_REG(hw, E1000_VFGPRC);
2827 stats->last_gorc = E1000_READ_REG(hw, E1000_VFGORC);
2828 stats->last_gptc = E1000_READ_REG(hw, E1000_VFGPTC);
2829 stats->last_gotc = E1000_READ_REG(hw, E1000_VFGOTC);
2830 stats->last_mprc = E1000_READ_REG(hw, E1000_VFMPRC);
2831}
2832
2833static void
2834igb_update_vf_stats_counters(struct igb_softc *sc)
2835{
2836 struct e1000_hw *hw = &sc->hw;
2837 struct e1000_vf_stats *stats;
2838
2839 if (sc->link_speed == 0)
2840 return;
2841
2842 stats = sc->stats;
2843 UPDATE_VF_REG(E1000_VFGPRC, stats->last_gprc, stats->gprc);
2844 UPDATE_VF_REG(E1000_VFGORC, stats->last_gorc, stats->gorc);
2845 UPDATE_VF_REG(E1000_VFGPTC, stats->last_gptc, stats->gptc);
2846 UPDATE_VF_REG(E1000_VFGOTC, stats->last_gotc, stats->gotc);
2847 UPDATE_VF_REG(E1000_VFMPRC, stats->last_mprc, stats->mprc);
2848}
2849
2850#ifdef DEVICE_POLLING
2851
2852static void
2853igb_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2854{
2855 struct igb_softc *sc = ifp->if_softc;
2856 uint32_t reg_icr;
2857
7d235eb5 2858 ASSERT_SERIALIZED(&sc->main_serialize);
1f7e3916
SZ		 2859
2860 switch (cmd) {
2861 case POLL_REGISTER:
2862 igb_disable_intr(sc);
2863 break;
2864
2865 case POLL_DEREGISTER:
2866 igb_enable_intr(sc);
2867 break;
2868
2869 case POLL_AND_CHECK_STATUS:
2870 reg_icr = E1000_READ_REG(&sc->hw, E1000_ICR);
2871 if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
2872 sc->hw.mac.get_link_status = 1;
2873 igb_update_link_status(sc);
2874 }
2875 /* FALL THROUGH */
2876 case POLL_ONLY:
2877 if (ifp->if_flags & IFF_RUNNING) {
7d235eb5
SZ		 2878 struct igb_tx_ring *txr;
2879 int i;
2880
2881 for (i = 0; i < sc->rx_ring_cnt; ++i) {
2882 struct igb_rx_ring *rxr = &sc->rx_rings[i];
2883
2884 lwkt_serialize_enter(&rxr->rx_serialize);
2885 igb_rxeof(rxr, count);
2886 lwkt_serialize_exit(&rxr->rx_serialize);
2887 }
1f7e3916 2888
7d235eb5
SZ		 2889 txr = &sc->tx_rings[0];
2890 lwkt_serialize_enter(&txr->tx_serialize);
2891 igb_txeof(txr);
1f7e3916
SZ		 2892 if (!ifq_is_empty(&ifp->if_snd))
2893 if_devstart(ifp);
7d235eb5 2894 lwkt_serialize_exit(&txr->tx_serialize);
1f7e3916
SZ		 2895 }
2896 break;
2897 }
2898}
2899
2900#endif /* DEVICE_POLLING */
2901
2902static void
2903igb_intr(void *xsc)
2904{
2905 struct igb_softc *sc = xsc;
2906 struct ifnet *ifp = &sc->arpcom.ac_if;
f6167a56
SZ		 2907 uint32_t eicr;
2908
7d235eb5 2909 ASSERT_SERIALIZED(&sc->main_serialize);
f6167a56
SZ		 2910
2911 eicr = E1000_READ_REG(&sc->hw, E1000_EICR);
2912
2913 if (eicr == 0)
2914 return;
2915
2916 if (ifp->if_flags & IFF_RUNNING) {
7d235eb5
SZ		 2917 struct igb_tx_ring *txr;
2918 int i;
f6167a56 2919
7d235eb5
SZ		 2920 for (i = 0; i < sc->rx_ring_cnt; ++i) {
2921 struct igb_rx_ring *rxr = &sc->rx_rings[i];
2922
2923 if (eicr & rxr->rx_intr_mask) {
2924 lwkt_serialize_enter(&rxr->rx_serialize);
2925 igb_rxeof(rxr, -1);
2926 lwkt_serialize_exit(&rxr->rx_serialize);
2927 }
2928 }
2929
2930 txr = &sc->tx_rings[0];
2931 if (eicr & txr->tx_intr_mask) {
2932 lwkt_serialize_enter(&txr->tx_serialize);
2933 igb_txeof(txr);
f6167a56
SZ		 2934 if (!ifq_is_empty(&ifp->if_snd))
2935 if_devstart(ifp);
7d235eb5 2936 lwkt_serialize_exit(&txr->tx_serialize);
f6167a56
SZ		 2937 }
2938 }
2939
2940 if (eicr & E1000_EICR_OTHER) {
2941 uint32_t icr = E1000_READ_REG(&sc->hw, E1000_ICR);
2942
2943 /* Link status change */
2944 if (icr & E1000_ICR_LSC) {
2945 sc->hw.mac.get_link_status = 1;
2946 igb_update_link_status(sc);
2947 }
2948 }
2949
2950 /*
2951 * Reading EICR has the side effect to clear interrupt mask,
2952 * so all interrupts need to be enabled here.
