xref: /titanic_41/usr/src/uts/common/io/igb/igb_main.c (revision 7a17cfad7ff3427e1ce7ecdbf566e442a7025ec9)
1 /*
2  * CDDL HEADER START
3  *
4  * Copyright(c) 2007-2009 Intel Corporation. All rights reserved.
5  * The contents of this file are subject to the terms of the
6  * Common Development and Distribution License (the "License").
7  * You may not use this file except in compliance with the License.
8  *
9  * You can obtain a copy of the license at:
10  *	http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When using or redistributing this file, you may do so under the
15  * License only. No other modification of this header is permitted.
16  *
17  * If applicable, add the following below this CDDL HEADER, with the
18  * fields enclosed by brackets "[]" replaced with your own identifying
19  * information: Portions Copyright [yyyy] [name of copyright owner]
20  *
21  * CDDL HEADER END
22  */
23 
24 /*
25  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
26  * Use is subject to license terms.
27  */
28 
29 #include "igb_sw.h"
30 
31 static char ident[] = "Intel 1Gb Ethernet";
32 static char igb_version[] = "igb 1.1.10";
33 
34 /*
35  * Local function protoypes
36  */
37 static int igb_register_mac(igb_t *);
38 static int igb_identify_hardware(igb_t *);
39 static int igb_regs_map(igb_t *);
40 static void igb_init_properties(igb_t *);
41 static int igb_init_driver_settings(igb_t *);
42 static void igb_init_locks(igb_t *);
43 static void igb_destroy_locks(igb_t *);
44 static int igb_init_mac_address(igb_t *);
45 static int igb_init(igb_t *);
46 static int igb_init_adapter(igb_t *);
47 static void igb_stop_adapter(igb_t *);
48 static int igb_reset(igb_t *);
49 static void igb_tx_clean(igb_t *);
50 static boolean_t igb_tx_drain(igb_t *);
51 static boolean_t igb_rx_drain(igb_t *);
52 static int igb_alloc_rings(igb_t *);
53 static void igb_free_rings(igb_t *);
54 static void igb_setup_rings(igb_t *);
55 static void igb_setup_rx(igb_t *);
56 static void igb_setup_tx(igb_t *);
57 static void igb_setup_rx_ring(igb_rx_ring_t *);
58 static void igb_setup_tx_ring(igb_tx_ring_t *);
59 static void igb_setup_rss(igb_t *);
60 static void igb_setup_mac_rss_classify(igb_t *);
61 static void igb_setup_mac_classify(igb_t *);
62 static void igb_init_unicst(igb_t *);
63 static void igb_setup_multicst(igb_t *);
64 static void igb_get_phy_state(igb_t *);
65 static void igb_get_conf(igb_t *);
66 static int igb_get_prop(igb_t *, char *, int, int, int);
67 static boolean_t igb_is_link_up(igb_t *);
68 static boolean_t igb_link_check(igb_t *);
69 static void igb_local_timer(void *);
70 static void igb_link_timer(void *);
71 static void igb_arm_watchdog_timer(igb_t *);
72 static void igb_start_watchdog_timer(igb_t *);
73 static void igb_restart_watchdog_timer(igb_t *);
74 static void igb_stop_watchdog_timer(igb_t *);
75 static void igb_start_link_timer(igb_t *);
76 static void igb_stop_link_timer(igb_t *);
77 static void igb_disable_adapter_interrupts(igb_t *);
78 static void igb_enable_adapter_interrupts_82575(igb_t *);
79 static void igb_enable_adapter_interrupts_82576(igb_t *);
80 static void igb_enable_adapter_interrupts_82580(igb_t *);
81 static boolean_t is_valid_mac_addr(uint8_t *);
82 static boolean_t igb_stall_check(igb_t *);
83 static boolean_t igb_set_loopback_mode(igb_t *, uint32_t);
84 static void igb_set_external_loopback(igb_t *);
85 static void igb_set_internal_mac_loopback(igb_t *);
86 static void igb_set_internal_phy_loopback(igb_t *);
87 static void igb_set_internal_serdes_loopback(igb_t *);
88 static boolean_t igb_find_mac_address(igb_t *);
89 static int igb_alloc_intrs(igb_t *);
90 static int igb_alloc_intr_handles(igb_t *, int);
91 static int igb_add_intr_handlers(igb_t *);
92 static void igb_rem_intr_handlers(igb_t *);
93 static void igb_rem_intrs(igb_t *);
94 static int igb_enable_intrs(igb_t *);
95 static int igb_disable_intrs(igb_t *);
96 static void igb_setup_msix_82575(igb_t *);
97 static void igb_setup_msix_82576(igb_t *);
98 static void igb_setup_msix_82580(igb_t *);
99 static uint_t igb_intr_legacy(void *, void *);
100 static uint_t igb_intr_msi(void *, void *);
101 static uint_t igb_intr_rx(void *, void *);
102 static uint_t igb_intr_tx(void *, void *);
103 static uint_t igb_intr_tx_other(void *, void *);
104 static void igb_intr_rx_work(igb_rx_ring_t *);
105 static void igb_intr_tx_work(igb_tx_ring_t *);
106 static void igb_intr_link_work(igb_t *);
107 static void igb_get_driver_control(struct e1000_hw *);
108 static void igb_release_driver_control(struct e1000_hw *);
109 
110 static int igb_attach(dev_info_t *, ddi_attach_cmd_t);
111 static int igb_detach(dev_info_t *, ddi_detach_cmd_t);
112 static int igb_resume(dev_info_t *);
113 static int igb_suspend(dev_info_t *);
114 static int igb_quiesce(dev_info_t *);
115 static void igb_unconfigure(dev_info_t *, igb_t *);
116 static int igb_fm_error_cb(dev_info_t *, ddi_fm_error_t *,
117     const void *);
118 static void igb_fm_init(igb_t *);
119 static void igb_fm_fini(igb_t *);
120 static void igb_release_multicast(igb_t *);
121 
122 static struct cb_ops igb_cb_ops = {
123 	nulldev,		/* cb_open */
124 	nulldev,		/* cb_close */
125 	nodev,			/* cb_strategy */
126 	nodev,			/* cb_print */
127 	nodev,			/* cb_dump */
128 	nodev,			/* cb_read */
129 	nodev,			/* cb_write */
130 	nodev,			/* cb_ioctl */
131 	nodev,			/* cb_devmap */
132 	nodev,			/* cb_mmap */
133 	nodev,			/* cb_segmap */
134 	nochpoll,		/* cb_chpoll */
135 	ddi_prop_op,		/* cb_prop_op */
136 	NULL,			/* cb_stream */
137 	D_MP | D_HOTPLUG,	/* cb_flag */
138 	CB_REV,			/* cb_rev */
139 	nodev,			/* cb_aread */
140 	nodev			/* cb_awrite */
141 };
142 
143 static struct dev_ops igb_dev_ops = {
144 	DEVO_REV,		/* devo_rev */
145 	0,			/* devo_refcnt */
146 	NULL,			/* devo_getinfo */
147 	nulldev,		/* devo_identify */
148 	nulldev,		/* devo_probe */
149 	igb_attach,		/* devo_attach */
150 	igb_detach,		/* devo_detach */
151 	nodev,			/* devo_reset */
152 	&igb_cb_ops,		/* devo_cb_ops */
153 	NULL,			/* devo_bus_ops */
154 	ddi_power,		/* devo_power */
155 	igb_quiesce,	/* devo_quiesce */
156 };
157 
158 static struct modldrv igb_modldrv = {
159 	&mod_driverops,		/* Type of module.  This one is a driver */
160 	ident,			/* Discription string */
161 	&igb_dev_ops,		/* driver ops */
162 };
163 
164 static struct modlinkage igb_modlinkage = {
165 	MODREV_1, &igb_modldrv, NULL
166 };
167 
168 /* Access attributes for register mapping */
169 ddi_device_acc_attr_t igb_regs_acc_attr = {
170 	DDI_DEVICE_ATTR_V1,
171 	DDI_STRUCTURE_LE_ACC,
172 	DDI_STRICTORDER_ACC,
173 	DDI_FLAGERR_ACC
174 };
175 
176 #define	IGB_M_CALLBACK_FLAGS	(MC_IOCTL | MC_GETCAPAB)
177 
178 static mac_callbacks_t igb_m_callbacks = {
179 	IGB_M_CALLBACK_FLAGS,
180 	igb_m_stat,
181 	igb_m_start,
182 	igb_m_stop,
183 	igb_m_promisc,
184 	igb_m_multicst,
185 	NULL,
186 	NULL,
187 	igb_m_ioctl,
188 	igb_m_getcapab
189 };
190 
191 /*
192  * Initialize capabilities of each supported adapter type
193  */
194 static adapter_info_t igb_82575_cap = {
195 	/* limits */
196 	4,		/* maximum number of rx queues */
197 	1,		/* minimum number of rx queues */
198 	4,		/* default number of rx queues */
199 	4,		/* maximum number of tx queues */
200 	1,		/* minimum number of tx queues */
201 	4,		/* default number of tx queues */
202 	65535,		/* maximum interrupt throttle rate */
203 	0,		/* minimum interrupt throttle rate */
204 	200,		/* default interrupt throttle rate */
205 
206 	/* function pointers */
207 	igb_enable_adapter_interrupts_82575,
208 	igb_setup_msix_82575,
209 
210 	/* capabilities */
211 	(IGB_FLAG_HAS_DCA |	/* capability flags */
212 	IGB_FLAG_VMDQ_POOL),
213 
214 	0xffc00000		/* mask for RXDCTL register */
215 };
216 
217 static adapter_info_t igb_82576_cap = {
218 	/* limits */
219 	16,		/* maximum number of rx queues */
220 	1,		/* minimum number of rx queues */
221 	4,		/* default number of rx queues */
222 	16,		/* maximum number of tx queues */
223 	1,		/* minimum number of tx queues */
224 	4,		/* default number of tx queues */
225 	65535,		/* maximum interrupt throttle rate */
226 	0,		/* minimum interrupt throttle rate */
227 	200,		/* default interrupt throttle rate */
228 
229 	/* function pointers */
230 	igb_enable_adapter_interrupts_82576,
231 	igb_setup_msix_82576,
232 
233 	/* capabilities */
234 	(IGB_FLAG_HAS_DCA |	/* capability flags */
235 	IGB_FLAG_VMDQ_POOL |
236 	IGB_FLAG_NEED_CTX_IDX),
237 
238 	0xffe00000		/* mask for RXDCTL register */
239 };
240 
241 static adapter_info_t igb_82580_cap = {
242 	/* limits */
243 	8,		/* maximum number of rx queues */
244 	1,		/* minimum number of rx queues */
245 	4,		/* default number of rx queues */
246 	8,		/* maximum number of tx queues */
247 	1,		/* minimum number of tx queues */
248 	4,		/* default number of tx queues */
249 	65535,		/* maximum interrupt throttle rate */
250 	0,		/* minimum interrupt throttle rate */
251 	200,		/* default interrupt throttle rate */
252 
253 	/* function pointers */
254 	igb_enable_adapter_interrupts_82580,
255 	igb_setup_msix_82580,
256 
257 	/* capabilities */
258 	(IGB_FLAG_HAS_DCA |	/* capability flags */
259 	IGB_FLAG_VMDQ_POOL |
260 	IGB_FLAG_NEED_CTX_IDX),
261 
262 	0xffe00000		/* mask for RXDCTL register */
263 };
264 
265 /*
266  * Module Initialization Functions
267  */
268 
269 int
270 _init(void)
271 {
272 	int status;
273 
274 	mac_init_ops(&igb_dev_ops, MODULE_NAME);
275 
276 	status = mod_install(&igb_modlinkage);
277 
278 	if (status != DDI_SUCCESS) {
279 		mac_fini_ops(&igb_dev_ops);
280 	}
281 
282 	return (status);
283 }
284 
285 int
286 _fini(void)
287 {
288 	int status;
289 
290 	status = mod_remove(&igb_modlinkage);
291 
292 	if (status == DDI_SUCCESS) {
293 		mac_fini_ops(&igb_dev_ops);
294 	}
295 
296 	return (status);
297 
298 }
299 
300 int
301 _info(struct modinfo *modinfop)
302 {
303 	int status;
304 
305 	status = mod_info(&igb_modlinkage, modinfop);
306 
307 	return (status);
308 }
309 
310 /*
311  * igb_attach - driver attach
312  *
313  * This function is the device specific initialization entry
314  * point. This entry point is required and must be written.
315  * The DDI_ATTACH command must be provided in the attach entry
316  * point. When attach() is called with cmd set to DDI_ATTACH,
317  * all normal kernel services (such as kmem_alloc(9F)) are
318  * available for use by the driver.
319  *
320  * The attach() function will be called once for each instance
321  * of  the  device  on  the  system with cmd set to DDI_ATTACH.
322  * Until attach() succeeds, the only driver entry points which
323  * may be called are open(9E) and getinfo(9E).
324  */
325 static int
326 igb_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
327 {
328 	igb_t *igb;
329 	struct igb_osdep *osdep;
330 	struct e1000_hw *hw;
331 	int instance;
332 
333 	/*
334 	 * Check the command and perform corresponding operations
335 	 */
336 	switch (cmd) {
337 	default:
338 		return (DDI_FAILURE);
339 
340 	case DDI_RESUME:
341 		return (igb_resume(devinfo));
342 
343 	case DDI_ATTACH:
344 		break;
345 	}
346 
347 	/* Get the device instance */
348 	instance = ddi_get_instance(devinfo);
349 
350 	/* Allocate memory for the instance data structure */
351 	igb = kmem_zalloc(sizeof (igb_t), KM_SLEEP);
352 
353 	igb->dip = devinfo;
354 	igb->instance = instance;
355 
356 	hw = &igb->hw;
357 	osdep = &igb->osdep;
358 	hw->back = osdep;
359 	osdep->igb = igb;
360 
361 	/* Attach the instance pointer to the dev_info data structure */
362 	ddi_set_driver_private(devinfo, igb);
363 
364 
365 	/* Initialize for fma support */
366 	igb->fm_capabilities = igb_get_prop(igb, "fm-capable",
367 	    0, 0x0f,
368 	    DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
369 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
370 	igb_fm_init(igb);
371 	igb->attach_progress |= ATTACH_PROGRESS_FMINIT;
372 
373 	/*
374 	 * Map PCI config space registers
375 	 */
376 	if (pci_config_setup(devinfo, &osdep->cfg_handle) != DDI_SUCCESS) {
377 		igb_error(igb, "Failed to map PCI configurations");
378 		goto attach_fail;
379 	}
380 	igb->attach_progress |= ATTACH_PROGRESS_PCI_CONFIG;
381 
382 	/*
383 	 * Identify the chipset family
384 	 */
385 	if (igb_identify_hardware(igb) != IGB_SUCCESS) {
386 		igb_error(igb, "Failed to identify hardware");
387 		goto attach_fail;
388 	}
389 
390 	/*
391 	 * Map device registers
392 	 */
393 	if (igb_regs_map(igb) != IGB_SUCCESS) {
394 		igb_error(igb, "Failed to map device registers");
395 		goto attach_fail;
396 	}
397 	igb->attach_progress |= ATTACH_PROGRESS_REGS_MAP;
398 
399 	/*
400 	 * Initialize driver parameters
401 	 */
402 	igb_init_properties(igb);
403 	igb->attach_progress |= ATTACH_PROGRESS_PROPS;
404 
405 	/*
406 	 * Allocate interrupts
407 	 */
408 	if (igb_alloc_intrs(igb) != IGB_SUCCESS) {
409 		igb_error(igb, "Failed to allocate interrupts");
410 		goto attach_fail;
411 	}
412 	igb->attach_progress |= ATTACH_PROGRESS_ALLOC_INTR;
413 
414 	/*
415 	 * Allocate rx/tx rings based on the ring numbers.
416 	 * The actual numbers of rx/tx rings are decided by the number of
417 	 * allocated interrupt vectors, so we should allocate the rings after
418 	 * interrupts are allocated.
419 	 */
420 	if (igb_alloc_rings(igb) != IGB_SUCCESS) {
421 		igb_error(igb, "Failed to allocate rx/tx rings or groups");
422 		goto attach_fail;
423 	}
424 	igb->attach_progress |= ATTACH_PROGRESS_ALLOC_RINGS;
425 
426 	/*
427 	 * Add interrupt handlers
428 	 */
429 	if (igb_add_intr_handlers(igb) != IGB_SUCCESS) {
430 		igb_error(igb, "Failed to add interrupt handlers");
431 		goto attach_fail;
432 	}
433 	igb->attach_progress |= ATTACH_PROGRESS_ADD_INTR;
434 
435 	/*
436 	 * Initialize driver parameters
437 	 */
438 	if (igb_init_driver_settings(igb) != IGB_SUCCESS) {
439 		igb_error(igb, "Failed to initialize driver settings");
440 		goto attach_fail;
441 	}
442 
443 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK) {
444 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
445 		goto attach_fail;
446 	}
447 
448 	/*
449 	 * Initialize mutexes for this device.
450 	 * Do this before enabling the interrupt handler and
451 	 * register the softint to avoid the condition where
452 	 * interrupt handler can try using uninitialized mutex
453 	 */
454 	igb_init_locks(igb);
455 	igb->attach_progress |= ATTACH_PROGRESS_LOCKS;
456 
457 	/*
458 	 * Allocate DMA resources
459 	 */
460 	if (igb_alloc_dma(igb) != IGB_SUCCESS) {
461 		igb_error(igb, "Failed to allocate DMA resources");
462 		goto attach_fail;
463 	}
464 	igb->attach_progress |= ATTACH_PROGRESS_ALLOC_DMA;
465 
466 	/*
467 	 * Initialize the adapter and setup the rx/tx rings
468 	 */
469 	if (igb_init(igb) != IGB_SUCCESS) {
470 		igb_error(igb, "Failed to initialize adapter");
471 		goto attach_fail;
472 	}
473 	igb->attach_progress |= ATTACH_PROGRESS_INIT_ADAPTER;
474 
475 	/*
476 	 * Initialize statistics
477 	 */
478 	if (igb_init_stats(igb) != IGB_SUCCESS) {
479 		igb_error(igb, "Failed to initialize statistics");
480 		goto attach_fail;
481 	}
482 	igb->attach_progress |= ATTACH_PROGRESS_STATS;
483 
484 	/*
485 	 * Initialize NDD parameters
486 	 */
487 	if (igb_nd_init(igb) != IGB_SUCCESS) {
488 		igb_error(igb, "Failed to initialize ndd");
489 		goto attach_fail;
490 	}
491 	igb->attach_progress |= ATTACH_PROGRESS_NDD;
492 
493 	/*
494 	 * Register the driver to the MAC
495 	 */
496 	if (igb_register_mac(igb) != IGB_SUCCESS) {
497 		igb_error(igb, "Failed to register MAC");
498 		goto attach_fail;
499 	}
500 	igb->attach_progress |= ATTACH_PROGRESS_MAC;
501 
502 	/*
503 	 * Now that mutex locks are initialized, and the chip is also
504 	 * initialized, enable interrupts.
505 	 */
506 	if (igb_enable_intrs(igb) != IGB_SUCCESS) {
507 		igb_error(igb, "Failed to enable DDI interrupts");
508 		goto attach_fail;
509 	}
510 	igb->attach_progress |= ATTACH_PROGRESS_ENABLE_INTR;
511 
512 	igb_log(igb, "%s", igb_version);
513 	atomic_or_32(&igb->igb_state, IGB_INITIALIZED);
514 
515 	return (DDI_SUCCESS);
516 
517 attach_fail:
518 	igb_unconfigure(devinfo, igb);
519 	return (DDI_FAILURE);
520 }
521 
522 /*
523  * igb_detach - driver detach
524  *
525  * The detach() function is the complement of the attach routine.
526  * If cmd is set to DDI_DETACH, detach() is used to remove  the
527  * state  associated  with  a  given  instance of a device node
528  * prior to the removal of that instance from the system.
529  *
530  * The detach() function will be called once for each  instance
531  * of the device for which there has been a successful attach()
532  * once there are no longer  any  opens  on  the  device.
533  *
534  * Interrupts routine are disabled, All memory allocated by this
535  * driver are freed.
536  */
537 static int
538 igb_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
539 {
540 	igb_t *igb;
541 
542 	/*
543 	 * Check detach command
544 	 */
545 	switch (cmd) {
546 	default:
547 		return (DDI_FAILURE);
548 
549 	case DDI_SUSPEND:
550 		return (igb_suspend(devinfo));
551 
552 	case DDI_DETACH:
553 		break;
554 	}
555 
556 
557 	/*
558 	 * Get the pointer to the driver private data structure
559 	 */
560 	igb = (igb_t *)ddi_get_driver_private(devinfo);
561 	if (igb == NULL)
562 		return (DDI_FAILURE);
563 
564 	/*
565 	 * Unregister MAC. If failed, we have to fail the detach
566 	 */
567 	if (mac_unregister(igb->mac_hdl) != 0) {
568 		igb_error(igb, "Failed to unregister MAC");
569 		return (DDI_FAILURE);
570 	}
571 	igb->attach_progress &= ~ATTACH_PROGRESS_MAC;
572 
573 	/*
574 	 * If the device is still running, it needs to be stopped first.
575 	 * This check is necessary because under some specific circumstances,
576 	 * the detach routine can be called without stopping the interface
577 	 * first.
578 	 */
579 	mutex_enter(&igb->gen_lock);
580 	if (igb->igb_state & IGB_STARTED) {
581 		atomic_and_32(&igb->igb_state, ~IGB_STARTED);
582 		igb_stop(igb);
583 		mutex_exit(&igb->gen_lock);
584 		/* Disable and stop the watchdog timer */
585 		igb_disable_watchdog_timer(igb);
586 	} else
587 		mutex_exit(&igb->gen_lock);
588 
589 	/*
590 	 * Check if there are still rx buffers held by the upper layer.
591 	 * If so, fail the detach.
592 	 */
593 	if (!igb_rx_drain(igb))
594 		return (DDI_FAILURE);
595 
596 	/*
597 	 * Do the remaining unconfigure routines
598 	 */
599 	igb_unconfigure(devinfo, igb);
600 
601 	return (DDI_SUCCESS);
602 }
603 
604 /*
605  * quiesce(9E) entry point.
606  *
607  * This function is called when the system is single-threaded at high
608  * PIL with preemption disabled. Therefore, this function must not be
609  * blocked.
610  *
611  * This function returns DDI_SUCCESS on success, or DDI_FAILURE on failure.
612  * DDI_FAILURE indicates an error condition and should almost never happen.
