xref: /titanic_51/usr/src/uts/common/io/igb/igb_main.c (revision 597bd30ba830d1e11c8efdb9a1b9de28e0599f5e)
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 2010 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 1.1.12";
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 int igb_alloc_rx_data(igb_t *);
54 static void igb_free_rx_data(igb_t *);
55 static void igb_free_rings(igb_t *);
56 static void igb_setup_rings(igb_t *);
57 static void igb_setup_rx(igb_t *);
58 static void igb_setup_tx(igb_t *);
59 static void igb_setup_rx_ring(igb_rx_ring_t *);
60 static void igb_setup_tx_ring(igb_tx_ring_t *);
61 static void igb_setup_rss(igb_t *);
62 static void igb_setup_mac_rss_classify(igb_t *);
63 static void igb_setup_mac_classify(igb_t *);
64 static void igb_init_unicst(igb_t *);
65 static void igb_setup_multicst(igb_t *);
66 static void igb_get_phy_state(igb_t *);
67 static void igb_param_sync(igb_t *);
68 static void igb_get_conf(igb_t *);
69 static int igb_get_prop(igb_t *, char *, int, int, int);
70 static boolean_t igb_is_link_up(igb_t *);
71 static boolean_t igb_link_check(igb_t *);
72 static void igb_local_timer(void *);
73 static void igb_link_timer(void *);
74 static void igb_arm_watchdog_timer(igb_t *);
75 static void igb_start_watchdog_timer(igb_t *);
76 static void igb_restart_watchdog_timer(igb_t *);
77 static void igb_stop_watchdog_timer(igb_t *);
78 static void igb_start_link_timer(igb_t *);
79 static void igb_stop_link_timer(igb_t *);
80 static void igb_disable_adapter_interrupts(igb_t *);
81 static void igb_enable_adapter_interrupts_82575(igb_t *);
82 static void igb_enable_adapter_interrupts_82576(igb_t *);
83 static void igb_enable_adapter_interrupts_82580(igb_t *);
84 static boolean_t is_valid_mac_addr(uint8_t *);
85 static boolean_t igb_stall_check(igb_t *);
86 static boolean_t igb_set_loopback_mode(igb_t *, uint32_t);
87 static void igb_set_external_loopback(igb_t *);
88 static void igb_set_internal_mac_loopback(igb_t *);
89 static void igb_set_internal_phy_loopback(igb_t *);
90 static void igb_set_internal_serdes_loopback(igb_t *);
91 static boolean_t igb_find_mac_address(igb_t *);
92 static int igb_alloc_intrs(igb_t *);
93 static int igb_alloc_intr_handles(igb_t *, int);
94 static int igb_add_intr_handlers(igb_t *);
95 static void igb_rem_intr_handlers(igb_t *);
96 static void igb_rem_intrs(igb_t *);
97 static int igb_enable_intrs(igb_t *);
98 static int igb_disable_intrs(igb_t *);
99 static void igb_setup_msix_82575(igb_t *);
100 static void igb_setup_msix_82576(igb_t *);
101 static void igb_setup_msix_82580(igb_t *);
102 static uint_t igb_intr_legacy(void *, void *);
103 static uint_t igb_intr_msi(void *, void *);
104 static uint_t igb_intr_rx(void *, void *);
105 static uint_t igb_intr_tx(void *, void *);
106 static uint_t igb_intr_tx_other(void *, void *);
107 static void igb_intr_rx_work(igb_rx_ring_t *);
108 static void igb_intr_tx_work(igb_tx_ring_t *);
109 static void igb_intr_link_work(igb_t *);
110 static void igb_get_driver_control(struct e1000_hw *);
111 static void igb_release_driver_control(struct e1000_hw *);
112 
113 static int igb_attach(dev_info_t *, ddi_attach_cmd_t);
114 static int igb_detach(dev_info_t *, ddi_detach_cmd_t);
115 static int igb_resume(dev_info_t *);
116 static int igb_suspend(dev_info_t *);
117 static int igb_quiesce(dev_info_t *);
118 static void igb_unconfigure(dev_info_t *, igb_t *);
119 static int igb_fm_error_cb(dev_info_t *, ddi_fm_error_t *,
120     const void *);
121 static void igb_fm_init(igb_t *);
122 static void igb_fm_fini(igb_t *);
123 static void igb_release_multicast(igb_t *);
124 
125 mac_priv_prop_t igb_priv_props[] = {
126 	{"_tx_copy_thresh", MAC_PROP_PERM_RW},
127 	{"_tx_recycle_thresh", MAC_PROP_PERM_RW},
128 	{"_tx_overload_thresh", MAC_PROP_PERM_RW},
129 	{"_tx_resched_thresh", MAC_PROP_PERM_RW},
130 	{"_rx_copy_thresh", MAC_PROP_PERM_RW},
131 	{"_rx_limit_per_intr", MAC_PROP_PERM_RW},
132 	{"_intr_throttling", MAC_PROP_PERM_RW},
133 	{"_adv_pause_cap", MAC_PROP_PERM_READ},
134 	{"_adv_asym_pause_cap", MAC_PROP_PERM_READ}
135 };
136 
137 #define	IGB_MAX_PRIV_PROPS \
138 	(sizeof (igb_priv_props) / sizeof (mac_priv_prop_t))
139 
140 static struct cb_ops igb_cb_ops = {
141 	nulldev,		/* cb_open */
142 	nulldev,		/* cb_close */
143 	nodev,			/* cb_strategy */
144 	nodev,			/* cb_print */
145 	nodev,			/* cb_dump */
146 	nodev,			/* cb_read */
147 	nodev,			/* cb_write */
148 	nodev,			/* cb_ioctl */
149 	nodev,			/* cb_devmap */
150 	nodev,			/* cb_mmap */
151 	nodev,			/* cb_segmap */
152 	nochpoll,		/* cb_chpoll */
153 	ddi_prop_op,		/* cb_prop_op */
154 	NULL,			/* cb_stream */
155 	D_MP | D_HOTPLUG,	/* cb_flag */
156 	CB_REV,			/* cb_rev */
157 	nodev,			/* cb_aread */
158 	nodev			/* cb_awrite */
159 };
160 
161 static struct dev_ops igb_dev_ops = {
162 	DEVO_REV,		/* devo_rev */
163 	0,			/* devo_refcnt */
164 	NULL,			/* devo_getinfo */
165 	nulldev,		/* devo_identify */
166 	nulldev,		/* devo_probe */
167 	igb_attach,		/* devo_attach */
168 	igb_detach,		/* devo_detach */
169 	nodev,			/* devo_reset */
170 	&igb_cb_ops,		/* devo_cb_ops */
171 	NULL,			/* devo_bus_ops */
172 	ddi_power,		/* devo_power */
173 	igb_quiesce,	/* devo_quiesce */
174 };
175 
176 static struct modldrv igb_modldrv = {
177 	&mod_driverops,		/* Type of module.  This one is a driver */
178 	ident,			/* Discription string */
179 	&igb_dev_ops,		/* driver ops */
180 };
181 
182 static struct modlinkage igb_modlinkage = {
183 	MODREV_1, &igb_modldrv, NULL
184 };
185 
186 /* Access attributes for register mapping */
187 ddi_device_acc_attr_t igb_regs_acc_attr = {
188 	DDI_DEVICE_ATTR_V1,
189 	DDI_STRUCTURE_LE_ACC,
190 	DDI_STRICTORDER_ACC,
191 	DDI_FLAGERR_ACC
192 };
193 
194 #define	IGB_M_CALLBACK_FLAGS \
195 	(MC_IOCTL | MC_GETCAPAB | MC_SETPROP | MC_GETPROP)
196 
197 static mac_callbacks_t igb_m_callbacks = {
198 	IGB_M_CALLBACK_FLAGS,
199 	igb_m_stat,
200 	igb_m_start,
201 	igb_m_stop,
202 	igb_m_promisc,
203 	igb_m_multicst,
204 	NULL,
205 	NULL,
206 	igb_m_ioctl,
207 	igb_m_getcapab,
208 	NULL,
209 	NULL,
210 	igb_m_setprop,
211 	igb_m_getprop
212 };
213 
214 /*
215  * Initialize capabilities of each supported adapter type
216  */
217 static adapter_info_t igb_82575_cap = {
218 	/* limits */
219 	4,		/* maximum number of rx queues */
220 	1,		/* minimum number of rx queues */
221 	4,		/* default number of rx queues */
222 	4,		/* 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_82575,
231 	igb_setup_msix_82575,
232 
233 	/* capabilities */
234 	(IGB_FLAG_HAS_DCA |	/* capability flags */
235 	IGB_FLAG_VMDQ_POOL),
236 
237 	0xffc00000		/* mask for RXDCTL register */
238 };
239 
240 static adapter_info_t igb_82576_cap = {
241 	/* limits */
242 	16,		/* maximum number of rx queues */
243 	1,		/* minimum number of rx queues */
244 	4,		/* default number of rx queues */
245 	16,		/* maximum number of tx queues */
246 	1,		/* minimum number of tx queues */
247 	4,		/* default number of tx queues */
248 	65535,		/* maximum interrupt throttle rate */
249 	0,		/* minimum interrupt throttle rate */
250 	200,		/* default interrupt throttle rate */
251 
252 	/* function pointers */
253 	igb_enable_adapter_interrupts_82576,
254 	igb_setup_msix_82576,
255 
256 	/* capabilities */
257 	(IGB_FLAG_HAS_DCA |	/* capability flags */
258 	IGB_FLAG_VMDQ_POOL |
259 	IGB_FLAG_NEED_CTX_IDX),
260 
261 	0xffe00000		/* mask for RXDCTL register */
262 };
263 
264 static adapter_info_t igb_82580_cap = {
265 	/* limits */
266 	8,		/* maximum number of rx queues */
267 	1,		/* minimum number of rx queues */
268 	4,		/* default number of rx queues */
269 	8,		/* maximum number of tx queues */
270 	1,		/* minimum number of tx queues */
271 	4,		/* default number of tx queues */
272 	65535,		/* maximum interrupt throttle rate */
273 	0,		/* minimum interrupt throttle rate */
274 	200,		/* default interrupt throttle rate */
275 
276 	/* function pointers */
277 	igb_enable_adapter_interrupts_82580,
278 	igb_setup_msix_82580,
279 
280 	/* capabilities */
281 	(IGB_FLAG_HAS_DCA |	/* capability flags */
282 	IGB_FLAG_VMDQ_POOL |
283 	IGB_FLAG_NEED_CTX_IDX),
284 
285 	0xffe00000		/* mask for RXDCTL register */
286 };
287 
288 /*
289  * Module Initialization Functions
290  */
291 
292 int
293 _init(void)
294 {
295 	int status;
296 
297 	mac_init_ops(&igb_dev_ops, MODULE_NAME);
298 
299 	status = mod_install(&igb_modlinkage);
300 
301 	if (status != DDI_SUCCESS) {
302 		mac_fini_ops(&igb_dev_ops);
303 	}
304 
305 	return (status);
306 }
307 
308 int
309 _fini(void)
310 {
311 	int status;
312 
313 	status = mod_remove(&igb_modlinkage);
314 
315 	if (status == DDI_SUCCESS) {
316 		mac_fini_ops(&igb_dev_ops);
317 	}
318 
319 	return (status);
320 
321 }
322 
323 int
324 _info(struct modinfo *modinfop)
325 {
326 	int status;
327 
328 	status = mod_info(&igb_modlinkage, modinfop);
329 
330 	return (status);
331 }
332 
333 /*
334  * igb_attach - driver attach
335  *
336  * This function is the device specific initialization entry
337  * point. This entry point is required and must be written.
338  * The DDI_ATTACH command must be provided in the attach entry
339  * point. When attach() is called with cmd set to DDI_ATTACH,
340  * all normal kernel services (such as kmem_alloc(9F)) are
341  * available for use by the driver.
342  *
343  * The attach() function will be called once for each instance
344  * of  the  device  on  the  system with cmd set to DDI_ATTACH.
345  * Until attach() succeeds, the only driver entry points which
346  * may be called are open(9E) and getinfo(9E).
347  */
348 static int
349 igb_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
350 {
351 	igb_t *igb;
352 	struct igb_osdep *osdep;
353 	struct e1000_hw *hw;
354 	int instance;
355 
356 	/*
357 	 * Check the command and perform corresponding operations
358 	 */
359 	switch (cmd) {
360 	default:
361 		return (DDI_FAILURE);
362 
363 	case DDI_RESUME:
364 		return (igb_resume(devinfo));
365 
366 	case DDI_ATTACH:
367 		break;
368 	}
369 
370 	/* Get the device instance */
371 	instance = ddi_get_instance(devinfo);
372 
373 	/* Allocate memory for the instance data structure */
374 	igb = kmem_zalloc(sizeof (igb_t), KM_SLEEP);
375 
376 	igb->dip = devinfo;
377 	igb->instance = instance;
378 
379 	hw = &igb->hw;
380 	osdep = &igb->osdep;
381 	hw->back = osdep;
382 	osdep->igb = igb;
383 
384 	/* Attach the instance pointer to the dev_info data structure */
385 	ddi_set_driver_private(devinfo, igb);
386 
387 
388 	/* Initialize for fma support */
389 	igb->fm_capabilities = igb_get_prop(igb, "fm-capable",
390 	    0, 0x0f,
391 	    DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
392 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
393 	igb_fm_init(igb);
394 	igb->attach_progress |= ATTACH_PROGRESS_FMINIT;
395 
396 	/*
397 	 * Map PCI config space registers
398 	 */
399 	if (pci_config_setup(devinfo, &osdep->cfg_handle) != DDI_SUCCESS) {
400 		igb_error(igb, "Failed to map PCI configurations");
401 		goto attach_fail;
402 	}
403 	igb->attach_progress |= ATTACH_PROGRESS_PCI_CONFIG;
404 
405 	/*
406 	 * Identify the chipset family
407 	 */
408 	if (igb_identify_hardware(igb) != IGB_SUCCESS) {
409 		igb_error(igb, "Failed to identify hardware");
410 		goto attach_fail;
411 	}
412 
413 	/*
414 	 * Map device registers
415 	 */
416 	if (igb_regs_map(igb) != IGB_SUCCESS) {
417 		igb_error(igb, "Failed to map device registers");
418 		goto attach_fail;
419 	}
420 	igb->attach_progress |= ATTACH_PROGRESS_REGS_MAP;
421 
422 	/*
423 	 * Initialize driver parameters
424 	 */
425 	igb_init_properties(igb);
426 	igb->attach_progress |= ATTACH_PROGRESS_PROPS;
427 
428 	/*
429 	 * Allocate interrupts
430 	 */
431 	if (igb_alloc_intrs(igb) != IGB_SUCCESS) {
432 		igb_error(igb, "Failed to allocate interrupts");
433 		goto attach_fail;
434 	}
435 	igb->attach_progress |= ATTACH_PROGRESS_ALLOC_INTR;
436 
437 	/*
438 	 * Allocate rx/tx rings based on the ring numbers.
439 	 * The actual numbers of rx/tx rings are decided by the number of
440 	 * allocated interrupt vectors, so we should allocate the rings after
441 	 * interrupts are allocated.
442 	 */
443 	if (igb_alloc_rings(igb) != IGB_SUCCESS) {
444 		igb_error(igb, "Failed to allocate rx/tx rings or groups");
445 		goto attach_fail;
446 	}
447 	igb->attach_progress |= ATTACH_PROGRESS_ALLOC_RINGS;
448 
449 	/*
450 	 * Add interrupt handlers
451 	 */
452 	if (igb_add_intr_handlers(igb) != IGB_SUCCESS) {
453 		igb_error(igb, "Failed to add interrupt handlers");
454 		goto attach_fail;
455 	}
456 	igb->attach_progress |= ATTACH_PROGRESS_ADD_INTR;
457 
458 	/*
459 	 * Initialize driver parameters
460 	 */
461 	if (igb_init_driver_settings(igb) != IGB_SUCCESS) {
462 		igb_error(igb, "Failed to initialize driver settings");
463 		goto attach_fail;
464 	}
465 
466 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK) {
467 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
468 		goto attach_fail;
469 	}
470 
471 	/*
472 	 * Initialize mutexes for this device.
473 	 * Do this before enabling the interrupt handler and
474 	 * register the softint to avoid the condition where
475 	 * interrupt handler can try using uninitialized mutex
476 	 */
477 	igb_init_locks(igb);
478 	igb->attach_progress |= ATTACH_PROGRESS_LOCKS;
479 
480 	/*
481 	 * Initialize the adapter
482 	 */
483 	if (igb_init(igb) != IGB_SUCCESS) {
484 		igb_error(igb, "Failed to initialize adapter");
485 		goto attach_fail;
486 	}
487 	igb->attach_progress |= ATTACH_PROGRESS_INIT_ADAPTER;
488 
489 	/*
490 	 * Initialize statistics
491 	 */
492 	if (igb_init_stats(igb) != IGB_SUCCESS) {
493 		igb_error(igb, "Failed to initialize statistics");
494 		goto attach_fail;
495 	}
496 	igb->attach_progress |= ATTACH_PROGRESS_STATS;
497 
498 	/*
499 	 * Register the driver to the MAC
500 	 */
501 	if (igb_register_mac(igb) != IGB_SUCCESS) {
502 		igb_error(igb, "Failed to register MAC");
503 		goto attach_fail;
504 	}
505 	igb->attach_progress |= ATTACH_PROGRESS_MAC;
506 
507 	/*
508 	 * Now that mutex locks are initialized, and the chip is also
509 	 * initialized, enable interrupts.
510 	 */
511 	if (igb_enable_intrs(igb) != IGB_SUCCESS) {
512 		igb_error(igb, "Failed to enable DDI interrupts");
513 		goto attach_fail;
514 	}
515 	igb->attach_progress |= ATTACH_PROGRESS_ENABLE_INTR;
516 
517 	igb_log(igb, "%s", igb_version);
518 	atomic_or_32(&igb->igb_state, IGB_INITIALIZED);
519 
520 	return (DDI_SUCCESS);
521 
522 attach_fail:
523 	igb_unconfigure(devinfo, igb);
524 	return (DDI_FAILURE);
525 }
526 
527 /*
528  * igb_detach - driver detach
529  *
530  * The detach() function is the complement of the attach routine.
531  * If cmd is set to DDI_DETACH, detach() is used to remove  the
532  * state  associated  with  a  given  instance of a device node
533  * prior to the removal of that instance from the system.
534  *
535  * The detach() function will be called once for each  instance
536  * of the device for which there has been a successful attach()
537  * once there are no longer  any  opens  on  the  device.
538  *
539  * Interrupts routine are disabled, All memory allocated by this
540  * driver are freed.
541  */
542 static int
543 igb_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
544 {
545 	igb_t *igb;
546 
547 	/*
548 	 * Check detach command
549 	 */
550 	switch (cmd) {
551 	default:
552 		return (DDI_FAILURE);
553 
554 	case DDI_SUSPEND:
555 		return (igb_suspend(devinfo));
556 
557 	case DDI_DETACH:
558 		break;
559 	}
560 
561 
562 	/*
563 	 * Get the pointer to the driver private data structure
564 	 */
565 	igb = (igb_t *)ddi_get_driver_private(devinfo);
566 	if (igb == NULL)
567 		return (DDI_FAILURE);
568 
569 	/*
570 	 * Unregister MAC. If failed, we have to fail the detach
571 	 */
572 	if (mac_unregister(igb->mac_hdl) != 0) {
573 		igb_error(igb, "Failed to unregister MAC");
574 		return (DDI_FAILURE);
575 	}
576 	igb->attach_progress &= ~ATTACH_PROGRESS_MAC;
577 
578 	/*
579 	 * If the device is still running, it needs to be stopped first.
580 	 * This check is necessary because under some specific circumstances,
581 	 * the detach routine can be called without stopping the interface
582 	 * first.
583 	 */
584 	mutex_enter(&igb->gen_lock);
585 	if (igb->igb_state & IGB_STARTED) {
586 		atomic_and_32(&igb->igb_state, ~IGB_STARTED);
587 		igb_stop(igb, B_TRUE);
588 		mutex_exit(&igb->gen_lock);
589 		/* Disable and stop the watchdog timer */
590 		igb_disable_watchdog_timer(igb);
591 	} else
592 		mutex_exit(&igb->gen_lock);
593 
594 	/*
595 	 * Check if there are still rx buffers held by the upper layer.
596 	 * If so, fail the detach.
597 	 */
598 	if (!igb_rx_drain(igb))
599 		return (DDI_FAILURE);
600 
601 	/*
602 	 * Do the remaining unconfigure routines
603 	 */
604 	igb_unconfigure(devinfo, igb);
605 
606 	return (DDI_SUCCESS);
607 }
608 
609 /*
610  * quiesce(9E) entry point.
611  *
612  * This function is called when the system is single-threaded at high
613  * PIL with preemption disabled. Therefore, this function must not be
614  * blocked.
615  *
616  * This function returns DDI_SUCCESS on success, or DDI_FAILURE on failure.
617  * DDI_FAILURE indicates an error condition and should almost never happen.