2953 */
2954 E1000_WRITE_REG(&sc->hw, E1000_EIMS, sc->intr_mask);
2955}
2956
2957static void
2958igb_shared_intr(void *xsc)
2959{
2960 struct igb_softc *sc = xsc;
2961 struct ifnet *ifp = &sc->arpcom.ac_if;
1f7e3916
SZ		 2962 uint32_t reg_icr;
2963
7d235eb5 2964 ASSERT_SERIALIZED(&sc->main_serialize);
1f7e3916
SZ		 2965
2966 reg_icr = E1000_READ_REG(&sc->hw, E1000_ICR);
2967
2968 /* Hot eject? */
2969 if (reg_icr == 0xffffffff)
2970 return;
2971
2972 /* Definitely not our interrupt. */
2973 if (reg_icr == 0x0)
2974 return;
2975
2976 if ((reg_icr & E1000_ICR_INT_ASSERTED) == 0)
2977 return;
2978
2979 if (ifp->if_flags & IFF_RUNNING) {
7d235eb5
SZ		 2980 struct igb_tx_ring *txr;
2981 int i;
2982
2983 for (i = 0; i < sc->rx_ring_cnt; ++i) {
2984 struct igb_rx_ring *rxr = &sc->rx_rings[i];
1f7e3916 2985
7d235eb5
SZ		 2986 lwkt_serialize_enter(&rxr->rx_serialize);
2987 igb_rxeof(rxr, -1);
2988 lwkt_serialize_exit(&rxr->rx_serialize);
2989 }
2990
2991 txr = &sc->tx_rings[0];
2992 lwkt_serialize_enter(&txr->tx_serialize);
2993 igb_txeof(txr);
1f7e3916
SZ		 2994 if (!ifq_is_empty(&ifp->if_snd))
2995 if_devstart(ifp);
7d235eb5 2996 lwkt_serialize_exit(&txr->tx_serialize);
1f7e3916
SZ		 2997 }
2998
2999 /* Link status change */
3000 if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3001 sc->hw.mac.get_link_status = 1;
3002 igb_update_link_status(sc);
3003 }
3004
3005 if (reg_icr & E1000_ICR_RXO)
3006 sc->rx_overruns++;
3007}
3008
3009static int
3010igb_txctx_pullup(struct igb_tx_ring *txr, struct mbuf **m0)
3011{
3012 struct mbuf *m = *m0;
3013 struct ether_header *eh;
3014 int len;
3015
3016 txr->ctx_try_pullup++;
3017
3018 len = ETHER_HDR_LEN + IGB_IPVHL_SIZE;
3019
3020 if (__predict_false(!M_WRITABLE(m))) {
3021 if (__predict_false(m->m_len < ETHER_HDR_LEN)) {
3022 txr->ctx_drop1++;
3023 m_freem(m);
3024 *m0 = NULL;
3025 return ENOBUFS;
3026 }
3027 eh = mtod(m, struct ether_header *);
3028
3029 if (eh->ether_type == htons(ETHERTYPE_VLAN))
3030 len += EVL_ENCAPLEN;
3031
3032 if (m->m_len < len) {
3033 txr->ctx_drop2++;
3034 m_freem(m);
3035 *m0 = NULL;
3036 return ENOBUFS;
3037 }
3038 return 0;
3039 }
3040
3041 if (__predict_false(m->m_len < ETHER_HDR_LEN)) {
3042 txr->ctx_pullup1++;
3043 m = m_pullup(m, ETHER_HDR_LEN);
3044 if (m == NULL) {
3045 txr->ctx_pullup1_failed++;
3046 *m0 = NULL;
3047 return ENOBUFS;
3048 }
3049 *m0 = m;
3050 }
3051 eh = mtod(m, struct ether_header *);
3052
3053 if (eh->ether_type == htons(ETHERTYPE_VLAN))