613  */
614 static int
615 igb_quiesce(dev_info_t *devinfo)
616 {
617 	igb_t *igb;
618 	struct e1000_hw *hw;
619 
620 	igb = (igb_t *)ddi_get_driver_private(devinfo);
621 
622 	if (igb == NULL)
623 		return (DDI_FAILURE);
624 
625 	hw = &igb->hw;
626 
627 	/*
628 	 * Disable the adapter interrupts
629 	 */
630 	igb_disable_adapter_interrupts(igb);
631 
632 	/* Tell firmware driver is no longer in control */
633 	igb_release_driver_control(hw);
634 
635 	/*
636 	 * Reset the chipset
637 	 */
638 	(void) e1000_reset_hw(hw);
639 
640 	/*
641 	 * Reset PHY if possible
642 	 */
643 	if (e1000_check_reset_block(hw) == E1000_SUCCESS)
644 		(void) e1000_phy_hw_reset(hw);
645 
646 	return (DDI_SUCCESS);
647 }
648 
649 /*
650  * igb_unconfigure - release all resources held by this instance
651  */
652 static void
653 igb_unconfigure(dev_info_t *devinfo, igb_t *igb)
654 {
655 	/*
656 	 * Disable interrupt
657 	 */
658 	if (igb->attach_progress & ATTACH_PROGRESS_ENABLE_INTR) {
659 		(void) igb_disable_intrs(igb);
660 	}
661 
662 	/*
663 	 * Unregister MAC
664 	 */
665 	if (igb->attach_progress & ATTACH_PROGRESS_MAC) {
666 		(void) mac_unregister(igb->mac_hdl);
667 	}
668 
669 	/*
670 	 * Free ndd parameters
671 	 */
672 	if (igb->attach_progress & ATTACH_PROGRESS_NDD) {
673 		igb_nd_cleanup(igb);
674 	}
675 
676 	/*
677 	 * Free statistics
678 	 */
679 	if (igb->attach_progress & ATTACH_PROGRESS_STATS) {
680 		kstat_delete((kstat_t *)igb->igb_ks);
681 	}
682 
683 	/*
684 	 * Remove interrupt handlers
685 	 */
686 	if (igb->attach_progress & ATTACH_PROGRESS_ADD_INTR) {
687 		igb_rem_intr_handlers(igb);
688 	}
689 
690 	/*
691 	 * Remove interrupts
692 	 */
693 	if (igb->attach_progress & ATTACH_PROGRESS_ALLOC_INTR) {
694 		igb_rem_intrs(igb);
695 	}
696 
697 	/*
698 	 * Remove driver properties
699 	 */
700 	if (igb->attach_progress & ATTACH_PROGRESS_PROPS) {
701 		(void) ddi_prop_remove_all(devinfo);
702 	}
703 
704 	/*
705 	 * Release the DMA resources of rx/tx rings
706 	 */
707 	if (igb->attach_progress & ATTACH_PROGRESS_ALLOC_DMA) {
708 		igb_free_dma(igb);
709 	}
710 
711 	/*
712 	 * Stop the adapter
713 	 */
714 	if (igb->attach_progress & ATTACH_PROGRESS_INIT_ADAPTER) {
715 		mutex_enter(&igb->gen_lock);
716 		igb_stop_adapter(igb);
717 		mutex_exit(&igb->gen_lock);
718 		if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
719 			ddi_fm_service_impact(igb->dip, DDI_SERVICE_UNAFFECTED);
720 	}
721 
722 	/*
723 	 * Free multicast table
724 	 */
725 	igb_release_multicast(igb);
726 
727 	/*
728 	 * Free register handle
729 	 */
730 	if (igb->attach_progress & ATTACH_PROGRESS_REGS_MAP) {
731 		if (igb->osdep.reg_handle != NULL)
732 			ddi_regs_map_free(&igb->osdep.reg_handle);
733 	}
734 
735 	/*
736 	 * Free PCI config handle
737 	 */
738 	if (igb->attach_progress & ATTACH_PROGRESS_PCI_CONFIG) {
739 		if (igb->osdep.cfg_handle != NULL)
740 			pci_config_teardown(&igb->osdep.cfg_handle);
741 	}
742 
743 	/*
744 	 * Free locks
745 	 */
746 	if (igb->attach_progress & ATTACH_PROGRESS_LOCKS) {
747 		igb_destroy_locks(igb);
748 	}
749 
750 	/*
751 	 * Free the rx/tx rings
752 	 */
753 	if (igb->attach_progress & ATTACH_PROGRESS_ALLOC_RINGS) {
754 		igb_free_rings(igb);
755 	}
756 
757 	/*
758 	 * Remove FMA
759 	 */
760 	if (igb->attach_progress & ATTACH_PROGRESS_FMINIT) {
761 		igb_fm_fini(igb);
762 	}
763 
764 	/*
765 	 * Free the driver data structure
766 	 */
767 	kmem_free(igb, sizeof (igb_t));
768 
769 	ddi_set_driver_private(devinfo, NULL);
770 }
771 
772 /*
773  * igb_register_mac - Register the driver and its function pointers with
774  * the GLD interface
775  */
776 static int
777 igb_register_mac(igb_t *igb)
778 {
779 	struct e1000_hw *hw = &igb->hw;
780 	mac_register_t *mac;
781 	int status;
782 
783 	if ((mac = mac_alloc(MAC_VERSION)) == NULL)
784 		return (IGB_FAILURE);
785 
786 	mac->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
787 	mac->m_driver = igb;
788 	mac->m_dip = igb->dip;
789 	mac->m_src_addr = hw->mac.addr;
790 	mac->m_callbacks = &igb_m_callbacks;
791 	mac->m_min_sdu = 0;
792 	mac->m_max_sdu = igb->max_frame_size -
793 	    sizeof (struct ether_vlan_header) - ETHERFCSL;
794 	mac->m_margin = VLAN_TAGSZ;
795 	mac->m_v12n = MAC_VIRT_LEVEL1;
796 
797 	status = mac_register(mac, &igb->mac_hdl);
798 
799 	mac_free(mac);
800 
801 	return ((status == 0) ? IGB_SUCCESS : IGB_FAILURE);
802 }
803 
804 /*
805  * igb_identify_hardware - Identify the type of the chipset
806  */
807 static int
808 igb_identify_hardware(igb_t *igb)
809 {
810 	struct e1000_hw *hw = &igb->hw;
811 	struct igb_osdep *osdep = &igb->osdep;
812 
813 	/*
814 	 * Get the device id
815 	 */
816 	hw->vendor_id =
817 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_VENID);
818 	hw->device_id =
819 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_DEVID);
820 	hw->revision_id =
821 	    pci_config_get8(osdep->cfg_handle, PCI_CONF_REVID);
822 	hw->subsystem_device_id =
823 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBSYSID);
824 	hw->subsystem_vendor_id =
825 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBVENID);
826 
827 	/*
828 	 * Set the mac type of the adapter based on the device id
829 	 */
830 	if (e1000_set_mac_type(hw) != E1000_SUCCESS) {
831 		return (IGB_FAILURE);
832 	}
833 
834 	/*
835 	 * Install adapter capabilities based on mac type
836 	 */
837 	switch (hw->mac.type) {
838 	case e1000_82575:
839 		igb->capab = &igb_82575_cap;
840 		break;
841 	case e1000_82576:
842 		igb->capab = &igb_82576_cap;
843 		break;
844 	case e1000_82580:
845 		igb->capab = &igb_82580_cap;
846 		break;
847 	default:
848 		return (IGB_FAILURE);
849 	}
850 
851 	return (IGB_SUCCESS);
852 }
853 
854 /*
855  * igb_regs_map - Map the device registers
856  */
857 static int
858 igb_regs_map(igb_t *igb)
859 {
860 	dev_info_t *devinfo = igb->dip;
861 	struct e1000_hw *hw = &igb->hw;
862 	struct igb_osdep *osdep = &igb->osdep;
863 	off_t mem_size;
864 
865 	/*
866 	 * First get the size of device registers to be mapped.
867 	 */
868 	if (ddi_dev_regsize(devinfo, IGB_ADAPTER_REGSET, &mem_size) !=
869 	    DDI_SUCCESS) {
870 		return (IGB_FAILURE);
871 	}
872 
873 	/*
874 	 * Call ddi_regs_map_setup() to map registers
875 	 */
876 	if ((ddi_regs_map_setup(devinfo, IGB_ADAPTER_REGSET,
877 	    (caddr_t *)&hw->hw_addr, 0,
878 	    mem_size, &igb_regs_acc_attr,
879 	    &osdep->reg_handle)) != DDI_SUCCESS) {
880 		return (IGB_FAILURE);
881 	}
882 
883 	return (IGB_SUCCESS);
884 }
885 
886 /*
887  * igb_init_properties - Initialize driver properties
888  */
889 static void
890 igb_init_properties(igb_t *igb)
891 {
892 	/*
893 	 * Get conf file properties, including link settings
894 	 * jumbo frames, ring number, descriptor number, etc.
895 	 */
896 	igb_get_conf(igb);
897 }
898 
899 /*
900  * igb_init_driver_settings - Initialize driver settings
901  *
902  * The settings include hardware function pointers, bus information,
903  * rx/tx rings settings, link state, and any other parameters that
904  * need to be setup during driver initialization.
905  */
906 static int
907 igb_init_driver_settings(igb_t *igb)
908 {
909 	struct e1000_hw *hw = &igb->hw;
910 	igb_rx_ring_t *rx_ring;
911 	igb_tx_ring_t *tx_ring;
912 	uint32_t rx_size;
913 	uint32_t tx_size;
914 	int i;
915 
916 	/*
917 	 * Initialize chipset specific hardware function pointers
918 	 */
919 	if (e1000_setup_init_funcs(hw, B_TRUE) != E1000_SUCCESS) {
920 		return (IGB_FAILURE);
921 	}
922 
923 	/*
924 	 * Get bus information
925 	 */
926 	if (e1000_get_bus_info(hw) != E1000_SUCCESS) {
927 		return (IGB_FAILURE);
928 	}
929 
930 	/*
931 	 * Get the system page size
932 	 */
933 	igb->page_size = ddi_ptob(igb->dip, (ulong_t)1);
934 
935 	/*
936 	 * Set rx buffer size
937 	 * The IP header alignment room is counted in the calculation.
938 	 * The rx buffer size is in unit of 1K that is required by the
939 	 * chipset hardware.
940 	 */
941 	rx_size = igb->max_frame_size + IPHDR_ALIGN_ROOM;
942 	igb->rx_buf_size = ((rx_size >> 10) +
943 	    ((rx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
944 
945 	/*
946 	 * Set tx buffer size
947 	 */
948 	tx_size = igb->max_frame_size;
949 	igb->tx_buf_size = ((tx_size >> 10) +
950 	    ((tx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
951 
952 	/*
953 	 * Initialize rx/tx rings parameters
954 	 */
955 	for (i = 0; i < igb->num_rx_rings; i++) {
956 		rx_ring = &igb->rx_rings[i];
957 		rx_ring->index = i;
958 		rx_ring->igb = igb;
959 
960 		rx_ring->ring_size = igb->rx_ring_size;
961 		rx_ring->free_list_size = igb->rx_ring_size;
962 		rx_ring->copy_thresh = igb->rx_copy_thresh;
963 		rx_ring->limit_per_intr = igb->rx_limit_per_intr;
964 	}
965 
966 	for (i = 0; i < igb->num_tx_rings; i++) {
967 		tx_ring = &igb->tx_rings[i];
968 		tx_ring->index = i;
969 		tx_ring->igb = igb;
970 		if (igb->tx_head_wb_enable)
971 			tx_ring->tx_recycle = igb_tx_recycle_head_wb;
972 		else
973 			tx_ring->tx_recycle = igb_tx_recycle_legacy;
974 
975 		tx_ring->ring_size = igb->tx_ring_size;
976 		tx_ring->free_list_size = igb->tx_ring_size +
977 		    (igb->tx_ring_size >> 1);
978 		tx_ring->copy_thresh = igb->tx_copy_thresh;
979 		tx_ring->recycle_thresh = igb->tx_recycle_thresh;
980 		tx_ring->overload_thresh = igb->tx_overload_thresh;
981 		tx_ring->resched_thresh = igb->tx_resched_thresh;
982 	}
983 
984 	/*
985 	 * Initialize values of interrupt throttling rates
986 	 */
987 	for (i = 1; i < MAX_NUM_EITR; i++)
988 		igb->intr_throttling[i] = igb->intr_throttling[0];
989 
990 	/*
991 	 * The initial link state should be "unknown"
992 	 */
993 	igb->link_state = LINK_STATE_UNKNOWN;
994 
995 	return (IGB_SUCCESS);
996 }
997 
998 /*
999  * igb_init_locks - Initialize locks
1000  */
1001 static void
1002 igb_init_locks(igb_t *igb)
1003 {
1004 	igb_rx_ring_t *rx_ring;
1005 	igb_tx_ring_t *tx_ring;
1006 	int i;
1007 
1008 	for (i = 0; i < igb->num_rx_rings; i++) {
1009 		rx_ring = &igb->rx_rings[i];
1010 		mutex_init(&rx_ring->rx_lock, NULL,
1011 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1012 		mutex_init(&rx_ring->recycle_lock, NULL,
1013 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1014 	}
1015 
1016 	for (i = 0; i < igb->num_tx_rings; i++) {
1017 		tx_ring = &igb->tx_rings[i];
1018 		mutex_init(&tx_ring->tx_lock, NULL,
1019 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1020 		mutex_init(&tx_ring->recycle_lock, NULL,
1021 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1022 		mutex_init(&tx_ring->tcb_head_lock, NULL,
1023 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1024 		mutex_init(&tx_ring->tcb_tail_lock, NULL,
1025 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1026 	}
1027 
1028 	mutex_init(&igb->gen_lock, NULL,
1029 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1030 
1031 	mutex_init(&igb->watchdog_lock, NULL,
1032 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1033 
1034 	mutex_init(&igb->link_lock, NULL,
1035 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1036 }
1037 
1038 /*
1039  * igb_destroy_locks - Destroy locks
1040  */
1041 static void
1042 igb_destroy_locks(igb_t *igb)
1043 {
1044 	igb_rx_ring_t *rx_ring;
1045 	igb_tx_ring_t *tx_ring;
1046 	int i;
1047 
1048 	for (i = 0; i < igb->num_rx_rings; i++) {
1049 		rx_ring = &igb->rx_rings[i];
1050 		mutex_destroy(&rx_ring->rx_lock);
1051 		mutex_destroy(&rx_ring->recycle_lock);
1052 	}
1053 
1054 	for (i = 0; i < igb->num_tx_rings; i++) {
1055 		tx_ring = &igb->tx_rings[i];
1056 		mutex_destroy(&tx_ring->tx_lock);
1057 		mutex_destroy(&tx_ring->recycle_lock);
1058 		mutex_destroy(&tx_ring->tcb_head_lock);
1059 		mutex_destroy(&tx_ring->tcb_tail_lock);
1060 	}
1061 
1062 	mutex_destroy(&igb->gen_lock);
1063 	mutex_destroy(&igb->watchdog_lock);
1064 	mutex_destroy(&igb->link_lock);
1065 }
1066 
1067 static int
1068 igb_resume(dev_info_t *devinfo)
1069 {
1070 	igb_t *igb;
1071 
1072 	igb = (igb_t *)ddi_get_driver_private(devinfo);
1073 	if (igb == NULL)
1074 		return (DDI_FAILURE);
1075 
1076 	mutex_enter(&igb->gen_lock);
1077 
1078 	if (igb->igb_state & IGB_STARTED) {
1079 		if (igb_start(igb) != IGB_SUCCESS) {
1080 			mutex_exit(&igb->gen_lock);
1081 			return (DDI_FAILURE);
1082 		}
1083 
1084 		/*
1085 		 * Enable and start the watchdog timer
1086 		 */
1087 		igb_enable_watchdog_timer(igb);
1088 	}
1089 
1090 	atomic_and_32(&igb->igb_state, ~IGB_SUSPENDED);
1091 
1092 	mutex_exit(&igb->gen_lock);
1093 
1094 	return (DDI_SUCCESS);
1095 }
1096 
1097 static int
1098 igb_suspend(dev_info_t *devinfo)
1099 {
1100 	igb_t *igb;
1101 
1102 	igb = (igb_t *)ddi_get_driver_private(devinfo);
1103 	if (igb == NULL)
1104 		return (DDI_FAILURE);
1105 
1106 	mutex_enter(&igb->gen_lock);
1107 
1108 	atomic_or_32(&igb->igb_state, IGB_SUSPENDED);
1109 
1110 	if (!(igb->igb_state & IGB_STARTED)) {
1111 		mutex_exit(&igb->gen_lock);
1112 		return (DDI_SUCCESS);
1113 	}
1114 
1115 	igb_stop(igb);
1116 
1117 	mutex_exit(&igb->gen_lock);
1118 
1119 	/*
1120 	 * Disable and stop the watchdog timer
1121 	 */
1122 	igb_disable_watchdog_timer(igb);
1123 
1124 	return (DDI_SUCCESS);
1125 }
1126 
1127 static int
1128 igb_init(igb_t *igb)
1129 {
1130 	int i;
1131 
1132 	mutex_enter(&igb->gen_lock);
1133 
1134 	/*
1135 	 * Initilize the adapter
1136 	 */
1137 	if (igb_init_adapter(igb) != IGB_SUCCESS) {
1138 		mutex_exit(&igb->gen_lock);
1139 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1140 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1141 		return (IGB_FAILURE);
1142 	}
1143 
1144 	/*
1145 	 * Setup the rx/tx rings
1146 	 */
1147 	for (i = 0; i < igb->num_rx_rings; i++)
1148 		mutex_enter(&igb->rx_rings[i].rx_lock);
1149 	for (i = 0; i < igb->num_tx_rings; i++)
1150 		mutex_enter(&igb->tx_rings[i].tx_lock);
1151 
1152 	igb_setup_rings(igb);
1153 
1154 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1155 		mutex_exit(&igb->tx_rings[i].tx_lock);
1156 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1157 		mutex_exit(&igb->rx_rings[i].rx_lock);
1158 
1159 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
1160 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1161 		return (IGB_FAILURE);
1162 	}
1163 
1164 	mutex_exit(&igb->gen_lock);
1165 
1166 	return (IGB_SUCCESS);
1167 }
1168 
1169 /*
1170  * igb_init_mac_address - Initialize the default MAC address
1171  *
1172  * On success, the MAC address is entered in the igb->hw.mac.addr
1173  * and hw->mac.perm_addr fields and the adapter's RAR(0) receive
1174  * address register.
1175  *
1176  * Important side effects:
1177  * 1. adapter is reset - this is required to put it in a known state.
1178  * 2. all of non-volatile memory (NVM) is read & checksummed - NVM is where
1179  * MAC address and all default settings are stored, so a valid checksum
1180  * is required.
1181  */
1182 static int
1183 igb_init_mac_address(igb_t *igb)
1184 {
1185 	struct e1000_hw *hw = &igb->hw;
1186 
1187 	ASSERT(mutex_owned(&igb->gen_lock));
1188 
1189 	/*
1190 	 * Reset chipset to put the hardware in a known state
1191 	 * before we try to get MAC address from NVM.
1192 	 */
1193 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1194 		igb_error(igb, "Adapter reset failed.");
1195 		goto init_mac_fail;
1196 	}
1197 
1198 	/*
1199 	 * NVM validation
1200 	 */
1201 	if (e1000_validate_nvm_checksum(hw) < 0) {
1202 		/*
1203 		 * Some PCI-E parts fail the first check due to
1204 		 * the link being in sleep state.  Call it again,
1205 		 * if it fails a second time its a real issue.
1206 		 */
1207 		if (e1000_validate_nvm_checksum(hw) < 0) {
1208 			igb_error(igb,
1209 			    "Invalid NVM checksum. Please contact "
1210 			    "the vendor to update the NVM.");
1211 			goto init_mac_fail;
1212 		}
1213 	}
1214 
1215 	/*
1216 	 * Get the mac address
1217 	 * This function should handle SPARC case correctly.
1218 	 */
1219 	if (!igb_find_mac_address(igb)) {
1220 		igb_error(igb, "Failed to get the mac address");
1221 		goto init_mac_fail;
1222 	}
1223 
1224 	/* Validate mac address */
1225 	if (!is_valid_mac_addr(hw->mac.addr)) {
1226 		igb_error(igb, "Invalid mac address");
1227 		goto init_mac_fail;
1228 	}
1229 
1230 	return (IGB_SUCCESS);
1231 
1232 init_mac_fail:
1233 	return (IGB_FAILURE);
1234 }
1235 
1236 /*
1237  * igb_init_adapter - Initialize the adapter
1238  */
1239 static int
1240 igb_init_adapter(igb_t *igb)
1241 {
1242 	struct e1000_hw *hw = &igb->hw;
1243 	uint32_t pba;
1244 	uint32_t high_water;
1245 	int i;
1246 
1247 	ASSERT(mutex_owned(&igb->gen_lock));
1248 
1249 	/*
1250 	 * In order to obtain the default MAC address, this will reset the
1251 	 * adapter and validate the NVM that the address and many other
1252 	 * default settings come from.
1253 	 */
1254 	if (igb_init_mac_address(igb) != IGB_SUCCESS) {
1255 		igb_error(igb, "Failed to initialize MAC address");
1256 		goto init_adapter_fail;
1257 	}
1258 
1259 	/*
1260 	 * Setup flow control
1261 	 *
1262 	 * These parameters set thresholds for the adapter's generation(Tx)
1263 	 * and response(Rx) to Ethernet PAUSE frames.  These are just threshold
1264 	 * settings.  Flow control is enabled or disabled in the configuration
1265 	 * file.
1266 	 * High-water mark is set down from the top of the rx fifo (not
1267 	 * sensitive to max_frame_size) and low-water is set just below
1268 	 * high-water mark.
1269 	 * The high water mark must be low enough to fit one full frame above
1270 	 * it in the rx FIFO.  Should be the lower of:
1271 	 * 90% of the Rx FIFO size, or the full Rx FIFO size minus one full
1272 	 * frame.
1273 	 */
1274 	/*
1275 	 * The default setting of PBA is correct for 82575 and other supported
1276 	 * adapters do not have the E1000_PBA register, so PBA value is only
1277 	 * used for calculation here and is never written to the adapter.
1278 	 */
1279 	if (hw->mac.type == e1000_82575) {
1280 		pba = E1000_PBA_34K;
1281 	} else {
1282 		pba = E1000_PBA_64K;
1283 	}
1284 
1285 	high_water = min(((pba << 10) * 9 / 10),
1286 	    ((pba << 10) - igb->max_frame_size));
1287 
1288 	if (hw->mac.type == e1000_82575) {
1289 		/* 8-byte granularity */
1290 		hw->fc.high_water = high_water & 0xFFF8;
1291 		hw->fc.low_water = hw->fc.high_water - 8;
1292 	} else {
1293 		/* 16-byte granularity */
1294 		hw->fc.high_water = high_water & 0xFFF0;
1295 		hw->fc.low_water = hw->fc.high_water - 16;
1296 	}
1297 
1298 	hw->fc.pause_time = E1000_FC_PAUSE_TIME;
1299 	hw->fc.send_xon = B_TRUE;
1300 
1301 	(void) e1000_validate_mdi_setting(hw);
1302 
1303 	/*
1304 	 * Reset the chipset hardware the second time to put PBA settings
1305 	 * into effect.
1306 	 */
1307 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1308 		igb_error(igb, "Second reset failed");
1309 		goto init_adapter_fail;
1310 	}
1311 
1312 	/*
1313 	 * Don't wait for auto-negotiation to complete
1314 	 */
1315 	hw->phy.autoneg_wait_to_complete = B_FALSE;
1316 
1317 	/*
1318 	 * Copper options
1319 	 */
1320 	if (hw->phy.media_type == e1000_media_type_copper) {
1321 		hw->phy.mdix = 0;	/* AUTO_ALL_MODES */
1322 		hw->phy.disable_polarity_correction = B_FALSE;
1323 		hw->phy.ms_type = e1000_ms_hw_default; /* E1000_MASTER_SLAVE */
1324 	}
1325 
1326 	/*
1327 	 * Initialize link settings
1328 	 */
1329 	(void) igb_setup_link(igb, B_FALSE);
1330 
1331 	/*
1332 	 * Configure/Initialize hardware
1333 	 */
1334 	if (e1000_init_hw(hw) != E1000_SUCCESS) {
1335 		igb_error(igb, "Failed to initialize hardware");
1336 		goto init_adapter_fail;
1337 	}
1338 
1339 	/*
1340 	 *  Start the link setup timer
1341 	 */
1342 	igb_start_link_timer(igb);
1343 
1344 	/*
1345 	 * Disable wakeup control by default
1346 	 */
1347 	E1000_WRITE_REG(hw, E1000_WUC, 0);
1348 
1349 	/*
1350 	 * Record phy info in hw struct
1351 	 */
1352 	(void) e1000_get_phy_info(hw);
1353 
1354 	/*
1355 	 * Make sure driver has control
1356 	 */
1357 	igb_get_driver_control(hw);
1358 
1359 	/*
1360 	 * Restore LED settings to the default from EEPROM
1361 	 * to meet the standard for Sun platforms.
1362 	 */
1363 	(void) e1000_cleanup_led(hw);
1364 
1365 	/*
1366 	 * Setup MSI-X interrupts
1367 	 */
1368 	if (igb->intr_type == DDI_INTR_TYPE_MSIX)
1369 		igb->capab->setup_msix(igb);
1370 
1371 	/*
1372 	 * Initialize unicast addresses.
