xref: /illumos-gate/usr/src/uts/common/io/igb/igb_main.c (revision df3cd224ef765c29101e4110546062199562f757)
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 1.1.13";
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_phy_loopback(igb_t *);
89 static void igb_set_internal_serdes_loopback(igb_t *);
90 static boolean_t igb_find_mac_address(igb_t *);
91 static int igb_alloc_intrs(igb_t *);
92 static int igb_alloc_intr_handles(igb_t *, int);
93 static int igb_add_intr_handlers(igb_t *);
94 static void igb_rem_intr_handlers(igb_t *);
95 static void igb_rem_intrs(igb_t *);
96 static int igb_enable_intrs(igb_t *);
97 static int igb_disable_intrs(igb_t *);
98 static void igb_setup_msix_82575(igb_t *);
99 static void igb_setup_msix_82576(igb_t *);
100 static void igb_setup_msix_82580(igb_t *);
101 static uint_t igb_intr_legacy(void *, void *);
102 static uint_t igb_intr_msi(void *, void *);
103 static uint_t igb_intr_rx(void *, void *);
104 static uint_t igb_intr_tx(void *, void *);
105 static uint_t igb_intr_tx_other(void *, void *);
106 static void igb_intr_rx_work(igb_rx_ring_t *);
107 static void igb_intr_tx_work(igb_tx_ring_t *);
108 static void igb_intr_link_work(igb_t *);
109 static void igb_get_driver_control(struct e1000_hw *);
110 static void igb_release_driver_control(struct e1000_hw *);
111 
112 static int igb_attach(dev_info_t *, ddi_attach_cmd_t);
113 static int igb_detach(dev_info_t *, ddi_detach_cmd_t);
114 static int igb_resume(dev_info_t *);
115 static int igb_suspend(dev_info_t *);
116 static int igb_quiesce(dev_info_t *);
117 static void igb_unconfigure(dev_info_t *, igb_t *);
118 static int igb_fm_error_cb(dev_info_t *, ddi_fm_error_t *,
119     const void *);
120 static void igb_fm_init(igb_t *);
121 static void igb_fm_fini(igb_t *);
122 static void igb_release_multicast(igb_t *);
123 
124 mac_priv_prop_t igb_priv_props[] = {
125 	{"_tx_copy_thresh", MAC_PROP_PERM_RW},
126 	{"_tx_recycle_thresh", MAC_PROP_PERM_RW},
127 	{"_tx_overload_thresh", MAC_PROP_PERM_RW},
128 	{"_tx_resched_thresh", MAC_PROP_PERM_RW},
129 	{"_rx_copy_thresh", MAC_PROP_PERM_RW},
130 	{"_rx_limit_per_intr", MAC_PROP_PERM_RW},
131 	{"_intr_throttling", MAC_PROP_PERM_RW},
132 	{"_adv_pause_cap", MAC_PROP_PERM_READ},
133 	{"_adv_asym_pause_cap", MAC_PROP_PERM_READ}
134 };
135 
136 #define	IGB_MAX_PRIV_PROPS \
137 	(sizeof (igb_priv_props) / sizeof (mac_priv_prop_t))
138 
139 static struct cb_ops igb_cb_ops = {
140 	nulldev,		/* cb_open */
141 	nulldev,		/* cb_close */
142 	nodev,			/* cb_strategy */
143 	nodev,			/* cb_print */
144 	nodev,			/* cb_dump */
145 	nodev,			/* cb_read */
146 	nodev,			/* cb_write */
147 	nodev,			/* cb_ioctl */
148 	nodev,			/* cb_devmap */
149 	nodev,			/* cb_mmap */
150 	nodev,			/* cb_segmap */
151 	nochpoll,		/* cb_chpoll */
152 	ddi_prop_op,		/* cb_prop_op */
153 	NULL,			/* cb_stream */
154 	D_MP | D_HOTPLUG,	/* cb_flag */
155 	CB_REV,			/* cb_rev */
156 	nodev,			/* cb_aread */
157 	nodev			/* cb_awrite */
158 };
159 
160 static struct dev_ops igb_dev_ops = {
161 	DEVO_REV,		/* devo_rev */
162 	0,			/* devo_refcnt */
163 	NULL,			/* devo_getinfo */
164 	nulldev,		/* devo_identify */
165 	nulldev,		/* devo_probe */
166 	igb_attach,		/* devo_attach */
167 	igb_detach,		/* devo_detach */
168 	nodev,			/* devo_reset */
169 	&igb_cb_ops,		/* devo_cb_ops */
170 	NULL,			/* devo_bus_ops */
171 	ddi_power,		/* devo_power */
172 	igb_quiesce,	/* devo_quiesce */
173 };
174 
175 static struct modldrv igb_modldrv = {
176 	&mod_driverops,		/* Type of module.  This one is a driver */
177 	ident,			/* Discription string */
178 	&igb_dev_ops,		/* driver ops */
179 };
180 
181 static struct modlinkage igb_modlinkage = {
182 	MODREV_1, &igb_modldrv, NULL
183 };
184 
185 /* Access attributes for register mapping */
186 ddi_device_acc_attr_t igb_regs_acc_attr = {
187 	DDI_DEVICE_ATTR_V1,
188 	DDI_STRUCTURE_LE_ACC,
189 	DDI_STRICTORDER_ACC,
190 	DDI_FLAGERR_ACC
191 };
192 
193 #define	IGB_M_CALLBACK_FLAGS \
194 	(MC_IOCTL | MC_GETCAPAB | MC_SETPROP | MC_GETPROP)
195 
196 static mac_callbacks_t igb_m_callbacks = {
197 	IGB_M_CALLBACK_FLAGS,
198 	igb_m_stat,
199 	igb_m_start,
200 	igb_m_stop,
201 	igb_m_promisc,
202 	igb_m_multicst,
203 	NULL,
204 	NULL,
205 	igb_m_ioctl,
206 	igb_m_getcapab,
207 	NULL,
208 	NULL,
209 	igb_m_setprop,
210 	igb_m_getprop
211 };
212 
213 /*
214  * Initialize capabilities of each supported adapter type
215  */
216 static adapter_info_t igb_82575_cap = {
217 	/* limits */
218 	4,		/* maximum number of rx queues */
219 	1,		/* minimum number of rx queues */
220 	4,		/* default number of rx queues */
221 	4,		/* maximum number of tx queues */
222 	1,		/* minimum number of tx queues */
223 	4,		/* default number of tx queues */
224 	65535,		/* maximum interrupt throttle rate */
225 	0,		/* minimum interrupt throttle rate */
226 	200,		/* default interrupt throttle rate */
227 
228 	/* function pointers */
229 	igb_enable_adapter_interrupts_82575,
230 	igb_setup_msix_82575,
231 
232 	/* capabilities */
233 	(IGB_FLAG_HAS_DCA |	/* capability flags */
234 	IGB_FLAG_VMDQ_POOL),
235 
236 	0xffc00000		/* mask for RXDCTL register */
237 };
238 
239 static adapter_info_t igb_82576_cap = {
240 	/* limits */
241 	16,		/* maximum number of rx queues */
242 	1,		/* minimum number of rx queues */
243 	4,		/* default number of rx queues */
244 	16,		/* maximum number of tx queues */
245 	1,		/* minimum number of tx queues */
246 	4,		/* default number of tx queues */
247 	65535,		/* maximum interrupt throttle rate */
248 	0,		/* minimum interrupt throttle rate */
249 	200,		/* default interrupt throttle rate */
250 
251 	/* function pointers */
252 	igb_enable_adapter_interrupts_82576,
253 	igb_setup_msix_82576,
254 
255 	/* capabilities */
256 	(IGB_FLAG_HAS_DCA |	/* capability flags */
257 	IGB_FLAG_VMDQ_POOL |
258 	IGB_FLAG_NEED_CTX_IDX),
259 
260 	0xffe00000		/* mask for RXDCTL register */
261 };
262 
263 static adapter_info_t igb_82580_cap = {
264 	/* limits */
265 	8,		/* maximum number of rx queues */
266 	1,		/* minimum number of rx queues */
267 	4,		/* default number of rx queues */
268 	8,		/* maximum number of tx queues */
269 	1,		/* minimum number of tx queues */
270 	4,		/* default number of tx queues */
271 	65535,		/* maximum interrupt throttle rate */
272 	0,		/* minimum interrupt throttle rate */
273 	200,		/* default interrupt throttle rate */
274 
275 	/* function pointers */
276 	igb_enable_adapter_interrupts_82580,
277 	igb_setup_msix_82580,
278 
279 	/* capabilities */
280 	(IGB_FLAG_HAS_DCA |	/* capability flags */
281 	IGB_FLAG_VMDQ_POOL |
282 	IGB_FLAG_NEED_CTX_IDX),
283 
284 	0xffe00000		/* mask for RXDCTL register */
285 };
286 
287 /*
288  * Module Initialization Functions
289  */
290 
291 int
292 _init(void)
293 {
294 	int status;
295 
296 	mac_init_ops(&igb_dev_ops, MODULE_NAME);
297 
298 	status = mod_install(&igb_modlinkage);
299 
300 	if (status != DDI_SUCCESS) {
301 		mac_fini_ops(&igb_dev_ops);
302 	}
303 
304 	return (status);
305 }
306 
307 int
308 _fini(void)
309 {
310 	int status;
311 
312 	status = mod_remove(&igb_modlinkage);
313 
314 	if (status == DDI_SUCCESS) {
315 		mac_fini_ops(&igb_dev_ops);
316 	}
317 
318 	return (status);
319 
320 }
321 
322 int
323 _info(struct modinfo *modinfop)
324 {
325 	int status;
326 
327 	status = mod_info(&igb_modlinkage, modinfop);
328 
329 	return (status);
330 }
331 
332 /*
333  * igb_attach - driver attach
334  *
335  * This function is the device specific initialization entry
336  * point. This entry point is required and must be written.
337  * The DDI_ATTACH command must be provided in the attach entry
338  * point. When attach() is called with cmd set to DDI_ATTACH,
339  * all normal kernel services (such as kmem_alloc(9F)) are
340  * available for use by the driver.
341  *
342  * The attach() function will be called once for each instance
343  * of  the  device  on  the  system with cmd set to DDI_ATTACH.
344  * Until attach() succeeds, the only driver entry points which
345  * may be called are open(9E) and getinfo(9E).
346  */
347 static int
348 igb_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
349 {
350 	igb_t *igb;
351 	struct igb_osdep *osdep;
352 	struct e1000_hw *hw;
353 	int instance;
354 
355 	/*
356 	 * Check the command and perform corresponding operations
357 	 */
358 	switch (cmd) {
359 	default:
360 		return (DDI_FAILURE);
361 
362 	case DDI_RESUME:
363 		return (igb_resume(devinfo));
364 
365 	case DDI_ATTACH:
366 		break;
367 	}
368 
369 	/* Get the device instance */
370 	instance = ddi_get_instance(devinfo);
371 
372 	/* Allocate memory for the instance data structure */
373 	igb = kmem_zalloc(sizeof (igb_t), KM_SLEEP);
374 
375 	igb->dip = devinfo;
376 	igb->instance = instance;
377 
378 	hw = &igb->hw;
379 	osdep = &igb->osdep;
380 	hw->back = osdep;
381 	osdep->igb = igb;
382 
383 	/* Attach the instance pointer to the dev_info data structure */
384 	ddi_set_driver_private(devinfo, igb);
385 
386 
387 	/* Initialize for fma support */
388 	igb->fm_capabilities = igb_get_prop(igb, "fm-capable",
389 	    0, 0x0f,
390 	    DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
391 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
392 	igb_fm_init(igb);
393 	igb->attach_progress |= ATTACH_PROGRESS_FMINIT;
394 
395 	/*
396 	 * Map PCI config space registers
397 	 */
398 	if (pci_config_setup(devinfo, &osdep->cfg_handle) != DDI_SUCCESS) {
399 		igb_error(igb, "Failed to map PCI configurations");
400 		goto attach_fail;
401 	}
402 	igb->attach_progress |= ATTACH_PROGRESS_PCI_CONFIG;
403 
404 	/*
405 	 * Identify the chipset family
406 	 */
407 	if (igb_identify_hardware(igb) != IGB_SUCCESS) {
408 		igb_error(igb, "Failed to identify hardware");
409 		goto attach_fail;
410 	}
411 
412 	/*
413 	 * Map device registers
414 	 */
415 	if (igb_regs_map(igb) != IGB_SUCCESS) {
416 		igb_error(igb, "Failed to map device registers");
417 		goto attach_fail;
418 	}
419 	igb->attach_progress |= ATTACH_PROGRESS_REGS_MAP;
420 
421 	/*
422 	 * Initialize driver parameters
423 	 */
424 	igb_init_properties(igb);
425 	igb->attach_progress |= ATTACH_PROGRESS_PROPS;
426 
427 	/*
428 	 * Allocate interrupts
429 	 */
430 	if (igb_alloc_intrs(igb) != IGB_SUCCESS) {
431 		igb_error(igb, "Failed to allocate interrupts");
432 		goto attach_fail;
433 	}
434 	igb->attach_progress |= ATTACH_PROGRESS_ALLOC_INTR;
435 
436 	/*
437 	 * Allocate rx/tx rings based on the ring numbers.