3054 len += EVL_ENCAPLEN;
3055
3056 if (m->m_len < len) {
3057 txr->ctx_pullup2++;
3058 m = m_pullup(m, len);
3059 if (m == NULL) {
3060 txr->ctx_pullup2_failed++;
3061 *m0 = NULL;
3062 return ENOBUFS;
3063 }
3064 *m0 = m;
3065 }
3066 return 0;
3067}
3068
3069static int
3070igb_encap(struct igb_tx_ring *txr, struct mbuf **m_headp)
3071{
3072 bus_dma_segment_t segs[IGB_MAX_SCATTER];
3073 bus_dmamap_t map;
3074 struct igb_tx_buf *tx_buf, *tx_buf_mapped;
3075 union e1000_adv_tx_desc *txd = NULL;
3076 struct mbuf *m_head = *m_headp;
b6220144
SZ		 3077 uint32_t olinfo_status = 0, cmd_type_len = 0, cmd_rs = 0;
3078 int maxsegs, nsegs, i, j, error, last = 0;
1f7e3916
SZ		 3079 uint32_t hdrlen = 0;
3080
3081 if (m_head->m_len < IGB_TXCSUM_MINHL &&
3082 ((m_head->m_pkthdr.csum_flags & IGB_CSUM_FEATURES) ||
3083 (m_head->m_flags & M_VLANTAG))) {
3084 /*
3085 * Make sure that ethernet header and ip.ip_hl are in
3086 * contiguous memory, since if TXCSUM or VLANTAG is
3087 * enabled, later TX context descriptor's setup need
3088 * to access ip.ip_hl.
3089 */
3090 error = igb_txctx_pullup(txr, m_headp);
3091 if (error) {
3092 KKASSERT(*m_headp == NULL);
3093 return error;
3094 }
3095 m_head = *m_headp;
3096 }
3097
3098 /* Set basic descriptor constants */
3099 cmd_type_len |= E1000_ADVTXD_DTYP_DATA;
3100 cmd_type_len |= E1000_ADVTXD_DCMD_IFCS | E1000_ADVTXD_DCMD_DEXT;
3101 if (m_head->m_flags & M_VLANTAG)
3102 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
3103
3104 /*
3105 * Map the packet for DMA.
1f7e3916 3106 */
b6220144 3107 tx_buf = &txr->tx_buf[txr->next_avail_desc];
1f7e3916
SZ		 3108 tx_buf_mapped = tx_buf;
3109 map = tx_buf->map;
3110
b6220144
SZ		 3111 maxsegs = txr->tx_avail - IGB_TX_RESERVED;
3112 KASSERT(maxsegs >= txr->spare_desc, ("not enough spare TX desc\n"));
1f7e3916
SZ		 3113 if (maxsegs > IGB_MAX_SCATTER)
3114 maxsegs = IGB_MAX_SCATTER;
3115
3116 error = bus_dmamap_load_mbuf_defrag(txr->tx_tag, map, m_headp,
3117 segs, maxsegs, &nsegs, BUS_DMA_NOWAIT);
3118 if (error) {
3119 if (error == ENOBUFS)
3120 txr->sc->mbuf_defrag_failed++;
3121 else
3122 txr->sc->no_tx_dma_setup++;
3123
3124 m_freem(*m_headp);
3125 *m_headp = NULL;
3126 return error;
3127 }
3128 bus_dmamap_sync(txr->tx_tag, map, BUS_DMASYNC_PREWRITE);
3129
3130 m_head = *m_headp;
3131
3132#if 0
3133 /*
3134 * Set up the context descriptor:
3135 * used when any hardware offload is done.
3136 * This includes CSUM, VLAN, and TSO. It
3137 * will use the first descriptor.