618  */
619 static int
620 igb_quiesce(dev_info_t *devinfo)
621 {
622 	igb_t *igb;
623 	struct e1000_hw *hw;
624 
625 	igb = (igb_t *)ddi_get_driver_private(devinfo);
626 
627 	if (igb == NULL)
628 		return (DDI_FAILURE);
629 
630 	hw = &igb->hw;
631 
632 	/*
633 	 * Disable the adapter interrupts
634 	 */
635 	igb_disable_adapter_interrupts(igb);
636 
637 	/* Tell firmware driver is no longer in control */
638 	igb_release_driver_control(hw);
639 
640 	/*
641 	 * Reset the chipset
642 	 */
643 	(void) e1000_reset_hw(hw);
644 
645 	/*
646 	 * Reset PHY if possible
647 	 */
648 	if (e1000_check_reset_block(hw) == E1000_SUCCESS)
649 		(void) e1000_phy_hw_reset(hw);
650 
651 	return (DDI_SUCCESS);
652 }
653 
654 /*
655  * igb_unconfigure - release all resources held by this instance
656  */
657 static void
658 igb_unconfigure(dev_info_t *devinfo, igb_t *igb)
659 {
660 	/*
661 	 * Disable interrupt
662 	 */
663 	if (igb->attach_progress & ATTACH_PROGRESS_ENABLE_INTR) {
664 		(void) igb_disable_intrs(igb);
665 	}
666 
667 	/*
668 	 * Unregister MAC
669 	 */
670 	if (igb->attach_progress & ATTACH_PROGRESS_MAC) {
671 		(void) mac_unregister(igb->mac_hdl);
672 	}
673 
674 	/*
675 	 * Free statistics
676 	 */
677 	if (igb->attach_progress & ATTACH_PROGRESS_STATS) {
678 		kstat_delete((kstat_t *)igb->igb_ks);
679 	}
680 
681 	/*
682 	 * Remove interrupt handlers
683 	 */
684 	if (igb->attach_progress & ATTACH_PROGRESS_ADD_INTR) {
685 		igb_rem_intr_handlers(igb);
686 	}
687 
688 	/*
689 	 * Remove interrupts
690 	 */
691 	if (igb->attach_progress & ATTACH_PROGRESS_ALLOC_INTR) {
692 		igb_rem_intrs(igb);
693 	}
694 
695 	/*
696 	 * Remove driver properties
697 	 */
698 	if (igb->attach_progress & ATTACH_PROGRESS_PROPS) {
699 		(void) ddi_prop_remove_all(devinfo);
700 	}
701 
702 	/*
703 	 * Stop the adapter
704 	 */
705 	if (igb->attach_progress & ATTACH_PROGRESS_INIT_ADAPTER) {
706 		mutex_enter(&igb->gen_lock);
707 		igb_stop_adapter(igb);
708 		mutex_exit(&igb->gen_lock);
709 		if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
710 			ddi_fm_service_impact(igb->dip, DDI_SERVICE_UNAFFECTED);
711 	}
712 
713 	/*
714 	 * Free multicast table
715 	 */
716 	igb_release_multicast(igb);
717 
718 	/*
719 	 * Free register handle
720 	 */
721 	if (igb->attach_progress & ATTACH_PROGRESS_REGS_MAP) {
722 		if (igb->osdep.reg_handle != NULL)
723 			ddi_regs_map_free(&igb->osdep.reg_handle);
724 	}
725 
726 	/*
727 	 * Free PCI config handle
728 	 */
729 	if (igb->attach_progress & ATTACH_PROGRESS_PCI_CONFIG) {
730 		if (igb->osdep.cfg_handle != NULL)
731 			pci_config_teardown(&igb->osdep.cfg_handle);
732 	}
733 
734 	/*
735 	 * Free locks
736 	 */
737 	if (igb->attach_progress & ATTACH_PROGRESS_LOCKS) {
738 		igb_destroy_locks(igb);
739 	}
740 
741 	/*
742 	 * Free the rx/tx rings
743 	 */
744 	if (igb->attach_progress & ATTACH_PROGRESS_ALLOC_RINGS) {
745 		igb_free_rings(igb);
746 	}
747 
748 	/*
749 	 * Remove FMA
750 	 */
751 	if (igb->attach_progress & ATTACH_PROGRESS_FMINIT) {
752 		igb_fm_fini(igb);
753 	}
754 
755 	/*
756 	 * Free the driver data structure
757 	 */
758 	kmem_free(igb, sizeof (igb_t));
759 
760 	ddi_set_driver_private(devinfo, NULL);
761 }
762 
763 /*
764  * igb_register_mac - Register the driver and its function pointers with
765  * the GLD interface
766  */
767 static int
768 igb_register_mac(igb_t *igb)
769 {
770 	struct e1000_hw *hw = &igb->hw;
771 	mac_register_t *mac;
772 	int status;
773 
774 	if ((mac = mac_alloc(MAC_VERSION)) == NULL)
775 		return (IGB_FAILURE);
776 
777 	mac->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
778 	mac->m_driver = igb;
779 	mac->m_dip = igb->dip;
780 	mac->m_src_addr = hw->mac.addr;
781 	mac->m_callbacks = &igb_m_callbacks;
782 	mac->m_min_sdu = 0;
783 	mac->m_max_sdu = igb->max_frame_size -
784 	    sizeof (struct ether_vlan_header) - ETHERFCSL;
785 	mac->m_margin = VLAN_TAGSZ;
786 	mac->m_priv_props = igb_priv_props;
787 	mac->m_priv_prop_count = IGB_MAX_PRIV_PROPS;
788 	mac->m_v12n = MAC_VIRT_LEVEL1;
789 
790 	status = mac_register(mac, &igb->mac_hdl);
791 
792 	mac_free(mac);
793 
794 	return ((status == 0) ? IGB_SUCCESS : IGB_FAILURE);
795 }
796 
797 /*
798  * igb_identify_hardware - Identify the type of the chipset
799  */
800 static int
801 igb_identify_hardware(igb_t *igb)
802 {
803 	struct e1000_hw *hw = &igb->hw;
804 	struct igb_osdep *osdep = &igb->osdep;
805 
806 	/*
807 	 * Get the device id
808 	 */
809 	hw->vendor_id =
810 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_VENID);
811 	hw->device_id =
812 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_DEVID);
813 	hw->revision_id =
814 	    pci_config_get8(osdep->cfg_handle, PCI_CONF_REVID);
815 	hw->subsystem_device_id =
816 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBSYSID);
817 	hw->subsystem_vendor_id =
818 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBVENID);
819 
820 	/*
821 	 * Set the mac type of the adapter based on the device id
822 	 */
823 	if (e1000_set_mac_type(hw) != E1000_SUCCESS) {
824 		return (IGB_FAILURE);
825 	}
826 
827 	/*
828 	 * Install adapter capabilities based on mac type
829 	 */
830 	switch (hw->mac.type) {
831 	case e1000_82575:
832 		igb->capab = &igb_82575_cap;
833 		break;
834 	case e1000_82576:
835 		igb->capab = &igb_82576_cap;
836 		break;
837 	case e1000_82580:
838 		igb->capab = &igb_82580_cap;
839 		break;
840 	default:
841 		return (IGB_FAILURE);
842 	}
843 
844 	return (IGB_SUCCESS);
845 }
846 
847 /*
848  * igb_regs_map - Map the device registers
849  */
850 static int
851 igb_regs_map(igb_t *igb)
852 {
853 	dev_info_t *devinfo = igb->dip;
854 	struct e1000_hw *hw = &igb->hw;
855 	struct igb_osdep *osdep = &igb->osdep;
856 	off_t mem_size;
857 
858 	/*
859 	 * First get the size of device registers to be mapped.
860 	 */
861 	if (ddi_dev_regsize(devinfo, IGB_ADAPTER_REGSET, &mem_size) !=
862 	    DDI_SUCCESS) {
863 		return (IGB_FAILURE);
864 	}
865 
866 	/*
867 	 * Call ddi_regs_map_setup() to map registers
868 	 */
869 	if ((ddi_regs_map_setup(devinfo, IGB_ADAPTER_REGSET,
870 	    (caddr_t *)&hw->hw_addr, 0,
871 	    mem_size, &igb_regs_acc_attr,
872 	    &osdep->reg_handle)) != DDI_SUCCESS) {
873 		return (IGB_FAILURE);
874 	}
875 
876 	return (IGB_SUCCESS);
877 }
878 
879 /*
880  * igb_init_properties - Initialize driver properties
881  */
882 static void
883 igb_init_properties(igb_t *igb)
884 {
885 	/*
886 	 * Get conf file properties, including link settings
887 	 * jumbo frames, ring number, descriptor number, etc.
888 	 */
889 	igb_get_conf(igb);
890 }
891 
892 /*
893  * igb_init_driver_settings - Initialize driver settings
894  *
895  * The settings include hardware function pointers, bus information,
896  * rx/tx rings settings, link state, and any other parameters that
897  * need to be setup during driver initialization.
898  */
899 static int
900 igb_init_driver_settings(igb_t *igb)
901 {
902 	struct e1000_hw *hw = &igb->hw;
903 	igb_rx_ring_t *rx_ring;
904 	igb_tx_ring_t *tx_ring;
905 	uint32_t rx_size;
906 	uint32_t tx_size;
907 	int i;
908 
909 	/*
910 	 * Initialize chipset specific hardware function pointers
911 	 */
912 	if (e1000_setup_init_funcs(hw, B_TRUE) != E1000_SUCCESS) {
913 		return (IGB_FAILURE);
914 	}
915 
916 	/*
917 	 * Get bus information
918 	 */
919 	if (e1000_get_bus_info(hw) != E1000_SUCCESS) {
920 		return (IGB_FAILURE);
921 	}
922 
923 	/*
924 	 * Get the system page size
925 	 */
926 	igb->page_size = ddi_ptob(igb->dip, (ulong_t)1);
927 
928 	/*
929 	 * Set rx buffer size
930 	 * The IP header alignment room is counted in the calculation.
931 	 * The rx buffer size is in unit of 1K that is required by the
932 	 * chipset hardware.
933 	 */
934 	rx_size = igb->max_frame_size + IPHDR_ALIGN_ROOM;
935 	igb->rx_buf_size = ((rx_size >> 10) +
936 	    ((rx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
937 
938 	/*
939 	 * Set tx buffer size
940 	 */
941 	tx_size = igb->max_frame_size;
942 	igb->tx_buf_size = ((tx_size >> 10) +
943 	    ((tx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
944 
945 	/*
946 	 * Initialize rx/tx rings parameters
947 	 */
948 	for (i = 0; i < igb->num_rx_rings; i++) {
949 		rx_ring = &igb->rx_rings[i];
950 		rx_ring->index = i;
951 		rx_ring->igb = igb;
952 	}
953 
954 	for (i = 0; i < igb->num_tx_rings; i++) {
955 		tx_ring = &igb->tx_rings[i];
956 		tx_ring->index = i;
957 		tx_ring->igb = igb;
958 		if (igb->tx_head_wb_enable)
959 			tx_ring->tx_recycle = igb_tx_recycle_head_wb;
960 		else
961 			tx_ring->tx_recycle = igb_tx_recycle_legacy;
962 
963 		tx_ring->ring_size = igb->tx_ring_size;
964 		tx_ring->free_list_size = igb->tx_ring_size +
965 		    (igb->tx_ring_size >> 1);
966 	}
967 
968 	/*
969 	 * Initialize values of interrupt throttling rates
970 	 */
971 	for (i = 1; i < MAX_NUM_EITR; i++)
972 		igb->intr_throttling[i] = igb->intr_throttling[0];
973 
974 	/*
975 	 * The initial link state should be "unknown"
976 	 */
977 	igb->link_state = LINK_STATE_UNKNOWN;
978 
979 	return (IGB_SUCCESS);
980 }
981 
982 /*
983  * igb_init_locks - Initialize locks
984  */
985 static void
986 igb_init_locks(igb_t *igb)
987 {
988 	igb_rx_ring_t *rx_ring;
989 	igb_tx_ring_t *tx_ring;
990 	int i;
991 
992 	for (i = 0; i < igb->num_rx_rings; i++) {
993 		rx_ring = &igb->rx_rings[i];
994 		mutex_init(&rx_ring->rx_lock, NULL,
995 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
996 	}
997 
998 	for (i = 0; i < igb->num_tx_rings; i++) {
999 		tx_ring = &igb->tx_rings[i];
1000 		mutex_init(&tx_ring->tx_lock, NULL,
1001 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1002 		mutex_init(&tx_ring->recycle_lock, NULL,
1003 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1004 		mutex_init(&tx_ring->tcb_head_lock, NULL,
1005 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1006 		mutex_init(&tx_ring->tcb_tail_lock, NULL,
1007 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1008 	}
1009 
1010 	mutex_init(&igb->gen_lock, NULL,
1011 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1012 
1013 	mutex_init(&igb->watchdog_lock, NULL,
1014 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1015 
1016 	mutex_init(&igb->link_lock, NULL,
1017 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1018 }
1019 
1020 /*
1021  * igb_destroy_locks - Destroy locks
1022  */
1023 static void
1024 igb_destroy_locks(igb_t *igb)
1025 {
1026 	igb_rx_ring_t *rx_ring;
1027 	igb_tx_ring_t *tx_ring;
1028 	int i;
1029 
1030 	for (i = 0; i < igb->num_rx_rings; i++) {
1031 		rx_ring = &igb->rx_rings[i];
1032 		mutex_destroy(&rx_ring->rx_lock);
1033 	}
1034 
1035 	for (i = 0; i < igb->num_tx_rings; i++) {
1036 		tx_ring = &igb->tx_rings[i];
1037 		mutex_destroy(&tx_ring->tx_lock);
1038 		mutex_destroy(&tx_ring->recycle_lock);
1039 		mutex_destroy(&tx_ring->tcb_head_lock);
1040 		mutex_destroy(&tx_ring->tcb_tail_lock);
1041 	}
1042 
1043 	mutex_destroy(&igb->gen_lock);
1044 	mutex_destroy(&igb->watchdog_lock);
1045 	mutex_destroy(&igb->link_lock);
1046 }
1047 
1048 static int
1049 igb_resume(dev_info_t *devinfo)
1050 {
1051 	igb_t *igb;
1052 
1053 	igb = (igb_t *)ddi_get_driver_private(devinfo);
1054 	if (igb == NULL)
1055 		return (DDI_FAILURE);
1056 
1057 	mutex_enter(&igb->gen_lock);
1058 
1059 	if (igb->igb_state & IGB_STARTED) {
1060 		if (igb_start(igb, B_FALSE) != IGB_SUCCESS) {
1061 			mutex_exit(&igb->gen_lock);
1062 			return (DDI_FAILURE);
1063 		}
1064 
1065 		/*
1066 		 * Enable and start the watchdog timer
1067 		 */
1068 		igb_enable_watchdog_timer(igb);
1069 	}
1070 
1071 	atomic_and_32(&igb->igb_state, ~IGB_SUSPENDED);
1072 
1073 	mutex_exit(&igb->gen_lock);
1074 
1075 	return (DDI_SUCCESS);
1076 }
1077 
1078 static int
1079 igb_suspend(dev_info_t *devinfo)
1080 {
1081 	igb_t *igb;
1082 
1083 	igb = (igb_t *)ddi_get_driver_private(devinfo);
1084 	if (igb == NULL)
1085 		return (DDI_FAILURE);
1086 
1087 	mutex_enter(&igb->gen_lock);
1088 
1089 	atomic_or_32(&igb->igb_state, IGB_SUSPENDED);
1090 
1091 	if (!(igb->igb_state & IGB_STARTED)) {
1092 		mutex_exit(&igb->gen_lock);
1093 		return (DDI_SUCCESS);
1094 	}
1095 
1096 	igb_stop(igb, B_FALSE);
1097 
1098 	mutex_exit(&igb->gen_lock);
1099 
1100 	/*
1101 	 * Disable and stop the watchdog timer
1102 	 */
1103 	igb_disable_watchdog_timer(igb);
1104 
1105 	return (DDI_SUCCESS);
1106 }
1107 
1108 static int
1109 igb_init(igb_t *igb)
1110 {
1111 	mutex_enter(&igb->gen_lock);
1112 
1113 	/*
1114 	 * Initilize the adapter
1115 	 */
1116 	if (igb_init_adapter(igb) != IGB_SUCCESS) {
1117 		mutex_exit(&igb->gen_lock);
1118 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1119 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1120 		return (IGB_FAILURE);
1121 	}
1122 
1123 	mutex_exit(&igb->gen_lock);
1124 
1125 	return (IGB_SUCCESS);
1126 }
1127 
1128 /*
1129  * igb_init_mac_address - Initialize the default MAC address
1130  *
1131  * On success, the MAC address is entered in the igb->hw.mac.addr
1132  * and hw->mac.perm_addr fields and the adapter's RAR(0) receive
1133  * address register.
1134  *
1135  * Important side effects:
1136  * 1. adapter is reset - this is required to put it in a known state.
1137  * 2. all of non-volatile memory (NVM) is read & checksummed - NVM is where
1138  * MAC address and all default settings are stored, so a valid checksum
1139  * is required.
1140  */
1141 static int
1142 igb_init_mac_address(igb_t *igb)
1143 {
1144 	struct e1000_hw *hw = &igb->hw;
1145 
1146 	ASSERT(mutex_owned(&igb->gen_lock));
1147 
1148 	/*
1149 	 * Reset chipset to put the hardware in a known state
1150 	 * before we try to get MAC address from NVM.
1151 	 */
1152 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1153 		igb_error(igb, "Adapter reset failed.");
1154 		goto init_mac_fail;
1155 	}
1156 
1157 	/*
1158 	 * NVM validation
1159 	 */
1160 	if (e1000_validate_nvm_checksum(hw) < 0) {
1161 		/*
1162 		 * Some PCI-E parts fail the first check due to
1163 		 * the link being in sleep state.  Call it again,
1164 		 * if it fails a second time its a real issue.
1165 		 */
1166 		if (e1000_validate_nvm_checksum(hw) < 0) {
1167 			igb_error(igb,
1168 			    "Invalid NVM checksum. Please contact "
1169 			    "the vendor to update the NVM.");
1170 			goto init_mac_fail;
1171 		}
1172 	}
1173 
1174 	/*
1175 	 * Get the mac address
1176 	 * This function should handle SPARC case correctly.
1177 	 */
1178 	if (!igb_find_mac_address(igb)) {
1179 		igb_error(igb, "Failed to get the mac address");
1180 		goto init_mac_fail;
1181 	}
1182 
1183 	/* Validate mac address */
1184 	if (!is_valid_mac_addr(hw->mac.addr)) {
1185 		igb_error(igb, "Invalid mac address");
1186 		goto init_mac_fail;
1187 	}
1188 
1189 	return (IGB_SUCCESS);
1190 
1191 init_mac_fail:
1192 	return (IGB_FAILURE);
1193 }
1194 
1195 /*
1196  * igb_init_adapter - Initialize the adapter
1197  */
1198 static int
1199 igb_init_adapter(igb_t *igb)
1200 {
1201 	struct e1000_hw *hw = &igb->hw;
1202 	uint32_t pba;
1203 	uint32_t high_water;
1204 	int i;
1205 
1206 	ASSERT(mutex_owned(&igb->gen_lock));
1207 
1208 	/*
1209 	 * In order to obtain the default MAC address, this will reset the
1210 	 * adapter and validate the NVM that the address and many other
1211 	 * default settings come from.
1212 	 */
1213 	if (igb_init_mac_address(igb) != IGB_SUCCESS) {
1214 		igb_error(igb, "Failed to initialize MAC address");
1215 		goto init_adapter_fail;
1216 	}
1217 
1218 	/*
1219 	 * Setup flow control
1220 	 *
1221 	 * These parameters set thresholds for the adapter's generation(Tx)
1222 	 * and response(Rx) to Ethernet PAUSE frames.  These are just threshold
1223 	 * settings.  Flow control is enabled or disabled in the configuration
1224 	 * file.
1225 	 * High-water mark is set down from the top of the rx fifo (not
1226 	 * sensitive to max_frame_size) and low-water is set just below
1227 	 * high-water mark.
1228 	 * The high water mark must be low enough to fit one full frame above
1229 	 * it in the rx FIFO.  Should be the lower of:
1230 	 * 90% of the Rx FIFO size, or the full Rx FIFO size minus one full
1231 	 * frame.
1232 	 */
1233 	/*
1234 	 * The default setting of PBA is correct for 82575 and other supported
1235 	 * adapters do not have the E1000_PBA register, so PBA value is only
1236 	 * used for calculation here and is never written to the adapter.
1237 	 */
1238 	if (hw->mac.type == e1000_82575) {
1239 		pba = E1000_PBA_34K;
1240 	} else {
1241 		pba = E1000_PBA_64K;
1242 	}
1243 
1244 	high_water = min(((pba << 10) * 9 / 10),
1245 	    ((pba << 10) - igb->max_frame_size));
1246 
1247 	if (hw->mac.type == e1000_82575) {
1248 		/* 8-byte granularity */
1249 		hw->fc.high_water = high_water & 0xFFF8;
1250 		hw->fc.low_water = hw->fc.high_water - 8;
1251 	} else {
1252 		/* 16-byte granularity */
1253 		hw->fc.high_water = high_water & 0xFFF0;
1254 		hw->fc.low_water = hw->fc.high_water - 16;
1255 	}
1256 
1257 	hw->fc.pause_time = E1000_FC_PAUSE_TIME;
1258 	hw->fc.send_xon = B_TRUE;
1259 
1260 	(void) e1000_validate_mdi_setting(hw);
1261 
1262 	/*
1263 	 * Reset the chipset hardware the second time to put PBA settings
1264 	 * into effect.
1265 	 */
1266 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1267 		igb_error(igb, "Second reset failed");
1268 		goto init_adapter_fail;
1269 	}
1270 
1271 	/*
1272 	 * Don't wait for auto-negotiation to complete
1273 	 */
1274 	hw->phy.autoneg_wait_to_complete = B_FALSE;
1275 
1276 	/*
1277 	 * Copper options
1278 	 */
1279 	if (hw->phy.media_type == e1000_media_type_copper) {
1280 		hw->phy.mdix = 0;	/* AUTO_ALL_MODES */
1281 		hw->phy.disable_polarity_correction = B_FALSE;
1282 		hw->phy.ms_type = e1000_ms_hw_default; /* E1000_MASTER_SLAVE */
1283 	}
1284 
1285 	/*
1286 	 * Initialize link settings
1287 	 */
1288 	(void) igb_setup_link(igb, B_FALSE);
1289 
1290 	/*
1291 	 * Configure/Initialize hardware
1292 	 */
1293 	if (e1000_init_hw(hw) != E1000_SUCCESS) {
1294 		igb_error(igb, "Failed to initialize hardware");
1295 		goto init_adapter_fail;
1296 	}
1297 
1298 	/*
1299 	 *  Start the link setup timer
1300 	 */
1301 	igb_start_link_timer(igb);
1302 
1303 	/*
1304 	 * Disable wakeup control by default
1305 	 */
1306 	E1000_WRITE_REG(hw, E1000_WUC, 0);
1307 
1308 	/*
1309 	 * Record phy info in hw struct
1310 	 */
1311 	(void) e1000_get_phy_info(hw);
1312 
1313 	/*
1314 	 * Make sure driver has control
1315 	 */
1316 	igb_get_driver_control(hw);
1317 
1318 	/*
1319 	 * Restore LED settings to the default from EEPROM
1320 	 * to meet the standard for Sun platforms.
1321 	 */
1322 	(void) e1000_cleanup_led(hw);
1323 
1324 	/*
1325 	 * Setup MSI-X interrupts
1326 	 */
1327 	if (igb->intr_type == DDI_INTR_TYPE_MSIX)
1328 		igb->capab->setup_msix(igb);
1329 
1330 	/*
1331 	 * Initialize unicast addresses.
1332 	 */
1333 	igb_init_unicst(igb);
1334 
1335 	/*
1336 	 * Setup and initialize the mctable structures.
1337 	 */
1338 	igb_setup_multicst(igb);
1339 
1340 	/*
1341 	 * Set interrupt throttling rate
1342 	 */
1343 	for (i = 0; i < igb->intr_cnt; i++)
1344 		E1000_WRITE_REG(hw, E1000_EITR(i), igb->intr_throttling[i]);
1345 
1346 	/*
1347 	 * Save the state of the phy
1348 	 */
1349 	igb_get_phy_state(igb);
1350 
1351 	igb_param_sync(igb);
1352 
1353 	return (IGB_SUCCESS);
1354 
1355 init_adapter_fail:
1356 	/*
1357 	 * Reset PHY if possible
1358 	 */
1359 	if (e1000_check_reset_block(hw) == E1000_SUCCESS)
1360 		(void) e1000_phy_hw_reset(hw);
1361 
1362 	return (IGB_FAILURE);
1363 }
1364 
1365 /*
1366  * igb_stop_adapter - Stop the adapter
1367  */
1368 static void
1369 igb_stop_adapter(igb_t *igb)
1370 {
1371 	struct e1000_hw *hw = &igb->hw;
1372 
1373 	ASSERT(mutex_owned(&igb->gen_lock));
1374 
1375 	/* Stop the link setup timer */
1376 	igb_stop_link_timer(igb);
1377 
1378 	/* Tell firmware driver is no longer in control */
1379 	igb_release_driver_control(hw);
1380 
1381 	/*
1382 	 * Reset the chipset
1383 	 */
1384 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1385 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1386 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1387 	}
1388 
1389 	/*
1390 	 * e1000_phy_hw_reset is not needed here, MAC reset above is sufficient
1391 	 */
1392 }
1393 
1394 /*
1395  * igb_reset - Reset the chipset and restart the driver.
1396  *
1397  * It involves stopping and re-starting the chipset,
1398  * and re-configuring the rx/tx rings.
1399  */
1400 static int
1401 igb_reset(igb_t *igb)
1402 {
1403 	int i;
1404 
1405 	mutex_enter(&igb->gen_lock);
1406 
1407 	ASSERT(igb->igb_state & IGB_STARTED);
1408 	atomic_and_32(&igb->igb_state, ~IGB_STARTED);
1409 
1410 	/*
1411 	 * Disable the adapter interrupts to stop any rx/tx activities
1412 	 * before draining pending data and resetting hardware.
1413 	 */
1414 	igb_disable_adapter_interrupts(igb);
1415 
1416 	/*
1417 	 * Drain the pending transmit packets
1418 	 */
1419 	(void) igb_tx_drain(igb);
1420 
1421 	for (i = 0; i < igb->num_rx_rings; i++)
1422 		mutex_enter(&igb->rx_rings[i].rx_lock);
1423 	for (i = 0; i < igb->num_tx_rings; i++)
1424 		mutex_enter(&igb->tx_rings[i].tx_lock);
1425 
1426 	/*
1427 	 * Stop the adapter
1428 	 */
1429 	igb_stop_adapter(igb);
1430 
1431 	/*
1432 	 * Clean the pending tx data/resources
1433 	 */
1434 	igb_tx_clean(igb);
1435 
1436 	/*
1437 	 * Start the adapter
1438 	 */
1439 	if (igb_init_adapter(igb) != IGB_SUCCESS) {
1440 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1441 		goto reset_failure;
1442 	}
1443 
1444 	/*
1445 	 * Setup the rx/tx rings
1446 	 */
1447 	igb->tx_ring_init = B_FALSE;
1448 	igb_setup_rings(igb);
1449 
1450 	atomic_and_32(&igb->igb_state, ~(IGB_ERROR | IGB_STALL));
1451 
1452 	/*
1453 	 * Enable adapter interrupts
1454 	 * The interrupts must be enabled after the driver state is START
1455 	 */
1456 	igb->capab->enable_intr(igb);
1457 
1458 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK)
1459 		goto reset_failure;
1460 
1461 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1462 		goto reset_failure;
1463 
1464 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1465 		mutex_exit(&igb->tx_rings[i].tx_lock);
1466 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1467 		mutex_exit(&igb->rx_rings[i].rx_lock);
1468 
1469 	atomic_or_32(&igb->igb_state, IGB_STARTED);
1470 
1471 	mutex_exit(&igb->gen_lock);
1472 
1473 	return (IGB_SUCCESS);
1474 
1475 reset_failure:
1476 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1477 		mutex_exit(&igb->tx_rings[i].tx_lock);
1478 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1479 		mutex_exit(&igb->rx_rings[i].rx_lock);
1480 
1481 	mutex_exit(&igb->gen_lock);
1482 
1483 	ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1484 
1485 	return (IGB_FAILURE);
1486 }
1487 
1488 /*
1489  * igb_tx_clean - Clean the pending transmit packets and DMA resources
1490  */
1491 static void
1492 igb_tx_clean(igb_t *igb)
1493 {
1494 	igb_tx_ring_t *tx_ring;
1495 	tx_control_block_t *tcb;
1496 	link_list_t pending_list;
1497 	uint32_t desc_num;
1498 	int i, j;
1499 
1500 	LINK_LIST_INIT(&pending_list);
1501 
1502 	for (i = 0; i < igb->num_tx_rings; i++) {
1503 		tx_ring = &igb->tx_rings[i];
1504 
1505 		mutex_enter(&tx_ring->recycle_lock);
1506 
1507 		/*
1508 		 * Clean the pending tx data - the pending packets in the
1509 		 * work_list that have no chances to be transmitted again.
