xref: /illumos-gate/usr/src/uts/common/io/igb/igb_main.c (revision 43da549e0419e2943d8bd143ebfb1db11a16c569)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 2007-2012 Intel Corporation. All rights reserved.
24  */
25 
26 /*
27  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Copyright 2013, Nexenta Systems, Inc. All rights reserved.
29  * Copyright 2016 Joyent, Inc.
30  */
31 
32 #include "igb_sw.h"
33 
34 static char ident[] = "Intel 1Gb Ethernet";
35 static char igb_version[] = "igb 2.3.8-ish";
36 
37 /*
38  * Local function protoypes
39  */
40 static int igb_register_mac(igb_t *);
41 static int igb_identify_hardware(igb_t *);
42 static int igb_regs_map(igb_t *);
43 static void igb_init_properties(igb_t *);
44 static int igb_init_driver_settings(igb_t *);
45 static void igb_init_locks(igb_t *);
46 static void igb_destroy_locks(igb_t *);
47 static int igb_init_mac_address(igb_t *);
48 static int igb_init(igb_t *);
49 static int igb_init_adapter(igb_t *);
50 static void igb_stop_adapter(igb_t *);
51 static int igb_reset(igb_t *);
52 static void igb_tx_clean(igb_t *);
53 static boolean_t igb_tx_drain(igb_t *);
54 static boolean_t igb_rx_drain(igb_t *);
55 static int igb_alloc_rings(igb_t *);
56 static int igb_alloc_rx_data(igb_t *);
57 static void igb_free_rx_data(igb_t *);
58 static void igb_free_rings(igb_t *);
59 static void igb_setup_rings(igb_t *);
60 static void igb_setup_rx(igb_t *);
61 static void igb_setup_tx(igb_t *);
62 static void igb_setup_rx_ring(igb_rx_ring_t *);
63 static void igb_setup_tx_ring(igb_tx_ring_t *);
64 static void igb_setup_rss(igb_t *);
65 static void igb_setup_mac_rss_classify(igb_t *);
66 static void igb_setup_mac_classify(igb_t *);
67 static void igb_init_unicst(igb_t *);
68 static void igb_setup_multicst(igb_t *);
69 static void igb_get_phy_state(igb_t *);
70 static void igb_param_sync(igb_t *);
71 static void igb_get_conf(igb_t *);
72 static int igb_get_prop(igb_t *, char *, int, int, int);
73 static boolean_t igb_is_link_up(igb_t *);
74 static boolean_t igb_link_check(igb_t *);
75 static void igb_local_timer(void *);
76 static void igb_link_timer(void *);
77 static void igb_arm_watchdog_timer(igb_t *);
78 static void igb_start_watchdog_timer(igb_t *);
79 static void igb_restart_watchdog_timer(igb_t *);
80 static void igb_stop_watchdog_timer(igb_t *);
81 static void igb_start_link_timer(igb_t *);
82 static void igb_stop_link_timer(igb_t *);
83 static void igb_disable_adapter_interrupts(igb_t *);
84 static void igb_enable_adapter_interrupts_82575(igb_t *);
85 static void igb_enable_adapter_interrupts_82576(igb_t *);
86 static void igb_enable_adapter_interrupts_82580(igb_t *);
87 static boolean_t is_valid_mac_addr(uint8_t *);
88 static boolean_t igb_stall_check(igb_t *);
89 static boolean_t igb_set_loopback_mode(igb_t *, uint32_t);
90 static void igb_set_external_loopback(igb_t *);
91 static void igb_set_internal_phy_loopback(igb_t *);
92 static void igb_set_internal_serdes_loopback(igb_t *);
93 static boolean_t igb_find_mac_address(igb_t *);
94 static int igb_alloc_intrs(igb_t *);
95 static int igb_alloc_intr_handles(igb_t *, int);
96 static int igb_add_intr_handlers(igb_t *);
97 static void igb_rem_intr_handlers(igb_t *);
98 static void igb_rem_intrs(igb_t *);
99 static int igb_enable_intrs(igb_t *);
100 static int igb_disable_intrs(igb_t *);
101 static void igb_setup_msix_82575(igb_t *);
102 static void igb_setup_msix_82576(igb_t *);
103 static void igb_setup_msix_82580(igb_t *);
104 static uint_t igb_intr_legacy(void *, void *);
105 static uint_t igb_intr_msi(void *, void *);
106 static uint_t igb_intr_rx(void *, void *);
107 static uint_t igb_intr_tx(void *, void *);
108 static uint_t igb_intr_tx_other(void *, void *);
109 static void igb_intr_rx_work(igb_rx_ring_t *);
110 static void igb_intr_tx_work(igb_tx_ring_t *);
111 static void igb_intr_link_work(igb_t *);
112 static void igb_get_driver_control(struct e1000_hw *);
113 static void igb_release_driver_control(struct e1000_hw *);
114 
115 static int igb_attach(dev_info_t *, ddi_attach_cmd_t);
116 static int igb_detach(dev_info_t *, ddi_detach_cmd_t);
117 static int igb_resume(dev_info_t *);
118 static int igb_suspend(dev_info_t *);
119 static int igb_quiesce(dev_info_t *);
120 static void igb_unconfigure(dev_info_t *, igb_t *);
121 static int igb_fm_error_cb(dev_info_t *, ddi_fm_error_t *,
122     const void *);
123 static void igb_fm_init(igb_t *);
124 static void igb_fm_fini(igb_t *);
125 static void igb_release_multicast(igb_t *);
126 
127 char *igb_priv_props[] = {
128 	"_eee_support",
129 	"_tx_copy_thresh",
130 	"_tx_recycle_thresh",
131 	"_tx_overload_thresh",
132 	"_tx_resched_thresh",
133 	"_rx_copy_thresh",
134 	"_rx_limit_per_intr",
135 	"_intr_throttling",
136 	"_adv_pause_cap",
137 	"_adv_asym_pause_cap",
138 	NULL
139 };
140 
141 static struct cb_ops igb_cb_ops = {
142 	nulldev,		/* cb_open */
143 	nulldev,		/* cb_close */
144 	nodev,			/* cb_strategy */
145 	nodev,			/* cb_print */
146 	nodev,			/* cb_dump */
147 	nodev,			/* cb_read */
148 	nodev,			/* cb_write */
149 	nodev,			/* cb_ioctl */
150 	nodev,			/* cb_devmap */
151 	nodev,			/* cb_mmap */
152 	nodev,			/* cb_segmap */
153 	nochpoll,		/* cb_chpoll */
154 	ddi_prop_op,		/* cb_prop_op */
155 	NULL,			/* cb_stream */
156 	D_MP | D_HOTPLUG,	/* cb_flag */
157 	CB_REV,			/* cb_rev */
158 	nodev,			/* cb_aread */
159 	nodev			/* cb_awrite */
160 };
161 
162 static struct dev_ops igb_dev_ops = {
163 	DEVO_REV,		/* devo_rev */
164 	0,			/* devo_refcnt */
165 	NULL,			/* devo_getinfo */
166 	nulldev,		/* devo_identify */
167 	nulldev,		/* devo_probe */
168 	igb_attach,		/* devo_attach */
169 	igb_detach,		/* devo_detach */
170 	nodev,			/* devo_reset */
171 	&igb_cb_ops,		/* devo_cb_ops */
172 	NULL,			/* devo_bus_ops */
173 	ddi_power,		/* devo_power */
174 	igb_quiesce,	/* devo_quiesce */
175 };
176 
177 static struct modldrv igb_modldrv = {
178 	&mod_driverops,		/* Type of module.  This one is a driver */
179 	ident,			/* Discription string */
180 	&igb_dev_ops,		/* driver ops */
181 };
182 
183 static struct modlinkage igb_modlinkage = {
184 	MODREV_1, &igb_modldrv, NULL
185 };
186 
187 /* Access attributes for register mapping */
188 ddi_device_acc_attr_t igb_regs_acc_attr = {
189 	DDI_DEVICE_ATTR_V1,
190 	DDI_STRUCTURE_LE_ACC,
191 	DDI_STRICTORDER_ACC,
192 	DDI_FLAGERR_ACC
193 };
194 
195 #define	IGB_M_CALLBACK_FLAGS \
196 	(MC_IOCTL | MC_GETCAPAB | MC_SETPROP | MC_GETPROP | MC_PROPINFO)
197 
198 static mac_callbacks_t igb_m_callbacks = {
199 	IGB_M_CALLBACK_FLAGS,
200 	igb_m_stat,
201 	igb_m_start,
202 	igb_m_stop,
203 	igb_m_promisc,
204 	igb_m_multicst,
205 	NULL,
206 	NULL,
207 	NULL,
208 	igb_m_ioctl,
209 	igb_m_getcapab,
210 	NULL,
211 	NULL,
212 	igb_m_setprop,
213 	igb_m_getprop,
214 	igb_m_propinfo
215 };
216 
217 /*
218  * Initialize capabilities of each supported adapter type
219  */
220 static adapter_info_t igb_82575_cap = {
221 	/* limits */
222 	4,		/* maximum number of rx queues */
223 	1,		/* minimum number of rx queues */
224 	4,		/* default number of rx queues */
225 	4,		/* maximum number of tx queues */
226 	1,		/* minimum number of tx queues */
227 	4,		/* default number of tx queues */
228 	65535,		/* maximum interrupt throttle rate */
229 	0,		/* minimum interrupt throttle rate */
230 	200,		/* default interrupt throttle rate */
231 
232 	/* function pointers */
233 	igb_enable_adapter_interrupts_82575,
234 	igb_setup_msix_82575,
235 
236 	/* capabilities */
237 	(IGB_FLAG_HAS_DCA |	/* capability flags */
238 	IGB_FLAG_VMDQ_POOL),
239 
240 	0xffc00000		/* mask for RXDCTL register */
241 };
242 
243 static adapter_info_t igb_82576_cap = {
244 	/* limits */
245 	16,		/* maximum number of rx queues */
246 	1,		/* minimum number of rx queues */
247 	4,		/* default number of rx queues */
248 	16,		/* maximum number of tx queues */
249 	1,		/* minimum number of tx queues */
250 	4,		/* default number of tx queues */
251 	65535,		/* maximum interrupt throttle rate */
252 	0,		/* minimum interrupt throttle rate */
253 	200,		/* default interrupt throttle rate */
254 
255 	/* function pointers */
256 	igb_enable_adapter_interrupts_82576,
257 	igb_setup_msix_82576,
258 
259 	/* capabilities */
260 	(IGB_FLAG_HAS_DCA |	/* capability flags */
261 	IGB_FLAG_VMDQ_POOL |
262 	IGB_FLAG_NEED_CTX_IDX),
263 
264 	0xffe00000		/* mask for RXDCTL register */
265 };
266 
267 static adapter_info_t igb_82580_cap = {
268 	/* limits */
269 	8,		/* maximum number of rx queues */
270 	1,		/* minimum number of rx queues */
271 	4,		/* default number of rx queues */
272 	8,		/* maximum number of tx queues */
273 	1,		/* minimum number of tx queues */
274 	4,		/* default number of tx queues */
275 	65535,		/* maximum interrupt throttle rate */
276 	0,		/* minimum interrupt throttle rate */
277 	200,		/* default interrupt throttle rate */
278 
279 	/* function pointers */
280 	igb_enable_adapter_interrupts_82580,
281 	igb_setup_msix_82580,
282 
283 	/* capabilities */
284 	(IGB_FLAG_HAS_DCA |	/* capability flags */
285 	IGB_FLAG_VMDQ_POOL |
286 	IGB_FLAG_NEED_CTX_IDX),
287 
288 	0xffe00000		/* mask for RXDCTL register */
289 };
290 
291 static adapter_info_t igb_i350_cap = {
292 	/* limits */
293 	8,		/* maximum number of rx queues */
294 	1,		/* minimum number of rx queues */
295 	4,		/* default number of rx queues */
296 	8,		/* maximum number of tx queues */
297 	1,		/* minimum number of tx queues */
298 	4,		/* default number of tx queues */
299 	65535,		/* maximum interrupt throttle rate */
300 	0,		/* minimum interrupt throttle rate */
301 	200,		/* default interrupt throttle rate */
302 
303 	/* function pointers */
304 	igb_enable_adapter_interrupts_82580,
305 	igb_setup_msix_82580,
306 
307 	/* capabilities */
308 	(IGB_FLAG_HAS_DCA |	/* capability flags */
309 	IGB_FLAG_VMDQ_POOL |
310 	IGB_FLAG_NEED_CTX_IDX),
311 
312 	0xffe00000		/* mask for RXDCTL register */
313 };
314 
315 static adapter_info_t igb_i210_cap = {
316 	/* limits */
317 	4,		/* maximum number of rx queues */
318 	1,		/* minimum number of rx queues */
319 	4,		/* default number of rx queues */
320 	4,		/* maximum number of tx queues */
321 	1,		/* minimum number of tx queues */
322 	4,		/* default number of tx queues */
323 	65535,		/* maximum interrupt throttle rate */
324 	0,		/* minimum interrupt throttle rate */
325 	200,		/* default interrupt throttle rate */
326 
327 	/* function pointers */
328 	igb_enable_adapter_interrupts_82580,
329 	igb_setup_msix_82580,
330 
331 	/* capabilities */
332 	(IGB_FLAG_HAS_DCA |	/* capability flags */
333 	IGB_FLAG_VMDQ_POOL |
334 	IGB_FLAG_NEED_CTX_IDX),
335 
336 	0xfff00000		/* mask for RXDCTL register */
337 };
338 
339 static adapter_info_t igb_i354_cap = {
340 	/* limits */
341 	8,		/* maximum number of rx queues */
342 	1,		/* minimum number of rx queues */
343 	4,		/* default number of rx queues */
344 	8,		/* maximum number of tx queues */
345 	1,		/* minimum number of tx queues */
346 	4,		/* default number of tx queues */
347 	65535,		/* maximum interrupt throttle rate */
348 	0,		/* minimum interrupt throttle rate */
349 	200,		/* default interrupt throttle rate */
350 
351 	/* function pointers */
352 	igb_enable_adapter_interrupts_82580,
353 	igb_setup_msix_82580,
354 
355 	/* capabilities */
356 	(IGB_FLAG_HAS_DCA |	/* capability flags */
357 	IGB_FLAG_VMDQ_POOL |
358 	IGB_FLAG_NEED_CTX_IDX),
359 
360 	0xfff00000		/* mask for RXDCTL register */
361 };
362 
363 /*
364  * Module Initialization Functions
365  */
366 
367 int
368 _init(void)
369 {
370 	int status;
371 
372 	mac_init_ops(&igb_dev_ops, MODULE_NAME);
373 
374 	status = mod_install(&igb_modlinkage);
375 
376 	if (status != DDI_SUCCESS) {
377 		mac_fini_ops(&igb_dev_ops);
378 	}
379 
380 	return (status);
381 }
382 
383 int
384 _fini(void)
385 {
386 	int status;
387 
388 	status = mod_remove(&igb_modlinkage);
389 
390 	if (status == DDI_SUCCESS) {
391 		mac_fini_ops(&igb_dev_ops);
392 	}
393 
394 	return (status);
395 
396 }
397 
398 int
399 _info(struct modinfo *modinfop)
400 {
401 	int status;
402 
403 	status = mod_info(&igb_modlinkage, modinfop);
404 
405 	return (status);
406 }
407 
408 /*
409  * igb_attach - driver attach
410  *
411  * This function is the device specific initialization entry
412  * point. This entry point is required and must be written.
413  * The DDI_ATTACH command must be provided in the attach entry
414  * point. When attach() is called with cmd set to DDI_ATTACH,
415  * all normal kernel services (such as kmem_alloc(9F)) are
416  * available for use by the driver.
417  *
418  * The attach() function will be called once for each instance
419  * of  the  device  on  the  system with cmd set to DDI_ATTACH.
420  * Until attach() succeeds, the only driver entry points which
421  * may be called are open(9E) and getinfo(9E).
422  */
423 static int
424 igb_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
425 {
426 	igb_t *igb;
427 	struct igb_osdep *osdep;
428 	struct e1000_hw *hw;
429 	int instance;
430 
431 	/*
432 	 * Check the command and perform corresponding operations
433 	 */
434 	switch (cmd) {
435 	default:
436 		return (DDI_FAILURE);
437 
438 	case DDI_RESUME:
439 		return (igb_resume(devinfo));
440 
441 	case DDI_ATTACH:
442 		break;
443 	}
444 
445 	/* Get the device instance */
446 	instance = ddi_get_instance(devinfo);
447 
448 	/* Allocate memory for the instance data structure */
449 	igb = kmem_zalloc(sizeof (igb_t), KM_SLEEP);
450 
451 	igb->dip = devinfo;
452 	igb->instance = instance;
453 
454 	hw = &igb->hw;
455 	osdep = &igb->osdep;
456 	hw->back = osdep;
457 	osdep->igb = igb;
458 
459 	/* Attach the instance pointer to the dev_info data structure */
460 	ddi_set_driver_private(devinfo, igb);
461 
462 
463 	/* Initialize for fma support */
464 	igb->fm_capabilities = igb_get_prop(igb, "fm-capable",
465 	    0, 0x0f,
466 	    DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
467 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
468 	igb_fm_init(igb);
469 	igb->attach_progress |= ATTACH_PROGRESS_FMINIT;
470 
471 	/*
472 	 * Map PCI config space registers
473 	 */
474 	if (pci_config_setup(devinfo, &osdep->cfg_handle) != DDI_SUCCESS) {
475 		igb_error(igb, "Failed to map PCI configurations");
476 		goto attach_fail;
477 	}
478 	igb->attach_progress |= ATTACH_PROGRESS_PCI_CONFIG;
479 
480 	/*
481 	 * Identify the chipset family
482 	 */
483 	if (igb_identify_hardware(igb) != IGB_SUCCESS) {
484 		igb_error(igb, "Failed to identify hardware");
485 		goto attach_fail;
486 	}
487 
488 	/*
489 	 * Map device registers
490 	 */
491 	if (igb_regs_map(igb) != IGB_SUCCESS) {
492 		igb_error(igb, "Failed to map device registers");
493 		goto attach_fail;
494 	}
495 	igb->attach_progress |= ATTACH_PROGRESS_REGS_MAP;
496 
497 	/*
498 	 * Initialize driver parameters
499 	 */
500 	igb_init_properties(igb);
501 	igb->attach_progress |= ATTACH_PROGRESS_PROPS;
502 
503 	/*
504 	 * Allocate interrupts
505 	 */
506 	if (igb_alloc_intrs(igb) != IGB_SUCCESS) {
507 		igb_error(igb, "Failed to allocate interrupts");
508 		goto attach_fail;
509 	}
510 	igb->attach_progress |= ATTACH_PROGRESS_ALLOC_INTR;
511 
512 	/*
513 	 * Allocate rx/tx rings based on the ring numbers.
514 	 * The actual numbers of rx/tx rings are decided by the number of
515 	 * allocated interrupt vectors, so we should allocate the rings after
516 	 * interrupts are allocated.
517 	 */
518 	if (igb_alloc_rings(igb) != IGB_SUCCESS) {
519 		igb_error(igb, "Failed to allocate rx/tx rings or groups");
520 		goto attach_fail;
521 	}
522 	igb->attach_progress |= ATTACH_PROGRESS_ALLOC_RINGS;
523 
524 	/*
525 	 * Add interrupt handlers
526 	 */
527 	if (igb_add_intr_handlers(igb) != IGB_SUCCESS) {
528 		igb_error(igb, "Failed to add interrupt handlers");
529 		goto attach_fail;
530 	}
531 	igb->attach_progress |= ATTACH_PROGRESS_ADD_INTR;
532 
533 	/*
534 	 * Initialize driver parameters
535 	 */
536 	if (igb_init_driver_settings(igb) != IGB_SUCCESS) {
537 		igb_error(igb, "Failed to initialize driver settings");
538 		goto attach_fail;
539 	}
540 
541 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK) {
542 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
543 		goto attach_fail;
544 	}
545 
546 	/*
547 	 * Initialize mutexes for this device.
548 	 * Do this before enabling the interrupt handler and
549 	 * register the softint to avoid the condition where
550 	 * interrupt handler can try using uninitialized mutex
551 	 */
552 	igb_init_locks(igb);
553 	igb->attach_progress |= ATTACH_PROGRESS_LOCKS;
554 
555 	/*
556 	 * Initialize the adapter
557 	 */
558 	if (igb_init(igb) != IGB_SUCCESS) {
559 		igb_error(igb, "Failed to initialize adapter");
560 		goto attach_fail;
561 	}
562 	igb->attach_progress |= ATTACH_PROGRESS_INIT_ADAPTER;
563 
564 	/*
565 	 * Initialize statistics
566 	 */
567 	if (igb_init_stats(igb) != IGB_SUCCESS) {
568 		igb_error(igb, "Failed to initialize statistics");
569 		goto attach_fail;
570 	}
571 	igb->attach_progress |= ATTACH_PROGRESS_STATS;
572 
573 	/*
574 	 * Register the driver to the MAC
575 	 */
576 	if (igb_register_mac(igb) != IGB_SUCCESS) {
577 		igb_error(igb, "Failed to register MAC");
578 		goto attach_fail;
579 	}
580 	igb->attach_progress |= ATTACH_PROGRESS_MAC;
581 
582 	/*
583 	 * Now that mutex locks are initialized, and the chip is also
584 	 * initialized, enable interrupts.
585 	 */
586 	if (igb_enable_intrs(igb) != IGB_SUCCESS) {
587 		igb_error(igb, "Failed to enable DDI interrupts");
588 		goto attach_fail;
589 	}
590 	igb->attach_progress |= ATTACH_PROGRESS_ENABLE_INTR;
591 
592 	igb_log(igb, "%s", igb_version);
593 	atomic_or_32(&igb->igb_state, IGB_INITIALIZED);
594 
595 	/*
596 	 * Newer models have Energy Efficient Ethernet, let's disable this by
597 	 * default.
598 	 */
599 	if (igb->hw.mac.type == e1000_i350)
600 		(void) e1000_set_eee_i350(&igb->hw, B_FALSE, B_FALSE);
601 	else if (igb->hw.mac.type == e1000_i354)
602 		(void) e1000_set_eee_i354(&igb->hw, B_FALSE, B_FALSE);
603 
604 	return (DDI_SUCCESS);
605 
606 attach_fail:
607 	igb_unconfigure(devinfo, igb);
608 	return (DDI_FAILURE);
609 }
610 
611 /*
612  * igb_detach - driver detach
613  *
614  * The detach() function is the complement of the attach routine.
615  * If cmd is set to DDI_DETACH, detach() is used to remove  the
616  * state  associated  with  a  given  instance of a device node
617  * prior to the removal of that instance from the system.
618  *
619  * The detach() function will be called once for each  instance
620  * of the device for which there has been a successful attach()
621  * once there are no longer  any  opens  on  the  device.
622  *
623  * Interrupts routine are disabled, All memory allocated by this
624  * driver are freed.
625  */
626 static int
627 igb_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
628 {
629 	igb_t *igb;
630 
631 	/*
632 	 * Check detach command
633 	 */
634 	switch (cmd) {
635 	default:
636 		return (DDI_FAILURE);
637 
638 	case DDI_SUSPEND:
639 		return (igb_suspend(devinfo));
640 
641 	case DDI_DETACH:
642 		break;
643 	}
644 
645 
646 	/*
647 	 * Get the pointer to the driver private data structure
648 	 */
649 	igb = (igb_t *)ddi_get_driver_private(devinfo);
650 	if (igb == NULL)
651 		return (DDI_FAILURE);
652 
653 	/*
654 	 * Unregister MAC. If failed, we have to fail the detach
655 	 */
656 	if (mac_unregister(igb->mac_hdl) != 0) {
657 		igb_error(igb, "Failed to unregister MAC");
658 		return (DDI_FAILURE);
659 	}
660 	igb->attach_progress &= ~ATTACH_PROGRESS_MAC;
661 
662 	/*
663 	 * If the device is still running, it needs to be stopped first.
664 	 * This check is necessary because under some specific circumstances,
665 	 * the detach routine can be called without stopping the interface
666 	 * first.
667 	 */
668 	mutex_enter(&igb->gen_lock);
669 	if (igb->igb_state & IGB_STARTED) {
670 		atomic_and_32(&igb->igb_state, ~IGB_STARTED);
671 		igb_stop(igb, B_TRUE);
672 		mutex_exit(&igb->gen_lock);
673 		/* Disable and stop the watchdog timer */
674 		igb_disable_watchdog_timer(igb);
675 	} else
676 		mutex_exit(&igb->gen_lock);
677 
678 	/*
679 	 * Check if there are still rx buffers held by the upper layer.
680 	 * If so, fail the detach.
681 	 */
682 	if (!igb_rx_drain(igb))
683 		return (DDI_FAILURE);
684 
685 	/*
686 	 * Do the remaining unconfigure routines
687 	 */
688 	igb_unconfigure(devinfo, igb);
689 
690 	return (DDI_SUCCESS);
691 }
692 
693 /*
694  * quiesce(9E) entry point.
695  *
696  * This function is called when the system is single-threaded at high
697  * PIL with preemption disabled. Therefore, this function must not be
698  * blocked.
699  *
700  * This function returns DDI_SUCCESS on success, or DDI_FAILURE on failure.