3138 */
3139 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
3140 if (igb_tso_setup(txr, m_head, &hdrlen)) {
3141 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
3142 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
3143 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3144 } else
3145 return (ENXIO);
3146 } else if (igb_tx_ctx_setup(txr, m_head))
3147 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3148#else
3149 if (igb_txctx(txr, m_head)) {
3150 olinfo_status |= (E1000_TXD_POPTS_IXSM << 8);
3151 if (m_head->m_pkthdr.csum_flags & (CSUM_UDP | CSUM_TCP))
3152 olinfo_status |= (E1000_TXD_POPTS_TXSM << 8);
b6220144 3153 txr->tx_nsegs++;
1f7e3916
SZ		 3154 }
3155#endif
3156
b6220144
SZ		 3157 txr->tx_nsegs += nsegs;
3158 if (txr->tx_nsegs >= txr->intr_nsegs) {
3159 /*
3160 * Report Status (RS) is turned on every intr_nsegs
3161 * descriptors (roughly).
3162 */
3163 txr->tx_nsegs = 0;
3164 cmd_rs = E1000_ADVTXD_DCMD_RS;
3165 }
3166
1f7e3916
SZ		 3167 /* Calculate payload length */
3168 olinfo_status |= ((m_head->m_pkthdr.len - hdrlen)
3169 << E1000_ADVTXD_PAYLEN_SHIFT);
3170
3171 /* 82575 needs the queue index added */
3172 if (txr->sc->hw.mac.type == e1000_82575)
3173 olinfo_status |= txr->me << 4;
3174
3175 /* Set up our transmit descriptors */
3176 i = txr->next_avail_desc;
3177 for (j = 0; j < nsegs; j++) {
3178 bus_size_t seg_len;
3179 bus_addr_t seg_addr;
3180
3181 tx_buf = &txr->tx_buf[i];
3182 txd = (union e1000_adv_tx_desc *)&txr->tx_base[i];
3183 seg_addr = segs[j].ds_addr;
3184 seg_len = segs[j].ds_len;
3185
3186 txd->read.buffer_addr = htole64(seg_addr);
3187 txd->read.cmd_type_len = htole32(cmd_type_len | seg_len);
3188 txd->read.olinfo_status = htole32(olinfo_status);
3189 last = i;
91b8700a 3190 if (++i == txr->num_tx_desc)
1f7e3916
SZ		 3191 i = 0;
3192 tx_buf->m_head = NULL;
1f7e3916
SZ		 3193 }
3194
3195 KASSERT(txr->tx_avail > nsegs, ("invalid avail TX desc\n"));
3196 txr->next_avail_desc = i;
3197 txr->tx_avail -= nsegs;
3198
3199 tx_buf->m_head = m_head;
3200 tx_buf_mapped->map = tx_buf->map;
3201 tx_buf->map = map;
3202
3203 /*
b6220144 3204 * Last Descriptor of Packet needs End Of Packet (EOP)
1f7e3916 3205 */
b6220144 3206 txd->read.cmd_type_len |= htole32(E1000_ADVTXD_DCMD_EOP | cmd_rs);
1f7e3916
SZ		 3207
3208 /*
3209 * Advance the Transmit Descriptor Tail (TDT), this tells the E1000
3210 * that this frame is available to transmit.
3211 */
3212 E1000_WRITE_REG(&txr->sc->hw, E1000_TDT(txr->me), i);
3213 ++txr->tx_packets;
3214
3215 return 0;
3216}
3217
3218static void
3219igb_start(struct ifnet *ifp)
3220{
3221 struct igb_softc *sc = ifp->if_softc;
27866bf1 3222 struct igb_tx_ring *txr = &sc->tx_rings[0];
1f7e3916
SZ		 3223 struct mbuf *m_head;
3224
7d235eb5 3225 ASSERT_SERIALIZED(&txr->tx_serialize);
1f7e3916
SZ		 3226
3227 if ((ifp->if_flags & (IFF_RUNNING|IFF_OACTIVE)) != IFF_RUNNING)
3228 return;
3229
3230 if (!sc->link_active) {
3231 ifq_purge(&ifp->if_snd);
3232 return;
3233 }
3234
b6220144 3235 if (!IGB_IS_NOT_OACTIVE(txr))
1f7e3916
SZ		 3236 igb_txeof(txr);
3237
3238 while (!ifq_is_empty(&ifp->if_snd)) {
b6220144 3239 if (IGB_IS_OACTIVE(txr)) {
1f7e3916
SZ		 3240 ifp->if_flags |= IFF_OACTIVE;
3241 /* Set watchdog on */
3242 ifp->if_timer = 5;
3243 break;
3244 }
3245
3246 m_head = ifq_dequeue(&ifp->if_snd, NULL);
3247 if (m_head == NULL)
3248 break;
3249
3250 if (igb_encap(txr, &m_head)) {
3251 ifp->if_oerrors++;
3252 continue;
3253 }
3254
3255 /* Send a copy of the frame to the BPF listener */
3256 ETHER_BPF_MTAP(ifp, m_head);
3257 }
3258}
3259
3260static void
3261igb_watchdog(struct ifnet *ifp)
3262{
3263 struct igb_softc *sc = ifp->if_softc;
27866bf1 3264 struct igb_tx_ring *txr = &sc->tx_rings[0];
1f7e3916
SZ		 3265
3266 ASSERT_IFNET_SERIALIZED_ALL(ifp);
3267
3268 /*
3269 * If flow control has paused us since last checking
3270 * it invalidates the watchdog timing, so dont run it.