1510 		 *
1511 		 * We must ensure the chipset is stopped or the link is down
1512 		 * before cleaning the transmit packets.
1513 		 */
1514 		desc_num = 0;
1515 		for (j = 0; j < tx_ring->ring_size; j++) {
1516 			tcb = tx_ring->work_list[j];
1517 			if (tcb != NULL) {
1518 				desc_num += tcb->desc_num;
1519 
1520 				tx_ring->work_list[j] = NULL;
1521 
1522 				igb_free_tcb(tcb);
1523 
1524 				LIST_PUSH_TAIL(&pending_list, &tcb->link);
1525 			}
1526 		}
1527 
1528 		if (desc_num > 0) {
1529 			atomic_add_32(&tx_ring->tbd_free, desc_num);
1530 			ASSERT(tx_ring->tbd_free == tx_ring->ring_size);
1531 
1532 			/*
1533 			 * Reset the head and tail pointers of the tbd ring;
1534 			 * Reset the head write-back if it is enabled.
1535 			 */
1536 			tx_ring->tbd_head = 0;
1537 			tx_ring->tbd_tail = 0;
1538 			if (igb->tx_head_wb_enable)
1539 				*tx_ring->tbd_head_wb = 0;
1540 
1541 			E1000_WRITE_REG(&igb->hw, E1000_TDH(tx_ring->index), 0);
1542 			E1000_WRITE_REG(&igb->hw, E1000_TDT(tx_ring->index), 0);
1543 		}
1544 
1545 		mutex_exit(&tx_ring->recycle_lock);
1546 
1547 		/*
1548 		 * Add the tx control blocks in the pending list to
1549 		 * the free list.
1550 		 */
1551 		igb_put_free_list(tx_ring, &pending_list);
1552 	}
1553 }
1554 
1555 /*
1556  * igb_tx_drain - Drain the tx rings to allow pending packets to be transmitted
1557  */
1558 static boolean_t
1559 igb_tx_drain(igb_t *igb)
1560 {
1561 	igb_tx_ring_t *tx_ring;
1562 	boolean_t done;
1563 	int i, j;
1564 
1565 	/*
1566 	 * Wait for a specific time to allow pending tx packets
1567 	 * to be transmitted.
1568 	 *
1569 	 * Check the counter tbd_free to see if transmission is done.
1570 	 * No lock protection is needed here.
1571 	 *
1572 	 * Return B_TRUE if all pending packets have been transmitted;
1573 	 * Otherwise return B_FALSE;
1574 	 */
1575 	for (i = 0; i < TX_DRAIN_TIME; i++) {
1576 
1577 		done = B_TRUE;
1578 		for (j = 0; j < igb->num_tx_rings; j++) {
1579 			tx_ring = &igb->tx_rings[j];
1580 			done = done &&
1581 			    (tx_ring->tbd_free == tx_ring->ring_size);
1582 		}
1583 
1584 		if (done)
1585 			break;
1586 
1587 		msec_delay(1);
1588 	}
1589 
1590 	return (done);
1591 }
1592 
1593 /*
1594  * igb_rx_drain - Wait for all rx buffers to be released by upper layer
1595  */
1596 static boolean_t
1597 igb_rx_drain(igb_t *igb)
1598 {
1599 	boolean_t done;
1600 	int i;
1601 
1602 	/*
1603 	 * Polling the rx free list to check if those rx buffers held by
1604 	 * the upper layer are released.
1605 	 *
1606 	 * Check the counter rcb_free to see if all pending buffers are
1607 	 * released. No lock protection is needed here.
1608 	 *
1609 	 * Return B_TRUE if all pending buffers have been released;
1610 	 * Otherwise return B_FALSE;
1611 	 */
1612 	for (i = 0; i < RX_DRAIN_TIME; i++) {
1613 		done = (igb->rcb_pending == 0);
1614 
1615 		if (done)
1616 			break;
1617 
1618 		msec_delay(1);
1619 	}
1620 
1621 	return (done);
1622 }
1623 
1624 /*
1625  * igb_start - Start the driver/chipset
1626  */
1627 int
1628 igb_start(igb_t *igb, boolean_t alloc_buffer)
1629 {
1630 	int i;
1631 
1632 	ASSERT(mutex_owned(&igb->gen_lock));
1633 
1634 	if (alloc_buffer) {
1635 		if (igb_alloc_rx_data(igb) != IGB_SUCCESS) {
1636 			igb_error(igb,
1637 			    "Failed to allocate software receive rings");
1638 			return (IGB_FAILURE);
1639 		}
1640 
1641 		/* Allocate buffers for all the rx/tx rings */
1642 		if (igb_alloc_dma(igb) != IGB_SUCCESS) {
1643 			igb_error(igb, "Failed to allocate DMA resource");
1644 			return (IGB_FAILURE);
1645 		}
1646 
1647 		igb->tx_ring_init = B_TRUE;
1648 	} else {
1649 		igb->tx_ring_init = B_FALSE;
1650 	}
1651 
1652 	for (i = 0; i < igb->num_rx_rings; i++)
1653 		mutex_enter(&igb->rx_rings[i].rx_lock);
1654 	for (i = 0; i < igb->num_tx_rings; i++)
1655 		mutex_enter(&igb->tx_rings[i].tx_lock);
1656 
1657 	/*
1658 	 * Start the adapter
1659 	 */
1660 	if ((igb->attach_progress & ATTACH_PROGRESS_INIT_ADAPTER) == 0) {
1661 		if (igb_init_adapter(igb) != IGB_SUCCESS) {
1662 			igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1663 			goto start_failure;
1664 		}
1665 		igb->attach_progress |= ATTACH_PROGRESS_INIT_ADAPTER;
1666 	}
1667 
1668 	/*
1669 	 * Setup the rx/tx rings
1670 	 */
1671 	igb_setup_rings(igb);
1672 
1673 	/*
1674 	 * Enable adapter interrupts
1675 	 * The interrupts must be enabled after the driver state is START
1676 	 */
1677 	igb->capab->enable_intr(igb);
1678 
1679 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK)
1680 		goto start_failure;
1681 
1682 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1683 		goto start_failure;
1684 
1685 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1686 		mutex_exit(&igb->tx_rings[i].tx_lock);
1687 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1688 		mutex_exit(&igb->rx_rings[i].rx_lock);
1689 
1690 	return (IGB_SUCCESS);
1691 
1692 start_failure:
1693 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1694 		mutex_exit(&igb->tx_rings[i].tx_lock);
1695 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1696 		mutex_exit(&igb->rx_rings[i].rx_lock);
1697 
1698 	ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1699 
1700 	return (IGB_FAILURE);
1701 }
1702 
1703 /*
1704  * igb_stop - Stop the driver/chipset
1705  */
1706 void
1707 igb_stop(igb_t *igb, boolean_t free_buffer)
1708 {
1709 	int i;
1710 
1711 	ASSERT(mutex_owned(&igb->gen_lock));
1712 
1713 	igb->attach_progress &= ~ATTACH_PROGRESS_INIT_ADAPTER;
1714 
1715 	/*
1716 	 * Disable the adapter interrupts
1717 	 */
1718 	igb_disable_adapter_interrupts(igb);
1719 
1720 	/*
1721 	 * Drain the pending tx packets
1722 	 */
1723 	(void) igb_tx_drain(igb);
1724 
1725 	for (i = 0; i < igb->num_rx_rings; i++)
1726 		mutex_enter(&igb->rx_rings[i].rx_lock);
1727 	for (i = 0; i < igb->num_tx_rings; i++)
1728 		mutex_enter(&igb->tx_rings[i].tx_lock);
1729 
1730 	/*
1731 	 * Stop the adapter
1732 	 */
1733 	igb_stop_adapter(igb);
1734 
1735 	/*
1736 	 * Clean the pending tx data/resources
1737 	 */
1738 	igb_tx_clean(igb);
1739 
1740 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1741 		mutex_exit(&igb->tx_rings[i].tx_lock);
1742 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1743 		mutex_exit(&igb->rx_rings[i].rx_lock);
1744 
1745 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1746 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1747 
1748 	if (igb->link_state == LINK_STATE_UP) {
1749 		igb->link_state = LINK_STATE_UNKNOWN;
1750 		mac_link_update(igb->mac_hdl, igb->link_state);
1751 	}
1752 
1753 	if (free_buffer) {
1754 		/*
1755 		 * Release the DMA/memory resources of rx/tx rings
1756 		 */
1757 		igb_free_dma(igb);
1758 		igb_free_rx_data(igb);
1759 	}
1760 }
1761 
1762 /*
1763  * igb_alloc_rings - Allocate memory space for rx/tx rings
1764  */
1765 static int
1766 igb_alloc_rings(igb_t *igb)
1767 {
1768 	/*
1769 	 * Allocate memory space for rx rings
1770 	 */
1771 	igb->rx_rings = kmem_zalloc(
1772 	    sizeof (igb_rx_ring_t) * igb->num_rx_rings,
1773 	    KM_NOSLEEP);
1774 
1775 	if (igb->rx_rings == NULL) {
1776 		return (IGB_FAILURE);
1777 	}
1778 
1779 	/*
1780 	 * Allocate memory space for tx rings
1781 	 */
1782 	igb->tx_rings = kmem_zalloc(
1783 	    sizeof (igb_tx_ring_t) * igb->num_tx_rings,
1784 	    KM_NOSLEEP);
1785 
1786 	if (igb->tx_rings == NULL) {
1787 		kmem_free(igb->rx_rings,
1788 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1789 		igb->rx_rings = NULL;
1790 		return (IGB_FAILURE);
1791 	}
1792 
1793 	/*
1794 	 * Allocate memory space for rx ring groups
1795 	 */
1796 	igb->rx_groups = kmem_zalloc(
1797 	    sizeof (igb_rx_group_t) * igb->num_rx_groups,
1798 	    KM_NOSLEEP);
1799 
1800 	if (igb->rx_groups == NULL) {
1801 		kmem_free(igb->rx_rings,
1802 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1803 		kmem_free(igb->tx_rings,
1804 		    sizeof (igb_tx_ring_t) * igb->num_tx_rings);
1805 		igb->rx_rings = NULL;
1806 		igb->tx_rings = NULL;
1807 		return (IGB_FAILURE);
1808 	}
1809 
1810 	return (IGB_SUCCESS);
1811 }
1812 
1813 /*
1814  * igb_free_rings - Free the memory space of rx/tx rings.
1815  */
1816 static void
1817 igb_free_rings(igb_t *igb)
1818 {
1819 	if (igb->rx_rings != NULL) {
1820 		kmem_free(igb->rx_rings,
1821 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1822 		igb->rx_rings = NULL;
1823 	}
1824 
1825 	if (igb->tx_rings != NULL) {
1826 		kmem_free(igb->tx_rings,
1827 		    sizeof (igb_tx_ring_t) * igb->num_tx_rings);
1828 		igb->tx_rings = NULL;
1829 	}
1830 
1831 	if (igb->rx_groups != NULL) {
1832 		kmem_free(igb->rx_groups,
1833 		    sizeof (igb_rx_group_t) * igb->num_rx_groups);
1834 		igb->rx_groups = NULL;
1835 	}
1836 }
1837 
1838 static int
1839 igb_alloc_rx_data(igb_t *igb)
1840 {
1841 	igb_rx_ring_t *rx_ring;
1842 	int i;
1843 
1844 	for (i = 0; i < igb->num_rx_rings; i++) {
1845 		rx_ring = &igb->rx_rings[i];
1846 		if (igb_alloc_rx_ring_data(rx_ring) != IGB_SUCCESS)
1847 			goto alloc_rx_rings_failure;
1848 	}
1849 	return (IGB_SUCCESS);
1850 
1851 alloc_rx_rings_failure:
1852 	igb_free_rx_data(igb);
1853 	return (IGB_FAILURE);
1854 }
1855 
1856 static void
1857 igb_free_rx_data(igb_t *igb)
1858 {
1859 	igb_rx_ring_t *rx_ring;
1860 	igb_rx_data_t *rx_data;
1861 	int i;
1862 
1863 	for (i = 0; i < igb->num_rx_rings; i++) {
1864 		rx_ring = &igb->rx_rings[i];
1865 
1866 		mutex_enter(&igb->rx_pending_lock);
1867 		rx_data = rx_ring->rx_data;
1868 
1869 		if (rx_data != NULL) {
1870 			rx_data->flag |= IGB_RX_STOPPED;
1871 
1872 			if (rx_data->rcb_pending == 0) {
1873 				igb_free_rx_ring_data(rx_data);
1874 				rx_ring->rx_data = NULL;
1875 			}
1876 		}
1877 
1878 		mutex_exit(&igb->rx_pending_lock);
1879 	}
1880 }
1881 
1882 /*
1883  * igb_setup_rings - Setup rx/tx rings
1884  */
1885 static void
1886 igb_setup_rings(igb_t *igb)
1887 {
1888 	/*
1889 	 * Setup the rx/tx rings, including the following:
1890 	 *
1891 	 * 1. Setup the descriptor ring and the control block buffers;
1892 	 * 2. Initialize necessary registers for receive/transmit;
1893 	 * 3. Initialize software pointers/parameters for receive/transmit;
1894 	 */
1895 	igb_setup_rx(igb);
1896 
1897 	igb_setup_tx(igb);
1898 }
1899 
1900 static void
1901 igb_setup_rx_ring(igb_rx_ring_t *rx_ring)
1902 {
1903 	igb_t *igb = rx_ring->igb;
1904 	igb_rx_data_t *rx_data = rx_ring->rx_data;
1905 	struct e1000_hw *hw = &igb->hw;
1906 	rx_control_block_t *rcb;
1907 	union e1000_adv_rx_desc	*rbd;
1908 	uint32_t size;
1909 	uint32_t buf_low;
1910 	uint32_t buf_high;
1911 	uint32_t rxdctl;
1912 	int i;
1913 
1914 	ASSERT(mutex_owned(&rx_ring->rx_lock));
1915 	ASSERT(mutex_owned(&igb->gen_lock));
1916 
1917 	/*
1918 	 * Initialize descriptor ring with buffer addresses
1919 	 */
1920 	for (i = 0; i < igb->rx_ring_size; i++) {
1921 		rcb = rx_data->work_list[i];
1922 		rbd = &rx_data->rbd_ring[i];
1923 
1924 		rbd->read.pkt_addr = rcb->rx_buf.dma_address;
1925 		rbd->read.hdr_addr = NULL;
1926 	}
1927 
1928 	/*
1929 	 * Initialize the base address registers
1930 	 */
1931 	buf_low = (uint32_t)rx_data->rbd_area.dma_address;
1932 	buf_high = (uint32_t)(rx_data->rbd_area.dma_address >> 32);
1933 	E1000_WRITE_REG(hw, E1000_RDBAH(rx_ring->index), buf_high);
1934 	E1000_WRITE_REG(hw, E1000_RDBAL(rx_ring->index), buf_low);
1935 
1936 	/*
1937 	 * Initialize the length register
1938 	 */
1939 	size = rx_data->ring_size * sizeof (union e1000_adv_rx_desc);
1940 	E1000_WRITE_REG(hw, E1000_RDLEN(rx_ring->index), size);
1941 
1942 	/*
1943 	 * Initialize buffer size & descriptor type
1944 	 */
1945 	E1000_WRITE_REG(hw, E1000_SRRCTL(rx_ring->index),
1946 	    ((igb->rx_buf_size >> E1000_SRRCTL_BSIZEPKT_SHIFT) |
1947 	    E1000_SRRCTL_DESCTYPE_ADV_ONEBUF));
1948 
1949 	/*
1950 	 * Setup the Receive Descriptor Control Register (RXDCTL)
1951 	 */
1952 	rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(rx_ring->index));
1953 	rxdctl &= igb->capab->rxdctl_mask;
1954 	rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1955 	rxdctl |= 16;		/* pthresh */
1956 	rxdctl |= 8 << 8;	/* hthresh */
1957 	rxdctl |= 1 << 16;	/* wthresh */
1958 	E1000_WRITE_REG(hw, E1000_RXDCTL(rx_ring->index), rxdctl);
1959 
1960 	rx_data->rbd_next = 0;
1961 }
1962 
1963 static void
1964 igb_setup_rx(igb_t *igb)
1965 {
1966 	igb_rx_ring_t *rx_ring;
1967 	igb_rx_data_t *rx_data;
1968 	igb_rx_group_t *rx_group;
1969 	struct e1000_hw *hw = &igb->hw;
1970 	uint32_t rctl, rxcsum;
1971 	uint32_t ring_per_group;
1972 	int i;
1973 
1974 	/*
1975 	 * Setup the Receive Control Register (RCTL), and enable the
1976 	 * receiver. The initial configuration is to: enable the receiver,
1977 	 * accept broadcasts, discard bad packets, accept long packets,
1978 	 * disable VLAN filter checking, and set receive buffer size to
1979 	 * 2k.  For 82575, also set the receive descriptor minimum
1980 	 * threshold size to 1/2 the ring.
1981 	 */
1982 	rctl = E1000_READ_REG(hw, E1000_RCTL);
1983 
1984 	/*
1985 	 * Clear the field used for wakeup control.  This driver doesn't do
1986 	 * wakeup but leave this here for completeness.
1987 	 */
1988 	rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1989 	rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
1990 
1991 	rctl |= (E1000_RCTL_EN |	/* Enable Receive Unit */
1992 	    E1000_RCTL_BAM |		/* Accept Broadcast Packets */
1993 	    E1000_RCTL_LPE |		/* Large Packet Enable */
1994 					/* Multicast filter offset */
1995 	    (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT) |
1996 	    E1000_RCTL_RDMTS_HALF |	/* rx descriptor threshold */
1997 	    E1000_RCTL_SECRC);		/* Strip Ethernet CRC */
1998 
1999 	for (i = 0; i < igb->num_rx_groups; i++) {
2000 		rx_group = &igb->rx_groups[i];
2001 		rx_group->index = i;
2002 		rx_group->igb = igb;
2003 	}
2004 
2005 	/*
2006 	 * Set up all rx descriptor rings - must be called before receive unit
2007 	 * enabled.
2008 	 */
2009 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
2010 	for (i = 0; i < igb->num_rx_rings; i++) {
2011 		rx_ring = &igb->rx_rings[i];
2012 		igb_setup_rx_ring(rx_ring);
2013 
2014 		/*
2015 		 * Map a ring to a group by assigning a group index
2016 		 */
2017 		rx_ring->group_index = i / ring_per_group;
2018 	}
2019 
2020 	/*
2021 	 * Setup the Rx Long Packet Max Length register
2022 	 */
2023 	E1000_WRITE_REG(hw, E1000_RLPML, igb->max_frame_size);
2024 
2025 	/*
2026 	 * Hardware checksum settings
2027 	 */
2028 	if (igb->rx_hcksum_enable) {
2029 		rxcsum =
2030 		    E1000_RXCSUM_TUOFL |	/* TCP/UDP checksum */
2031 		    E1000_RXCSUM_IPOFL;		/* IP checksum */
2032 
2033 		E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2034 	}
2035 
2036 	/*
2037 	 * Setup classify and RSS for multiple receive queues
2038 	 */
2039 	switch (igb->vmdq_mode) {
2040 	case E1000_VMDQ_OFF:
2041 		/*
2042 		 * One ring group, only RSS is needed when more than
2043 		 * one ring enabled.
2044 		 */
2045 		if (igb->num_rx_rings > 1)
2046 			igb_setup_rss(igb);
2047 		break;
2048 	case E1000_VMDQ_MAC:
2049 		/*
2050 		 * Multiple groups, each group has one ring,
2051 		 * only the MAC classification is needed.
2052 		 */
2053 		igb_setup_mac_classify(igb);
2054 		break;
2055 	case E1000_VMDQ_MAC_RSS:
2056 		/*
2057 		 * Multiple groups and multiple rings, both
2058 		 * MAC classification and RSS are needed.
2059 		 */
2060 		igb_setup_mac_rss_classify(igb);
2061 		break;
2062 	}
2063 
2064 	/*
2065 	 * Enable the receive unit - must be done after all
2066 	 * the rx setup above.
2067 	 */
2068 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2069 
2070 	/*
2071 	 * Initialize all adapter ring head & tail pointers - must
2072 	 * be done after receive unit is enabled
2073 	 */
2074 	for (i = 0; i < igb->num_rx_rings; i++) {
2075 		rx_ring = &igb->rx_rings[i];
2076 		rx_data = rx_ring->rx_data;
2077 		E1000_WRITE_REG(hw, E1000_RDH(i), 0);
2078 		E1000_WRITE_REG(hw, E1000_RDT(i), rx_data->ring_size - 1);
2079 	}
2080 
2081 	/*
2082 	 * 82575 with manageability enabled needs a special flush to make
2083 	 * sure the fifos start clean.
2084 	 */
2085 	if ((hw->mac.type == e1000_82575) &&
2086 	    (E1000_READ_REG(hw, E1000_MANC) & E1000_MANC_RCV_TCO_EN)) {
2087 		e1000_rx_fifo_flush_82575(hw);
2088 	}
2089 }
2090 
2091 static void
2092 igb_setup_tx_ring(igb_tx_ring_t *tx_ring)
2093 {
2094 	igb_t *igb = tx_ring->igb;
2095 	struct e1000_hw *hw = &igb->hw;
2096 	uint32_t size;
2097 	uint32_t buf_low;
2098 	uint32_t buf_high;
2099 	uint32_t reg_val;
2100 
2101 	ASSERT(mutex_owned(&tx_ring->tx_lock));
2102 	ASSERT(mutex_owned(&igb->gen_lock));
2103 
2104 
2105 	/*
2106 	 * Initialize the length register
2107 	 */
2108 	size = tx_ring->ring_size * sizeof (union e1000_adv_tx_desc);
2109 	E1000_WRITE_REG(hw, E1000_TDLEN(tx_ring->index), size);
2110 
2111 	/*
2112 	 * Initialize the base address registers
2113 	 */
2114 	buf_low = (uint32_t)tx_ring->tbd_area.dma_address;
2115 	buf_high = (uint32_t)(tx_ring->tbd_area.dma_address >> 32);
2116 	E1000_WRITE_REG(hw, E1000_TDBAL(tx_ring->index), buf_low);
2117 	E1000_WRITE_REG(hw, E1000_TDBAH(tx_ring->index), buf_high);
2118 
2119 	/*
2120 	 * Setup head & tail pointers
2121 	 */
2122 	E1000_WRITE_REG(hw, E1000_TDH(tx_ring->index), 0);
2123 	E1000_WRITE_REG(hw, E1000_TDT(tx_ring->index), 0);
2124 
2125 	/*
2126 	 * Setup head write-back
2127 	 */
2128 	if (igb->tx_head_wb_enable) {
2129 		/*
2130 		 * The memory of the head write-back is allocated using
2131 		 * the extra tbd beyond the tail of the tbd ring.