438 	 * The actual numbers of rx/tx rings are decided by the number of
439 	 * allocated interrupt vectors, so we should allocate the rings after
440 	 * interrupts are allocated.
441 	 */
442 	if (igb_alloc_rings(igb) != IGB_SUCCESS) {
443 		igb_error(igb, "Failed to allocate rx/tx rings or groups");
444 		goto attach_fail;
445 	}
446 	igb->attach_progress |= ATTACH_PROGRESS_ALLOC_RINGS;
447 
448 	/*
449 	 * Add interrupt handlers
450 	 */
451 	if (igb_add_intr_handlers(igb) != IGB_SUCCESS) {
452 		igb_error(igb, "Failed to add interrupt handlers");
453 		goto attach_fail;
454 	}
455 	igb->attach_progress |= ATTACH_PROGRESS_ADD_INTR;
456 
457 	/*
458 	 * Initialize driver parameters
459 	 */
460 	if (igb_init_driver_settings(igb) != IGB_SUCCESS) {
461 		igb_error(igb, "Failed to initialize driver settings");
462 		goto attach_fail;
463 	}
464 
465 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK) {
466 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
467 		goto attach_fail;
468 	}
469 
470 	/*
471 	 * Initialize mutexes for this device.
472 	 * Do this before enabling the interrupt handler and
473 	 * register the softint to avoid the condition where
474 	 * interrupt handler can try using uninitialized mutex
475 	 */
476 	igb_init_locks(igb);
477 	igb->attach_progress |= ATTACH_PROGRESS_LOCKS;
478 
479 	/*
480 	 * Initialize the adapter
481 	 */
482 	if (igb_init(igb) != IGB_SUCCESS) {
483 		igb_error(igb, "Failed to initialize adapter");
484 		goto attach_fail;
485 	}
486 	igb->attach_progress |= ATTACH_PROGRESS_INIT_ADAPTER;
487 
488 	/*
489 	 * Initialize statistics
490 	 */
491 	if (igb_init_stats(igb) != IGB_SUCCESS) {
492 		igb_error(igb, "Failed to initialize statistics");
493 		goto attach_fail;
494 	}
495 	igb->attach_progress |= ATTACH_PROGRESS_STATS;
496 
497 	/*
498 	 * Register the driver to the MAC
499 	 */
500 	if (igb_register_mac(igb) != IGB_SUCCESS) {
501 		igb_error(igb, "Failed to register MAC");
502 		goto attach_fail;
503 	}
504 	igb->attach_progress |= ATTACH_PROGRESS_MAC;
505 
506 	/*
507 	 * Now that mutex locks are initialized, and the chip is also
508 	 * initialized, enable interrupts.
509 	 */
510 	if (igb_enable_intrs(igb) != IGB_SUCCESS) {
511 		igb_error(igb, "Failed to enable DDI interrupts");
512 		goto attach_fail;
513 	}
514 	igb->attach_progress |= ATTACH_PROGRESS_ENABLE_INTR;
515 
516 	igb_log(igb, "%s", igb_version);
517 	atomic_or_32(&igb->igb_state, IGB_INITIALIZED);
518 
519 	return (DDI_SUCCESS);
520 
521 attach_fail:
522 	igb_unconfigure(devinfo, igb);
523 	return (DDI_FAILURE);
524 }
525 
526 /*
527  * igb_detach - driver detach
528  *
529  * The detach() function is the complement of the attach routine.
530  * If cmd is set to DDI_DETACH, detach() is used to remove  the
531  * state  associated  with  a  given  instance of a device node
532  * prior to the removal of that instance from the system.
533  *
534  * The detach() function will be called once for each  instance
535  * of the device for which there has been a successful attach()
536  * once there are no longer  any  opens  on  the  device.
537  *
538  * Interrupts routine are disabled, All memory allocated by this
539  * driver are freed.
540  */
541 static int
542 igb_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
543 {
544 	igb_t *igb;
545 
546 	/*
547 	 * Check detach command
548 	 */
549 	switch (cmd) {
550 	default:
551 		return (DDI_FAILURE);
552 
553 	case DDI_SUSPEND:
554 		return (igb_suspend(devinfo));
555 
556 	case DDI_DETACH:
557 		break;
558 	}
559 
560 
561 	/*
562 	 * Get the pointer to the driver private data structure
563 	 */
564 	igb = (igb_t *)ddi_get_driver_private(devinfo);
565 	if (igb == NULL)
566 		return (DDI_FAILURE);
567 
568 	/*
569 	 * Unregister MAC. If failed, we have to fail the detach
570 	 */
571 	if (mac_unregister(igb->mac_hdl) != 0) {
572 		igb_error(igb, "Failed to unregister MAC");
573 		return (DDI_FAILURE);
574 	}
575 	igb->attach_progress &= ~ATTACH_PROGRESS_MAC;
576 
577 	/*
578 	 * If the device is still running, it needs to be stopped first.
579 	 * This check is necessary because under some specific circumstances,
580 	 * the detach routine can be called without stopping the interface
581 	 * first.
582 	 */
583 	mutex_enter(&igb->gen_lock);
584 	if (igb->igb_state & IGB_STARTED) {
585 		atomic_and_32(&igb->igb_state, ~IGB_STARTED);
586 		igb_stop(igb, B_TRUE);
587 		mutex_exit(&igb->gen_lock);
588 		/* Disable and stop the watchdog timer */
589 		igb_disable_watchdog_timer(igb);
590 	} else
591 		mutex_exit(&igb->gen_lock);
592 
593 	/*
594 	 * Check if there are still rx buffers held by the upper layer.
595 	 * If so, fail the detach.
596 	 */
597 	if (!igb_rx_drain(igb))
598 		return (DDI_FAILURE);
599 
600 	/*
601 	 * Do the remaining unconfigure routines
602 	 */
603 	igb_unconfigure(devinfo, igb);
604 
605 	return (DDI_SUCCESS);
606 }
607 
608 /*
609  * quiesce(9E) entry point.
610  *
611  * This function is called when the system is single-threaded at high
612  * PIL with preemption disabled. Therefore, this function must not be
613  * blocked.
614  *
615  * This function returns DDI_SUCCESS on success, or DDI_FAILURE on failure.
616  * DDI_FAILURE indicates an error condition and should almost never happen.
617  */
618 static int
619 igb_quiesce(dev_info_t *devinfo)
620 {
621 	igb_t *igb;
622 	struct e1000_hw *hw;
623 
624 	igb = (igb_t *)ddi_get_driver_private(devinfo);
625 
626 	if (igb == NULL)
627 		return (DDI_FAILURE);
628 
629 	hw = &igb->hw;
630 
631 	/*
632 	 * Disable the adapter interrupts
633 	 */
634 	igb_disable_adapter_interrupts(igb);
635 
636 	/* Tell firmware driver is no longer in control */
637 	igb_release_driver_control(hw);
638 
639 	/*
640 	 * Reset the chipset
641 	 */
642 	(void) e1000_reset_hw(hw);
643 
644 	/*
645 	 * Reset PHY if possible
646 	 */
647 	if (e1000_check_reset_block(hw) == E1000_SUCCESS)
648 		(void) e1000_phy_hw_reset(hw);
649 
650 	return (DDI_SUCCESS);
651 }
652 
653 /*
654  * igb_unconfigure - release all resources held by this instance
655  */
656 static void
657 igb_unconfigure(dev_info_t *devinfo, igb_t *igb)
658 {
659 	/*
660 	 * Disable interrupt
661 	 */
662 	if (igb->attach_progress & ATTACH_PROGRESS_ENABLE_INTR) {
663 		(void) igb_disable_intrs(igb);
664 	}
665 
666 	/*
667 	 * Unregister MAC
668 	 */
669 	if (igb->attach_progress & ATTACH_PROGRESS_MAC) {
670 		(void) mac_unregister(igb->mac_hdl);
671 	}
672 
673 	/*
674 	 * Free statistics
675 	 */
676 	if (igb->attach_progress & ATTACH_PROGRESS_STATS) {
677 		kstat_delete((kstat_t *)igb->igb_ks);
678 	}
679 
680 	/*
681 	 * Remove interrupt handlers
682 	 */
683 	if (igb->attach_progress & ATTACH_PROGRESS_ADD_INTR) {
684 		igb_rem_intr_handlers(igb);
685 	}
686 
687 	/*
688 	 * Remove interrupts
689 	 */
690 	if (igb->attach_progress & ATTACH_PROGRESS_ALLOC_INTR) {
691 		igb_rem_intrs(igb);
692 	}
693 
694 	/*
695 	 * Remove driver properties
696 	 */
697 	if (igb->attach_progress & ATTACH_PROGRESS_PROPS) {
698 		(void) ddi_prop_remove_all(devinfo);
699 	}
700 
701 	/*
702 	 * Stop the adapter
703 	 */
704 	if (igb->attach_progress & ATTACH_PROGRESS_INIT_ADAPTER) {
705 		mutex_enter(&igb->gen_lock);
706 		igb_stop_adapter(igb);
707 		mutex_exit(&igb->gen_lock);
708 		if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
709 			ddi_fm_service_impact(igb->dip, DDI_SERVICE_UNAFFECTED);
710 	}
711 
712 	/*
713 	 * Free multicast table
714 	 */
715 	igb_release_multicast(igb);
716 
717 	/*
718 	 * Free register handle
719 	 */
720 	if (igb->attach_progress & ATTACH_PROGRESS_REGS_MAP) {
721 		if (igb->osdep.reg_handle != NULL)
722 			ddi_regs_map_free(&igb->osdep.reg_handle);
723 	}
724 
725 	/*
726 	 * Free PCI config handle
727 	 */
728 	if (igb->attach_progress & ATTACH_PROGRESS_PCI_CONFIG) {
729 		if (igb->osdep.cfg_handle != NULL)
730 			pci_config_teardown(&igb->osdep.cfg_handle);
731 	}
732 
733 	/*
734 	 * Free locks
735 	 */
736 	if (igb->attach_progress & ATTACH_PROGRESS_LOCKS) {
737 		igb_destroy_locks(igb);
738 	}
739 
740 	/*
741 	 * Free the rx/tx rings
742 	 */
743 	if (igb->attach_progress & ATTACH_PROGRESS_ALLOC_RINGS) {
744 		igb_free_rings(igb);
745 	}
746 
747 	/*
748 	 * Remove FMA
749 	 */
750 	if (igb->attach_progress & ATTACH_PROGRESS_FMINIT) {
751 		igb_fm_fini(igb);
752 	}
753 
754 	/*
755 	 * Free the driver data structure
756 	 */
757 	kmem_free(igb, sizeof (igb_t));
758 
759 	ddi_set_driver_private(devinfo, NULL);
760 }
761 
762 /*
763  * igb_register_mac - Register the driver and its function pointers with
764  * the GLD interface
765  */
766 static int
767 igb_register_mac(igb_t *igb)
768 {
769 	struct e1000_hw *hw = &igb->hw;
770 	mac_register_t *mac;
771 	int status;
772 
773 	if ((mac = mac_alloc(MAC_VERSION)) == NULL)
774 		return (IGB_FAILURE);
775 
776 	mac->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
777 	mac->m_driver = igb;
778 	mac->m_dip = igb->dip;
779 	mac->m_src_addr = hw->mac.addr;
780 	mac->m_callbacks = &igb_m_callbacks;
781 	mac->m_min_sdu = 0;
782 	mac->m_max_sdu = igb->max_frame_size -
783 	    sizeof (struct ether_vlan_header) - ETHERFCSL;
784 	mac->m_margin = VLAN_TAGSZ;
785 	mac->m_priv_props = igb_priv_props;
786 	mac->m_priv_prop_count = IGB_MAX_PRIV_PROPS;
787 	mac->m_v12n = MAC_VIRT_LEVEL1;
788 
789 	status = mac_register(mac, &igb->mac_hdl);
790 
791 	mac_free(mac);
792 
793 	return ((status == 0) ? IGB_SUCCESS : IGB_FAILURE);
794 }
795 
796 /*
797  * igb_identify_hardware - Identify the type of the chipset
798  */
799 static int
800 igb_identify_hardware(igb_t *igb)
801 {
802 	struct e1000_hw *hw = &igb->hw;
803 	struct igb_osdep *osdep = &igb->osdep;
804 
805 	/*
806 	 * Get the device id
807 	 */
808 	hw->vendor_id =
809 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_VENID);
810 	hw->device_id =
811 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_DEVID);
812 	hw->revision_id =
813 	    pci_config_get8(osdep->cfg_handle, PCI_CONF_REVID);
814 	hw->subsystem_device_id =
815 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBSYSID);
816 	hw->subsystem_vendor_id =
817 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBVENID);
818 
819 	/*
820 	 * Set the mac type of the adapter based on the device id
821 	 */
822 	if (e1000_set_mac_type(hw) != E1000_SUCCESS) {
823 		return (IGB_FAILURE);
824 	}
825 
826 	/*
827 	 * Install adapter capabilities based on mac type
828 	 */
829 	switch (hw->mac.type) {
830 	case e1000_82575:
831 		igb->capab = &igb_82575_cap;
832 		break;
833 	case e1000_82576:
834 		igb->capab = &igb_82576_cap;
835 		break;
836 	case e1000_82580:
837 		igb->capab = &igb_82580_cap;
838 		break;
839 	default:
840 		return (IGB_FAILURE);
841 	}
842 
843 	return (IGB_SUCCESS);
844 }
845 
846 /*
847  * igb_regs_map - Map the device registers
848  */
849 static int
850 igb_regs_map(igb_t *igb)
851 {
852 	dev_info_t *devinfo = igb->dip;
853 	struct e1000_hw *hw = &igb->hw;
854 	struct igb_osdep *osdep = &igb->osdep;
855 	off_t mem_size;
856 
857 	/*
858 	 * First get the size of device registers to be mapped.