701  * DDI_FAILURE indicates an error condition and should almost never happen.
702  */
703 static int
704 igb_quiesce(dev_info_t *devinfo)
705 {
706 	igb_t *igb;
707 	struct e1000_hw *hw;
708 
709 	igb = (igb_t *)ddi_get_driver_private(devinfo);
710 
711 	if (igb == NULL)
712 		return (DDI_FAILURE);
713 
714 	hw = &igb->hw;
715 
716 	/*
717 	 * Disable the adapter interrupts
718 	 */
719 	igb_disable_adapter_interrupts(igb);
720 
721 	/* Tell firmware driver is no longer in control */
722 	igb_release_driver_control(hw);
723 
724 	/*
725 	 * Reset the chipset
726 	 */
727 	(void) e1000_reset_hw(hw);
728 
729 	/*
730 	 * Reset PHY if possible
731 	 */
732 	if (e1000_check_reset_block(hw) == E1000_SUCCESS)
733 		(void) e1000_phy_hw_reset(hw);
734 
735 	return (DDI_SUCCESS);
736 }
737 
738 /*
739  * igb_unconfigure - release all resources held by this instance
740  */
741 static void
742 igb_unconfigure(dev_info_t *devinfo, igb_t *igb)
743 {
744 	/*
745 	 * Disable interrupt
746 	 */
747 	if (igb->attach_progress & ATTACH_PROGRESS_ENABLE_INTR) {
748 		(void) igb_disable_intrs(igb);
749 	}
750 
751 	/*
752 	 * Unregister MAC
753 	 */
754 	if (igb->attach_progress & ATTACH_PROGRESS_MAC) {
755 		(void) mac_unregister(igb->mac_hdl);
756 	}
757 
758 	/*
759 	 * Free statistics
760 	 */
761 	if (igb->attach_progress & ATTACH_PROGRESS_STATS) {
762 		kstat_delete((kstat_t *)igb->igb_ks);
763 	}
764 
765 	/*
766 	 * Remove interrupt handlers
767 	 */
768 	if (igb->attach_progress & ATTACH_PROGRESS_ADD_INTR) {
769 		igb_rem_intr_handlers(igb);
770 	}
771 
772 	/*
773 	 * Remove interrupts
774 	 */
775 	if (igb->attach_progress & ATTACH_PROGRESS_ALLOC_INTR) {
776 		igb_rem_intrs(igb);
777 	}
778 
779 	/*
780 	 * Remove driver properties
781 	 */
782 	if (igb->attach_progress & ATTACH_PROGRESS_PROPS) {
783 		(void) ddi_prop_remove_all(devinfo);
784 	}
785 
786 	/*
787 	 * Stop the adapter
788 	 */
789 	if (igb->attach_progress & ATTACH_PROGRESS_INIT_ADAPTER) {
790 		mutex_enter(&igb->gen_lock);
791 		igb_stop_adapter(igb);
792 		mutex_exit(&igb->gen_lock);
793 		if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
794 			ddi_fm_service_impact(igb->dip, DDI_SERVICE_UNAFFECTED);
795 	}
796 
797 	/*
798 	 * Free multicast table
799 	 */
800 	igb_release_multicast(igb);
801 
802 	/*
803 	 * Free register handle
804 	 */
805 	if (igb->attach_progress & ATTACH_PROGRESS_REGS_MAP) {
806 		if (igb->osdep.reg_handle != NULL)
807 			ddi_regs_map_free(&igb->osdep.reg_handle);
808 	}
809 
810 	/*
811 	 * Free PCI config handle
812 	 */
813 	if (igb->attach_progress & ATTACH_PROGRESS_PCI_CONFIG) {
814 		if (igb->osdep.cfg_handle != NULL)
815 			pci_config_teardown(&igb->osdep.cfg_handle);
816 	}
817 
818 	/*
819 	 * Free locks
820 	 */
821 	if (igb->attach_progress & ATTACH_PROGRESS_LOCKS) {
822 		igb_destroy_locks(igb);
823 	}
824 
825 	/*
826 	 * Free the rx/tx rings
827 	 */
828 	if (igb->attach_progress & ATTACH_PROGRESS_ALLOC_RINGS) {
829 		igb_free_rings(igb);
830 	}
831 
832 	/*
833 	 * Remove FMA
834 	 */
835 	if (igb->attach_progress & ATTACH_PROGRESS_FMINIT) {
836 		igb_fm_fini(igb);
837 	}
838 
839 	/*
840 	 * Free the driver data structure
841 	 */
842 	kmem_free(igb, sizeof (igb_t));
843 
844 	ddi_set_driver_private(devinfo, NULL);
845 }
846 
847 /*
848  * igb_register_mac - Register the driver and its function pointers with
849  * the GLD interface
850  */
851 static int
852 igb_register_mac(igb_t *igb)
853 {
854 	struct e1000_hw *hw = &igb->hw;
855 	mac_register_t *mac;
856 	int status;
857 
858 	if ((mac = mac_alloc(MAC_VERSION)) == NULL)
859 		return (IGB_FAILURE);
860 
861 	mac->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
862 	mac->m_driver = igb;
863 	mac->m_dip = igb->dip;
864 	mac->m_src_addr = hw->mac.addr;
865 	mac->m_callbacks = &igb_m_callbacks;
866 	mac->m_min_sdu = 0;
867 	mac->m_max_sdu = igb->max_frame_size -
868 	    sizeof (struct ether_vlan_header) - ETHERFCSL;
869 	mac->m_margin = VLAN_TAGSZ;
870 	mac->m_priv_props = igb_priv_props;
871 	mac->m_v12n = MAC_VIRT_LEVEL1;
872 
873 	status = mac_register(mac, &igb->mac_hdl);
874 
875 	mac_free(mac);
876 
877 	return ((status == 0) ? IGB_SUCCESS : IGB_FAILURE);
878 }
879 
880 /*
881  * igb_identify_hardware - Identify the type of the chipset
882  */
883 static int
884 igb_identify_hardware(igb_t *igb)
885 {
886 	struct e1000_hw *hw = &igb->hw;
887 	struct igb_osdep *osdep = &igb->osdep;
888 
889 	/*
890 	 * Get the device id
891 	 */
892 	hw->vendor_id =
893 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_VENID);
894 	hw->device_id =
895 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_DEVID);
896 	hw->revision_id =
897 	    pci_config_get8(osdep->cfg_handle, PCI_CONF_REVID);
898 	hw->subsystem_device_id =
899 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBSYSID);
900 	hw->subsystem_vendor_id =
901 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBVENID);
902 
903 	/*
904 	 * Set the mac type of the adapter based on the device id
905 	 */
906 	if (e1000_set_mac_type(hw) != E1000_SUCCESS) {
907 		return (IGB_FAILURE);
908 	}
909 
910 	/*
911 	 * Install adapter capabilities based on mac type
912 	 */
913 	switch (hw->mac.type) {
914 	case e1000_82575:
915 		igb->capab = &igb_82575_cap;
916 		break;
917 	case e1000_82576:
918 		igb->capab = &igb_82576_cap;
919 		break;
920 	case e1000_82580:
921 		igb->capab = &igb_82580_cap;
922 		break;
923 	case e1000_i350:
924 		igb->capab = &igb_i350_cap;
925 		break;
926 	case e1000_i210:
927 	case e1000_i211:
928 		igb->capab = &igb_i210_cap;
929 		break;
930 	case e1000_i354:
931 		igb->capab = &igb_i354_cap;
932 		break;
933 	default:
934 		return (IGB_FAILURE);
935 	}
936 
937 	return (IGB_SUCCESS);
938 }
939 
940 /*
941  * igb_regs_map - Map the device registers
942  */
943 static int
944 igb_regs_map(igb_t *igb)
945 {
946 	dev_info_t *devinfo = igb->dip;
947 	struct e1000_hw *hw = &igb->hw;
948 	struct igb_osdep *osdep = &igb->osdep;
949 	off_t mem_size;
950 
951 	/*
952 	 * First get the size of device registers to be mapped.
953 	 */
954 	if (ddi_dev_regsize(devinfo, IGB_ADAPTER_REGSET, &mem_size) !=
955 	    DDI_SUCCESS) {
956 		return (IGB_FAILURE);
957 	}
958 
959 	/*
960 	 * Call ddi_regs_map_setup() to map registers
961 	 */
962 	if ((ddi_regs_map_setup(devinfo, IGB_ADAPTER_REGSET,
963 	    (caddr_t *)&hw->hw_addr, 0,
964 	    mem_size, &igb_regs_acc_attr,
965 	    &osdep->reg_handle)) != DDI_SUCCESS) {
966 		return (IGB_FAILURE);
967 	}
968 
969 	return (IGB_SUCCESS);
970 }
971 
972 /*
973  * igb_init_properties - Initialize driver properties
974  */
975 static void
976 igb_init_properties(igb_t *igb)
977 {
978 	/*
979 	 * Get conf file properties, including link settings
980 	 * jumbo frames, ring number, descriptor number, etc.
981 	 */
982 	igb_get_conf(igb);
983 }
984 
985 /*
986  * igb_init_driver_settings - Initialize driver settings
987  *
988  * The settings include hardware function pointers, bus information,
989  * rx/tx rings settings, link state, and any other parameters that
990  * need to be setup during driver initialization.
991  */
992 static int
993 igb_init_driver_settings(igb_t *igb)
994 {
995 	struct e1000_hw *hw = &igb->hw;
996 	igb_rx_ring_t *rx_ring;
997 	igb_tx_ring_t *tx_ring;
998 	uint32_t rx_size;
999 	uint32_t tx_size;
1000 	int i;
1001 
1002 	/*
1003 	 * Initialize chipset specific hardware function pointers
1004 	 */
1005 	if (e1000_setup_init_funcs(hw, B_TRUE) != E1000_SUCCESS) {
1006 		return (IGB_FAILURE);
1007 	}
1008 
1009 	/*
1010 	 * Get bus information
1011 	 */
1012 	if (e1000_get_bus_info(hw) != E1000_SUCCESS) {
1013 		return (IGB_FAILURE);
1014 	}
1015 
1016 	/*
1017 	 * Get the system page size
1018 	 */
1019 	igb->page_size = ddi_ptob(igb->dip, (ulong_t)1);
1020 
1021 	/*
1022 	 * Set rx buffer size
1023 	 * The IP header alignment room is counted in the calculation.
1024 	 * The rx buffer size is in unit of 1K that is required by the
1025 	 * chipset hardware.
1026 	 */
1027 	rx_size = igb->max_frame_size + IPHDR_ALIGN_ROOM;
1028 	igb->rx_buf_size = ((rx_size >> 10) +
1029 	    ((rx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
1030 
1031 	/*
1032 	 * Set tx buffer size
1033 	 */
1034 	tx_size = igb->max_frame_size;
1035 	igb->tx_buf_size = ((tx_size >> 10) +
1036 	    ((tx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
1037 
1038 	/*
1039 	 * Initialize rx/tx rings parameters
1040 	 */
1041 	for (i = 0; i < igb->num_rx_rings; i++) {
1042 		rx_ring = &igb->rx_rings[i];
1043 		rx_ring->index = i;
1044 		rx_ring->igb = igb;
1045 	}
1046 
1047 	for (i = 0; i < igb->num_tx_rings; i++) {
1048 		tx_ring = &igb->tx_rings[i];
1049 		tx_ring->index = i;
1050 		tx_ring->igb = igb;
1051 		if (igb->tx_head_wb_enable)
1052 			tx_ring->tx_recycle = igb_tx_recycle_head_wb;
1053 		else
1054 			tx_ring->tx_recycle = igb_tx_recycle_legacy;
1055 
1056 		tx_ring->ring_size = igb->tx_ring_size;
1057 		tx_ring->free_list_size = igb->tx_ring_size +
1058 		    (igb->tx_ring_size >> 1);
1059 	}
1060 
1061 	/*
1062 	 * Initialize values of interrupt throttling rates
1063 	 */
1064 	for (i = 1; i < MAX_NUM_EITR; i++)
1065 		igb->intr_throttling[i] = igb->intr_throttling[0];
1066 
1067 	/*
1068 	 * The initial link state should be "unknown"
1069 	 */
1070 	igb->link_state = LINK_STATE_UNKNOWN;
1071 
1072 	return (IGB_SUCCESS);
1073 }
1074 
1075 /*
1076  * igb_init_locks - Initialize locks
1077  */
1078 static void
1079 igb_init_locks(igb_t *igb)
1080 {
1081 	igb_rx_ring_t *rx_ring;
1082 	igb_tx_ring_t *tx_ring;
1083 	int i;
1084 
1085 	for (i = 0; i < igb->num_rx_rings; i++) {
1086 		rx_ring = &igb->rx_rings[i];
1087 		mutex_init(&rx_ring->rx_lock, NULL,
1088 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1089 	}
1090 
1091 	for (i = 0; i < igb->num_tx_rings; i++) {
1092 		tx_ring = &igb->tx_rings[i];
1093 		mutex_init(&tx_ring->tx_lock, NULL,
1094 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1095 		mutex_init(&tx_ring->recycle_lock, NULL,
1096 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1097 		mutex_init(&tx_ring->tcb_head_lock, NULL,
1098 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1099 		mutex_init(&tx_ring->tcb_tail_lock, NULL,
1100 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1101 	}
1102 
1103 	mutex_init(&igb->gen_lock, NULL,
1104 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1105 
1106 	mutex_init(&igb->watchdog_lock, NULL,
1107 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1108 
1109 	mutex_init(&igb->link_lock, NULL,
1110 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1111 }
1112 
1113 /*
1114  * igb_destroy_locks - Destroy locks
1115  */
1116 static void
1117 igb_destroy_locks(igb_t *igb)
1118 {
1119 	igb_rx_ring_t *rx_ring;
1120 	igb_tx_ring_t *tx_ring;
1121 	int i;
1122 
1123 	for (i = 0; i < igb->num_rx_rings; i++) {
1124 		rx_ring = &igb->rx_rings[i];
1125 		mutex_destroy(&rx_ring->rx_lock);
1126 	}
1127 
1128 	for (i = 0; i < igb->num_tx_rings; i++) {
1129 		tx_ring = &igb->tx_rings[i];
1130 		mutex_destroy(&tx_ring->tx_lock);
1131 		mutex_destroy(&tx_ring->recycle_lock);
1132 		mutex_destroy(&tx_ring->tcb_head_lock);
1133 		mutex_destroy(&tx_ring->tcb_tail_lock);
1134 	}
1135 
1136 	mutex_destroy(&igb->gen_lock);
1137 	mutex_destroy(&igb->watchdog_lock);
1138 	mutex_destroy(&igb->link_lock);
1139 }
1140 
1141 static int
1142 igb_resume(dev_info_t *devinfo)
1143 {
1144 	igb_t *igb;
1145 
1146 	igb = (igb_t *)ddi_get_driver_private(devinfo);
1147 	if (igb == NULL)
1148 		return (DDI_FAILURE);
1149 
1150 	mutex_enter(&igb->gen_lock);
1151 
1152 	/*
1153 	 * Enable interrupts
1154 	 */
1155 	if (igb->attach_progress & ATTACH_PROGRESS_ENABLE_INTR) {
1156 		if (igb_enable_intrs(igb) != IGB_SUCCESS) {
1157 			igb_error(igb, "Failed to enable DDI interrupts");
1158 			mutex_exit(&igb->gen_lock);
1159 			return (DDI_FAILURE);
1160 		}
1161 	}
1162 
1163 	if (igb->igb_state & IGB_STARTED) {
1164 		if (igb_start(igb, B_FALSE) != IGB_SUCCESS) {
1165 			mutex_exit(&igb->gen_lock);
1166 			return (DDI_FAILURE);
1167 		}
1168 
1169 		/*
1170 		 * Enable and start the watchdog timer
1171 		 */
1172 		igb_enable_watchdog_timer(igb);
1173 	}
1174 
1175 	atomic_and_32(&igb->igb_state, ~IGB_SUSPENDED);
1176 
1177 	mutex_exit(&igb->gen_lock);
1178 
1179 	return (DDI_SUCCESS);
1180 }
1181 
1182 static int
1183 igb_suspend(dev_info_t *devinfo)
1184 {
1185 	igb_t *igb;
1186 
1187 	igb = (igb_t *)ddi_get_driver_private(devinfo);
1188 	if (igb == NULL)
1189 		return (DDI_FAILURE);
1190 
1191 	mutex_enter(&igb->gen_lock);
1192 
1193 	atomic_or_32(&igb->igb_state, IGB_SUSPENDED);
1194 
1195 	/*
1196 	 * Disable interrupts
1197 	 */
1198 	if (igb->attach_progress & ATTACH_PROGRESS_ENABLE_INTR) {
1199 		(void) igb_disable_intrs(igb);
1200 	}
1201 
1202 	if (!(igb->igb_state & IGB_STARTED)) {
1203 		mutex_exit(&igb->gen_lock);
1204 		return (DDI_SUCCESS);
1205 	}
1206 
1207 	igb_stop(igb, B_FALSE);
1208 
1209 	mutex_exit(&igb->gen_lock);
1210 
1211 	/*
1212 	 * Disable and stop the watchdog timer
1213 	 */
1214 	igb_disable_watchdog_timer(igb);
1215 
1216 	return (DDI_SUCCESS);
1217 }
1218 
1219 static int
1220 igb_init(igb_t *igb)
1221 {
1222 	mutex_enter(&igb->gen_lock);
1223 
1224 	/*
1225 	 * Initilize the adapter
1226 	 */
1227 	if (igb_init_adapter(igb) != IGB_SUCCESS) {
1228 		mutex_exit(&igb->gen_lock);
1229 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1230 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1231 		return (IGB_FAILURE);
1232 	}
1233 
1234 	mutex_exit(&igb->gen_lock);
1235 
1236 	return (IGB_SUCCESS);
1237 }
1238 
1239 /*
1240  * igb_init_mac_address - Initialize the default MAC address
1241  *
1242  * On success, the MAC address is entered in the igb->hw.mac.addr
1243  * and hw->mac.perm_addr fields and the adapter's RAR(0) receive
1244  * address register.
1245  *
1246  * Important side effects:
1247  * 1. adapter is reset - this is required to put it in a known state.
1248  * 2. all of non-volatile memory (NVM) is read & checksummed - NVM is where
1249  * MAC address and all default settings are stored, so a valid checksum
1250  * is required.
1251  */
1252 static int
1253 igb_init_mac_address(igb_t *igb)
1254 {
1255 	struct e1000_hw *hw = &igb->hw;
1256 
1257 	ASSERT(mutex_owned(&igb->gen_lock));
1258 
1259 	/*
1260 	 * Reset chipset to put the hardware in a known state
1261 	 * before we try to get MAC address from NVM.
1262 	 */
1263 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1264 		igb_error(igb, "Adapter reset failed.");
1265 		goto init_mac_fail;
1266 	}
1267 
1268 	/*
1269 	 * NVM validation
1270 	 */
1271 	if (((igb->hw.mac.type != e1000_i210) &&
1272 	    (igb->hw.mac.type != e1000_i211)) &&
1273 	    (e1000_validate_nvm_checksum(hw) < 0)) {
1274 		/*
1275 		 * Some PCI-E parts fail the first check due to
1276 		 * the link being in sleep state.  Call it again,
1277 		 * if it fails a second time its a real issue.
1278 		 */
1279 		if (e1000_validate_nvm_checksum(hw) < 0) {
1280 			igb_error(igb,
1281 			    "Invalid NVM checksum. Please contact "
1282 			    "the vendor to update the NVM.");
1283 			goto init_mac_fail;
1284 		}
1285 	}
1286 
1287 	/*
1288 	 * Get the mac address
1289 	 * This function should handle SPARC case correctly.
1290 	 */
1291 	if (!igb_find_mac_address(igb)) {
1292 		igb_error(igb, "Failed to get the mac address");
1293 		goto init_mac_fail;
1294 	}
1295 
1296 	/* Validate mac address */
1297 	if (!is_valid_mac_addr(hw->mac.addr)) {
1298 		igb_error(igb, "Invalid mac address");
1299 		goto init_mac_fail;
1300 	}
1301 
1302 	return (IGB_SUCCESS);
1303 
1304 init_mac_fail:
1305 	return (IGB_FAILURE);
1306 }
1307 
1308 /*
1309  * igb_init_adapter - Initialize the adapter
1310  */
1311 static int
1312 igb_init_adapter(igb_t *igb)
1313 {
1314 	struct e1000_hw *hw = &igb->hw;
1315 	uint32_t pba;
1316 	int oemid[2];
1317 	uint16_t nvmword;
1318 	uint32_t hwm;
1319 	uint32_t default_mtu;
1320 	u8 pbanum[E1000_PBANUM_LENGTH];
1321 	char eepromver[5];	/* f.ff */
1322 	int i;
1323 
1324 	ASSERT(mutex_owned(&igb->gen_lock));
1325 
1326 	/*
1327 	 * In order to obtain the default MAC address, this will reset the
1328 	 * adapter and validate the NVM that the address and many other
1329 	 * default settings come from.
1330 	 */
1331 	if (igb_init_mac_address(igb) != IGB_SUCCESS) {
1332 		igb_error(igb, "Failed to initialize MAC address");
1333 		goto init_adapter_fail;
1334 	}
1335 
1336 	/*
1337 	 * Packet Buffer Allocation (PBA)
1338 	 * Writing PBA sets the receive portion of the buffer
1339 	 * the remainder is used for the transmit buffer.
1340 	 */
1341 	switch (hw->mac.type) {
1342 	case e1000_82575:
1343 		pba = E1000_PBA_32K;
1344 		break;
1345 	case e1000_82576:
1346 		pba = E1000_READ_REG(hw, E1000_RXPBS);
1347 		pba &= E1000_RXPBS_SIZE_MASK_82576;
1348 		break;
1349 	case e1000_82580:
1350 	case e1000_i350:
1351 	case e1000_i354:
1352 		pba = E1000_READ_REG(hw, E1000_RXPBS);
1353 		pba = e1000_rxpbs_adjust_82580(pba);
1354 		break;
1355 	case e1000_i210:
1356 	case e1000_i211:
1357 		pba = E1000_PBA_34K;
1358 	default:
1359 		break;
1360 	}
1361 
1362 	/* Special needs in case of Jumbo frames */
1363 	default_mtu = igb_get_prop(igb, PROP_DEFAULT_MTU,
1364 	    MIN_MTU, MAX_MTU, DEFAULT_MTU);
1365 	if ((hw->mac.type == e1000_82575) && (default_mtu > ETHERMTU)) {
1366 		u32 tx_space, min_tx, min_rx;
1367 		pba = E1000_READ_REG(hw, E1000_PBA);
1368 		tx_space = pba >> 16;
1369 		pba &= 0xffff;
1370 		min_tx = (igb->max_frame_size +
1371 		    sizeof (struct e1000_tx_desc) - ETHERNET_FCS_SIZE) * 2;
1372 		min_tx = roundup(min_tx, 1024);
1373 		min_tx >>= 10;
1374 		min_rx = igb->max_frame_size;
1375 		min_rx = roundup(min_rx, 1024);
1376 		min_rx >>= 10;
1377 		if (tx_space < min_tx &&
1378 		    ((min_tx - tx_space) < pba)) {
1379 			pba = pba - (min_tx - tx_space);
1380 			/*
1381 			 * if short on rx space, rx wins
1382 			 * and must trump tx adjustment
1383 			 */
1384 			if (pba < min_rx)
1385 				pba = min_rx;
1386 		}
1387 		E1000_WRITE_REG(hw, E1000_PBA, pba);
1388 	}
1389 
1390 	DEBUGOUT1("igb_init: pba=%dK", pba);
1391 
1392 	/*
1393 	 * These parameters control the automatic generation (Tx) and
1394 	 * response (Rx) to Ethernet PAUSE frames.
1395 	 * - High water mark should allow for at least two frames to be
1396 	 *   received after sending an XOFF.
1397 	 * - Low water mark works best when it is very near the high water mark.
1398 	 *   This allows the receiver to restart by sending XON when it has
1399 	 *   drained a bit.
1400 	 */
1401 	hwm = min(((pba << 10) * 9 / 10),
1402 	    ((pba << 10) - 2 * igb->max_frame_size));
1403 
1404 	if (hw->mac.type < e1000_82576) {
1405 		hw->fc.high_water = hwm & 0xFFF8;  /* 8-byte granularity */
1406 		hw->fc.low_water = hw->fc.high_water - 8;
1407 	} else {
1408 		hw->fc.high_water = hwm & 0xFFF0;  /* 16-byte granularity */
1409 		hw->fc.low_water = hw->fc.high_water - 16;
1410 	}
1411 
1412 	hw->fc.pause_time = E1000_FC_PAUSE_TIME;
1413 	hw->fc.send_xon = B_TRUE;
1414 
1415 	(void) e1000_validate_mdi_setting(hw);
1416 
1417 	/*
1418 	 * Reset the chipset hardware the second time to put PBA settings
1419 	 * into effect.