3271 */
3272 if (sc->pause_frames) {
3273 sc->pause_frames = 0;
3274 ifp->if_timer = 5;
3275 return;
3276 }
3277
3278 if_printf(ifp, "Watchdog timeout -- resetting\n");
3279 if_printf(ifp, "Queue(%d) tdh = %d, hw tdt = %d\n", txr->me,
3280 E1000_READ_REG(&sc->hw, E1000_TDH(txr->me)),
3281 E1000_READ_REG(&sc->hw, E1000_TDT(txr->me)));
3282 if_printf(ifp, "TX(%d) desc avail = %d, "
3283 "Next TX to Clean = %d\n",
3284 txr->me, txr->tx_avail, txr->next_to_clean);
3285
3286 ifp->if_oerrors++;
3287 sc->watchdog_events++;
3288
3289 igb_init(sc);
3290 if (!ifq_is_empty(&ifp->if_snd))
3291 if_devstart(ifp);
3292}
3293
3294static void
f6167a56 3295igb_set_eitr(struct igb_softc *sc)
1f7e3916
SZ		 3296{
3297 uint32_t itr = 0;
3298
3299 if (sc->intr_rate > 0) {
3300 if (sc->hw.mac.type == e1000_82575) {
3301 itr = 1000000000 / 256 / sc->intr_rate;
3302 /*
3303 * NOTE:
3304 * Document is wrong on the 2 bits left shift
3305 */
3306 } else {
3307 itr = 1000000 / sc->intr_rate;
3308 itr <<= 2;
3309 }
3310 itr &= 0x7FFC;
3311 }
3312 if (sc->hw.mac.type == e1000_82575)
3313 itr |= itr << 16;
3314 else
3315 itr |= E1000_EITR_CNT_IGNR;
3316 E1000_WRITE_REG(&sc->hw, E1000_EITR(0), itr);
3317}
3318
3319static int
3320igb_sysctl_intr_rate(SYSCTL_HANDLER_ARGS)
3321{
3322 struct igb_softc *sc = (void *)arg1;
3323 struct ifnet *ifp = &sc->arpcom.ac_if;
3324 int error, intr_rate;
3325
3326 intr_rate = sc->intr_rate;
3327 error = sysctl_handle_int(oidp, &intr_rate, 0, req);
3328 if (error || req->newptr == NULL)
3329 return error;
3330 if (intr_rate < 0)
3331 return EINVAL;
3332
3333 ifnet_serialize_all(ifp);
3334
3335 sc->intr_rate = intr_rate;
3336 if (ifp->if_flags & IFF_RUNNING)
f6167a56 3337 igb_set_eitr(sc);
1f7e3916
SZ		 3338
3339 ifnet_deserialize_all(ifp);
3340
3341 if (bootverbose)
3342 if_printf(ifp, "Interrupt rate set to %d/sec\n", sc->intr_rate);
3343 return 0;
3344}
b6220144
SZ		 3345
3346static int
3347igb_sysctl_tx_intr_nsegs(SYSCTL_HANDLER_ARGS)
3348{
3349 struct igb_softc *sc = (void *)arg1;
3350 struct ifnet *ifp = &sc->arpcom.ac_if;
27866bf1 3351 struct igb_tx_ring *txr = &sc->tx_rings[0];
b6220144
SZ		 3352 int error, nsegs;
3353
3354 nsegs = txr->intr_nsegs;
3355 error = sysctl_handle_int(oidp, &nsegs, 0, req);
3356 if (error || req->newptr == NULL)
3357 return error;
3358 if (nsegs <= 0)
3359 return EINVAL;
3360
3361 ifnet_serialize_all(ifp);
3362
91b8700a 3363 if (nsegs >= txr->num_tx_desc - txr->oact_lo_desc ||
b6220144
SZ		 3364 nsegs >= txr->oact_hi_desc - IGB_MAX_SCATTER) {
3365 error = EINVAL;
3366 } else {
3367 error = 0;
3368 txr->intr_nsegs = nsegs;
3369 }
3370
3371 ifnet_deserialize_all(ifp);
3372
3373 return error;
3374}
f6167a56
SZ		 3375
3376static void
3377igb_init_intr(struct igb_softc *sc)
3378{
3379 if (sc->flags & IGB_FLAG_SHARED_INTR)
3380 igb_set_eitr(sc);
3381 else
3382 igb_init_unshared_intr(sc);
3383}
3384
3385static void
3386igb_init_unshared_intr(struct igb_softc *sc)
3387{