2132 		 */
2133 		tx_ring->tbd_head_wb = (uint32_t *)
2134 		    ((uintptr_t)tx_ring->tbd_area.address + size);
2135 		*tx_ring->tbd_head_wb = 0;
2136 
2137 		buf_low = (uint32_t)
2138 		    (tx_ring->tbd_area.dma_address + size);
2139 		buf_high = (uint32_t)
2140 		    ((tx_ring->tbd_area.dma_address + size) >> 32);
2141 
2142 		/* Set the head write-back enable bit */
2143 		buf_low |= E1000_TX_HEAD_WB_ENABLE;
2144 
2145 		E1000_WRITE_REG(hw, E1000_TDWBAL(tx_ring->index), buf_low);
2146 		E1000_WRITE_REG(hw, E1000_TDWBAH(tx_ring->index), buf_high);
2147 
2148 		/*
2149 		 * Turn off relaxed ordering for head write back or it will
2150 		 * cause problems with the tx recycling
2151 		 */
2152 		reg_val = E1000_READ_REG(hw,
2153 		    E1000_DCA_TXCTRL(tx_ring->index));
2154 		reg_val &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
2155 		E1000_WRITE_REG(hw,
2156 		    E1000_DCA_TXCTRL(tx_ring->index), reg_val);
2157 	} else {
2158 		tx_ring->tbd_head_wb = NULL;
2159 	}
2160 
2161 	tx_ring->tbd_head = 0;
2162 	tx_ring->tbd_tail = 0;
2163 	tx_ring->tbd_free = tx_ring->ring_size;
2164 
2165 	if (igb->tx_ring_init == B_TRUE) {
2166 		tx_ring->tcb_head = 0;
2167 		tx_ring->tcb_tail = 0;
2168 		tx_ring->tcb_free = tx_ring->free_list_size;
2169 	}
2170 
2171 	/*
2172 	 * Enable TXDCTL per queue
2173 	 */
2174 	reg_val = E1000_READ_REG(hw, E1000_TXDCTL(tx_ring->index));
2175 	reg_val |= E1000_TXDCTL_QUEUE_ENABLE;
2176 	E1000_WRITE_REG(hw, E1000_TXDCTL(tx_ring->index), reg_val);
2177 
2178 	/*
2179 	 * Initialize hardware checksum offload settings
2180 	 */
2181 	bzero(&tx_ring->tx_context, sizeof (tx_context_t));
2182 }
2183 
2184 static void
2185 igb_setup_tx(igb_t *igb)
2186 {
2187 	igb_tx_ring_t *tx_ring;
2188 	struct e1000_hw *hw = &igb->hw;
2189 	uint32_t reg_val;
2190 	int i;
2191 
2192 	for (i = 0; i < igb->num_tx_rings; i++) {
2193 		tx_ring = &igb->tx_rings[i];
2194 		igb_setup_tx_ring(tx_ring);
2195 	}
2196 
2197 	/*
2198 	 * Setup the Transmit Control Register (TCTL)
2199 	 */
2200 	reg_val = E1000_READ_REG(hw, E1000_TCTL);
2201 	reg_val &= ~E1000_TCTL_CT;
2202 	reg_val |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2203 	    (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2204 
2205 	/* Enable transmits */
2206 	reg_val |= E1000_TCTL_EN;
2207 
2208 	E1000_WRITE_REG(hw, E1000_TCTL, reg_val);
2209 }
2210 
2211 /*
2212  * igb_setup_rss - Setup receive-side scaling feature
2213  */
2214 static void
2215 igb_setup_rss(igb_t *igb)
2216 {
2217 	struct e1000_hw *hw = &igb->hw;
2218 	uint32_t i, mrqc, rxcsum;
2219 	int shift = 0;
2220 	uint32_t random;
2221 	union e1000_reta {
2222 		uint32_t	dword;
2223 		uint8_t		bytes[4];
2224 	} reta;
2225 
2226 	/* Setup the Redirection Table */
2227 	if (hw->mac.type == e1000_82576) {
2228 		shift = 3;
2229 	} else if (hw->mac.type == e1000_82575) {
2230 		shift = 6;
2231 	}
2232 	for (i = 0; i < (32 * 4); i++) {
2233 		reta.bytes[i & 3] = (i % igb->num_rx_rings) << shift;
2234 		if ((i & 3) == 3) {
2235 			E1000_WRITE_REG(hw,
2236 			    (E1000_RETA(0) + (i & ~3)), reta.dword);
2237 		}
2238 	}
2239 
2240 	/* Fill out hash function seeds */
2241 	for (i = 0; i < 10; i++) {
2242 		(void) random_get_pseudo_bytes((uint8_t *)&random,
2243 		    sizeof (uint32_t));
2244 		E1000_WRITE_REG(hw, E1000_RSSRK(i), random);
2245 	}
2246 
2247 	/* Setup the Multiple Receive Queue Control register */
2248 	mrqc = E1000_MRQC_ENABLE_RSS_4Q;
2249 	mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2250 	    E1000_MRQC_RSS_FIELD_IPV4_TCP |
2251 	    E1000_MRQC_RSS_FIELD_IPV6 |
2252 	    E1000_MRQC_RSS_FIELD_IPV6_TCP |
2253 	    E1000_MRQC_RSS_FIELD_IPV4_UDP |
2254 	    E1000_MRQC_RSS_FIELD_IPV6_UDP |
2255 	    E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2256 	    E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2257 
2258 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2259 
2260 	/*
2261 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2262 	 *
2263 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2264 	 * checksum offloading provided by the 82575 chipset besides the IP
2265 	 * header checksum offloading and the TCP/UDP checksum offloading.
2266 	 * The Packet Checksum is by default computed over the entire packet
2267 	 * from the first byte of the DA through the last byte of the CRC,
2268 	 * including the Ethernet and IP headers.
2269 	 *
2270 	 * It is a hardware limitation that Packet Checksum is mutually
2271 	 * exclusive with RSS.
2272 	 */
2273 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2274 	rxcsum |= E1000_RXCSUM_PCSD;
2275 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2276 }
2277 
2278 /*
2279  * igb_setup_mac_rss_classify - Setup MAC classification and rss
2280  */
2281 static void
2282 igb_setup_mac_rss_classify(igb_t *igb)
2283 {
2284 	struct e1000_hw *hw = &igb->hw;
2285 	uint32_t i, mrqc, vmdctl, rxcsum;
2286 	uint32_t ring_per_group;
2287 	int shift_group0, shift_group1;
2288 	uint32_t random;
2289 	union e1000_reta {
2290 		uint32_t	dword;
2291 		uint8_t		bytes[4];
2292 	} reta;
2293 
2294 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
2295 
2296 	/* Setup the Redirection Table, it is shared between two groups */
2297 	shift_group0 = 2;
2298 	shift_group1 = 6;
2299 	for (i = 0; i < (32 * 4); i++) {
2300 		reta.bytes[i & 3] = ((i % ring_per_group) << shift_group0) |
2301 		    ((ring_per_group + (i % ring_per_group)) << shift_group1);
2302 		if ((i & 3) == 3) {
2303 			E1000_WRITE_REG(hw,
2304 			    (E1000_RETA(0) + (i & ~3)), reta.dword);
2305 		}
2306 	}
2307 
2308 	/* Fill out hash function seeds */
2309 	for (i = 0; i < 10; i++) {
2310 		(void) random_get_pseudo_bytes((uint8_t *)&random,
2311 		    sizeof (uint32_t));
2312 		E1000_WRITE_REG(hw, E1000_RSSRK(i), random);
2313 	}
2314 
2315 	/*
2316 	 * Setup the Multiple Receive Queue Control register,
2317 	 * enable VMDq based on packet destination MAC address and RSS.
2318 	 */
2319 	mrqc = E1000_MRQC_ENABLE_VMDQ_MAC_RSS_GROUP;
2320 	mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2321 	    E1000_MRQC_RSS_FIELD_IPV4_TCP |
2322 	    E1000_MRQC_RSS_FIELD_IPV6 |
2323 	    E1000_MRQC_RSS_FIELD_IPV6_TCP |
2324 	    E1000_MRQC_RSS_FIELD_IPV4_UDP |
2325 	    E1000_MRQC_RSS_FIELD_IPV6_UDP |
2326 	    E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2327 	    E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2328 
2329 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2330 
2331 
2332 	/* Define the default group and default queues */
2333 	vmdctl = E1000_VMDQ_MAC_GROUP_DEFAULT_QUEUE;
2334 	E1000_WRITE_REG(hw, E1000_VT_CTL, vmdctl);
2335 
2336 	/*
2337 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2338 	 *
2339 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2340 	 * checksum offloading provided by the 82575 chipset besides the IP
2341 	 * header checksum offloading and the TCP/UDP checksum offloading.
2342 	 * The Packet Checksum is by default computed over the entire packet
2343 	 * from the first byte of the DA through the last byte of the CRC,
2344 	 * including the Ethernet and IP headers.
2345 	 *
2346 	 * It is a hardware limitation that Packet Checksum is mutually
2347 	 * exclusive with RSS.
2348 	 */
2349 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2350 	rxcsum |= E1000_RXCSUM_PCSD;
2351 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2352 }
2353 
2354 /*
2355  * igb_setup_mac_classify - Setup MAC classification feature
2356  */
2357 static void
2358 igb_setup_mac_classify(igb_t *igb)
2359 {
2360 	struct e1000_hw *hw = &igb->hw;
2361 	uint32_t mrqc, rxcsum;
2362 
2363 	/*
2364 	 * Setup the Multiple Receive Queue Control register,
2365 	 * enable VMDq based on packet destination MAC address.
2366 	 */
2367 	mrqc = E1000_MRQC_ENABLE_VMDQ_MAC_GROUP;
2368 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2369 
2370 	/*
2371 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2372 	 *
2373 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2374 	 * checksum offloading provided by the 82575 chipset besides the IP
2375 	 * header checksum offloading and the TCP/UDP checksum offloading.
2376 	 * The Packet Checksum is by default computed over the entire packet
2377 	 * from the first byte of the DA through the last byte of the CRC,
2378 	 * including the Ethernet and IP headers.
2379 	 *
2380 	 * It is a hardware limitation that Packet Checksum is mutually
2381 	 * exclusive with RSS.
2382 	 */
2383 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2384 	rxcsum |= E1000_RXCSUM_PCSD;
2385 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2386 
2387 }
2388 
2389 /*
2390  * igb_init_unicst - Initialize the unicast addresses
2391  */
2392 static void
2393 igb_init_unicst(igb_t *igb)
2394 {
2395 	struct e1000_hw *hw = &igb->hw;
2396 	int slot;
2397 
2398 	/*
2399 	 * Here we should consider two situations:
2400 	 *
2401 	 * 1. Chipset is initialized the first time
2402 	 *    Initialize the multiple unicast addresses, and
2403 	 *    save the default MAC address.
2404 	 *
2405 	 * 2. Chipset is reset
2406 	 *    Recover the multiple unicast addresses from the
2407 	 *    software data structure to the RAR registers.
2408 	 */
2409 
2410 	/*
2411 	 * Clear the default MAC address in the RAR0 rgister,
2412 	 * which is loaded from EEPROM when system boot or chipreset,
2413 	 * this will cause the conficts with add_mac/rem_mac entry
2414 	 * points when VMDq is enabled. For this reason, the RAR0
2415 	 * must be cleared for both cases mentioned above.
2416 	 */
2417 	e1000_rar_clear(hw, 0);
2418 
2419 	if (!igb->unicst_init) {
2420 
2421 		/* Initialize the multiple unicast addresses */
2422 		igb->unicst_total = MAX_NUM_UNICAST_ADDRESSES;
2423 		igb->unicst_avail = igb->unicst_total;
2424 
2425 		for (slot = 0; slot < igb->unicst_total; slot++)
2426 			igb->unicst_addr[slot].mac.set = 0;
2427 
2428 		igb->unicst_init = B_TRUE;
2429 	} else {
2430 		/* Re-configure the RAR registers */
2431 		for (slot = 0; slot < igb->unicst_total; slot++) {
2432 			e1000_rar_set_vmdq(hw, igb->unicst_addr[slot].mac.addr,
2433 			    slot, igb->vmdq_mode,
2434 			    igb->unicst_addr[slot].mac.group_index);
2435 		}
2436 	}
2437 }
2438 
2439 /*
2440  * igb_unicst_find - Find the slot for the specified unicast address
2441  */
2442 int
2443 igb_unicst_find(igb_t *igb, const uint8_t *mac_addr)
2444 {
2445 	int slot;
2446 
2447 	ASSERT(mutex_owned(&igb->gen_lock));
2448 
2449 	for (slot = 0; slot < igb->unicst_total; slot++) {
2450 		if (bcmp(igb->unicst_addr[slot].mac.addr,
2451 		    mac_addr, ETHERADDRL) == 0)
2452 			return (slot);
2453 	}
2454 
2455 	return (-1);
2456 }
2457 
2458 /*
2459  * igb_unicst_set - Set the unicast address to the specified slot
2460  */
2461 int
2462 igb_unicst_set(igb_t *igb, const uint8_t *mac_addr,
2463     int slot)
2464 {
2465 	struct e1000_hw *hw = &igb->hw;
2466 
2467 	ASSERT(mutex_owned(&igb->gen_lock));
2468 
2469 	/*
2470 	 * Save the unicast address in the software data structure
2471 	 */
2472 	bcopy(mac_addr, igb->unicst_addr[slot].mac.addr, ETHERADDRL);
2473 
2474 	/*
2475 	 * Set the unicast address to the RAR register
2476 	 */
2477 	e1000_rar_set(hw, (uint8_t *)mac_addr, slot);
2478 
2479 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2480 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2481 		return (EIO);
2482 	}
2483 
2484 	return (0);
2485 }
2486 
2487 /*
2488  * igb_multicst_add - Add a multicst address
2489  */
2490 int
2491 igb_multicst_add(igb_t *igb, const uint8_t *multiaddr)
2492 {
2493 	struct ether_addr *new_table;
2494 	size_t new_len;
2495 	size_t old_len;
2496 
2497 	ASSERT(mutex_owned(&igb->gen_lock));
2498 
2499 	if ((multiaddr[0] & 01) == 0) {
2500 		igb_error(igb, "Illegal multicast address");
2501 		return (EINVAL);
2502 	}
2503 
2504 	if (igb->mcast_count >= igb->mcast_max_num) {
2505 		igb_error(igb, "Adapter requested more than %d mcast addresses",
2506 		    igb->mcast_max_num);
2507 		return (ENOENT);
2508 	}
2509 
2510 	if (igb->mcast_count == igb->mcast_alloc_count) {
2511 		old_len = igb->mcast_alloc_count *
2512 		    sizeof (struct ether_addr);
2513 		new_len = (igb->mcast_alloc_count + MCAST_ALLOC_COUNT) *
2514 		    sizeof (struct ether_addr);
2515 
2516 		new_table = kmem_alloc(new_len, KM_NOSLEEP);
2517 		if (new_table == NULL) {
2518 			igb_error(igb,
2519 			    "Not enough memory to alloc mcast table");
2520 			return (ENOMEM);
2521 		}
2522 
2523 		if (igb->mcast_table != NULL) {
2524 			bcopy(igb->mcast_table, new_table, old_len);
2525 			kmem_free(igb->mcast_table, old_len);
2526 		}
2527 		igb->mcast_alloc_count += MCAST_ALLOC_COUNT;
2528 		igb->mcast_table = new_table;
2529 	}
2530 
2531 	bcopy(multiaddr,
2532 	    &igb->mcast_table[igb->mcast_count], ETHERADDRL);
2533 	igb->mcast_count++;
2534 
2535 	/*
2536 	 * Update the multicast table in the hardware
2537 	 */
2538 	igb_setup_multicst(igb);
2539 
2540 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2541 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2542 		return (EIO);
2543 	}
2544 
2545 	return (0);
2546 }
2547 
2548 /*
2549  * igb_multicst_remove - Remove a multicst address
2550  */
2551 int
2552 igb_multicst_remove(igb_t *igb, const uint8_t *multiaddr)
2553 {
2554 	struct ether_addr *new_table;
2555 	size_t new_len;
2556 	size_t old_len;
2557 	int i;
2558 
2559 	ASSERT(mutex_owned(&igb->gen_lock));
2560 
2561 	for (i = 0; i < igb->mcast_count; i++) {
2562 		if (bcmp(multiaddr, &igb->mcast_table[i],
2563 		    ETHERADDRL) == 0) {
2564 			for (i++; i < igb->mcast_count; i++) {
2565 				igb->mcast_table[i - 1] =
2566 				    igb->mcast_table[i];
2567 			}
2568 			igb->mcast_count--;
2569 			break;
2570 		}
2571 	}
2572 
2573 	if ((igb->mcast_alloc_count - igb->mcast_count) >
2574 	    MCAST_ALLOC_COUNT) {
2575 		old_len = igb->mcast_alloc_count *
2576 		    sizeof (struct ether_addr);
2577 		new_len = (igb->mcast_alloc_count - MCAST_ALLOC_COUNT) *
2578 		    sizeof (struct ether_addr);
2579 
2580 		new_table = kmem_alloc(new_len, KM_NOSLEEP);
2581 		if (new_table != NULL) {
2582 			bcopy(igb->mcast_table, new_table, new_len);
2583 			kmem_free(igb->mcast_table, old_len);
2584 			igb->mcast_alloc_count -= MCAST_ALLOC_COUNT;
2585 			igb->mcast_table = new_table;
2586 		}
2587 	}
2588 
2589 	/*
2590 	 * Update the multicast table in the hardware
2591 	 */
2592 	igb_setup_multicst(igb);
2593 
2594 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2595 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2596 		return (EIO);
2597 	}
2598 
2599 	return (0);
2600 }
2601 
2602 static void
2603 igb_release_multicast(igb_t *igb)
2604 {
2605 	if (igb->mcast_table != NULL) {
2606 		kmem_free(igb->mcast_table,
2607 		    igb->mcast_alloc_count * sizeof (struct ether_addr));
2608 		igb->mcast_table = NULL;
2609 	}
2610 }
2611 
2612 /*
2613  * igb_setup_multicast - setup multicast data structures
2614  *
2615  * This routine initializes all of the multicast related structures
2616  * and save them in the hardware registers.
2617  */
2618 static void
2619 igb_setup_multicst(igb_t *igb)
2620 {
2621 	uint8_t *mc_addr_list;
2622 	uint32_t mc_addr_count;
2623 	struct e1000_hw *hw = &igb->hw;
2624 
2625 	ASSERT(mutex_owned(&igb->gen_lock));
2626 	ASSERT(igb->mcast_count <= igb->mcast_max_num);
2627 
2628 	mc_addr_list = (uint8_t *)igb->mcast_table;
2629 	mc_addr_count = igb->mcast_count;
2630 
2631 	/*
2632 	 * Update the multicase addresses to the MTA registers
2633 	 */
2634 	e1000_update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
2635 }
2636 
2637 /*
2638  * igb_get_conf - Get driver configurations set in driver.conf
2639  *
2640  * This routine gets user-configured values out of the configuration
2641  * file igb.conf.
2642  *
2643  * For each configurable value, there is a minimum, a maximum, and a
2644  * default.
2645  * If user does not configure a value, use the default.
2646  * If user configures below the minimum, use the minumum.
2647  * If user configures above the maximum, use the maxumum.
2648  */
2649 static void
2650 igb_get_conf(igb_t *igb)
2651 {
2652 	struct e1000_hw *hw = &igb->hw;
2653 	uint32_t default_mtu;
2654 	uint32_t flow_control;
2655 	uint32_t ring_per_group;
2656 	int i;
2657 
2658 	/*
2659 	 * igb driver supports the following user configurations:
2660 	 *
2661 	 * Link configurations:
2662 	 *    adv_autoneg_cap
2663 	 *    adv_1000fdx_cap
2664 	 *    adv_100fdx_cap
2665 	 *    adv_100hdx_cap
2666 	 *    adv_10fdx_cap
2667 	 *    adv_10hdx_cap
2668 	 * Note: 1000hdx is not supported.
2669 	 *
2670 	 * Jumbo frame configuration:
2671 	 *    default_mtu
2672 	 *
2673 	 * Ethernet flow control configuration:
2674 	 *    flow_control
2675 	 *
2676 	 * Multiple rings configurations:
2677 	 *    tx_queue_number
2678 	 *    tx_ring_size
2679 	 *    rx_queue_number
2680 	 *    rx_ring_size
2681 	 *
2682 	 * Call igb_get_prop() to get the value for a specific
2683 	 * configuration parameter.
2684 	 */
2685 
2686 	/*
2687 	 * Link configurations
2688 	 */
2689 	igb->param_adv_autoneg_cap = igb_get_prop(igb,
2690 	    PROP_ADV_AUTONEG_CAP, 0, 1, 1);
2691 	igb->param_adv_1000fdx_cap = igb_get_prop(igb,
2692 	    PROP_ADV_1000FDX_CAP, 0, 1, 1);
2693 	igb->param_adv_100fdx_cap = igb_get_prop(igb,
2694 	    PROP_ADV_100FDX_CAP, 0, 1, 1);
2695 	igb->param_adv_100hdx_cap = igb_get_prop(igb,
2696 	    PROP_ADV_100HDX_CAP, 0, 1, 1);
2697 	igb->param_adv_10fdx_cap = igb_get_prop(igb,
2698 	    PROP_ADV_10FDX_CAP, 0, 1, 1);
2699 	igb->param_adv_10hdx_cap = igb_get_prop(igb,
2700 	    PROP_ADV_10HDX_CAP, 0, 1, 1);
2701 
2702 	/*
2703 	 * Jumbo frame configurations
2704 	 */
2705 	default_mtu = igb_get_prop(igb, PROP_DEFAULT_MTU,
2706 	    MIN_MTU, MAX_MTU, DEFAULT_MTU);
2707 
2708 	igb->max_frame_size = default_mtu +
2709 	    sizeof (struct ether_vlan_header) + ETHERFCSL;
2710 
2711 	/*
2712 	 * Ethernet flow control configuration
2713 	 */
2714 	flow_control = igb_get_prop(igb, PROP_FLOW_CONTROL,
2715 	    e1000_fc_none, 4, e1000_fc_full);
2716 	if (flow_control == 4)
2717 		flow_control = e1000_fc_default;
2718 
2719 	hw->fc.requested_mode = flow_control;
2720 
2721 	/*
2722 	 * Multiple rings configurations
2723 	 */
2724 	igb->tx_ring_size = igb_get_prop(igb, PROP_TX_RING_SIZE,
2725 	    MIN_TX_RING_SIZE, MAX_TX_RING_SIZE, DEFAULT_TX_RING_SIZE);
2726 	igb->rx_ring_size = igb_get_prop(igb, PROP_RX_RING_SIZE,
2727 	    MIN_RX_RING_SIZE, MAX_RX_RING_SIZE, DEFAULT_RX_RING_SIZE);
2728 
2729 	igb->mr_enable = igb_get_prop(igb, PROP_MR_ENABLE, 0, 1, 0);
2730 	igb->num_rx_groups = igb_get_prop(igb, PROP_RX_GROUP_NUM,
2731 	    MIN_RX_GROUP_NUM, MAX_RX_GROUP_NUM, DEFAULT_RX_GROUP_NUM);
2732 	/*
2733 	 * Currently we do not support VMDq for 82576 and 82580.
2734 	 * If it is e1000_82576, set num_rx_groups to 1.
2735 	 */
2736 	if (hw->mac.type >= e1000_82576)
2737 		igb->num_rx_groups = 1;
2738 
2739 	if (igb->mr_enable) {
2740 		igb->num_tx_rings = igb->capab->def_tx_que_num;
2741 		igb->num_rx_rings = igb->capab->def_rx_que_num;
2742 	} else {
2743 		igb->num_tx_rings = 1;
2744 		igb->num_rx_rings = 1;
2745 
2746 		if (igb->num_rx_groups > 1) {
2747 			igb_error(igb,
2748 			    "Invalid rx groups number. Please enable multiple "
2749 			    "rings first");
2750 			igb->num_rx_groups = 1;
2751 		}
2752 	}
2753 
2754 	/*
2755 	 * Check the divisibility between rx rings and rx groups.
2756 	 */
2757 	for (i = igb->num_rx_groups; i > 0; i--) {
2758 		if ((igb->num_rx_rings % i) == 0)
2759 			break;
2760 	}
2761 	if (i != igb->num_rx_groups) {
2762 		igb_error(igb,
2763 		    "Invalid rx groups number. Downgrade the rx group "
2764 		    "number to %d.", i);
2765 		igb->num_rx_groups = i;
2766 	}
2767 
2768 	/*
2769 	 * Get the ring number per group.