1420 	 */
1421 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1422 		igb_error(igb, "Second reset failed");
1423 		goto init_adapter_fail;
1424 	}
1425 
1426 	/*
1427 	 * Don't wait for auto-negotiation to complete
1428 	 */
1429 	hw->phy.autoneg_wait_to_complete = B_FALSE;
1430 
1431 	/*
1432 	 * Copper options
1433 	 */
1434 	if (hw->phy.media_type == e1000_media_type_copper) {
1435 		hw->phy.mdix = 0;	/* AUTO_ALL_MODES */
1436 		hw->phy.disable_polarity_correction = B_FALSE;
1437 		hw->phy.ms_type = e1000_ms_hw_default; /* E1000_MASTER_SLAVE */
1438 	}
1439 
1440 	/*
1441 	 * Initialize link settings
1442 	 */
1443 	(void) igb_setup_link(igb, B_FALSE);
1444 
1445 	/*
1446 	 * Configure/Initialize hardware
1447 	 */
1448 	if (e1000_init_hw(hw) != E1000_SUCCESS) {
1449 		igb_error(igb, "Failed to initialize hardware");
1450 		goto init_adapter_fail;
1451 	}
1452 
1453 	/*
1454 	 *  Start the link setup timer
1455 	 */
1456 	igb_start_link_timer(igb);
1457 
1458 	/*
1459 	 * Disable wakeup control by default
1460 	 */
1461 	E1000_WRITE_REG(hw, E1000_WUC, 0);
1462 
1463 	/*
1464 	 * Record phy info in hw struct
1465 	 */
1466 	(void) e1000_get_phy_info(hw);
1467 
1468 	/*
1469 	 * Make sure driver has control
1470 	 */
1471 	igb_get_driver_control(hw);
1472 
1473 	/*
1474 	 * Restore LED settings to the default from EEPROM
1475 	 * to meet the standard for Sun platforms.
1476 	 */
1477 	(void) e1000_cleanup_led(hw);
1478 
1479 	/*
1480 	 * Setup MSI-X interrupts
1481 	 */
1482 	if (igb->intr_type == DDI_INTR_TYPE_MSIX)
1483 		igb->capab->setup_msix(igb);
1484 
1485 	/*
1486 	 * Initialize unicast addresses.
1487 	 */
1488 	igb_init_unicst(igb);
1489 
1490 	/*
1491 	 * Setup and initialize the mctable structures.
1492 	 */
1493 	igb_setup_multicst(igb);
1494 
1495 	/*
1496 	 * Set interrupt throttling rate
1497 	 */
1498 	for (i = 0; i < igb->intr_cnt; i++)
1499 		E1000_WRITE_REG(hw, E1000_EITR(i), igb->intr_throttling[i]);
1500 
1501 	/*
1502 	 * Read identifying information and place in devinfo.
1503 	 */
1504 	nvmword = 0xffff;
1505 	(void) e1000_read_nvm(&igb->hw, NVM_OEM_OFFSET_0, 1, &nvmword);
1506 	oemid[0] = (int)nvmword;
1507 	(void) e1000_read_nvm(&igb->hw, NVM_OEM_OFFSET_1, 1, &nvmword);
1508 	oemid[1] = (int)nvmword;
1509 	(void) ddi_prop_update_int_array(DDI_DEV_T_NONE, igb->dip,
1510 	    "oem-identifier", oemid, 2);
1511 
1512 	pbanum[0] = '\0';
1513 	(void) e1000_read_pba_string(&igb->hw, pbanum, sizeof (pbanum));
1514 	if (*pbanum != '\0') {
1515 		(void) ddi_prop_update_string(DDI_DEV_T_NONE, igb->dip,
1516 		    "printed-board-assembly", (char *)pbanum);
1517 	}
1518 
1519 	nvmword = 0xffff;
1520 	(void) e1000_read_nvm(&igb->hw, NVM_VERSION, 1, &nvmword);
1521 	if ((nvmword & 0xf00) == 0) {
1522 		(void) snprintf(eepromver, sizeof (eepromver), "%x.%x",
1523 		    (nvmword & 0xf000) >> 12, (nvmword & 0xff));
1524 		(void) ddi_prop_update_string(DDI_DEV_T_NONE, igb->dip,
1525 		    "nvm-version", eepromver);
1526 	}
1527 
1528 	/*
1529 	 * Save the state of the phy
1530 	 */
1531 	igb_get_phy_state(igb);
1532 
1533 	igb_param_sync(igb);
1534 
1535 	return (IGB_SUCCESS);
1536 
1537 init_adapter_fail:
1538 	/*
1539 	 * Reset PHY if possible
1540 	 */
1541 	if (e1000_check_reset_block(hw) == E1000_SUCCESS)
1542 		(void) e1000_phy_hw_reset(hw);
1543 
1544 	return (IGB_FAILURE);
1545 }
1546 
1547 /*
1548  * igb_stop_adapter - Stop the adapter
1549  */
1550 static void
1551 igb_stop_adapter(igb_t *igb)
1552 {
1553 	struct e1000_hw *hw = &igb->hw;
1554 
1555 	ASSERT(mutex_owned(&igb->gen_lock));
1556 
1557 	/* Stop the link setup timer */
1558 	igb_stop_link_timer(igb);
1559 
1560 	/* Tell firmware driver is no longer in control */
1561 	igb_release_driver_control(hw);
1562 
1563 	/*
1564 	 * Reset the chipset
1565 	 */
1566 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1567 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1568 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1569 	}
1570 
1571 	/*
1572 	 * e1000_phy_hw_reset is not needed here, MAC reset above is sufficient
1573 	 */
1574 }
1575 
1576 /*
1577  * igb_reset - Reset the chipset and restart the driver.
1578  *
1579  * It involves stopping and re-starting the chipset,
1580  * and re-configuring the rx/tx rings.
1581  */
1582 static int
1583 igb_reset(igb_t *igb)
1584 {
1585 	int i;
1586 
1587 	mutex_enter(&igb->gen_lock);
1588 
1589 	ASSERT(igb->igb_state & IGB_STARTED);
1590 	atomic_and_32(&igb->igb_state, ~IGB_STARTED);
1591 
1592 	/*
1593 	 * Disable the adapter interrupts to stop any rx/tx activities
1594 	 * before draining pending data and resetting hardware.
1595 	 */
1596 	igb_disable_adapter_interrupts(igb);
1597 
1598 	/*
1599 	 * Drain the pending transmit packets
1600 	 */
1601 	(void) igb_tx_drain(igb);
1602 
1603 	for (i = 0; i < igb->num_rx_rings; i++)
1604 		mutex_enter(&igb->rx_rings[i].rx_lock);
1605 	for (i = 0; i < igb->num_tx_rings; i++)
1606 		mutex_enter(&igb->tx_rings[i].tx_lock);
1607 
1608 	/*
1609 	 * Stop the adapter
1610 	 */
1611 	igb_stop_adapter(igb);
1612 
1613 	/*
1614 	 * Clean the pending tx data/resources
1615 	 */
1616 	igb_tx_clean(igb);
1617 
1618 	/*
1619 	 * Start the adapter
1620 	 */
1621 	if (igb_init_adapter(igb) != IGB_SUCCESS) {
1622 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1623 		goto reset_failure;
1624 	}
1625 
1626 	/*
1627 	 * Setup the rx/tx rings
1628 	 */
1629 	igb->tx_ring_init = B_FALSE;
1630 	igb_setup_rings(igb);
1631 
1632 	atomic_and_32(&igb->igb_state, ~(IGB_ERROR | IGB_STALL));
1633 
1634 	/*
1635 	 * Enable adapter interrupts
1636 	 * The interrupts must be enabled after the driver state is START
1637 	 */
1638 	igb->capab->enable_intr(igb);
1639 
1640 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK)
1641 		goto reset_failure;
1642 
1643 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1644 		goto reset_failure;
1645 
1646 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1647 		mutex_exit(&igb->tx_rings[i].tx_lock);
1648 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1649 		mutex_exit(&igb->rx_rings[i].rx_lock);
1650 
1651 	atomic_or_32(&igb->igb_state, IGB_STARTED);
1652 
1653 	mutex_exit(&igb->gen_lock);
1654 
1655 	return (IGB_SUCCESS);
1656 
1657 reset_failure:
1658 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1659 		mutex_exit(&igb->tx_rings[i].tx_lock);
1660 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1661 		mutex_exit(&igb->rx_rings[i].rx_lock);
1662 
1663 	mutex_exit(&igb->gen_lock);
1664 
1665 	ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1666 
1667 	return (IGB_FAILURE);
1668 }
1669 
1670 /*
1671  * igb_tx_clean - Clean the pending transmit packets and DMA resources
1672  */
1673 static void
1674 igb_tx_clean(igb_t *igb)
1675 {
1676 	igb_tx_ring_t *tx_ring;
1677 	tx_control_block_t *tcb;
1678 	link_list_t pending_list;
1679 	uint32_t desc_num;
1680 	int i, j;
1681 
1682 	LINK_LIST_INIT(&pending_list);
1683 
1684 	for (i = 0; i < igb->num_tx_rings; i++) {
1685 		tx_ring = &igb->tx_rings[i];
1686 
1687 		mutex_enter(&tx_ring->recycle_lock);
1688 
1689 		/*
1690 		 * Clean the pending tx data - the pending packets in the
1691 		 * work_list that have no chances to be transmitted again.
1692 		 *
1693 		 * We must ensure the chipset is stopped or the link is down
1694 		 * before cleaning the transmit packets.
1695 		 */
1696 		desc_num = 0;
1697 		for (j = 0; j < tx_ring->ring_size; j++) {
1698 			tcb = tx_ring->work_list[j];
1699 			if (tcb != NULL) {
1700 				desc_num += tcb->desc_num;
1701 
1702 				tx_ring->work_list[j] = NULL;
1703 
1704 				igb_free_tcb(tcb);
1705 
1706 				LIST_PUSH_TAIL(&pending_list, &tcb->link);
1707 			}
1708 		}
1709 
1710 		if (desc_num > 0) {
1711 			atomic_add_32(&tx_ring->tbd_free, desc_num);
1712 			ASSERT(tx_ring->tbd_free == tx_ring->ring_size);
1713 
1714 			/*
1715 			 * Reset the head and tail pointers of the tbd ring;
1716 			 * Reset the head write-back if it is enabled.
1717 			 */
1718 			tx_ring->tbd_head = 0;
1719 			tx_ring->tbd_tail = 0;
1720 			if (igb->tx_head_wb_enable)
1721 				*tx_ring->tbd_head_wb = 0;
1722 
1723 			E1000_WRITE_REG(&igb->hw, E1000_TDH(tx_ring->index), 0);
1724 			E1000_WRITE_REG(&igb->hw, E1000_TDT(tx_ring->index), 0);
1725 		}
1726 
1727 		mutex_exit(&tx_ring->recycle_lock);
1728 
1729 		/*
1730 		 * Add the tx control blocks in the pending list to
1731 		 * the free list.
1732 		 */
1733 		igb_put_free_list(tx_ring, &pending_list);
1734 	}
1735 }
1736 
1737 /*
1738  * igb_tx_drain - Drain the tx rings to allow pending packets to be transmitted
1739  */
1740 static boolean_t
1741 igb_tx_drain(igb_t *igb)
1742 {
1743 	igb_tx_ring_t *tx_ring;
1744 	boolean_t done;
1745 	int i, j;
1746 
1747 	/*
1748 	 * Wait for a specific time to allow pending tx packets
1749 	 * to be transmitted.
1750 	 *
1751 	 * Check the counter tbd_free to see if transmission is done.
1752 	 * No lock protection is needed here.
1753 	 *
1754 	 * Return B_TRUE if all pending packets have been transmitted;
1755 	 * Otherwise return B_FALSE;
1756 	 */
1757 	for (i = 0; i < TX_DRAIN_TIME; i++) {
1758 
1759 		done = B_TRUE;
1760 		for (j = 0; j < igb->num_tx_rings; j++) {
1761 			tx_ring = &igb->tx_rings[j];
1762 			done = done &&
1763 			    (tx_ring->tbd_free == tx_ring->ring_size);
1764 		}
1765 
1766 		if (done)
1767 			break;
1768 
1769 		msec_delay(1);
1770 	}
1771 
1772 	return (done);
1773 }
1774 
1775 /*
1776  * igb_rx_drain - Wait for all rx buffers to be released by upper layer
1777  */
1778 static boolean_t
1779 igb_rx_drain(igb_t *igb)
1780 {
1781 	boolean_t done;
1782 	int i;
1783 
1784 	/*
1785 	 * Polling the rx free list to check if those rx buffers held by
1786 	 * the upper layer are released.
1787 	 *
1788 	 * Check the counter rcb_free to see if all pending buffers are
1789 	 * released. No lock protection is needed here.
1790 	 *
1791 	 * Return B_TRUE if all pending buffers have been released;
1792 	 * Otherwise return B_FALSE;
1793 	 */
1794 	for (i = 0; i < RX_DRAIN_TIME; i++) {
1795 		done = (igb->rcb_pending == 0);
1796 
1797 		if (done)
1798 			break;
1799 
1800 		msec_delay(1);
1801 	}
1802 
1803 	return (done);
1804 }
1805 
1806 /*
1807  * igb_start - Start the driver/chipset
1808  */
1809 int
1810 igb_start(igb_t *igb, boolean_t alloc_buffer)
1811 {
1812 	int i;
1813 
1814 	ASSERT(mutex_owned(&igb->gen_lock));
1815 
1816 	if (alloc_buffer) {
1817 		if (igb_alloc_rx_data(igb) != IGB_SUCCESS) {
1818 			igb_error(igb,
1819 			    "Failed to allocate software receive rings");
1820 			return (IGB_FAILURE);
1821 		}
1822 
1823 		/* Allocate buffers for all the rx/tx rings */
1824 		if (igb_alloc_dma(igb) != IGB_SUCCESS) {
1825 			igb_error(igb, "Failed to allocate DMA resource");
1826 			return (IGB_FAILURE);
1827 		}
1828 
1829 		igb->tx_ring_init = B_TRUE;
1830 	} else {
1831 		igb->tx_ring_init = B_FALSE;
1832 	}
1833 
1834 	for (i = 0; i < igb->num_rx_rings; i++)
1835 		mutex_enter(&igb->rx_rings[i].rx_lock);
1836 	for (i = 0; i < igb->num_tx_rings; i++)
1837 		mutex_enter(&igb->tx_rings[i].tx_lock);
1838 
1839 	/*
1840 	 * Start the adapter
1841 	 */
1842 	if ((igb->attach_progress & ATTACH_PROGRESS_INIT_ADAPTER) == 0) {
1843 		if (igb_init_adapter(igb) != IGB_SUCCESS) {
1844 			igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1845 			goto start_failure;
1846 		}
1847 		igb->attach_progress |= ATTACH_PROGRESS_INIT_ADAPTER;
1848 	}
1849 
1850 	/*
1851 	 * Setup the rx/tx rings
1852 	 */
1853 	igb_setup_rings(igb);
1854 
1855 	/*
1856 	 * Enable adapter interrupts
1857 	 * The interrupts must be enabled after the driver state is START
1858 	 */
1859 	igb->capab->enable_intr(igb);
1860 
1861 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK)
1862 		goto start_failure;
1863 
1864 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1865 		goto start_failure;
1866 
1867 	if (igb->hw.mac.type == e1000_i350)
1868 		(void) e1000_set_eee_i350(&igb->hw, B_FALSE, B_FALSE);
1869 	else if (igb->hw.mac.type == e1000_i354)
1870 		(void) e1000_set_eee_i354(&igb->hw, B_FALSE, B_FALSE);
1871 
1872 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1873 		mutex_exit(&igb->tx_rings[i].tx_lock);
1874 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1875 		mutex_exit(&igb->rx_rings[i].rx_lock);
1876 
1877 	return (IGB_SUCCESS);
1878 
1879 start_failure:
1880 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1881 		mutex_exit(&igb->tx_rings[i].tx_lock);
1882 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1883 		mutex_exit(&igb->rx_rings[i].rx_lock);
1884 
1885 	ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1886 
1887 	return (IGB_FAILURE);
1888 }
1889 
1890 /*
1891  * igb_stop - Stop the driver/chipset
1892  */
1893 void
1894 igb_stop(igb_t *igb, boolean_t free_buffer)
1895 {
1896 	int i;
1897 
1898 	ASSERT(mutex_owned(&igb->gen_lock));
1899 
1900 	igb->attach_progress &= ~ATTACH_PROGRESS_INIT_ADAPTER;
1901 
1902 	/*
1903 	 * Disable the adapter interrupts
1904 	 */
1905 	igb_disable_adapter_interrupts(igb);
1906 
1907 	/*
1908 	 * Drain the pending tx packets
1909 	 */
1910 	(void) igb_tx_drain(igb);
1911 
1912 	for (i = 0; i < igb->num_rx_rings; i++)
1913 		mutex_enter(&igb->rx_rings[i].rx_lock);
1914 	for (i = 0; i < igb->num_tx_rings; i++)
1915 		mutex_enter(&igb->tx_rings[i].tx_lock);
1916 
1917 	/*
1918 	 * Stop the adapter
1919 	 */
1920 	igb_stop_adapter(igb);
1921 
1922 	/*
1923 	 * Clean the pending tx data/resources
1924 	 */
1925 	igb_tx_clean(igb);
1926 
1927 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1928 		mutex_exit(&igb->tx_rings[i].tx_lock);
1929 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1930 		mutex_exit(&igb->rx_rings[i].rx_lock);
1931 
1932 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1933 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1934 
1935 	if (igb->link_state == LINK_STATE_UP) {
1936 		igb->link_state = LINK_STATE_UNKNOWN;
1937 		mac_link_update(igb->mac_hdl, igb->link_state);
1938 	}
1939 
1940 	if (free_buffer) {
1941 		/*
1942 		 * Release the DMA/memory resources of rx/tx rings
1943 		 */
1944 		igb_free_dma(igb);
1945 		igb_free_rx_data(igb);
1946 	}
1947 }
1948 
1949 /*
1950  * igb_alloc_rings - Allocate memory space for rx/tx rings
1951  */
1952 static int
1953 igb_alloc_rings(igb_t *igb)
1954 {
1955 	/*
1956 	 * Allocate memory space for rx rings
1957 	 */
1958 	igb->rx_rings = kmem_zalloc(
1959 	    sizeof (igb_rx_ring_t) * igb->num_rx_rings,
1960 	    KM_NOSLEEP);
1961 
1962 	if (igb->rx_rings == NULL) {
1963 		return (IGB_FAILURE);
1964 	}
1965 
1966 	/*
1967 	 * Allocate memory space for tx rings
1968 	 */
1969 	igb->tx_rings = kmem_zalloc(
1970 	    sizeof (igb_tx_ring_t) * igb->num_tx_rings,
1971 	    KM_NOSLEEP);
1972 
1973 	if (igb->tx_rings == NULL) {
1974 		kmem_free(igb->rx_rings,
1975 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1976 		igb->rx_rings = NULL;
1977 		return (IGB_FAILURE);
1978 	}
1979 
1980 	/*
1981 	 * Allocate memory space for rx ring groups
1982 	 */
1983 	igb->rx_groups = kmem_zalloc(
1984 	    sizeof (igb_rx_group_t) * igb->num_rx_groups,
1985 	    KM_NOSLEEP);
1986 
1987 	if (igb->rx_groups == NULL) {
1988 		kmem_free(igb->rx_rings,
1989 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1990 		kmem_free(igb->tx_rings,
1991 		    sizeof (igb_tx_ring_t) * igb->num_tx_rings);
1992 		igb->rx_rings = NULL;
1993 		igb->tx_rings = NULL;
1994 		return (IGB_FAILURE);
1995 	}
1996 
1997 	return (IGB_SUCCESS);
1998 }
1999 
2000 /*
2001  * igb_free_rings - Free the memory space of rx/tx rings.
2002  */
2003 static void
2004 igb_free_rings(igb_t *igb)
2005 {
2006 	if (igb->rx_rings != NULL) {
2007 		kmem_free(igb->rx_rings,
2008 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
2009 		igb->rx_rings = NULL;
2010 	}
2011 
2012 	if (igb->tx_rings != NULL) {
2013 		kmem_free(igb->tx_rings,
2014 		    sizeof (igb_tx_ring_t) * igb->num_tx_rings);
2015 		igb->tx_rings = NULL;
2016 	}
2017 
2018 	if (igb->rx_groups != NULL) {
2019 		kmem_free(igb->rx_groups,
2020 		    sizeof (igb_rx_group_t) * igb->num_rx_groups);
2021 		igb->rx_groups = NULL;
2022 	}
2023 }
2024 
2025 static int
2026 igb_alloc_rx_data(igb_t *igb)
2027 {
2028 	igb_rx_ring_t *rx_ring;
2029 	int i;
2030 
2031 	for (i = 0; i < igb->num_rx_rings; i++) {
2032 		rx_ring = &igb->rx_rings[i];
2033 		if (igb_alloc_rx_ring_data(rx_ring) != IGB_SUCCESS)
2034 			goto alloc_rx_rings_failure;
2035 	}
2036 	return (IGB_SUCCESS);
2037 
2038 alloc_rx_rings_failure:
2039 	igb_free_rx_data(igb);
2040 	return (IGB_FAILURE);
2041 }
2042 
2043 static void
2044 igb_free_rx_data(igb_t *igb)
2045 {
2046 	igb_rx_ring_t *rx_ring;
2047 	igb_rx_data_t *rx_data;
2048 	int i;
2049 
2050 	for (i = 0; i < igb->num_rx_rings; i++) {
2051 		rx_ring = &igb->rx_rings[i];
2052 
2053 		mutex_enter(&igb->rx_pending_lock);
2054 		rx_data = rx_ring->rx_data;
2055 
2056 		if (rx_data != NULL) {
2057 			rx_data->flag |= IGB_RX_STOPPED;
2058 
2059 			if (rx_data->rcb_pending == 0) {
2060 				igb_free_rx_ring_data(rx_data);
2061 				rx_ring->rx_data = NULL;
2062 			}
2063 		}
2064 
2065 		mutex_exit(&igb->rx_pending_lock);
2066 	}
2067 }
2068 
2069 /*
2070  * igb_setup_rings - Setup rx/tx rings
2071  */
2072 static void
2073 igb_setup_rings(igb_t *igb)
2074 {
2075 	/*
2076 	 * Setup the rx/tx rings, including the following:
2077 	 *
2078 	 * 1. Setup the descriptor ring and the control block buffers;
2079 	 * 2. Initialize necessary registers for receive/transmit;
2080 	 * 3. Initialize software pointers/parameters for receive/transmit;
2081 	 */
2082 	igb_setup_rx(igb);
2083 
2084 	igb_setup_tx(igb);
2085 }
2086 
2087 static void
2088 igb_setup_rx_ring(igb_rx_ring_t *rx_ring)
2089 {
2090 	igb_t *igb = rx_ring->igb;
2091 	igb_rx_data_t *rx_data = rx_ring->rx_data;
2092 	struct e1000_hw *hw = &igb->hw;
2093 	rx_control_block_t *rcb;
2094 	union e1000_adv_rx_desc	*rbd;
2095 	uint32_t size;
2096 	uint32_t buf_low;
2097 	uint32_t buf_high;
2098 	uint32_t rxdctl;
2099 	int i;
2100 
2101 	ASSERT(mutex_owned(&rx_ring->rx_lock));
2102 	ASSERT(mutex_owned(&igb->gen_lock));
2103 
2104 	/*
2105 	 * Initialize descriptor ring with buffer addresses
2106 	 */
2107 	for (i = 0; i < igb->rx_ring_size; i++) {
2108 		rcb = rx_data->work_list[i];
2109 		rbd = &rx_data->rbd_ring[i];
2110 
2111 		rbd->read.pkt_addr = rcb->rx_buf.dma_address;
2112 		rbd->read.hdr_addr = NULL;
2113 	}
2114 
2115 	/*
2116 	 * Initialize the base address registers
2117 	 */
2118 	buf_low = (uint32_t)rx_data->rbd_area.dma_address;
2119 	buf_high = (uint32_t)(rx_data->rbd_area.dma_address >> 32);
2120 	E1000_WRITE_REG(hw, E1000_RDBAH(rx_ring->index), buf_high);
2121 	E1000_WRITE_REG(hw, E1000_RDBAL(rx_ring->index), buf_low);
2122 
2123 	/*
2124 	 * Initialize the length register
2125 	 */
2126 	size = rx_data->ring_size * sizeof (union e1000_adv_rx_desc);
2127 	E1000_WRITE_REG(hw, E1000_RDLEN(rx_ring->index), size);
2128 
2129 	/*
2130 	 * Initialize buffer size & descriptor type
2131 	 */
2132 	E1000_WRITE_REG(hw, E1000_SRRCTL(rx_ring->index),
2133 	    ((igb->rx_buf_size >> E1000_SRRCTL_BSIZEPKT_SHIFT) |
2134 	    E1000_SRRCTL_DESCTYPE_ADV_ONEBUF));
2135 
2136 	/*
2137 	 * Setup the Receive Descriptor Control Register (RXDCTL)
2138 	 */
2139 	rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(rx_ring->index));
2140 	rxdctl &= igb->capab->rxdctl_mask;
2141 	rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
2142 	rxdctl |= 16;		/* pthresh */
2143 	rxdctl |= 8 << 8;	/* hthresh */
2144 	rxdctl |= 1 << 16;	/* wthresh */
2145 	E1000_WRITE_REG(hw, E1000_RXDCTL(rx_ring->index), rxdctl);
2146 
2147 	rx_data->rbd_next = 0;
2148 }
2149 
2150 static void
2151 igb_setup_rx(igb_t *igb)
2152 {
2153 	igb_rx_ring_t *rx_ring;
2154 	igb_rx_data_t *rx_data;
2155 	igb_rx_group_t *rx_group;
2156 	struct e1000_hw *hw = &igb->hw;
2157 	uint32_t rctl, rxcsum;
2158 	uint32_t ring_per_group;
2159 	int i;
2160 
2161 	/*
2162 	 * Setup the Receive Control Register (RCTL), and enable the
2163 	 * receiver. The initial configuration is to: enable the receiver,
2164 	 * accept broadcasts, discard bad packets, accept long packets,
2165 	 * disable VLAN filter checking, and set receive buffer size to
2166 	 * 2k.  For 82575, also set the receive descriptor minimum
2167 	 * threshold size to 1/2 the ring.