3388 struct e1000_hw *hw = &sc->hw;
3389 const struct igb_rx_ring *rxr;
3390 const struct igb_tx_ring *txr;
3391 uint32_t ivar, index;
3392 int i;
3393
3394 /*
3395 * Enable extended mode
3396 */
3397 if (sc->hw.mac.type != e1000_82575) {
3398 E1000_WRITE_REG(hw, E1000_GPIE, E1000_GPIE_NSICR);
3399 } else {
3400 uint32_t tmp;
3401
3402 tmp = E1000_READ_REG(hw, E1000_CTRL_EXT);
3403 tmp |= E1000_CTRL_EXT_IRCA;
3404 E1000_WRITE_REG(hw, E1000_CTRL_EXT, tmp);
3405 }
3406
3407 /*
3408 * Map TX/RX interrupts to EICR
3409 */
3410 switch (sc->hw.mac.type) {
3411 case e1000_82580:
3412 case e1000_i350:
3413 case e1000_vfadapt:
3414 case e1000_vfadapt_i350:
3415 /* RX entries */
27866bf1 3416 for (i = 0; i < sc->rx_ring_cnt; ++i) {
f6167a56
SZ		 3417 rxr = &sc->rx_rings[i];
3418
3419 index = i >> 1;
3420 ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
3421
3422 if (i & 1) {
3423 ivar &= 0xff00ffff;
3424 ivar |=
3425 (rxr->rx_intr_bit | E1000_IVAR_VALID) << 16;
3426 } else {
3427 ivar &= 0xffffff00;
3428 ivar |=
3429 (rxr->rx_intr_bit | E1000_IVAR_VALID);
3430 }
3431 E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
3432 }
3433 /* TX entries */
27866bf1 3434 for (i = 0; i < sc->tx_ring_cnt; ++i) {
f6167a56
SZ		 3435 txr = &sc->tx_rings[i];
3436
3437 index = i >> 1;
3438 ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
3439
3440 if (i & 1) {
3441 ivar &= 0x00ffffff;
3442 ivar |=
3443 (txr->tx_intr_bit | E1000_IVAR_VALID) << 24;
3444 } else {
3445 ivar &= 0xffff00ff;
3446 ivar |=
3447 (txr->tx_intr_bit | E1000_IVAR_VALID) << 8;
3448 }
3449 E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
3450 }
3451 /* Clear unused IVAR_MISC */
3452 E1000_WRITE_REG(hw, E1000_IVAR_MISC, 0);
3453 break;
3454
3455 case e1000_82576:
3456 /* RX entries */
27866bf1 3457 for (i = 0; i < sc->rx_ring_cnt; ++i) {
f6167a56
SZ		 3458 rxr = &sc->rx_rings[i];
3459
3460 index = i & 0x7; /* Each IVAR has two entries */
3461 ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
3462
3463 if (i < 8) {
3464 ivar &= 0xffffff00;
3465 ivar |=
3466 (rxr->rx_intr_bit | E1000_IVAR_VALID);
3467 } else {
3468 ivar &= 0xff00ffff;
3469 ivar |=
3470 (rxr->rx_intr_bit | E1000_IVAR_VALID) << 16;
3471 }
3472 E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
3473 }
3474 /* TX entries */
27866bf1 3475 for (i = 0; i < sc->tx_ring_cnt; ++i) {
f6167a56
SZ		 3476 txr = &sc->tx_rings[i];
3477
3478 index = i & 0x7; /* Each IVAR has two entries */
3479 ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
3480
3481 if (i < 8) {
3482 ivar &= 0xffff00ff;
3483 ivar |=
3484 (txr->tx_intr_bit | E1000_IVAR_VALID) << 8;
3485 } else {
3486 ivar &= 0x00ffffff;
3487 ivar |=
3488 (txr->tx_intr_bit | E1000_IVAR_VALID) << 24;
3489 }
3490 E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
3491 }
3492 /* Clear unused IVAR_MISC */
3493 E1000_WRITE_REG(hw, E1000_IVAR_MISC, 0);
3494 break;
3495
3496 case e1000_82575:
3497 /*
3498 * Enable necessary interrupt bits.