2770 	 */
2771 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
2772 
2773 	if (igb->num_rx_groups == 1) {
2774 		/*
2775 		 * One rx ring group, the rx ring number is num_rx_rings.
2776 		 */
2777 		igb->vmdq_mode = E1000_VMDQ_OFF;
2778 	} else if (ring_per_group == 1) {
2779 		/*
2780 		 * Multiple rx groups, each group has one rx ring.
2781 		 */
2782 		igb->vmdq_mode = E1000_VMDQ_MAC;
2783 	} else {
2784 		/*
2785 		 * Multiple groups and multiple rings.
2786 		 */
2787 		igb->vmdq_mode = E1000_VMDQ_MAC_RSS;
2788 	}
2789 
2790 	/*
2791 	 * Tunable used to force an interrupt type. The only use is
2792 	 * for testing of the lesser interrupt types.
2793 	 * 0 = don't force interrupt type
2794 	 * 1 = force interrupt type MSIX
2795 	 * 2 = force interrupt type MSI
2796 	 * 3 = force interrupt type Legacy
2797 	 */
2798 	igb->intr_force = igb_get_prop(igb, PROP_INTR_FORCE,
2799 	    IGB_INTR_NONE, IGB_INTR_LEGACY, IGB_INTR_NONE);
2800 
2801 	igb->tx_hcksum_enable = igb_get_prop(igb, PROP_TX_HCKSUM_ENABLE,
2802 	    0, 1, 1);
2803 	igb->rx_hcksum_enable = igb_get_prop(igb, PROP_RX_HCKSUM_ENABLE,
2804 	    0, 1, 1);
2805 	igb->lso_enable = igb_get_prop(igb, PROP_LSO_ENABLE,
2806 	    0, 1, 1);
2807 	igb->tx_head_wb_enable = igb_get_prop(igb, PROP_TX_HEAD_WB_ENABLE,
2808 	    0, 1, 1);
2809 
2810 	/*
2811 	 * igb LSO needs the tx h/w checksum support.
2812 	 * Here LSO will be disabled if tx h/w checksum has been disabled.
2813 	 */
2814 	if (igb->tx_hcksum_enable == B_FALSE)
2815 		igb->lso_enable = B_FALSE;
2816 
2817 	igb->tx_copy_thresh = igb_get_prop(igb, PROP_TX_COPY_THRESHOLD,
2818 	    MIN_TX_COPY_THRESHOLD, MAX_TX_COPY_THRESHOLD,
2819 	    DEFAULT_TX_COPY_THRESHOLD);
2820 	igb->tx_recycle_thresh = igb_get_prop(igb, PROP_TX_RECYCLE_THRESHOLD,
2821 	    MIN_TX_RECYCLE_THRESHOLD, MAX_TX_RECYCLE_THRESHOLD,
2822 	    DEFAULT_TX_RECYCLE_THRESHOLD);
2823 	igb->tx_overload_thresh = igb_get_prop(igb, PROP_TX_OVERLOAD_THRESHOLD,
2824 	    MIN_TX_OVERLOAD_THRESHOLD, MAX_TX_OVERLOAD_THRESHOLD,
2825 	    DEFAULT_TX_OVERLOAD_THRESHOLD);
2826 	igb->tx_resched_thresh = igb_get_prop(igb, PROP_TX_RESCHED_THRESHOLD,
2827 	    MIN_TX_RESCHED_THRESHOLD, MAX_TX_RESCHED_THRESHOLD,
2828 	    DEFAULT_TX_RESCHED_THRESHOLD);
2829 
2830 	igb->rx_copy_thresh = igb_get_prop(igb, PROP_RX_COPY_THRESHOLD,
2831 	    MIN_RX_COPY_THRESHOLD, MAX_RX_COPY_THRESHOLD,
2832 	    DEFAULT_RX_COPY_THRESHOLD);
2833 	igb->rx_limit_per_intr = igb_get_prop(igb, PROP_RX_LIMIT_PER_INTR,
2834 	    MIN_RX_LIMIT_PER_INTR, MAX_RX_LIMIT_PER_INTR,
2835 	    DEFAULT_RX_LIMIT_PER_INTR);
2836 
2837 	igb->intr_throttling[0] = igb_get_prop(igb, PROP_INTR_THROTTLING,
2838 	    igb->capab->min_intr_throttle,
2839 	    igb->capab->max_intr_throttle,
2840 	    igb->capab->def_intr_throttle);
2841 
2842 	/*
2843 	 * Max number of multicast addresses
2844 	 */
2845 	igb->mcast_max_num =
2846 	    igb_get_prop(igb, PROP_MCAST_MAX_NUM,
2847 	    MIN_MCAST_NUM, MAX_MCAST_NUM, DEFAULT_MCAST_NUM);
2848 }
2849 
2850 /*
2851  * igb_get_prop - Get a property value out of the configuration file igb.conf
2852  *
2853  * Caller provides the name of the property, a default value, a minimum
2854  * value, and a maximum value.
2855  *
2856  * Return configured value of the property, with default, minimum and
2857  * maximum properly applied.
2858  */
2859 static int
2860 igb_get_prop(igb_t *igb,
2861     char *propname,	/* name of the property */
2862     int minval,		/* minimum acceptable value */
2863     int maxval,		/* maximim acceptable value */
2864     int defval)		/* default value */
2865 {
2866 	int value;
2867 
2868 	/*
2869 	 * Call ddi_prop_get_int() to read the conf settings
2870 	 */
2871 	value = ddi_prop_get_int(DDI_DEV_T_ANY, igb->dip,
2872 	    DDI_PROP_DONTPASS, propname, defval);
2873 
2874 	if (value > maxval)
2875 		value = maxval;
2876 
2877 	if (value < minval)
2878 		value = minval;
2879 
2880 	return (value);
2881 }
2882 
2883 /*
2884  * igb_setup_link - Using the link properties to setup the link
2885  */
2886 int
2887 igb_setup_link(igb_t *igb, boolean_t setup_hw)
2888 {
2889 	struct e1000_mac_info *mac;
2890 	struct e1000_phy_info *phy;
2891 	boolean_t invalid;
2892 
2893 	mac = &igb->hw.mac;
2894 	phy = &igb->hw.phy;
2895 	invalid = B_FALSE;
2896 
2897 	if (igb->param_adv_autoneg_cap == 1) {
2898 		mac->autoneg = B_TRUE;
2899 		phy->autoneg_advertised = 0;
2900 
2901 		/*
2902 		 * 1000hdx is not supported for autonegotiation
2903 		 */
2904 		if (igb->param_adv_1000fdx_cap == 1)
2905 			phy->autoneg_advertised |= ADVERTISE_1000_FULL;
2906 
2907 		if (igb->param_adv_100fdx_cap == 1)
2908 			phy->autoneg_advertised |= ADVERTISE_100_FULL;
2909 
2910 		if (igb->param_adv_100hdx_cap == 1)
2911 			phy->autoneg_advertised |= ADVERTISE_100_HALF;
2912 
2913 		if (igb->param_adv_10fdx_cap == 1)
2914 			phy->autoneg_advertised |= ADVERTISE_10_FULL;
2915 
2916 		if (igb->param_adv_10hdx_cap == 1)
2917 			phy->autoneg_advertised |= ADVERTISE_10_HALF;
2918 
2919 		if (phy->autoneg_advertised == 0)
2920 			invalid = B_TRUE;
2921 	} else {
2922 		mac->autoneg = B_FALSE;
2923 
2924 		/*
2925 		 * 1000fdx and 1000hdx are not supported for forced link
2926 		 */
2927 		if (igb->param_adv_100fdx_cap == 1)
2928 			mac->forced_speed_duplex = ADVERTISE_100_FULL;
2929 		else if (igb->param_adv_100hdx_cap == 1)
2930 			mac->forced_speed_duplex = ADVERTISE_100_HALF;
2931 		else if (igb->param_adv_10fdx_cap == 1)
2932 			mac->forced_speed_duplex = ADVERTISE_10_FULL;
2933 		else if (igb->param_adv_10hdx_cap == 1)
2934 			mac->forced_speed_duplex = ADVERTISE_10_HALF;
2935 		else
2936 			invalid = B_TRUE;
2937 	}
2938 
2939 	if (invalid) {
2940 		igb_notice(igb, "Invalid link settings. Setup link to "
2941 		    "autonegotiation with full link capabilities.");
2942 		mac->autoneg = B_TRUE;
2943 		phy->autoneg_advertised = ADVERTISE_1000_FULL |
2944 		    ADVERTISE_100_FULL | ADVERTISE_100_HALF |
2945 		    ADVERTISE_10_FULL | ADVERTISE_10_HALF;
2946 	}
2947 
2948 	if (setup_hw) {
2949 		if (e1000_setup_link(&igb->hw) != E1000_SUCCESS)
2950 			return (IGB_FAILURE);
2951 	}
2952 
2953 	return (IGB_SUCCESS);
2954 }
2955 
2956 
2957 /*
2958  * igb_is_link_up - Check if the link is up
2959  */
2960 static boolean_t
2961 igb_is_link_up(igb_t *igb)
2962 {
2963 	struct e1000_hw *hw = &igb->hw;
2964 	boolean_t link_up = B_FALSE;
2965 
2966 	ASSERT(mutex_owned(&igb->gen_lock));
2967 
2968 	/*
2969 	 * get_link_status is set in the interrupt handler on link-status-change
2970 	 * or rx sequence error interrupt.  get_link_status will stay
2971 	 * false until the e1000_check_for_link establishes link only
2972 	 * for copper adapters.
2973 	 */
2974 	switch (hw->phy.media_type) {
2975 	case e1000_media_type_copper:
2976 		if (hw->mac.get_link_status) {
2977 			(void) e1000_check_for_link(hw);
2978 			link_up = !hw->mac.get_link_status;
2979 		} else {
2980 			link_up = B_TRUE;
2981 		}
2982 		break;
2983 	case e1000_media_type_fiber:
2984 		(void) e1000_check_for_link(hw);
2985 		link_up = (E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU);
2986 		break;
2987 	case e1000_media_type_internal_serdes:
2988 		(void) e1000_check_for_link(hw);
2989 		link_up = hw->mac.serdes_has_link;
2990 		break;
2991 	}
2992 
2993 	return (link_up);
2994 }
2995 
2996 /*
2997  * igb_link_check - Link status processing
2998  */
2999 static boolean_t
3000 igb_link_check(igb_t *igb)
3001 {
3002 	struct e1000_hw *hw = &igb->hw;
3003 	uint16_t speed = 0, duplex = 0;
3004 	boolean_t link_changed = B_FALSE;
3005 
3006 	ASSERT(mutex_owned(&igb->gen_lock));
3007 
3008 	if (igb_is_link_up(igb)) {
3009 		/*
3010 		 * The Link is up, check whether it was marked as down earlier
3011 		 */
3012 		if (igb->link_state != LINK_STATE_UP) {
3013 			(void) e1000_get_speed_and_duplex(hw, &speed, &duplex);
3014 			igb->link_speed = speed;
3015 			igb->link_duplex = duplex;
3016 			igb->link_state = LINK_STATE_UP;
3017 			igb->link_down_timeout = 0;
3018 			link_changed = B_TRUE;
3019 			if (!igb->link_complete)
3020 				igb_stop_link_timer(igb);
3021 		}
3022 	} else if (igb->link_complete) {
3023 		if (igb->link_state != LINK_STATE_DOWN) {
3024 			igb->link_speed = 0;
3025 			igb->link_duplex = 0;
3026 			igb->link_state = LINK_STATE_DOWN;
3027 			link_changed = B_TRUE;
3028 		}
3029 
3030 		if (igb->igb_state & IGB_STARTED) {
3031 			if (igb->link_down_timeout < MAX_LINK_DOWN_TIMEOUT) {
3032 				igb->link_down_timeout++;
3033 			} else if (igb->link_down_timeout ==
3034 			    MAX_LINK_DOWN_TIMEOUT) {
3035 				igb_tx_clean(igb);
3036 				igb->link_down_timeout++;
3037 			}
3038 		}
3039 	}
3040 
3041 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
3042 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
3043 		return (B_FALSE);
3044 	}
3045 
3046 	return (link_changed);
3047 }
3048 
3049 /*
3050  * igb_local_timer - driver watchdog function
3051  *
3052  * This function will handle the hardware stall check, link status
3053  * check and other routines.
3054  */
3055 static void
3056 igb_local_timer(void *arg)
3057 {
3058 	igb_t *igb = (igb_t *)arg;
3059 	boolean_t link_changed = B_FALSE;
3060 
3061 	if (igb->igb_state & IGB_ERROR) {
3062 		igb->reset_count++;
3063 		if (igb_reset(igb) == IGB_SUCCESS)
3064 			ddi_fm_service_impact(igb->dip, DDI_SERVICE_RESTORED);
3065 
3066 		igb_restart_watchdog_timer(igb);
3067 		return;
3068 	}
3069 
3070 	if (igb_stall_check(igb) || (igb->igb_state & IGB_STALL)) {
3071 		igb_fm_ereport(igb, DDI_FM_DEVICE_STALL);
3072 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
3073 		igb->reset_count++;
3074 		if (igb_reset(igb) == IGB_SUCCESS)
3075 			ddi_fm_service_impact(igb->dip, DDI_SERVICE_RESTORED);
3076 
3077 		igb_restart_watchdog_timer(igb);
3078 		return;
3079 	}
3080 
3081 	mutex_enter(&igb->gen_lock);
3082 	if (!(igb->igb_state & IGB_SUSPENDED) && (igb->igb_state & IGB_STARTED))
3083 		link_changed = igb_link_check(igb);
3084 	mutex_exit(&igb->gen_lock);
3085 
3086 	if (link_changed)
3087 		mac_link_update(igb->mac_hdl, igb->link_state);
3088 
3089 	igb_restart_watchdog_timer(igb);
3090 }
3091 
3092 /*
3093  * igb_link_timer - link setup timer function
3094  *
3095  * It is called when the timer for link setup is expired, which indicates
3096  * the completion of the link setup. The link state will not be updated
3097  * until the link setup is completed. And the link state will not be sent
3098  * to the upper layer through mac_link_update() in this function. It will
3099  * be updated in the local timer routine or the interrupts service routine
3100  * after the interface is started (plumbed).
3101  */
3102 static void
3103 igb_link_timer(void *arg)
3104 {
3105 	igb_t *igb = (igb_t *)arg;
3106 
3107 	mutex_enter(&igb->link_lock);
3108 	igb->link_complete = B_TRUE;
3109 	igb->link_tid = 0;
3110 	mutex_exit(&igb->link_lock);
3111 }
3112 /*
3113  * igb_stall_check - check for transmit stall
3114  *
3115  * This function checks if the adapter is stalled (in transmit).
3116  *
3117  * It is called each time the watchdog timeout is invoked.
3118  * If the transmit descriptor reclaim continuously fails,
3119  * the watchdog value will increment by 1. If the watchdog
3120  * value exceeds the threshold, the igb is assumed to
3121  * have stalled and need to be reset.
3122  */
3123 static boolean_t
3124 igb_stall_check(igb_t *igb)
3125 {
3126 	igb_tx_ring_t *tx_ring;
3127 	struct e1000_hw *hw = &igb->hw;
3128 	boolean_t result;
3129 	int i;
3130 
3131 	if (igb->link_state != LINK_STATE_UP)
3132 		return (B_FALSE);
3133 
3134 	/*
3135 	 * If any tx ring is stalled, we'll reset the chipset
3136 	 */
3137 	result = B_FALSE;
3138 	for (i = 0; i < igb->num_tx_rings; i++) {
3139 		tx_ring = &igb->tx_rings[i];
3140 
3141 		if (tx_ring->recycle_fail > 0)
3142 			tx_ring->stall_watchdog++;
3143 		else
3144 			tx_ring->stall_watchdog = 0;
3145 
3146 		if (tx_ring->stall_watchdog >= STALL_WATCHDOG_TIMEOUT) {
3147 			result = B_TRUE;
3148 			if (hw->mac.type == e1000_82580) {
3149 				hw->dev_spec._82575.global_device_reset
3150 				    = B_TRUE;
3151 			}
3152 			break;
3153 		}
3154 	}
3155 
3156 	if (result) {
3157 		tx_ring->stall_watchdog = 0;
3158 		tx_ring->recycle_fail = 0;
3159 	}
3160 
3161 	return (result);
3162 }
3163 
3164 
3165 /*
3166  * is_valid_mac_addr - Check if the mac address is valid
3167  */
3168 static boolean_t
3169 is_valid_mac_addr(uint8_t *mac_addr)
3170 {
3171 	const uint8_t addr_test1[6] = { 0, 0, 0, 0, 0, 0 };
3172 	const uint8_t addr_test2[6] =
3173 	    { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF };
3174 
3175 	if (!(bcmp(addr_test1, mac_addr, ETHERADDRL)) ||
3176 	    !(bcmp(addr_test2, mac_addr, ETHERADDRL)))
3177 		return (B_FALSE);
3178 
3179 	return (B_TRUE);
3180 }
3181 
3182 static boolean_t
3183 igb_find_mac_address(igb_t *igb)
3184 {
3185 	struct e1000_hw *hw = &igb->hw;
3186 #ifdef __sparc
3187 	uchar_t *bytes;
3188 	struct ether_addr sysaddr;
3189 	uint_t nelts;
3190 	int err;
3191 	boolean_t found = B_FALSE;
3192 
3193 	/*
3194 	 * The "vendor's factory-set address" may already have
3195 	 * been extracted from the chip, but if the property
3196 	 * "local-mac-address" is set we use that instead.
3197 	 *
3198 	 * We check whether it looks like an array of 6
3199 	 * bytes (which it should, if OBP set it).  If we can't
3200 	 * make sense of it this way, we'll ignore it.
3201 	 */
3202 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip,
3203 	    DDI_PROP_DONTPASS, "local-mac-address", &bytes, &nelts);
3204 	if (err == DDI_PROP_SUCCESS) {
3205 		if (nelts == ETHERADDRL) {
3206 			while (nelts--)
3207 				hw->mac.addr[nelts] = bytes[nelts];
3208 			found = B_TRUE;
3209 		}
3210 		ddi_prop_free(bytes);
3211 	}
3212 
3213 	/*
3214 	 * Look up the OBP property "local-mac-address?". If the user has set
3215 	 * 'local-mac-address? = false', use "the system address" instead.
3216 	 */
3217 	if (ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip, 0,
3218 	    "local-mac-address?", &bytes, &nelts) == DDI_PROP_SUCCESS) {
3219 		if (strncmp("false", (caddr_t)bytes, (size_t)nelts) == 0) {
3220 			if (localetheraddr(NULL, &sysaddr) != 0) {
3221 				bcopy(&sysaddr, hw->mac.addr, ETHERADDRL);
3222 				found = B_TRUE;
3223 			}
3224 		}
3225 		ddi_prop_free(bytes);
3226 	}
3227 
3228 	/*
3229 	 * Finally(!), if there's a valid "mac-address" property (created
3230 	 * if we netbooted from this interface), we must use this instead
3231 	 * of any of the above to ensure that the NFS/install server doesn't
3232 	 * get confused by the address changing as Solaris takes over!
3233 	 */
3234 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip,
3235 	    DDI_PROP_DONTPASS, "mac-address", &bytes, &nelts);
3236 	if (err == DDI_PROP_SUCCESS) {
3237 		if (nelts == ETHERADDRL) {
3238 			while (nelts--)
3239 				hw->mac.addr[nelts] = bytes[nelts];
3240 			found = B_TRUE;
3241 		}
3242 		ddi_prop_free(bytes);
3243 	}
3244 
3245 	if (found) {
3246 		bcopy(hw->mac.addr, hw->mac.perm_addr, ETHERADDRL);
3247 		return (B_TRUE);
3248 	}
3249 #endif
3250 
3251 	/*
3252 	 * Read the device MAC address from the EEPROM
3253 	 */
3254 	if (e1000_read_mac_addr(hw) != E1000_SUCCESS)
3255 		return (B_FALSE);
3256 
3257 	return (B_TRUE);
3258 }
3259 
3260 #pragma inline(igb_arm_watchdog_timer)
3261 
3262 static void
3263 igb_arm_watchdog_timer(igb_t *igb)
3264 {
3265 	/*
3266 	 * Fire a watchdog timer
3267 	 */
3268 	igb->watchdog_tid =
3269 	    timeout(igb_local_timer,
3270 	    (void *)igb, 1 * drv_usectohz(1000000));
3271 
3272 }
3273 
3274 /*
3275  * igb_enable_watchdog_timer - Enable and start the driver watchdog timer
3276  */
3277 void
3278 igb_enable_watchdog_timer(igb_t *igb)
3279 {
3280 	mutex_enter(&igb->watchdog_lock);
3281 
3282 	if (!igb->watchdog_enable) {
3283 		igb->watchdog_enable = B_TRUE;
3284 		igb->watchdog_start = B_TRUE;
3285 		igb_arm_watchdog_timer(igb);
3286 	}
3287 
3288 	mutex_exit(&igb->watchdog_lock);
3289 
3290 }
3291 
3292 /*
3293  * igb_disable_watchdog_timer - Disable and stop the driver watchdog timer
3294  */
3295 void
3296 igb_disable_watchdog_timer(igb_t *igb)
3297 {
3298 	timeout_id_t tid;
3299 
3300 	mutex_enter(&igb->watchdog_lock);
3301 
3302 	igb->watchdog_enable = B_FALSE;
3303 	igb->watchdog_start = B_FALSE;
3304 	tid = igb->watchdog_tid;
3305 	igb->watchdog_tid = 0;
3306 
3307 	mutex_exit(&igb->watchdog_lock);
3308 
3309 	if (tid != 0)
3310 		(void) untimeout(tid);
3311 
3312 }
3313 
3314 /*
3315  * igb_start_watchdog_timer - Start the driver watchdog timer
3316  */
3317 static void
3318 igb_start_watchdog_timer(igb_t *igb)
3319 {
3320 	mutex_enter(&igb->watchdog_lock);
3321 
3322 	if (igb->watchdog_enable) {
3323 		if (!igb->watchdog_start) {
3324 			igb->watchdog_start = B_TRUE;
3325 			igb_arm_watchdog_timer(igb);
3326 		}
3327 	}
3328 
3329 	mutex_exit(&igb->watchdog_lock);
3330 }
3331 
3332 /*
3333  * igb_restart_watchdog_timer - Restart the driver watchdog timer
3334  */
3335 static void
3336 igb_restart_watchdog_timer(igb_t *igb)
3337 {
3338 	mutex_enter(&igb->watchdog_lock);
3339 
3340 	if (igb->watchdog_start)
3341 		igb_arm_watchdog_timer(igb);
3342 
3343 	mutex_exit(&igb->watchdog_lock);
3344 }
3345 
3346 /*
3347  * igb_stop_watchdog_timer - Stop the driver watchdog timer
3348  */
3349 static void
3350 igb_stop_watchdog_timer(igb_t *igb)
3351 {
3352 	timeout_id_t tid;
3353 
3354 	mutex_enter(&igb->watchdog_lock);
3355 
3356 	igb->watchdog_start = B_FALSE;
3357 	tid = igb->watchdog_tid;
3358 	igb->watchdog_tid = 0;
3359 
3360 	mutex_exit(&igb->watchdog_lock);
3361 
3362 	if (tid != 0)
3363 		(void) untimeout(tid);
3364 }
3365 
3366 /*
3367  * igb_start_link_timer - Start the link setup timer
3368  */
3369 static void
3370 igb_start_link_timer(struct igb *igb)
3371 {
3372 	struct e1000_hw *hw = &igb->hw;
3373 	clock_t link_timeout;
3374 
3375 	if (hw->mac.autoneg)
3376 		link_timeout = PHY_AUTO_NEG_LIMIT *
3377 		    drv_usectohz(100000);
3378 	else
3379 		link_timeout = PHY_FORCE_LIMIT * drv_usectohz(100000);
3380 
3381 	mutex_enter(&igb->link_lock);
3382 	if (hw->phy.autoneg_wait_to_complete) {
3383 		igb->link_complete = B_TRUE;
3384 	} else {
3385 		igb->link_complete = B_FALSE;
3386 		igb->link_tid = timeout(igb_link_timer, (void *)igb,
3387 		    link_timeout);
3388 	}
3389 	mutex_exit(&igb->link_lock);
3390 }
3391 
3392 /*
3393  * igb_stop_link_timer - Stop the link setup timer
3394  */
3395 static void
3396 igb_stop_link_timer(struct igb *igb)
3397 {
3398 	timeout_id_t tid;
3399 
3400 	mutex_enter(&igb->link_lock);
3401 	igb->link_complete = B_TRUE;
3402 	tid = igb->link_tid;
3403 	igb->link_tid = 0;
3404 	mutex_exit(&igb->link_lock);
3405 
3406 	if (tid != 0)
3407 		(void) untimeout(tid);
3408 }
3409 
3410 /*
3411  * igb_disable_adapter_interrupts - Clear/disable all hardware interrupts
3412  */
3413 static void
3414 igb_disable_adapter_interrupts(igb_t *igb)
3415 {
3416 	struct e1000_hw *hw = &igb->hw;
3417 
3418 	/*
3419 	 * Set the IMC register to mask all the interrupts,
3420 	 * including the tx interrupts.