2168 	 */
2169 	rctl = E1000_READ_REG(hw, E1000_RCTL);
2170 
2171 	/*
2172 	 * Clear the field used for wakeup control.  This driver doesn't do
2173 	 * wakeup but leave this here for completeness.
2174 	 */
2175 	rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2176 	rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
2177 
2178 	rctl |= (E1000_RCTL_EN |	/* Enable Receive Unit */
2179 	    E1000_RCTL_BAM |		/* Accept Broadcast Packets */
2180 	    E1000_RCTL_LPE |		/* Large Packet Enable */
2181 					/* Multicast filter offset */
2182 	    (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT) |
2183 	    E1000_RCTL_RDMTS_HALF |	/* rx descriptor threshold */
2184 	    E1000_RCTL_SECRC);		/* Strip Ethernet CRC */
2185 
2186 	for (i = 0; i < igb->num_rx_groups; i++) {
2187 		rx_group = &igb->rx_groups[i];
2188 		rx_group->index = i;
2189 		rx_group->igb = igb;
2190 	}
2191 
2192 	/*
2193 	 * Set up all rx descriptor rings - must be called before receive unit
2194 	 * enabled.
2195 	 */
2196 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
2197 	for (i = 0; i < igb->num_rx_rings; i++) {
2198 		rx_ring = &igb->rx_rings[i];
2199 		igb_setup_rx_ring(rx_ring);
2200 
2201 		/*
2202 		 * Map a ring to a group by assigning a group index
2203 		 */
2204 		rx_ring->group_index = i / ring_per_group;
2205 	}
2206 
2207 	/*
2208 	 * Setup the Rx Long Packet Max Length register
2209 	 */
2210 	E1000_WRITE_REG(hw, E1000_RLPML, igb->max_frame_size);
2211 
2212 	/*
2213 	 * Hardware checksum settings
2214 	 */
2215 	if (igb->rx_hcksum_enable) {
2216 		rxcsum =
2217 		    E1000_RXCSUM_TUOFL |	/* TCP/UDP checksum */
2218 		    E1000_RXCSUM_IPOFL;		/* IP checksum */
2219 
2220 		E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2221 	}
2222 
2223 	/*
2224 	 * Setup classify and RSS for multiple receive queues
2225 	 */
2226 	switch (igb->vmdq_mode) {
2227 	case E1000_VMDQ_OFF:
2228 		/*
2229 		 * One ring group, only RSS is needed when more than
2230 		 * one ring enabled.
2231 		 */
2232 		if (igb->num_rx_rings > 1)
2233 			igb_setup_rss(igb);
2234 		break;
2235 	case E1000_VMDQ_MAC:
2236 		/*
2237 		 * Multiple groups, each group has one ring,
2238 		 * only the MAC classification is needed.
2239 		 */
2240 		igb_setup_mac_classify(igb);
2241 		break;
2242 	case E1000_VMDQ_MAC_RSS:
2243 		/*
2244 		 * Multiple groups and multiple rings, both
2245 		 * MAC classification and RSS are needed.
2246 		 */
2247 		igb_setup_mac_rss_classify(igb);
2248 		break;
2249 	}
2250 
2251 	/*
2252 	 * Enable the receive unit - must be done after all
2253 	 * the rx setup above.
2254 	 */
2255 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2256 
2257 	/*
2258 	 * Initialize all adapter ring head & tail pointers - must
2259 	 * be done after receive unit is enabled
2260 	 */
2261 	for (i = 0; i < igb->num_rx_rings; i++) {
2262 		rx_ring = &igb->rx_rings[i];
2263 		rx_data = rx_ring->rx_data;
2264 		E1000_WRITE_REG(hw, E1000_RDH(i), 0);
2265 		E1000_WRITE_REG(hw, E1000_RDT(i), rx_data->ring_size - 1);
2266 	}
2267 
2268 	/*
2269 	 * 82575 with manageability enabled needs a special flush to make
2270 	 * sure the fifos start clean.
2271 	 */
2272 	if ((hw->mac.type == e1000_82575) &&
2273 	    (E1000_READ_REG(hw, E1000_MANC) & E1000_MANC_RCV_TCO_EN)) {
2274 		e1000_rx_fifo_flush_82575(hw);
2275 	}
2276 }
2277 
2278 static void
2279 igb_setup_tx_ring(igb_tx_ring_t *tx_ring)
2280 {
2281 	igb_t *igb = tx_ring->igb;
2282 	struct e1000_hw *hw = &igb->hw;
2283 	uint32_t size;
2284 	uint32_t buf_low;
2285 	uint32_t buf_high;
2286 	uint32_t reg_val;
2287 
2288 	ASSERT(mutex_owned(&tx_ring->tx_lock));
2289 	ASSERT(mutex_owned(&igb->gen_lock));
2290 
2291 
2292 	/*
2293 	 * Initialize the length register
2294 	 */
2295 	size = tx_ring->ring_size * sizeof (union e1000_adv_tx_desc);
2296 	E1000_WRITE_REG(hw, E1000_TDLEN(tx_ring->index), size);
2297 
2298 	/*
2299 	 * Initialize the base address registers
2300 	 */
2301 	buf_low = (uint32_t)tx_ring->tbd_area.dma_address;
2302 	buf_high = (uint32_t)(tx_ring->tbd_area.dma_address >> 32);
2303 	E1000_WRITE_REG(hw, E1000_TDBAL(tx_ring->index), buf_low);
2304 	E1000_WRITE_REG(hw, E1000_TDBAH(tx_ring->index), buf_high);
2305 
2306 	/*
2307 	 * Setup head & tail pointers
2308 	 */
2309 	E1000_WRITE_REG(hw, E1000_TDH(tx_ring->index), 0);
2310 	E1000_WRITE_REG(hw, E1000_TDT(tx_ring->index), 0);
2311 
2312 	/*
2313 	 * Setup head write-back
2314 	 */
2315 	if (igb->tx_head_wb_enable) {
2316 		/*
2317 		 * The memory of the head write-back is allocated using
2318 		 * the extra tbd beyond the tail of the tbd ring.
2319 		 */
2320 		tx_ring->tbd_head_wb = (uint32_t *)
2321 		    ((uintptr_t)tx_ring->tbd_area.address + size);
2322 		*tx_ring->tbd_head_wb = 0;
2323 
2324 		buf_low = (uint32_t)
2325 		    (tx_ring->tbd_area.dma_address + size);
2326 		buf_high = (uint32_t)
2327 		    ((tx_ring->tbd_area.dma_address + size) >> 32);
2328 
2329 		/* Set the head write-back enable bit */
2330 		buf_low |= E1000_TX_HEAD_WB_ENABLE;
2331 
2332 		E1000_WRITE_REG(hw, E1000_TDWBAL(tx_ring->index), buf_low);
2333 		E1000_WRITE_REG(hw, E1000_TDWBAH(tx_ring->index), buf_high);
2334 
2335 		/*
2336 		 * Turn off relaxed ordering for head write back or it will
2337 		 * cause problems with the tx recycling
2338 		 */
2339 		reg_val = E1000_READ_REG(hw,
2340 		    E1000_DCA_TXCTRL(tx_ring->index));
2341 		reg_val &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
2342 		E1000_WRITE_REG(hw,
2343 		    E1000_DCA_TXCTRL(tx_ring->index), reg_val);
2344 	} else {
2345 		tx_ring->tbd_head_wb = NULL;
2346 	}
2347 
2348 	tx_ring->tbd_head = 0;
2349 	tx_ring->tbd_tail = 0;
2350 	tx_ring->tbd_free = tx_ring->ring_size;
2351 
2352 	if (igb->tx_ring_init == B_TRUE) {
2353 		tx_ring->tcb_head = 0;
2354 		tx_ring->tcb_tail = 0;
2355 		tx_ring->tcb_free = tx_ring->free_list_size;
2356 	}
2357 
2358 	/*
2359 	 * Enable TXDCTL per queue
2360 	 */
2361 	reg_val = E1000_READ_REG(hw, E1000_TXDCTL(tx_ring->index));
2362 	reg_val |= E1000_TXDCTL_QUEUE_ENABLE;
2363 	E1000_WRITE_REG(hw, E1000_TXDCTL(tx_ring->index), reg_val);
2364 
2365 	/*
2366 	 * Initialize hardware checksum offload settings
2367 	 */
2368 	bzero(&tx_ring->tx_context, sizeof (tx_context_t));
2369 }
2370 
2371 static void
2372 igb_setup_tx(igb_t *igb)
2373 {
2374 	igb_tx_ring_t *tx_ring;
2375 	struct e1000_hw *hw = &igb->hw;
2376 	uint32_t reg_val;
2377 	int i;
2378 
2379 	for (i = 0; i < igb->num_tx_rings; i++) {
2380 		tx_ring = &igb->tx_rings[i];
2381 		igb_setup_tx_ring(tx_ring);
2382 	}
2383 
2384 	/*
2385 	 * Setup the Transmit Control Register (TCTL)
2386 	 */
2387 	reg_val = E1000_READ_REG(hw, E1000_TCTL);
2388 	reg_val &= ~E1000_TCTL_CT;
2389 	reg_val |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2390 	    (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2391 
2392 	/* Enable transmits */
2393 	reg_val |= E1000_TCTL_EN;
2394 
2395 	E1000_WRITE_REG(hw, E1000_TCTL, reg_val);
2396 }
2397 
2398 /*
2399  * igb_setup_rss - Setup receive-side scaling feature
2400  */
2401 static void
2402 igb_setup_rss(igb_t *igb)
2403 {
2404 	struct e1000_hw *hw = &igb->hw;
2405 	uint32_t i, mrqc, rxcsum;
2406 	int shift = 0;
2407 	uint32_t random;
2408 	union e1000_reta {
2409 		uint32_t	dword;
2410 		uint8_t		bytes[4];
2411 	} reta;
2412 
2413 	/* Setup the Redirection Table */
2414 	if (hw->mac.type == e1000_82576) {
2415 		shift = 3;
2416 	} else if (hw->mac.type == e1000_82575) {
2417 		shift = 6;
2418 	}
2419 	for (i = 0; i < (32 * 4); i++) {
2420 		reta.bytes[i & 3] = (i % igb->num_rx_rings) << shift;
2421 		if ((i & 3) == 3) {
2422 			E1000_WRITE_REG(hw,
2423 			    (E1000_RETA(0) + (i & ~3)), reta.dword);
2424 		}
2425 	}
2426 
2427 	/* Fill out hash function seeds */
2428 	for (i = 0; i < 10; i++) {
2429 		(void) random_get_pseudo_bytes((uint8_t *)&random,
2430 		    sizeof (uint32_t));
2431 		E1000_WRITE_REG(hw, E1000_RSSRK(i), random);
2432 	}
2433 
2434 	/* Setup the Multiple Receive Queue Control register */
2435 	mrqc = E1000_MRQC_ENABLE_RSS_4Q;
2436 	mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2437 	    E1000_MRQC_RSS_FIELD_IPV4_TCP |
2438 	    E1000_MRQC_RSS_FIELD_IPV6 |
2439 	    E1000_MRQC_RSS_FIELD_IPV6_TCP |
2440 	    E1000_MRQC_RSS_FIELD_IPV4_UDP |
2441 	    E1000_MRQC_RSS_FIELD_IPV6_UDP |
2442 	    E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2443 	    E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2444 
2445 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2446 
2447 	/*
2448 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2449 	 *
2450 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2451 	 * checksum offloading provided by the 82575 chipset besides the IP
2452 	 * header checksum offloading and the TCP/UDP checksum offloading.
2453 	 * The Packet Checksum is by default computed over the entire packet
2454 	 * from the first byte of the DA through the last byte of the CRC,
2455 	 * including the Ethernet and IP headers.
2456 	 *
2457 	 * It is a hardware limitation that Packet Checksum is mutually
2458 	 * exclusive with RSS.
2459 	 */
2460 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2461 	rxcsum |= E1000_RXCSUM_PCSD;
2462 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2463 }
2464 
2465 /*
2466  * igb_setup_mac_rss_classify - Setup MAC classification and rss
2467  */
2468 static void
2469 igb_setup_mac_rss_classify(igb_t *igb)
2470 {
2471 	struct e1000_hw *hw = &igb->hw;
2472 	uint32_t i, mrqc, vmdctl, rxcsum;
2473 	uint32_t ring_per_group;
2474 	int shift_group0, shift_group1;
2475 	uint32_t random;
2476 	union e1000_reta {
2477 		uint32_t	dword;
2478 		uint8_t		bytes[4];
2479 	} reta;
2480 
2481 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
2482 
2483 	/* Setup the Redirection Table, it is shared between two groups */
2484 	shift_group0 = 2;
2485 	shift_group1 = 6;
2486 	for (i = 0; i < (32 * 4); i++) {
2487 		reta.bytes[i & 3] = ((i % ring_per_group) << shift_group0) |
2488 		    ((ring_per_group + (i % ring_per_group)) << shift_group1);
2489 		if ((i & 3) == 3) {
2490 			E1000_WRITE_REG(hw,
2491 			    (E1000_RETA(0) + (i & ~3)), reta.dword);
2492 		}
2493 	}
2494 
2495 	/* Fill out hash function seeds */
2496 	for (i = 0; i < 10; i++) {
2497 		(void) random_get_pseudo_bytes((uint8_t *)&random,
2498 		    sizeof (uint32_t));
2499 		E1000_WRITE_REG(hw, E1000_RSSRK(i), random);
2500 	}
2501 
2502 	/*
2503 	 * Setup the Multiple Receive Queue Control register,
2504 	 * enable VMDq based on packet destination MAC address and RSS.
2505 	 */
2506 	mrqc = E1000_MRQC_ENABLE_VMDQ_MAC_RSS_GROUP;
2507 	mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2508 	    E1000_MRQC_RSS_FIELD_IPV4_TCP |
2509 	    E1000_MRQC_RSS_FIELD_IPV6 |
2510 	    E1000_MRQC_RSS_FIELD_IPV6_TCP |
2511 	    E1000_MRQC_RSS_FIELD_IPV4_UDP |
2512 	    E1000_MRQC_RSS_FIELD_IPV6_UDP |
2513 	    E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2514 	    E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2515 
2516 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2517 
2518 
2519 	/* Define the default group and default queues */
2520 	vmdctl = E1000_VMDQ_MAC_GROUP_DEFAULT_QUEUE;
2521 	E1000_WRITE_REG(hw, E1000_VT_CTL, vmdctl);
2522 
2523 	/*
2524 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2525 	 *
2526 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2527 	 * checksum offloading provided by the 82575 chipset besides the IP
2528 	 * header checksum offloading and the TCP/UDP checksum offloading.
2529 	 * The Packet Checksum is by default computed over the entire packet
2530 	 * from the first byte of the DA through the last byte of the CRC,
2531 	 * including the Ethernet and IP headers.
2532 	 *
2533 	 * It is a hardware limitation that Packet Checksum is mutually
2534 	 * exclusive with RSS.
2535 	 */
2536 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2537 	rxcsum |= E1000_RXCSUM_PCSD;
2538 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2539 }
2540 
2541 /*
2542  * igb_setup_mac_classify - Setup MAC classification feature
2543  */
2544 static void
2545 igb_setup_mac_classify(igb_t *igb)
2546 {
2547 	struct e1000_hw *hw = &igb->hw;
2548 	uint32_t mrqc, rxcsum;
2549 
2550 	/*
2551 	 * Setup the Multiple Receive Queue Control register,
2552 	 * enable VMDq based on packet destination MAC address.
2553 	 */
2554 	mrqc = E1000_MRQC_ENABLE_VMDQ_MAC_GROUP;
2555 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2556 
2557 	/*
2558 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2559 	 *
2560 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2561 	 * checksum offloading provided by the 82575 chipset besides the IP
2562 	 * header checksum offloading and the TCP/UDP checksum offloading.
2563 	 * The Packet Checksum is by default computed over the entire packet
2564 	 * from the first byte of the DA through the last byte of the CRC,
2565 	 * including the Ethernet and IP headers.
2566 	 *
2567 	 * It is a hardware limitation that Packet Checksum is mutually
2568 	 * exclusive with RSS.
2569 	 */
2570 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2571 	rxcsum |= E1000_RXCSUM_PCSD;
2572 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2573 
2574 }
2575 
2576 /*
2577  * igb_init_unicst - Initialize the unicast addresses
2578  */
2579 static void
2580 igb_init_unicst(igb_t *igb)
2581 {
2582 	struct e1000_hw *hw = &igb->hw;
2583 	int slot;
2584 
2585 	/*
2586 	 * Here we should consider two situations:
2587 	 *
2588 	 * 1. Chipset is initialized the first time
2589 	 *    Initialize the multiple unicast addresses, and
2590 	 *    save the default MAC address.
2591 	 *
2592 	 * 2. Chipset is reset
2593 	 *    Recover the multiple unicast addresses from the
2594 	 *    software data structure to the RAR registers.
2595 	 */
2596 
2597 	/*
2598 	 * Clear the default MAC address in the RAR0 rgister,
2599 	 * which is loaded from EEPROM when system boot or chipreset,
2600 	 * this will cause the conficts with add_mac/rem_mac entry
2601 	 * points when VMDq is enabled. For this reason, the RAR0
2602 	 * must be cleared for both cases mentioned above.
2603 	 */
2604 	e1000_rar_clear(hw, 0);
2605 
2606 	if (!igb->unicst_init) {
2607 
2608 		/* Initialize the multiple unicast addresses */
2609 		igb->unicst_total = MAX_NUM_UNICAST_ADDRESSES;
2610 		igb->unicst_avail = igb->unicst_total;
2611 
2612 		for (slot = 0; slot < igb->unicst_total; slot++)
2613 			igb->unicst_addr[slot].mac.set = 0;
2614 
2615 		igb->unicst_init = B_TRUE;
2616 	} else {
2617 		/* Re-configure the RAR registers */
2618 		for (slot = 0; slot < igb->unicst_total; slot++) {
2619 			(void) e1000_rar_set_vmdq(hw,
2620 			    igb->unicst_addr[slot].mac.addr,
2621 			    slot, igb->vmdq_mode,
2622 			    igb->unicst_addr[slot].mac.group_index);
2623 		}
2624 	}
2625 }
2626 
2627 /*
2628  * igb_unicst_find - Find the slot for the specified unicast address
2629  */
2630 int
2631 igb_unicst_find(igb_t *igb, const uint8_t *mac_addr)
2632 {
2633 	int slot;
2634 
2635 	ASSERT(mutex_owned(&igb->gen_lock));
2636 
2637 	for (slot = 0; slot < igb->unicst_total; slot++) {
2638 		if (bcmp(igb->unicst_addr[slot].mac.addr,
2639 		    mac_addr, ETHERADDRL) == 0)
2640 			return (slot);
2641 	}
2642 
2643 	return (-1);
2644 }
2645 
2646 /*
2647  * igb_unicst_set - Set the unicast address to the specified slot
2648  */
2649 int
2650 igb_unicst_set(igb_t *igb, const uint8_t *mac_addr,
2651     int slot)
2652 {
2653 	struct e1000_hw *hw = &igb->hw;
2654 
2655 	ASSERT(mutex_owned(&igb->gen_lock));
2656 
2657 	/*
2658 	 * Save the unicast address in the software data structure
2659 	 */
2660 	bcopy(mac_addr, igb->unicst_addr[slot].mac.addr, ETHERADDRL);
2661 
2662 	/*
2663 	 * Set the unicast address to the RAR register
2664 	 */
2665 	(void) e1000_rar_set(hw, (uint8_t *)mac_addr, slot);
2666 
2667 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2668 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2669 		return (EIO);
2670 	}
2671 
2672 	return (0);
2673 }
2674 
2675 /*
2676  * igb_multicst_add - Add a multicst address
2677  */
2678 int
2679 igb_multicst_add(igb_t *igb, const uint8_t *multiaddr)
2680 {
2681 	struct ether_addr *new_table;
2682 	size_t new_len;
2683 	size_t old_len;
2684 
2685 	ASSERT(mutex_owned(&igb->gen_lock));
2686 
2687 	if ((multiaddr[0] & 01) == 0) {
2688 		igb_error(igb, "Illegal multicast address");
2689 		return (EINVAL);
2690 	}
2691 
2692 	if (igb->mcast_count >= igb->mcast_max_num) {
2693 		igb_error(igb, "Adapter requested more than %d mcast addresses",
2694 		    igb->mcast_max_num);
2695 		return (ENOENT);
2696 	}
2697 
2698 	if (igb->mcast_count == igb->mcast_alloc_count) {
2699 		old_len = igb->mcast_alloc_count *
2700 		    sizeof (struct ether_addr);
2701 		new_len = (igb->mcast_alloc_count + MCAST_ALLOC_COUNT) *
2702 		    sizeof (struct ether_addr);
2703 
2704 		new_table = kmem_alloc(new_len, KM_NOSLEEP);
2705 		if (new_table == NULL) {
2706 			igb_error(igb,
2707 			    "Not enough memory to alloc mcast table");
2708 			return (ENOMEM);
2709 		}
2710 
2711 		if (igb->mcast_table != NULL) {
2712 			bcopy(igb->mcast_table, new_table, old_len);
2713 			kmem_free(igb->mcast_table, old_len);
2714 		}
2715 		igb->mcast_alloc_count += MCAST_ALLOC_COUNT;
2716 		igb->mcast_table = new_table;
2717 	}
2718 
2719 	bcopy(multiaddr,
2720 	    &igb->mcast_table[igb->mcast_count], ETHERADDRL);
2721 	igb->mcast_count++;
2722 
2723 	/*
2724 	 * Update the multicast table in the hardware
2725 	 */
2726 	igb_setup_multicst(igb);
2727 
2728 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2729 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2730 		return (EIO);
2731 	}
2732 
2733 	return (0);
2734 }
2735 
2736 /*
2737  * igb_multicst_remove - Remove a multicst address
2738  */
2739 int
2740 igb_multicst_remove(igb_t *igb, const uint8_t *multiaddr)
2741 {
2742 	struct ether_addr *new_table;
2743 	size_t new_len;
2744 	size_t old_len;
2745 	int i;
2746 
2747 	ASSERT(mutex_owned(&igb->gen_lock));
2748 
2749 	for (i = 0; i < igb->mcast_count; i++) {
2750 		if (bcmp(multiaddr, &igb->mcast_table[i],
2751 		    ETHERADDRL) == 0) {
2752 			for (i++; i < igb->mcast_count; i++) {
2753 				igb->mcast_table[i - 1] =
2754 				    igb->mcast_table[i];
2755 			}
2756 			igb->mcast_count--;
2757 			break;
2758 		}
2759 	}
2760 
2761 	if ((igb->mcast_alloc_count - igb->mcast_count) >
2762 	    MCAST_ALLOC_COUNT) {
2763 		old_len = igb->mcast_alloc_count *
2764 		    sizeof (struct ether_addr);
2765 		new_len = (igb->mcast_alloc_count - MCAST_ALLOC_COUNT) *
2766 		    sizeof (struct ether_addr);
2767 
2768 		new_table = kmem_alloc(new_len, KM_NOSLEEP);
2769 		if (new_table != NULL) {
2770 			bcopy(igb->mcast_table, new_table, new_len);
2771 			kmem_free(igb->mcast_table, old_len);
2772 			igb->mcast_alloc_count -= MCAST_ALLOC_COUNT;
2773 			igb->mcast_table = new_table;
2774 		}
2775 	}
2776 
2777 	/*
2778 	 * Update the multicast table in the hardware
2779 	 */
2780 	igb_setup_multicst(igb);
2781 
2782 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2783 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2784 		return (EIO);
2785 	}
2786 
2787 	return (0);
2788 }
2789 
2790 static void
2791 igb_release_multicast(igb_t *igb)
2792 {
2793 	if (igb->mcast_table != NULL) {
2794 		kmem_free(igb->mcast_table,
2795 		    igb->mcast_alloc_count * sizeof (struct ether_addr));
2796 		igb->mcast_table = NULL;
2797 	}
2798 }
2799 
2800 /*
2801  * igb_setup_multicast - setup multicast data structures
2802  *
2803  * This routine initializes all of the multicast related structures
2804  * and save them in the hardware registers.