859 	 */
860 	if (ddi_dev_regsize(devinfo, IGB_ADAPTER_REGSET, &mem_size) !=
861 	    DDI_SUCCESS) {
862 		return (IGB_FAILURE);
863 	}
864 
865 	/*
866 	 * Call ddi_regs_map_setup() to map registers
867 	 */
868 	if ((ddi_regs_map_setup(devinfo, IGB_ADAPTER_REGSET,
869 	    (caddr_t *)&hw->hw_addr, 0,
870 	    mem_size, &igb_regs_acc_attr,
871 	    &osdep->reg_handle)) != DDI_SUCCESS) {
872 		return (IGB_FAILURE);
873 	}
874 
875 	return (IGB_SUCCESS);
876 }
877 
878 /*
879  * igb_init_properties - Initialize driver properties
880  */
881 static void
882 igb_init_properties(igb_t *igb)
883 {
884 	/*
885 	 * Get conf file properties, including link settings
886 	 * jumbo frames, ring number, descriptor number, etc.
887 	 */
888 	igb_get_conf(igb);
889 }
890 
891 /*
892  * igb_init_driver_settings - Initialize driver settings
893  *
894  * The settings include hardware function pointers, bus information,
895  * rx/tx rings settings, link state, and any other parameters that
896  * need to be setup during driver initialization.
897  */
898 static int
899 igb_init_driver_settings(igb_t *igb)
900 {
901 	struct e1000_hw *hw = &igb->hw;
902 	igb_rx_ring_t *rx_ring;
903 	igb_tx_ring_t *tx_ring;
904 	uint32_t rx_size;
905 	uint32_t tx_size;
906 	int i;
907 
908 	/*
909 	 * Initialize chipset specific hardware function pointers
910 	 */
911 	if (e1000_setup_init_funcs(hw, B_TRUE) != E1000_SUCCESS) {
912 		return (IGB_FAILURE);
913 	}
914 
915 	/*
916 	 * Get bus information
917 	 */
918 	if (e1000_get_bus_info(hw) != E1000_SUCCESS) {
919 		return (IGB_FAILURE);
920 	}
921 
922 	/*
923 	 * Get the system page size
924 	 */
925 	igb->page_size = ddi_ptob(igb->dip, (ulong_t)1);
926 
927 	/*
928 	 * Set rx buffer size
929 	 * The IP header alignment room is counted in the calculation.
930 	 * The rx buffer size is in unit of 1K that is required by the
931 	 * chipset hardware.
932 	 */
933 	rx_size = igb->max_frame_size + IPHDR_ALIGN_ROOM;
934 	igb->rx_buf_size = ((rx_size >> 10) +
935 	    ((rx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
936 
937 	/*
938 	 * Set tx buffer size
939 	 */
940 	tx_size = igb->max_frame_size;
941 	igb->tx_buf_size = ((tx_size >> 10) +
942 	    ((tx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
943 
944 	/*
945 	 * Initialize rx/tx rings parameters
946 	 */
947 	for (i = 0; i < igb->num_rx_rings; i++) {
948 		rx_ring = &igb->rx_rings[i];
949 		rx_ring->index = i;
950 		rx_ring->igb = igb;
951 	}
952 
953 	for (i = 0; i < igb->num_tx_rings; i++) {
954 		tx_ring = &igb->tx_rings[i];
955 		tx_ring->index = i;
956 		tx_ring->igb = igb;
957 		if (igb->tx_head_wb_enable)
958 			tx_ring->tx_recycle = igb_tx_recycle_head_wb;
959 		else
960 			tx_ring->tx_recycle = igb_tx_recycle_legacy;
961 
962 		tx_ring->ring_size = igb->tx_ring_size;
963 		tx_ring->free_list_size = igb->tx_ring_size +
964 		    (igb->tx_ring_size >> 1);
965 	}
966 
967 	/*
968 	 * Initialize values of interrupt throttling rates
969 	 */
970 	for (i = 1; i < MAX_NUM_EITR; i++)
971 		igb->intr_throttling[i] = igb->intr_throttling[0];
972 
973 	/*
974 	 * The initial link state should be "unknown"
975 	 */
976 	igb->link_state = LINK_STATE_UNKNOWN;
977 
978 	return (IGB_SUCCESS);
979 }
980 
981 /*
982  * igb_init_locks - Initialize locks
983  */
984 static void
985 igb_init_locks(igb_t *igb)
986 {
987 	igb_rx_ring_t *rx_ring;
988 	igb_tx_ring_t *tx_ring;
989 	int i;
990 
991 	for (i = 0; i < igb->num_rx_rings; i++) {
992 		rx_ring = &igb->rx_rings[i];
993 		mutex_init(&rx_ring->rx_lock, NULL,
994 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
995 	}
996 
997 	for (i = 0; i < igb->num_tx_rings; i++) {
998 		tx_ring = &igb->tx_rings[i];
999 		mutex_init(&tx_ring->tx_lock, NULL,
1000 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1001 		mutex_init(&tx_ring->recycle_lock, NULL,
1002 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1003 		mutex_init(&tx_ring->tcb_head_lock, NULL,
1004 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1005 		mutex_init(&tx_ring->tcb_tail_lock, NULL,
1006 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1007 	}
1008 
1009 	mutex_init(&igb->gen_lock, NULL,
1010 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1011 
1012 	mutex_init(&igb->watchdog_lock, NULL,
1013 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1014 
1015 	mutex_init(&igb->link_lock, NULL,
1016 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
1017 }
1018 
1019 /*
1020  * igb_destroy_locks - Destroy locks
1021  */
1022 static void
1023 igb_destroy_locks(igb_t *igb)
1024 {
1025 	igb_rx_ring_t *rx_ring;
1026 	igb_tx_ring_t *tx_ring;
1027 	int i;
1028 
1029 	for (i = 0; i < igb->num_rx_rings; i++) {
1030 		rx_ring = &igb->rx_rings[i];
1031 		mutex_destroy(&rx_ring->rx_lock);
1032 	}
1033 
1034 	for (i = 0; i < igb->num_tx_rings; i++) {
1035 		tx_ring = &igb->tx_rings[i];
1036 		mutex_destroy(&tx_ring->tx_lock);
1037 		mutex_destroy(&tx_ring->recycle_lock);
1038 		mutex_destroy(&tx_ring->tcb_head_lock);
1039 		mutex_destroy(&tx_ring->tcb_tail_lock);
1040 	}
1041 
1042 	mutex_destroy(&igb->gen_lock);
1043 	mutex_destroy(&igb->watchdog_lock);
1044 	mutex_destroy(&igb->link_lock);
1045 }
1046 
1047 static int
1048 igb_resume(dev_info_t *devinfo)
1049 {
1050 	igb_t *igb;
1051 
1052 	igb = (igb_t *)ddi_get_driver_private(devinfo);
1053 	if (igb == NULL)
1054 		return (DDI_FAILURE);
1055 
1056 	mutex_enter(&igb->gen_lock);
1057 
1058 	if (igb->igb_state & IGB_STARTED) {
1059 		if (igb_start(igb, B_FALSE) != IGB_SUCCESS) {
1060 			mutex_exit(&igb->gen_lock);
1061 			return (DDI_FAILURE);
1062 		}
1063 
1064 		/*
1065 		 * Enable and start the watchdog timer
1066 		 */
1067 		igb_enable_watchdog_timer(igb);
1068 	}
1069 
1070 	atomic_and_32(&igb->igb_state, ~IGB_SUSPENDED);
1071 
1072 	mutex_exit(&igb->gen_lock);
1073 
1074 	return (DDI_SUCCESS);
1075 }
1076 
1077 static int
1078 igb_suspend(dev_info_t *devinfo)
1079 {
1080 	igb_t *igb;
1081 
1082 	igb = (igb_t *)ddi_get_driver_private(devinfo);
1083 	if (igb == NULL)
1084 		return (DDI_FAILURE);
1085 
1086 	mutex_enter(&igb->gen_lock);
1087 
1088 	atomic_or_32(&igb->igb_state, IGB_SUSPENDED);
1089 
1090 	if (!(igb->igb_state & IGB_STARTED)) {
1091 		mutex_exit(&igb->gen_lock);
1092 		return (DDI_SUCCESS);
1093 	}
1094 
1095 	igb_stop(igb, B_FALSE);
1096 
1097 	mutex_exit(&igb->gen_lock);
1098 
1099 	/*
1100 	 * Disable and stop the watchdog timer
1101 	 */
1102 	igb_disable_watchdog_timer(igb);
1103 
1104 	return (DDI_SUCCESS);
1105 }
1106 
1107 static int
1108 igb_init(igb_t *igb)
1109 {
1110 	mutex_enter(&igb->gen_lock);
1111 
1112 	/*
1113 	 * Initilize the adapter
1114 	 */
1115 	if (igb_init_adapter(igb) != IGB_SUCCESS) {
1116 		mutex_exit(&igb->gen_lock);
1117 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1118 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1119 		return (IGB_FAILURE);
1120 	}
1121 
1122 	mutex_exit(&igb->gen_lock);
1123 
1124 	return (IGB_SUCCESS);
1125 }
1126 
1127 /*
1128  * igb_init_mac_address - Initialize the default MAC address
1129  *
1130  * On success, the MAC address is entered in the igb->hw.mac.addr
1131  * and hw->mac.perm_addr fields and the adapter's RAR(0) receive
1132  * address register.
1133  *
1134  * Important side effects:
1135  * 1. adapter is reset - this is required to put it in a known state.
1136  * 2. all of non-volatile memory (NVM) is read & checksummed - NVM is where
1137  * MAC address and all default settings are stored, so a valid checksum
1138  * is required.
1139  */
1140 static int
1141 igb_init_mac_address(igb_t *igb)
1142 {
1143 	struct e1000_hw *hw = &igb->hw;
1144 
1145 	ASSERT(mutex_owned(&igb->gen_lock));
1146 
1147 	/*
1148 	 * Reset chipset to put the hardware in a known state
1149 	 * before we try to get MAC address from NVM.
1150 	 */
1151 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1152 		igb_error(igb, "Adapter reset failed.");
1153 		goto init_mac_fail;
1154 	}
1155 
1156 	/*
1157 	 * NVM validation
1158 	 */
1159 	if (e1000_validate_nvm_checksum(hw) < 0) {
1160 		/*
1161 		 * Some PCI-E parts fail the first check due to
1162 		 * the link being in sleep state.  Call it again,
1163 		 * if it fails a second time its a real issue.
1164 		 */
1165 		if (e1000_validate_nvm_checksum(hw) < 0) {
1166 			igb_error(igb,
1167 			    "Invalid NVM checksum. Please contact "
1168 			    "the vendor to update the NVM.");
1169 			goto init_mac_fail;
1170 		}
1171 	}
1172 
1173 	/*
1174 	 * Get the mac address
1175 	 * This function should handle SPARC case correctly.