1373 	 */
1374 	igb_init_unicst(igb);
1375 
1376 	/*
1377 	 * Setup and initialize the mctable structures.
1378 	 */
1379 	igb_setup_multicst(igb);
1380 
1381 	/*
1382 	 * Set interrupt throttling rate
1383 	 */
1384 	for (i = 0; i < igb->intr_cnt; i++)
1385 		E1000_WRITE_REG(hw, E1000_EITR(i), igb->intr_throttling[i]);
1386 
1387 	/*
1388 	 * Save the state of the phy
1389 	 */
1390 	igb_get_phy_state(igb);
1391 
1392 	return (IGB_SUCCESS);
1393 
1394 init_adapter_fail:
1395 	/*
1396 	 * Reset PHY if possible
1397 	 */
1398 	if (e1000_check_reset_block(hw) == E1000_SUCCESS)
1399 		(void) e1000_phy_hw_reset(hw);
1400 
1401 	return (IGB_FAILURE);
1402 }
1403 
1404 /*
1405  * igb_stop_adapter - Stop the adapter
1406  */
1407 static void
1408 igb_stop_adapter(igb_t *igb)
1409 {
1410 	struct e1000_hw *hw = &igb->hw;
1411 
1412 	ASSERT(mutex_owned(&igb->gen_lock));
1413 
1414 	/* Stop the link setup timer */
1415 	igb_stop_link_timer(igb);
1416 
1417 	/* Tell firmware driver is no longer in control */
1418 	igb_release_driver_control(hw);
1419 
1420 	/*
1421 	 * Reset the chipset
1422 	 */
1423 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1424 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1425 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1426 	}
1427 
1428 	/*
1429 	 * e1000_phy_hw_reset is not needed here, MAC reset above is sufficient
1430 	 */
1431 }
1432 
1433 /*
1434  * igb_reset - Reset the chipset and restart the driver.
1435  *
1436  * It involves stopping and re-starting the chipset,
1437  * and re-configuring the rx/tx rings.
1438  */
1439 static int
1440 igb_reset(igb_t *igb)
1441 {
1442 	int i;
1443 
1444 	mutex_enter(&igb->gen_lock);
1445 
1446 	ASSERT(igb->igb_state & IGB_STARTED);
1447 	atomic_and_32(&igb->igb_state, ~IGB_STARTED);
1448 
1449 	/*
1450 	 * Disable the adapter interrupts to stop any rx/tx activities
1451 	 * before draining pending data and resetting hardware.
1452 	 */
1453 	igb_disable_adapter_interrupts(igb);
1454 
1455 	/*
1456 	 * Drain the pending transmit packets
1457 	 */
1458 	(void) igb_tx_drain(igb);
1459 
1460 	for (i = 0; i < igb->num_rx_rings; i++)
1461 		mutex_enter(&igb->rx_rings[i].rx_lock);
1462 	for (i = 0; i < igb->num_tx_rings; i++)
1463 		mutex_enter(&igb->tx_rings[i].tx_lock);
1464 
1465 	/*
1466 	 * Stop the adapter
1467 	 */
1468 	igb_stop_adapter(igb);
1469 
1470 	/*
1471 	 * Clean the pending tx data/resources
1472 	 */
1473 	igb_tx_clean(igb);
1474 
1475 	/*
1476 	 * Start the adapter
1477 	 */
1478 	if (igb_init_adapter(igb) != IGB_SUCCESS) {
1479 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1480 		goto reset_failure;
1481 	}
1482 
1483 	/*
1484 	 * Setup the rx/tx rings
1485 	 */
1486 	igb_setup_rings(igb);
1487 
1488 	atomic_and_32(&igb->igb_state, ~(IGB_ERROR | IGB_STALL));
1489 
1490 	/*
1491 	 * Enable adapter interrupts
1492 	 * The interrupts must be enabled after the driver state is START
1493 	 */
1494 	igb->capab->enable_intr(igb);
1495 
1496 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK)
1497 		goto reset_failure;
1498 
1499 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1500 		goto reset_failure;
1501 
1502 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1503 		mutex_exit(&igb->tx_rings[i].tx_lock);
1504 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1505 		mutex_exit(&igb->rx_rings[i].rx_lock);
1506 
1507 	atomic_or_32(&igb->igb_state, IGB_STARTED);
1508 
1509 	mutex_exit(&igb->gen_lock);
1510 
1511 	return (IGB_SUCCESS);
1512 
1513 reset_failure:
1514 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1515 		mutex_exit(&igb->tx_rings[i].tx_lock);
1516 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1517 		mutex_exit(&igb->rx_rings[i].rx_lock);
1518 
1519 	mutex_exit(&igb->gen_lock);
1520 
1521 	ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1522 
1523 	return (IGB_FAILURE);
1524 }
1525 
1526 /*
1527  * igb_tx_clean - Clean the pending transmit packets and DMA resources
1528  */
1529 static void
1530 igb_tx_clean(igb_t *igb)
1531 {
1532 	igb_tx_ring_t *tx_ring;
1533 	tx_control_block_t *tcb;
1534 	link_list_t pending_list;
1535 	uint32_t desc_num;
1536 	int i, j;
1537 
1538 	LINK_LIST_INIT(&pending_list);
1539 
1540 	for (i = 0; i < igb->num_tx_rings; i++) {
1541 		tx_ring = &igb->tx_rings[i];
1542 
1543 		mutex_enter(&tx_ring->recycle_lock);
1544 
1545 		/*
1546 		 * Clean the pending tx data - the pending packets in the
1547 		 * work_list that have no chances to be transmitted again.
1548 		 *
1549 		 * We must ensure the chipset is stopped or the link is down
1550 		 * before cleaning the transmit packets.
1551 		 */
1552 		desc_num = 0;
1553 		for (j = 0; j < tx_ring->ring_size; j++) {
1554 			tcb = tx_ring->work_list[j];
1555 			if (tcb != NULL) {
1556 				desc_num += tcb->desc_num;
1557 
1558 				tx_ring->work_list[j] = NULL;
1559 
1560 				igb_free_tcb(tcb);
1561 
1562 				LIST_PUSH_TAIL(&pending_list, &tcb->link);
1563 			}
1564 		}
1565 
1566 		if (desc_num > 0) {
1567 			atomic_add_32(&tx_ring->tbd_free, desc_num);
1568 			ASSERT(tx_ring->tbd_free == tx_ring->ring_size);
1569 
1570 			/*
1571 			 * Reset the head and tail pointers of the tbd ring;
1572 			 * Reset the head write-back if it is enabled.
1573 			 */
1574 			tx_ring->tbd_head = 0;
1575 			tx_ring->tbd_tail = 0;
1576 			if (igb->tx_head_wb_enable)
1577 				*tx_ring->tbd_head_wb = 0;
1578 
1579 			E1000_WRITE_REG(&igb->hw, E1000_TDH(tx_ring->index), 0);
1580 			E1000_WRITE_REG(&igb->hw, E1000_TDT(tx_ring->index), 0);
1581 		}
1582 
1583 		mutex_exit(&tx_ring->recycle_lock);
1584 
1585 		/*
1586 		 * Add the tx control blocks in the pending list to
1587 		 * the free list.
1588 		 */
1589 		igb_put_free_list(tx_ring, &pending_list);
1590 	}
1591 }
1592 
1593 /*
1594  * igb_tx_drain - Drain the tx rings to allow pending packets to be transmitted
1595  */
1596 static boolean_t
1597 igb_tx_drain(igb_t *igb)
1598 {
1599 	igb_tx_ring_t *tx_ring;
1600 	boolean_t done;
1601 	int i, j;
1602 
1603 	/*
1604 	 * Wait for a specific time to allow pending tx packets
1605 	 * to be transmitted.
1606 	 *
1607 	 * Check the counter tbd_free to see if transmission is done.
1608 	 * No lock protection is needed here.
1609 	 *
1610 	 * Return B_TRUE if all pending packets have been transmitted;
1611 	 * Otherwise return B_FALSE;
1612 	 */
1613 	for (i = 0; i < TX_DRAIN_TIME; i++) {
1614 
1615 		done = B_TRUE;
1616 		for (j = 0; j < igb->num_tx_rings; j++) {
1617 			tx_ring = &igb->tx_rings[j];
1618 			done = done &&
1619 			    (tx_ring->tbd_free == tx_ring->ring_size);
1620 		}
1621 
1622 		if (done)
1623 			break;
1624 
1625 		msec_delay(1);
1626 	}
1627 
1628 	return (done);
1629 }
1630 
1631 /*
1632  * igb_rx_drain - Wait for all rx buffers to be released by upper layer
1633  */
1634 static boolean_t
1635 igb_rx_drain(igb_t *igb)
1636 {
1637 	igb_rx_ring_t *rx_ring;
1638 	boolean_t done;
1639 	int i, j;
1640 
1641 	/*
1642 	 * Polling the rx free list to check if those rx buffers held by
1643 	 * the upper layer are released.
1644 	 *
1645 	 * Check the counter rcb_free to see if all pending buffers are
1646 	 * released. No lock protection is needed here.
1647 	 *
1648 	 * Return B_TRUE if all pending buffers have been released;
1649 	 * Otherwise return B_FALSE;
1650 	 */
1651 	for (i = 0; i < RX_DRAIN_TIME; i++) {
1652 
1653 		done = B_TRUE;
1654 		for (j = 0; j < igb->num_rx_rings; j++) {
1655 			rx_ring = &igb->rx_rings[j];
1656 			done = done &&
1657 			    (rx_ring->rcb_free == rx_ring->free_list_size);
1658 		}
1659 
1660 		if (done)
1661 			break;
1662 
1663 		msec_delay(1);
1664 	}
1665 
1666 	return (done);
1667 }
1668 
1669 /*
1670  * igb_start - Start the driver/chipset
1671  */
1672 int
1673 igb_start(igb_t *igb)
1674 {
1675 	int i;
1676 
1677 	ASSERT(mutex_owned(&igb->gen_lock));
1678 
1679 	for (i = 0; i < igb->num_rx_rings; i++)
1680 		mutex_enter(&igb->rx_rings[i].rx_lock);
1681 	for (i = 0; i < igb->num_tx_rings; i++)
1682 		mutex_enter(&igb->tx_rings[i].tx_lock);
1683 
1684 	/*
1685 	 * Start the adapter
1686 	 */
1687 	if ((igb->attach_progress & ATTACH_PROGRESS_INIT_ADAPTER) == 0) {
1688 		if (igb_init_adapter(igb) != IGB_SUCCESS) {
1689 			igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1690 			goto start_failure;
1691 		}
1692 		igb->attach_progress |= ATTACH_PROGRESS_INIT_ADAPTER;
1693 
1694 		/*
1695 		 * Setup the rx/tx rings
1696 		 */
1697 		igb_setup_rings(igb);
1698 	}
1699 
1700 	/*
1701 	 * Enable adapter interrupts
1702 	 * The interrupts must be enabled after the driver state is START
1703 	 */
1704 	igb->capab->enable_intr(igb);
1705 
1706 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK)
1707 		goto start_failure;
1708 
1709 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1710 		goto start_failure;
1711 
1712 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1713 		mutex_exit(&igb->tx_rings[i].tx_lock);
1714 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1715 		mutex_exit(&igb->rx_rings[i].rx_lock);
1716 
1717 	return (IGB_SUCCESS);
1718 
1719 start_failure:
1720 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1721 		mutex_exit(&igb->tx_rings[i].tx_lock);
1722 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1723 		mutex_exit(&igb->rx_rings[i].rx_lock);
1724 
1725 	ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1726 
1727 	return (IGB_FAILURE);
1728 }
1729 
1730 /*
1731  * igb_stop - Stop the driver/chipset
1732  */
1733 void
1734 igb_stop(igb_t *igb)
1735 {
1736 	int i;
1737 
1738 	ASSERT(mutex_owned(&igb->gen_lock));
1739 
1740 	igb->attach_progress &= ~ATTACH_PROGRESS_INIT_ADAPTER;
1741 
1742 	/*
1743 	 * Disable the adapter interrupts
1744 	 */
1745 	igb_disable_adapter_interrupts(igb);
1746 
1747 	/*
1748 	 * Drain the pending tx packets
1749 	 */
1750 	(void) igb_tx_drain(igb);
1751 
1752 	for (i = 0; i < igb->num_rx_rings; i++)
1753 		mutex_enter(&igb->rx_rings[i].rx_lock);
1754 	for (i = 0; i < igb->num_tx_rings; i++)
1755 		mutex_enter(&igb->tx_rings[i].tx_lock);
1756 
1757 	/*
1758 	 * Stop the adapter
1759 	 */
1760 	igb_stop_adapter(igb);
1761 
1762 	/*
1763 	 * Clean the pending tx data/resources
1764 	 */
1765 	igb_tx_clean(igb);
1766 
1767 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1768 		mutex_exit(&igb->tx_rings[i].tx_lock);
1769 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1770 		mutex_exit(&igb->rx_rings[i].rx_lock);
1771 
1772 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1773 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1774 }
1775 
1776 /*
1777  * igb_alloc_rings - Allocate memory space for rx/tx rings
1778  */
1779 static int
1780 igb_alloc_rings(igb_t *igb)
1781 {
1782 	/*
1783 	 * Allocate memory space for rx rings
1784 	 */
1785 	igb->rx_rings = kmem_zalloc(
1786 	    sizeof (igb_rx_ring_t) * igb->num_rx_rings,
1787 	    KM_NOSLEEP);
1788 
1789 	if (igb->rx_rings == NULL) {
1790 		return (IGB_FAILURE);
1791 	}
1792 
1793 	/*
1794 	 * Allocate memory space for tx rings
1795 	 */
1796 	igb->tx_rings = kmem_zalloc(
1797 	    sizeof (igb_tx_ring_t) * igb->num_tx_rings,
1798 	    KM_NOSLEEP);
1799 
1800 	if (igb->tx_rings == NULL) {
1801 		kmem_free(igb->rx_rings,
1802 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1803 		igb->rx_rings = NULL;
1804 		return (IGB_FAILURE);
1805 	}
1806 
1807 	/*
1808 	 * Allocate memory space for rx ring groups
1809 	 */
1810 	igb->rx_groups = kmem_zalloc(
1811 	    sizeof (igb_rx_group_t) * igb->num_rx_groups,
1812 	    KM_NOSLEEP);
1813 
1814 	if (igb->rx_groups == NULL) {
1815 		kmem_free(igb->rx_rings,
1816 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1817 		kmem_free(igb->tx_rings,
1818 		    sizeof (igb_tx_ring_t) * igb->num_tx_rings);
1819 		igb->rx_rings = NULL;
1820 		igb->tx_rings = NULL;
1821 		return (IGB_FAILURE);
1822 	}
1823 
1824 	return (IGB_SUCCESS);
1825 }
1826 
1827 /*
1828  * igb_free_rings - Free the memory space of rx/tx rings.
1829  */
1830 static void
1831 igb_free_rings(igb_t *igb)
1832 {
1833 	if (igb->rx_rings != NULL) {
1834 		kmem_free(igb->rx_rings,
1835 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1836 		igb->rx_rings = NULL;
1837 	}
1838 
1839 	if (igb->tx_rings != NULL) {
1840 		kmem_free(igb->tx_rings,
1841 		    sizeof (igb_tx_ring_t) * igb->num_tx_rings);
1842 		igb->tx_rings = NULL;
1843 	}
1844 
1845 	if (igb->rx_groups != NULL) {
1846 		kmem_free(igb->rx_groups,
1847 		    sizeof (igb_rx_group_t) * igb->num_rx_groups);
1848 		igb->rx_groups = NULL;
1849 	}
1850 }
1851 
1852 /*
1853  * igb_setup_rings - Setup rx/tx rings
1854  */
1855 static void
1856 igb_setup_rings(igb_t *igb)
1857 {
1858 	/*
1859 	 * Setup the rx/tx rings, including the following:
1860 	 *
1861 	 * 1. Setup the descriptor ring and the control block buffers;
1862 	 * 2. Initialize necessary registers for receive/transmit;
1863 	 * 3. Initialize software pointers/parameters for receive/transmit;
1864 	 */
1865 	igb_setup_rx(igb);
1866 
1867 	igb_setup_tx(igb);
1868 }
1869 
1870 static void
1871 igb_setup_rx_ring(igb_rx_ring_t *rx_ring)
1872 {
1873 	igb_t *igb = rx_ring->igb;
1874 	struct e1000_hw *hw = &igb->hw;
1875 	rx_control_block_t *rcb;
1876 	union e1000_adv_rx_desc	*rbd;
1877 	uint32_t size;
1878 	uint32_t buf_low;
1879 	uint32_t buf_high;
1880 	uint32_t rxdctl;
1881 	int i;
1882 
1883 	ASSERT(mutex_owned(&rx_ring->rx_lock));
1884 	ASSERT(mutex_owned(&igb->gen_lock));
1885 
1886 	/*
1887 	 * Initialize descriptor ring with buffer addresses
1888 	 */
1889 	for (i = 0; i < igb->rx_ring_size; i++) {
1890 		rcb = rx_ring->work_list[i];
1891 		rbd = &rx_ring->rbd_ring[i];
1892 
1893 		rbd->read.pkt_addr = rcb->rx_buf.dma_address;
1894 		rbd->read.hdr_addr = NULL;
1895 	}
1896 
1897 	/*
1898 	 * Initialize the base address registers
1899 	 */
1900 	buf_low = (uint32_t)rx_ring->rbd_area.dma_address;
1901 	buf_high = (uint32_t)(rx_ring->rbd_area.dma_address >> 32);
1902 	E1000_WRITE_REG(hw, E1000_RDBAH(rx_ring->index), buf_high);
1903 	E1000_WRITE_REG(hw, E1000_RDBAL(rx_ring->index), buf_low);
1904 
1905 	/*
1906 	 * Initialize the length register
1907 	 */
1908 	size = rx_ring->ring_size * sizeof (union e1000_adv_rx_desc);
1909 	E1000_WRITE_REG(hw, E1000_RDLEN(rx_ring->index), size);
1910 
1911 	/*
1912 	 * Initialize buffer size & descriptor type
1913 	 */
1914 	E1000_WRITE_REG(hw, E1000_SRRCTL(rx_ring->index),
1915 	    ((igb->rx_buf_size >> E1000_SRRCTL_BSIZEPKT_SHIFT) |
1916 	    E1000_SRRCTL_DESCTYPE_ADV_ONEBUF));
1917 
1918 	/*
1919 	 * Setup the Receive Descriptor Control Register (RXDCTL)
1920 	 */
1921 	rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(rx_ring->index));
1922 	rxdctl &= igb->capab->rxdctl_mask;
1923 	rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1924 	rxdctl |= 16;		/* pthresh */
1925 	rxdctl |= 8 << 8;	/* hthresh */
1926 	rxdctl |= 1 << 16;	/* wthresh */
1927 	E1000_WRITE_REG(hw, E1000_RXDCTL(rx_ring->index), rxdctl);
1928 
1929 	rx_ring->rbd_next = 0;
1930 
1931 	/*
1932 	 * Note: Considering the case that the chipset is being reset
1933 	 * and there are still some buffers held by the upper layer,
1934 	 * we should not reset the values of rcb_head, rcb_tail and
1935 	 * rcb_free;
1936 	 */
1937 	if (igb->igb_state == IGB_UNKNOWN) {
1938 		rx_ring->rcb_head = 0;
1939 		rx_ring->rcb_tail = 0;
1940 		rx_ring->rcb_free = rx_ring->free_list_size;
1941 	}
1942 }
1943 
1944 static void
1945 igb_setup_rx(igb_t *igb)
1946 {
1947 	igb_rx_ring_t *rx_ring;
1948 	igb_rx_group_t *rx_group;
1949 	struct e1000_hw *hw = &igb->hw;
1950 	uint32_t rctl, rxcsum;
1951 	uint32_t ring_per_group;
1952 	int i;
1953 
1954 	/*
1955 	 * Setup the Receive Control Register (RCTL), and enable the
1956 	 * receiver. The initial configuration is to: enable the receiver,
1957 	 * accept broadcasts, discard bad packets, accept long packets,
1958 	 * disable VLAN filter checking, and set receive buffer size to
1959 	 * 2k.  For 82575, also set the receive descriptor minimum
1960 	 * threshold size to 1/2 the ring.
1961 	 */
1962 	rctl = E1000_READ_REG(hw, E1000_RCTL);
1963 
1964 	/*
1965 	 * Clear the field used for wakeup control.  This driver doesn't do
1966 	 * wakeup but leave this here for completeness.
1967 	 */
1968 	rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1969 	rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
1970 
1971 	rctl |= (E1000_RCTL_EN |	/* Enable Receive Unit */
1972 	    E1000_RCTL_BAM |		/* Accept Broadcast Packets */
1973 	    E1000_RCTL_LPE |		/* Large Packet Enable */
1974 					/* Multicast filter offset */
1975 	    (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT) |
1976 	    E1000_RCTL_RDMTS_HALF |	/* rx descriptor threshold */
1977 	    E1000_RCTL_SECRC);		/* Strip Ethernet CRC */
1978 
1979 	for (i = 0; i < igb->num_rx_groups; i++) {
1980 		rx_group = &igb->rx_groups[i];
1981 		rx_group->index = i;
1982 		rx_group->igb = igb;
1983 	}
1984 
1985 	/*
1986 	 * Set up all rx descriptor rings - must be called before receive unit
1987 	 * enabled.
1988 	 */
1989 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
1990 	for (i = 0; i < igb->num_rx_rings; i++) {
1991 		rx_ring = &igb->rx_rings[i];
1992 		igb_setup_rx_ring(rx_ring);
1993 
1994 		/*
1995 		 * Map a ring to a group by assigning a group index
1996 		 */
1997 		rx_ring->group_index = i / ring_per_group;
1998 	}
1999 
2000 	/*
2001 	 * Setup the Rx Long Packet Max Length register
2002 	 */
2003 	E1000_WRITE_REG(hw, E1000_RLPML, igb->max_frame_size);
2004 
2005 	/*
2006 	 * Hardware checksum settings
2007 	 */
2008 	if (igb->rx_hcksum_enable) {
2009 		rxcsum =
2010 		    E1000_RXCSUM_TUOFL |	/* TCP/UDP checksum */
2011 		    E1000_RXCSUM_IPOFL;		/* IP checksum */
2012 
2013 		E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2014 	}
2015 
2016 	/*
2017 	 * Setup classify and RSS for multiple receive queues
2018 	 */
2019 	switch (igb->vmdq_mode) {
2020 	case E1000_VMDQ_OFF:
2021 		/*
2022 		 * One ring group, only RSS is needed when more than
2023 		 * one ring enabled.
2024 		 */
2025 		if (igb->num_rx_rings > 1)
2026 			igb_setup_rss(igb);
2027 		break;
2028 	case E1000_VMDQ_MAC:
2029 		/*
2030 		 * Multiple groups, each group has one ring,
2031 		 * only the MAC classification is needed.
2032 		 */
2033 		igb_setup_mac_classify(igb);
2034 		break;
2035 	case E1000_VMDQ_MAC_RSS:
2036 		/*
2037 		 * Multiple groups and multiple rings, both
2038 		 * MAC classification and RSS are needed.
2039 		 */
2040 		igb_setup_mac_rss_classify(igb);
2041 		break;
2042 	}
2043 
2044 	/*
2045 	 * Enable the receive unit - must be done after all
2046 	 * the rx setup above.
2047 	 */
2048 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2049 
2050 	/*
2051 	 * Initialize all adapter ring head & tail pointers - must
2052 	 * be done after receive unit is enabled
2053 	 */
2054 	for (i = 0; i < igb->num_rx_rings; i++) {
2055 		rx_ring = &igb->rx_rings[i];
2056 		E1000_WRITE_REG(hw, E1000_RDH(i), 0);
2057 		E1000_WRITE_REG(hw, E1000_RDT(i), rx_ring->ring_size - 1);
2058 	}
2059 
2060 	/*
2061 	 * 82575 with manageability enabled needs a special flush to make
2062 	 * sure the fifos start clean.