2805  */
2806 static void
2807 igb_setup_multicst(igb_t *igb)
2808 {
2809 	uint8_t *mc_addr_list;
2810 	uint32_t mc_addr_count;
2811 	struct e1000_hw *hw = &igb->hw;
2812 
2813 	ASSERT(mutex_owned(&igb->gen_lock));
2814 	ASSERT(igb->mcast_count <= igb->mcast_max_num);
2815 
2816 	mc_addr_list = (uint8_t *)igb->mcast_table;
2817 	mc_addr_count = igb->mcast_count;
2818 
2819 	/*
2820 	 * Update the multicase addresses to the MTA registers
2821 	 */
2822 	e1000_update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
2823 }
2824 
2825 /*
2826  * igb_get_conf - Get driver configurations set in driver.conf
2827  *
2828  * This routine gets user-configured values out of the configuration
2829  * file igb.conf.
2830  *
2831  * For each configurable value, there is a minimum, a maximum, and a
2832  * default.
2833  * If user does not configure a value, use the default.
2834  * If user configures below the minimum, use the minumum.
2835  * If user configures above the maximum, use the maxumum.
2836  */
2837 static void
2838 igb_get_conf(igb_t *igb)
2839 {
2840 	struct e1000_hw *hw = &igb->hw;
2841 	uint32_t default_mtu;
2842 	uint32_t flow_control;
2843 	uint32_t ring_per_group;
2844 	int i;
2845 
2846 	/*
2847 	 * igb driver supports the following user configurations:
2848 	 *
2849 	 * Link configurations:
2850 	 *    adv_autoneg_cap
2851 	 *    adv_1000fdx_cap
2852 	 *    adv_100fdx_cap
2853 	 *    adv_100hdx_cap
2854 	 *    adv_10fdx_cap
2855 	 *    adv_10hdx_cap
2856 	 * Note: 1000hdx is not supported.
2857 	 *
2858 	 * Jumbo frame configuration:
2859 	 *    default_mtu
2860 	 *
2861 	 * Ethernet flow control configuration:
2862 	 *    flow_control
2863 	 *
2864 	 * Multiple rings configurations:
2865 	 *    tx_queue_number
2866 	 *    tx_ring_size
2867 	 *    rx_queue_number
2868 	 *    rx_ring_size
2869 	 *
2870 	 * Call igb_get_prop() to get the value for a specific
2871 	 * configuration parameter.
2872 	 */
2873 
2874 	/*
2875 	 * Link configurations
2876 	 */
2877 	igb->param_adv_autoneg_cap = igb_get_prop(igb,
2878 	    PROP_ADV_AUTONEG_CAP, 0, 1, 1);
2879 	igb->param_adv_1000fdx_cap = igb_get_prop(igb,
2880 	    PROP_ADV_1000FDX_CAP, 0, 1, 1);
2881 	igb->param_adv_100fdx_cap = igb_get_prop(igb,
2882 	    PROP_ADV_100FDX_CAP, 0, 1, 1);
2883 	igb->param_adv_100hdx_cap = igb_get_prop(igb,
2884 	    PROP_ADV_100HDX_CAP, 0, 1, 1);
2885 	igb->param_adv_10fdx_cap = igb_get_prop(igb,
2886 	    PROP_ADV_10FDX_CAP, 0, 1, 1);
2887 	igb->param_adv_10hdx_cap = igb_get_prop(igb,
2888 	    PROP_ADV_10HDX_CAP, 0, 1, 1);
2889 
2890 	/*
2891 	 * Jumbo frame configurations
2892 	 */
2893 	default_mtu = igb_get_prop(igb, PROP_DEFAULT_MTU,
2894 	    MIN_MTU, MAX_MTU, DEFAULT_MTU);
2895 
2896 	igb->max_frame_size = default_mtu +
2897 	    sizeof (struct ether_vlan_header) + ETHERFCSL;
2898 
2899 	/*
2900 	 * Ethernet flow control configuration
2901 	 */
2902 	flow_control = igb_get_prop(igb, PROP_FLOW_CONTROL,
2903 	    e1000_fc_none, 4, e1000_fc_full);
2904 	if (flow_control == 4)
2905 		flow_control = e1000_fc_default;
2906 
2907 	hw->fc.requested_mode = flow_control;
2908 
2909 	/*
2910 	 * Multiple rings configurations
2911 	 */
2912 	igb->tx_ring_size = igb_get_prop(igb, PROP_TX_RING_SIZE,
2913 	    MIN_TX_RING_SIZE, MAX_TX_RING_SIZE, DEFAULT_TX_RING_SIZE);
2914 	igb->rx_ring_size = igb_get_prop(igb, PROP_RX_RING_SIZE,
2915 	    MIN_RX_RING_SIZE, MAX_RX_RING_SIZE, DEFAULT_RX_RING_SIZE);
2916 
2917 	igb->mr_enable = igb_get_prop(igb, PROP_MR_ENABLE, 0, 1, 0);
2918 	igb->num_rx_groups = igb_get_prop(igb, PROP_RX_GROUP_NUM,
2919 	    MIN_RX_GROUP_NUM, MAX_RX_GROUP_NUM, DEFAULT_RX_GROUP_NUM);
2920 	/*
2921 	 * Currently we do not support VMDq for 82576 and 82580.
2922 	 * If it is e1000_82576, set num_rx_groups to 1.
2923 	 */
2924 	if (hw->mac.type >= e1000_82576)
2925 		igb->num_rx_groups = 1;
2926 
2927 	if (igb->mr_enable) {
2928 		igb->num_tx_rings = igb->capab->def_tx_que_num;
2929 		igb->num_rx_rings = igb->capab->def_rx_que_num;
2930 	} else {
2931 		igb->num_tx_rings = 1;
2932 		igb->num_rx_rings = 1;
2933 
2934 		if (igb->num_rx_groups > 1) {
2935 			igb_error(igb,
2936 			    "Invalid rx groups number. Please enable multiple "
2937 			    "rings first");
2938 			igb->num_rx_groups = 1;
2939 		}
2940 	}
2941 
2942 	/*
2943 	 * Check the divisibility between rx rings and rx groups.
2944 	 */
2945 	for (i = igb->num_rx_groups; i > 0; i--) {
2946 		if ((igb->num_rx_rings % i) == 0)
2947 			break;
2948 	}
2949 	if (i != igb->num_rx_groups) {
2950 		igb_error(igb,
2951 		    "Invalid rx groups number. Downgrade the rx group "
2952 		    "number to %d.", i);
2953 		igb->num_rx_groups = i;
2954 	}
2955 
2956 	/*
2957 	 * Get the ring number per group.
2958 	 */
2959 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
2960 
2961 	if (igb->num_rx_groups == 1) {
2962 		/*
2963 		 * One rx ring group, the rx ring number is num_rx_rings.
2964 		 */
2965 		igb->vmdq_mode = E1000_VMDQ_OFF;
2966 	} else if (ring_per_group == 1) {
2967 		/*
2968 		 * Multiple rx groups, each group has one rx ring.
2969 		 */
2970 		igb->vmdq_mode = E1000_VMDQ_MAC;
2971 	} else {
2972 		/*
2973 		 * Multiple groups and multiple rings.
2974 		 */
2975 		igb->vmdq_mode = E1000_VMDQ_MAC_RSS;
2976 	}
2977 
2978 	/*
2979 	 * Tunable used to force an interrupt type. The only use is
2980 	 * for testing of the lesser interrupt types.
2981 	 * 0 = don't force interrupt type
2982 	 * 1 = force interrupt type MSIX
2983 	 * 2 = force interrupt type MSI
2984 	 * 3 = force interrupt type Legacy
2985 	 */
2986 	igb->intr_force = igb_get_prop(igb, PROP_INTR_FORCE,
2987 	    IGB_INTR_NONE, IGB_INTR_LEGACY, IGB_INTR_NONE);
2988 
2989 	igb->tx_hcksum_enable = igb_get_prop(igb, PROP_TX_HCKSUM_ENABLE,
2990 	    0, 1, 1);
2991 	igb->rx_hcksum_enable = igb_get_prop(igb, PROP_RX_HCKSUM_ENABLE,
2992 	    0, 1, 1);
2993 	igb->lso_enable = igb_get_prop(igb, PROP_LSO_ENABLE,
2994 	    0, 1, 1);
2995 	igb->tx_head_wb_enable = igb_get_prop(igb, PROP_TX_HEAD_WB_ENABLE,
2996 	    0, 1, 1);
2997 
2998 	/*
2999 	 * igb LSO needs the tx h/w checksum support.
3000 	 * Here LSO will be disabled if tx h/w checksum has been disabled.
3001 	 */
3002 	if (igb->tx_hcksum_enable == B_FALSE)
3003 		igb->lso_enable = B_FALSE;
3004 
3005 	igb->tx_copy_thresh = igb_get_prop(igb, PROP_TX_COPY_THRESHOLD,
3006 	    MIN_TX_COPY_THRESHOLD, MAX_TX_COPY_THRESHOLD,
3007 	    DEFAULT_TX_COPY_THRESHOLD);
3008 	igb->tx_recycle_thresh = igb_get_prop(igb, PROP_TX_RECYCLE_THRESHOLD,
3009 	    MIN_TX_RECYCLE_THRESHOLD, MAX_TX_RECYCLE_THRESHOLD,
3010 	    DEFAULT_TX_RECYCLE_THRESHOLD);
3011 	igb->tx_overload_thresh = igb_get_prop(igb, PROP_TX_OVERLOAD_THRESHOLD,
3012 	    MIN_TX_OVERLOAD_THRESHOLD, MAX_TX_OVERLOAD_THRESHOLD,
3013 	    DEFAULT_TX_OVERLOAD_THRESHOLD);
3014 	igb->tx_resched_thresh = igb_get_prop(igb, PROP_TX_RESCHED_THRESHOLD,
3015 	    MIN_TX_RESCHED_THRESHOLD,
3016 	    MIN(igb->tx_ring_size, MAX_TX_RESCHED_THRESHOLD),
3017 	    igb->tx_ring_size > DEFAULT_TX_RESCHED_THRESHOLD ?
3018 	    DEFAULT_TX_RESCHED_THRESHOLD : DEFAULT_TX_RESCHED_THRESHOLD_LOW);
3019 
3020 	igb->rx_copy_thresh = igb_get_prop(igb, PROP_RX_COPY_THRESHOLD,
3021 	    MIN_RX_COPY_THRESHOLD, MAX_RX_COPY_THRESHOLD,
3022 	    DEFAULT_RX_COPY_THRESHOLD);
3023 	igb->rx_limit_per_intr = igb_get_prop(igb, PROP_RX_LIMIT_PER_INTR,
3024 	    MIN_RX_LIMIT_PER_INTR, MAX_RX_LIMIT_PER_INTR,
3025 	    DEFAULT_RX_LIMIT_PER_INTR);
3026 
3027 	igb->intr_throttling[0] = igb_get_prop(igb, PROP_INTR_THROTTLING,
3028 	    igb->capab->min_intr_throttle,
3029 	    igb->capab->max_intr_throttle,
3030 	    igb->capab->def_intr_throttle);
3031 
3032 	/*
3033 	 * Max number of multicast addresses
3034 	 */
3035 	igb->mcast_max_num =
3036 	    igb_get_prop(igb, PROP_MCAST_MAX_NUM,
3037 	    MIN_MCAST_NUM, MAX_MCAST_NUM, DEFAULT_MCAST_NUM);
3038 }
3039 
3040 /*
3041  * igb_get_prop - Get a property value out of the configuration file igb.conf
3042  *
3043  * Caller provides the name of the property, a default value, a minimum
3044  * value, and a maximum value.
3045  *
3046  * Return configured value of the property, with default, minimum and
3047  * maximum properly applied.
3048  */
3049 static int
3050 igb_get_prop(igb_t *igb,
3051     char *propname,	/* name of the property */
3052     int minval,		/* minimum acceptable value */
3053     int maxval,		/* maximim acceptable value */
3054     int defval)		/* default value */
3055 {
3056 	int value;
3057 
3058 	/*
3059 	 * Call ddi_prop_get_int() to read the conf settings
3060 	 */
3061 	value = ddi_prop_get_int(DDI_DEV_T_ANY, igb->dip,
3062 	    DDI_PROP_DONTPASS, propname, defval);
3063 
3064 	if (value > maxval)
3065 		value = maxval;
3066 
3067 	if (value < minval)
3068 		value = minval;
3069 
3070 	return (value);
3071 }
3072 
3073 /*
3074  * igb_setup_link - Using the link properties to setup the link
3075  */
3076 int
3077 igb_setup_link(igb_t *igb, boolean_t setup_hw)
3078 {
3079 	struct e1000_mac_info *mac;
3080 	struct e1000_phy_info *phy;
3081 	boolean_t invalid;
3082 
3083 	mac = &igb->hw.mac;
3084 	phy = &igb->hw.phy;
3085 	invalid = B_FALSE;
3086 
3087 	if (igb->param_adv_autoneg_cap == 1) {
3088 		mac->autoneg = B_TRUE;
3089 		phy->autoneg_advertised = 0;
3090 
3091 		/*
3092 		 * 1000hdx is not supported for autonegotiation
3093 		 */
3094 		if (igb->param_adv_1000fdx_cap == 1)
3095 			phy->autoneg_advertised |= ADVERTISE_1000_FULL;
3096 
3097 		if (igb->param_adv_100fdx_cap == 1)
3098 			phy->autoneg_advertised |= ADVERTISE_100_FULL;
3099 
3100 		if (igb->param_adv_100hdx_cap == 1)
3101 			phy->autoneg_advertised |= ADVERTISE_100_HALF;
3102 
3103 		if (igb->param_adv_10fdx_cap == 1)
3104 			phy->autoneg_advertised |= ADVERTISE_10_FULL;
3105 
3106 		if (igb->param_adv_10hdx_cap == 1)
3107 			phy->autoneg_advertised |= ADVERTISE_10_HALF;
3108 
3109 		if (phy->autoneg_advertised == 0)
3110 			invalid = B_TRUE;
3111 	} else {
3112 		mac->autoneg = B_FALSE;
3113 
3114 		/*
3115 		 * 1000fdx and 1000hdx are not supported for forced link
3116 		 */
3117 		if (igb->param_adv_100fdx_cap == 1)
3118 			mac->forced_speed_duplex = ADVERTISE_100_FULL;
3119 		else if (igb->param_adv_100hdx_cap == 1)
3120 			mac->forced_speed_duplex = ADVERTISE_100_HALF;
3121 		else if (igb->param_adv_10fdx_cap == 1)
3122 			mac->forced_speed_duplex = ADVERTISE_10_FULL;
3123 		else if (igb->param_adv_10hdx_cap == 1)
3124 			mac->forced_speed_duplex = ADVERTISE_10_HALF;
3125 		else
3126 			invalid = B_TRUE;
3127 	}
3128 
3129 	if (invalid) {
3130 		igb_notice(igb, "Invalid link settings. Setup link to "
3131 		    "autonegotiation with full link capabilities.");
3132 		mac->autoneg = B_TRUE;
3133 		phy->autoneg_advertised = ADVERTISE_1000_FULL |
3134 		    ADVERTISE_100_FULL | ADVERTISE_100_HALF |
3135 		    ADVERTISE_10_FULL | ADVERTISE_10_HALF;
3136 	}
3137 
3138 	if (setup_hw) {
3139 		if (e1000_setup_link(&igb->hw) != E1000_SUCCESS)
3140 			return (IGB_FAILURE);
3141 	}
3142 
3143 	return (IGB_SUCCESS);
3144 }
3145 
3146 
3147 /*
3148  * igb_is_link_up - Check if the link is up
3149  */
3150 static boolean_t
3151 igb_is_link_up(igb_t *igb)
3152 {
3153 	struct e1000_hw *hw = &igb->hw;
3154 	boolean_t link_up = B_FALSE;
3155 
3156 	ASSERT(mutex_owned(&igb->gen_lock));
3157 
3158 	/*
3159 	 * get_link_status is set in the interrupt handler on link-status-change
3160 	 * or rx sequence error interrupt.  get_link_status will stay
3161 	 * false until the e1000_check_for_link establishes link only
3162 	 * for copper adapters.
3163 	 */
3164 	switch (hw->phy.media_type) {
3165 	case e1000_media_type_copper:
3166 		if (hw->mac.get_link_status) {
3167 			(void) e1000_check_for_link(hw);
3168 			link_up = !hw->mac.get_link_status;
3169 		} else {
3170 			link_up = B_TRUE;
3171 		}
3172 		break;
3173 	case e1000_media_type_fiber:
3174 		(void) e1000_check_for_link(hw);
3175 		link_up = (E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU);
3176 		break;
3177 	case e1000_media_type_internal_serdes:
3178 		(void) e1000_check_for_link(hw);
3179 		link_up = hw->mac.serdes_has_link;
3180 		break;
3181 	}
3182 
3183 	return (link_up);
3184 }
3185 
3186 /*
3187  * igb_link_check - Link status processing
3188  */
3189 static boolean_t
3190 igb_link_check(igb_t *igb)
3191 {
3192 	struct e1000_hw *hw = &igb->hw;
3193 	uint16_t speed = 0, duplex = 0;
3194 	boolean_t link_changed = B_FALSE;
3195 
3196 	ASSERT(mutex_owned(&igb->gen_lock));
3197 
3198 	if (igb_is_link_up(igb)) {
3199 		/*
3200 		 * The Link is up, check whether it was marked as down earlier
3201 		 */
3202 		if (igb->link_state != LINK_STATE_UP) {
3203 			(void) e1000_get_speed_and_duplex(hw, &speed, &duplex);
3204 			igb->link_speed = speed;
3205 			igb->link_duplex = duplex;
3206 			igb->link_state = LINK_STATE_UP;
3207 			link_changed = B_TRUE;
3208 			if (!igb->link_complete)
3209 				igb_stop_link_timer(igb);
3210 		}
3211 	} else if (igb->link_complete) {
3212 		if (igb->link_state != LINK_STATE_DOWN) {
3213 			igb->link_speed = 0;
3214 			igb->link_duplex = 0;
3215 			igb->link_state = LINK_STATE_DOWN;
3216 			link_changed = B_TRUE;
3217 		}
3218 	}
3219 
3220 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
3221 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
3222 		return (B_FALSE);
3223 	}
3224 
3225 	return (link_changed);
3226 }
3227 
3228 /*
3229  * igb_local_timer - driver watchdog function
3230  *
3231  * This function will handle the hardware stall check, link status
3232  * check and other routines.
3233  */
3234 static void
3235 igb_local_timer(void *arg)
3236 {
3237 	igb_t *igb = (igb_t *)arg;
3238 	boolean_t link_changed = B_FALSE;
3239 
3240 	if (igb->igb_state & IGB_ERROR) {
3241 		igb->reset_count++;
3242 		if (igb_reset(igb) == IGB_SUCCESS)
3243 			ddi_fm_service_impact(igb->dip, DDI_SERVICE_RESTORED);
3244 
3245 		igb_restart_watchdog_timer(igb);
3246 		return;
3247 	}
3248 
3249 	if (igb_stall_check(igb) || (igb->igb_state & IGB_STALL)) {
3250 		igb_fm_ereport(igb, DDI_FM_DEVICE_STALL);
3251 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
3252 		igb->reset_count++;
3253 		if (igb_reset(igb) == IGB_SUCCESS)
3254 			ddi_fm_service_impact(igb->dip, DDI_SERVICE_RESTORED);
3255 
3256 		igb_restart_watchdog_timer(igb);
3257 		return;
3258 	}
3259 
3260 	mutex_enter(&igb->gen_lock);
3261 	if (!(igb->igb_state & IGB_SUSPENDED) && (igb->igb_state & IGB_STARTED))
3262 		link_changed = igb_link_check(igb);
3263 	mutex_exit(&igb->gen_lock);
3264 
3265 	if (link_changed)
3266 		mac_link_update(igb->mac_hdl, igb->link_state);
3267 
3268 	igb_restart_watchdog_timer(igb);
3269 }
3270 
3271 /*
3272  * igb_link_timer - link setup timer function
3273  *
3274  * It is called when the timer for link setup is expired, which indicates
3275  * the completion of the link setup. The link state will not be updated
3276  * until the link setup is completed. And the link state will not be sent
3277  * to the upper layer through mac_link_update() in this function. It will
3278  * be updated in the local timer routine or the interrupts service routine
3279  * after the interface is started (plumbed).
3280  */
3281 static void
3282 igb_link_timer(void *arg)
3283 {
3284 	igb_t *igb = (igb_t *)arg;
3285 
3286 	mutex_enter(&igb->link_lock);
3287 	igb->link_complete = B_TRUE;
3288 	igb->link_tid = 0;
3289 	mutex_exit(&igb->link_lock);
3290 }
3291 /*
3292  * igb_stall_check - check for transmit stall
3293  *
3294  * This function checks if the adapter is stalled (in transmit).
3295  *
3296  * It is called each time the watchdog timeout is invoked.
3297  * If the transmit descriptor reclaim continuously fails,
3298  * the watchdog value will increment by 1. If the watchdog
3299  * value exceeds the threshold, the igb is assumed to
3300  * have stalled and need to be reset.
3301  */
3302 static boolean_t
3303 igb_stall_check(igb_t *igb)
3304 {
3305 	igb_tx_ring_t *tx_ring;
3306 	struct e1000_hw *hw = &igb->hw;
3307 	boolean_t result;
3308 	int i;
3309 
3310 	if (igb->link_state != LINK_STATE_UP)
3311 		return (B_FALSE);
3312 
3313 	/*
3314 	 * If any tx ring is stalled, we'll reset the chipset
3315 	 */
3316 	result = B_FALSE;
3317 	for (i = 0; i < igb->num_tx_rings; i++) {
3318 		tx_ring = &igb->tx_rings[i];
3319 
3320 		if (tx_ring->recycle_fail > 0)
3321 			tx_ring->stall_watchdog++;
3322 		else
3323 			tx_ring->stall_watchdog = 0;
3324 
3325 		if (tx_ring->stall_watchdog >= STALL_WATCHDOG_TIMEOUT) {
3326 			result = B_TRUE;
3327 			if (hw->mac.type == e1000_82580) {
3328 				hw->dev_spec._82575.global_device_reset
3329 				    = B_TRUE;
3330 			}
3331 			break;
3332 		}
3333 	}
3334 
3335 	if (result) {
3336 		tx_ring->stall_watchdog = 0;
3337 		tx_ring->recycle_fail = 0;
3338 	}
3339 
3340 	return (result);
3341 }
3342 
3343 
3344 /*
3345  * is_valid_mac_addr - Check if the mac address is valid
3346  */
3347 static boolean_t
3348 is_valid_mac_addr(uint8_t *mac_addr)
3349 {
3350 	const uint8_t addr_test1[6] = { 0, 0, 0, 0, 0, 0 };
3351 	const uint8_t addr_test2[6] =
3352 	    { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF };
3353 
3354 	if (!(bcmp(addr_test1, mac_addr, ETHERADDRL)) ||
3355 	    !(bcmp(addr_test2, mac_addr, ETHERADDRL)))
3356 		return (B_FALSE);
3357 
3358 	return (B_TRUE);
3359 }
3360 
3361 static boolean_t
3362 igb_find_mac_address(igb_t *igb)
3363 {
3364 	struct e1000_hw *hw = &igb->hw;
3365 #ifdef __sparc
3366 	uchar_t *bytes;
3367 	struct ether_addr sysaddr;
3368 	uint_t nelts;
3369 	int err;
3370 	boolean_t found = B_FALSE;
3371 
3372 	/*
3373 	 * The "vendor's factory-set address" may already have
3374 	 * been extracted from the chip, but if the property
3375 	 * "local-mac-address" is set we use that instead.
3376 	 *
3377 	 * We check whether it looks like an array of 6
3378 	 * bytes (which it should, if OBP set it).  If we can't
3379 	 * make sense of it this way, we'll ignore it.
3380 	 */
3381 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip,
3382 	    DDI_PROP_DONTPASS, "local-mac-address", &bytes, &nelts);
3383 	if (err == DDI_PROP_SUCCESS) {
3384 		if (nelts == ETHERADDRL) {
3385 			while (nelts--)
3386 				hw->mac.addr[nelts] = bytes[nelts];
3387 			found = B_TRUE;
3388 		}
3389 		ddi_prop_free(bytes);
3390 	}
3391 
3392 	/*
3393 	 * Look up the OBP property "local-mac-address?". If the user has set
3394 	 * 'local-mac-address? = false', use "the system address" instead.
3395 	 */
3396 	if (ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip, 0,
3397 	    "local-mac-address?", &bytes, &nelts) == DDI_PROP_SUCCESS) {
3398 		if (strncmp("false", (caddr_t)bytes, (size_t)nelts) == 0) {
3399 			if (localetheraddr(NULL, &sysaddr) != 0) {
3400 				bcopy(&sysaddr, hw->mac.addr, ETHERADDRL);
3401 				found = B_TRUE;
3402 			}
3403 		}
3404 		ddi_prop_free(bytes);
3405 	}
3406 
3407 	/*
3408 	 * Finally(!), if there's a valid "mac-address" property (created
3409 	 * if we netbooted from this interface), we must use this instead
3410 	 * of any of the above to ensure that the NFS/install server doesn't
3411 	 * get confused by the address changing as Solaris takes over!