3499 *
3500 * The name of the register is confusing; in addition to
3501 * configuring the first vector of MSI-X, it also configures
3502 * which bits of EICR could be set by the hardware even when
3503 * MSI or line interrupt is used; it thus controls interrupt
3504 * generation. It MUST be configured explicitly; the default
3505 * value mentioned in the datasheet is wrong: RX queue0 and
3506 * TX queue0 are NOT enabled by default.
3507 */
3508 E1000_WRITE_REG(&sc->hw, E1000_MSIXBM(0), sc->intr_mask);
3509 break;
3510
3511 default:
3512 break;
3513 }
3514
3515 /*
3516 * Configure interrupt moderation
3517 */
3518 igb_set_eitr(sc);
3519}
3520
3521static int
3522igb_setup_intr(struct igb_softc *sc)
3523{
3524 struct ifnet *ifp = &sc->arpcom.ac_if;
8d6600da 3525 int error, i, intr_bit, intr_bitmax;
f6167a56
SZ		 3526
3527 /*
3528 * Setup interrupt mask
3529 */
8d6600da
SZ		 3530 switch (sc->hw.mac.type) {
3531 case e1000_82575:
3532 intr_bitmax = IGB_MAX_TXRXINT_82575;
3533 break;
3534 case e1000_82580:
3535 intr_bitmax = IGB_MAX_TXRXINT_82580;
3536 break;
3537 case e1000_i350:
3538 intr_bitmax = IGB_MAX_TXRXINT_I350;
3539 break;
3540 case e1000_82576:
3541 intr_bitmax = IGB_MAX_TXRXINT_82576;
3542 break;
3543 default:
3544 intr_bitmax = IGB_MIN_TXRXINT;
3545 break;
3546 }
3547 intr_bit = 0;
27866bf1 3548 for (i = 0; i < sc->tx_ring_cnt; ++i)
8d6600da 3549 igb_setup_tx_intr(&sc->tx_rings[i], &intr_bit, intr_bitmax);
27866bf1 3550 for (i = 0; i < sc->rx_ring_cnt; ++i)
8d6600da 3551 igb_setup_rx_intr(&sc->rx_rings[i], &intr_bit, intr_bitmax);
f6167a56
SZ		 3552
3553 sc->intr_mask = E1000_EICR_OTHER;
27866bf1 3554 for (i = 0; i < sc->rx_ring_cnt; ++i)
f6167a56 3555 sc->intr_mask |= sc->rx_rings[i].rx_intr_mask;
27866bf1 3556 for (i = 0; i < sc->tx_ring_cnt; ++i)
f6167a56
SZ		 3557 sc->intr_mask |= sc->tx_rings[i].tx_intr_mask;
3558
3559 if (sc->intr_type == PCI_INTR_TYPE_LEGACY) {
3560 int unshared;
3561
3562 unshared = device_getenv_int(sc->dev, "irq.unshared", 0);
3563 if (!unshared) {
3564 sc->flags |= IGB_FLAG_SHARED_INTR;
3565 if (bootverbose)
3566 device_printf(sc->dev, "IRQ shared\n");
3567 } else if (bootverbose) {
3568 device_printf(sc->dev, "IRQ unshared\n");
3569 }
3570 }
3571
3572 error = bus_setup_intr(sc->dev, sc->intr_res, INTR_MPSAFE,
3573 (sc->flags & IGB_FLAG_SHARED_INTR) ? igb_shared_intr : igb_intr,
7d235eb5 3574 sc, &sc->intr_tag, &sc->main_serialize);
f6167a56
SZ		 3575 if (error) {
3576 device_printf(sc->dev, "Failed to register interrupt handler");
3577 return error;
3578 }
3579
3580 ifp->if_cpuid = rman_get_cpuid(sc->intr_res);
3581 KKASSERT(ifp->if_cpuid >= 0 && ifp->if_cpuid < ncpus);
3582
3583 return 0;
3584}
3585
3586static void
8d6600da 3587igb_setup_tx_intr(struct igb_tx_ring *txr, int *intr_bit0, int intr_bitmax)
f6167a56