1176 	 */
1177 	if (!igb_find_mac_address(igb)) {
1178 		igb_error(igb, "Failed to get the mac address");
1179 		goto init_mac_fail;
1180 	}
1181 
1182 	/* Validate mac address */
1183 	if (!is_valid_mac_addr(hw->mac.addr)) {
1184 		igb_error(igb, "Invalid mac address");
1185 		goto init_mac_fail;
1186 	}
1187 
1188 	return (IGB_SUCCESS);
1189 
1190 init_mac_fail:
1191 	return (IGB_FAILURE);
1192 }
1193 
1194 /*
1195  * igb_init_adapter - Initialize the adapter
1196  */
1197 static int
1198 igb_init_adapter(igb_t *igb)
1199 {
1200 	struct e1000_hw *hw = &igb->hw;
1201 	uint32_t pba;
1202 	uint32_t high_water;
1203 	int i;
1204 
1205 	ASSERT(mutex_owned(&igb->gen_lock));
1206 
1207 	/*
1208 	 * In order to obtain the default MAC address, this will reset the
1209 	 * adapter and validate the NVM that the address and many other
1210 	 * default settings come from.
1211 	 */
1212 	if (igb_init_mac_address(igb) != IGB_SUCCESS) {
1213 		igb_error(igb, "Failed to initialize MAC address");
1214 		goto init_adapter_fail;
1215 	}
1216 
1217 	/*
1218 	 * Setup flow control
1219 	 *
1220 	 * These parameters set thresholds for the adapter's generation(Tx)
1221 	 * and response(Rx) to Ethernet PAUSE frames.  These are just threshold
1222 	 * settings.  Flow control is enabled or disabled in the configuration
1223 	 * file.
1224 	 * High-water mark is set down from the top of the rx fifo (not
1225 	 * sensitive to max_frame_size) and low-water is set just below
1226 	 * high-water mark.
1227 	 * The high water mark must be low enough to fit one full frame above
1228 	 * it in the rx FIFO.  Should be the lower of:
1229 	 * 90% of the Rx FIFO size, or the full Rx FIFO size minus one full
1230 	 * frame.
1231 	 */
1232 	/*
1233 	 * The default setting of PBA is correct for 82575 and other supported
1234 	 * adapters do not have the E1000_PBA register, so PBA value is only
1235 	 * used for calculation here and is never written to the adapter.
1236 	 */
1237 	if (hw->mac.type == e1000_82575) {
1238 		pba = E1000_PBA_34K;
1239 	} else {
1240 		pba = E1000_PBA_64K;
1241 	}
1242 
1243 	high_water = min(((pba << 10) * 9 / 10),
1244 	    ((pba << 10) - igb->max_frame_size));
1245 
1246 	if (hw->mac.type == e1000_82575) {
1247 		/* 8-byte granularity */
1248 		hw->fc.high_water = high_water & 0xFFF8;
1249 		hw->fc.low_water = hw->fc.high_water - 8;
1250 	} else {
1251 		/* 16-byte granularity */
1252 		hw->fc.high_water = high_water & 0xFFF0;
1253 		hw->fc.low_water = hw->fc.high_water - 16;
1254 	}
1255 
1256 	hw->fc.pause_time = E1000_FC_PAUSE_TIME;
1257 	hw->fc.send_xon = B_TRUE;
1258 
1259 	(void) e1000_validate_mdi_setting(hw);
1260 
1261 	/*
1262 	 * Reset the chipset hardware the second time to put PBA settings
1263 	 * into effect.
1264 	 */
1265 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1266 		igb_error(igb, "Second reset failed");
1267 		goto init_adapter_fail;
1268 	}
1269 
1270 	/*
1271 	 * Don't wait for auto-negotiation to complete
1272 	 */
1273 	hw->phy.autoneg_wait_to_complete = B_FALSE;
1274 
1275 	/*
1276 	 * Copper options
1277 	 */
1278 	if (hw->phy.media_type == e1000_media_type_copper) {
1279 		hw->phy.mdix = 0;	/* AUTO_ALL_MODES */
1280 		hw->phy.disable_polarity_correction = B_FALSE;
1281 		hw->phy.ms_type = e1000_ms_hw_default; /* E1000_MASTER_SLAVE */
1282 	}
1283 
1284 	/*
1285 	 * Initialize link settings
1286 	 */
1287 	(void) igb_setup_link(igb, B_FALSE);
1288 
1289 	/*
1290 	 * Configure/Initialize hardware
1291 	 */
1292 	if (e1000_init_hw(hw) != E1000_SUCCESS) {
1293 		igb_error(igb, "Failed to initialize hardware");
1294 		goto init_adapter_fail;
1295 	}
1296 
1297 	/*
1298 	 *  Start the link setup timer
1299 	 */
1300 	igb_start_link_timer(igb);
1301 
1302 	/*
1303 	 * Disable wakeup control by default
1304 	 */
1305 	E1000_WRITE_REG(hw, E1000_WUC, 0);
1306 
1307 	/*
1308 	 * Record phy info in hw struct
1309 	 */
1310 	(void) e1000_get_phy_info(hw);
1311 
1312 	/*
1313 	 * Make sure driver has control
1314 	 */
1315 	igb_get_driver_control(hw);
1316 
1317 	/*
1318 	 * Restore LED settings to the default from EEPROM
1319 	 * to meet the standard for Sun platforms.
1320 	 */
1321 	(void) e1000_cleanup_led(hw);
1322 
1323 	/*
1324 	 * Setup MSI-X interrupts
1325 	 */
1326 	if (igb->intr_type == DDI_INTR_TYPE_MSIX)
1327 		igb->capab->setup_msix(igb);
1328 
1329 	/*
1330 	 * Initialize unicast addresses.
1331 	 */
1332 	igb_init_unicst(igb);
1333 
1334 	/*
1335 	 * Setup and initialize the mctable structures.
1336 	 */
1337 	igb_setup_multicst(igb);
1338 
1339 	/*
1340 	 * Set interrupt throttling rate
1341 	 */
1342 	for (i = 0; i < igb->intr_cnt; i++)
1343 		E1000_WRITE_REG(hw, E1000_EITR(i), igb->intr_throttling[i]);
1344 
1345 	/*
1346 	 * Save the state of the phy
1347 	 */
1348 	igb_get_phy_state(igb);
1349 
1350 	igb_param_sync(igb);
1351 
1352 	return (IGB_SUCCESS);
1353 
1354 init_adapter_fail:
1355 	/*
1356 	 * Reset PHY if possible
1357 	 */
1358 	if (e1000_check_reset_block(hw) == E1000_SUCCESS)
1359 		(void) e1000_phy_hw_reset(hw);
1360 
1361 	return (IGB_FAILURE);
1362 }
1363 
1364 /*
1365  * igb_stop_adapter - Stop the adapter
1366  */
1367 static void
1368 igb_stop_adapter(igb_t *igb)
1369 {
1370 	struct e1000_hw *hw = &igb->hw;
1371 
1372 	ASSERT(mutex_owned(&igb->gen_lock));
1373 
1374 	/* Stop the link setup timer */
1375 	igb_stop_link_timer(igb);
1376 
1377 	/* Tell firmware driver is no longer in control */
1378 	igb_release_driver_control(hw);
1379 
1380 	/*
1381 	 * Reset the chipset
1382 	 */
1383 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1384 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1385 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1386 	}
1387 
1388 	/*
1389 	 * e1000_phy_hw_reset is not needed here, MAC reset above is sufficient
1390 	 */
1391 }
1392 
1393 /*
1394  * igb_reset - Reset the chipset and restart the driver.
1395  *
1396  * It involves stopping and re-starting the chipset,
1397  * and re-configuring the rx/tx rings.
1398  */
1399 static int
1400 igb_reset(igb_t *igb)
1401 {
1402 	int i;
1403 
1404 	mutex_enter(&igb->gen_lock);
1405 
1406 	ASSERT(igb->igb_state & IGB_STARTED);
1407 	atomic_and_32(&igb->igb_state, ~IGB_STARTED);
1408 
1409 	/*
1410 	 * Disable the adapter interrupts to stop any rx/tx activities
1411 	 * before draining pending data and resetting hardware.
1412 	 */
1413 	igb_disable_adapter_interrupts(igb);
1414 
1415 	/*
1416 	 * Drain the pending transmit packets
1417 	 */
1418 	(void) igb_tx_drain(igb);
1419 
1420 	for (i = 0; i < igb->num_rx_rings; i++)
1421 		mutex_enter(&igb->rx_rings[i].rx_lock);
1422 	for (i = 0; i < igb->num_tx_rings; i++)
1423 		mutex_enter(&igb->tx_rings[i].tx_lock);
1424 
1425 	/*
1426 	 * Stop the adapter
1427 	 */
1428 	igb_stop_adapter(igb);
1429 
1430 	/*
1431 	 * Clean the pending tx data/resources
1432 	 */
1433 	igb_tx_clean(igb);
1434 
1435 	/*
1436 	 * Start the adapter
1437 	 */
1438 	if (igb_init_adapter(igb) != IGB_SUCCESS) {
1439 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1440 		goto reset_failure;
1441 	}
1442 
1443 	/*
1444 	 * Setup the rx/tx rings
1445 	 */
1446 	igb->tx_ring_init = B_FALSE;
1447 	igb_setup_rings(igb);
1448 
1449 	atomic_and_32(&igb->igb_state, ~(IGB_ERROR | IGB_STALL));
1450 
1451 	/*
1452 	 * Enable adapter interrupts
1453 	 * The interrupts must be enabled after the driver state is START
1454 	 */
1455 	igb->capab->enable_intr(igb);
1456 
1457 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK)
1458 		goto reset_failure;
1459 
1460 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1461 		goto reset_failure;
1462 
1463 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1464 		mutex_exit(&igb->tx_rings[i].tx_lock);
1465 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1466 		mutex_exit(&igb->rx_rings[i].rx_lock);
1467 
1468 	atomic_or_32(&igb->igb_state, IGB_STARTED);
1469 
1470 	mutex_exit(&igb->gen_lock);
1471 
1472 	return (IGB_SUCCESS);
1473 
1474 reset_failure:
1475 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1476 		mutex_exit(&igb->tx_rings[i].tx_lock);
1477 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1478 		mutex_exit(&igb->rx_rings[i].rx_lock);
1479 
1480 	mutex_exit(&igb->gen_lock);
1481 
1482 	ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1483 
1484 	return (IGB_FAILURE);
1485 }
1486 
1487 /*
1488  * igb_tx_clean - Clean the pending transmit packets and DMA resources
1489  */
1490 static void
1491 igb_tx_clean(igb_t *igb)
1492 {
1493 	igb_tx_ring_t *tx_ring;
1494 	tx_control_block_t *tcb;
1495 	link_list_t pending_list;
1496 	uint32_t desc_num;
1497 	int i, j;
1498 
1499 	LINK_LIST_INIT(&pending_list);
1500 
1501 	for (i = 0; i < igb->num_tx_rings; i++) {
1502 		tx_ring = &igb->tx_rings[i];
1503 
1504 		mutex_enter(&tx_ring->recycle_lock);
1505 
1506 		/*
1507 		 * Clean the pending tx data - the pending packets in the
1508 		 * work_list that have no chances to be transmitted again.
1509 		 *
1510 		 * We must ensure the chipset is stopped or the link is down
1511 		 * before cleaning the transmit packets.
1512 		 */
1513 		desc_num = 0;
1514 		for (j = 0; j < tx_ring->ring_size; j++) {
1515 			tcb = tx_ring->work_list[j];
1516 			if (tcb != NULL) {
1517 				desc_num += tcb->desc_num;
1518 
1519 				tx_ring->work_list[j] = NULL;
1520 
1521 				igb_free_tcb(tcb);
1522 
1523 				LIST_PUSH_TAIL(&pending_list, &tcb->link);
1524 			}
1525 		}
1526 
1527 		if (desc_num > 0) {
1528 			atomic_add_32(&tx_ring->tbd_free, desc_num);
1529 			ASSERT(tx_ring->tbd_free == tx_ring->ring_size);
1530 
1531 			/*
1532 			 * Reset the head and tail pointers of the tbd ring;
1533 			 * Reset the head write-back if it is enabled.
1534 			 */
1535 			tx_ring->tbd_head = 0;
1536 			tx_ring->tbd_tail = 0;
1537 			if (igb->tx_head_wb_enable)
1538 				*tx_ring->tbd_head_wb = 0;
1539 
1540 			E1000_WRITE_REG(&igb->hw, E1000_TDH(tx_ring->index), 0);
1541 			E1000_WRITE_REG(&igb->hw, E1000_TDT(tx_ring->index), 0);
1542 		}
1543 
1544 		mutex_exit(&tx_ring->recycle_lock);
1545 
1546 		/*
1547 		 * Add the tx control blocks in the pending list to
1548 		 * the free list.