2063 	 */
2064 	if ((hw->mac.type == e1000_82575) &&
2065 	    (E1000_READ_REG(hw, E1000_MANC) & E1000_MANC_RCV_TCO_EN)) {
2066 		e1000_rx_fifo_flush_82575(hw);
2067 	}
2068 }
2069 
2070 static void
2071 igb_setup_tx_ring(igb_tx_ring_t *tx_ring)
2072 {
2073 	igb_t *igb = tx_ring->igb;
2074 	struct e1000_hw *hw = &igb->hw;
2075 	uint32_t size;
2076 	uint32_t buf_low;
2077 	uint32_t buf_high;
2078 	uint32_t reg_val;
2079 
2080 	ASSERT(mutex_owned(&tx_ring->tx_lock));
2081 	ASSERT(mutex_owned(&igb->gen_lock));
2082 
2083 
2084 	/*
2085 	 * Initialize the length register
2086 	 */
2087 	size = tx_ring->ring_size * sizeof (union e1000_adv_tx_desc);
2088 	E1000_WRITE_REG(hw, E1000_TDLEN(tx_ring->index), size);
2089 
2090 	/*
2091 	 * Initialize the base address registers
2092 	 */
2093 	buf_low = (uint32_t)tx_ring->tbd_area.dma_address;
2094 	buf_high = (uint32_t)(tx_ring->tbd_area.dma_address >> 32);
2095 	E1000_WRITE_REG(hw, E1000_TDBAL(tx_ring->index), buf_low);
2096 	E1000_WRITE_REG(hw, E1000_TDBAH(tx_ring->index), buf_high);
2097 
2098 	/*
2099 	 * Setup head & tail pointers
2100 	 */
2101 	E1000_WRITE_REG(hw, E1000_TDH(tx_ring->index), 0);
2102 	E1000_WRITE_REG(hw, E1000_TDT(tx_ring->index), 0);
2103 
2104 	/*
2105 	 * Setup head write-back
2106 	 */
2107 	if (igb->tx_head_wb_enable) {
2108 		/*
2109 		 * The memory of the head write-back is allocated using
2110 		 * the extra tbd beyond the tail of the tbd ring.
2111 		 */
2112 		tx_ring->tbd_head_wb = (uint32_t *)
2113 		    ((uintptr_t)tx_ring->tbd_area.address + size);
2114 		*tx_ring->tbd_head_wb = 0;
2115 
2116 		buf_low = (uint32_t)
2117 		    (tx_ring->tbd_area.dma_address + size);
2118 		buf_high = (uint32_t)
2119 		    ((tx_ring->tbd_area.dma_address + size) >> 32);
2120 
2121 		/* Set the head write-back enable bit */
2122 		buf_low |= E1000_TX_HEAD_WB_ENABLE;
2123 
2124 		E1000_WRITE_REG(hw, E1000_TDWBAL(tx_ring->index), buf_low);
2125 		E1000_WRITE_REG(hw, E1000_TDWBAH(tx_ring->index), buf_high);
2126 
2127 		/*
2128 		 * Turn off relaxed ordering for head write back or it will
2129 		 * cause problems with the tx recycling
2130 		 */
2131 		reg_val = E1000_READ_REG(hw,
2132 		    E1000_DCA_TXCTRL(tx_ring->index));
2133 		reg_val &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
2134 		E1000_WRITE_REG(hw,
2135 		    E1000_DCA_TXCTRL(tx_ring->index), reg_val);
2136 	} else {
2137 		tx_ring->tbd_head_wb = NULL;
2138 	}
2139 
2140 	tx_ring->tbd_head = 0;
2141 	tx_ring->tbd_tail = 0;
2142 	tx_ring->tbd_free = tx_ring->ring_size;
2143 
2144 	/*
2145 	 * Note: for the case that the chipset is being reset, we should not
2146 	 * reset the values of tcb_head, tcb_tail. And considering there might
2147 	 * still be some packets kept in the pending_list, we should not assert
2148 	 * (tcb_free == free_list_size) here.
2149 	 */
2150 	if (igb->igb_state == IGB_UNKNOWN) {
2151 		tx_ring->tcb_head = 0;
2152 		tx_ring->tcb_tail = 0;
2153 		tx_ring->tcb_free = tx_ring->free_list_size;
2154 	}
2155 
2156 	/*
2157 	 * Enable TXDCTL per queue
2158 	 */
2159 	reg_val = E1000_READ_REG(hw, E1000_TXDCTL(tx_ring->index));
2160 	reg_val |= E1000_TXDCTL_QUEUE_ENABLE;
2161 	E1000_WRITE_REG(hw, E1000_TXDCTL(tx_ring->index), reg_val);
2162 
2163 	/*
2164 	 * Initialize hardware checksum offload settings
2165 	 */
2166 	bzero(&tx_ring->tx_context, sizeof (tx_context_t));
2167 }
2168 
2169 static void
2170 igb_setup_tx(igb_t *igb)
2171 {
2172 	igb_tx_ring_t *tx_ring;
2173 	struct e1000_hw *hw = &igb->hw;
2174 	uint32_t reg_val;
2175 	int i;
2176 
2177 	for (i = 0; i < igb->num_tx_rings; i++) {
2178 		tx_ring = &igb->tx_rings[i];
2179 		igb_setup_tx_ring(tx_ring);
2180 	}
2181 
2182 	/*
2183 	 * Setup the Transmit Control Register (TCTL)
2184 	 */
2185 	reg_val = E1000_READ_REG(hw, E1000_TCTL);
2186 	reg_val &= ~E1000_TCTL_CT;
2187 	reg_val |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2188 	    (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2189 
2190 	/* Enable transmits */
2191 	reg_val |= E1000_TCTL_EN;
2192 
2193 	E1000_WRITE_REG(hw, E1000_TCTL, reg_val);
2194 }
2195 
2196 /*
2197  * igb_setup_rss - Setup receive-side scaling feature
2198  */
2199 static void
2200 igb_setup_rss(igb_t *igb)
2201 {
2202 	struct e1000_hw *hw = &igb->hw;
2203 	uint32_t i, mrqc, rxcsum;
2204 	int shift = 0;
2205 	uint32_t random;
2206 	union e1000_reta {
2207 		uint32_t	dword;
2208 		uint8_t		bytes[4];
2209 	} reta;
2210 
2211 	/* Setup the Redirection Table */
2212 	if (hw->mac.type == e1000_82576) {
2213 		shift = 3;
2214 	} else if (hw->mac.type == e1000_82575) {
2215 		shift = 6;
2216 	}
2217 	for (i = 0; i < (32 * 4); i++) {
2218 		reta.bytes[i & 3] = (i % igb->num_rx_rings) << shift;
2219 		if ((i & 3) == 3) {
2220 			E1000_WRITE_REG(hw,
2221 			    (E1000_RETA(0) + (i & ~3)), reta.dword);
2222 		}
2223 	}
2224 
2225 	/* Fill out hash function seeds */
2226 	for (i = 0; i < 10; i++) {
2227 		(void) random_get_pseudo_bytes((uint8_t *)&random,
2228 		    sizeof (uint32_t));
2229 		E1000_WRITE_REG(hw, E1000_RSSRK(i), random);
2230 	}
2231 
2232 	/* Setup the Multiple Receive Queue Control register */
2233 	mrqc = E1000_MRQC_ENABLE_RSS_4Q;
2234 	mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2235 	    E1000_MRQC_RSS_FIELD_IPV4_TCP |
2236 	    E1000_MRQC_RSS_FIELD_IPV6 |
2237 	    E1000_MRQC_RSS_FIELD_IPV6_TCP |
2238 	    E1000_MRQC_RSS_FIELD_IPV4_UDP |
2239 	    E1000_MRQC_RSS_FIELD_IPV6_UDP |
2240 	    E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2241 	    E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2242 
2243 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2244 
2245 	/*
2246 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2247 	 *
2248 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2249 	 * checksum offloading provided by the 82575 chipset besides the IP
2250 	 * header checksum offloading and the TCP/UDP checksum offloading.
2251 	 * The Packet Checksum is by default computed over the entire packet
2252 	 * from the first byte of the DA through the last byte of the CRC,
2253 	 * including the Ethernet and IP headers.
2254 	 *
2255 	 * It is a hardware limitation that Packet Checksum is mutually
2256 	 * exclusive with RSS.
2257 	 */
2258 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2259 	rxcsum |= E1000_RXCSUM_PCSD;
2260 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2261 }
2262 
2263 /*
2264  * igb_setup_mac_rss_classify - Setup MAC classification and rss
2265  */
2266 static void
2267 igb_setup_mac_rss_classify(igb_t *igb)
2268 {
2269 	struct e1000_hw *hw = &igb->hw;
2270 	uint32_t i, mrqc, vmdctl, rxcsum;
2271 	uint32_t ring_per_group;
2272 	int shift_group0, shift_group1;
2273 	uint32_t random;
2274 	union e1000_reta {
2275 		uint32_t	dword;
2276 		uint8_t		bytes[4];
2277 	} reta;
2278 
2279 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
2280 
2281 	/* Setup the Redirection Table, it is shared between two groups */
2282 	shift_group0 = 2;
2283 	shift_group1 = 6;
2284 	for (i = 0; i < (32 * 4); i++) {
2285 		reta.bytes[i & 3] = ((i % ring_per_group) << shift_group0) |
2286 		    ((ring_per_group + (i % ring_per_group)) << shift_group1);
2287 		if ((i & 3) == 3) {
2288 			E1000_WRITE_REG(hw,
2289 			    (E1000_RETA(0) + (i & ~3)), reta.dword);
2290 		}
2291 	}
2292 
2293 	/* Fill out hash function seeds */
2294 	for (i = 0; i < 10; i++) {
2295 		(void) random_get_pseudo_bytes((uint8_t *)&random,
2296 		    sizeof (uint32_t));
2297 		E1000_WRITE_REG(hw, E1000_RSSRK(i), random);
2298 	}
2299 
2300 	/*
2301 	 * Setup the Multiple Receive Queue Control register,
2302 	 * enable VMDq based on packet destination MAC address and RSS.
2303 	 */
2304 	mrqc = E1000_MRQC_ENABLE_VMDQ_MAC_RSS_GROUP;
2305 	mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2306 	    E1000_MRQC_RSS_FIELD_IPV4_TCP |
2307 	    E1000_MRQC_RSS_FIELD_IPV6 |
2308 	    E1000_MRQC_RSS_FIELD_IPV6_TCP |
2309 	    E1000_MRQC_RSS_FIELD_IPV4_UDP |
2310 	    E1000_MRQC_RSS_FIELD_IPV6_UDP |
2311 	    E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2312 	    E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2313 
2314 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2315 
2316 
2317 	/* Define the default group and default queues */
2318 	vmdctl = E1000_VMDQ_MAC_GROUP_DEFAULT_QUEUE;
2319 	E1000_WRITE_REG(hw, E1000_VT_CTL, vmdctl);
2320 
2321 	/*
2322 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2323 	 *
2324 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2325 	 * checksum offloading provided by the 82575 chipset besides the IP
2326 	 * header checksum offloading and the TCP/UDP checksum offloading.
2327 	 * The Packet Checksum is by default computed over the entire packet
2328 	 * from the first byte of the DA through the last byte of the CRC,
2329 	 * including the Ethernet and IP headers.
2330 	 *
2331 	 * It is a hardware limitation that Packet Checksum is mutually
2332 	 * exclusive with RSS.
2333 	 */
2334 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2335 	rxcsum |= E1000_RXCSUM_PCSD;
2336 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2337 }
2338 
2339 /*
2340  * igb_setup_mac_classify - Setup MAC classification feature
2341  */
2342 static void
2343 igb_setup_mac_classify(igb_t *igb)
2344 {
2345 	struct e1000_hw *hw = &igb->hw;
2346 	uint32_t mrqc, rxcsum;
2347 
2348 	/*
2349 	 * Setup the Multiple Receive Queue Control register,
2350 	 * enable VMDq based on packet destination MAC address.
2351 	 */
2352 	mrqc = E1000_MRQC_ENABLE_VMDQ_MAC_GROUP;
2353 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2354 
2355 	/*
2356 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2357 	 *
2358 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2359 	 * checksum offloading provided by the 82575 chipset besides the IP
2360 	 * header checksum offloading and the TCP/UDP checksum offloading.
2361 	 * The Packet Checksum is by default computed over the entire packet
2362 	 * from the first byte of the DA through the last byte of the CRC,
2363 	 * including the Ethernet and IP headers.
2364 	 *
2365 	 * It is a hardware limitation that Packet Checksum is mutually
2366 	 * exclusive with RSS.
2367 	 */
2368 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2369 	rxcsum |= E1000_RXCSUM_PCSD;
2370 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2371 
2372 }
2373 
2374 /*
2375  * igb_init_unicst - Initialize the unicast addresses
2376  */
2377 static void
2378 igb_init_unicst(igb_t *igb)
2379 {
2380 	struct e1000_hw *hw = &igb->hw;
2381 	int slot;
2382 
2383 	/*
2384 	 * Here we should consider two situations:
2385 	 *
2386 	 * 1. Chipset is initialized the first time
2387 	 *    Initialize the multiple unicast addresses, and
2388 	 *    save the default MAC address.
2389 	 *
2390 	 * 2. Chipset is reset
2391 	 *    Recover the multiple unicast addresses from the
2392 	 *    software data structure to the RAR registers.
2393 	 */
2394 
2395 	/*
2396 	 * Clear the default MAC address in the RAR0 rgister,
2397 	 * which is loaded from EEPROM when system boot or chipreset,
2398 	 * this will cause the conficts with add_mac/rem_mac entry
2399 	 * points when VMDq is enabled. For this reason, the RAR0
2400 	 * must be cleared for both cases mentioned above.
2401 	 */
2402 	e1000_rar_clear(hw, 0);
2403 
2404 	if (!igb->unicst_init) {
2405 
2406 		/* Initialize the multiple unicast addresses */
2407 		igb->unicst_total = MAX_NUM_UNICAST_ADDRESSES;
2408 		igb->unicst_avail = igb->unicst_total;
2409 
2410 		for (slot = 0; slot < igb->unicst_total; slot++)
2411 			igb->unicst_addr[slot].mac.set = 0;
2412 
2413 		igb->unicst_init = B_TRUE;
2414 	} else {
2415 		/* Re-configure the RAR registers */
2416 		for (slot = 0; slot < igb->unicst_total; slot++) {
2417 			e1000_rar_set_vmdq(hw, igb->unicst_addr[slot].mac.addr,
2418 			    slot, igb->vmdq_mode,
2419 			    igb->unicst_addr[slot].mac.group_index);
2420 		}
2421 	}
2422 }
2423 
2424 /*
2425  * igb_unicst_find - Find the slot for the specified unicast address
2426  */
2427 int
2428 igb_unicst_find(igb_t *igb, const uint8_t *mac_addr)
2429 {
2430 	int slot;
2431 
2432 	ASSERT(mutex_owned(&igb->gen_lock));
2433 
2434 	for (slot = 0; slot < igb->unicst_total; slot++) {
2435 		if (bcmp(igb->unicst_addr[slot].mac.addr,
2436 		    mac_addr, ETHERADDRL) == 0)
2437 			return (slot);
2438 	}
2439 
2440 	return (-1);
2441 }
2442 
2443 /*
2444  * igb_unicst_set - Set the unicast address to the specified slot
2445  */
2446 int
2447 igb_unicst_set(igb_t *igb, const uint8_t *mac_addr,
2448     int slot)
2449 {
2450 	struct e1000_hw *hw = &igb->hw;
2451 
2452 	ASSERT(mutex_owned(&igb->gen_lock));
2453 
2454 	/*
2455 	 * Save the unicast address in the software data structure
2456 	 */
2457 	bcopy(mac_addr, igb->unicst_addr[slot].mac.addr, ETHERADDRL);
2458 
2459 	/*
2460 	 * Set the unicast address to the RAR register
2461 	 */
2462 	e1000_rar_set(hw, (uint8_t *)mac_addr, slot);
2463 
2464 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2465 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2466 		return (EIO);
2467 	}
2468 
2469 	return (0);
2470 }
2471 
2472 /*
2473  * igb_multicst_add - Add a multicst address
2474  */
2475 int
2476 igb_multicst_add(igb_t *igb, const uint8_t *multiaddr)
2477 {
2478 	struct ether_addr *new_table;
2479 	size_t new_len;
2480 	size_t old_len;
2481 
2482 	ASSERT(mutex_owned(&igb->gen_lock));
2483 
2484 	if ((multiaddr[0] & 01) == 0) {
2485 		igb_error(igb, "Illegal multicast address");
2486 		return (EINVAL);
2487 	}
2488 
2489 	if (igb->mcast_count >= igb->mcast_max_num) {
2490 		igb_error(igb, "Adapter requested more than %d mcast addresses",
2491 		    igb->mcast_max_num);
2492 		return (ENOENT);
2493 	}
2494 
2495 	if (igb->mcast_count == igb->mcast_alloc_count) {
2496 		old_len = igb->mcast_alloc_count *
2497 		    sizeof (struct ether_addr);
2498 		new_len = (igb->mcast_alloc_count + MCAST_ALLOC_COUNT) *
2499 		    sizeof (struct ether_addr);
2500 
2501 		new_table = kmem_alloc(new_len, KM_NOSLEEP);
2502 		if (new_table == NULL) {
2503 			igb_error(igb,
2504 			    "Not enough memory to alloc mcast table");
2505 			return (ENOMEM);
2506 		}
2507 
2508 		if (igb->mcast_table != NULL) {
2509 			bcopy(igb->mcast_table, new_table, old_len);
2510 			kmem_free(igb->mcast_table, old_len);
2511 		}
2512 		igb->mcast_alloc_count += MCAST_ALLOC_COUNT;
2513 		igb->mcast_table = new_table;
2514 	}
2515 
2516 	bcopy(multiaddr,
2517 	    &igb->mcast_table[igb->mcast_count], ETHERADDRL);
2518 	igb->mcast_count++;
2519 
2520 	/*
2521 	 * Update the multicast table in the hardware
2522 	 */
2523 	igb_setup_multicst(igb);
2524 
2525 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2526 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2527 		return (EIO);
2528 	}
2529 
2530 	return (0);
2531 }
2532 
2533 /*
2534  * igb_multicst_remove - Remove a multicst address
2535  */
2536 int
2537 igb_multicst_remove(igb_t *igb, const uint8_t *multiaddr)
2538 {
2539 	struct ether_addr *new_table;
2540 	size_t new_len;
2541 	size_t old_len;
2542 	int i;
2543 
2544 	ASSERT(mutex_owned(&igb->gen_lock));
2545 
2546 	for (i = 0; i < igb->mcast_count; i++) {
2547 		if (bcmp(multiaddr, &igb->mcast_table[i],
2548 		    ETHERADDRL) == 0) {
2549 			for (i++; i < igb->mcast_count; i++) {
2550 				igb->mcast_table[i - 1] =
2551 				    igb->mcast_table[i];
2552 			}
2553 			igb->mcast_count--;
2554 			break;
2555 		}
2556 	}
2557 
2558 	if ((igb->mcast_alloc_count - igb->mcast_count) >
2559 	    MCAST_ALLOC_COUNT) {
2560 		old_len = igb->mcast_alloc_count *
2561 		    sizeof (struct ether_addr);
2562 		new_len = (igb->mcast_alloc_count - MCAST_ALLOC_COUNT) *
2563 		    sizeof (struct ether_addr);
2564 
2565 		new_table = kmem_alloc(new_len, KM_NOSLEEP);
2566 		if (new_table != NULL) {
2567 			bcopy(igb->mcast_table, new_table, new_len);
2568 			kmem_free(igb->mcast_table, old_len);
2569 			igb->mcast_alloc_count -= MCAST_ALLOC_COUNT;
2570 			igb->mcast_table = new_table;
2571 		}
2572 	}
2573 
2574 	/*
2575 	 * Update the multicast table in the hardware
2576 	 */
2577 	igb_setup_multicst(igb);
2578 
2579 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2580 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2581 		return (EIO);
2582 	}
2583 
2584 	return (0);
2585 }
2586 
2587 static void
2588 igb_release_multicast(igb_t *igb)
2589 {
2590 	if (igb->mcast_table != NULL) {
2591 		kmem_free(igb->mcast_table,
2592 		    igb->mcast_alloc_count * sizeof (struct ether_addr));
2593 		igb->mcast_table = NULL;
2594 	}
2595 }
2596 
2597 /*
2598  * igb_setup_multicast - setup multicast data structures
2599  *
2600  * This routine initializes all of the multicast related structures
2601  * and save them in the hardware registers.
2602  */
2603 static void
2604 igb_setup_multicst(igb_t *igb)
2605 {
2606 	uint8_t *mc_addr_list;
2607 	uint32_t mc_addr_count;
2608 	struct e1000_hw *hw = &igb->hw;
2609 
2610 	ASSERT(mutex_owned(&igb->gen_lock));
2611 	ASSERT(igb->mcast_count <= igb->mcast_max_num);
2612 
2613 	mc_addr_list = (uint8_t *)igb->mcast_table;
2614 	mc_addr_count = igb->mcast_count;
2615 
2616 	/*
2617 	 * Update the multicase addresses to the MTA registers
2618 	 */
2619 	e1000_update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
2620 }
2621 
2622 /*
2623  * igb_get_conf - Get driver configurations set in driver.conf
2624  *
2625  * This routine gets user-configured values out of the configuration
2626  * file igb.conf.
2627  *
2628  * For each configurable value, there is a minimum, a maximum, and a
2629  * default.
2630  * If user does not configure a value, use the default.
2631  * If user configures below the minimum, use the minumum.
2632  * If user configures above the maximum, use the maxumum.
2633  */
2634 static void
2635 igb_get_conf(igb_t *igb)
2636 {
2637 	struct e1000_hw *hw = &igb->hw;
2638 	uint32_t default_mtu;
2639 	uint32_t flow_control;
2640 	uint32_t ring_per_group;
2641 	int i;
2642 
2643 	/*
2644 	 * igb driver supports the following user configurations:
2645 	 *
2646 	 * Link configurations:
2647 	 *    adv_autoneg_cap
2648 	 *    adv_1000fdx_cap
2649 	 *    adv_100fdx_cap
2650 	 *    adv_100hdx_cap
2651 	 *    adv_10fdx_cap
2652 	 *    adv_10hdx_cap
2653 	 * Note: 1000hdx is not supported.
2654 	 *
2655 	 * Jumbo frame configuration:
2656 	 *    default_mtu
2657 	 *
2658 	 * Ethernet flow control configuration:
2659 	 *    flow_control
2660 	 *
2661 	 * Multiple rings configurations:
2662 	 *    tx_queue_number
2663 	 *    tx_ring_size
2664 	 *    rx_queue_number
2665 	 *    rx_ring_size
2666 	 *
2667 	 * Call igb_get_prop() to get the value for a specific
2668 	 * configuration parameter.