3412 	 */
3413 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip,
3414 	    DDI_PROP_DONTPASS, "mac-address", &bytes, &nelts);
3415 	if (err == DDI_PROP_SUCCESS) {
3416 		if (nelts == ETHERADDRL) {
3417 			while (nelts--)
3418 				hw->mac.addr[nelts] = bytes[nelts];
3419 			found = B_TRUE;
3420 		}
3421 		ddi_prop_free(bytes);
3422 	}
3423 
3424 	if (found) {
3425 		bcopy(hw->mac.addr, hw->mac.perm_addr, ETHERADDRL);
3426 		return (B_TRUE);
3427 	}
3428 #endif
3429 
3430 	/*
3431 	 * Read the device MAC address from the EEPROM
3432 	 */
3433 	if (e1000_read_mac_addr(hw) != E1000_SUCCESS)
3434 		return (B_FALSE);
3435 
3436 	return (B_TRUE);
3437 }
3438 
3439 #pragma inline(igb_arm_watchdog_timer)
3440 
3441 static void
3442 igb_arm_watchdog_timer(igb_t *igb)
3443 {
3444 	/*
3445 	 * Fire a watchdog timer
3446 	 */
3447 	igb->watchdog_tid =
3448 	    timeout(igb_local_timer,
3449 	    (void *)igb, 1 * drv_usectohz(1000000));
3450 
3451 }
3452 
3453 /*
3454  * igb_enable_watchdog_timer - Enable and start the driver watchdog timer
3455  */
3456 void
3457 igb_enable_watchdog_timer(igb_t *igb)
3458 {
3459 	mutex_enter(&igb->watchdog_lock);
3460 
3461 	if (!igb->watchdog_enable) {
3462 		igb->watchdog_enable = B_TRUE;
3463 		igb->watchdog_start = B_TRUE;
3464 		igb_arm_watchdog_timer(igb);
3465 	}
3466 
3467 	mutex_exit(&igb->watchdog_lock);
3468 
3469 }
3470 
3471 /*
3472  * igb_disable_watchdog_timer - Disable and stop the driver watchdog timer
3473  */
3474 void
3475 igb_disable_watchdog_timer(igb_t *igb)
3476 {
3477 	timeout_id_t tid;
3478 
3479 	mutex_enter(&igb->watchdog_lock);
3480 
3481 	igb->watchdog_enable = B_FALSE;
3482 	igb->watchdog_start = B_FALSE;
3483 	tid = igb->watchdog_tid;
3484 	igb->watchdog_tid = 0;
3485 
3486 	mutex_exit(&igb->watchdog_lock);
3487 
3488 	if (tid != 0)
3489 		(void) untimeout(tid);
3490 
3491 }
3492 
3493 /*
3494  * igb_start_watchdog_timer - Start the driver watchdog timer
3495  */
3496 static void
3497 igb_start_watchdog_timer(igb_t *igb)
3498 {
3499 	mutex_enter(&igb->watchdog_lock);
3500 
3501 	if (igb->watchdog_enable) {
3502 		if (!igb->watchdog_start) {
3503 			igb->watchdog_start = B_TRUE;
3504 			igb_arm_watchdog_timer(igb);
3505 		}
3506 	}
3507 
3508 	mutex_exit(&igb->watchdog_lock);
3509 }
3510 
3511 /*
3512  * igb_restart_watchdog_timer - Restart the driver watchdog timer
3513  */
3514 static void
3515 igb_restart_watchdog_timer(igb_t *igb)
3516 {
3517 	mutex_enter(&igb->watchdog_lock);
3518 
3519 	if (igb->watchdog_start)
3520 		igb_arm_watchdog_timer(igb);
3521 
3522 	mutex_exit(&igb->watchdog_lock);
3523 }
3524 
3525 /*
3526  * igb_stop_watchdog_timer - Stop the driver watchdog timer
3527  */
3528 static void
3529 igb_stop_watchdog_timer(igb_t *igb)
3530 {
3531 	timeout_id_t tid;
3532 
3533 	mutex_enter(&igb->watchdog_lock);
3534 
3535 	igb->watchdog_start = B_FALSE;
3536 	tid = igb->watchdog_tid;
3537 	igb->watchdog_tid = 0;
3538 
3539 	mutex_exit(&igb->watchdog_lock);
3540 
3541 	if (tid != 0)
3542 		(void) untimeout(tid);
3543 }
3544 
3545 /*
3546  * igb_start_link_timer - Start the link setup timer
3547  */
3548 static void
3549 igb_start_link_timer(struct igb *igb)
3550 {
3551 	struct e1000_hw *hw = &igb->hw;
3552 	clock_t link_timeout;
3553 
3554 	if (hw->mac.autoneg)
3555 		link_timeout = PHY_AUTO_NEG_LIMIT *
3556 		    drv_usectohz(100000);
3557 	else
3558 		link_timeout = PHY_FORCE_LIMIT * drv_usectohz(100000);
3559 
3560 	mutex_enter(&igb->link_lock);
3561 	if (hw->phy.autoneg_wait_to_complete) {
3562 		igb->link_complete = B_TRUE;
3563 	} else {
3564 		igb->link_complete = B_FALSE;
3565 		igb->link_tid = timeout(igb_link_timer, (void *)igb,
3566 		    link_timeout);
3567 	}
3568 	mutex_exit(&igb->link_lock);
3569 }
3570 
3571 /*
3572  * igb_stop_link_timer - Stop the link setup timer
3573  */
3574 static void
3575 igb_stop_link_timer(struct igb *igb)
3576 {
3577 	timeout_id_t tid;
3578 
3579 	mutex_enter(&igb->link_lock);
3580 	igb->link_complete = B_TRUE;
3581 	tid = igb->link_tid;
3582 	igb->link_tid = 0;
3583 	mutex_exit(&igb->link_lock);
3584 
3585 	if (tid != 0)
3586 		(void) untimeout(tid);
3587 }
3588 
3589 /*
3590  * igb_disable_adapter_interrupts - Clear/disable all hardware interrupts
3591  */
3592 static void
3593 igb_disable_adapter_interrupts(igb_t *igb)
3594 {
3595 	struct e1000_hw *hw = &igb->hw;
3596 
3597 	/*
3598 	 * Set the IMC register to mask all the interrupts,
3599 	 * including the tx interrupts.
3600 	 */
3601 	E1000_WRITE_REG(hw, E1000_IMC, ~0);
3602 	E1000_WRITE_REG(hw, E1000_IAM, 0);
3603 
3604 	/*
3605 	 * Additional disabling for MSI-X
3606 	 */
3607 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3608 		E1000_WRITE_REG(hw, E1000_EIMC, ~0);
3609 		E1000_WRITE_REG(hw, E1000_EIAC, 0);
3610 		E1000_WRITE_REG(hw, E1000_EIAM, 0);
3611 	}
3612 
3613 	E1000_WRITE_FLUSH(hw);
3614 }
3615 
3616 /*
3617  * igb_enable_adapter_interrupts_82580 - Enable NIC interrupts for 82580
3618  */
3619 static void
3620 igb_enable_adapter_interrupts_82580(igb_t *igb)
3621 {
3622 	struct e1000_hw *hw = &igb->hw;
3623 
3624 	/* Clear any pending interrupts */
3625 	(void) E1000_READ_REG(hw, E1000_ICR);
3626 	igb->ims_mask |= E1000_IMS_DRSTA;
3627 
3628 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3629 
3630 		/* Interrupt enabling for MSI-X */
3631 		E1000_WRITE_REG(hw, E1000_EIMS, igb->eims_mask);
3632 		E1000_WRITE_REG(hw, E1000_EIAC, igb->eims_mask);
3633 		igb->ims_mask = (E1000_IMS_LSC | E1000_IMS_DRSTA);
3634 		E1000_WRITE_REG(hw, E1000_IMS, igb->ims_mask);
3635 	} else { /* Interrupt enabling for MSI and legacy */
3636 		E1000_WRITE_REG(hw, E1000_IVAR0, E1000_IVAR_VALID);
3637 		igb->ims_mask = IMS_ENABLE_MASK | E1000_IMS_TXQE;
3638 		igb->ims_mask |= E1000_IMS_DRSTA;
3639 		E1000_WRITE_REG(hw, E1000_IMS, igb->ims_mask);
3640 	}
3641 
3642 	/* Disable auto-mask for ICR interrupt bits */
3643 	E1000_WRITE_REG(hw, E1000_IAM, 0);
3644 
3645 	E1000_WRITE_FLUSH(hw);
3646 }
3647 
3648 /*
3649  * igb_enable_adapter_interrupts_82576 - Enable NIC interrupts for 82576
3650  */
3651 static void
3652 igb_enable_adapter_interrupts_82576(igb_t *igb)
3653 {
3654 	struct e1000_hw *hw = &igb->hw;
3655 
3656 	/* Clear any pending interrupts */
3657 	(void) E1000_READ_REG(hw, E1000_ICR);
3658 
3659 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3660 
3661 		/* Interrupt enabling for MSI-X */
3662 		E1000_WRITE_REG(hw, E1000_EIMS, igb->eims_mask);
3663 		E1000_WRITE_REG(hw, E1000_EIAC, igb->eims_mask);
3664 		igb->ims_mask = E1000_IMS_LSC;
3665 		E1000_WRITE_REG(hw, E1000_IMS, E1000_IMS_LSC);
3666 	} else {
3667 		/* Interrupt enabling for MSI and legacy */
3668 		E1000_WRITE_REG(hw, E1000_IVAR0, E1000_IVAR_VALID);
3669 		igb->ims_mask = IMS_ENABLE_MASK | E1000_IMS_TXQE;
3670 		E1000_WRITE_REG(hw, E1000_IMS,
3671 		    (IMS_ENABLE_MASK | E1000_IMS_TXQE));
3672 	}
3673 
3674 	/* Disable auto-mask for ICR interrupt bits */
3675 	E1000_WRITE_REG(hw, E1000_IAM, 0);
3676 
3677 	E1000_WRITE_FLUSH(hw);
3678 }
3679 
3680 /*
3681  * igb_enable_adapter_interrupts_82575 - Enable NIC interrupts for 82575
3682  */
3683 static void
3684 igb_enable_adapter_interrupts_82575(igb_t *igb)
3685 {
3686 	struct e1000_hw *hw = &igb->hw;
3687 	uint32_t reg;
3688 
3689 	/* Clear any pending interrupts */
3690 	(void) E1000_READ_REG(hw, E1000_ICR);
3691 
3692 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3693 		/* Interrupt enabling for MSI-X */
3694 		E1000_WRITE_REG(hw, E1000_EIMS, igb->eims_mask);
3695 		E1000_WRITE_REG(hw, E1000_EIAC, igb->eims_mask);
3696 		igb->ims_mask = E1000_IMS_LSC;
3697 		E1000_WRITE_REG(hw, E1000_IMS, E1000_IMS_LSC);
3698 
3699 		/* Enable MSI-X PBA support */
3700 		reg = E1000_READ_REG(hw, E1000_CTRL_EXT);
3701 		reg |= E1000_CTRL_EXT_PBA_CLR;
3702 
3703 		/* Non-selective interrupt clear-on-read */
3704 		reg |= E1000_CTRL_EXT_IRCA;	/* Called NSICR in the EAS */
3705 
3706 		E1000_WRITE_REG(hw, E1000_CTRL_EXT, reg);
3707 	} else {
3708 		/* Interrupt enabling for MSI and legacy */
3709 		igb->ims_mask = IMS_ENABLE_MASK;
3710 		E1000_WRITE_REG(hw, E1000_IMS, IMS_ENABLE_MASK);
3711 	}
3712 
3713 	E1000_WRITE_FLUSH(hw);
3714 }
3715 
3716 /*
3717  * Loopback Support
3718  */
3719 static lb_property_t lb_normal =
3720 	{ normal,	"normal",	IGB_LB_NONE		};
3721 static lb_property_t lb_external =
3722 	{ external,	"External",	IGB_LB_EXTERNAL		};
3723 static lb_property_t lb_phy =
3724 	{ internal,	"PHY",		IGB_LB_INTERNAL_PHY	};
3725 static lb_property_t lb_serdes =
3726 	{ internal,	"SerDes",	IGB_LB_INTERNAL_SERDES	};
3727 
3728 enum ioc_reply
3729 igb_loopback_ioctl(igb_t *igb, struct iocblk *iocp, mblk_t *mp)
3730 {
3731 	lb_info_sz_t *lbsp;
3732 	lb_property_t *lbpp;
3733 	struct e1000_hw *hw;
3734 	uint32_t *lbmp;
3735 	uint32_t size;
3736 	uint32_t value;
3737 
3738 	hw = &igb->hw;
3739 
3740 	if (mp->b_cont == NULL)
3741 		return (IOC_INVAL);
3742 
3743 	switch (iocp->ioc_cmd) {
3744 	default:
3745 		return (IOC_INVAL);
3746 
3747 	case LB_GET_INFO_SIZE:
3748 		size = sizeof (lb_info_sz_t);
3749 		if (iocp->ioc_count != size)
3750 			return (IOC_INVAL);
3751 
3752 		value = sizeof (lb_normal);
3753 		if (hw->phy.media_type == e1000_media_type_copper)
3754 			value += sizeof (lb_phy);
3755 		else
3756 			value += sizeof (lb_serdes);
3757 		value += sizeof (lb_external);
3758 
3759 		lbsp = (lb_info_sz_t *)(uintptr_t)mp->b_cont->b_rptr;
3760 		*lbsp = value;
3761 		break;
3762 
3763 	case LB_GET_INFO:
3764 		value = sizeof (lb_normal);
3765 		if (hw->phy.media_type == e1000_media_type_copper)
3766 			value += sizeof (lb_phy);
3767 		else
3768 			value += sizeof (lb_serdes);
3769 		value += sizeof (lb_external);
3770 
3771 		size = value;
3772 		if (iocp->ioc_count != size)
3773 			return (IOC_INVAL);
3774 
3775 		value = 0;
3776 		lbpp = (lb_property_t *)(uintptr_t)mp->b_cont->b_rptr;
3777 
3778 		lbpp[value++] = lb_normal;
3779 		if (hw->phy.media_type == e1000_media_type_copper)
3780 			lbpp[value++] = lb_phy;
3781 		else
3782 			lbpp[value++] = lb_serdes;
3783 		lbpp[value++] = lb_external;
3784 		break;
3785 
3786 	case LB_GET_MODE:
3787 		size = sizeof (uint32_t);
3788 		if (iocp->ioc_count != size)
3789 			return (IOC_INVAL);
3790 
3791 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
3792 		*lbmp = igb->loopback_mode;
3793 		break;
3794 
3795 	case LB_SET_MODE:
3796 		size = 0;
3797 		if (iocp->ioc_count != sizeof (uint32_t))
3798 			return (IOC_INVAL);
3799 
3800 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
3801 		if (!igb_set_loopback_mode(igb, *lbmp))
3802 			return (IOC_INVAL);
3803 		break;
3804 	}
3805 
3806 	iocp->ioc_count = size;
3807 	iocp->ioc_error = 0;
3808 
3809 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
3810 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
3811 		return (IOC_INVAL);
3812 	}
3813 
3814 	return (IOC_REPLY);
3815 }
3816 
3817 /*
3818  * igb_set_loopback_mode - Setup loopback based on the loopback mode
3819  */
3820 static boolean_t
3821 igb_set_loopback_mode(igb_t *igb, uint32_t mode)
3822 {
3823 	struct e1000_hw *hw;
3824 	int i;
3825 
3826 	if (mode == igb->loopback_mode)
3827 		return (B_TRUE);
3828 
3829 	hw = &igb->hw;
3830 
3831 	igb->loopback_mode = mode;
3832 
3833 	if (mode == IGB_LB_NONE) {
3834 		/* Reset the chip */
3835 		hw->phy.autoneg_wait_to_complete = B_TRUE;
3836 		(void) igb_reset(igb);
3837 		hw->phy.autoneg_wait_to_complete = B_FALSE;
3838 		return (B_TRUE);
3839 	}
3840 
3841 	mutex_enter(&igb->gen_lock);
3842 
3843 	switch (mode) {
3844 	default:
3845 		mutex_exit(&igb->gen_lock);
3846 		return (B_FALSE);
3847 
3848 	case IGB_LB_EXTERNAL:
3849 		igb_set_external_loopback(igb);
3850 		break;
3851 
3852 	case IGB_LB_INTERNAL_PHY:
3853 		igb_set_internal_phy_loopback(igb);
3854 		break;
3855 
3856 	case IGB_LB_INTERNAL_SERDES:
3857 		igb_set_internal_serdes_loopback(igb);
3858 		break;
3859 	}
3860 
3861 	mutex_exit(&igb->gen_lock);
3862 
3863 	/*
3864 	 * When external loopback is set, wait up to 1000ms to get the link up.
3865 	 * According to test, 1000ms can work and it's an experimental value.
3866 	 */
3867 	if (mode == IGB_LB_EXTERNAL) {
3868 		for (i = 0; i <= 10; i++) {
3869 			mutex_enter(&igb->gen_lock);
3870 			(void) igb_link_check(igb);
3871 			mutex_exit(&igb->gen_lock);
3872 
3873 			if (igb->link_state == LINK_STATE_UP)
3874 				break;
3875 
3876 			msec_delay(100);
3877 		}
3878 
3879 		if (igb->link_state != LINK_STATE_UP) {
3880 			/*
3881 			 * Does not support external loopback.
3882 			 * Reset driver to loopback none.
3883 			 */
3884 			igb->loopback_mode = IGB_LB_NONE;
3885 
3886 			/* Reset the chip */
3887 			hw->phy.autoneg_wait_to_complete = B_TRUE;
3888 			(void) igb_reset(igb);
3889 			hw->phy.autoneg_wait_to_complete = B_FALSE;
3890 
3891 			IGB_DEBUGLOG_0(igb, "Set external loopback failed, "
3892 			    "reset to loopback none.");
3893 
3894 			return (B_FALSE);
3895 		}
3896 	}
3897 
3898 	return (B_TRUE);
3899 }
3900 
3901 /*
3902  * igb_set_external_loopback - Set the external loopback mode
3903  */
3904 static void
3905 igb_set_external_loopback(igb_t *igb)
3906 {
3907 	struct e1000_hw *hw;
3908 	uint32_t ctrl_ext;
3909 
3910 	hw = &igb->hw;
3911 
3912 	/* Set link mode to PHY (00b) in the Extended Control register */
3913 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3914 	ctrl_ext &= ~E1000_CTRL_EXT_LINK_MODE_MASK;
3915 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3916 
3917 	(void) e1000_write_phy_reg(hw, 0x0, 0x0140);
3918 	(void) e1000_write_phy_reg(hw, 0x9, 0x1a00);
3919 	(void) e1000_write_phy_reg(hw, 0x12, 0x1610);
3920 	(void) e1000_write_phy_reg(hw, 0x1f37, 0x3f1c);
3921 }
3922 
3923 /*
3924  * igb_set_internal_phy_loopback - Set the internal PHY loopback mode
3925  */
3926 static void
3927 igb_set_internal_phy_loopback(igb_t *igb)
3928 {
3929 	struct e1000_hw *hw;
3930 	uint32_t ctrl_ext;
3931 	uint16_t phy_ctrl;
3932 	uint16_t phy_pconf;
3933 
3934 	hw = &igb->hw;
3935 
3936 	/* Set link mode to PHY (00b) in the Extended Control register */
3937 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3938 	ctrl_ext &= ~E1000_CTRL_EXT_LINK_MODE_MASK;
3939 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3940 
3941 	/*
3942 	 * Set PHY control register (0x4140):
3943 	 *    Set full duplex mode
3944 	 *    Set loopback bit
3945 	 *    Clear auto-neg enable bit
3946 	 *    Set PHY speed
3947 	 */
3948 	phy_ctrl = MII_CR_FULL_DUPLEX | MII_CR_SPEED_1000 | MII_CR_LOOPBACK;
3949 	(void) e1000_write_phy_reg(hw, PHY_CONTROL, phy_ctrl);
3950 
3951 	/* Set the link disable bit in the Port Configuration register */
3952 	(void) e1000_read_phy_reg(hw, 0x10, &phy_pconf);
3953 	phy_pconf |= (uint16_t)1 << 14;
3954 	(void) e1000_write_phy_reg(hw, 0x10, phy_pconf);
3955 }
3956 
3957 /*
3958  * igb_set_internal_serdes_loopback - Set the internal SerDes loopback mode
3959  */
3960 static void
3961 igb_set_internal_serdes_loopback(igb_t *igb)
3962 {
3963 	struct e1000_hw *hw;
3964 	uint32_t ctrl_ext;
3965 	uint32_t ctrl;
3966 	uint32_t pcs_lctl;
3967 	uint32_t connsw;
3968 
3969 	hw = &igb->hw;
3970 
3971 	/* Set link mode to SerDes (11b) in the Extended Control register */
3972 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3973 	ctrl_ext |= E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
3974 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3975 
3976 	/* Configure the SerDes to loopback */
3977 	E1000_WRITE_REG(hw, E1000_SCTL, 0x410);
3978 
3979 	/* Set Device Control register */
3980 	ctrl = E1000_READ_REG(hw, E1000_CTRL);
3981 	ctrl |= (E1000_CTRL_FD |	/* Force full duplex */
3982 	    E1000_CTRL_SLU);		/* Force link up */
3983 	ctrl &= ~(E1000_CTRL_RFCE |	/* Disable receive flow control */
3984 	    E1000_CTRL_TFCE |		/* Disable transmit flow control */
3985 	    E1000_CTRL_LRST);		/* Clear link reset */
3986 	E1000_WRITE_REG(hw, E1000_CTRL, ctrl);
3987 
3988 	/* Set PCS Link Control register */
3989 	pcs_lctl = E1000_READ_REG(hw, E1000_PCS_LCTL);
3990 	pcs_lctl |= (E1000_PCS_LCTL_FORCE_LINK |
3991 	    E1000_PCS_LCTL_FSD |
3992 	    E1000_PCS_LCTL_FDV_FULL |
3993 	    E1000_PCS_LCTL_FLV_LINK_UP);
3994 	pcs_lctl &= ~E1000_PCS_LCTL_AN_ENABLE;
3995 	E1000_WRITE_REG(hw, E1000_PCS_LCTL, pcs_lctl);
3996 
3997 	/* Set the Copper/Fiber Switch Control - CONNSW register */
3998 	connsw = E1000_READ_REG(hw, E1000_CONNSW);
3999 	connsw &= ~E1000_CONNSW_ENRGSRC;
4000 	E1000_WRITE_REG(hw, E1000_CONNSW, connsw);
4001 }
4002 
4003 #pragma inline(igb_intr_rx_work)
4004 /*
4005  * igb_intr_rx_work - rx processing of ISR
4006  */
4007 static void
4008 igb_intr_rx_work(igb_rx_ring_t *rx_ring)
4009 {
4010 	mblk_t *mp;
4011 
4012 	mutex_enter(&rx_ring->rx_lock);
4013 	mp = igb_rx(rx_ring, IGB_NO_POLL);
4014 	mutex_exit(&rx_ring->rx_lock);
4015 
4016 	if (mp != NULL)
4017 		mac_rx_ring(rx_ring->igb->mac_hdl, rx_ring->ring_handle, mp,
4018 		    rx_ring->ring_gen_num);
4019 }
4020 
4021 #pragma inline(igb_intr_tx_work)
4022 /*
4023  * igb_intr_tx_work - tx processing of ISR
4024  */
4025 static void
4026 igb_intr_tx_work(igb_tx_ring_t *tx_ring)
4027 {
4028 	igb_t *igb = tx_ring->igb;
4029 
4030 	/* Recycle the tx descriptors */
4031 	tx_ring->tx_recycle(tx_ring);
4032 
4033 	/* Schedule the re-transmit */
4034 	if (tx_ring->reschedule &&
4035 	    (tx_ring->tbd_free >= igb->tx_resched_thresh)) {
4036 		tx_ring->reschedule = B_FALSE;
4037 		mac_tx_ring_update(tx_ring->igb->mac_hdl, tx_ring->ring_handle);
4038 		IGB_DEBUG_STAT(tx_ring->stat_reschedule);
4039 	}
4040 }
4041 
4042 #pragma inline(igb_intr_link_work)
4043 /*
4044  * igb_intr_link_work - link-status-change processing of ISR
4045  */
4046 static void
4047 igb_intr_link_work(igb_t *igb)
4048 {
4049 	boolean_t link_changed;
4050 
4051 	igb_stop_watchdog_timer(igb);
4052 
4053 	mutex_enter(&igb->gen_lock);
4054 
4055 	/*
4056 	 * Because we got a link-status-change interrupt, force
4057 	 * e1000_check_for_link() to look at phy
4058 	 */
4059 	igb->hw.mac.get_link_status = B_TRUE;
4060 
4061 	/* igb_link_check takes care of link status change */
4062 	link_changed = igb_link_check(igb);
4063 
4064 	/* Get new phy state */
4065 	igb_get_phy_state(igb);
4066 
4067 	mutex_exit(&igb->gen_lock);
4068 
4069 	if (link_changed)
4070 		mac_link_update(igb->mac_hdl, igb->link_state);
4071 
4072 	igb_start_watchdog_timer(igb);
4073 }
4074 
4075 /*
4076  * igb_intr_legacy - Interrupt handler for legacy interrupts
4077  */
4078 static uint_t
4079 igb_intr_legacy(void *arg1, void *arg2)
4080 {
4081 	igb_t *igb = (igb_t *)arg1;
4082 	igb_tx_ring_t *tx_ring;
4083 	uint32_t icr;
4084 	mblk_t *mp;
4085 	boolean_t tx_reschedule;
4086 	boolean_t link_changed;
4087 	uint_t result;
4088 
4089 	_NOTE(ARGUNUSED(arg2));
4090 
4091 	mutex_enter(&igb->gen_lock);
4092 
4093 	if (igb->igb_state & IGB_SUSPENDED) {
4094 		mutex_exit(&igb->gen_lock);
4095 		return (DDI_INTR_UNCLAIMED);
4096 	}
4097 
4098 	mp = NULL;
4099 	tx_reschedule = B_FALSE;
4100 	link_changed = B_FALSE;
4101 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
4102 
4103 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
4104 		mutex_exit(&igb->gen_lock);
4105 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
4106 		atomic_or_32(&igb->igb_state, IGB_ERROR);
4107 		return (DDI_INTR_UNCLAIMED);
4108 	}
4109 
4110 	if (icr & E1000_ICR_INT_ASSERTED) {
4111 		/*
4112 		 * E1000_ICR_INT_ASSERTED bit was set:
4113 		 * Read(Clear) the ICR, claim this interrupt,
4114 		 * look for work to do.