1549 		 */
1550 		igb_put_free_list(tx_ring, &pending_list);
1551 	}
1552 }
1553 
1554 /*
1555  * igb_tx_drain - Drain the tx rings to allow pending packets to be transmitted
1556  */
1557 static boolean_t
1558 igb_tx_drain(igb_t *igb)
1559 {
1560 	igb_tx_ring_t *tx_ring;
1561 	boolean_t done;
1562 	int i, j;
1563 
1564 	/*
1565 	 * Wait for a specific time to allow pending tx packets
1566 	 * to be transmitted.
1567 	 *
1568 	 * Check the counter tbd_free to see if transmission is done.
1569 	 * No lock protection is needed here.
1570 	 *
1571 	 * Return B_TRUE if all pending packets have been transmitted;
1572 	 * Otherwise return B_FALSE;
1573 	 */
1574 	for (i = 0; i < TX_DRAIN_TIME; i++) {
1575 
1576 		done = B_TRUE;
1577 		for (j = 0; j < igb->num_tx_rings; j++) {
1578 			tx_ring = &igb->tx_rings[j];
1579 			done = done &&
1580 			    (tx_ring->tbd_free == tx_ring->ring_size);
1581 		}
1582 
1583 		if (done)
1584 			break;
1585 
1586 		msec_delay(1);
1587 	}
1588 
1589 	return (done);
1590 }
1591 
1592 /*
1593  * igb_rx_drain - Wait for all rx buffers to be released by upper layer
1594  */
1595 static boolean_t
1596 igb_rx_drain(igb_t *igb)
1597 {
1598 	boolean_t done;
1599 	int i;
1600 
1601 	/*
1602 	 * Polling the rx free list to check if those rx buffers held by
1603 	 * the upper layer are released.
1604 	 *
1605 	 * Check the counter rcb_free to see if all pending buffers are
1606 	 * released. No lock protection is needed here.
1607 	 *
1608 	 * Return B_TRUE if all pending buffers have been released;
1609 	 * Otherwise return B_FALSE;
1610 	 */
1611 	for (i = 0; i < RX_DRAIN_TIME; i++) {
1612 		done = (igb->rcb_pending == 0);
1613 
1614 		if (done)
1615 			break;
1616 
1617 		msec_delay(1);
1618 	}
1619 
1620 	return (done);
1621 }
1622 
1623 /*
1624  * igb_start - Start the driver/chipset
1625  */
1626 int
1627 igb_start(igb_t *igb, boolean_t alloc_buffer)
1628 {
1629 	int i;
1630 
1631 	ASSERT(mutex_owned(&igb->gen_lock));
1632 
1633 	if (alloc_buffer) {
1634 		if (igb_alloc_rx_data(igb) != IGB_SUCCESS) {
1635 			igb_error(igb,
1636 			    "Failed to allocate software receive rings");
1637 			return (IGB_FAILURE);
1638 		}
1639 
1640 		/* Allocate buffers for all the rx/tx rings */
1641 		if (igb_alloc_dma(igb) != IGB_SUCCESS) {
1642 			igb_error(igb, "Failed to allocate DMA resource");
1643 			return (IGB_FAILURE);
1644 		}
1645 
1646 		igb->tx_ring_init = B_TRUE;
1647 	} else {
1648 		igb->tx_ring_init = B_FALSE;
1649 	}
1650 
1651 	for (i = 0; i < igb->num_rx_rings; i++)
1652 		mutex_enter(&igb->rx_rings[i].rx_lock);
1653 	for (i = 0; i < igb->num_tx_rings; i++)
1654 		mutex_enter(&igb->tx_rings[i].tx_lock);
1655 
1656 	/*
1657 	 * Start the adapter
1658 	 */
1659 	if ((igb->attach_progress & ATTACH_PROGRESS_INIT_ADAPTER) == 0) {
1660 		if (igb_init_adapter(igb) != IGB_SUCCESS) {
1661 			igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1662 			goto start_failure;
1663 		}
1664 		igb->attach_progress |= ATTACH_PROGRESS_INIT_ADAPTER;
1665 	}
1666 
1667 	/*
1668 	 * Setup the rx/tx rings
1669 	 */
1670 	igb_setup_rings(igb);
1671 
1672 	/*
1673 	 * Enable adapter interrupts
1674 	 * The interrupts must be enabled after the driver state is START
1675 	 */
1676 	igb->capab->enable_intr(igb);
1677 
1678 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK)
1679 		goto start_failure;
1680 
1681 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1682 		goto start_failure;
1683 
1684 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1685 		mutex_exit(&igb->tx_rings[i].tx_lock);
1686 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1687 		mutex_exit(&igb->rx_rings[i].rx_lock);
1688 
1689 	return (IGB_SUCCESS);
1690 
1691 start_failure:
1692 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1693 		mutex_exit(&igb->tx_rings[i].tx_lock);
1694 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1695 		mutex_exit(&igb->rx_rings[i].rx_lock);
1696 
1697 	ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1698 
1699 	return (IGB_FAILURE);
1700 }
1701 
1702 /*
1703  * igb_stop - Stop the driver/chipset
1704  */
1705 void
1706 igb_stop(igb_t *igb, boolean_t free_buffer)
1707 {
1708 	int i;
1709 
1710 	ASSERT(mutex_owned(&igb->gen_lock));
1711 
1712 	igb->attach_progress &= ~ATTACH_PROGRESS_INIT_ADAPTER;
1713 
1714 	/*
1715 	 * Disable the adapter interrupts
1716 	 */
1717 	igb_disable_adapter_interrupts(igb);
1718 
1719 	/*
1720 	 * Drain the pending tx packets
1721 	 */
1722 	(void) igb_tx_drain(igb);
1723 
1724 	for (i = 0; i < igb->num_rx_rings; i++)
1725 		mutex_enter(&igb->rx_rings[i].rx_lock);
1726 	for (i = 0; i < igb->num_tx_rings; i++)
1727 		mutex_enter(&igb->tx_rings[i].tx_lock);
1728 
1729 	/*
1730 	 * Stop the adapter
1731 	 */
1732 	igb_stop_adapter(igb);
1733 
1734 	/*
1735 	 * Clean the pending tx data/resources
1736 	 */
1737 	igb_tx_clean(igb);
1738 
1739 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1740 		mutex_exit(&igb->tx_rings[i].tx_lock);
1741 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1742 		mutex_exit(&igb->rx_rings[i].rx_lock);
1743 
1744 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1745 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1746 
1747 	if (igb->link_state == LINK_STATE_UP) {
1748 		igb->link_state = LINK_STATE_UNKNOWN;
1749 		mac_link_update(igb->mac_hdl, igb->link_state);
1750 	}
1751 
1752 	if (free_buffer) {
1753 		/*
1754 		 * Release the DMA/memory resources of rx/tx rings
1755 		 */
1756 		igb_free_dma(igb);
1757 		igb_free_rx_data(igb);
1758 	}
1759 }
1760 
1761 /*
1762  * igb_alloc_rings - Allocate memory space for rx/tx rings
1763  */
1764 static int
1765 igb_alloc_rings(igb_t *igb)
1766 {
1767 	/*
1768 	 * Allocate memory space for rx rings
1769 	 */
1770 	igb->rx_rings = kmem_zalloc(
1771 	    sizeof (igb_rx_ring_t) * igb->num_rx_rings,
1772 	    KM_NOSLEEP);
1773 
1774 	if (igb->rx_rings == NULL) {
1775 		return (IGB_FAILURE);
1776 	}
1777 
1778 	/*
1779 	 * Allocate memory space for tx rings
1780 	 */
1781 	igb->tx_rings = kmem_zalloc(
1782 	    sizeof (igb_tx_ring_t) * igb->num_tx_rings,
1783 	    KM_NOSLEEP);
1784 
1785 	if (igb->tx_rings == NULL) {
1786 		kmem_free(igb->rx_rings,
1787 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1788 		igb->rx_rings = NULL;
1789 		return (IGB_FAILURE);
1790 	}
1791 
1792 	/*
1793 	 * Allocate memory space for rx ring groups
1794 	 */
1795 	igb->rx_groups = kmem_zalloc(
1796 	    sizeof (igb_rx_group_t) * igb->num_rx_groups,
1797 	    KM_NOSLEEP);
1798 
1799 	if (igb->rx_groups == NULL) {
1800 		kmem_free(igb->rx_rings,
1801 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1802 		kmem_free(igb->tx_rings,
1803 		    sizeof (igb_tx_ring_t) * igb->num_tx_rings);
1804 		igb->rx_rings = NULL;
1805 		igb->tx_rings = NULL;
1806 		return (IGB_FAILURE);
1807 	}
1808 
1809 	return (IGB_SUCCESS);
1810 }
1811 
1812 /*
1813  * igb_free_rings - Free the memory space of rx/tx rings.
1814  */
1815 static void
1816 igb_free_rings(igb_t *igb)
1817 {
1818 	if (igb->rx_rings != NULL) {
1819 		kmem_free(igb->rx_rings,
1820 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1821 		igb->rx_rings = NULL;
1822 	}
1823 
1824 	if (igb->tx_rings != NULL) {
1825 		kmem_free(igb->tx_rings,
1826 		    sizeof (igb_tx_ring_t) * igb->num_tx_rings);
1827 		igb->tx_rings = NULL;
1828 	}
1829 
1830 	if (igb->rx_groups != NULL) {
1831 		kmem_free(igb->rx_groups,
1832 		    sizeof (igb_rx_group_t) * igb->num_rx_groups);
1833 		igb->rx_groups = NULL;
1834 	}
1835 }
1836 
1837 static int
1838 igb_alloc_rx_data(igb_t *igb)
1839 {
1840 	igb_rx_ring_t *rx_ring;
1841 	int i;
1842 
1843 	for (i = 0; i < igb->num_rx_rings; i++) {
1844 		rx_ring = &igb->rx_rings[i];
1845 		if (igb_alloc_rx_ring_data(rx_ring) != IGB_SUCCESS)
1846 			goto alloc_rx_rings_failure;
1847 	}
1848 	return (IGB_SUCCESS);
1849 
1850 alloc_rx_rings_failure:
1851 	igb_free_rx_data(igb);
1852 	return (IGB_FAILURE);
1853 }
1854 
1855 static void
1856 igb_free_rx_data(igb_t *igb)
1857 {
1858 	igb_rx_ring_t *rx_ring;
1859 	igb_rx_data_t *rx_data;
1860 	int i;
1861 
1862 	for (i = 0; i < igb->num_rx_rings; i++) {
1863 		rx_ring = &igb->rx_rings[i];
1864 
1865 		mutex_enter(&igb->rx_pending_lock);
1866 		rx_data = rx_ring->rx_data;
1867 
1868 		if (rx_data != NULL) {
1869 			rx_data->flag |= IGB_RX_STOPPED;
1870 
1871 			if (rx_data->rcb_pending == 0) {
1872 				igb_free_rx_ring_data(rx_data);
1873 				rx_ring->rx_data = NULL;
1874 			}
1875 		}
1876 
1877 		mutex_exit(&igb->rx_pending_lock);
1878 	}
1879 }
1880 
1881 /*
1882  * igb_setup_rings - Setup rx/tx rings
1883  */
1884 static void
1885 igb_setup_rings(igb_t *igb)
1886 {
1887 	/*
1888 	 * Setup the rx/tx rings, including the following:
1889 	 *
1890 	 * 1. Setup the descriptor ring and the control block buffers;
1891 	 * 2. Initialize necessary registers for receive/transmit;
1892 	 * 3. Initialize software pointers/parameters for receive/transmit;
1893 	 */
1894 	igb_setup_rx(igb);
1895 
1896 	igb_setup_tx(igb);
1897 }
1898 
1899 static void
1900 igb_setup_rx_ring(igb_rx_ring_t *rx_ring)
1901 {
1902 	igb_t *igb = rx_ring->igb;
1903 	igb_rx_data_t *rx_data = rx_ring->rx_data;
1904 	struct e1000_hw *hw = &igb->hw;
1905 	rx_control_block_t *rcb;
1906 	union e1000_adv_rx_desc	*rbd;
1907 	uint32_t size;
1908 	uint32_t buf_low;
1909 	uint32_t buf_high;
1910 	uint32_t rxdctl;
1911 	int i;
1912 
1913 	ASSERT(mutex_owned(&rx_ring->rx_lock));
1914 	ASSERT(mutex_owned(&igb->gen_lock));
1915 
1916 	/*
1917 	 * Initialize descriptor ring with buffer addresses
1918 	 */
1919 	for (i = 0; i < igb->rx_ring_size; i++) {
1920 		rcb = rx_data->work_list[i];
1921 		rbd = &rx_data->rbd_ring[i];
1922 
1923 		rbd->read.pkt_addr = rcb->rx_buf.dma_address;
1924 		rbd->read.hdr_addr = NULL;
1925 	}
1926 
1927 	/*
1928 	 * Initialize the base address registers
1929 	 */
1930 	buf_low = (uint32_t)rx_data->rbd_area.dma_address;
1931 	buf_high = (uint32_t)(rx_data->rbd_area.dma_address >> 32);
1932 	E1000_WRITE_REG(hw, E1000_RDBAH(rx_ring->index), buf_high);
1933 	E1000_WRITE_REG(hw, E1000_RDBAL(rx_ring->index), buf_low);
1934 
1935 	/*
1936 	 * Initialize the length register
1937 	 */
1938 	size = rx_data->ring_size * sizeof (union e1000_adv_rx_desc);
1939 	E1000_WRITE_REG(hw, E1000_RDLEN(rx_ring->index), size);
1940 
1941 	/*
1942 	 * Initialize buffer size & descriptor type
1943 	 */
1944 	E1000_WRITE_REG(hw, E1000_SRRCTL(rx_ring->index),
1945 	    ((igb->rx_buf_size >> E1000_SRRCTL_BSIZEPKT_SHIFT) |
1946 	    E1000_SRRCTL_DESCTYPE_ADV_ONEBUF));
1947 
1948 	/*
1949 	 * Setup the Receive Descriptor Control Register (RXDCTL)
1950 	 */
1951 	rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(rx_ring->index));
1952 	rxdctl &= igb->capab->rxdctl_mask;
1953 	rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1954 	rxdctl |= 16;		/* pthresh */
1955 	rxdctl |= 8 << 8;	/* hthresh */
1956 	rxdctl |= 1 << 16;	/* wthresh */
1957 	E1000_WRITE_REG(hw, E1000_RXDCTL(rx_ring->index), rxdctl);
1958 
1959 	rx_data->rbd_next = 0;
1960 }
1961 
1962 static void
1963 igb_setup_rx(igb_t *igb)
1964 {
1965 	igb_rx_ring_t *rx_ring;
1966 	igb_rx_data_t *rx_data;
1967 	igb_rx_group_t *rx_group;
1968 	struct e1000_hw *hw = &igb->hw;
1969 	uint32_t rctl, rxcsum;
1970 	uint32_t ring_per_group;
1971 	int i;
1972 
1973 	/*
1974 	 * Setup the Receive Control Register (RCTL), and enable the
1975 	 * receiver. The initial configuration is to: enable the receiver,
1976 	 * accept broadcasts, discard bad packets, accept long packets,
1977 	 * disable VLAN filter checking, and set receive buffer size to
1978 	 * 2k.  For 82575, also set the receive descriptor minimum
1979 	 * threshold size to 1/2 the ring.