3421 	 */
3422 	E1000_WRITE_REG(hw, E1000_IMC, ~0);
3423 	E1000_WRITE_REG(hw, E1000_IAM, 0);
3424 
3425 	/*
3426 	 * Additional disabling for MSI-X
3427 	 */
3428 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3429 		E1000_WRITE_REG(hw, E1000_EIMC, ~0);
3430 		E1000_WRITE_REG(hw, E1000_EIAC, 0);
3431 		E1000_WRITE_REG(hw, E1000_EIAM, 0);
3432 	}
3433 
3434 	E1000_WRITE_FLUSH(hw);
3435 }
3436 
3437 /*
3438  * igb_enable_adapter_interrupts_82580 - Enable NIC interrupts for 82580
3439  */
3440 static void
3441 igb_enable_adapter_interrupts_82580(igb_t *igb)
3442 {
3443 	struct e1000_hw *hw = &igb->hw;
3444 
3445 	/* Clear any pending interrupts */
3446 	(void) E1000_READ_REG(hw, E1000_ICR);
3447 	igb->ims_mask |= E1000_IMS_DRSTA;
3448 
3449 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3450 
3451 		/* Interrupt enabling for MSI-X */
3452 		E1000_WRITE_REG(hw, E1000_EIMS, igb->eims_mask);
3453 		E1000_WRITE_REG(hw, E1000_EIAC, igb->eims_mask);
3454 		igb->ims_mask = (E1000_IMS_LSC | E1000_IMS_DRSTA);
3455 		E1000_WRITE_REG(hw, E1000_IMS, igb->ims_mask);
3456 	} else { /* Interrupt enabling for MSI and legacy */
3457 		E1000_WRITE_REG(hw, E1000_IVAR0, E1000_IVAR_VALID);
3458 		igb->ims_mask = IMS_ENABLE_MASK | E1000_IMS_TXQE;
3459 		igb->ims_mask |= E1000_IMS_DRSTA;
3460 		E1000_WRITE_REG(hw, E1000_IMS, igb->ims_mask);
3461 	}
3462 
3463 	/* Disable auto-mask for ICR interrupt bits */
3464 	E1000_WRITE_REG(hw, E1000_IAM, 0);
3465 
3466 	E1000_WRITE_FLUSH(hw);
3467 }
3468 
3469 /*
3470  * igb_enable_adapter_interrupts_82576 - Enable NIC interrupts for 82576
3471  */
3472 static void
3473 igb_enable_adapter_interrupts_82576(igb_t *igb)
3474 {
3475 	struct e1000_hw *hw = &igb->hw;
3476 
3477 	/* Clear any pending interrupts */
3478 	(void) E1000_READ_REG(hw, E1000_ICR);
3479 
3480 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3481 
3482 		/* Interrupt enabling for MSI-X */
3483 		E1000_WRITE_REG(hw, E1000_EIMS, igb->eims_mask);
3484 		E1000_WRITE_REG(hw, E1000_EIAC, igb->eims_mask);
3485 		igb->ims_mask = E1000_IMS_LSC;
3486 		E1000_WRITE_REG(hw, E1000_IMS, E1000_IMS_LSC);
3487 	} else {
3488 		/* Interrupt enabling for MSI and legacy */
3489 		E1000_WRITE_REG(hw, E1000_IVAR0, E1000_IVAR_VALID);
3490 		igb->ims_mask = IMS_ENABLE_MASK | E1000_IMS_TXQE;
3491 		E1000_WRITE_REG(hw, E1000_IMS,
3492 		    (IMS_ENABLE_MASK | E1000_IMS_TXQE));
3493 	}
3494 
3495 	/* Disable auto-mask for ICR interrupt bits */
3496 	E1000_WRITE_REG(hw, E1000_IAM, 0);
3497 
3498 	E1000_WRITE_FLUSH(hw);
3499 }
3500 
3501 /*
3502  * igb_enable_adapter_interrupts_82575 - Enable NIC interrupts for 82575
3503  */
3504 static void
3505 igb_enable_adapter_interrupts_82575(igb_t *igb)
3506 {
3507 	struct e1000_hw *hw = &igb->hw;
3508 	uint32_t reg;
3509 
3510 	/* Clear any pending interrupts */
3511 	(void) E1000_READ_REG(hw, E1000_ICR);
3512 
3513 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3514 		/* Interrupt enabling for MSI-X */
3515 		E1000_WRITE_REG(hw, E1000_EIMS, igb->eims_mask);
3516 		E1000_WRITE_REG(hw, E1000_EIAC, igb->eims_mask);
3517 		igb->ims_mask = E1000_IMS_LSC;
3518 		E1000_WRITE_REG(hw, E1000_IMS, E1000_IMS_LSC);
3519 
3520 		/* Enable MSI-X PBA support */
3521 		reg = E1000_READ_REG(hw, E1000_CTRL_EXT);
3522 		reg |= E1000_CTRL_EXT_PBA_CLR;
3523 
3524 		/* Non-selective interrupt clear-on-read */
3525 		reg |= E1000_CTRL_EXT_IRCA;	/* Called NSICR in the EAS */
3526 
3527 		E1000_WRITE_REG(hw, E1000_CTRL_EXT, reg);
3528 	} else {
3529 		/* Interrupt enabling for MSI and legacy */
3530 		igb->ims_mask = IMS_ENABLE_MASK;
3531 		E1000_WRITE_REG(hw, E1000_IMS, IMS_ENABLE_MASK);
3532 	}
3533 
3534 	E1000_WRITE_FLUSH(hw);
3535 }
3536 
3537 /*
3538  * Loopback Support
3539  */
3540 static lb_property_t lb_normal =
3541 	{ normal,	"normal",	IGB_LB_NONE		};
3542 static lb_property_t lb_external =
3543 	{ external,	"External",	IGB_LB_EXTERNAL		};
3544 static lb_property_t lb_mac =
3545 	{ internal,	"MAC",		IGB_LB_INTERNAL_MAC	};
3546 static lb_property_t lb_phy =
3547 	{ internal,	"PHY",		IGB_LB_INTERNAL_PHY	};
3548 static lb_property_t lb_serdes =
3549 	{ internal,	"SerDes",	IGB_LB_INTERNAL_SERDES	};
3550 
3551 enum ioc_reply
3552 igb_loopback_ioctl(igb_t *igb, struct iocblk *iocp, mblk_t *mp)
3553 {
3554 	lb_info_sz_t *lbsp;
3555 	lb_property_t *lbpp;
3556 	struct e1000_hw *hw;
3557 	uint32_t *lbmp;
3558 	uint32_t size;
3559 	uint32_t value;
3560 
3561 	hw = &igb->hw;
3562 
3563 	if (mp->b_cont == NULL)
3564 		return (IOC_INVAL);
3565 
3566 	switch (iocp->ioc_cmd) {
3567 	default:
3568 		return (IOC_INVAL);
3569 
3570 	case LB_GET_INFO_SIZE:
3571 		size = sizeof (lb_info_sz_t);
3572 		if (iocp->ioc_count != size)
3573 			return (IOC_INVAL);
3574 
3575 		value = sizeof (lb_normal);
3576 		value += sizeof (lb_mac);
3577 		if (hw->phy.media_type == e1000_media_type_copper)
3578 			value += sizeof (lb_phy);
3579 		else
3580 			value += sizeof (lb_serdes);
3581 		value += sizeof (lb_external);
3582 
3583 		lbsp = (lb_info_sz_t *)(uintptr_t)mp->b_cont->b_rptr;
3584 		*lbsp = value;
3585 		break;
3586 
3587 	case LB_GET_INFO:
3588 		value = sizeof (lb_normal);
3589 		value += sizeof (lb_mac);
3590 		if (hw->phy.media_type == e1000_media_type_copper)
3591 			value += sizeof (lb_phy);
3592 		else
3593 			value += sizeof (lb_serdes);
3594 		value += sizeof (lb_external);
3595 
3596 		size = value;
3597 		if (iocp->ioc_count != size)
3598 			return (IOC_INVAL);
3599 
3600 		value = 0;
3601 		lbpp = (lb_property_t *)(uintptr_t)mp->b_cont->b_rptr;
3602 
3603 		lbpp[value++] = lb_normal;
3604 		lbpp[value++] = lb_mac;
3605 		if (hw->phy.media_type == e1000_media_type_copper)
3606 			lbpp[value++] = lb_phy;
3607 		else
3608 			lbpp[value++] = lb_serdes;
3609 		lbpp[value++] = lb_external;
3610 		break;
3611 
3612 	case LB_GET_MODE:
3613 		size = sizeof (uint32_t);
3614 		if (iocp->ioc_count != size)
3615 			return (IOC_INVAL);
3616 
3617 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
3618 		*lbmp = igb->loopback_mode;
3619 		break;
3620 
3621 	case LB_SET_MODE:
3622 		size = 0;
3623 		if (iocp->ioc_count != sizeof (uint32_t))
3624 			return (IOC_INVAL);
3625 
3626 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
3627 		if (!igb_set_loopback_mode(igb, *lbmp))
3628 			return (IOC_INVAL);
3629 		break;
3630 	}
3631 
3632 	iocp->ioc_count = size;
3633 	iocp->ioc_error = 0;
3634 
3635 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
3636 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
3637 		return (IOC_INVAL);
3638 	}
3639 
3640 	return (IOC_REPLY);
3641 }
3642 
3643 /*
3644  * igb_set_loopback_mode - Setup loopback based on the loopback mode
3645  */
3646 static boolean_t
3647 igb_set_loopback_mode(igb_t *igb, uint32_t mode)
3648 {
3649 	struct e1000_hw *hw;
3650 	int i;
3651 
3652 	if (mode == igb->loopback_mode)
3653 		return (B_TRUE);
3654 
3655 	hw = &igb->hw;
3656 
3657 	igb->loopback_mode = mode;
3658 
3659 	if (mode == IGB_LB_NONE) {
3660 		/* Reset the chip */
3661 		hw->phy.autoneg_wait_to_complete = B_TRUE;
3662 		(void) igb_reset(igb);
3663 		hw->phy.autoneg_wait_to_complete = B_FALSE;
3664 		return (B_TRUE);
3665 	}
3666 
3667 	mutex_enter(&igb->gen_lock);
3668 
3669 	switch (mode) {
3670 	default:
3671 		mutex_exit(&igb->gen_lock);
3672 		return (B_FALSE);
3673 
3674 	case IGB_LB_EXTERNAL:
3675 		igb_set_external_loopback(igb);
3676 		break;
3677 
3678 	case IGB_LB_INTERNAL_MAC:
3679 		igb_set_internal_mac_loopback(igb);
3680 		break;
3681 
3682 	case IGB_LB_INTERNAL_PHY:
3683 		igb_set_internal_phy_loopback(igb);
3684 		break;
3685 
3686 	case IGB_LB_INTERNAL_SERDES:
3687 		igb_set_internal_serdes_loopback(igb);
3688 		break;
3689 	}
3690 
3691 	mutex_exit(&igb->gen_lock);
3692 
3693 	/*
3694 	 * When external loopback is set, wait up to 1000ms to get the link up.
3695 	 * According to test, 1000ms can work and it's an experimental value.
3696 	 */
3697 	if (mode == IGB_LB_EXTERNAL) {
3698 		for (i = 0; i <= 10; i++) {
3699 			mutex_enter(&igb->gen_lock);
3700 			(void) igb_link_check(igb);
3701 			mutex_exit(&igb->gen_lock);
3702 
3703 			if (igb->link_state == LINK_STATE_UP)
3704 				break;
3705 
3706 			msec_delay(100);
3707 		}
3708 
3709 		if (igb->link_state != LINK_STATE_UP) {
3710 			/*
3711 			 * Does not support external loopback.
3712 			 * Reset driver to loopback none.
3713 			 */
3714 			igb->loopback_mode = IGB_LB_NONE;
3715 
3716 			/* Reset the chip */
3717 			hw->phy.autoneg_wait_to_complete = B_TRUE;
3718 			(void) igb_reset(igb);
3719 			hw->phy.autoneg_wait_to_complete = B_FALSE;
3720 
3721 			IGB_DEBUGLOG_0(igb, "Set external loopback failed, "
3722 			    "reset to loopback none.");
3723 
3724 			return (B_FALSE);
3725 		}
3726 	}
3727 
3728 	return (B_TRUE);
3729 }
3730 
3731 /*
3732  * igb_set_external_loopback - Set the external loopback mode
3733  */
3734 static void
3735 igb_set_external_loopback(igb_t *igb)
3736 {
3737 	struct e1000_hw *hw;
3738 
3739 	hw = &igb->hw;
3740 
3741 	/* Set phy to known state */
3742 	(void) e1000_phy_hw_reset(hw);
3743 
3744 	(void) e1000_write_phy_reg(hw, 0x0, 0x0140);
3745 	(void) e1000_write_phy_reg(hw, 0x9, 0x1b00);
3746 	(void) e1000_write_phy_reg(hw, 0x12, 0x1610);
3747 	(void) e1000_write_phy_reg(hw, 0x1f37, 0x3f1c);
3748 }
3749 
3750 /*
3751  * igb_set_internal_mac_loopback - Set the internal MAC loopback mode
3752  */
3753 static void
3754 igb_set_internal_mac_loopback(igb_t *igb)
3755 {
3756 	struct e1000_hw *hw;
3757 	uint32_t ctrl;
3758 	uint32_t rctl;
3759 	uint32_t ctrl_ext;
3760 	uint16_t phy_ctrl;
3761 	uint16_t phy_status;
3762 
3763 	hw = &igb->hw;
3764 
3765 	(void) e1000_read_phy_reg(hw, PHY_CONTROL, &phy_ctrl);
3766 	phy_ctrl &= ~MII_CR_AUTO_NEG_EN;
3767 	(void) e1000_write_phy_reg(hw, PHY_CONTROL, phy_ctrl);
3768 
3769 	(void) e1000_read_phy_reg(hw, PHY_STATUS, &phy_status);
3770 
3771 	/* Set link mode to PHY (00b) in the Extended Control register */
3772 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3773 	ctrl_ext &= ~E1000_CTRL_EXT_LINK_MODE_MASK;
3774 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3775 
3776 	/* Set the Device Control register */
3777 	ctrl = E1000_READ_REG(hw, E1000_CTRL);
3778 	if (!(phy_status & MII_SR_LINK_STATUS))
3779 		ctrl |= E1000_CTRL_ILOS; /* Set ILOS when the link is down */
3780 	ctrl &= ~E1000_CTRL_SPD_SEL;	/* Clear the speed sel bits */
3781 	ctrl |= (E1000_CTRL_SLU |	/* Force link up */
3782 	    E1000_CTRL_FRCSPD |		/* Force speed */
3783 	    E1000_CTRL_FRCDPX |		/* Force duplex */
3784 	    E1000_CTRL_SPD_1000 |	/* Force speed to 1000 */
3785 	    E1000_CTRL_FD);		/* Force full duplex */
3786 
3787 	E1000_WRITE_REG(hw, E1000_CTRL, ctrl);
3788 
3789 	/* Set the Receive Control register */
3790 	rctl = E1000_READ_REG(hw, E1000_RCTL);
3791 	rctl &= ~E1000_RCTL_LBM_TCVR;
3792 	rctl |= E1000_RCTL_LBM_MAC;
3793 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
3794 }
3795 
3796 /*
3797  * igb_set_internal_phy_loopback - Set the internal PHY loopback mode
3798  */
3799 static void
3800 igb_set_internal_phy_loopback(igb_t *igb)
3801 {
3802 	struct e1000_hw *hw;
3803 	uint32_t ctrl_ext;
3804 	uint16_t phy_ctrl;
3805 	uint16_t phy_pconf;
3806 
3807 	hw = &igb->hw;
3808 
3809 	/* Set link mode to PHY (00b) in the Extended Control register */
3810 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3811 	ctrl_ext &= ~E1000_CTRL_EXT_LINK_MODE_MASK;
3812 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3813 
3814 	/*
3815 	 * Set PHY control register (0x4140):
3816 	 *    Set full duplex mode
3817 	 *    Set loopback bit
3818 	 *    Clear auto-neg enable bit
3819 	 *    Set PHY speed
3820 	 */
3821 	phy_ctrl = MII_CR_FULL_DUPLEX | MII_CR_SPEED_1000 | MII_CR_LOOPBACK;
3822 	(void) e1000_write_phy_reg(hw, PHY_CONTROL, phy_ctrl);
3823 
3824 	/* Set the link disable bit in the Port Configuration register */
3825 	(void) e1000_read_phy_reg(hw, 0x10, &phy_pconf);
3826 	phy_pconf |= (uint16_t)1 << 14;
3827 	(void) e1000_write_phy_reg(hw, 0x10, phy_pconf);
3828 }
3829 
3830 /*
3831  * igb_set_internal_serdes_loopback - Set the internal SerDes loopback mode
3832  */
3833 static void
3834 igb_set_internal_serdes_loopback(igb_t *igb)
3835 {
3836 	struct e1000_hw *hw;
3837 	uint32_t ctrl_ext;
3838 	uint32_t ctrl;
3839 	uint32_t pcs_lctl;
3840 	uint32_t connsw;
3841 
3842 	hw = &igb->hw;
3843 
3844 	/* Set link mode to SerDes (11b) in the Extended Control register */
3845 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3846 	ctrl_ext |= E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
3847 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3848 
3849 	/* Configure the SerDes to loopback */
3850 	E1000_WRITE_REG(hw, E1000_SCTL, 0x410);
3851 
3852 	/* Set Device Control register */
3853 	ctrl = E1000_READ_REG(hw, E1000_CTRL);
3854 	ctrl |= (E1000_CTRL_FD |	/* Force full duplex */
3855 	    E1000_CTRL_SLU);		/* Force link up */
3856 	ctrl &= ~(E1000_CTRL_RFCE |	/* Disable receive flow control */
3857 	    E1000_CTRL_TFCE |		/* Disable transmit flow control */
3858 	    E1000_CTRL_LRST);		/* Clear link reset */
3859 	E1000_WRITE_REG(hw, E1000_CTRL, ctrl);
3860 
3861 	/* Set PCS Link Control register */
3862 	pcs_lctl = E1000_READ_REG(hw, E1000_PCS_LCTL);
3863 	pcs_lctl |= (E1000_PCS_LCTL_FORCE_LINK |
3864 	    E1000_PCS_LCTL_FSD |
3865 	    E1000_PCS_LCTL_FDV_FULL |
3866 	    E1000_PCS_LCTL_FLV_LINK_UP);
3867 	pcs_lctl &= ~E1000_PCS_LCTL_AN_ENABLE;
3868 	E1000_WRITE_REG(hw, E1000_PCS_LCTL, pcs_lctl);
3869 
3870 	/* Set the Copper/Fiber Switch Control - CONNSW register */
3871 	connsw = E1000_READ_REG(hw, E1000_CONNSW);
3872 	connsw &= ~E1000_CONNSW_ENRGSRC;
3873 	E1000_WRITE_REG(hw, E1000_CONNSW, connsw);
3874 }
3875 
3876 #pragma inline(igb_intr_rx_work)
3877 /*
3878  * igb_intr_rx_work - rx processing of ISR
3879  */
3880 static void
3881 igb_intr_rx_work(igb_rx_ring_t *rx_ring)
3882 {
3883 	mblk_t *mp;
3884 
3885 	mutex_enter(&rx_ring->rx_lock);
3886 	mp = igb_rx(rx_ring, IGB_NO_POLL);
3887 	mutex_exit(&rx_ring->rx_lock);
3888 
3889 	if (mp != NULL)
3890 		mac_rx_ring(rx_ring->igb->mac_hdl, rx_ring->ring_handle, mp,
3891 		    rx_ring->ring_gen_num);
3892 }
3893 
3894 #pragma inline(igb_intr_tx_work)
3895 /*
3896  * igb_intr_tx_work - tx processing of ISR
3897  */
3898 static void
3899 igb_intr_tx_work(igb_tx_ring_t *tx_ring)
3900 {
3901 	igb_t *igb = tx_ring->igb;
3902 
3903 	/* Recycle the tx descriptors */
3904 	tx_ring->tx_recycle(tx_ring);
3905 
3906 	/* Schedule the re-transmit */
3907 	if (tx_ring->reschedule &&
3908 	    (tx_ring->tbd_free >= igb->tx_resched_thresh)) {
3909 		tx_ring->reschedule = B_FALSE;
3910 		mac_tx_ring_update(tx_ring->igb->mac_hdl, tx_ring->ring_handle);
3911 		IGB_DEBUG_STAT(tx_ring->stat_reschedule);
3912 	}
3913 }
3914 
3915 #pragma inline(igb_intr_link_work)
3916 /*
3917  * igb_intr_link_work - link-status-change processing of ISR
3918  */
3919 static void
3920 igb_intr_link_work(igb_t *igb)
3921 {
3922 	boolean_t link_changed;
3923 
3924 	igb_stop_watchdog_timer(igb);
3925 
3926 	mutex_enter(&igb->gen_lock);
3927 
3928 	/*
3929 	 * Because we got a link-status-change interrupt, force
3930 	 * e1000_check_for_link() to look at phy
3931 	 */
3932 	igb->hw.mac.get_link_status = B_TRUE;
3933 
3934 	/* igb_link_check takes care of link status change */
3935 	link_changed = igb_link_check(igb);
3936 
3937 	/* Get new phy state */
3938 	igb_get_phy_state(igb);
3939 
3940 	mutex_exit(&igb->gen_lock);
3941 
3942 	if (link_changed)
3943 		mac_link_update(igb->mac_hdl, igb->link_state);
3944 
3945 	igb_start_watchdog_timer(igb);
3946 }
3947 
3948 /*
3949  * igb_intr_legacy - Interrupt handler for legacy interrupts
3950  */
3951 static uint_t
3952 igb_intr_legacy(void *arg1, void *arg2)
3953 {
3954 	igb_t *igb = (igb_t *)arg1;
3955 	igb_tx_ring_t *tx_ring;
3956 	uint32_t icr;
3957 	mblk_t *mp;
3958 	boolean_t tx_reschedule;
3959 	boolean_t link_changed;
3960 	uint_t result;
3961 
3962 	_NOTE(ARGUNUSED(arg2));
3963 
3964 	mutex_enter(&igb->gen_lock);
3965 
3966 	if (igb->igb_state & IGB_SUSPENDED) {
3967 		mutex_exit(&igb->gen_lock);
3968 		return (DDI_INTR_UNCLAIMED);
3969 	}
3970 
3971 	mp = NULL;
3972 	tx_reschedule = B_FALSE;
3973 	link_changed = B_FALSE;
3974 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
3975 
3976 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
3977 		mutex_exit(&igb->gen_lock);
3978 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
3979 		atomic_or_32(&igb->igb_state, IGB_ERROR);
3980 		return (DDI_INTR_UNCLAIMED);
3981 	}
3982 
3983 	if (icr & E1000_ICR_INT_ASSERTED) {
3984 		/*
3985 		 * E1000_ICR_INT_ASSERTED bit was set:
3986 		 * Read(Clear) the ICR, claim this interrupt,
3987 		 * look for work to do.