4115 		 */
4116 		ASSERT(igb->num_rx_rings == 1);
4117 		ASSERT(igb->num_tx_rings == 1);
4118 
4119 		/* Make sure all interrupt causes cleared */
4120 		(void) E1000_READ_REG(&igb->hw, E1000_EICR);
4121 
4122 		if (icr & E1000_ICR_RXT0) {
4123 			mp = igb_rx(&igb->rx_rings[0], IGB_NO_POLL);
4124 		}
4125 
4126 		if (icr & E1000_ICR_TXDW) {
4127 			tx_ring = &igb->tx_rings[0];
4128 
4129 			/* Recycle the tx descriptors */
4130 			tx_ring->tx_recycle(tx_ring);
4131 
4132 			/* Schedule the re-transmit */
4133 			tx_reschedule = (tx_ring->reschedule &&
4134 			    (tx_ring->tbd_free >= igb->tx_resched_thresh));
4135 		}
4136 
4137 		if (icr & E1000_ICR_LSC) {
4138 			/*
4139 			 * Because we got a link-status-change interrupt, force
4140 			 * e1000_check_for_link() to look at phy
4141 			 */
4142 			igb->hw.mac.get_link_status = B_TRUE;
4143 
4144 			/* igb_link_check takes care of link status change */
4145 			link_changed = igb_link_check(igb);
4146 
4147 			/* Get new phy state */
4148 			igb_get_phy_state(igb);
4149 		}
4150 
4151 		if (icr & E1000_ICR_DRSTA) {
4152 			/* 82580 Full Device Reset needed */
4153 			atomic_or_32(&igb->igb_state, IGB_STALL);
4154 		}
4155 
4156 		result = DDI_INTR_CLAIMED;
4157 	} else {
4158 		/*
4159 		 * E1000_ICR_INT_ASSERTED bit was not set:
4160 		 * Don't claim this interrupt.
4161 		 */
4162 		result = DDI_INTR_UNCLAIMED;
4163 	}
4164 
4165 	mutex_exit(&igb->gen_lock);
4166 
4167 	/*
4168 	 * Do the following work outside of the gen_lock
4169 	 */
4170 	if (mp != NULL)
4171 		mac_rx(igb->mac_hdl, NULL, mp);
4172 
4173 	if (tx_reschedule)  {
4174 		tx_ring->reschedule = B_FALSE;
4175 		mac_tx_ring_update(igb->mac_hdl, tx_ring->ring_handle);
4176 		IGB_DEBUG_STAT(tx_ring->stat_reschedule);
4177 	}
4178 
4179 	if (link_changed)
4180 		mac_link_update(igb->mac_hdl, igb->link_state);
4181 
4182 	return (result);
4183 }
4184 
4185 /*
4186  * igb_intr_msi - Interrupt handler for MSI
4187  */
4188 static uint_t
4189 igb_intr_msi(void *arg1, void *arg2)
4190 {
4191 	igb_t *igb = (igb_t *)arg1;
4192 	uint32_t icr;
4193 
4194 	_NOTE(ARGUNUSED(arg2));
4195 
4196 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
4197 
4198 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
4199 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
4200 		atomic_or_32(&igb->igb_state, IGB_ERROR);
4201 		return (DDI_INTR_CLAIMED);
4202 	}
4203 
4204 	/* Make sure all interrupt causes cleared */
4205 	(void) E1000_READ_REG(&igb->hw, E1000_EICR);
4206 
4207 	/*
4208 	 * For MSI interrupt, we have only one vector,
4209 	 * so we have only one rx ring and one tx ring enabled.
4210 	 */
4211 	ASSERT(igb->num_rx_rings == 1);
4212 	ASSERT(igb->num_tx_rings == 1);
4213 
4214 	if (icr & E1000_ICR_RXT0) {
4215 		igb_intr_rx_work(&igb->rx_rings[0]);
4216 	}
4217 
4218 	if (icr & E1000_ICR_TXDW) {
4219 		igb_intr_tx_work(&igb->tx_rings[0]);
4220 	}
4221 
4222 	if (icr & E1000_ICR_LSC) {
4223 		igb_intr_link_work(igb);
4224 	}
4225 
4226 	if (icr & E1000_ICR_DRSTA) {
4227 		/* 82580 Full Device Reset needed */
4228 		atomic_or_32(&igb->igb_state, IGB_STALL);
4229 	}
4230 
4231 	return (DDI_INTR_CLAIMED);
4232 }
4233 
4234 /*
4235  * igb_intr_rx - Interrupt handler for rx
4236  */
4237 static uint_t
4238 igb_intr_rx(void *arg1, void *arg2)
4239 {
4240 	igb_rx_ring_t *rx_ring = (igb_rx_ring_t *)arg1;
4241 
4242 	_NOTE(ARGUNUSED(arg2));
4243 
4244 	/*
4245 	 * Only used via MSI-X vector so don't check cause bits
4246 	 * and only clean the given ring.
4247 	 */
4248 	igb_intr_rx_work(rx_ring);
4249 
4250 	return (DDI_INTR_CLAIMED);
4251 }
4252 
4253 /*
4254  * igb_intr_tx - Interrupt handler for tx
4255  */
4256 static uint_t
4257 igb_intr_tx(void *arg1, void *arg2)
4258 {
4259 	igb_tx_ring_t *tx_ring = (igb_tx_ring_t *)arg1;
4260 
4261 	_NOTE(ARGUNUSED(arg2));
4262 
4263 	/*
4264 	 * Only used via MSI-X vector so don't check cause bits
4265 	 * and only clean the given ring.
4266 	 */
4267 	igb_intr_tx_work(tx_ring);
4268 
4269 	return (DDI_INTR_CLAIMED);
4270 }
4271 
4272 /*
4273  * igb_intr_tx_other - Interrupt handler for both tx and other
4274  *
4275  */
4276 static uint_t
4277 igb_intr_tx_other(void *arg1, void *arg2)
4278 {
4279 	igb_t *igb = (igb_t *)arg1;
4280 	uint32_t icr;
4281 
4282 	_NOTE(ARGUNUSED(arg2));
4283 
4284 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
4285 
4286 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
4287 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
4288 		atomic_or_32(&igb->igb_state, IGB_ERROR);
4289 		return (DDI_INTR_CLAIMED);
4290 	}
4291 
4292 	/*
4293 	 * Look for tx reclaiming work first. Remember, in the
4294 	 * case of only interrupt sharing, only one tx ring is
4295 	 * used
4296 	 */
4297 	igb_intr_tx_work(&igb->tx_rings[0]);
4298 
4299 	/*
4300 	 * Check for "other" causes.
4301 	 */
4302 	if (icr & E1000_ICR_LSC) {
4303 		igb_intr_link_work(igb);
4304 	}
4305 
4306 	/*
4307 	 * The DOUTSYNC bit indicates a tx packet dropped because
4308 	 * DMA engine gets "out of sync". There isn't a real fix
4309 	 * for this. The Intel recommendation is to count the number
4310 	 * of occurrences so user can detect when it is happening.
4311 	 * The issue is non-fatal and there's no recovery action
4312 	 * available.
4313 	 */
4314 	if (icr & E1000_ICR_DOUTSYNC) {
4315 		IGB_STAT(igb->dout_sync);
4316 	}
4317 
4318 	if (icr & E1000_ICR_DRSTA) {
4319 		/* 82580 Full Device Reset needed */
4320 		atomic_or_32(&igb->igb_state, IGB_STALL);
4321 	}
4322 
4323 	return (DDI_INTR_CLAIMED);
4324 }
4325 
4326 /*
4327  * igb_alloc_intrs - Allocate interrupts for the driver
4328  *
4329  * Normal sequence is to try MSI-X; if not sucessful, try MSI;
4330  * if not successful, try Legacy.
4331  * igb->intr_force can be used to force sequence to start with
4332  * any of the 3 types.
4333  * If MSI-X is not used, number of tx/rx rings is forced to 1.
4334  */
4335 static int
4336 igb_alloc_intrs(igb_t *igb)
4337 {
4338 	dev_info_t *devinfo;
4339 	int intr_types;
4340 	int rc;
4341 
4342 	devinfo = igb->dip;
4343 
4344 	/* Get supported interrupt types */
4345 	rc = ddi_intr_get_supported_types(devinfo, &intr_types);
4346 
4347 	if (rc != DDI_SUCCESS) {
4348 		igb_log(igb,
4349 		    "Get supported interrupt types failed: %d", rc);
4350 		return (IGB_FAILURE);
4351 	}
4352 	IGB_DEBUGLOG_1(igb, "Supported interrupt types: %x", intr_types);
4353 
4354 	igb->intr_type = 0;
4355 
4356 	/* Install MSI-X interrupts */
4357 	if ((intr_types & DDI_INTR_TYPE_MSIX) &&
4358 	    (igb->intr_force <= IGB_INTR_MSIX)) {
4359 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_MSIX);
4360 
4361 		if (rc == IGB_SUCCESS)
4362 			return (IGB_SUCCESS);
4363 
4364 		igb_log(igb,
4365 		    "Allocate MSI-X failed, trying MSI interrupts...");
4366 	}
4367 
4368 	/* MSI-X not used, force rings to 1 */
4369 	igb->num_rx_rings = 1;
4370 	igb->num_tx_rings = 1;
4371 	igb_log(igb,
4372 	    "MSI-X not used, force rx and tx queue number to 1");
4373 
4374 	/* Install MSI interrupts */
4375 	if ((intr_types & DDI_INTR_TYPE_MSI) &&
4376 	    (igb->intr_force <= IGB_INTR_MSI)) {
4377 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_MSI);
4378 
4379 		if (rc == IGB_SUCCESS)
4380 			return (IGB_SUCCESS);
4381 
4382 		igb_log(igb,
4383 		    "Allocate MSI failed, trying Legacy interrupts...");
4384 	}
4385 
4386 	/* Install legacy interrupts */
4387 	if (intr_types & DDI_INTR_TYPE_FIXED) {
4388 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_FIXED);
4389 
4390 		if (rc == IGB_SUCCESS)
4391 			return (IGB_SUCCESS);
4392 
4393 		igb_log(igb,
4394 		    "Allocate Legacy interrupts failed");
4395 	}
4396 
4397 	/* If none of the 3 types succeeded, return failure */
4398 	return (IGB_FAILURE);
4399 }
4400 
4401 /*
4402  * igb_alloc_intr_handles - Allocate interrupt handles.
4403  *
4404  * For legacy and MSI, only 1 handle is needed.  For MSI-X,
4405  * if fewer than 2 handles are available, return failure.
4406  * Upon success, this sets the number of Rx rings to a number that
4407  * matches the handles available for Rx interrupts.
4408  */
4409 static int
4410 igb_alloc_intr_handles(igb_t *igb, int intr_type)
4411 {
4412 	dev_info_t *devinfo;
4413 	int orig, request, count, avail, actual;
4414 	int diff, minimum;
4415 	int rc;
4416 
4417 	devinfo = igb->dip;
4418 
4419 	switch (intr_type) {
4420 	case DDI_INTR_TYPE_FIXED:
4421 		request = 1;	/* Request 1 legacy interrupt handle */
4422 		minimum = 1;
4423 		IGB_DEBUGLOG_0(igb, "interrupt type: legacy");
4424 		break;
4425 
4426 	case DDI_INTR_TYPE_MSI:
4427 		request = 1;	/* Request 1 MSI interrupt handle */
4428 		minimum = 1;
4429 		IGB_DEBUGLOG_0(igb, "interrupt type: MSI");
4430 		break;
4431 
4432 	case DDI_INTR_TYPE_MSIX:
4433 		/*
4434 		 * Number of vectors for the adapter is
4435 		 * # rx rings + # tx rings
4436 		 * One of tx vectors is for tx & other
4437 		 */
4438 		request = igb->num_rx_rings + igb->num_tx_rings;
4439 		orig = request;
4440 		minimum = 2;
4441 		IGB_DEBUGLOG_0(igb, "interrupt type: MSI-X");
4442 		break;
4443 
4444 	default:
4445 		igb_log(igb,
4446 		    "invalid call to igb_alloc_intr_handles(): %d\n",
4447 		    intr_type);
4448 		return (IGB_FAILURE);
4449 	}
4450 	IGB_DEBUGLOG_2(igb, "interrupt handles requested: %d  minimum: %d",
4451 	    request, minimum);
4452 
4453 	/*
4454 	 * Get number of supported interrupts
4455 	 */
4456 	rc = ddi_intr_get_nintrs(devinfo, intr_type, &count);
4457 	if ((rc != DDI_SUCCESS) || (count < minimum)) {
4458 		igb_log(igb,
4459 		    "Get supported interrupt number failed. "
4460 		    "Return: %d, count: %d", rc, count);
4461 		return (IGB_FAILURE);
4462 	}
4463 	IGB_DEBUGLOG_1(igb, "interrupts supported: %d", count);
4464 
4465 	/*
4466 	 * Get number of available interrupts
4467 	 */
4468 	rc = ddi_intr_get_navail(devinfo, intr_type, &avail);
4469 	if ((rc != DDI_SUCCESS) || (avail < minimum)) {
4470 		igb_log(igb,
4471 		    "Get available interrupt number failed. "
4472 		    "Return: %d, available: %d", rc, avail);
4473 		return (IGB_FAILURE);
4474 	}
4475 	IGB_DEBUGLOG_1(igb, "interrupts available: %d", avail);
4476 
4477 	if (avail < request) {
4478 		igb_log(igb, "Request %d handles, %d available",
4479 		    request, avail);
4480 		request = avail;
4481 	}
4482 
4483 	actual = 0;
4484 	igb->intr_cnt = 0;
4485 
4486 	/*
4487 	 * Allocate an array of interrupt handles
4488 	 */
4489 	igb->intr_size = request * sizeof (ddi_intr_handle_t);
4490 	igb->htable = kmem_alloc(igb->intr_size, KM_SLEEP);
4491 
4492 	rc = ddi_intr_alloc(devinfo, igb->htable, intr_type, 0,
4493 	    request, &actual, DDI_INTR_ALLOC_NORMAL);
4494 	if (rc != DDI_SUCCESS) {
4495 		igb_log(igb, "Allocate interrupts failed. "
4496 		    "return: %d, request: %d, actual: %d",
4497 		    rc, request, actual);
4498 		goto alloc_handle_fail;
4499 	}
4500 	IGB_DEBUGLOG_1(igb, "interrupts actually allocated: %d", actual);
4501 
4502 	igb->intr_cnt = actual;
4503 
4504 	if (actual < minimum) {
4505 		igb_log(igb, "Insufficient interrupt handles allocated: %d",
4506 		    actual);
4507 		goto alloc_handle_fail;
4508 	}
4509 
4510 	/*
4511 	 * For MSI-X, actual might force us to reduce number of tx & rx rings
4512 	 */
4513 	if ((intr_type == DDI_INTR_TYPE_MSIX) && (orig > actual)) {
4514 		diff = orig - actual;
4515 		if (diff < igb->num_tx_rings) {
4516 			igb_log(igb,
4517 			    "MSI-X vectors force Tx queue number to %d",
4518 			    igb->num_tx_rings - diff);
4519 			igb->num_tx_rings -= diff;
4520 		} else {
4521 			igb_log(igb,
4522 			    "MSI-X vectors force Tx queue number to 1");
4523 			igb->num_tx_rings = 1;
4524 
4525 			igb_log(igb,
4526 			    "MSI-X vectors force Rx queue number to %d",
4527 			    actual - 1);
4528 			igb->num_rx_rings = actual - 1;
4529 		}
4530 	}
4531 
4532 	/*
4533 	 * Get priority for first vector, assume remaining are all the same
4534 	 */
4535 	rc = ddi_intr_get_pri(igb->htable[0], &igb->intr_pri);
4536 	if (rc != DDI_SUCCESS) {
4537 		igb_log(igb,
4538 		    "Get interrupt priority failed: %d", rc);
4539 		goto alloc_handle_fail;
4540 	}
4541 
4542 	rc = ddi_intr_get_cap(igb->htable[0], &igb->intr_cap);
4543 	if (rc != DDI_SUCCESS) {
4544 		igb_log(igb,
4545 		    "Get interrupt cap failed: %d", rc);
4546 		goto alloc_handle_fail;
4547 	}
4548 
4549 	igb->intr_type = intr_type;
4550 
4551 	return (IGB_SUCCESS);
4552 
4553 alloc_handle_fail:
4554 	igb_rem_intrs(igb);
4555 
4556 	return (IGB_FAILURE);
4557 }
4558 
4559 /*
4560  * igb_add_intr_handlers - Add interrupt handlers based on the interrupt type
4561  *
4562  * Before adding the interrupt handlers, the interrupt vectors have
4563  * been allocated, and the rx/tx rings have also been allocated.
4564  */
4565 static int
4566 igb_add_intr_handlers(igb_t *igb)
4567 {
4568 	igb_rx_ring_t *rx_ring;
4569 	igb_tx_ring_t *tx_ring;
4570 	int vector;
4571 	int rc;
4572 	int i;
4573 
4574 	vector = 0;
4575 
4576 	switch (igb->intr_type) {
4577 	case DDI_INTR_TYPE_MSIX:
4578 		/* Add interrupt handler for tx + other */
4579 		tx_ring = &igb->tx_rings[0];
4580 		rc = ddi_intr_add_handler(igb->htable[vector],
4581 		    (ddi_intr_handler_t *)igb_intr_tx_other,
4582 		    (void *)igb, NULL);
4583 
4584 		if (rc != DDI_SUCCESS) {
4585 			igb_log(igb,
4586 			    "Add tx/other interrupt handler failed: %d", rc);
4587 			return (IGB_FAILURE);
4588 		}
4589 		tx_ring->intr_vector = vector;
4590 		vector++;
4591 
4592 		/* Add interrupt handler for each rx ring */
4593 		for (i = 0; i < igb->num_rx_rings; i++) {
4594 			rx_ring = &igb->rx_rings[i];
4595 
4596 			rc = ddi_intr_add_handler(igb->htable[vector],
4597 			    (ddi_intr_handler_t *)igb_intr_rx,
4598 			    (void *)rx_ring, NULL);
4599 
4600 			if (rc != DDI_SUCCESS) {
4601 				igb_log(igb,
4602 				    "Add rx interrupt handler failed. "
4603 				    "return: %d, rx ring: %d", rc, i);
4604 				for (vector--; vector >= 0; vector--) {
4605 					(void) ddi_intr_remove_handler(
4606 					    igb->htable[vector]);
4607 				}
4608 				return (IGB_FAILURE);
4609 			}
4610 
4611 			rx_ring->intr_vector = vector;
4612 
4613 			vector++;
4614 		}
4615 
4616 		/* Add interrupt handler for each tx ring from 2nd ring */
4617 		for (i = 1; i < igb->num_tx_rings; i++) {
4618 			tx_ring = &igb->tx_rings[i];
4619 
4620 			rc = ddi_intr_add_handler(igb->htable[vector],
4621 			    (ddi_intr_handler_t *)igb_intr_tx,
4622 			    (void *)tx_ring, NULL);
4623 
4624 			if (rc != DDI_SUCCESS) {
4625 				igb_log(igb,
4626 				    "Add tx interrupt handler failed. "
4627 				    "return: %d, tx ring: %d", rc, i);
4628 				for (vector--; vector >= 0; vector--) {
4629 					(void) ddi_intr_remove_handler(
4630 					    igb->htable[vector]);
4631 				}
4632 				return (IGB_FAILURE);
4633 			}
4634 
4635 			tx_ring->intr_vector = vector;
4636 
4637 			vector++;
4638 		}
4639 
4640 		break;
4641 
4642 	case DDI_INTR_TYPE_MSI:
4643 		/* Add interrupt handlers for the only vector */
4644 		rc = ddi_intr_add_handler(igb->htable[vector],
4645 		    (ddi_intr_handler_t *)igb_intr_msi,
4646 		    (void *)igb, NULL);
4647 
4648 		if (rc != DDI_SUCCESS) {
4649 			igb_log(igb,
4650 			    "Add MSI interrupt handler failed: %d", rc);
4651 			return (IGB_FAILURE);
4652 		}
4653 
4654 		rx_ring = &igb->rx_rings[0];
4655 		rx_ring->intr_vector = vector;
4656 
4657 		vector++;
4658 		break;
4659 
4660 	case DDI_INTR_TYPE_FIXED:
4661 		/* Add interrupt handlers for the only vector */
4662 		rc = ddi_intr_add_handler(igb->htable[vector],
4663 		    (ddi_intr_handler_t *)igb_intr_legacy,
4664 		    (void *)igb, NULL);
4665 
4666 		if (rc != DDI_SUCCESS) {
4667 			igb_log(igb,
4668 			    "Add legacy interrupt handler failed: %d", rc);
4669 			return (IGB_FAILURE);
4670 		}
4671 
4672 		rx_ring = &igb->rx_rings[0];
4673 		rx_ring->intr_vector = vector;
4674 
4675 		vector++;
4676 		break;
4677 
4678 	default:
4679 		return (IGB_FAILURE);
4680 	}
4681 
4682 	ASSERT(vector == igb->intr_cnt);
4683 
4684 	return (IGB_SUCCESS);
4685 }
4686 
4687 /*
4688  * igb_setup_msix_82575 - setup 82575 adapter to use MSI-X interrupts
4689  *
4690  * For each vector enabled on the adapter, Set the MSIXBM register accordingly
4691  */
4692 static void
4693 igb_setup_msix_82575(igb_t *igb)
4694 {
4695 	uint32_t eims = 0;
4696 	int i, vector;
4697 	struct e1000_hw *hw = &igb->hw;
4698 
4699 	/*
4700 	 * Set vector for tx ring 0 and other causes.
4701 	 * NOTE assumption that it is vector 0.
4702 	 */
4703 	vector = 0;
4704 
4705 	igb->eims_mask = E1000_EICR_TX_QUEUE0 | E1000_EICR_OTHER;
4706 	E1000_WRITE_REG(hw, E1000_MSIXBM(vector), igb->eims_mask);
4707 	vector++;
4708 
4709 	for (i = 0; i < igb->num_rx_rings; i++) {
4710 		/*
4711 		 * Set vector for each rx ring
4712 		 */
4713 		eims = (E1000_EICR_RX_QUEUE0 << i);
4714 		E1000_WRITE_REG(hw, E1000_MSIXBM(vector), eims);
4715 
4716 		/*
4717 		 * Accumulate bits to enable in
4718 		 * igb_enable_adapter_interrupts_82575()
4719 		 */
4720 		igb->eims_mask |= eims;
4721 
4722 		vector++;
4723 	}
4724 
4725 	for (i = 1; i < igb->num_tx_rings; i++) {
4726 		/*
4727 		 * Set vector for each tx ring from 2nd tx ring
4728 		 */
4729 		eims = (E1000_EICR_TX_QUEUE0 << i);
4730 		E1000_WRITE_REG(hw, E1000_MSIXBM(vector), eims);
4731 
4732 		/*
4733 		 * Accumulate bits to enable in
4734 		 * igb_enable_adapter_interrupts_82575()
4735 		 */
4736 		igb->eims_mask |= eims;
4737 
4738 		vector++;
4739 	}
4740 
4741 	ASSERT(vector == igb->intr_cnt);
4742 
4743 	/*
4744 	 * Disable IAM for ICR interrupt bits
4745 	 */
4746 	E1000_WRITE_REG(hw, E1000_IAM, 0);
4747 	E1000_WRITE_FLUSH(hw);
4748 }
4749 
4750 /*
4751  * igb_setup_msix_82576 - setup 82576 adapter to use MSI-X interrupts
4752  *
4753  * 82576 uses a table based method for assigning vectors.  Each queue has a
4754  * single entry in the table to which we write a vector number along with a
4755  * "valid" bit.  The entry is a single byte in a 4-byte register.  Vectors
4756  * take a different position in the 4-byte register depending on whether
4757  * they are numbered above or below 8.