1980 	 */
1981 	rctl = E1000_READ_REG(hw, E1000_RCTL);
1982 
1983 	/*
1984 	 * Clear the field used for wakeup control.  This driver doesn't do
1985 	 * wakeup but leave this here for completeness.
1986 	 */
1987 	rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1988 	rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
1989 
1990 	rctl |= (E1000_RCTL_EN |	/* Enable Receive Unit */
1991 	    E1000_RCTL_BAM |		/* Accept Broadcast Packets */
1992 	    E1000_RCTL_LPE |		/* Large Packet Enable */
1993 					/* Multicast filter offset */
1994 	    (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT) |
1995 	    E1000_RCTL_RDMTS_HALF |	/* rx descriptor threshold */
1996 	    E1000_RCTL_SECRC);		/* Strip Ethernet CRC */
1997 
1998 	for (i = 0; i < igb->num_rx_groups; i++) {
1999 		rx_group = &igb->rx_groups[i];
2000 		rx_group->index = i;
2001 		rx_group->igb = igb;
2002 	}
2003 
2004 	/*
2005 	 * Set up all rx descriptor rings - must be called before receive unit
2006 	 * enabled.
2007 	 */
2008 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
2009 	for (i = 0; i < igb->num_rx_rings; i++) {
2010 		rx_ring = &igb->rx_rings[i];
2011 		igb_setup_rx_ring(rx_ring);
2012 
2013 		/*
2014 		 * Map a ring to a group by assigning a group index
2015 		 */
2016 		rx_ring->group_index = i / ring_per_group;
2017 	}
2018 
2019 	/*
2020 	 * Setup the Rx Long Packet Max Length register
2021 	 */
2022 	E1000_WRITE_REG(hw, E1000_RLPML, igb->max_frame_size);
2023 
2024 	/*
2025 	 * Hardware checksum settings
2026 	 */
2027 	if (igb->rx_hcksum_enable) {
2028 		rxcsum =
2029 		    E1000_RXCSUM_TUOFL |	/* TCP/UDP checksum */
2030 		    E1000_RXCSUM_IPOFL;		/* IP checksum */
2031 
2032 		E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2033 	}
2034 
2035 	/*
2036 	 * Setup classify and RSS for multiple receive queues
2037 	 */
2038 	switch (igb->vmdq_mode) {
2039 	case E1000_VMDQ_OFF:
2040 		/*
2041 		 * One ring group, only RSS is needed when more than
2042 		 * one ring enabled.
2043 		 */
2044 		if (igb->num_rx_rings > 1)
2045 			igb_setup_rss(igb);
2046 		break;
2047 	case E1000_VMDQ_MAC:
2048 		/*
2049 		 * Multiple groups, each group has one ring,
2050 		 * only the MAC classification is needed.
2051 		 */
2052 		igb_setup_mac_classify(igb);
2053 		break;
2054 	case E1000_VMDQ_MAC_RSS:
2055 		/*
2056 		 * Multiple groups and multiple rings, both
2057 		 * MAC classification and RSS are needed.
2058 		 */
2059 		igb_setup_mac_rss_classify(igb);
2060 		break;
2061 	}
2062 
2063 	/*
2064 	 * Enable the receive unit - must be done after all
2065 	 * the rx setup above.
2066 	 */
2067 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
2068 
2069 	/*
2070 	 * Initialize all adapter ring head & tail pointers - must
2071 	 * be done after receive unit is enabled
2072 	 */
2073 	for (i = 0; i < igb->num_rx_rings; i++) {
2074 		rx_ring = &igb->rx_rings[i];
2075 		rx_data = rx_ring->rx_data;
2076 		E1000_WRITE_REG(hw, E1000_RDH(i), 0);
2077 		E1000_WRITE_REG(hw, E1000_RDT(i), rx_data->ring_size - 1);
2078 	}
2079 
2080 	/*
2081 	 * 82575 with manageability enabled needs a special flush to make
2082 	 * sure the fifos start clean.
2083 	 */
2084 	if ((hw->mac.type == e1000_82575) &&
2085 	    (E1000_READ_REG(hw, E1000_MANC) & E1000_MANC_RCV_TCO_EN)) {
2086 		e1000_rx_fifo_flush_82575(hw);
2087 	}
2088 }
2089 
2090 static void
2091 igb_setup_tx_ring(igb_tx_ring_t *tx_ring)
2092 {
2093 	igb_t *igb = tx_ring->igb;
2094 	struct e1000_hw *hw = &igb->hw;
2095 	uint32_t size;
2096 	uint32_t buf_low;
2097 	uint32_t buf_high;
2098 	uint32_t reg_val;
2099 
2100 	ASSERT(mutex_owned(&tx_ring->tx_lock));
2101 	ASSERT(mutex_owned(&igb->gen_lock));
2102 
2103 
2104 	/*
2105 	 * Initialize the length register
2106 	 */
2107 	size = tx_ring->ring_size * sizeof (union e1000_adv_tx_desc);
2108 	E1000_WRITE_REG(hw, E1000_TDLEN(tx_ring->index), size);
2109 
2110 	/*
2111 	 * Initialize the base address registers
2112 	 */
2113 	buf_low = (uint32_t)tx_ring->tbd_area.dma_address;
2114 	buf_high = (uint32_t)(tx_ring->tbd_area.dma_address >> 32);
2115 	E1000_WRITE_REG(hw, E1000_TDBAL(tx_ring->index), buf_low);
2116 	E1000_WRITE_REG(hw, E1000_TDBAH(tx_ring->index), buf_high);
2117 
2118 	/*
2119 	 * Setup head & tail pointers
2120 	 */
2121 	E1000_WRITE_REG(hw, E1000_TDH(tx_ring->index), 0);
2122 	E1000_WRITE_REG(hw, E1000_TDT(tx_ring->index), 0);
2123 
2124 	/*
2125 	 * Setup head write-back
2126 	 */
2127 	if (igb->tx_head_wb_enable) {
2128 		/*
2129 		 * The memory of the head write-back is allocated using
2130 		 * the extra tbd beyond the tail of the tbd ring.
2131 		 */
2132 		tx_ring->tbd_head_wb = (uint32_t *)
2133 		    ((uintptr_t)tx_ring->tbd_area.address + size);
2134 		*tx_ring->tbd_head_wb = 0;
2135 
2136 		buf_low = (uint32_t)
2137 		    (tx_ring->tbd_area.dma_address + size);
2138 		buf_high = (uint32_t)
2139 		    ((tx_ring->tbd_area.dma_address + size) >> 32);
2140 
2141 		/* Set the head write-back enable bit */
2142 		buf_low |= E1000_TX_HEAD_WB_ENABLE;
2143 
2144 		E1000_WRITE_REG(hw, E1000_TDWBAL(tx_ring->index), buf_low);
2145 		E1000_WRITE_REG(hw, E1000_TDWBAH(tx_ring->index), buf_high);
2146 
2147 		/*
2148 		 * Turn off relaxed ordering for head write back or it will
2149 		 * cause problems with the tx recycling
2150 		 */
2151 		reg_val = E1000_READ_REG(hw,
2152 		    E1000_DCA_TXCTRL(tx_ring->index));
2153 		reg_val &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
2154 		E1000_WRITE_REG(hw,
2155 		    E1000_DCA_TXCTRL(tx_ring->index), reg_val);
2156 	} else {
2157 		tx_ring->tbd_head_wb = NULL;
2158 	}
2159 
2160 	tx_ring->tbd_head = 0;
2161 	tx_ring->tbd_tail = 0;
2162 	tx_ring->tbd_free = tx_ring->ring_size;
2163 
2164 	if (igb->tx_ring_init == B_TRUE) {
2165 		tx_ring->tcb_head = 0;
2166 		tx_ring->tcb_tail = 0;
2167 		tx_ring->tcb_free = tx_ring->free_list_size;
2168 	}
2169 
2170 	/*
2171 	 * Enable TXDCTL per queue
2172 	 */
2173 	reg_val = E1000_READ_REG(hw, E1000_TXDCTL(tx_ring->index));
2174 	reg_val |= E1000_TXDCTL_QUEUE_ENABLE;
2175 	E1000_WRITE_REG(hw, E1000_TXDCTL(tx_ring->index), reg_val);
2176 
2177 	/*
2178 	 * Initialize hardware checksum offload settings
2179 	 */
2180 	bzero(&tx_ring->tx_context, sizeof (tx_context_t));
2181 }
2182 
2183 static void
2184 igb_setup_tx(igb_t *igb)
2185 {
2186 	igb_tx_ring_t *tx_ring;
2187 	struct e1000_hw *hw = &igb->hw;
2188 	uint32_t reg_val;
2189 	int i;
2190 
2191 	for (i = 0; i < igb->num_tx_rings; i++) {
2192 		tx_ring = &igb->tx_rings[i];
2193 		igb_setup_tx_ring(tx_ring);
2194 	}
2195 
2196 	/*
2197 	 * Setup the Transmit Control Register (TCTL)
2198 	 */
2199 	reg_val = E1000_READ_REG(hw, E1000_TCTL);
2200 	reg_val &= ~E1000_TCTL_CT;
2201 	reg_val |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2202 	    (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2203 
2204 	/* Enable transmits */
2205 	reg_val |= E1000_TCTL_EN;
2206 
2207 	E1000_WRITE_REG(hw, E1000_TCTL, reg_val);
2208 }
2209 
2210 /*
2211  * igb_setup_rss - Setup receive-side scaling feature
2212  */
2213 static void
2214 igb_setup_rss(igb_t *igb)
2215 {
2216 	struct e1000_hw *hw = &igb->hw;
2217 	uint32_t i, mrqc, rxcsum;
2218 	int shift = 0;
2219 	uint32_t random;
2220 	union e1000_reta {
2221 		uint32_t	dword;
2222 		uint8_t		bytes[4];
2223 	} reta;
2224 
2225 	/* Setup the Redirection Table */
2226 	if (hw->mac.type == e1000_82576) {
2227 		shift = 3;
2228 	} else if (hw->mac.type == e1000_82575) {
2229 		shift = 6;
2230 	}
2231 	for (i = 0; i < (32 * 4); i++) {
2232 		reta.bytes[i & 3] = (i % igb->num_rx_rings) << shift;
2233 		if ((i & 3) == 3) {
2234 			E1000_WRITE_REG(hw,
2235 			    (E1000_RETA(0) + (i & ~3)), reta.dword);
2236 		}
2237 	}
2238 
2239 	/* Fill out hash function seeds */
2240 	for (i = 0; i < 10; i++) {
2241 		(void) random_get_pseudo_bytes((uint8_t *)&random,
2242 		    sizeof (uint32_t));
2243 		E1000_WRITE_REG(hw, E1000_RSSRK(i), random);
2244 	}
2245 
2246 	/* Setup the Multiple Receive Queue Control register */
2247 	mrqc = E1000_MRQC_ENABLE_RSS_4Q;
2248 	mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2249 	    E1000_MRQC_RSS_FIELD_IPV4_TCP |
2250 	    E1000_MRQC_RSS_FIELD_IPV6 |
2251 	    E1000_MRQC_RSS_FIELD_IPV6_TCP |
2252 	    E1000_MRQC_RSS_FIELD_IPV4_UDP |
2253 	    E1000_MRQC_RSS_FIELD_IPV6_UDP |
2254 	    E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2255 	    E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2256 
2257 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2258 
2259 	/*
2260 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2261 	 *
2262 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2263 	 * checksum offloading provided by the 82575 chipset besides the IP
2264 	 * header checksum offloading and the TCP/UDP checksum offloading.