3988 		 */
3989 		ASSERT(igb->num_rx_rings == 1);
3990 		ASSERT(igb->num_tx_rings == 1);
3991 
3992 		/* Make sure all interrupt causes cleared */
3993 		(void) E1000_READ_REG(&igb->hw, E1000_EICR);
3994 
3995 		if (icr & E1000_ICR_RXT0) {
3996 			mp = igb_rx(&igb->rx_rings[0], IGB_NO_POLL);
3997 		}
3998 
3999 		if (icr & E1000_ICR_TXDW) {
4000 			tx_ring = &igb->tx_rings[0];
4001 
4002 			/* Recycle the tx descriptors */
4003 			tx_ring->tx_recycle(tx_ring);
4004 
4005 			/* Schedule the re-transmit */
4006 			tx_reschedule = (tx_ring->reschedule &&
4007 			    (tx_ring->tbd_free >= igb->tx_resched_thresh));
4008 		}
4009 
4010 		if (icr & E1000_ICR_LSC) {
4011 			/*
4012 			 * Because we got a link-status-change interrupt, force
4013 			 * e1000_check_for_link() to look at phy
4014 			 */
4015 			igb->hw.mac.get_link_status = B_TRUE;
4016 
4017 			/* igb_link_check takes care of link status change */
4018 			link_changed = igb_link_check(igb);
4019 
4020 			/* Get new phy state */
4021 			igb_get_phy_state(igb);
4022 		}
4023 
4024 		if (icr & E1000_ICR_DRSTA) {
4025 			/* 82580 Full Device Reset needed */
4026 			atomic_or_32(&igb->igb_state, IGB_STALL);
4027 		}
4028 
4029 		result = DDI_INTR_CLAIMED;
4030 	} else {
4031 		/*
4032 		 * E1000_ICR_INT_ASSERTED bit was not set:
4033 		 * Don't claim this interrupt.
4034 		 */
4035 		result = DDI_INTR_UNCLAIMED;
4036 	}
4037 
4038 	mutex_exit(&igb->gen_lock);
4039 
4040 	/*
4041 	 * Do the following work outside of the gen_lock
4042 	 */
4043 	if (mp != NULL)
4044 		mac_rx(igb->mac_hdl, NULL, mp);
4045 
4046 	if (tx_reschedule)  {
4047 		tx_ring->reschedule = B_FALSE;
4048 		mac_tx_ring_update(igb->mac_hdl, tx_ring->ring_handle);
4049 		IGB_DEBUG_STAT(tx_ring->stat_reschedule);
4050 	}
4051 
4052 	if (link_changed)
4053 		mac_link_update(igb->mac_hdl, igb->link_state);
4054 
4055 	return (result);
4056 }
4057 
4058 /*
4059  * igb_intr_msi - Interrupt handler for MSI
4060  */
4061 static uint_t
4062 igb_intr_msi(void *arg1, void *arg2)
4063 {
4064 	igb_t *igb = (igb_t *)arg1;
4065 	uint32_t icr;
4066 
4067 	_NOTE(ARGUNUSED(arg2));
4068 
4069 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
4070 
4071 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
4072 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
4073 		atomic_or_32(&igb->igb_state, IGB_ERROR);
4074 		return (DDI_INTR_CLAIMED);
4075 	}
4076 
4077 	/* Make sure all interrupt causes cleared */
4078 	(void) E1000_READ_REG(&igb->hw, E1000_EICR);
4079 
4080 	/*
4081 	 * For MSI interrupt, we have only one vector,
4082 	 * so we have only one rx ring and one tx ring enabled.
4083 	 */
4084 	ASSERT(igb->num_rx_rings == 1);
4085 	ASSERT(igb->num_tx_rings == 1);
4086 
4087 	if (icr & E1000_ICR_RXT0) {
4088 		igb_intr_rx_work(&igb->rx_rings[0]);
4089 	}
4090 
4091 	if (icr & E1000_ICR_TXDW) {
4092 		igb_intr_tx_work(&igb->tx_rings[0]);
4093 	}
4094 
4095 	if (icr & E1000_ICR_LSC) {
4096 		igb_intr_link_work(igb);
4097 	}
4098 
4099 	if (icr & E1000_ICR_DRSTA) {
4100 		/* 82580 Full Device Reset needed */
4101 		atomic_or_32(&igb->igb_state, IGB_STALL);
4102 	}
4103 
4104 	return (DDI_INTR_CLAIMED);
4105 }
4106 
4107 /*
4108  * igb_intr_rx - Interrupt handler for rx
4109  */
4110 static uint_t
4111 igb_intr_rx(void *arg1, void *arg2)
4112 {
4113 	igb_rx_ring_t *rx_ring = (igb_rx_ring_t *)arg1;
4114 
4115 	_NOTE(ARGUNUSED(arg2));
4116 
4117 	/*
4118 	 * Only used via MSI-X vector so don't check cause bits
4119 	 * and only clean the given ring.
4120 	 */
4121 	igb_intr_rx_work(rx_ring);
4122 
4123 	return (DDI_INTR_CLAIMED);
4124 }
4125 
4126 /*
4127  * igb_intr_tx - Interrupt handler for tx
4128  */
4129 static uint_t
4130 igb_intr_tx(void *arg1, void *arg2)
4131 {
4132 	igb_tx_ring_t *tx_ring = (igb_tx_ring_t *)arg1;
4133 
4134 	_NOTE(ARGUNUSED(arg2));
4135 
4136 	/*
4137 	 * Only used via MSI-X vector so don't check cause bits
4138 	 * and only clean the given ring.
4139 	 */
4140 	igb_intr_tx_work(tx_ring);
4141 
4142 	return (DDI_INTR_CLAIMED);
4143 }
4144 
4145 /*
4146  * igb_intr_tx_other - Interrupt handler for both tx and other
4147  *
4148  */
4149 static uint_t
4150 igb_intr_tx_other(void *arg1, void *arg2)
4151 {
4152 	igb_t *igb = (igb_t *)arg1;
4153 	uint32_t icr;
4154 
4155 	_NOTE(ARGUNUSED(arg2));
4156 
4157 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
4158 
4159 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
4160 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
4161 		atomic_or_32(&igb->igb_state, IGB_ERROR);
4162 		return (DDI_INTR_CLAIMED);
4163 	}
4164 
4165 	/*
4166 	 * Look for tx reclaiming work first. Remember, in the
4167 	 * case of only interrupt sharing, only one tx ring is
4168 	 * used
4169 	 */
4170 	igb_intr_tx_work(&igb->tx_rings[0]);
4171 
4172 	/*
4173 	 * Check for "other" causes.
4174 	 */
4175 	if (icr & E1000_ICR_LSC) {
4176 		igb_intr_link_work(igb);
4177 	}
4178 
4179 	/*
4180 	 * The DOUTSYNC bit indicates a tx packet dropped because
4181 	 * DMA engine gets "out of sync". There isn't a real fix
4182 	 * for this. The Intel recommendation is to count the number
4183 	 * of occurrences so user can detect when it is happening.
4184 	 * The issue is non-fatal and there's no recovery action
4185 	 * available.
4186 	 */
4187 	if (icr & E1000_ICR_DOUTSYNC) {
4188 		IGB_STAT(igb->dout_sync);
4189 	}
4190 
4191 	if (icr & E1000_ICR_DRSTA) {
4192 		/* 82580 Full Device Reset needed */
4193 		atomic_or_32(&igb->igb_state, IGB_STALL);
4194 	}
4195 
4196 	return (DDI_INTR_CLAIMED);
4197 }
4198 
4199 /*
4200  * igb_alloc_intrs - Allocate interrupts for the driver
4201  *
4202  * Normal sequence is to try MSI-X; if not sucessful, try MSI;
4203  * if not successful, try Legacy.
4204  * igb->intr_force can be used to force sequence to start with
4205  * any of the 3 types.
4206  * If MSI-X is not used, number of tx/rx rings is forced to 1.
4207  */
4208 static int
4209 igb_alloc_intrs(igb_t *igb)
4210 {
4211 	dev_info_t *devinfo;
4212 	int intr_types;
4213 	int rc;
4214 
4215 	devinfo = igb->dip;
4216 
4217 	/* Get supported interrupt types */
4218 	rc = ddi_intr_get_supported_types(devinfo, &intr_types);
4219 
4220 	if (rc != DDI_SUCCESS) {
4221 		igb_log(igb,
4222 		    "Get supported interrupt types failed: %d", rc);
4223 		return (IGB_FAILURE);
4224 	}
4225 	IGB_DEBUGLOG_1(igb, "Supported interrupt types: %x", intr_types);
4226 
4227 	igb->intr_type = 0;
4228 
4229 	/* Install MSI-X interrupts */
4230 	if ((intr_types & DDI_INTR_TYPE_MSIX) &&
4231 	    (igb->intr_force <= IGB_INTR_MSIX)) {
4232 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_MSIX);
4233 
4234 		if (rc == IGB_SUCCESS)
4235 			return (IGB_SUCCESS);
4236 
4237 		igb_log(igb,
4238 		    "Allocate MSI-X failed, trying MSI interrupts...");
4239 	}
4240 
4241 	/* MSI-X not used, force rings to 1 */
4242 	igb->num_rx_rings = 1;
4243 	igb->num_tx_rings = 1;
4244 	igb_log(igb,
4245 	    "MSI-X not used, force rx and tx queue number to 1");
4246 
4247 	/* Install MSI interrupts */
4248 	if ((intr_types & DDI_INTR_TYPE_MSI) &&
4249 	    (igb->intr_force <= IGB_INTR_MSI)) {
4250 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_MSI);
4251 
4252 		if (rc == IGB_SUCCESS)
4253 			return (IGB_SUCCESS);
4254 
4255 		igb_log(igb,
4256 		    "Allocate MSI failed, trying Legacy interrupts...");
4257 	}
4258 
4259 	/* Install legacy interrupts */
4260 	if (intr_types & DDI_INTR_TYPE_FIXED) {
4261 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_FIXED);
4262 
4263 		if (rc == IGB_SUCCESS)
4264 			return (IGB_SUCCESS);
4265 
4266 		igb_log(igb,
4267 		    "Allocate Legacy interrupts failed");
4268 	}
4269 
4270 	/* If none of the 3 types succeeded, return failure */
4271 	return (IGB_FAILURE);
4272 }
4273 
4274 /*
4275  * igb_alloc_intr_handles - Allocate interrupt handles.
4276  *
4277  * For legacy and MSI, only 1 handle is needed.  For MSI-X,
4278  * if fewer than 2 handles are available, return failure.
4279  * Upon success, this sets the number of Rx rings to a number that
4280  * matches the handles available for Rx interrupts.
4281  */
4282 static int
4283 igb_alloc_intr_handles(igb_t *igb, int intr_type)
4284 {
4285 	dev_info_t *devinfo;
4286 	int orig, request, count, avail, actual;
4287 	int diff, minimum;
4288 	int rc;
4289 
4290 	devinfo = igb->dip;
4291 
4292 	switch (intr_type) {
4293 	case DDI_INTR_TYPE_FIXED:
4294 		request = 1;	/* Request 1 legacy interrupt handle */
4295 		minimum = 1;
4296 		IGB_DEBUGLOG_0(igb, "interrupt type: legacy");
4297 		break;
4298 
4299 	case DDI_INTR_TYPE_MSI:
4300 		request = 1;	/* Request 1 MSI interrupt handle */
4301 		minimum = 1;
4302 		IGB_DEBUGLOG_0(igb, "interrupt type: MSI");
4303 		break;
4304 
4305 	case DDI_INTR_TYPE_MSIX:
4306 		/*
4307 		 * Number of vectors for the adapter is
4308 		 * # rx rings + # tx rings
4309 		 * One of tx vectors is for tx & other
4310 		 */
4311 		request = igb->num_rx_rings + igb->num_tx_rings;
4312 		orig = request;
4313 		minimum = 2;
4314 		IGB_DEBUGLOG_0(igb, "interrupt type: MSI-X");
4315 		break;
4316 
4317 	default:
4318 		igb_log(igb,
4319 		    "invalid call to igb_alloc_intr_handles(): %d\n",
4320 		    intr_type);
4321 		return (IGB_FAILURE);
4322 	}
4323 	IGB_DEBUGLOG_2(igb, "interrupt handles requested: %d  minimum: %d",
4324 	    request, minimum);
4325 
4326 	/*
4327 	 * Get number of supported interrupts
4328 	 */
4329 	rc = ddi_intr_get_nintrs(devinfo, intr_type, &count);
4330 	if ((rc != DDI_SUCCESS) || (count < minimum)) {
4331 		igb_log(igb,
4332 		    "Get supported interrupt number failed. "
4333 		    "Return: %d, count: %d", rc, count);
4334 		return (IGB_FAILURE);
4335 	}
4336 	IGB_DEBUGLOG_1(igb, "interrupts supported: %d", count);
4337 
4338 	/*
4339 	 * Get number of available interrupts
4340 	 */
4341 	rc = ddi_intr_get_navail(devinfo, intr_type, &avail);
4342 	if ((rc != DDI_SUCCESS) || (avail < minimum)) {
4343 		igb_log(igb,
4344 		    "Get available interrupt number failed. "
4345 		    "Return: %d, available: %d", rc, avail);
4346 		return (IGB_FAILURE);
4347 	}
4348 	IGB_DEBUGLOG_1(igb, "interrupts available: %d", avail);
4349 
4350 	if (avail < request) {
4351 		igb_log(igb, "Request %d handles, %d available",
4352 		    request, avail);
4353 		request = avail;
4354 	}
4355 
4356 	actual = 0;
4357 	igb->intr_cnt = 0;
4358 
4359 	/*
4360 	 * Allocate an array of interrupt handles
4361 	 */
4362 	igb->intr_size = request * sizeof (ddi_intr_handle_t);
4363 	igb->htable = kmem_alloc(igb->intr_size, KM_SLEEP);
4364 
4365 	rc = ddi_intr_alloc(devinfo, igb->htable, intr_type, 0,
4366 	    request, &actual, DDI_INTR_ALLOC_NORMAL);
4367 	if (rc != DDI_SUCCESS) {
4368 		igb_log(igb, "Allocate interrupts failed. "
4369 		    "return: %d, request: %d, actual: %d",
4370 		    rc, request, actual);
4371 		goto alloc_handle_fail;
4372 	}
4373 	IGB_DEBUGLOG_1(igb, "interrupts actually allocated: %d", actual);
4374 
4375 	igb->intr_cnt = actual;
4376 
4377 	if (actual < minimum) {
4378 		igb_log(igb, "Insufficient interrupt handles allocated: %d",
4379 		    actual);
4380 		goto alloc_handle_fail;
4381 	}
4382 
4383 	/*
4384 	 * For MSI-X, actual might force us to reduce number of tx & rx rings
4385 	 */
4386 	if ((intr_type == DDI_INTR_TYPE_MSIX) && (orig > actual)) {
4387 		diff = orig - actual;
4388 		if (diff < igb->num_tx_rings) {
4389 			igb_log(igb,
4390 			    "MSI-X vectors force Tx queue number to %d",
4391 			    igb->num_tx_rings - diff);
4392 			igb->num_tx_rings -= diff;
4393 		} else {
4394 			igb_log(igb,
4395 			    "MSI-X vectors force Tx queue number to 1");
4396 			igb->num_tx_rings = 1;
4397 
4398 			igb_log(igb,
4399 			    "MSI-X vectors force Rx queue number to %d",
4400 			    actual - 1);
4401 			igb->num_rx_rings = actual - 1;
4402 		}
4403 	}
4404 
4405 	/*
4406 	 * Get priority for first vector, assume remaining are all the same
4407 	 */
4408 	rc = ddi_intr_get_pri(igb->htable[0], &igb->intr_pri);
4409 	if (rc != DDI_SUCCESS) {
4410 		igb_log(igb,
4411 		    "Get interrupt priority failed: %d", rc);
4412 		goto alloc_handle_fail;
4413 	}
4414 
4415 	rc = ddi_intr_get_cap(igb->htable[0], &igb->intr_cap);
4416 	if (rc != DDI_SUCCESS) {
4417 		igb_log(igb,
4418 		    "Get interrupt cap failed: %d", rc);
4419 		goto alloc_handle_fail;
4420 	}
4421 
4422 	igb->intr_type = intr_type;
4423 
4424 	return (IGB_SUCCESS);
4425 
4426 alloc_handle_fail:
4427 	igb_rem_intrs(igb);
4428 
4429 	return (IGB_FAILURE);
4430 }
4431 
4432 /*
4433  * igb_add_intr_handlers - Add interrupt handlers based on the interrupt type
4434  *
4435  * Before adding the interrupt handlers, the interrupt vectors have
4436  * been allocated, and the rx/tx rings have also been allocated.
4437  */
4438 static int
4439 igb_add_intr_handlers(igb_t *igb)
4440 {
4441 	igb_rx_ring_t *rx_ring;
4442 	igb_tx_ring_t *tx_ring;
4443 	int vector;
4444 	int rc;
4445 	int i;
4446 
4447 	vector = 0;
4448 
4449 	switch (igb->intr_type) {
4450 	case DDI_INTR_TYPE_MSIX:
4451 		/* Add interrupt handler for tx + other */
4452 		tx_ring = &igb->tx_rings[0];
4453 		rc = ddi_intr_add_handler(igb->htable[vector],
4454 		    (ddi_intr_handler_t *)igb_intr_tx_other,
4455 		    (void *)igb, NULL);
4456 
4457 		if (rc != DDI_SUCCESS) {
4458 			igb_log(igb,
4459 			    "Add tx/other interrupt handler failed: %d", rc);
4460 			return (IGB_FAILURE);
4461 		}
4462 		tx_ring->intr_vector = vector;
4463 		vector++;
4464 
4465 		/* Add interrupt handler for each rx ring */
4466 		for (i = 0; i < igb->num_rx_rings; i++) {
4467 			rx_ring = &igb->rx_rings[i];
4468 
4469 			rc = ddi_intr_add_handler(igb->htable[vector],
4470 			    (ddi_intr_handler_t *)igb_intr_rx,
4471 			    (void *)rx_ring, NULL);
4472 
4473 			if (rc != DDI_SUCCESS) {
4474 				igb_log(igb,
4475 				    "Add rx interrupt handler failed. "
4476 				    "return: %d, rx ring: %d", rc, i);
4477 				for (vector--; vector >= 0; vector--) {
4478 					(void) ddi_intr_remove_handler(
4479 					    igb->htable[vector]);
4480 				}
4481 				return (IGB_FAILURE);
4482 			}
4483 
4484 			rx_ring->intr_vector = vector;
4485 
4486 			vector++;
4487 		}
4488 
4489 		/* Add interrupt handler for each tx ring from 2nd ring */
4490 		for (i = 1; i < igb->num_tx_rings; i++) {
4491 			tx_ring = &igb->tx_rings[i];
4492 
4493 			rc = ddi_intr_add_handler(igb->htable[vector],
4494 			    (ddi_intr_handler_t *)igb_intr_tx,
4495 			    (void *)tx_ring, NULL);
4496 
4497 			if (rc != DDI_SUCCESS) {
4498 				igb_log(igb,
4499 				    "Add tx interrupt handler failed. "
4500 				    "return: %d, tx ring: %d", rc, i);
4501 				for (vector--; vector >= 0; vector--) {
4502 					(void) ddi_intr_remove_handler(
4503 					    igb->htable[vector]);
4504 				}
4505 				return (IGB_FAILURE);
4506 			}
4507 
4508 			tx_ring->intr_vector = vector;
4509 
4510 			vector++;
4511 		}
4512 
4513 		break;
4514 
4515 	case DDI_INTR_TYPE_MSI:
4516 		/* Add interrupt handlers for the only vector */
4517 		rc = ddi_intr_add_handler(igb->htable[vector],
4518 		    (ddi_intr_handler_t *)igb_intr_msi,
4519 		    (void *)igb, NULL);
4520 
4521 		if (rc != DDI_SUCCESS) {
4522 			igb_log(igb,
4523 			    "Add MSI interrupt handler failed: %d", rc);
4524 			return (IGB_FAILURE);
4525 		}
4526 
4527 		rx_ring = &igb->rx_rings[0];
4528 		rx_ring->intr_vector = vector;
4529 
4530 		vector++;
4531 		break;
4532 
4533 	case DDI_INTR_TYPE_FIXED:
4534 		/* Add interrupt handlers for the only vector */
4535 		rc = ddi_intr_add_handler(igb->htable[vector],
4536 		    (ddi_intr_handler_t *)igb_intr_legacy,
4537 		    (void *)igb, NULL);
4538 
4539 		if (rc != DDI_SUCCESS) {
4540 			igb_log(igb,
4541 			    "Add legacy interrupt handler failed: %d", rc);
4542 			return (IGB_FAILURE);
4543 		}
4544 
4545 		rx_ring = &igb->rx_rings[0];
4546 		rx_ring->intr_vector = vector;
4547 
4548 		vector++;
4549 		break;
4550 
4551 	default:
4552 		return (IGB_FAILURE);
4553 	}
4554 
4555 	ASSERT(vector == igb->intr_cnt);
4556 
4557 	return (IGB_SUCCESS);
4558 }
4559 
4560 /*
4561  * igb_setup_msix_82575 - setup 82575 adapter to use MSI-X interrupts
4562  *
4563  * For each vector enabled on the adapter, Set the MSIXBM register accordingly
4564  */
4565 static void
4566 igb_setup_msix_82575(igb_t *igb)
4567 {
4568 	uint32_t eims = 0;
4569 	int i, vector;
4570 	struct e1000_hw *hw = &igb->hw;
4571 
4572 	/*
4573 	 * Set vector for tx ring 0 and other causes.
4574 	 * NOTE assumption that it is vector 0.
4575 	 */
4576 	vector = 0;
4577 
4578 	igb->eims_mask = E1000_EICR_TX_QUEUE0 | E1000_EICR_OTHER;
4579 	E1000_WRITE_REG(hw, E1000_MSIXBM(vector), igb->eims_mask);
4580 	vector++;
4581 
4582 	for (i = 0; i < igb->num_rx_rings; i++) {
4583 		/*
4584 		 * Set vector for each rx ring
4585 		 */
4586 		eims = (E1000_EICR_RX_QUEUE0 << i);
4587 		E1000_WRITE_REG(hw, E1000_MSIXBM(vector), eims);
4588 
4589 		/*
4590 		 * Accumulate bits to enable in
4591 		 * igb_enable_adapter_interrupts_82575()
4592 		 */
4593 		igb->eims_mask |= eims;
4594 
4595 		vector++;
4596 	}
4597 
4598 	for (i = 1; i < igb->num_tx_rings; i++) {
4599 		/*
4600 		 * Set vector for each tx ring from 2nd tx ring
4601 		 */
4602 		eims = (E1000_EICR_TX_QUEUE0 << i);
4603 		E1000_WRITE_REG(hw, E1000_MSIXBM(vector), eims);
4604 
4605 		/*
4606 		 * Accumulate bits to enable in
4607 		 * igb_enable_adapter_interrupts_82575()
4608 		 */
4609 		igb->eims_mask |= eims;
4610 
4611 		vector++;
4612 	}
4613 
4614 	ASSERT(vector == igb->intr_cnt);
4615 
4616 	/*
4617 	 * Disable IAM for ICR interrupt bits
4618 	 */
4619 	E1000_WRITE_REG(hw, E1000_IAM, 0);
4620 	E1000_WRITE_FLUSH(hw);
4621 }
4622 
4623 /*
4624  * igb_setup_msix_82576 - setup 82576 adapter to use MSI-X interrupts
4625  *
4626  * 82576 uses a table based method for assigning vectors.  Each queue has a
4627  * single entry in the table to which we write a vector number along with a
4628  * "valid" bit.  The entry is a single byte in a 4-byte register.  Vectors
4629  * take a different position in the 4-byte register depending on whether
4630  * they are numbered above or below 8.
4631  */
4632 static void
4633 igb_setup_msix_82576(igb_t *igb)
4634 {
4635 	struct e1000_hw *hw = &igb->hw;
4636 	uint32_t ivar, index, vector;
4637 	int i;
4638 
4639 	/* must enable msi-x capability before IVAR settings */
4640 	E1000_WRITE_REG(hw, E1000_GPIE,
4641 	    (E1000_GPIE_MSIX_MODE | E1000_GPIE_PBA | E1000_GPIE_NSICR));
4642 
4643 	/*
4644 	 * Set vector for tx ring 0 and other causes.
4645 	 * NOTE assumption that it is vector 0.
4646 	 * This is also interdependent with installation of interrupt service
4647 	 * routines in igb_add_intr_handlers().