2669 	 */
2670 
2671 	/*
2672 	 * Link configurations
2673 	 */
2674 	igb->param_adv_autoneg_cap = igb_get_prop(igb,
2675 	    PROP_ADV_AUTONEG_CAP, 0, 1, 1);
2676 	igb->param_adv_1000fdx_cap = igb_get_prop(igb,
2677 	    PROP_ADV_1000FDX_CAP, 0, 1, 1);
2678 	igb->param_adv_100fdx_cap = igb_get_prop(igb,
2679 	    PROP_ADV_100FDX_CAP, 0, 1, 1);
2680 	igb->param_adv_100hdx_cap = igb_get_prop(igb,
2681 	    PROP_ADV_100HDX_CAP, 0, 1, 1);
2682 	igb->param_adv_10fdx_cap = igb_get_prop(igb,
2683 	    PROP_ADV_10FDX_CAP, 0, 1, 1);
2684 	igb->param_adv_10hdx_cap = igb_get_prop(igb,
2685 	    PROP_ADV_10HDX_CAP, 0, 1, 1);
2686 
2687 	/*
2688 	 * Jumbo frame configurations
2689 	 */
2690 	default_mtu = igb_get_prop(igb, PROP_DEFAULT_MTU,
2691 	    MIN_MTU, MAX_MTU, DEFAULT_MTU);
2692 
2693 	igb->max_frame_size = default_mtu +
2694 	    sizeof (struct ether_vlan_header) + ETHERFCSL;
2695 
2696 	/*
2697 	 * Ethernet flow control configuration
2698 	 */
2699 	flow_control = igb_get_prop(igb, PROP_FLOW_CONTROL,
2700 	    e1000_fc_none, 4, e1000_fc_full);
2701 	if (flow_control == 4)
2702 		flow_control = e1000_fc_default;
2703 
2704 	hw->fc.requested_mode = flow_control;
2705 
2706 	/*
2707 	 * Multiple rings configurations
2708 	 */
2709 	igb->tx_ring_size = igb_get_prop(igb, PROP_TX_RING_SIZE,
2710 	    MIN_TX_RING_SIZE, MAX_TX_RING_SIZE, DEFAULT_TX_RING_SIZE);
2711 	igb->rx_ring_size = igb_get_prop(igb, PROP_RX_RING_SIZE,
2712 	    MIN_RX_RING_SIZE, MAX_RX_RING_SIZE, DEFAULT_RX_RING_SIZE);
2713 
2714 	igb->mr_enable = igb_get_prop(igb, PROP_MR_ENABLE, 0, 1, 0);
2715 	igb->num_rx_groups = igb_get_prop(igb, PROP_RX_GROUP_NUM,
2716 	    MIN_RX_GROUP_NUM, MAX_RX_GROUP_NUM, DEFAULT_RX_GROUP_NUM);
2717 	/*
2718 	 * Currently we do not support VMDq for 82576 and 82580.
2719 	 * If it is e1000_82576, set num_rx_groups to 1.
2720 	 */
2721 	if (hw->mac.type >= e1000_82576)
2722 		igb->num_rx_groups = 1;
2723 
2724 	if (igb->mr_enable) {
2725 		igb->num_tx_rings = igb->capab->def_tx_que_num;
2726 		igb->num_rx_rings = igb->capab->def_rx_que_num;
2727 	} else {
2728 		igb->num_tx_rings = 1;
2729 		igb->num_rx_rings = 1;
2730 
2731 		if (igb->num_rx_groups > 1) {
2732 			igb_error(igb,
2733 			    "Invalid rx groups number. Please enable multiple "
2734 			    "rings first");
2735 			igb->num_rx_groups = 1;
2736 		}
2737 	}
2738 
2739 	/*
2740 	 * Check the divisibility between rx rings and rx groups.
2741 	 */
2742 	for (i = igb->num_rx_groups; i > 0; i--) {
2743 		if ((igb->num_rx_rings % i) == 0)
2744 			break;
2745 	}
2746 	if (i != igb->num_rx_groups) {
2747 		igb_error(igb,
2748 		    "Invalid rx groups number. Downgrade the rx group "
2749 		    "number to %d.", i);
2750 		igb->num_rx_groups = i;
2751 	}
2752 
2753 	/*
2754 	 * Get the ring number per group.
2755 	 */
2756 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
2757 
2758 	if (igb->num_rx_groups == 1) {
2759 		/*
2760 		 * One rx ring group, the rx ring number is num_rx_rings.
2761 		 */
2762 		igb->vmdq_mode = E1000_VMDQ_OFF;
2763 	} else if (ring_per_group == 1) {
2764 		/*
2765 		 * Multiple rx groups, each group has one rx ring.
2766 		 */
2767 		igb->vmdq_mode = E1000_VMDQ_MAC;
2768 	} else {
2769 		/*
2770 		 * Multiple groups and multiple rings.
2771 		 */
2772 		igb->vmdq_mode = E1000_VMDQ_MAC_RSS;
2773 	}
2774 
2775 	/*
2776 	 * Tunable used to force an interrupt type. The only use is
2777 	 * for testing of the lesser interrupt types.
2778 	 * 0 = don't force interrupt type
2779 	 * 1 = force interrupt type MSIX
2780 	 * 2 = force interrupt type MSI
2781 	 * 3 = force interrupt type Legacy
2782 	 */
2783 	igb->intr_force = igb_get_prop(igb, PROP_INTR_FORCE,
2784 	    IGB_INTR_NONE, IGB_INTR_LEGACY, IGB_INTR_NONE);
2785 
2786 	igb->tx_hcksum_enable = igb_get_prop(igb, PROP_TX_HCKSUM_ENABLE,
2787 	    0, 1, 1);
2788 	igb->rx_hcksum_enable = igb_get_prop(igb, PROP_RX_HCKSUM_ENABLE,
2789 	    0, 1, 1);
2790 	igb->lso_enable = igb_get_prop(igb, PROP_LSO_ENABLE,
2791 	    0, 1, 1);
2792 	igb->tx_head_wb_enable = igb_get_prop(igb, PROP_TX_HEAD_WB_ENABLE,
2793 	    0, 1, 1);
2794 
2795 	/*
2796 	 * igb LSO needs the tx h/w checksum support.
2797 	 * Here LSO will be disabled if tx h/w checksum has been disabled.
2798 	 */
2799 	if (igb->tx_hcksum_enable == B_FALSE)
2800 		igb->lso_enable = B_FALSE;
2801 
2802 	igb->tx_copy_thresh = igb_get_prop(igb, PROP_TX_COPY_THRESHOLD,
2803 	    MIN_TX_COPY_THRESHOLD, MAX_TX_COPY_THRESHOLD,
2804 	    DEFAULT_TX_COPY_THRESHOLD);
2805 	igb->tx_recycle_thresh = igb_get_prop(igb, PROP_TX_RECYCLE_THRESHOLD,
2806 	    MIN_TX_RECYCLE_THRESHOLD, MAX_TX_RECYCLE_THRESHOLD,
2807 	    DEFAULT_TX_RECYCLE_THRESHOLD);
2808 	igb->tx_overload_thresh = igb_get_prop(igb, PROP_TX_OVERLOAD_THRESHOLD,
2809 	    MIN_TX_OVERLOAD_THRESHOLD, MAX_TX_OVERLOAD_THRESHOLD,
2810 	    DEFAULT_TX_OVERLOAD_THRESHOLD);
2811 	igb->tx_resched_thresh = igb_get_prop(igb, PROP_TX_RESCHED_THRESHOLD,
2812 	    MIN_TX_RESCHED_THRESHOLD, MAX_TX_RESCHED_THRESHOLD,
2813 	    DEFAULT_TX_RESCHED_THRESHOLD);
2814 
2815 	igb->rx_copy_thresh = igb_get_prop(igb, PROP_RX_COPY_THRESHOLD,
2816 	    MIN_RX_COPY_THRESHOLD, MAX_RX_COPY_THRESHOLD,
2817 	    DEFAULT_RX_COPY_THRESHOLD);
2818 	igb->rx_limit_per_intr = igb_get_prop(igb, PROP_RX_LIMIT_PER_INTR,
2819 	    MIN_RX_LIMIT_PER_INTR, MAX_RX_LIMIT_PER_INTR,
2820 	    DEFAULT_RX_LIMIT_PER_INTR);
2821 
2822 	igb->intr_throttling[0] = igb_get_prop(igb, PROP_INTR_THROTTLING,
2823 	    igb->capab->min_intr_throttle,
2824 	    igb->capab->max_intr_throttle,
2825 	    igb->capab->def_intr_throttle);
2826 
2827 	/*
2828 	 * Max number of multicast addresses
2829 	 */
2830 	igb->mcast_max_num =
2831 	    igb_get_prop(igb, PROP_MCAST_MAX_NUM,
2832 	    MIN_MCAST_NUM, MAX_MCAST_NUM, DEFAULT_MCAST_NUM);
2833 }
2834 
2835 /*
2836  * igb_get_prop - Get a property value out of the configuration file igb.conf
2837  *
2838  * Caller provides the name of the property, a default value, a minimum
2839  * value, and a maximum value.
2840  *
2841  * Return configured value of the property, with default, minimum and
2842  * maximum properly applied.
2843  */
2844 static int
2845 igb_get_prop(igb_t *igb,
2846     char *propname,	/* name of the property */
2847     int minval,		/* minimum acceptable value */
2848     int maxval,		/* maximim acceptable value */
2849     int defval)		/* default value */
2850 {
2851 	int value;
2852 
2853 	/*
2854 	 * Call ddi_prop_get_int() to read the conf settings
2855 	 */
2856 	value = ddi_prop_get_int(DDI_DEV_T_ANY, igb->dip,
2857 	    DDI_PROP_DONTPASS, propname, defval);
2858 
2859 	if (value > maxval)
2860 		value = maxval;
2861 
2862 	if (value < minval)
2863 		value = minval;
2864 
2865 	return (value);
2866 }
2867 
2868 /*
2869  * igb_setup_link - Using the link properties to setup the link
2870  */
2871 int
2872 igb_setup_link(igb_t *igb, boolean_t setup_hw)
2873 {
2874 	struct e1000_mac_info *mac;
2875 	struct e1000_phy_info *phy;
2876 	boolean_t invalid;
2877 
2878 	mac = &igb->hw.mac;
2879 	phy = &igb->hw.phy;
2880 	invalid = B_FALSE;
2881 
2882 	if (igb->param_adv_autoneg_cap == 1) {
2883 		mac->autoneg = B_TRUE;
2884 		phy->autoneg_advertised = 0;
2885 
2886 		/*
2887 		 * 1000hdx is not supported for autonegotiation
2888 		 */
2889 		if (igb->param_adv_1000fdx_cap == 1)
2890 			phy->autoneg_advertised |= ADVERTISE_1000_FULL;
2891 
2892 		if (igb->param_adv_100fdx_cap == 1)
2893 			phy->autoneg_advertised |= ADVERTISE_100_FULL;
2894 
2895 		if (igb->param_adv_100hdx_cap == 1)
2896 			phy->autoneg_advertised |= ADVERTISE_100_HALF;
2897 
2898 		if (igb->param_adv_10fdx_cap == 1)
2899 			phy->autoneg_advertised |= ADVERTISE_10_FULL;
2900 
2901 		if (igb->param_adv_10hdx_cap == 1)
2902 			phy->autoneg_advertised |= ADVERTISE_10_HALF;
2903 
2904 		if (phy->autoneg_advertised == 0)
2905 			invalid = B_TRUE;
2906 	} else {
2907 		mac->autoneg = B_FALSE;
2908 
2909 		/*
2910 		 * 1000fdx and 1000hdx are not supported for forced link
2911 		 */
2912 		if (igb->param_adv_100fdx_cap == 1)
2913 			mac->forced_speed_duplex = ADVERTISE_100_FULL;
2914 		else if (igb->param_adv_100hdx_cap == 1)
2915 			mac->forced_speed_duplex = ADVERTISE_100_HALF;
2916 		else if (igb->param_adv_10fdx_cap == 1)
2917 			mac->forced_speed_duplex = ADVERTISE_10_FULL;
2918 		else if (igb->param_adv_10hdx_cap == 1)
2919 			mac->forced_speed_duplex = ADVERTISE_10_HALF;
2920 		else
2921 			invalid = B_TRUE;
2922 	}
2923 
2924 	if (invalid) {
2925 		igb_notice(igb, "Invalid link settings. Setup link to "
2926 		    "autonegotiation with full link capabilities.");
2927 		mac->autoneg = B_TRUE;
2928 		phy->autoneg_advertised = ADVERTISE_1000_FULL |
2929 		    ADVERTISE_100_FULL | ADVERTISE_100_HALF |
2930 		    ADVERTISE_10_FULL | ADVERTISE_10_HALF;
2931 	}
2932 
2933 	if (setup_hw) {
2934 		if (e1000_setup_link(&igb->hw) != E1000_SUCCESS)
2935 			return (IGB_FAILURE);
2936 	}
2937 
2938 	return (IGB_SUCCESS);
2939 }
2940 
2941 
2942 /*
2943  * igb_is_link_up - Check if the link is up
2944  */
2945 static boolean_t
2946 igb_is_link_up(igb_t *igb)
2947 {
2948 	struct e1000_hw *hw = &igb->hw;
2949 	boolean_t link_up = B_FALSE;
2950 
2951 	ASSERT(mutex_owned(&igb->gen_lock));
2952 
2953 	/*
2954 	 * get_link_status is set in the interrupt handler on link-status-change
2955 	 * or rx sequence error interrupt.  get_link_status will stay
2956 	 * false until the e1000_check_for_link establishes link only
2957 	 * for copper adapters.
2958 	 */
2959 	switch (hw->phy.media_type) {
2960 	case e1000_media_type_copper:
2961 		if (hw->mac.get_link_status) {
2962 			(void) e1000_check_for_link(hw);
2963 			link_up = !hw->mac.get_link_status;
2964 		} else {
2965 			link_up = B_TRUE;
2966 		}
2967 		break;
2968 	case e1000_media_type_fiber:
2969 		(void) e1000_check_for_link(hw);
2970 		link_up = (E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU);
2971 		break;
2972 	case e1000_media_type_internal_serdes:
2973 		(void) e1000_check_for_link(hw);
2974 		link_up = hw->mac.serdes_has_link;
2975 		break;
2976 	}
2977 
2978 	return (link_up);
2979 }
2980 
2981 /*
2982  * igb_link_check - Link status processing
2983  */
2984 static boolean_t
2985 igb_link_check(igb_t *igb)
2986 {
2987 	struct e1000_hw *hw = &igb->hw;
2988 	uint16_t speed = 0, duplex = 0;
2989 	boolean_t link_changed = B_FALSE;
2990 
2991 	ASSERT(mutex_owned(&igb->gen_lock));
2992 
2993 	if (igb_is_link_up(igb)) {
2994 		/*
2995 		 * The Link is up, check whether it was marked as down earlier
2996 		 */
2997 		if (igb->link_state != LINK_STATE_UP) {
2998 			(void) e1000_get_speed_and_duplex(hw, &speed, &duplex);
2999 			igb->link_speed = speed;
3000 			igb->link_duplex = duplex;
3001 			igb->link_state = LINK_STATE_UP;
3002 			igb->link_down_timeout = 0;
3003 			link_changed = B_TRUE;
3004 			if (!igb->link_complete)
3005 				igb_stop_link_timer(igb);
3006 		}
3007 	} else if (igb->link_complete) {
3008 		if (igb->link_state != LINK_STATE_DOWN) {
3009 			igb->link_speed = 0;
3010 			igb->link_duplex = 0;
3011 			igb->link_state = LINK_STATE_DOWN;
3012 			link_changed = B_TRUE;
3013 		}
3014 
3015 		if (igb->igb_state & IGB_STARTED) {
3016 			if (igb->link_down_timeout < MAX_LINK_DOWN_TIMEOUT) {
3017 				igb->link_down_timeout++;
3018 			} else if (igb->link_down_timeout ==
3019 			    MAX_LINK_DOWN_TIMEOUT) {
3020 				igb_tx_clean(igb);
3021 				igb->link_down_timeout++;
3022 			}
3023 		}
3024 	}
3025 
3026 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
3027 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
3028 		return (B_FALSE);
3029 	}
3030 
3031 	return (link_changed);
3032 }
3033 
3034 /*
3035  * igb_local_timer - driver watchdog function
3036  *
3037  * This function will handle the hardware stall check, link status
3038  * check and other routines.
3039  */
3040 static void
3041 igb_local_timer(void *arg)
3042 {
3043 	igb_t *igb = (igb_t *)arg;
3044 	boolean_t link_changed = B_FALSE;
3045 
3046 	if (igb->igb_state & IGB_ERROR) {
3047 		igb->reset_count++;
3048 		if (igb_reset(igb) == IGB_SUCCESS)
3049 			ddi_fm_service_impact(igb->dip, DDI_SERVICE_RESTORED);
3050 
3051 		igb_restart_watchdog_timer(igb);
3052 		return;
3053 	}
3054 
3055 	if (igb_stall_check(igb) || (igb->igb_state & IGB_STALL)) {
3056 		igb_fm_ereport(igb, DDI_FM_DEVICE_STALL);
3057 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
3058 		igb->reset_count++;
3059 		if (igb_reset(igb) == IGB_SUCCESS)
3060 			ddi_fm_service_impact(igb->dip, DDI_SERVICE_RESTORED);
3061 
3062 		igb_restart_watchdog_timer(igb);
3063 		return;
3064 	}
3065 
3066 	mutex_enter(&igb->gen_lock);
3067 	if (!(igb->igb_state & IGB_SUSPENDED) && (igb->igb_state & IGB_STARTED))
3068 		link_changed = igb_link_check(igb);
3069 	mutex_exit(&igb->gen_lock);
3070 
3071 	if (link_changed)
3072 		mac_link_update(igb->mac_hdl, igb->link_state);
3073 
3074 	igb_restart_watchdog_timer(igb);
3075 }
3076 
3077 /*
3078  * igb_link_timer - link setup timer function
3079  *
3080  * It is called when the timer for link setup is expired, which indicates
3081  * the completion of the link setup. The link state will not be updated
3082  * until the link setup is completed. And the link state will not be sent
3083  * to the upper layer through mac_link_update() in this function. It will
3084  * be updated in the local timer routine or the interrupts service routine
3085  * after the interface is started (plumbed).
3086  */
3087 static void
3088 igb_link_timer(void *arg)
3089 {
3090 	igb_t *igb = (igb_t *)arg;
3091 
3092 	mutex_enter(&igb->link_lock);
3093 	igb->link_complete = B_TRUE;
3094 	igb->link_tid = 0;
3095 	mutex_exit(&igb->link_lock);
3096 }
3097 /*
3098  * igb_stall_check - check for transmit stall
3099  *
3100  * This function checks if the adapter is stalled (in transmit).
3101  *
3102  * It is called each time the watchdog timeout is invoked.
3103  * If the transmit descriptor reclaim continuously fails,
3104  * the watchdog value will increment by 1. If the watchdog
3105  * value exceeds the threshold, the igb is assumed to
3106  * have stalled and need to be reset.
3107  */
3108 static boolean_t
3109 igb_stall_check(igb_t *igb)
3110 {
3111 	igb_tx_ring_t *tx_ring;
3112 	struct e1000_hw *hw = &igb->hw;
3113 	boolean_t result;
3114 	int i;
3115 
3116 	if (igb->link_state != LINK_STATE_UP)
3117 		return (B_FALSE);
3118 
3119 	/*
3120 	 * If any tx ring is stalled, we'll reset the chipset
3121 	 */
3122 	result = B_FALSE;
3123 	for (i = 0; i < igb->num_tx_rings; i++) {
3124 		tx_ring = &igb->tx_rings[i];
3125 
3126 		if (tx_ring->recycle_fail > 0)
3127 			tx_ring->stall_watchdog++;
3128 		else
3129 			tx_ring->stall_watchdog = 0;
3130 
3131 		if (tx_ring->stall_watchdog >= STALL_WATCHDOG_TIMEOUT) {
3132 			result = B_TRUE;
3133 			if (hw->mac.type == e1000_82580) {
3134 				hw->dev_spec._82575.global_device_reset
3135 				    = B_TRUE;
3136 			}
3137 			break;
3138 		}
3139 	}
3140 
3141 	if (result) {
3142 		tx_ring->stall_watchdog = 0;
3143 		tx_ring->recycle_fail = 0;
3144 	}
3145 
3146 	return (result);
3147 }
3148 
3149 
3150 /*
3151  * is_valid_mac_addr - Check if the mac address is valid
3152  */
3153 static boolean_t
3154 is_valid_mac_addr(uint8_t *mac_addr)
3155 {
3156 	const uint8_t addr_test1[6] = { 0, 0, 0, 0, 0, 0 };
3157 	const uint8_t addr_test2[6] =
3158 	    { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF };
3159 
3160 	if (!(bcmp(addr_test1, mac_addr, ETHERADDRL)) ||
3161 	    !(bcmp(addr_test2, mac_addr, ETHERADDRL)))
3162 		return (B_FALSE);
3163 
3164 	return (B_TRUE);
3165 }
3166 
3167 static boolean_t
3168 igb_find_mac_address(igb_t *igb)
3169 {
3170 	struct e1000_hw *hw = &igb->hw;
3171 #ifdef __sparc
3172 	uchar_t *bytes;
3173 	struct ether_addr sysaddr;
3174 	uint_t nelts;
3175 	int err;
3176 	boolean_t found = B_FALSE;
3177 
3178 	/*
3179 	 * The "vendor's factory-set address" may already have
3180 	 * been extracted from the chip, but if the property
3181 	 * "local-mac-address" is set we use that instead.
3182 	 *
3183 	 * We check whether it looks like an array of 6
3184 	 * bytes (which it should, if OBP set it).  If we can't
3185 	 * make sense of it this way, we'll ignore it.
3186 	 */
3187 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip,
3188 	    DDI_PROP_DONTPASS, "local-mac-address", &bytes, &nelts);
3189 	if (err == DDI_PROP_SUCCESS) {
3190 		if (nelts == ETHERADDRL) {
3191 			while (nelts--)
3192 				hw->mac.addr[nelts] = bytes[nelts];
3193 			found = B_TRUE;
3194 		}
3195 		ddi_prop_free(bytes);
3196 	}
3197 
3198 	/*
3199 	 * Look up the OBP property "local-mac-address?". If the user has set
3200 	 * 'local-mac-address? = false', use "the system address" instead.
3201 	 */
3202 	if (ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip, 0,
3203 	    "local-mac-address?", &bytes, &nelts) == DDI_PROP_SUCCESS) {
3204 		if (strncmp("false", (caddr_t)bytes, (size_t)nelts) == 0) {
3205 			if (localetheraddr(NULL, &sysaddr) != 0) {
3206 				bcopy(&sysaddr, hw->mac.addr, ETHERADDRL);
3207 				found = B_TRUE;
3208 			}
3209 		}
3210 		ddi_prop_free(bytes);
3211 	}
3212 
3213 	/*
3214 	 * Finally(!), if there's a valid "mac-address" property (created
3215 	 * if we netbooted from this interface), we must use this instead
3216 	 * of any of the above to ensure that the NFS/install server doesn't
3217 	 * get confused by the address changing as Solaris takes over!