4758  */
4759 static void
4760 igb_setup_msix_82576(igb_t *igb)
4761 {
4762 	struct e1000_hw *hw = &igb->hw;
4763 	uint32_t ivar, index, vector;
4764 	int i;
4765 
4766 	/* must enable msi-x capability before IVAR settings */
4767 	E1000_WRITE_REG(hw, E1000_GPIE,
4768 	    (E1000_GPIE_MSIX_MODE | E1000_GPIE_PBA | E1000_GPIE_NSICR));
4769 
4770 	/*
4771 	 * Set vector for tx ring 0 and other causes.
4772 	 * NOTE assumption that it is vector 0.
4773 	 * This is also interdependent with installation of interrupt service
4774 	 * routines in igb_add_intr_handlers().
4775 	 */
4776 
4777 	/* assign "other" causes to vector 0 */
4778 	vector = 0;
4779 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4780 	E1000_WRITE_REG(hw, E1000_IVAR_MISC, ivar);
4781 
4782 	/* assign tx ring 0 to vector 0 */
4783 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4784 	E1000_WRITE_REG(hw, E1000_IVAR0, ivar);
4785 
4786 	/* prepare to enable tx & other interrupt causes */
4787 	igb->eims_mask = (1 << vector);
4788 
4789 	vector ++;
4790 	for (i = 0; i < igb->num_rx_rings; i++) {
4791 		/*
4792 		 * Set vector for each rx ring
4793 		 */
4794 		index = (i & 0x7);
4795 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4796 
4797 		if (i < 8) {
4798 			/* vector goes into low byte of register */
4799 			ivar = ivar & 0xFFFFFF00;
4800 			ivar |= (vector | E1000_IVAR_VALID);
4801 		} else {
4802 			/* vector goes into third byte of register */
4803 			ivar = ivar & 0xFF00FFFF;
4804 			ivar |= ((vector | E1000_IVAR_VALID) << 16);
4805 		}
4806 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4807 
4808 		/* Accumulate interrupt-cause bits to enable */
4809 		igb->eims_mask |= (1 << vector);
4810 
4811 		vector ++;
4812 	}
4813 
4814 	for (i = 1; i < igb->num_tx_rings; i++) {
4815 		/*
4816 		 * Set vector for each tx ring from 2nd tx ring.
4817 		 * Note assumption that tx vectors numericall follow rx vectors.
4818 		 */
4819 		index = (i & 0x7);
4820 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4821 
4822 		if (i < 8) {
4823 			/* vector goes into second byte of register */
4824 			ivar = ivar & 0xFFFF00FF;
4825 			ivar |= ((vector | E1000_IVAR_VALID) << 8);
4826 		} else {
4827 			/* vector goes into fourth byte of register */
4828 			ivar = ivar & 0x00FFFFFF;
4829 			ivar |= (vector | E1000_IVAR_VALID) << 24;
4830 		}
4831 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4832 
4833 		/* Accumulate interrupt-cause bits to enable */
4834 		igb->eims_mask |= (1 << vector);
4835 
4836 		vector ++;
4837 	}
4838 
4839 	ASSERT(vector == igb->intr_cnt);
4840 }
4841 
4842 /*
4843  * igb_setup_msix_82580 - setup 82580 adapter to use MSI-X interrupts
4844  *
4845  * 82580 uses same table approach at 82576 but has fewer entries.  Each
4846  * queue has a single entry in the table to which we write a vector number
4847  * along with a "valid" bit.  Vectors take a different position in the
4848  * register depending on * whether * they are numbered above or below 4.
4849  */
4850 static void
4851 igb_setup_msix_82580(igb_t *igb)
4852 {
4853 	struct e1000_hw *hw = &igb->hw;
4854 	uint32_t ivar, index, vector;
4855 	int i;
4856 
4857 	/* must enable msi-x capability before IVAR settings */
4858 	E1000_WRITE_REG(hw, E1000_GPIE, (E1000_GPIE_MSIX_MODE |
4859 	    E1000_GPIE_PBA | E1000_GPIE_NSICR | E1000_GPIE_EIAME));
4860 	/*
4861 	 * Set vector for tx ring 0 and other causes.
4862 	 * NOTE assumption that it is vector 0.
4863 	 * This is also interdependent with installation of interrupt service
4864 	 * routines in igb_add_intr_handlers().
4865 	 */
4866 
4867 	/* assign "other" causes to vector 0 */
4868 	vector = 0;
4869 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4870 	E1000_WRITE_REG(hw, E1000_IVAR_MISC, ivar);
4871 
4872 	/* assign tx ring 0 to vector 0 */
4873 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4874 	E1000_WRITE_REG(hw, E1000_IVAR0, ivar);
4875 
4876 	/* prepare to enable tx & other interrupt causes */
4877 	igb->eims_mask = (1 << vector);
4878 
4879 	vector ++;
4880 
4881 	for (i = 0; i < igb->num_rx_rings; i++) {
4882 		/*
4883 		 * Set vector for each rx ring
4884 		 */
4885 		index = (i >> 1);
4886 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4887 
4888 		if (i & 1) {
4889 			/* vector goes into third byte of register */
4890 			ivar = ivar & 0xFF00FFFF;
4891 			ivar |= ((vector | E1000_IVAR_VALID) << 16);
4892 		} else {
4893 			/* vector goes into low byte of register */
4894 			ivar = ivar & 0xFFFFFF00;
4895 			ivar |= (vector | E1000_IVAR_VALID);
4896 		}
4897 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4898 
4899 		/* Accumulate interrupt-cause bits to enable */
4900 		igb->eims_mask |= (1 << vector);
4901 
4902 		vector ++;
4903 	}
4904 
4905 	for (i = 1; i < igb->num_tx_rings; i++) {
4906 		/*
4907 		 * Set vector for each tx ring from 2nd tx ring.
4908 		 * Note assumption that tx vectors numericall follow rx vectors.
4909 		 */
4910 		index = (i >> 1);
4911 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4912 
4913 		if (i & 1) {
4914 			/* vector goes into high byte of register */
4915 			ivar = ivar & 0x00FFFFFF;
4916 			ivar |= ((vector | E1000_IVAR_VALID) << 24);
4917 		} else {
4918 			/* vector goes into second byte of register */
4919 			ivar = ivar & 0xFFFF00FF;
4920 			ivar |= (vector | E1000_IVAR_VALID) << 8;
4921 		}
4922 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4923 
4924 		/* Accumulate interrupt-cause bits to enable */
4925 		igb->eims_mask |= (1 << vector);
4926 
4927 		vector ++;
4928 	}
4929 	ASSERT(vector == igb->intr_cnt);
4930 }
4931 
4932 /*
4933  * igb_rem_intr_handlers - remove the interrupt handlers
4934  */
4935 static void
4936 igb_rem_intr_handlers(igb_t *igb)
4937 {
4938 	int i;
4939 	int rc;
4940 
4941 	for (i = 0; i < igb->intr_cnt; i++) {
4942 		rc = ddi_intr_remove_handler(igb->htable[i]);
4943 		if (rc != DDI_SUCCESS) {
4944 			IGB_DEBUGLOG_1(igb,
4945 			    "Remove intr handler failed: %d", rc);
4946 		}
4947 	}
4948 }
4949 
4950 /*
4951  * igb_rem_intrs - remove the allocated interrupts
4952  */
4953 static void
4954 igb_rem_intrs(igb_t *igb)
4955 {
4956 	int i;
4957 	int rc;
4958 
4959 	for (i = 0; i < igb->intr_cnt; i++) {
4960 		rc = ddi_intr_free(igb->htable[i]);
4961 		if (rc != DDI_SUCCESS) {
4962 			IGB_DEBUGLOG_1(igb,
4963 			    "Free intr failed: %d", rc);
4964 		}
4965 	}
4966 
4967 	kmem_free(igb->htable, igb->intr_size);
4968 	igb->htable = NULL;
4969 }
4970 
4971 /*
4972  * igb_enable_intrs - enable all the ddi interrupts
4973  */
4974 static int
4975 igb_enable_intrs(igb_t *igb)
4976 {
4977 	int i;
4978 	int rc;
4979 
4980 	/* Enable interrupts */
4981 	if (igb->intr_cap & DDI_INTR_FLAG_BLOCK) {
4982 		/* Call ddi_intr_block_enable() for MSI */
4983 		rc = ddi_intr_block_enable(igb->htable, igb->intr_cnt);
4984 		if (rc != DDI_SUCCESS) {
4985 			igb_log(igb,
4986 			    "Enable block intr failed: %d", rc);
4987 			return (IGB_FAILURE);
4988 		}
4989 	} else {
4990 		/* Call ddi_intr_enable() for Legacy/MSI non block enable */
4991 		for (i = 0; i < igb->intr_cnt; i++) {
4992 			rc = ddi_intr_enable(igb->htable[i]);
4993 			if (rc != DDI_SUCCESS) {
4994 				igb_log(igb,
4995 				    "Enable intr failed: %d", rc);
4996 				return (IGB_FAILURE);
4997 			}
4998 		}
4999 	}
5000 
5001 	return (IGB_SUCCESS);
5002 }
5003 
5004 /*
5005  * igb_disable_intrs - disable all the ddi interrupts
5006  */
5007 static int
5008 igb_disable_intrs(igb_t *igb)
5009 {
5010 	int i;
5011 	int rc;
5012 
5013 	/* Disable all interrupts */
5014 	if (igb->intr_cap & DDI_INTR_FLAG_BLOCK) {
5015 		rc = ddi_intr_block_disable(igb->htable, igb->intr_cnt);
5016 		if (rc != DDI_SUCCESS) {
5017 			igb_log(igb,
5018 			    "Disable block intr failed: %d", rc);
5019 			return (IGB_FAILURE);
5020 		}
5021 	} else {
5022 		for (i = 0; i < igb->intr_cnt; i++) {
5023 			rc = ddi_intr_disable(igb->htable[i]);
5024 			if (rc != DDI_SUCCESS) {
5025 				igb_log(igb,
5026 				    "Disable intr failed: %d", rc);
5027 				return (IGB_FAILURE);
5028 			}
5029 		}
5030 	}
5031 
5032 	return (IGB_SUCCESS);
5033 }
5034 
5035 /*
5036  * igb_get_phy_state - Get and save the parameters read from PHY registers
5037  */
5038 static void
5039 igb_get_phy_state(igb_t *igb)
5040 {
5041 	struct e1000_hw *hw = &igb->hw;
5042 	uint16_t phy_ctrl;
5043 	uint16_t phy_status;
5044 	uint16_t phy_an_adv;
5045 	uint16_t phy_an_exp;
5046 	uint16_t phy_ext_status;
5047 	uint16_t phy_1000t_ctrl;
5048 	uint16_t phy_1000t_status;
5049 	uint16_t phy_lp_able;
5050 
5051 	ASSERT(mutex_owned(&igb->gen_lock));
5052 
5053 	if (hw->phy.media_type == e1000_media_type_copper) {
5054 		(void) e1000_read_phy_reg(hw, PHY_CONTROL, &phy_ctrl);
5055 		(void) e1000_read_phy_reg(hw, PHY_STATUS, &phy_status);
5056 		(void) e1000_read_phy_reg(hw, PHY_AUTONEG_ADV, &phy_an_adv);
5057 		(void) e1000_read_phy_reg(hw, PHY_AUTONEG_EXP, &phy_an_exp);
5058 		(void) e1000_read_phy_reg(hw, PHY_EXT_STATUS, &phy_ext_status);
5059 		(void) e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_1000t_ctrl);
5060 		(void) e1000_read_phy_reg(hw,
5061 		    PHY_1000T_STATUS, &phy_1000t_status);
5062 		(void) e1000_read_phy_reg(hw, PHY_LP_ABILITY, &phy_lp_able);
5063 
5064 		igb->param_autoneg_cap =
5065 		    (phy_status & MII_SR_AUTONEG_CAPS) ? 1 : 0;
5066 		igb->param_pause_cap =
5067 		    (phy_an_adv & NWAY_AR_PAUSE) ? 1 : 0;
5068 		igb->param_asym_pause_cap =
5069 		    (phy_an_adv & NWAY_AR_ASM_DIR) ? 1 : 0;
5070 		igb->param_1000fdx_cap =
5071 		    ((phy_ext_status & IEEE_ESR_1000T_FD_CAPS) ||
5072 		    (phy_ext_status & IEEE_ESR_1000X_FD_CAPS)) ? 1 : 0;
5073 		igb->param_1000hdx_cap =
5074 		    ((phy_ext_status & IEEE_ESR_1000T_HD_CAPS) ||
5075 		    (phy_ext_status & IEEE_ESR_1000X_HD_CAPS)) ? 1 : 0;
5076 		igb->param_100t4_cap =
5077 		    (phy_status & MII_SR_100T4_CAPS) ? 1 : 0;
5078 		igb->param_100fdx_cap = ((phy_status & MII_SR_100X_FD_CAPS) ||
5079 		    (phy_status & MII_SR_100T2_FD_CAPS)) ? 1 : 0;
5080 		igb->param_100hdx_cap = ((phy_status & MII_SR_100X_HD_CAPS) ||
5081 		    (phy_status & MII_SR_100T2_HD_CAPS)) ? 1 : 0;
5082 		igb->param_10fdx_cap =
5083 		    (phy_status & MII_SR_10T_FD_CAPS) ? 1 : 0;
5084 		igb->param_10hdx_cap =
5085 		    (phy_status & MII_SR_10T_HD_CAPS) ? 1 : 0;
5086 		igb->param_rem_fault =
5087 		    (phy_status & MII_SR_REMOTE_FAULT) ? 1 : 0;
5088 
5089 		igb->param_adv_autoneg_cap = hw->mac.autoneg;
5090 		igb->param_adv_pause_cap =
5091 		    (phy_an_adv & NWAY_AR_PAUSE) ? 1 : 0;
5092 		igb->param_adv_asym_pause_cap =
5093 		    (phy_an_adv & NWAY_AR_ASM_DIR) ? 1 : 0;
5094 		igb->param_adv_1000hdx_cap =
5095 		    (phy_1000t_ctrl & CR_1000T_HD_CAPS) ? 1 : 0;
5096 		igb->param_adv_100t4_cap =
5097 		    (phy_an_adv & NWAY_AR_100T4_CAPS) ? 1 : 0;
5098 		igb->param_adv_rem_fault =
5099 		    (phy_an_adv & NWAY_AR_REMOTE_FAULT) ? 1 : 0;
5100 		if (igb->param_adv_autoneg_cap == 1) {
5101 			igb->param_adv_1000fdx_cap =
5102 			    (phy_1000t_ctrl & CR_1000T_FD_CAPS) ? 1 : 0;
5103 			igb->param_adv_100fdx_cap =
5104 			    (phy_an_adv & NWAY_AR_100TX_FD_CAPS) ? 1 : 0;
5105 			igb->param_adv_100hdx_cap =
5106 			    (phy_an_adv & NWAY_AR_100TX_HD_CAPS) ? 1 : 0;
5107 			igb->param_adv_10fdx_cap =
5108 			    (phy_an_adv & NWAY_AR_10T_FD_CAPS) ? 1 : 0;
5109 			igb->param_adv_10hdx_cap =
5110 			    (phy_an_adv & NWAY_AR_10T_HD_CAPS) ? 1 : 0;
5111 		}
5112 
5113 		igb->param_lp_autoneg_cap =
5114 		    (phy_an_exp & NWAY_ER_LP_NWAY_CAPS) ? 1 : 0;
5115 		igb->param_lp_pause_cap =
5116 		    (phy_lp_able & NWAY_LPAR_PAUSE) ? 1 : 0;
5117 		igb->param_lp_asym_pause_cap =
5118 		    (phy_lp_able & NWAY_LPAR_ASM_DIR) ? 1 : 0;
5119 		igb->param_lp_1000fdx_cap =
5120 		    (phy_1000t_status & SR_1000T_LP_FD_CAPS) ? 1 : 0;
5121 		igb->param_lp_1000hdx_cap =
5122 		    (phy_1000t_status & SR_1000T_LP_HD_CAPS) ? 1 : 0;
5123 		igb->param_lp_100t4_cap =
5124 		    (phy_lp_able & NWAY_LPAR_100T4_CAPS) ? 1 : 0;
5125 		igb->param_lp_100fdx_cap =
5126 		    (phy_lp_able & NWAY_LPAR_100TX_FD_CAPS) ? 1 : 0;
5127 		igb->param_lp_100hdx_cap =
5128 		    (phy_lp_able & NWAY_LPAR_100TX_HD_CAPS) ? 1 : 0;
5129 		igb->param_lp_10fdx_cap =
5130 		    (phy_lp_able & NWAY_LPAR_10T_FD_CAPS) ? 1 : 0;
5131 		igb->param_lp_10hdx_cap =
5132 		    (phy_lp_able & NWAY_LPAR_10T_HD_CAPS) ? 1 : 0;
5133 		igb->param_lp_rem_fault =
5134 		    (phy_lp_able & NWAY_LPAR_REMOTE_FAULT) ? 1 : 0;
5135 	} else {
5136 		/*
5137 		 * 1Gig Fiber adapter only offers 1Gig Full Duplex.
5138 		 */
5139 		igb->param_autoneg_cap = 0;
5140 		igb->param_pause_cap = 1;
5141 		igb->param_asym_pause_cap = 1;
5142 		igb->param_1000fdx_cap = 1;
5143 		igb->param_1000hdx_cap = 0;
5144 		igb->param_100t4_cap = 0;
5145 		igb->param_100fdx_cap = 0;
5146 		igb->param_100hdx_cap = 0;
5147 		igb->param_10fdx_cap = 0;
5148 		igb->param_10hdx_cap = 0;
5149 
5150 		igb->param_adv_autoneg_cap = 0;
5151 		igb->param_adv_pause_cap = 1;
5152 		igb->param_adv_asym_pause_cap = 1;
5153 		igb->param_adv_1000fdx_cap = 1;
5154 		igb->param_adv_1000hdx_cap = 0;
5155 		igb->param_adv_100t4_cap = 0;
5156 		igb->param_adv_100fdx_cap = 0;
5157 		igb->param_adv_100hdx_cap = 0;
5158 		igb->param_adv_10fdx_cap = 0;
5159 		igb->param_adv_10hdx_cap = 0;
5160 
5161 		igb->param_lp_autoneg_cap = 0;
5162 		igb->param_lp_pause_cap = 0;
5163 		igb->param_lp_asym_pause_cap = 0;
5164 		igb->param_lp_1000fdx_cap = 0;
5165 		igb->param_lp_1000hdx_cap = 0;
5166 		igb->param_lp_100t4_cap = 0;
5167 		igb->param_lp_100fdx_cap = 0;
5168 		igb->param_lp_100hdx_cap = 0;
5169 		igb->param_lp_10fdx_cap = 0;
5170 		igb->param_lp_10hdx_cap = 0;
5171 		igb->param_lp_rem_fault = 0;
5172 	}
5173 }
5174 
5175 /*
5176  * synchronize the adv* and en* parameters.
5177  *
5178  * See comments in <sys/dld.h> for details of the *_en_*
5179  * parameters. The usage of ndd for setting adv parameters will
5180  * synchronize all the en parameters with the e1000g parameters,
5181  * implicitly disabling any settings made via dladm.
5182  */
5183 static void
5184 igb_param_sync(igb_t *igb)
5185 {
5186 	igb->param_en_1000fdx_cap = igb->param_adv_1000fdx_cap;
5187 	igb->param_en_1000hdx_cap = igb->param_adv_1000hdx_cap;
5188 	igb->param_en_100t4_cap = igb->param_adv_100t4_cap;
5189 	igb->param_en_100fdx_cap = igb->param_adv_100fdx_cap;
5190 	igb->param_en_100hdx_cap = igb->param_adv_100hdx_cap;
5191 	igb->param_en_10fdx_cap = igb->param_adv_10fdx_cap;
5192 	igb->param_en_10hdx_cap = igb->param_adv_10hdx_cap;
5193 }
5194 
5195 /*
5196  * igb_get_driver_control
5197  */
5198 static void
5199 igb_get_driver_control(struct e1000_hw *hw)
5200 {
5201 	uint32_t ctrl_ext;
5202 
5203 	/* Notify firmware that driver is in control of device */
5204 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
5205 	ctrl_ext |= E1000_CTRL_EXT_DRV_LOAD;
5206 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
5207 }
5208 
5209 /*
5210  * igb_release_driver_control
5211  */
5212 static void
5213 igb_release_driver_control(struct e1000_hw *hw)
5214 {
5215 	uint32_t ctrl_ext;
5216 
5217 	/* Notify firmware that driver is no longer in control of device */
5218 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
5219 	ctrl_ext &= ~E1000_CTRL_EXT_DRV_LOAD;
5220 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
5221 }
5222 
5223 /*
5224  * igb_atomic_reserve - Atomic decrease operation
5225  */
5226 int
5227 igb_atomic_reserve(uint32_t *count_p, uint32_t n)
5228 {
5229 	uint32_t oldval;
5230 	uint32_t newval;
5231 
5232 	/* ATOMICALLY */
5233 	do {
5234 		oldval = *count_p;
5235 		if (oldval < n)
5236 			return (-1);
5237 		newval = oldval - n;
5238 	} while (atomic_cas_32(count_p, oldval, newval) != oldval);
5239 
5240 	return (newval);
5241 }
5242 
5243 /*
5244  * FMA support
5245  */
5246 
5247 int
5248 igb_check_acc_handle(ddi_acc_handle_t handle)
5249 {
5250 	ddi_fm_error_t de;
5251 
5252 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
5253 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
5254 	return (de.fme_status);
5255 }
5256 
5257 int
5258 igb_check_dma_handle(ddi_dma_handle_t handle)
5259 {
5260 	ddi_fm_error_t de;
5261 
5262 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
5263 	return (de.fme_status);
5264 }
5265 
5266 /*
5267  * The IO fault service error handling callback function
5268  */
5269 /*ARGSUSED*/
5270 static int
5271 igb_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
5272 {
5273 	/*
5274 	 * as the driver can always deal with an error in any dma or
5275 	 * access handle, we can just return the fme_status value.
5276 	 */
5277 	pci_ereport_post(dip, err, NULL);
5278 	return (err->fme_status);
5279 }
5280 
5281 static void
5282 igb_fm_init(igb_t *igb)
5283 {
5284 	ddi_iblock_cookie_t iblk;
5285 	int fma_dma_flag;
5286 
5287 	/* Only register with IO Fault Services if we have some capability */
5288 	if (igb->fm_capabilities & DDI_FM_ACCCHK_CAPABLE) {
5289 		igb_regs_acc_attr.devacc_attr_access = DDI_FLAGERR_ACC;
5290 	} else {
5291 		igb_regs_acc_attr.devacc_attr_access = DDI_DEFAULT_ACC;
5292 	}
5293 
5294 	if (igb->fm_capabilities & DDI_FM_DMACHK_CAPABLE) {
5295 		fma_dma_flag = 1;
5296 	} else {
5297 		fma_dma_flag = 0;
5298 	}
5299 
5300 	(void) igb_set_fma_flags(fma_dma_flag);
5301 
5302 	if (igb->fm_capabilities) {
5303 
5304 		/* Register capabilities with IO Fault Services */
5305 		ddi_fm_init(igb->dip, &igb->fm_capabilities, &iblk);
5306 
5307 		/*
5308 		 * Initialize pci ereport capabilities if ereport capable
5309 		 */
5310 		if (DDI_FM_EREPORT_CAP(igb->fm_capabilities) ||
5311 		    DDI_FM_ERRCB_CAP(igb->fm_capabilities))
5312 			pci_ereport_setup(igb->dip);
5313 
5314 		/*
5315 		 * Register error callback if error callback capable
5316 		 */
5317 		if (DDI_FM_ERRCB_CAP(igb->fm_capabilities))
5318 			ddi_fm_handler_register(igb->dip,
5319 			    igb_fm_error_cb, (void*) igb);
5320 	}
5321 }
5322 
5323 static void
5324 igb_fm_fini(igb_t *igb)
5325 {
5326 	/* Only unregister FMA capabilities if we registered some */
5327 	if (igb->fm_capabilities) {
5328 
5329 		/*
5330 		 * Release any resources allocated by pci_ereport_setup()
5331 		 */
5332 		if (DDI_FM_EREPORT_CAP(igb->fm_capabilities) ||
5333 		    DDI_FM_ERRCB_CAP(igb->fm_capabilities))
5334 			pci_ereport_teardown(igb->dip);
5335 
5336 		/*
5337 		 * Un-register error callback if error callback capable
5338 		 */
5339 		if (DDI_FM_ERRCB_CAP(igb->fm_capabilities))
5340 			ddi_fm_handler_unregister(igb->dip);
5341 
5342 		/* Unregister from IO Fault Services */
5343 		ddi_fm_fini(igb->dip);
5344 	}
5345 }
5346 
5347 void
5348 igb_fm_ereport(igb_t *igb, char *detail)
5349 {
5350 	uint64_t ena;
5351 	char buf[FM_MAX_CLASS];
5352 
5353 	(void) snprintf(buf, FM_MAX_CLASS, "%s.%s", DDI_FM_DEVICE, detail);
5354 	ena = fm_ena_generate(0, FM_ENA_FMT1);
5355 	if (DDI_FM_EREPORT_CAP(igb->fm_capabilities)) {
5356 		ddi_fm_ereport_post(igb->dip, buf, ena, DDI_NOSLEEP,
5357 		    FM_VERSION, DATA_TYPE_UINT8, FM_EREPORT_VERS0, NULL);
5358 	}
5359 }
5360