4648 	 */
4649 
4650 	/* assign "other" causes to vector 0 */
4651 	vector = 0;
4652 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4653 	E1000_WRITE_REG(hw, E1000_IVAR_MISC, ivar);
4654 
4655 	/* assign tx ring 0 to vector 0 */
4656 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4657 	E1000_WRITE_REG(hw, E1000_IVAR0, ivar);
4658 
4659 	/* prepare to enable tx & other interrupt causes */
4660 	igb->eims_mask = (1 << vector);
4661 
4662 	vector ++;
4663 	for (i = 0; i < igb->num_rx_rings; i++) {
4664 		/*
4665 		 * Set vector for each rx ring
4666 		 */
4667 		index = (i & 0x7);
4668 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4669 
4670 		if (i < 8) {
4671 			/* vector goes into low byte of register */
4672 			ivar = ivar & 0xFFFFFF00;
4673 			ivar |= (vector | E1000_IVAR_VALID);
4674 		} else {
4675 			/* vector goes into third byte of register */
4676 			ivar = ivar & 0xFF00FFFF;
4677 			ivar |= ((vector | E1000_IVAR_VALID) << 16);
4678 		}
4679 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4680 
4681 		/* Accumulate interrupt-cause bits to enable */
4682 		igb->eims_mask |= (1 << vector);
4683 
4684 		vector ++;
4685 	}
4686 
4687 	for (i = 1; i < igb->num_tx_rings; i++) {
4688 		/*
4689 		 * Set vector for each tx ring from 2nd tx ring.
4690 		 * Note assumption that tx vectors numericall follow rx vectors.
4691 		 */
4692 		index = (i & 0x7);
4693 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4694 
4695 		if (i < 8) {
4696 			/* vector goes into second byte of register */
4697 			ivar = ivar & 0xFFFF00FF;
4698 			ivar |= ((vector | E1000_IVAR_VALID) << 8);
4699 		} else {
4700 			/* vector goes into fourth byte of register */
4701 			ivar = ivar & 0x00FFFFFF;
4702 			ivar |= (vector | E1000_IVAR_VALID) << 24;
4703 		}
4704 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4705 
4706 		/* Accumulate interrupt-cause bits to enable */
4707 		igb->eims_mask |= (1 << vector);
4708 
4709 		vector ++;
4710 	}
4711 
4712 	ASSERT(vector == igb->intr_cnt);
4713 }
4714 
4715 /*
4716  * igb_setup_msix_82580 - setup 82580 adapter to use MSI-X interrupts
4717  *
4718  * 82580 uses same table approach at 82576 but has fewer entries.  Each
4719  * queue has a single entry in the table to which we write a vector number
4720  * along with a "valid" bit.  Vectors take a different position in the
4721  * register depending on * whether * they are numbered above or below 4.
4722  */
4723 static void
4724 igb_setup_msix_82580(igb_t *igb)
4725 {
4726 	struct e1000_hw *hw = &igb->hw;
4727 	uint32_t ivar, index, vector;
4728 	int i;
4729 
4730 	/* must enable msi-x capability before IVAR settings */
4731 	E1000_WRITE_REG(hw, E1000_GPIE, (E1000_GPIE_MSIX_MODE |
4732 	    E1000_GPIE_PBA | E1000_GPIE_NSICR | E1000_GPIE_EIAME));
4733 	/*
4734 	 * Set vector for tx ring 0 and other causes.
4735 	 * NOTE assumption that it is vector 0.
4736 	 * This is also interdependent with installation of interrupt service
4737 	 * routines in igb_add_intr_handlers().
4738 	 */
4739 
4740 	/* assign "other" causes to vector 0 */
4741 	vector = 0;
4742 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4743 	E1000_WRITE_REG(hw, E1000_IVAR_MISC, ivar);
4744 
4745 	/* assign tx ring 0 to vector 0 */
4746 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4747 	E1000_WRITE_REG(hw, E1000_IVAR0, ivar);
4748 
4749 	/* prepare to enable tx & other interrupt causes */
4750 	igb->eims_mask = (1 << vector);
4751 
4752 	vector ++;
4753 
4754 	for (i = 0; i < igb->num_rx_rings; i++) {
4755 		/*
4756 		 * Set vector for each rx ring
4757 		 */
4758 		index = (i >> 1);
4759 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4760 
4761 		if (i & 1) {
4762 			/* vector goes into third byte of register */
4763 			ivar = ivar & 0xFF00FFFF;
4764 			ivar |= ((vector | E1000_IVAR_VALID) << 16);
4765 		} else {
4766 			/* vector goes into low byte of register */
4767 			ivar = ivar & 0xFFFFFF00;
4768 			ivar |= (vector | E1000_IVAR_VALID);
4769 		}
4770 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4771 
4772 		/* Accumulate interrupt-cause bits to enable */
4773 		igb->eims_mask |= (1 << vector);
4774 
4775 		vector ++;
4776 	}
4777 
4778 	for (i = 1; i < igb->num_tx_rings; i++) {
4779 		/*
4780 		 * Set vector for each tx ring from 2nd tx ring.
4781 		 * Note assumption that tx vectors numericall follow rx vectors.
4782 		 */
4783 		index = (i >> 1);
4784 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4785 
4786 		if (i & 1) {
4787 			/* vector goes into high byte of register */
4788 			ivar = ivar & 0x00FFFFFF;
4789 			ivar |= ((vector | E1000_IVAR_VALID) << 24);
4790 		} else {
4791 			/* vector goes into second byte of register */
4792 			ivar = ivar & 0xFFFF00FF;
4793 			ivar |= (vector | E1000_IVAR_VALID) << 8;
4794 		}
4795 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4796 
4797 		/* Accumulate interrupt-cause bits to enable */
4798 		igb->eims_mask |= (1 << vector);
4799 
4800 		vector ++;
4801 	}
4802 	ASSERT(vector == igb->intr_cnt);
4803 }
4804 
4805 /*
4806  * igb_rem_intr_handlers - remove the interrupt handlers
4807  */
4808 static void
4809 igb_rem_intr_handlers(igb_t *igb)
4810 {
4811 	int i;
4812 	int rc;
4813 
4814 	for (i = 0; i < igb->intr_cnt; i++) {
4815 		rc = ddi_intr_remove_handler(igb->htable[i]);
4816 		if (rc != DDI_SUCCESS) {
4817 			IGB_DEBUGLOG_1(igb,
4818 			    "Remove intr handler failed: %d", rc);
4819 		}
4820 	}
4821 }
4822 
4823 /*
4824  * igb_rem_intrs - remove the allocated interrupts
4825  */
4826 static void
4827 igb_rem_intrs(igb_t *igb)
4828 {
4829 	int i;
4830 	int rc;
4831 
4832 	for (i = 0; i < igb->intr_cnt; i++) {
4833 		rc = ddi_intr_free(igb->htable[i]);
4834 		if (rc != DDI_SUCCESS) {
4835 			IGB_DEBUGLOG_1(igb,
4836 			    "Free intr failed: %d", rc);
4837 		}
4838 	}
4839 
4840 	kmem_free(igb->htable, igb->intr_size);
4841 	igb->htable = NULL;
4842 }
4843 
4844 /*
4845  * igb_enable_intrs - enable all the ddi interrupts
4846  */
4847 static int
4848 igb_enable_intrs(igb_t *igb)
4849 {
4850 	int i;
4851 	int rc;
4852 
4853 	/* Enable interrupts */
4854 	if (igb->intr_cap & DDI_INTR_FLAG_BLOCK) {
4855 		/* Call ddi_intr_block_enable() for MSI */
4856 		rc = ddi_intr_block_enable(igb->htable, igb->intr_cnt);
4857 		if (rc != DDI_SUCCESS) {
4858 			igb_log(igb,
4859 			    "Enable block intr failed: %d", rc);
4860 			return (IGB_FAILURE);
4861 		}
4862 	} else {
4863 		/* Call ddi_intr_enable() for Legacy/MSI non block enable */
4864 		for (i = 0; i < igb->intr_cnt; i++) {
4865 			rc = ddi_intr_enable(igb->htable[i]);
4866 			if (rc != DDI_SUCCESS) {
4867 				igb_log(igb,
4868 				    "Enable intr failed: %d", rc);
4869 				return (IGB_FAILURE);
4870 			}
4871 		}
4872 	}
4873 
4874 	return (IGB_SUCCESS);
4875 }
4876 
4877 /*
4878  * igb_disable_intrs - disable all the ddi interrupts
4879  */
4880 static int
4881 igb_disable_intrs(igb_t *igb)
4882 {
4883 	int i;
4884 	int rc;
4885 
4886 	/* Disable all interrupts */
4887 	if (igb->intr_cap & DDI_INTR_FLAG_BLOCK) {
4888 		rc = ddi_intr_block_disable(igb->htable, igb->intr_cnt);
4889 		if (rc != DDI_SUCCESS) {
4890 			igb_log(igb,
4891 			    "Disable block intr failed: %d", rc);
4892 			return (IGB_FAILURE);
4893 		}
4894 	} else {
4895 		for (i = 0; i < igb->intr_cnt; i++) {
4896 			rc = ddi_intr_disable(igb->htable[i]);
4897 			if (rc != DDI_SUCCESS) {
4898 				igb_log(igb,
4899 				    "Disable intr failed: %d", rc);
4900 				return (IGB_FAILURE);
4901 			}
4902 		}
4903 	}
4904 
4905 	return (IGB_SUCCESS);
4906 }
4907 
4908 /*
4909  * igb_get_phy_state - Get and save the parameters read from PHY registers
4910  */
4911 static void
4912 igb_get_phy_state(igb_t *igb)
4913 {
4914 	struct e1000_hw *hw = &igb->hw;
4915 	uint16_t phy_ctrl;
4916 	uint16_t phy_status;
4917 	uint16_t phy_an_adv;
4918 	uint16_t phy_an_exp;
4919 	uint16_t phy_ext_status;
4920 	uint16_t phy_1000t_ctrl;
4921 	uint16_t phy_1000t_status;
4922 	uint16_t phy_lp_able;
4923 
4924 	ASSERT(mutex_owned(&igb->gen_lock));
4925 
4926 	if (hw->phy.media_type == e1000_media_type_copper) {
4927 		(void) e1000_read_phy_reg(hw, PHY_CONTROL, &phy_ctrl);
4928 		(void) e1000_read_phy_reg(hw, PHY_STATUS, &phy_status);
4929 		(void) e1000_read_phy_reg(hw, PHY_AUTONEG_ADV, &phy_an_adv);
4930 		(void) e1000_read_phy_reg(hw, PHY_AUTONEG_EXP, &phy_an_exp);
4931 		(void) e1000_read_phy_reg(hw, PHY_EXT_STATUS, &phy_ext_status);
4932 		(void) e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_1000t_ctrl);
4933 		(void) e1000_read_phy_reg(hw,
4934 		    PHY_1000T_STATUS, &phy_1000t_status);
4935 		(void) e1000_read_phy_reg(hw, PHY_LP_ABILITY, &phy_lp_able);
4936 
4937 		igb->param_autoneg_cap =
4938 		    (phy_status & MII_SR_AUTONEG_CAPS) ? 1 : 0;
4939 		igb->param_pause_cap =
4940 		    (phy_an_adv & NWAY_AR_PAUSE) ? 1 : 0;
4941 		igb->param_asym_pause_cap =
4942 		    (phy_an_adv & NWAY_AR_ASM_DIR) ? 1 : 0;
4943 		igb->param_1000fdx_cap =
4944 		    ((phy_ext_status & IEEE_ESR_1000T_FD_CAPS) ||
4945 		    (phy_ext_status & IEEE_ESR_1000X_FD_CAPS)) ? 1 : 0;
4946 		igb->param_1000hdx_cap =
4947 		    ((phy_ext_status & IEEE_ESR_1000T_HD_CAPS) ||
4948 		    (phy_ext_status & IEEE_ESR_1000X_HD_CAPS)) ? 1 : 0;
4949 		igb->param_100t4_cap =
4950 		    (phy_status & MII_SR_100T4_CAPS) ? 1 : 0;
4951 		igb->param_100fdx_cap = ((phy_status & MII_SR_100X_FD_CAPS) ||
4952 		    (phy_status & MII_SR_100T2_FD_CAPS)) ? 1 : 0;
4953 		igb->param_100hdx_cap = ((phy_status & MII_SR_100X_HD_CAPS) ||
4954 		    (phy_status & MII_SR_100T2_HD_CAPS)) ? 1 : 0;
4955 		igb->param_10fdx_cap =
4956 		    (phy_status & MII_SR_10T_FD_CAPS) ? 1 : 0;
4957 		igb->param_10hdx_cap =
4958 		    (phy_status & MII_SR_10T_HD_CAPS) ? 1 : 0;
4959 		igb->param_rem_fault =
4960 		    (phy_status & MII_SR_REMOTE_FAULT) ? 1 : 0;
4961 
4962 		igb->param_adv_autoneg_cap = hw->mac.autoneg;
4963 		igb->param_adv_pause_cap =
4964 		    (phy_an_adv & NWAY_AR_PAUSE) ? 1 : 0;
4965 		igb->param_adv_asym_pause_cap =
4966 		    (phy_an_adv & NWAY_AR_ASM_DIR) ? 1 : 0;
4967 		igb->param_adv_1000hdx_cap =
4968 		    (phy_1000t_ctrl & CR_1000T_HD_CAPS) ? 1 : 0;
4969 		igb->param_adv_100t4_cap =
4970 		    (phy_an_adv & NWAY_AR_100T4_CAPS) ? 1 : 0;
4971 		igb->param_adv_rem_fault =
4972 		    (phy_an_adv & NWAY_AR_REMOTE_FAULT) ? 1 : 0;
4973 		if (igb->param_adv_autoneg_cap == 1) {
4974 			igb->param_adv_1000fdx_cap =
4975 			    (phy_1000t_ctrl & CR_1000T_FD_CAPS) ? 1 : 0;
4976 			igb->param_adv_100fdx_cap =
4977 			    (phy_an_adv & NWAY_AR_100TX_FD_CAPS) ? 1 : 0;
4978 			igb->param_adv_100hdx_cap =
4979 			    (phy_an_adv & NWAY_AR_100TX_HD_CAPS) ? 1 : 0;
4980 			igb->param_adv_10fdx_cap =
4981 			    (phy_an_adv & NWAY_AR_10T_FD_CAPS) ? 1 : 0;
4982 			igb->param_adv_10hdx_cap =
4983 			    (phy_an_adv & NWAY_AR_10T_HD_CAPS) ? 1 : 0;
4984 		}
4985 
4986 		igb->param_lp_autoneg_cap =
4987 		    (phy_an_exp & NWAY_ER_LP_NWAY_CAPS) ? 1 : 0;
4988 		igb->param_lp_pause_cap =
4989 		    (phy_lp_able & NWAY_LPAR_PAUSE) ? 1 : 0;
4990 		igb->param_lp_asym_pause_cap =
4991 		    (phy_lp_able & NWAY_LPAR_ASM_DIR) ? 1 : 0;
4992 		igb->param_lp_1000fdx_cap =
4993 		    (phy_1000t_status & SR_1000T_LP_FD_CAPS) ? 1 : 0;
4994 		igb->param_lp_1000hdx_cap =
4995 		    (phy_1000t_status & SR_1000T_LP_HD_CAPS) ? 1 : 0;
4996 		igb->param_lp_100t4_cap =
4997 		    (phy_lp_able & NWAY_LPAR_100T4_CAPS) ? 1 : 0;
4998 		igb->param_lp_100fdx_cap =
4999 		    (phy_lp_able & NWAY_LPAR_100TX_FD_CAPS) ? 1 : 0;
5000 		igb->param_lp_100hdx_cap =
5001 		    (phy_lp_able & NWAY_LPAR_100TX_HD_CAPS) ? 1 : 0;
5002 		igb->param_lp_10fdx_cap =
5003 		    (phy_lp_able & NWAY_LPAR_10T_FD_CAPS) ? 1 : 0;
5004 		igb->param_lp_10hdx_cap =
5005 		    (phy_lp_able & NWAY_LPAR_10T_HD_CAPS) ? 1 : 0;
5006 		igb->param_lp_rem_fault =
5007 		    (phy_lp_able & NWAY_LPAR_REMOTE_FAULT) ? 1 : 0;
5008 	} else {
5009 		/*
5010 		 * 1Gig Fiber adapter only offers 1Gig Full Duplex.
5011 		 */
5012 		igb->param_autoneg_cap = 0;
5013 		igb->param_pause_cap = 1;
5014 		igb->param_asym_pause_cap = 1;
5015 		igb->param_1000fdx_cap = 1;
5016 		igb->param_1000hdx_cap = 0;
5017 		igb->param_100t4_cap = 0;
5018 		igb->param_100fdx_cap = 0;
5019 		igb->param_100hdx_cap = 0;
5020 		igb->param_10fdx_cap = 0;
5021 		igb->param_10hdx_cap = 0;
5022 
5023 		igb->param_adv_autoneg_cap = 0;
5024 		igb->param_adv_pause_cap = 1;
5025 		igb->param_adv_asym_pause_cap = 1;
5026 		igb->param_adv_1000fdx_cap = 1;
5027 		igb->param_adv_1000hdx_cap = 0;
5028 		igb->param_adv_100t4_cap = 0;
5029 		igb->param_adv_100fdx_cap = 0;
5030 		igb->param_adv_100hdx_cap = 0;
5031 		igb->param_adv_10fdx_cap = 0;
5032 		igb->param_adv_10hdx_cap = 0;
5033 
5034 		igb->param_lp_autoneg_cap = 0;
5035 		igb->param_lp_pause_cap = 0;
5036 		igb->param_lp_asym_pause_cap = 0;
5037 		igb->param_lp_1000fdx_cap = 0;
5038 		igb->param_lp_1000hdx_cap = 0;
5039 		igb->param_lp_100t4_cap = 0;
5040 		igb->param_lp_100fdx_cap = 0;
5041 		igb->param_lp_100hdx_cap = 0;
5042 		igb->param_lp_10fdx_cap = 0;
5043 		igb->param_lp_10hdx_cap = 0;
5044 		igb->param_lp_rem_fault = 0;
5045 	}
5046 }
5047 
5048 /*
5049  * synchronize the adv* and en* parameters.
5050  *
5051  * See comments in <sys/dld.h> for details of the *_en_*
5052  * parameters. The usage of ndd for setting adv parameters will
5053  * synchronize all the en parameters with the e1000g parameters,
5054  * implicitly disabling any settings made via dladm.
5055  */
5056 static void
5057 igb_param_sync(igb_t *igb)
5058 {
5059 	igb->param_en_1000fdx_cap = igb->param_adv_1000fdx_cap;
5060 	igb->param_en_1000hdx_cap = igb->param_adv_1000hdx_cap;
5061 	igb->param_en_100t4_cap = igb->param_adv_100t4_cap;
5062 	igb->param_en_100fdx_cap = igb->param_adv_100fdx_cap;
5063 	igb->param_en_100hdx_cap = igb->param_adv_100hdx_cap;
5064 	igb->param_en_10fdx_cap = igb->param_adv_10fdx_cap;
5065 	igb->param_en_10hdx_cap = igb->param_adv_10hdx_cap;
5066 }
5067 
5068 /*
5069  * igb_get_driver_control
5070  */
5071 static void
5072 igb_get_driver_control(struct e1000_hw *hw)
5073 {
5074 	uint32_t ctrl_ext;
5075 
5076 	/* Notify firmware that driver is in control of device */
5077 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
5078 	ctrl_ext |= E1000_CTRL_EXT_DRV_LOAD;
5079 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
5080 }
5081 
5082 /*
5083  * igb_release_driver_control
5084  */
5085 static void
5086 igb_release_driver_control(struct e1000_hw *hw)
5087 {
5088 	uint32_t ctrl_ext;
5089 
5090 	/* Notify firmware that driver is no longer in control of device */
5091 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
5092 	ctrl_ext &= ~E1000_CTRL_EXT_DRV_LOAD;
5093 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
5094 }
5095 
5096 /*
5097  * igb_atomic_reserve - Atomic decrease operation
5098  */
5099 int
5100 igb_atomic_reserve(uint32_t *count_p, uint32_t n)
5101 {
5102 	uint32_t oldval;
5103 	uint32_t newval;
5104 
5105 	/* ATOMICALLY */
5106 	do {
5107 		oldval = *count_p;
5108 		if (oldval < n)
5109 			return (-1);
5110 		newval = oldval - n;
5111 	} while (atomic_cas_32(count_p, oldval, newval) != oldval);
5112 
5113 	return (newval);
5114 }
5115 
5116 /*
5117  * FMA support
5118  */
5119 
5120 int
5121 igb_check_acc_handle(ddi_acc_handle_t handle)
5122 {
5123 	ddi_fm_error_t de;
5124 
5125 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
5126 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
5127 	return (de.fme_status);
5128 }
5129 
5130 int
5131 igb_check_dma_handle(ddi_dma_handle_t handle)
5132 {
5133 	ddi_fm_error_t de;
5134 
5135 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
5136 	return (de.fme_status);
5137 }
5138 
5139 /*
5140  * The IO fault service error handling callback function
5141  */
5142 /*ARGSUSED*/
5143 static int
5144 igb_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
5145 {
5146 	/*
5147 	 * as the driver can always deal with an error in any dma or
5148 	 * access handle, we can just return the fme_status value.
5149 	 */
5150 	pci_ereport_post(dip, err, NULL);
5151 	return (err->fme_status);
5152 }
5153 
5154 static void
5155 igb_fm_init(igb_t *igb)
5156 {
5157 	ddi_iblock_cookie_t iblk;
5158 	int fma_dma_flag;
5159 
5160 	/* Only register with IO Fault Services if we have some capability */
5161 	if (igb->fm_capabilities & DDI_FM_ACCCHK_CAPABLE) {
5162 		igb_regs_acc_attr.devacc_attr_access = DDI_FLAGERR_ACC;
5163 	} else {
5164 		igb_regs_acc_attr.devacc_attr_access = DDI_DEFAULT_ACC;
5165 	}
5166 
5167 	if (igb->fm_capabilities & DDI_FM_DMACHK_CAPABLE) {
5168 		fma_dma_flag = 1;
5169 	} else {
5170 		fma_dma_flag = 0;
5171 	}
5172 
5173 	(void) igb_set_fma_flags(fma_dma_flag);
5174 
5175 	if (igb->fm_capabilities) {
5176 
5177 		/* Register capabilities with IO Fault Services */
5178 		ddi_fm_init(igb->dip, &igb->fm_capabilities, &iblk);
5179 
5180 		/*
5181 		 * Initialize pci ereport capabilities if ereport capable
5182 		 */
5183 		if (DDI_FM_EREPORT_CAP(igb->fm_capabilities) ||
5184 		    DDI_FM_ERRCB_CAP(igb->fm_capabilities))
5185 			pci_ereport_setup(igb->dip);
5186 
5187 		/*
5188 		 * Register error callback if error callback capable
5189 		 */
5190 		if (DDI_FM_ERRCB_CAP(igb->fm_capabilities))
5191 			ddi_fm_handler_register(igb->dip,
5192 			    igb_fm_error_cb, (void*) igb);
5193 	}
5194 }
5195 
5196 static void
5197 igb_fm_fini(igb_t *igb)
5198 {
5199 	/* Only unregister FMA capabilities if we registered some */
5200 	if (igb->fm_capabilities) {
5201 
5202 		/*
5203 		 * Release any resources allocated by pci_ereport_setup()
5204 		 */
5205 		if (DDI_FM_EREPORT_CAP(igb->fm_capabilities) ||
5206 		    DDI_FM_ERRCB_CAP(igb->fm_capabilities))
5207 			pci_ereport_teardown(igb->dip);
5208 
5209 		/*
5210 		 * Un-register error callback if error callback capable
5211 		 */
5212 		if (DDI_FM_ERRCB_CAP(igb->fm_capabilities))
5213 			ddi_fm_handler_unregister(igb->dip);
5214 
5215 		/* Unregister from IO Fault Services */
5216 		ddi_fm_fini(igb->dip);
5217 	}
5218 }
5219 
5220 void
5221 igb_fm_ereport(igb_t *igb, char *detail)
5222 {
5223 	uint64_t ena;
5224 	char buf[FM_MAX_CLASS];
5225 
5226 	(void) snprintf(buf, FM_MAX_CLASS, "%s.%s", DDI_FM_DEVICE, detail);
5227 	ena = fm_ena_generate(0, FM_ENA_FMT1);
5228 	if (DDI_FM_EREPORT_CAP(igb->fm_capabilities)) {
5229 		ddi_fm_ereport_post(igb->dip, buf, ena, DDI_NOSLEEP,
5230 		    FM_VERSION, DATA_TYPE_UINT8, FM_EREPORT_VERS0, NULL);
5231 	}
5232 }
5233