3218 	 */
3219 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip,
3220 	    DDI_PROP_DONTPASS, "mac-address", &bytes, &nelts);
3221 	if (err == DDI_PROP_SUCCESS) {
3222 		if (nelts == ETHERADDRL) {
3223 			while (nelts--)
3224 				hw->mac.addr[nelts] = bytes[nelts];
3225 			found = B_TRUE;
3226 		}
3227 		ddi_prop_free(bytes);
3228 	}
3229 
3230 	if (found) {
3231 		bcopy(hw->mac.addr, hw->mac.perm_addr, ETHERADDRL);
3232 		return (B_TRUE);
3233 	}
3234 #endif
3235 
3236 	/*
3237 	 * Read the device MAC address from the EEPROM
3238 	 */
3239 	if (e1000_read_mac_addr(hw) != E1000_SUCCESS)
3240 		return (B_FALSE);
3241 
3242 	return (B_TRUE);
3243 }
3244 
3245 #pragma inline(igb_arm_watchdog_timer)
3246 
3247 static void
3248 igb_arm_watchdog_timer(igb_t *igb)
3249 {
3250 	/*
3251 	 * Fire a watchdog timer
3252 	 */
3253 	igb->watchdog_tid =
3254 	    timeout(igb_local_timer,
3255 	    (void *)igb, 1 * drv_usectohz(1000000));
3256 
3257 }
3258 
3259 /*
3260  * igb_enable_watchdog_timer - Enable and start the driver watchdog timer
3261  */
3262 void
3263 igb_enable_watchdog_timer(igb_t *igb)
3264 {
3265 	mutex_enter(&igb->watchdog_lock);
3266 
3267 	if (!igb->watchdog_enable) {
3268 		igb->watchdog_enable = B_TRUE;
3269 		igb->watchdog_start = B_TRUE;
3270 		igb_arm_watchdog_timer(igb);
3271 	}
3272 
3273 	mutex_exit(&igb->watchdog_lock);
3274 
3275 }
3276 
3277 /*
3278  * igb_disable_watchdog_timer - Disable and stop the driver watchdog timer
3279  */
3280 void
3281 igb_disable_watchdog_timer(igb_t *igb)
3282 {
3283 	timeout_id_t tid;
3284 
3285 	mutex_enter(&igb->watchdog_lock);
3286 
3287 	igb->watchdog_enable = B_FALSE;
3288 	igb->watchdog_start = B_FALSE;
3289 	tid = igb->watchdog_tid;
3290 	igb->watchdog_tid = 0;
3291 
3292 	mutex_exit(&igb->watchdog_lock);
3293 
3294 	if (tid != 0)
3295 		(void) untimeout(tid);
3296 
3297 }
3298 
3299 /*
3300  * igb_start_watchdog_timer - Start the driver watchdog timer
3301  */
3302 static void
3303 igb_start_watchdog_timer(igb_t *igb)
3304 {
3305 	mutex_enter(&igb->watchdog_lock);
3306 
3307 	if (igb->watchdog_enable) {
3308 		if (!igb->watchdog_start) {
3309 			igb->watchdog_start = B_TRUE;
3310 			igb_arm_watchdog_timer(igb);
3311 		}
3312 	}
3313 
3314 	mutex_exit(&igb->watchdog_lock);
3315 }
3316 
3317 /*
3318  * igb_restart_watchdog_timer - Restart the driver watchdog timer
3319  */
3320 static void
3321 igb_restart_watchdog_timer(igb_t *igb)
3322 {
3323 	mutex_enter(&igb->watchdog_lock);
3324 
3325 	if (igb->watchdog_start)
3326 		igb_arm_watchdog_timer(igb);
3327 
3328 	mutex_exit(&igb->watchdog_lock);
3329 }
3330 
3331 /*
3332  * igb_stop_watchdog_timer - Stop the driver watchdog timer
3333  */
3334 static void
3335 igb_stop_watchdog_timer(igb_t *igb)
3336 {
3337 	timeout_id_t tid;
3338 
3339 	mutex_enter(&igb->watchdog_lock);
3340 
3341 	igb->watchdog_start = B_FALSE;
3342 	tid = igb->watchdog_tid;
3343 	igb->watchdog_tid = 0;
3344 
3345 	mutex_exit(&igb->watchdog_lock);
3346 
3347 	if (tid != 0)
3348 		(void) untimeout(tid);
3349 }
3350 
3351 /*
3352  * igb_start_link_timer - Start the link setup timer
3353  */
3354 static void
3355 igb_start_link_timer(struct igb *igb)
3356 {
3357 	struct e1000_hw *hw = &igb->hw;
3358 	clock_t link_timeout;
3359 
3360 	if (hw->mac.autoneg)
3361 		link_timeout = PHY_AUTO_NEG_LIMIT *
3362 		    drv_usectohz(100000);
3363 	else
3364 		link_timeout = PHY_FORCE_LIMIT * drv_usectohz(100000);
3365 
3366 	mutex_enter(&igb->link_lock);
3367 	if (hw->phy.autoneg_wait_to_complete) {
3368 		igb->link_complete = B_TRUE;
3369 	} else {
3370 		igb->link_complete = B_FALSE;
3371 		igb->link_tid = timeout(igb_link_timer, (void *)igb,
3372 		    link_timeout);
3373 	}
3374 	mutex_exit(&igb->link_lock);
3375 }
3376 
3377 /*
3378  * igb_stop_link_timer - Stop the link setup timer
3379  */
3380 static void
3381 igb_stop_link_timer(struct igb *igb)
3382 {
3383 	timeout_id_t tid;
3384 
3385 	mutex_enter(&igb->link_lock);
3386 	igb->link_complete = B_TRUE;
3387 	tid = igb->link_tid;
3388 	igb->link_tid = 0;
3389 	mutex_exit(&igb->link_lock);
3390 
3391 	if (tid != 0)
3392 		(void) untimeout(tid);
3393 }
3394 
3395 /*
3396  * igb_disable_adapter_interrupts - Clear/disable all hardware interrupts
3397  */
3398 static void
3399 igb_disable_adapter_interrupts(igb_t *igb)
3400 {
3401 	struct e1000_hw *hw = &igb->hw;
3402 
3403 	/*
3404 	 * Set the IMC register to mask all the interrupts,
3405 	 * including the tx interrupts.
3406 	 */
3407 	E1000_WRITE_REG(hw, E1000_IMC, ~0);
3408 	E1000_WRITE_REG(hw, E1000_IAM, 0);
3409 
3410 	/*
3411 	 * Additional disabling for MSI-X
3412 	 */
3413 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3414 		E1000_WRITE_REG(hw, E1000_EIMC, ~0);
3415 		E1000_WRITE_REG(hw, E1000_EIAC, 0);
3416 		E1000_WRITE_REG(hw, E1000_EIAM, 0);
3417 	}
3418 
3419 	E1000_WRITE_FLUSH(hw);
3420 }
3421 
3422 /*
3423  * igb_enable_adapter_interrupts_82580 - Enable NIC interrupts for 82580
3424  */
3425 static void
3426 igb_enable_adapter_interrupts_82580(igb_t *igb)
3427 {
3428 	struct e1000_hw *hw = &igb->hw;
3429 
3430 	/* Clear any pending interrupts */
3431 	(void) E1000_READ_REG(hw, E1000_ICR);
3432 	igb->ims_mask |= E1000_IMS_DRSTA;
3433 
3434 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3435 
3436 		/* Interrupt enabling for MSI-X */
3437 		E1000_WRITE_REG(hw, E1000_EIMS, igb->eims_mask);
3438 		E1000_WRITE_REG(hw, E1000_EIAC, igb->eims_mask);
3439 		igb->ims_mask = (E1000_IMS_LSC | E1000_IMS_DRSTA);
3440 		E1000_WRITE_REG(hw, E1000_IMS, igb->ims_mask);
3441 	} else { /* Interrupt enabling for MSI and legacy */
3442 		E1000_WRITE_REG(hw, E1000_IVAR0, E1000_IVAR_VALID);
3443 		igb->ims_mask = IMS_ENABLE_MASK | E1000_IMS_TXQE;
3444 		igb->ims_mask |= E1000_IMS_DRSTA;
3445 		E1000_WRITE_REG(hw, E1000_IMS, igb->ims_mask);
3446 	}
3447 
3448 	/* Disable auto-mask for ICR interrupt bits */
3449 	E1000_WRITE_REG(hw, E1000_IAM, 0);
3450 
3451 	E1000_WRITE_FLUSH(hw);
3452 }
3453 
3454 /*
3455  * igb_enable_adapter_interrupts_82576 - Enable NIC interrupts for 82576
3456  */
3457 static void
3458 igb_enable_adapter_interrupts_82576(igb_t *igb)
3459 {
3460 	struct e1000_hw *hw = &igb->hw;
3461 
3462 	/* Clear any pending interrupts */
3463 	(void) E1000_READ_REG(hw, E1000_ICR);
3464 
3465 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3466 
3467 		/* Interrupt enabling for MSI-X */
3468 		E1000_WRITE_REG(hw, E1000_EIMS, igb->eims_mask);
3469 		E1000_WRITE_REG(hw, E1000_EIAC, igb->eims_mask);
3470 		igb->ims_mask = E1000_IMS_LSC;
3471 		E1000_WRITE_REG(hw, E1000_IMS, E1000_IMS_LSC);
3472 	} else {
3473 		/* Interrupt enabling for MSI and legacy */
3474 		E1000_WRITE_REG(hw, E1000_IVAR0, E1000_IVAR_VALID);
3475 		igb->ims_mask = IMS_ENABLE_MASK | E1000_IMS_TXQE;
3476 		E1000_WRITE_REG(hw, E1000_IMS,
3477 		    (IMS_ENABLE_MASK | E1000_IMS_TXQE));
3478 	}
3479 
3480 	/* Disable auto-mask for ICR interrupt bits */
3481 	E1000_WRITE_REG(hw, E1000_IAM, 0);
3482 
3483 	E1000_WRITE_FLUSH(hw);
3484 }
3485 
3486 /*
3487  * igb_enable_adapter_interrupts_82575 - Enable NIC interrupts for 82575
3488  */
3489 static void
3490 igb_enable_adapter_interrupts_82575(igb_t *igb)
3491 {
3492 	struct e1000_hw *hw = &igb->hw;
3493 	uint32_t reg;
3494 
3495 	/* Clear any pending interrupts */
3496 	(void) E1000_READ_REG(hw, E1000_ICR);
3497 
3498 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3499 		/* Interrupt enabling for MSI-X */
3500 		E1000_WRITE_REG(hw, E1000_EIMS, igb->eims_mask);
3501 		E1000_WRITE_REG(hw, E1000_EIAC, igb->eims_mask);
3502 		igb->ims_mask = E1000_IMS_LSC;
3503 		E1000_WRITE_REG(hw, E1000_IMS, E1000_IMS_LSC);
3504 
3505 		/* Enable MSI-X PBA support */
3506 		reg = E1000_READ_REG(hw, E1000_CTRL_EXT);
3507 		reg |= E1000_CTRL_EXT_PBA_CLR;
3508 
3509 		/* Non-selective interrupt clear-on-read */
3510 		reg |= E1000_CTRL_EXT_IRCA;	/* Called NSICR in the EAS */
3511 
3512 		E1000_WRITE_REG(hw, E1000_CTRL_EXT, reg);
3513 	} else {
3514 		/* Interrupt enabling for MSI and legacy */
3515 		igb->ims_mask = IMS_ENABLE_MASK;
3516 		E1000_WRITE_REG(hw, E1000_IMS, IMS_ENABLE_MASK);
3517 	}
3518 
3519 	E1000_WRITE_FLUSH(hw);
3520 }
3521 
3522 /*
3523  * Loopback Support
3524  */
3525 static lb_property_t lb_normal =
3526 	{ normal,	"normal",	IGB_LB_NONE		};
3527 static lb_property_t lb_external =
3528 	{ external,	"External",	IGB_LB_EXTERNAL		};
3529 static lb_property_t lb_mac =
3530 	{ internal,	"MAC",		IGB_LB_INTERNAL_MAC	};
3531 static lb_property_t lb_phy =
3532 	{ internal,	"PHY",		IGB_LB_INTERNAL_PHY	};
3533 static lb_property_t lb_serdes =
3534 	{ internal,	"SerDes",	IGB_LB_INTERNAL_SERDES	};
3535 
3536 enum ioc_reply
3537 igb_loopback_ioctl(igb_t *igb, struct iocblk *iocp, mblk_t *mp)
3538 {
3539 	lb_info_sz_t *lbsp;
3540 	lb_property_t *lbpp;
3541 	struct e1000_hw *hw;
3542 	uint32_t *lbmp;
3543 	uint32_t size;
3544 	uint32_t value;
3545 
3546 	hw = &igb->hw;
3547 
3548 	if (mp->b_cont == NULL)
3549 		return (IOC_INVAL);
3550 
3551 	switch (iocp->ioc_cmd) {
3552 	default:
3553 		return (IOC_INVAL);
3554 
3555 	case LB_GET_INFO_SIZE:
3556 		size = sizeof (lb_info_sz_t);
3557 		if (iocp->ioc_count != size)
3558 			return (IOC_INVAL);
3559 
3560 		value = sizeof (lb_normal);
3561 		value += sizeof (lb_mac);
3562 		if (hw->phy.media_type == e1000_media_type_copper)
3563 			value += sizeof (lb_phy);
3564 		else
3565 			value += sizeof (lb_serdes);
3566 		value += sizeof (lb_external);
3567 
3568 		lbsp = (lb_info_sz_t *)(uintptr_t)mp->b_cont->b_rptr;
3569 		*lbsp = value;
3570 		break;
3571 
3572 	case LB_GET_INFO:
3573 		value = sizeof (lb_normal);
3574 		value += sizeof (lb_mac);
3575 		if (hw->phy.media_type == e1000_media_type_copper)
3576 			value += sizeof (lb_phy);
3577 		else
3578 			value += sizeof (lb_serdes);
3579 		value += sizeof (lb_external);
3580 
3581 		size = value;
3582 		if (iocp->ioc_count != size)
3583 			return (IOC_INVAL);
3584 
3585 		value = 0;
3586 		lbpp = (lb_property_t *)(uintptr_t)mp->b_cont->b_rptr;
3587 
3588 		lbpp[value++] = lb_normal;
3589 		lbpp[value++] = lb_mac;
3590 		if (hw->phy.media_type == e1000_media_type_copper)
3591 			lbpp[value++] = lb_phy;
3592 		else
3593 			lbpp[value++] = lb_serdes;
3594 		lbpp[value++] = lb_external;
3595 		break;
3596 
3597 	case LB_GET_MODE:
3598 		size = sizeof (uint32_t);
3599 		if (iocp->ioc_count != size)
3600 			return (IOC_INVAL);
3601 
3602 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
3603 		*lbmp = igb->loopback_mode;
3604 		break;
3605 
3606 	case LB_SET_MODE:
3607 		size = 0;
3608 		if (iocp->ioc_count != sizeof (uint32_t))
3609 			return (IOC_INVAL);
3610 
3611 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
3612 		if (!igb_set_loopback_mode(igb, *lbmp))
3613 			return (IOC_INVAL);
3614 		break;
3615 	}
3616 
3617 	iocp->ioc_count = size;
3618 	iocp->ioc_error = 0;
3619 
3620 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
3621 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
3622 		return (IOC_INVAL);
3623 	}
3624 
3625 	return (IOC_REPLY);
3626 }
3627 
3628 /*
3629  * igb_set_loopback_mode - Setup loopback based on the loopback mode
3630  */
3631 static boolean_t
3632 igb_set_loopback_mode(igb_t *igb, uint32_t mode)
3633 {
3634 	struct e1000_hw *hw;
3635 
3636 	if (mode == igb->loopback_mode)
3637 		return (B_TRUE);
3638 
3639 	hw = &igb->hw;
3640 
3641 	igb->loopback_mode = mode;
3642 
3643 	if (mode == IGB_LB_NONE) {
3644 		/* Reset the chip */
3645 		hw->phy.autoneg_wait_to_complete = B_TRUE;
3646 		(void) igb_reset(igb);
3647 		hw->phy.autoneg_wait_to_complete = B_FALSE;
3648 		return (B_TRUE);
3649 	}
3650 
3651 	mutex_enter(&igb->gen_lock);
3652 
3653 	switch (mode) {
3654 	default:
3655 		mutex_exit(&igb->gen_lock);
3656 		return (B_FALSE);
3657 
3658 	case IGB_LB_EXTERNAL:
3659 		igb_set_external_loopback(igb);
3660 		break;
3661 
3662 	case IGB_LB_INTERNAL_MAC:
3663 		igb_set_internal_mac_loopback(igb);
3664 		break;
3665 
3666 	case IGB_LB_INTERNAL_PHY:
3667 		igb_set_internal_phy_loopback(igb);
3668 		break;
3669 
3670 	case IGB_LB_INTERNAL_SERDES:
3671 		igb_set_internal_serdes_loopback(igb);
3672 		break;
3673 	}
3674 
3675 	mutex_exit(&igb->gen_lock);
3676 
3677 	return (B_TRUE);
3678 }
3679 
3680 /*
3681  * igb_set_external_loopback - Set the external loopback mode
3682  */
3683 static void
3684 igb_set_external_loopback(igb_t *igb)
3685 {
3686 	struct e1000_hw *hw;
3687 
3688 	hw = &igb->hw;
3689 
3690 	/* Set phy to known state */
3691 	(void) e1000_phy_hw_reset(hw);
3692 
3693 	(void) e1000_write_phy_reg(hw, 0x0, 0x0140);
3694 	(void) e1000_write_phy_reg(hw, 0x9, 0x1b00);
3695 	(void) e1000_write_phy_reg(hw, 0x12, 0x1610);
3696 	(void) e1000_write_phy_reg(hw, 0x1f37, 0x3f1c);
3697 }
3698 
3699 /*
3700  * igb_set_internal_mac_loopback - Set the internal MAC loopback mode
3701  */
3702 static void
3703 igb_set_internal_mac_loopback(igb_t *igb)
3704 {
3705 	struct e1000_hw *hw;
3706 	uint32_t ctrl;
3707 	uint32_t rctl;
3708 	uint32_t ctrl_ext;
3709 	uint16_t phy_ctrl;
3710 	uint16_t phy_status;
3711 
3712 	hw = &igb->hw;
3713 
3714 	(void) e1000_read_phy_reg(hw, PHY_CONTROL, &phy_ctrl);
3715 	phy_ctrl &= ~MII_CR_AUTO_NEG_EN;
3716 	(void) e1000_write_phy_reg(hw, PHY_CONTROL, phy_ctrl);
3717 
3718 	(void) e1000_read_phy_reg(hw, PHY_STATUS, &phy_status);
3719 
3720 	/* Set link mode to PHY (00b) in the Extended Control register */
3721 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3722 	ctrl_ext &= ~E1000_CTRL_EXT_LINK_MODE_MASK;
3723 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3724 
3725 	/* Set the Device Control register */
3726 	ctrl = E1000_READ_REG(hw, E1000_CTRL);
3727 	if (!(phy_status & MII_SR_LINK_STATUS))
3728 		ctrl |= E1000_CTRL_ILOS; /* Set ILOS when the link is down */
3729 	ctrl &= ~E1000_CTRL_SPD_SEL;	/* Clear the speed sel bits */
3730 	ctrl |= (E1000_CTRL_SLU |	/* Force link up */
3731 	    E1000_CTRL_FRCSPD |		/* Force speed */
3732 	    E1000_CTRL_FRCDPX |		/* Force duplex */
3733 	    E1000_CTRL_SPD_1000 |	/* Force speed to 1000 */
3734 	    E1000_CTRL_FD);		/* Force full duplex */
3735 
3736 	E1000_WRITE_REG(hw, E1000_CTRL, ctrl);
3737 
3738 	/* Set the Receive Control register */
3739 	rctl = E1000_READ_REG(hw, E1000_RCTL);
3740 	rctl &= ~E1000_RCTL_LBM_TCVR;
3741 	rctl |= E1000_RCTL_LBM_MAC;
3742 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
3743 }
3744 
3745 /*
3746  * igb_set_internal_phy_loopback - Set the internal PHY loopback mode
3747  */
3748 static void
3749 igb_set_internal_phy_loopback(igb_t *igb)
3750 {
3751 	struct e1000_hw *hw;
3752 	uint32_t ctrl_ext;
3753 	uint16_t phy_ctrl;
3754 	uint16_t phy_pconf;
3755 
3756 	hw = &igb->hw;
3757 
3758 	/* Set link mode to PHY (00b) in the Extended Control register */
3759 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3760 	ctrl_ext &= ~E1000_CTRL_EXT_LINK_MODE_MASK;
3761 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3762 
3763 	/*
3764 	 * Set PHY control register (0x4140):
3765 	 *    Set full duplex mode
3766 	 *    Set loopback bit
3767 	 *    Clear auto-neg enable bit
3768 	 *    Set PHY speed
3769 	 */
3770 	phy_ctrl = MII_CR_FULL_DUPLEX | MII_CR_SPEED_1000 | MII_CR_LOOPBACK;
3771 	(void) e1000_write_phy_reg(hw, PHY_CONTROL, phy_ctrl);
3772 
3773 	/* Set the link disable bit in the Port Configuration register */
3774 	(void) e1000_read_phy_reg(hw, 0x10, &phy_pconf);
3775 	phy_pconf |= (uint16_t)1 << 14;
3776 	(void) e1000_write_phy_reg(hw, 0x10, phy_pconf);
3777 }
3778 
3779 /*
3780  * igb_set_internal_serdes_loopback - Set the internal SerDes loopback mode
3781  */
3782 static void
3783 igb_set_internal_serdes_loopback(igb_t *igb)
3784 {
3785 	struct e1000_hw *hw;
3786 	uint32_t ctrl_ext;
3787 	uint32_t ctrl;
3788 	uint32_t pcs_lctl;
3789 	uint32_t connsw;
3790 
3791 	hw = &igb->hw;
3792 
3793 	/* Set link mode to SerDes (11b) in the Extended Control register */
3794 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3795 	ctrl_ext |= E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
3796 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3797 
3798 	/* Configure the SerDes to loopback */
3799 	E1000_WRITE_REG(hw, E1000_SCTL, 0x410);
3800 
3801 	/* Set Device Control register */
3802 	ctrl = E1000_READ_REG(hw, E1000_CTRL);
3803 	ctrl |= (E1000_CTRL_FD |	/* Force full duplex */
3804 	    E1000_CTRL_SLU);		/* Force link up */
3805 	ctrl &= ~(E1000_CTRL_RFCE |	/* Disable receive flow control */
3806 	    E1000_CTRL_TFCE |		/* Disable transmit flow control */
3807 	    E1000_CTRL_LRST);		/* Clear link reset */
3808 	E1000_WRITE_REG(hw, E1000_CTRL, ctrl);
3809 
3810 	/* Set PCS Link Control register */
3811 	pcs_lctl = E1000_READ_REG(hw, E1000_PCS_LCTL);
3812 	pcs_lctl |= (E1000_PCS_LCTL_FORCE_LINK |
3813 	    E1000_PCS_LCTL_FSD |
3814 	    E1000_PCS_LCTL_FDV_FULL |
3815 	    E1000_PCS_LCTL_FLV_LINK_UP);
3816 	pcs_lctl &= ~E1000_PCS_LCTL_AN_ENABLE;
3817 	E1000_WRITE_REG(hw, E1000_PCS_LCTL, pcs_lctl);
3818 
3819 	/* Set the Copper/Fiber Switch Control - CONNSW register */
3820 	connsw = E1000_READ_REG(hw, E1000_CONNSW);
3821 	connsw &= ~E1000_CONNSW_ENRGSRC;
3822 	E1000_WRITE_REG(hw, E1000_CONNSW, connsw);
3823 }
3824 
3825 #pragma inline(igb_intr_rx_work)
3826 /*
3827  * igb_intr_rx_work - rx processing of ISR
3828  */
3829 static void
3830 igb_intr_rx_work(igb_rx_ring_t *rx_ring)
3831 {
3832 	mblk_t *mp;
3833 
3834 	mutex_enter(&rx_ring->rx_lock);
3835 	mp = igb_rx(rx_ring, IGB_NO_POLL);
3836 	mutex_exit(&rx_ring->rx_lock);
3837 
3838 	if (mp != NULL)
3839 		mac_rx_ring(rx_ring->igb->mac_hdl, rx_ring->ring_handle, mp,
3840 		    rx_ring->ring_gen_num);
3841 }
3842 
3843 #pragma inline(igb_intr_tx_work)
3844 /*
3845  * igb_intr_tx_work - tx processing of ISR
3846  */
3847 static void
3848 igb_intr_tx_work(igb_tx_ring_t *tx_ring)
3849 {
3850 	/* Recycle the tx descriptors */
3851 	tx_ring->tx_recycle(tx_ring);
3852 
3853 	/* Schedule the re-transmit */
3854 	if (tx_ring->reschedule &&
3855 	    (tx_ring->tbd_free >= tx_ring->resched_thresh)) {
3856 		tx_ring->reschedule = B_FALSE;
3857 		mac_tx_ring_update(tx_ring->igb->mac_hdl, tx_ring->ring_handle);
3858 		IGB_DEBUG_STAT(tx_ring->stat_reschedule);
3859 	}
3860 }
3861 
3862 #pragma inline(igb_intr_link_work)
3863 /*
3864  * igb_intr_link_work - link-status-change processing of ISR
3865  */
3866 static void
3867 igb_intr_link_work(igb_t *igb)
3868 {
3869 	boolean_t link_changed;
3870 
3871 	igb_stop_watchdog_timer(igb);
3872 
3873 	mutex_enter(&igb->gen_lock);
3874 
3875 	/*
3876 	 * Because we got a link-status-change interrupt, force
3877 	 * e1000_check_for_link() to look at phy
3878 	 */
3879 	igb->hw.mac.get_link_status = B_TRUE;
3880 
3881 	/* igb_link_check takes care of link status change */
3882 	link_changed = igb_link_check(igb);
3883 
3884 	/* Get new phy state */
3885 	igb_get_phy_state(igb);
3886 
3887 	mutex_exit(&igb->gen_lock);
3888 
3889 	if (link_changed)
3890 		mac_link_update(igb->mac_hdl, igb->link_state);
3891 
3892 	igb_start_watchdog_timer(igb);
3893 }
3894 
3895 /*
3896  * igb_intr_legacy - Interrupt handler for legacy interrupts
3897  */
3898 static uint_t
3899 igb_intr_legacy(void *arg1, void *arg2)
3900 {
3901 	igb_t *igb = (igb_t *)arg1;
3902 	igb_tx_ring_t *tx_ring;
3903 	uint32_t icr;
3904 	mblk_t *mp;
3905 	boolean_t tx_reschedule;
3906 	boolean_t link_changed;
3907 	uint_t result;
3908 
3909 	_NOTE(ARGUNUSED(arg2));
3910 
3911 	mutex_enter(&igb->gen_lock);
3912 
3913 	if (igb->igb_state & IGB_SUSPENDED) {
3914 		mutex_exit(&igb->gen_lock);
3915 		return (DDI_INTR_UNCLAIMED);
3916 	}
3917 
3918 	mp = NULL;
3919 	tx_reschedule = B_FALSE;
3920 	link_changed = B_FALSE;
3921 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
3922 
3923 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
3924 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
3925 		atomic_or_32(&igb->igb_state, IGB_ERROR);
3926 		return (DDI_INTR_UNCLAIMED);
3927 	}
3928 
3929 	if (icr & E1000_ICR_INT_ASSERTED) {
3930 		/*
3931 		 * E1000_ICR_INT_ASSERTED bit was set:
3932 		 * Read(Clear) the ICR, claim this interrupt,
3933 		 * look for work to do.