2265 	 * The Packet Checksum is by default computed over the entire packet
2266 	 * from the first byte of the DA through the last byte of the CRC,
2267 	 * including the Ethernet and IP headers.
2268 	 *
2269 	 * It is a hardware limitation that Packet Checksum is mutually
2270 	 * exclusive with RSS.
2271 	 */
2272 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2273 	rxcsum |= E1000_RXCSUM_PCSD;
2274 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2275 }
2276 
2277 /*
2278  * igb_setup_mac_rss_classify - Setup MAC classification and rss
2279  */
2280 static void
2281 igb_setup_mac_rss_classify(igb_t *igb)
2282 {
2283 	struct e1000_hw *hw = &igb->hw;
2284 	uint32_t i, mrqc, vmdctl, rxcsum;
2285 	uint32_t ring_per_group;
2286 	int shift_group0, shift_group1;
2287 	uint32_t random;
2288 	union e1000_reta {
2289 		uint32_t	dword;
2290 		uint8_t		bytes[4];
2291 	} reta;
2292 
2293 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
2294 
2295 	/* Setup the Redirection Table, it is shared between two groups */
2296 	shift_group0 = 2;
2297 	shift_group1 = 6;
2298 	for (i = 0; i < (32 * 4); i++) {
2299 		reta.bytes[i & 3] = ((i % ring_per_group) << shift_group0) |
2300 		    ((ring_per_group + (i % ring_per_group)) << shift_group1);
2301 		if ((i & 3) == 3) {
2302 			E1000_WRITE_REG(hw,
2303 			    (E1000_RETA(0) + (i & ~3)), reta.dword);
2304 		}
2305 	}
2306 
2307 	/* Fill out hash function seeds */
2308 	for (i = 0; i < 10; i++) {
2309 		(void) random_get_pseudo_bytes((uint8_t *)&random,
2310 		    sizeof (uint32_t));
2311 		E1000_WRITE_REG(hw, E1000_RSSRK(i), random);
2312 	}
2313 
2314 	/*
2315 	 * Setup the Multiple Receive Queue Control register,
2316 	 * enable VMDq based on packet destination MAC address and RSS.
2317 	 */
2318 	mrqc = E1000_MRQC_ENABLE_VMDQ_MAC_RSS_GROUP;
2319 	mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2320 	    E1000_MRQC_RSS_FIELD_IPV4_TCP |
2321 	    E1000_MRQC_RSS_FIELD_IPV6 |
2322 	    E1000_MRQC_RSS_FIELD_IPV6_TCP |
2323 	    E1000_MRQC_RSS_FIELD_IPV4_UDP |
2324 	    E1000_MRQC_RSS_FIELD_IPV6_UDP |
2325 	    E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2326 	    E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2327 
2328 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2329 
2330 
2331 	/* Define the default group and default queues */
2332 	vmdctl = E1000_VMDQ_MAC_GROUP_DEFAULT_QUEUE;
2333 	E1000_WRITE_REG(hw, E1000_VT_CTL, vmdctl);
2334 
2335 	/*
2336 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2337 	 *
2338 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2339 	 * checksum offloading provided by the 82575 chipset besides the IP
2340 	 * header checksum offloading and the TCP/UDP checksum offloading.
2341 	 * The Packet Checksum is by default computed over the entire packet
2342 	 * from the first byte of the DA through the last byte of the CRC,
2343 	 * including the Ethernet and IP headers.
2344 	 *
2345 	 * It is a hardware limitation that Packet Checksum is mutually
2346 	 * exclusive with RSS.
2347 	 */
2348 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2349 	rxcsum |= E1000_RXCSUM_PCSD;
2350 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2351 }
2352 
2353 /*
2354  * igb_setup_mac_classify - Setup MAC classification feature
2355  */
2356 static void
2357 igb_setup_mac_classify(igb_t *igb)
2358 {
2359 	struct e1000_hw *hw = &igb->hw;
2360 	uint32_t mrqc, rxcsum;
2361 
2362 	/*
2363 	 * Setup the Multiple Receive Queue Control register,
2364 	 * enable VMDq based on packet destination MAC address.
2365 	 */
2366 	mrqc = E1000_MRQC_ENABLE_VMDQ_MAC_GROUP;
2367 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2368 
2369 	/*
2370 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2371 	 *
2372 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2373 	 * checksum offloading provided by the 82575 chipset besides the IP
2374 	 * header checksum offloading and the TCP/UDP checksum offloading.
2375 	 * The Packet Checksum is by default computed over the entire packet
2376 	 * from the first byte of the DA through the last byte of the CRC,
2377 	 * including the Ethernet and IP headers.
2378 	 *
2379 	 * It is a hardware limitation that Packet Checksum is mutually
2380 	 * exclusive with RSS.
2381 	 */
2382 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2383 	rxcsum |= E1000_RXCSUM_PCSD;
2384 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2385 
2386 }
2387 
2388 /*
2389  * igb_init_unicst - Initialize the unicast addresses
2390  */
2391 static void
2392 igb_init_unicst(igb_t *igb)
2393 {
2394 	struct e1000_hw *hw = &igb->hw;
2395 	int slot;
2396 
2397 	/*
2398 	 * Here we should consider two situations:
2399 	 *
2400 	 * 1. Chipset is initialized the first time
2401 	 *    Initialize the multiple unicast addresses, and
2402 	 *    save the default MAC address.
2403 	 *
2404 	 * 2. Chipset is reset
2405 	 *    Recover the multiple unicast addresses from the
2406 	 *    software data structure to the RAR registers.
2407 	 */
2408 
2409 	/*
2410 	 * Clear the default MAC address in the RAR0 rgister,
2411 	 * which is loaded from EEPROM when system boot or chipreset,
2412 	 * this will cause the conficts with add_mac/rem_mac entry
2413 	 * points when VMDq is enabled. For this reason, the RAR0
2414 	 * must be cleared for both cases mentioned above.
2415 	 */
2416 	e1000_rar_clear(hw, 0);
2417 
2418 	if (!igb->unicst_init) {
2419 
2420 		/* Initialize the multiple unicast addresses */
2421 		igb->unicst_total = MAX_NUM_UNICAST_ADDRESSES;
2422 		igb->unicst_avail = igb->unicst_total;
2423 
2424 		for (slot = 0; slot < igb->unicst_total; slot++)
2425 			igb->unicst_addr[slot].mac.set = 0;
2426 
2427 		igb->unicst_init = B_TRUE;
2428 	} else {
2429 		/* Re-configure the RAR registers */
2430 		for (slot = 0; slot < igb->unicst_total; slot++) {
2431 			e1000_rar_set_vmdq(hw, igb->unicst_addr[slot].mac.addr,
2432 			    slot, igb->vmdq_mode,
2433 			    igb->unicst_addr[slot].mac.group_index);
2434 		}
2435 	}
2436 }
2437 
2438 /*
2439  * igb_unicst_find - Find the slot for the specified unicast address
2440  */
2441 int
2442 igb_unicst_find(igb_t *igb, const uint8_t *mac_addr)
2443 {
2444 	int slot;
2445 
2446 	ASSERT(mutex_owned(&igb->gen_lock));
2447 
2448 	for (slot = 0; slot < igb->unicst_total; slot++) {
2449 		if (bcmp(igb->unicst_addr[slot].mac.addr,
2450 		    mac_addr, ETHERADDRL) == 0)
2451 			return (slot);
2452 	}
2453 
2454 	return (-1);
2455 }
2456 
2457 /*
2458  * igb_unicst_set - Set the unicast address to the specified slot
2459  */
2460 int
2461 igb_unicst_set(igb_t *igb, const uint8_t *mac_addr,
2462     int slot)
2463 {
2464 	struct e1000_hw *hw = &igb->hw;
2465 
2466 	ASSERT(mutex_owned(&igb->gen_lock));
2467 
2468 	/*
2469 	 * Save the unicast address in the software data structure
2470 	 */
2471 	bcopy(mac_addr, igb->unicst_addr[slot].mac.addr, ETHERADDRL);
2472 
2473 	/*
2474 	 * Set the unicast address to the RAR register
2475 	 */
2476 	e1000_rar_set(hw, (uint8_t *)mac_addr, slot);
2477 
2478 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2479 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2480 		return (EIO);
2481 	}
2482 
2483 	return (0);
2484 }
2485 
2486 /*
2487  * igb_multicst_add - Add a multicst address
2488  */
2489 int
2490 igb_multicst_add(igb_t *igb, const uint8_t *multiaddr)
2491 {
2492 	struct ether_addr *new_table;
2493 	size_t new_len;
2494 	size_t old_len;
2495 
2496 	ASSERT(mutex_owned(&igb->gen_lock));
2497 
2498 	if ((multiaddr[0] & 01) == 0) {
2499 		igb_error(igb, "Illegal multicast address");
2500 		return (EINVAL);
2501 	}
2502 
2503 	if (igb->mcast_count >= igb->mcast_max_num) {
2504 		igb_error(igb, "Adapter requested more than %d mcast addresses",
2505 		    igb->mcast_max_num);
2506 		return (ENOENT);
2507 	}
2508 
2509 	if (igb->mcast_count == igb->mcast_alloc_count) {
2510 		old_len = igb->mcast_alloc_count *
2511 		    sizeof (struct ether_addr);
2512 		new_len = (igb->mcast_alloc_count + MCAST_ALLOC_COUNT) *
2513 		    sizeof (struct ether_addr);
2514 
2515 		new_table = kmem_alloc(new_len, KM_NOSLEEP);
2516 		if (new_table == NULL) {
2517 			igb_error(igb,
2518 			    "Not enough memory to alloc mcast table");
2519 			return (ENOMEM);
2520 		}
2521 
2522 		if (igb->mcast_table != NULL) {
2523 			bcopy(igb->mcast_table, new_table, old_len);
2524 			kmem_free(igb->mcast_table, old_len);
2525 		}
2526 		igb->mcast_alloc_count += MCAST_ALLOC_COUNT;
2527 		igb->mcast_table = new_table;
2528 	}
2529 
2530 	bcopy(multiaddr,
2531 	    &igb->mcast_table[igb->mcast_count], ETHERADDRL);
2532 	igb->mcast_count++;
2533 
2534 	/*
2535 	 * Update the multicast table in the hardware
2536 	 */
2537 	igb_setup_multicst(igb);
2538 
2539 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2540 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2541 		return (EIO);
2542 	}
2543 
2544 	return (0);
2545 }
2546 
2547 /*
2548  * igb_multicst_remove - Remove a multicst address
2549  */
2550 int
2551 igb_multicst_remove(igb_t *igb, const uint8_t *multiaddr)
2552 {
2553 	struct ether_addr *new_table;
2554 	size_t new_len;
2555 	size_t old_len;
2556 	int i;
2557 
2558 	ASSERT(mutex_owned(&igb->gen_lock));
2559 
2560 	for (i = 0; i < igb->mcast_count; i++) {
2561 		if (bcmp(multiaddr, &igb->mcast_table[i],
2562 		    ETHERADDRL) == 0) {
2563 			for (i++; i < igb->mcast_count; i++) {
2564 				igb->mcast_table[i - 1] =
2565 				    igb->mcast_table[i];
2566 			}
2567 			igb->mcast_count--;
2568 			break;
2569 		}
2570 	}
2571 
2572 	if ((igb->mcast_alloc_count - igb->mcast_count) >
2573 	    MCAST_ALLOC_COUNT) {
2574 		old_len = igb->mcast_alloc_count *
2575 		    sizeof (struct ether_addr);
2576 		new_len = (igb->mcast_alloc_count - MCAST_ALLOC_COUNT) *
2577 		    sizeof (struct ether_addr);
2578 
2579 		new_table = kmem_alloc(new_len, KM_NOSLEEP);
2580 		if (new_table != NULL) {
2581 			bcopy(igb->mcast_table, new_table, new_len);
2582 			kmem_free(igb->mcast_table, old_len);
2583 			igb->mcast_alloc_count -= MCAST_ALLOC_COUNT;
2584 			igb->mcast_table = new_table;
2585 		}
2586 	}
2587 
2588 	/*
2589 	 * Update the multicast table in the hardware
2590 	 */
2591 	igb_setup_multicst(igb);
2592 
2593 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2594 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2595 		return (EIO);
2596 	}
2597 
2598 	return (0);
2599 }
2600 
2601 static void
2602 igb_release_multicast(igb_t *igb)
2603 {
2604 	if (igb->mcast_table != NULL) {
2605 		kmem_free(igb->mcast_table,
2606 		    igb->mcast_alloc_count * sizeof (struct ether_addr));
2607 		igb->mcast_table = NULL;
2608 	}
2609 }
2610 
2611 /*
2612  * igb_setup_multicast - setup multicast data structures
2613  *
2614  * This routine initializes all of the multicast related structures
2615  * and save them in the hardware registers.