SZ		 3588{
3589 if (txr->sc->hw.mac.type == e1000_82575) {
3590 txr->tx_intr_bit = 0; /* unused */
3591 switch (txr->me) {
3592 case 0:
3593 txr->tx_intr_mask = E1000_EICR_TX_QUEUE0;
3594 break;
3595 case 1:
3596 txr->tx_intr_mask = E1000_EICR_TX_QUEUE1;
3597 break;
3598 case 2:
3599 txr->tx_intr_mask = E1000_EICR_TX_QUEUE2;
3600 break;
3601 case 3:
3602 txr->tx_intr_mask = E1000_EICR_TX_QUEUE3;
3603 break;
3604 default:
3605 panic("unsupported # of TX ring, %d\n", txr->me);
3606 }
3607 } else {
8d6600da
SZ		 3608 int intr_bit = *intr_bit0;
3609
3610 txr->tx_intr_bit = intr_bit % intr_bitmax;
f6167a56 3611 txr->tx_intr_mask = 1 << txr->tx_intr_bit;
8d6600da
SZ		 3612
3613 *intr_bit0 = intr_bit + 1;
f6167a56
SZ		 3614 }
3615}
3616
3617static void
8d6600da 3618igb_setup_rx_intr(struct igb_rx_ring *rxr, int *intr_bit0, int intr_bitmax)
f6167a56
SZ		 3619{
3620 if (rxr->sc->hw.mac.type == e1000_82575) {
3621 rxr->rx_intr_bit = 0; /* unused */
3622 switch (rxr->me) {
3623 case 0:
3624 rxr->rx_intr_mask = E1000_EICR_RX_QUEUE0;
3625 break;
3626 case 1:
3627 rxr->rx_intr_mask = E1000_EICR_RX_QUEUE1;
3628 break;
3629 case 2:
3630 rxr->rx_intr_mask = E1000_EICR_RX_QUEUE2;
3631 break;
3632 case 3:
3633 rxr->rx_intr_mask = E1000_EICR_RX_QUEUE3;
3634 break;
3635 default:
3636 panic("unsupported # of RX ring, %d\n", rxr->me);
3637 }
3638 } else {
8d6600da
SZ		 3639 int intr_bit = *intr_bit0;
3640
3641 rxr->rx_intr_bit = intr_bit % intr_bitmax;
f6167a56 3642 rxr->rx_intr_mask = 1 << rxr->rx_intr_bit;
8d6600da
SZ		 3643
3644 *intr_bit0 = intr_bit + 1;
f6167a56
SZ		 3645 }
3646}
7d235eb5
SZ		 3647
3648static void
3649igb_serialize(struct ifnet *ifp, enum ifnet_serialize slz)
3650{
3651 struct igb_softc *sc = ifp->if_softc;
3652
3653 ifnet_serialize_array_enter(sc->serializes, sc->serialize_cnt,
3654 sc->tx_serialize, sc->rx_serialize, slz);
3655}
3656
3657static void
3658igb_deserialize(struct ifnet *ifp, enum ifnet_serialize slz)
3659{
3660 struct igb_softc *sc = ifp->if_softc;
3661
3662 ifnet_serialize_array_exit(sc->serializes, sc->serialize_cnt,
3663 sc->tx_serialize, sc->rx_serialize, slz);
3664}
3665
3666static int
3667igb_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz)
3668{
3669 struct igb_softc *sc = ifp->if_softc;
3670
3671 return ifnet_serialize_array_try(sc->serializes, sc->serialize_cnt,
3672 sc->tx_serialize, sc->rx_serialize, slz);
3673}
3674
7d235eb5
SZ		 3675#ifdef INVARIANTS
3676
3677static void
3678igb_serialize_assert(struct ifnet *ifp, enum ifnet_serialize slz,
3679 boolean_t serialized)
3680{
3681 struct igb_softc *sc = ifp->if_softc;
3682
3683 ifnet_serialize_array_assert(sc->serializes, sc->serialize_cnt,
3684 sc->tx_serialize, sc->rx_serialize, slz, serialized);
3685}
3686
3687#endif /* INVARIANTS */
DragonFly Project GIT Source repo