3934 		 */
3935 		ASSERT(igb->num_rx_rings == 1);
3936 		ASSERT(igb->num_tx_rings == 1);
3937 
3938 		/* Make sure all interrupt causes cleared */
3939 		(void) E1000_READ_REG(&igb->hw, E1000_EICR);
3940 
3941 		if (icr & E1000_ICR_RXT0) {
3942 			mp = igb_rx(&igb->rx_rings[0], IGB_NO_POLL);
3943 		}
3944 
3945 		if (icr & E1000_ICR_TXDW) {
3946 			tx_ring = &igb->tx_rings[0];
3947 
3948 			/* Recycle the tx descriptors */
3949 			tx_ring->tx_recycle(tx_ring);
3950 
3951 			/* Schedule the re-transmit */
3952 			tx_reschedule = (tx_ring->reschedule &&
3953 			    (tx_ring->tbd_free >= tx_ring->resched_thresh));
3954 		}
3955 
3956 		if (icr & E1000_ICR_LSC) {
3957 			/*
3958 			 * Because we got a link-status-change interrupt, force
3959 			 * e1000_check_for_link() to look at phy
3960 			 */
3961 			igb->hw.mac.get_link_status = B_TRUE;
3962 
3963 			/* igb_link_check takes care of link status change */
3964 			link_changed = igb_link_check(igb);
3965 
3966 			/* Get new phy state */
3967 			igb_get_phy_state(igb);
3968 		}
3969 
3970 		if (icr & E1000_ICR_DRSTA) {
3971 			/* 82580 Full Device Reset needed */
3972 			atomic_or_32(&igb->igb_state, IGB_STALL);
3973 		}
3974 
3975 		result = DDI_INTR_CLAIMED;
3976 	} else {
3977 		/*
3978 		 * E1000_ICR_INT_ASSERTED bit was not set:
3979 		 * Don't claim this interrupt.
3980 		 */
3981 		result = DDI_INTR_UNCLAIMED;
3982 	}
3983 
3984 	mutex_exit(&igb->gen_lock);
3985 
3986 	/*
3987 	 * Do the following work outside of the gen_lock
3988 	 */
3989 	if (mp != NULL)
3990 		mac_rx(igb->mac_hdl, NULL, mp);
3991 
3992 	if (tx_reschedule)  {
3993 		tx_ring->reschedule = B_FALSE;
3994 		mac_tx_ring_update(igb->mac_hdl, tx_ring->ring_handle);
3995 		IGB_DEBUG_STAT(tx_ring->stat_reschedule);
3996 	}
3997 
3998 	if (link_changed)
3999 		mac_link_update(igb->mac_hdl, igb->link_state);
4000 
4001 	return (result);
4002 }
4003 
4004 /*
4005  * igb_intr_msi - Interrupt handler for MSI
4006  */
4007 static uint_t
4008 igb_intr_msi(void *arg1, void *arg2)
4009 {
4010 	igb_t *igb = (igb_t *)arg1;
4011 	uint32_t icr;
4012 
4013 	_NOTE(ARGUNUSED(arg2));
4014 
4015 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
4016 
4017 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
4018 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
4019 		atomic_or_32(&igb->igb_state, IGB_ERROR);
4020 		return (DDI_INTR_CLAIMED);
4021 	}
4022 
4023 	/* Make sure all interrupt causes cleared */
4024 	(void) E1000_READ_REG(&igb->hw, E1000_EICR);
4025 
4026 	/*
4027 	 * For MSI interrupt, we have only one vector,
4028 	 * so we have only one rx ring and one tx ring enabled.
4029 	 */
4030 	ASSERT(igb->num_rx_rings == 1);
4031 	ASSERT(igb->num_tx_rings == 1);
4032 
4033 	if (icr & E1000_ICR_RXT0) {
4034 		igb_intr_rx_work(&igb->rx_rings[0]);
4035 	}
4036 
4037 	if (icr & E1000_ICR_TXDW) {
4038 		igb_intr_tx_work(&igb->tx_rings[0]);
4039 	}
4040 
4041 	if (icr & E1000_ICR_LSC) {
4042 		igb_intr_link_work(igb);
4043 	}
4044 
4045 	if (icr & E1000_ICR_DRSTA) {
4046 		/* 82580 Full Device Reset needed */
4047 		atomic_or_32(&igb->igb_state, IGB_STALL);
4048 	}
4049 
4050 	return (DDI_INTR_CLAIMED);
4051 }
4052 
4053 /*
4054  * igb_intr_rx - Interrupt handler for rx
4055  */
4056 static uint_t
4057 igb_intr_rx(void *arg1, void *arg2)
4058 {
4059 	igb_rx_ring_t *rx_ring = (igb_rx_ring_t *)arg1;
4060 
4061 	_NOTE(ARGUNUSED(arg2));
4062 
4063 	/*
4064 	 * Only used via MSI-X vector so don't check cause bits
4065 	 * and only clean the given ring.
4066 	 */
4067 	igb_intr_rx_work(rx_ring);
4068 
4069 	return (DDI_INTR_CLAIMED);
4070 }
4071 
4072 /*
4073  * igb_intr_tx - Interrupt handler for tx
4074  */
4075 static uint_t
4076 igb_intr_tx(void *arg1, void *arg2)
4077 {
4078 	igb_tx_ring_t *tx_ring = (igb_tx_ring_t *)arg1;
4079 
4080 	_NOTE(ARGUNUSED(arg2));
4081 
4082 	/*
4083 	 * Only used via MSI-X vector so don't check cause bits
4084 	 * and only clean the given ring.
4085 	 */
4086 	igb_intr_tx_work(tx_ring);
4087 
4088 	return (DDI_INTR_CLAIMED);
4089 }
4090 
4091 /*
4092  * igb_intr_tx_other - Interrupt handler for both tx and other
4093  *
4094  */
4095 static uint_t
4096 igb_intr_tx_other(void *arg1, void *arg2)
4097 {
4098 	igb_t *igb = (igb_t *)arg1;
4099 	uint32_t icr;
4100 
4101 	_NOTE(ARGUNUSED(arg2));
4102 
4103 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
4104 
4105 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
4106 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
4107 		atomic_or_32(&igb->igb_state, IGB_ERROR);
4108 		return (DDI_INTR_CLAIMED);
4109 	}
4110 
4111 	/*
4112 	 * Look for tx reclaiming work first. Remember, in the
4113 	 * case of only interrupt sharing, only one tx ring is
4114 	 * used
4115 	 */
4116 	igb_intr_tx_work(&igb->tx_rings[0]);
4117 
4118 	/*
4119 	 * Check for "other" causes.
4120 	 */
4121 	if (icr & E1000_ICR_LSC) {
4122 		igb_intr_link_work(igb);
4123 	}
4124 
4125 	/*
4126 	 * The DOUTSYNC bit indicates a tx packet dropped because
4127 	 * DMA engine gets "out of sync". There isn't a real fix
4128 	 * for this. The Intel recommendation is to count the number
4129 	 * of occurrences so user can detect when it is happening.
4130 	 * The issue is non-fatal and there's no recovery action
4131 	 * available.
4132 	 */
4133 	if (icr & E1000_ICR_DOUTSYNC) {
4134 		IGB_STAT(igb->dout_sync);
4135 	}
4136 
4137 	if (icr & E1000_ICR_DRSTA) {
4138 		/* 82580 Full Device Reset needed */
4139 		atomic_or_32(&igb->igb_state, IGB_STALL);
4140 	}
4141 
4142 	return (DDI_INTR_CLAIMED);
4143 }
4144 
4145 /*
4146  * igb_alloc_intrs - Allocate interrupts for the driver
4147  *
4148  * Normal sequence is to try MSI-X; if not sucessful, try MSI;
4149  * if not successful, try Legacy.
4150  * igb->intr_force can be used to force sequence to start with
4151  * any of the 3 types.
4152  * If MSI-X is not used, number of tx/rx rings is forced to 1.
4153  */
4154 static int
4155 igb_alloc_intrs(igb_t *igb)
4156 {
4157 	dev_info_t *devinfo;
4158 	int intr_types;
4159 	int rc;
4160 
4161 	devinfo = igb->dip;
4162 
4163 	/* Get supported interrupt types */
4164 	rc = ddi_intr_get_supported_types(devinfo, &intr_types);
4165 
4166 	if (rc != DDI_SUCCESS) {
4167 		igb_log(igb,
4168 		    "Get supported interrupt types failed: %d", rc);
4169 		return (IGB_FAILURE);
4170 	}
4171 	IGB_DEBUGLOG_1(igb, "Supported interrupt types: %x", intr_types);
4172 
4173 	igb->intr_type = 0;
4174 
4175 	/* Install MSI-X interrupts */
4176 	if ((intr_types & DDI_INTR_TYPE_MSIX) &&
4177 	    (igb->intr_force <= IGB_INTR_MSIX)) {
4178 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_MSIX);
4179 
4180 		if (rc == IGB_SUCCESS)
4181 			return (IGB_SUCCESS);
4182 
4183 		igb_log(igb,
4184 		    "Allocate MSI-X failed, trying MSI interrupts...");
4185 	}
4186 
4187 	/* MSI-X not used, force rings to 1 */
4188 	igb->num_rx_rings = 1;
4189 	igb->num_tx_rings = 1;
4190 	igb_log(igb,
4191 	    "MSI-X not used, force rx and tx queue number to 1");
4192 
4193 	/* Install MSI interrupts */
4194 	if ((intr_types & DDI_INTR_TYPE_MSI) &&
4195 	    (igb->intr_force <= IGB_INTR_MSI)) {
4196 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_MSI);
4197 
4198 		if (rc == IGB_SUCCESS)
4199 			return (IGB_SUCCESS);
4200 
4201 		igb_log(igb,
4202 		    "Allocate MSI failed, trying Legacy interrupts...");
4203 	}
4204 
4205 	/* Install legacy interrupts */
4206 	if (intr_types & DDI_INTR_TYPE_FIXED) {
4207 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_FIXED);
4208 
4209 		if (rc == IGB_SUCCESS)
4210 			return (IGB_SUCCESS);
4211 
4212 		igb_log(igb,
4213 		    "Allocate Legacy interrupts failed");
4214 	}
4215 
4216 	/* If none of the 3 types succeeded, return failure */
4217 	return (IGB_FAILURE);
4218 }
4219 
4220 /*
4221  * igb_alloc_intr_handles - Allocate interrupt handles.
4222  *
4223  * For legacy and MSI, only 1 handle is needed.  For MSI-X,
4224  * if fewer than 2 handles are available, return failure.
4225  * Upon success, this sets the number of Rx rings to a number that
4226  * matches the handles available for Rx interrupts.
4227  */
4228 static int
4229 igb_alloc_intr_handles(igb_t *igb, int intr_type)
4230 {
4231 	dev_info_t *devinfo;
4232 	int orig, request, count, avail, actual;
4233 	int diff, minimum;
4234 	int rc;
4235 
4236 	devinfo = igb->dip;
4237 
4238 	switch (intr_type) {
4239 	case DDI_INTR_TYPE_FIXED:
4240 		request = 1;	/* Request 1 legacy interrupt handle */
4241 		minimum = 1;
4242 		IGB_DEBUGLOG_0(igb, "interrupt type: legacy");
4243 		break;
4244 
4245 	case DDI_INTR_TYPE_MSI:
4246 		request = 1;	/* Request 1 MSI interrupt handle */
4247 		minimum = 1;
4248 		IGB_DEBUGLOG_0(igb, "interrupt type: MSI");
4249 		break;
4250 
4251 	case DDI_INTR_TYPE_MSIX:
4252 		/*
4253 		 * Number of vectors for the adapter is
4254 		 * # rx rings + # tx rings
4255 		 * One of tx vectors is for tx & other
4256 		 */
4257 		request = igb->num_rx_rings + igb->num_tx_rings;
4258 		orig = request;
4259 		minimum = 2;
4260 		IGB_DEBUGLOG_0(igb, "interrupt type: MSI-X");
4261 		break;
4262 
4263 	default:
4264 		igb_log(igb,
4265 		    "invalid call to igb_alloc_intr_handles(): %d\n",
4266 		    intr_type);
4267 		return (IGB_FAILURE);
4268 	}
4269 	IGB_DEBUGLOG_2(igb, "interrupt handles requested: %d  minimum: %d",
4270 	    request, minimum);
4271 
4272 	/*
4273 	 * Get number of supported interrupts
4274 	 */
4275 	rc = ddi_intr_get_nintrs(devinfo, intr_type, &count);
4276 	if ((rc != DDI_SUCCESS) || (count < minimum)) {
4277 		igb_log(igb,
4278 		    "Get supported interrupt number failed. "
4279 		    "Return: %d, count: %d", rc, count);
4280 		return (IGB_FAILURE);
4281 	}
4282 	IGB_DEBUGLOG_1(igb, "interrupts supported: %d", count);
4283 
4284 	/*
4285 	 * Get number of available interrupts
4286 	 */
4287 	rc = ddi_intr_get_navail(devinfo, intr_type, &avail);
4288 	if ((rc != DDI_SUCCESS) || (avail < minimum)) {
4289 		igb_log(igb,
4290 		    "Get available interrupt number failed. "
4291 		    "Return: %d, available: %d", rc, avail);
4292 		return (IGB_FAILURE);
4293 	}
4294 	IGB_DEBUGLOG_1(igb, "interrupts available: %d", avail);
4295 
4296 	if (avail < request) {
4297 		igb_log(igb, "Request %d handles, %d available",
4298 		    request, avail);
4299 		request = avail;
4300 	}
4301 
4302 	actual = 0;
4303 	igb->intr_cnt = 0;
4304 
4305 	/*
4306 	 * Allocate an array of interrupt handles
4307 	 */
4308 	igb->intr_size = request * sizeof (ddi_intr_handle_t);
4309 	igb->htable = kmem_alloc(igb->intr_size, KM_SLEEP);
4310 
4311 	rc = ddi_intr_alloc(devinfo, igb->htable, intr_type, 0,
4312 	    request, &actual, DDI_INTR_ALLOC_NORMAL);
4313 	if (rc != DDI_SUCCESS) {
4314 		igb_log(igb, "Allocate interrupts failed. "
4315 		    "return: %d, request: %d, actual: %d",
4316 		    rc, request, actual);
4317 		goto alloc_handle_fail;
4318 	}
4319 	IGB_DEBUGLOG_1(igb, "interrupts actually allocated: %d", actual);
4320 
4321 	igb->intr_cnt = actual;
4322 
4323 	if (actual < minimum) {
4324 		igb_log(igb, "Insufficient interrupt handles allocated: %d",
4325 		    actual);
4326 		goto alloc_handle_fail;
4327 	}
4328 
4329 	/*
4330 	 * For MSI-X, actual might force us to reduce number of tx & rx rings
4331 	 */
4332 	if ((intr_type == DDI_INTR_TYPE_MSIX) && (orig > actual)) {
4333 		diff = orig - actual;
4334 		if (diff < igb->num_tx_rings) {
4335 			igb_log(igb,
4336 			    "MSI-X vectors force Tx queue number to %d",
4337 			    igb->num_tx_rings - diff);
4338 			igb->num_tx_rings -= diff;
4339 		} else {
4340 			igb_log(igb,
4341 			    "MSI-X vectors force Tx queue number to 1");
4342 			igb->num_tx_rings = 1;
4343 
4344 			igb_log(igb,
4345 			    "MSI-X vectors force Rx queue number to %d",
4346 			    actual - 1);
4347 			igb->num_rx_rings = actual - 1;
4348 		}
4349 	}
4350 
4351 	/*
4352 	 * Get priority for first vector, assume remaining are all the same
4353 	 */
4354 	rc = ddi_intr_get_pri(igb->htable[0], &igb->intr_pri);
4355 	if (rc != DDI_SUCCESS) {
4356 		igb_log(igb,
4357 		    "Get interrupt priority failed: %d", rc);
4358 		goto alloc_handle_fail;
4359 	}
4360 
4361 	rc = ddi_intr_get_cap(igb->htable[0], &igb->intr_cap);
4362 	if (rc != DDI_SUCCESS) {
4363 		igb_log(igb,
4364 		    "Get interrupt cap failed: %d", rc);
4365 		goto alloc_handle_fail;
4366 	}
4367 
4368 	igb->intr_type = intr_type;
4369 
4370 	return (IGB_SUCCESS);
4371 
4372 alloc_handle_fail:
4373 	igb_rem_intrs(igb);
4374 
4375 	return (IGB_FAILURE);
4376 }
4377 
4378 /*
4379  * igb_add_intr_handlers - Add interrupt handlers based on the interrupt type
4380  *
4381  * Before adding the interrupt handlers, the interrupt vectors have
4382  * been allocated, and the rx/tx rings have also been allocated.
4383  */
4384 static int
4385 igb_add_intr_handlers(igb_t *igb)
4386 {
4387 	igb_rx_ring_t *rx_ring;
4388 	igb_tx_ring_t *tx_ring;
4389 	int vector;
4390 	int rc;
4391 	int i;
4392 
4393 	vector = 0;
4394 
4395 	switch (igb->intr_type) {
4396 	case DDI_INTR_TYPE_MSIX:
4397 		/* Add interrupt handler for tx + other */
4398 		tx_ring = &igb->tx_rings[0];
4399 		rc = ddi_intr_add_handler(igb->htable[vector],
4400 		    (ddi_intr_handler_t *)igb_intr_tx_other,
4401 		    (void *)igb, NULL);
4402 
4403 		if (rc != DDI_SUCCESS) {
4404 			igb_log(igb,
4405 			    "Add tx/other interrupt handler failed: %d", rc);
4406 			return (IGB_FAILURE);
4407 		}
4408 		tx_ring->intr_vector = vector;
4409 		vector++;
4410 
4411 		/* Add interrupt handler for each rx ring */
4412 		for (i = 0; i < igb->num_rx_rings; i++) {
4413 			rx_ring = &igb->rx_rings[i];
4414 
4415 			rc = ddi_intr_add_handler(igb->htable[vector],
4416 			    (ddi_intr_handler_t *)igb_intr_rx,
4417 			    (void *)rx_ring, NULL);
4418 
4419 			if (rc != DDI_SUCCESS) {
4420 				igb_log(igb,
4421 				    "Add rx interrupt handler failed. "
4422 				    "return: %d, rx ring: %d", rc, i);
4423 				for (vector--; vector >= 0; vector--) {
4424 					(void) ddi_intr_remove_handler(
4425 					    igb->htable[vector]);
4426 				}
4427 				return (IGB_FAILURE);
4428 			}
4429 
4430 			rx_ring->intr_vector = vector;
4431 
4432 			vector++;
4433 		}
4434 
4435 		/* Add interrupt handler for each tx ring from 2nd ring */
4436 		for (i = 1; i < igb->num_tx_rings; i++) {
4437 			tx_ring = &igb->tx_rings[i];
4438 
4439 			rc = ddi_intr_add_handler(igb->htable[vector],
4440 			    (ddi_intr_handler_t *)igb_intr_tx,
4441 			    (void *)tx_ring, NULL);
4442 
4443 			if (rc != DDI_SUCCESS) {
4444 				igb_log(igb,
4445 				    "Add tx interrupt handler failed. "
4446 				    "return: %d, tx ring: %d", rc, i);
4447 				for (vector--; vector >= 0; vector--) {
4448 					(void) ddi_intr_remove_handler(
4449 					    igb->htable[vector]);
4450 				}
4451 				return (IGB_FAILURE);
4452 			}
4453 
4454 			tx_ring->intr_vector = vector;
4455 
4456 			vector++;
4457 		}
4458 
4459 		break;
4460 
4461 	case DDI_INTR_TYPE_MSI:
4462 		/* Add interrupt handlers for the only vector */
4463 		rc = ddi_intr_add_handler(igb->htable[vector],
4464 		    (ddi_intr_handler_t *)igb_intr_msi,
4465 		    (void *)igb, NULL);
4466 
4467 		if (rc != DDI_SUCCESS) {
4468 			igb_log(igb,
4469 			    "Add MSI interrupt handler failed: %d", rc);
4470 			return (IGB_FAILURE);
4471 		}
4472 
4473 		rx_ring = &igb->rx_rings[0];
4474 		rx_ring->intr_vector = vector;
4475 
4476 		vector++;
4477 		break;
4478 
4479 	case DDI_INTR_TYPE_FIXED:
4480 		/* Add interrupt handlers for the only vector */
4481 		rc = ddi_intr_add_handler(igb->htable[vector],
4482 		    (ddi_intr_handler_t *)igb_intr_legacy,
4483 		    (void *)igb, NULL);
4484 
4485 		if (rc != DDI_SUCCESS) {
4486 			igb_log(igb,
4487 			    "Add legacy interrupt handler failed: %d", rc);
4488 			return (IGB_FAILURE);
4489 		}
4490 
4491 		rx_ring = &igb->rx_rings[0];
4492 		rx_ring->intr_vector = vector;
4493 
4494 		vector++;
4495 		break;
4496 
4497 	default:
4498 		return (IGB_FAILURE);
4499 	}
4500 
4501 	ASSERT(vector == igb->intr_cnt);
4502 
4503 	return (IGB_SUCCESS);
4504 }
4505 
4506 /*
4507  * igb_setup_msix_82575 - setup 82575 adapter to use MSI-X interrupts
4508  *
4509  * For each vector enabled on the adapter, Set the MSIXBM register accordingly
4510  */
4511 static void
4512 igb_setup_msix_82575(igb_t *igb)
4513 {
4514 	uint32_t eims = 0;
4515 	int i, vector;
4516 	struct e1000_hw *hw = &igb->hw;
4517 
4518 	/*
4519 	 * Set vector for tx ring 0 and other causes.
4520 	 * NOTE assumption that it is vector 0.
4521 	 */
4522 	vector = 0;
4523 
4524 	igb->eims_mask = E1000_EICR_TX_QUEUE0 | E1000_EICR_OTHER;
4525 	E1000_WRITE_REG(hw, E1000_MSIXBM(vector), igb->eims_mask);
4526 	vector++;
4527 
4528 	for (i = 0; i < igb->num_rx_rings; i++) {
4529 		/*
4530 		 * Set vector for each rx ring
4531 		 */
4532 		eims = (E1000_EICR_RX_QUEUE0 << i);
4533 		E1000_WRITE_REG(hw, E1000_MSIXBM(vector), eims);
4534 
4535 		/*
4536 		 * Accumulate bits to enable in
4537 		 * igb_enable_adapter_interrupts_82575()
4538 		 */
4539 		igb->eims_mask |= eims;
4540 
4541 		vector++;
4542 	}
4543 
4544 	for (i = 1; i < igb->num_tx_rings; i++) {
4545 		/*
4546 		 * Set vector for each tx ring from 2nd tx ring
4547 		 */
4548 		eims = (E1000_EICR_TX_QUEUE0 << i);
4549 		E1000_WRITE_REG(hw, E1000_MSIXBM(vector), eims);
4550 
4551 		/*
4552 		 * Accumulate bits to enable in
4553 		 * igb_enable_adapter_interrupts_82575()
4554 		 */
4555 		igb->eims_mask |= eims;
4556 
4557 		vector++;
4558 	}
4559 
4560 	ASSERT(vector == igb->intr_cnt);
4561 
4562 	/*
4563 	 * Disable IAM for ICR interrupt bits
4564 	 */
4565 	E1000_WRITE_REG(hw, E1000_IAM, 0);
4566 	E1000_WRITE_FLUSH(hw);
4567 }
4568 
4569 /*
4570  * igb_setup_msix_82576 - setup 82576 adapter to use MSI-X interrupts
4571  *
4572  * 82576 uses a table based method for assigning vectors.  Each queue has a
4573  * single entry in the table to which we write a vector number along with a
4574  * "valid" bit.  The entry is a single byte in a 4-byte register.  Vectors
4575  * take a different position in the 4-byte register depending on whether
4576  * they are numbered above or below 8.
4577  */
4578 static void
4579 igb_setup_msix_82576(igb_t *igb)
4580 {
4581 	struct e1000_hw *hw = &igb->hw;
4582 	uint32_t ivar, index, vector;
4583 	int i;
4584 
4585 	/* must enable msi-x capability before IVAR settings */
4586 	E1000_WRITE_REG(hw, E1000_GPIE,
4587 	    (E1000_GPIE_MSIX_MODE | E1000_GPIE_PBA | E1000_GPIE_NSICR));
4588 
4589 	/*
4590 	 * Set vector for tx ring 0 and other causes.
4591 	 * NOTE assumption that it is vector 0.
4592 	 * This is also interdependent with installation of interrupt service
4593 	 * routines in igb_add_intr_handlers().
4594 	 */
4595 
4596 	/* assign "other" causes to vector 0 */
4597 	vector = 0;
4598 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4599 	E1000_WRITE_REG(hw, E1000_IVAR_MISC, ivar);
4600 
4601 	/* assign tx ring 0 to vector 0 */
4602 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4603 	E1000_WRITE_REG(hw, E1000_IVAR0, ivar);
4604 
4605 	/* prepare to enable tx & other interrupt causes */
4606 	igb->eims_mask = (1 << vector);
4607 
4608 	vector ++;
4609 	for (i = 0; i < igb->num_rx_rings; i++) {
4610 		/*
4611 		 * Set vector for each rx ring
4612 		 */
4613 		index = (i & 0x7);
4614 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4615 
4616 		if (i < 8) {
4617 			/* vector goes into low byte of register */
4618 			ivar = ivar & 0xFFFFFF00;
4619 			ivar |= (vector | E1000_IVAR_VALID);
4620 		} else {
4621 			/* vector goes into third byte of register */
4622 			ivar = ivar & 0xFF00FFFF;
4623 			ivar |= ((vector | E1000_IVAR_VALID) << 16);
4624 		}
4625 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4626 
4627 		/* Accumulate interrupt-cause bits to enable */
4628 		igb->eims_mask |= (1 << vector);
4629 
4630 		vector ++;
4631 	}
4632 
4633 	for (i = 1; i < igb->num_tx_rings; i++) {
4634 		/*
4635 		 * Set vector for each tx ring from 2nd tx ring.
4636 		 * Note assumption that tx vectors numericall follow rx vectors.
4637 		 */
4638 		index = (i & 0x7);
4639 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4640 
4641 		if (i < 8) {
4642 			/* vector goes into second byte of register */
4643 			ivar = ivar & 0xFFFF00FF;
4644 			ivar |= ((vector | E1000_IVAR_VALID) << 8);
4645 		} else {
4646 			/* vector goes into fourth byte of register */
4647 			ivar = ivar & 0x00FFFFFF;
4648 			ivar |= (vector | E1000_IVAR_VALID) << 24;
4649 		}
4650 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4651 
4652 		/* Accumulate interrupt-cause bits to enable */
4653 		igb->eims_mask |= (1 << vector);
4654 
4655 		vector ++;
4656 	}
4657 
4658 	ASSERT(vector == igb->intr_cnt);
4659 }
4660 
4661 /*
4662  * igb_setup_msix_82580 - setup 82580 adapter to use MSI-X interrupts
4663  *
4664  * 82580 uses same table approach at 82576 but has fewer entries.  Each
4665  * queue has a single entry in the table to which we write a vector number
4666  * along with a "valid" bit.  Vectors take a different position in the
4667  * register depending on * whether * they are numbered above or below 4.
4668  */
4669 static void
4670 igb_setup_msix_82580(igb_t *igb)
4671 {
4672 	struct e1000_hw *hw = &igb->hw;
4673 	uint32_t ivar, index, vector;
4674 	int i;
4675 
4676 	/* must enable msi-x capability before IVAR settings */
4677 	E1000_WRITE_REG(hw, E1000_GPIE, (E1000_GPIE_MSIX_MODE |
4678 	    E1000_GPIE_PBA | E1000_GPIE_NSICR | E1000_GPIE_EIAME));
4679 	/*
4680 	 * Set vector for tx ring 0 and other causes.
4681 	 * NOTE assumption that it is vector 0.
4682 	 * This is also interdependent with installation of interrupt service
4683 	 * routines in igb_add_intr_handlers().
4684 	 */
4685 
4686 	/* assign "other" causes to vector 0 */
4687 	vector = 0;
4688 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4689 	E1000_WRITE_REG(hw, E1000_IVAR_MISC, ivar);
4690 
4691 	/* assign tx ring 0 to vector 0 */
4692 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4693 	E1000_WRITE_REG(hw, E1000_IVAR0, ivar);
4694 
4695 	/* prepare to enable tx & other interrupt causes */
4696 	igb->eims_mask = (1 << vector);
4697 
4698 	vector ++;
4699 
4700 	for (i = 0; i < igb->num_rx_rings; i++) {
4701 		/*
4702 		 * Set vector for each rx ring
4703 		 */
4704 		index = (i >> 1);
4705 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4706 
4707 		if (i & 1) {
4708 			/* vector goes into third byte of register */
4709 			ivar = ivar & 0xFF00FFFF;
4710 			ivar |= ((vector | E1000_IVAR_VALID) << 16);
4711 		} else {
4712 			/* vector goes into low byte of register */
4713 			ivar = ivar & 0xFFFFFF00;
4714 			ivar |= (vector | E1000_IVAR_VALID);
4715 		}
4716 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4717 
4718 		/* Accumulate interrupt-cause bits to enable */
4719 		igb->eims_mask |= (1 << vector);
4720 
4721 		vector ++;
4722 	}
4723 
4724 	for (i = 1; i < igb->num_tx_rings; i++) {
4725 		/*
4726 		 * Set vector for each tx ring from 2nd tx ring.
4727 		 * Note assumption that tx vectors numericall follow rx vectors.
4728 		 */
4729 		index = (i >> 1);
4730 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4731 
4732 		if (i & 1) {
4733 			/* vector goes into high byte of register */
4734 			ivar = ivar & 0x00FFFFFF;
4735 			ivar |= ((vector | E1000_IVAR_VALID) << 24);
4736 		} else {
4737 			/* vector goes into second byte of register */
4738 			ivar = ivar & 0xFFFF00FF;
4739 			ivar |= (vector | E1000_IVAR_VALID) << 8;
4740 		}
4741 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4742 
4743 		/* Accumulate interrupt-cause bits to enable */
4744 		igb->eims_mask |= (1 << vector);
4745 
4746 		vector ++;
4747 	}
4748 	ASSERT(vector == igb->intr_cnt);
4749 }
4750 
4751 /*
4752  * igb_rem_intr_handlers - remove the interrupt handlers
4753  */
4754 static void
4755 igb_rem_intr_handlers(igb_t *igb)
4756 {
4757 	int i;
4758 	int rc;
4759 
4760 	for (i = 0; i < igb->intr_cnt; i++) {
4761 		rc = ddi_intr_remove_handler(igb->htable[i]);
4762 		if (rc != DDI_SUCCESS) {
4763 			IGB_DEBUGLOG_1(igb,
4764 			    "Remove intr handler failed: %d", rc);
4765 		}
4766 	}
4767 }
4768 
4769 /*
4770  * igb_rem_intrs - remove the allocated interrupts
4771  */
4772 static void
4773 igb_rem_intrs(igb_t *igb)
4774 {
4775 	int i;
4776 	int rc;
4777 
4778 	for (i = 0; i < igb->intr_cnt; i++) {
4779 		rc = ddi_intr_free(igb->htable[i]);
4780 		if (rc != DDI_SUCCESS) {
4781 			IGB_DEBUGLOG_1(igb,
4782 			    "Free intr failed: %d", rc);
4783 		}
4784 	}
4785 
4786 	kmem_free(igb->htable, igb->intr_size);
4787 	igb->htable = NULL;
4788 }
4789 
4790 /*
4791  * igb_enable_intrs - enable all the ddi interrupts
4792  */
4793 static int
4794 igb_enable_intrs(igb_t *igb)
4795 {
4796 	int i;
4797 	int rc;
4798 
4799 	/* Enable interrupts */
4800 	if (igb->intr_cap & DDI_INTR_FLAG_BLOCK) {
4801 		/* Call ddi_intr_block_enable() for MSI */
4802 		rc = ddi_intr_block_enable(igb->htable, igb->intr_cnt);
4803 		if (rc != DDI_SUCCESS) {
4804 			igb_log(igb,
4805 			    "Enable block intr failed: %d", rc);
4806 			return (IGB_FAILURE);
4807 		}
4808 	} else {
4809 		/* Call ddi_intr_enable() for Legacy/MSI non block enable */
4810 		for (i = 0; i < igb->intr_cnt; i++) {
4811 			rc = ddi_intr_enable(igb->htable[i]);
4812 			if (rc != DDI_SUCCESS) {
4813 				igb_log(igb,
4814 				    "Enable intr failed: %d", rc);
4815 				return (IGB_FAILURE);
4816 			}
4817 		}
4818 	}
4819 
4820 	return (IGB_SUCCESS);
4821 }
4822 
4823 /*
4824  * igb_disable_intrs - disable all the ddi interrupts
4825  */
4826 static int
4827 igb_disable_intrs(igb_t *igb)
4828 {
4829 	int i;
4830 	int rc;
4831 
4832 	/* Disable all interrupts */
4833 	if (igb->intr_cap & DDI_INTR_FLAG_BLOCK) {
4834 		rc = ddi_intr_block_disable(igb->htable, igb->intr_cnt);
4835 		if (rc != DDI_SUCCESS) {
4836 			igb_log(igb,
4837 			    "Disable block intr failed: %d", rc);
4838 			return (IGB_FAILURE);
4839 		}
4840 	} else {
4841 		for (i = 0; i < igb->intr_cnt; i++) {
4842 			rc = ddi_intr_disable(igb->htable[i]);
4843 			if (rc != DDI_SUCCESS) {
4844 				igb_log(igb,
4845 				    "Disable intr failed: %d", rc);
4846 				return (IGB_FAILURE);
4847 			}
4848 		}
4849 	}
4850 
4851 	return (IGB_SUCCESS);
4852 }
4853 
4854 /*
4855  * igb_get_phy_state - Get and save the parameters read from PHY registers
4856  */
4857 static void
4858 igb_get_phy_state(igb_t *igb)
4859 {
4860 	struct e1000_hw *hw = &igb->hw;
4861 	uint16_t phy_ctrl;
4862 	uint16_t phy_status;
4863 	uint16_t phy_an_adv;
4864 	uint16_t phy_an_exp;
4865 	uint16_t phy_ext_status;
4866 	uint16_t phy_1000t_ctrl;
4867 	uint16_t phy_1000t_status;
4868 	uint16_t phy_lp_able;
4869 
4870 	ASSERT(mutex_owned(&igb->gen_lock));
4871 
4872 	(void) e1000_read_phy_reg(hw, PHY_CONTROL, &phy_ctrl);
4873 	(void) e1000_read_phy_reg(hw, PHY_STATUS, &phy_status);
4874 	(void) e1000_read_phy_reg(hw, PHY_AUTONEG_ADV, &phy_an_adv);
4875 	(void) e1000_read_phy_reg(hw, PHY_AUTONEG_EXP, &phy_an_exp);
4876 	(void) e1000_read_phy_reg(hw, PHY_EXT_STATUS, &phy_ext_status);
4877 	(void) e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_1000t_ctrl);
4878 	(void) e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_1000t_status);
4879 	(void) e1000_read_phy_reg(hw, PHY_LP_ABILITY, &phy_lp_able);
4880 
4881 	igb->param_autoneg_cap =
4882 	    (phy_status & MII_SR_AUTONEG_CAPS) ? 1 : 0;
4883 	igb->param_pause_cap =
4884 	    (phy_an_adv & NWAY_AR_PAUSE) ? 1 : 0;
4885 	igb->param_asym_pause_cap =
4886 	    (phy_an_adv & NWAY_AR_ASM_DIR) ? 1 : 0;
4887 	igb->param_1000fdx_cap = ((phy_ext_status & IEEE_ESR_1000T_FD_CAPS) ||
4888 	    (phy_ext_status & IEEE_ESR_1000X_FD_CAPS)) ? 1 : 0;
4889 	igb->param_1000hdx_cap = ((phy_ext_status & IEEE_ESR_1000T_HD_CAPS) ||
4890 	    (phy_ext_status & IEEE_ESR_1000X_HD_CAPS)) ? 1 : 0;
4891 	igb->param_100t4_cap =
4892 	    (phy_status & MII_SR_100T4_CAPS) ? 1 : 0;
4893 	igb->param_100fdx_cap = ((phy_status & MII_SR_100X_FD_CAPS) ||
4894 	    (phy_status & MII_SR_100T2_FD_CAPS)) ? 1 : 0;
4895 	igb->param_100hdx_cap = ((phy_status & MII_SR_100X_HD_CAPS) ||
4896 	    (phy_status & MII_SR_100T2_HD_CAPS)) ? 1 : 0;
4897 	igb->param_10fdx_cap =
4898 	    (phy_status & MII_SR_10T_FD_CAPS) ? 1 : 0;
4899 	igb->param_10hdx_cap =
4900 	    (phy_status & MII_SR_10T_HD_CAPS) ? 1 : 0;
4901 	igb->param_rem_fault =
4902 	    (phy_status & MII_SR_REMOTE_FAULT) ? 1 : 0;
4903 
4904 	igb->param_adv_autoneg_cap = hw->mac.autoneg;
4905 	igb->param_adv_pause_cap =
4906 	    (phy_an_adv & NWAY_AR_PAUSE) ? 1 : 0;
4907 	igb->param_adv_asym_pause_cap =
4908 	    (phy_an_adv & NWAY_AR_ASM_DIR) ? 1 : 0;
4909 	igb->param_adv_1000hdx_cap =
4910 	    (phy_1000t_ctrl & CR_1000T_HD_CAPS) ? 1 : 0;
4911 	igb->param_adv_100t4_cap =
4912 	    (phy_an_adv & NWAY_AR_100T4_CAPS) ? 1 : 0;
4913 	igb->param_adv_rem_fault =
4914 	    (phy_an_adv & NWAY_AR_REMOTE_FAULT) ? 1 : 0;
4915 	if (igb->param_adv_autoneg_cap == 1) {
4916 		igb->param_adv_1000fdx_cap =
4917 		    (phy_1000t_ctrl & CR_1000T_FD_CAPS) ? 1 : 0;
4918 		igb->param_adv_100fdx_cap =
4919 		    (phy_an_adv & NWAY_AR_100TX_FD_CAPS) ? 1 : 0;
4920 		igb->param_adv_100hdx_cap =
4921 		    (phy_an_adv & NWAY_AR_100TX_HD_CAPS) ? 1 : 0;
4922 		igb->param_adv_10fdx_cap =
4923 		    (phy_an_adv & NWAY_AR_10T_FD_CAPS) ? 1 : 0;
4924 		igb->param_adv_10hdx_cap =
4925 		    (phy_an_adv & NWAY_AR_10T_HD_CAPS) ? 1 : 0;
4926 	}
4927 
4928 	igb->param_lp_autoneg_cap =
4929 	    (phy_an_exp & NWAY_ER_LP_NWAY_CAPS) ? 1 : 0;
4930 	igb->param_lp_pause_cap =
4931 	    (phy_lp_able & NWAY_LPAR_PAUSE) ? 1 : 0;
4932 	igb->param_lp_asym_pause_cap =
4933 	    (phy_lp_able & NWAY_LPAR_ASM_DIR) ? 1 : 0;
4934 	igb->param_lp_1000fdx_cap =
4935 	    (phy_1000t_status & SR_1000T_LP_FD_CAPS) ? 1 : 0;
4936 	igb->param_lp_1000hdx_cap =
4937 	    (phy_1000t_status & SR_1000T_LP_HD_CAPS) ? 1 : 0;
4938 	igb->param_lp_100t4_cap =
4939 	    (phy_lp_able & NWAY_LPAR_100T4_CAPS) ? 1 : 0;
4940 	igb->param_lp_100fdx_cap =
4941 	    (phy_lp_able & NWAY_LPAR_100TX_FD_CAPS) ? 1 : 0;
4942 	igb->param_lp_100hdx_cap =
4943 	    (phy_lp_able & NWAY_LPAR_100TX_HD_CAPS) ? 1 : 0;
4944 	igb->param_lp_10fdx_cap =
4945 	    (phy_lp_able & NWAY_LPAR_10T_FD_CAPS) ? 1 : 0;
4946 	igb->param_lp_10hdx_cap =
4947 	    (phy_lp_able & NWAY_LPAR_10T_HD_CAPS) ? 1 : 0;
4948 	igb->param_lp_rem_fault =
4949 	    (phy_lp_able & NWAY_LPAR_REMOTE_FAULT) ? 1 : 0;
4950 }
4951 
4952 /*
4953  * igb_get_driver_control
4954  */
4955 static void
4956 igb_get_driver_control(struct e1000_hw *hw)
4957 {
4958 	uint32_t ctrl_ext;
4959 
4960 	/* Notify firmware that driver is in control of device */
4961 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
4962 	ctrl_ext |= E1000_CTRL_EXT_DRV_LOAD;
4963 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
4964 }
4965 
4966 /*
4967  * igb_release_driver_control
4968  */
4969 static void
4970 igb_release_driver_control(struct e1000_hw *hw)
4971 {
4972 	uint32_t ctrl_ext;
4973 
4974 	/* Notify firmware that driver is no longer in control of device */
4975 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
4976 	ctrl_ext &= ~E1000_CTRL_EXT_DRV_LOAD;
4977 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
4978 }
4979 
4980 /*
4981  * igb_atomic_reserve - Atomic decrease operation
4982  */
4983 int
4984 igb_atomic_reserve(uint32_t *count_p, uint32_t n)
4985 {
4986 	uint32_t oldval;
4987 	uint32_t newval;
4988 
4989 	/* ATOMICALLY */
4990 	do {
4991 		oldval = *count_p;
4992 		if (oldval < n)
4993 			return (-1);
4994 		newval = oldval - n;
4995 	} while (atomic_cas_32(count_p, oldval, newval) != oldval);
4996 
4997 	return (newval);
4998 }
4999 
5000 /*
5001  * FMA support
5002  */
5003 
5004 int
5005 igb_check_acc_handle(ddi_acc_handle_t handle)
5006 {
5007 	ddi_fm_error_t de;
5008 
5009 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
5010 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
5011 	return (de.fme_status);
5012 }
5013 
5014 int
5015 igb_check_dma_handle(ddi_dma_handle_t handle)
5016 {
5017 	ddi_fm_error_t de;
5018 
5019 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
5020 	return (de.fme_status);
5021 }
5022 
5023 /*
5024  * The IO fault service error handling callback function
5025  */
5026 /*ARGSUSED*/
5027 static int
5028 igb_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
5029 {
5030 	/*
5031 	 * as the driver can always deal with an error in any dma or
5032 	 * access handle, we can just return the fme_status value.
5033 	 */
5034 	pci_ereport_post(dip, err, NULL);
5035 	return (err->fme_status);
5036 }
5037 
5038 static void
5039 igb_fm_init(igb_t *igb)
5040 {
5041 	ddi_iblock_cookie_t iblk;
5042 	int fma_dma_flag;
5043 
5044 	/* Only register with IO Fault Services if we have some capability */
5045 	if (igb->fm_capabilities & DDI_FM_ACCCHK_CAPABLE) {
5046 		igb_regs_acc_attr.devacc_attr_access = DDI_FLAGERR_ACC;
5047 	} else {
5048 		igb_regs_acc_attr.devacc_attr_access = DDI_DEFAULT_ACC;
5049 	}
5050 
5051 	if (igb->fm_capabilities & DDI_FM_DMACHK_CAPABLE) {
5052 		fma_dma_flag = 1;
5053 	} else {
5054 		fma_dma_flag = 0;
5055 	}
5056 
5057 	(void) igb_set_fma_flags(fma_dma_flag);
5058 
5059 	if (igb->fm_capabilities) {
5060 
5061 		/* Register capabilities with IO Fault Services */
5062 		ddi_fm_init(igb->dip, &igb->fm_capabilities, &iblk);
5063 
5064 		/*
5065 		 * Initialize pci ereport capabilities if ereport capable
5066 		 */
5067 		if (DDI_FM_EREPORT_CAP(igb->fm_capabilities) ||
5068 		    DDI_FM_ERRCB_CAP(igb->fm_capabilities))
5069 			pci_ereport_setup(igb->dip);
5070 
5071 		/*
5072 		 * Register error callback if error callback capable
5073 		 */
5074 		if (DDI_FM_ERRCB_CAP(igb->fm_capabilities))
5075 			ddi_fm_handler_register(igb->dip,
5076 			    igb_fm_error_cb, (void*) igb);
5077 	}
5078 }
5079 
5080 static void
5081 igb_fm_fini(igb_t *igb)
5082 {
5083 	/* Only unregister FMA capabilities if we registered some */
5084 	if (igb->fm_capabilities) {
5085 
5086 		/*
5087 		 * Release any resources allocated by pci_ereport_setup()
5088 		 */
5089 		if (DDI_FM_EREPORT_CAP(igb->fm_capabilities) ||
5090 		    DDI_FM_ERRCB_CAP(igb->fm_capabilities))
5091 			pci_ereport_teardown(igb->dip);
5092 
5093 		/*
5094 		 * Un-register error callback if error callback capable
5095 		 */
5096 		if (DDI_FM_ERRCB_CAP(igb->fm_capabilities))
5097 			ddi_fm_handler_unregister(igb->dip);
5098 
5099 		/* Unregister from IO Fault Services */
5100 		ddi_fm_fini(igb->dip);
5101 	}
5102 }
5103 
5104 void
5105 igb_fm_ereport(igb_t *igb, char *detail)
5106 {
5107 	uint64_t ena;
5108 	char buf[FM_MAX_CLASS];
5109 
5110 	(void) snprintf(buf, FM_MAX_CLASS, "%s.%s", DDI_FM_DEVICE, detail);
5111 	ena = fm_ena_generate(0, FM_ENA_FMT1);
5112 	if (DDI_FM_EREPORT_CAP(igb->fm_capabilities)) {
5113 		ddi_fm_ereport_post(igb->dip, buf, ena, DDI_NOSLEEP,
5114 		    FM_VERSION, DATA_TYPE_UINT8, FM_EREPORT_VERS0, NULL);
5115 	}
5116 }
5117