2616  */
2617 static void
2618 igb_setup_multicst(igb_t *igb)
2619 {
2620 	uint8_t *mc_addr_list;
2621 	uint32_t mc_addr_count;
2622 	struct e1000_hw *hw = &igb->hw;
2623 
2624 	ASSERT(mutex_owned(&igb->gen_lock));
2625 	ASSERT(igb->mcast_count <= igb->mcast_max_num);
2626 
2627 	mc_addr_list = (uint8_t *)igb->mcast_table;
2628 	mc_addr_count = igb->mcast_count;
2629 
2630 	/*
2631 	 * Update the multicase addresses to the MTA registers
2632 	 */
2633 	e1000_update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
2634 }
2635 
2636 /*
2637  * igb_get_conf - Get driver configurations set in driver.conf
2638  *
2639  * This routine gets user-configured values out of the configuration
2640  * file igb.conf.
2641  *
2642  * For each configurable value, there is a minimum, a maximum, and a
2643  * default.
2644  * If user does not configure a value, use the default.
2645  * If user configures below the minimum, use the minumum.
2646  * If user configures above the maximum, use the maxumum.
2647  */
2648 static void
2649 igb_get_conf(igb_t *igb)
2650 {
2651 	struct e1000_hw *hw = &igb->hw;
2652 	uint32_t default_mtu;
2653 	uint32_t flow_control;
2654 	uint32_t ring_per_group;
2655 	int i;
2656 
2657 	/*
2658 	 * igb driver supports the following user configurations:
2659 	 *
2660 	 * Link configurations:
2661 	 *    adv_autoneg_cap
2662 	 *    adv_1000fdx_cap
2663 	 *    adv_100fdx_cap
2664 	 *    adv_100hdx_cap
2665 	 *    adv_10fdx_cap
2666 	 *    adv_10hdx_cap
2667 	 * Note: 1000hdx is not supported.
2668 	 *
2669 	 * Jumbo frame configuration:
2670 	 *    default_mtu
2671 	 *
2672 	 * Ethernet flow control configuration:
2673 	 *    flow_control
2674 	 *
2675 	 * Multiple rings configurations:
2676 	 *    tx_queue_number
2677 	 *    tx_ring_size
2678 	 *    rx_queue_number
2679 	 *    rx_ring_size
2680 	 *
2681 	 * Call igb_get_prop() to get the value for a specific
2682 	 * configuration parameter.
2683 	 */
2684 
2685 	/*
2686 	 * Link configurations
2687 	 */
2688 	igb->param_adv_autoneg_cap = igb_get_prop(igb,
2689 	    PROP_ADV_AUTONEG_CAP, 0, 1, 1);
2690 	igb->param_adv_1000fdx_cap = igb_get_prop(igb,
2691 	    PROP_ADV_1000FDX_CAP, 0, 1, 1);
2692 	igb->param_adv_100fdx_cap = igb_get_prop(igb,
2693 	    PROP_ADV_100FDX_CAP, 0, 1, 1);
2694 	igb->param_adv_100hdx_cap = igb_get_prop(igb,
2695 	    PROP_ADV_100HDX_CAP, 0, 1, 1);
2696 	igb->param_adv_10fdx_cap = igb_get_prop(igb,
2697 	    PROP_ADV_10FDX_CAP, 0, 1, 1);
2698 	igb->param_adv_10hdx_cap = igb_get_prop(igb,
2699 	    PROP_ADV_10HDX_CAP, 0, 1, 1);
2700 
2701 	/*
2702 	 * Jumbo frame configurations
2703 	 */
2704 	default_mtu = igb_get_prop(igb, PROP_DEFAULT_MTU,
2705 	    MIN_MTU, MAX_MTU, DEFAULT_MTU);
2706 
2707 	igb->max_frame_size = default_mtu +
2708 	    sizeof (struct ether_vlan_header) + ETHERFCSL;
2709 
2710 	/*
2711 	 * Ethernet flow control configuration
2712 	 */
2713 	flow_control = igb_get_prop(igb, PROP_FLOW_CONTROL,
2714 	    e1000_fc_none, 4, e1000_fc_full);
2715 	if (flow_control == 4)
2716 		flow_control = e1000_fc_default;
2717 
2718 	hw->fc.requested_mode = flow_control;
2719 
2720 	/*
2721 	 * Multiple rings configurations
2722 	 */
2723 	igb->tx_ring_size = igb_get_prop(igb, PROP_TX_RING_SIZE,
2724 	    MIN_TX_RING_SIZE, MAX_TX_RING_SIZE, DEFAULT_TX_RING_SIZE);
2725 	igb->rx_ring_size = igb_get_prop(igb, PROP_RX_RING_SIZE,
2726 	    MIN_RX_RING_SIZE, MAX_RX_RING_SIZE, DEFAULT_RX_RING_SIZE);
2727 
2728 	igb->mr_enable = igb_get_prop(igb, PROP_MR_ENABLE, 0, 1, 0);
2729 	igb->num_rx_groups = igb_get_prop(igb, PROP_RX_GROUP_NUM,
2730 	    MIN_RX_GROUP_NUM, MAX_RX_GROUP_NUM, DEFAULT_RX_GROUP_NUM);
2731 	/*
2732 	 * Currently we do not support VMDq for 82576 and 82580.
2733 	 * If it is e1000_82576, set num_rx_groups to 1.
2734 	 */
2735 	if (hw->mac.type >= e1000_82576)
2736 		igb->num_rx_groups = 1;
2737 
2738 	if (igb->mr_enable) {
2739 		igb->num_tx_rings = igb->capab->def_tx_que_num;
2740 		igb->num_rx_rings = igb->capab->def_rx_que_num;
2741 	} else {
2742 		igb->num_tx_rings = 1;
2743 		igb->num_rx_rings = 1;
2744 
2745 		if (igb->num_rx_groups > 1) {
2746 			igb_error(igb,
2747 			    "Invalid rx groups number. Please enable multiple "
2748 			    "rings first");
2749 			igb->num_rx_groups = 1;
2750 		}
2751 	}
2752 
2753 	/*
2754 	 * Check the divisibility between rx rings and rx groups.
2755 	 */
2756 	for (i = igb->num_rx_groups; i > 0; i--) {
2757 		if ((igb->num_rx_rings % i) == 0)
2758 			break;
2759 	}
2760 	if (i != igb->num_rx_groups) {
2761 		igb_error(igb,
2762 		    "Invalid rx groups number. Downgrade the rx group "
2763 		    "number to %d.", i);
2764 		igb->num_rx_groups = i;
2765 	}
2766 
2767 	/*
2768 	 * Get the ring number per group.
2769 	 */
2770 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
2771 
2772 	if (igb->num_rx_groups == 1) {
2773 		/*
2774 		 * One rx ring group, the rx ring number is num_rx_rings.
2775 		 */
2776 		igb->vmdq_mode = E1000_VMDQ_OFF;
2777 	} else if (ring_per_group == 1) {
2778 		/*
2779 		 * Multiple rx groups, each group has one rx ring.
2780 		 */
2781 		igb->vmdq_mode = E1000_VMDQ_MAC;
2782 	} else {
2783 		/*
2784 		 * Multiple groups and multiple rings.
2785 		 */
2786 		igb->vmdq_mode = E1000_VMDQ_MAC_RSS;
2787 	}
2788 
2789 	/*
2790 	 * Tunable used to force an interrupt type. The only use is
2791 	 * for testing of the lesser interrupt types.
2792 	 * 0 = don't force interrupt type
2793 	 * 1 = force interrupt type MSIX
2794 	 * 2 = force interrupt type MSI
2795 	 * 3 = force interrupt type Legacy
2796 	 */
2797 	igb->intr_force = igb_get_prop(igb, PROP_INTR_FORCE,
2798 	    IGB_INTR_NONE, IGB_INTR_LEGACY, IGB_INTR_NONE);
2799 
2800 	igb->tx_hcksum_enable = igb_get_prop(igb, PROP_TX_HCKSUM_ENABLE,
2801 	    0, 1, 1);
2802 	igb->rx_hcksum_enable = igb_get_prop(igb, PROP_RX_HCKSUM_ENABLE,
2803 	    0, 1, 1);
2804 	igb->lso_enable = igb_get_prop(igb, PROP_LSO_ENABLE,
2805 	    0, 1, 1);
2806 	igb->tx_head_wb_enable = igb_get_prop(igb, PROP_TX_HEAD_WB_ENABLE,
2807 	    0, 1, 1);
2808 
2809 	/*
2810 	 * igb LSO needs the tx h/w checksum support.
2811 	 * Here LSO will be disabled if tx h/w checksum has been disabled.
2812 	 */
2813 	if (igb->tx_hcksum_enable == B_FALSE)
2814 		igb->lso_enable = B_FALSE;
2815 
2816 	igb->tx_copy_thresh = igb_get_prop(igb, PROP_TX_COPY_THRESHOLD,
2817 	    MIN_TX_COPY_THRESHOLD, MAX_TX_COPY_THRESHOLD,
2818 	    DEFAULT_TX_COPY_THRESHOLD);
2819 	igb->tx_recycle_thresh = igb_get_prop(igb, PROP_TX_RECYCLE_THRESHOLD,
2820 	    MIN_TX_RECYCLE_THRESHOLD, MAX_TX_RECYCLE_THRESHOLD,
2821 	    DEFAULT_TX_RECYCLE_THRESHOLD);
2822 	igb->tx_overload_thresh = igb_get_prop(igb, PROP_TX_OVERLOAD_THRESHOLD,
2823 	    MIN_TX_OVERLOAD_THRESHOLD, MAX_TX_OVERLOAD_THRESHOLD,
2824 	    DEFAULT_TX_OVERLOAD_THRESHOLD);
2825 	igb->tx_resched_thresh = igb_get_prop(igb, PROP_TX_RESCHED_THRESHOLD,
2826 	    MIN_TX_RESCHED_THRESHOLD, MAX_TX_RESCHED_THRESHOLD,
2827 	    DEFAULT_TX_RESCHED_THRESHOLD);
2828 
2829 	igb->rx_copy_thresh = igb_get_prop(igb, PROP_RX_COPY_THRESHOLD,
2830 	    MIN_RX_COPY_THRESHOLD, MAX_RX_COPY_THRESHOLD,
2831 	    DEFAULT_RX_COPY_THRESHOLD);
2832 	igb->rx_limit_per_intr = igb_get_prop(igb, PROP_RX_LIMIT_PER_INTR,
2833 	    MIN_RX_LIMIT_PER_INTR, MAX_RX_LIMIT_PER_INTR,
2834 	    DEFAULT_RX_LIMIT_PER_INTR);
2835 
2836 	igb->intr_throttling[0] = igb_get_prop(igb, PROP_INTR_THROTTLING,
2837 	    igb->capab->min_intr_throttle,
2838 	    igb->capab->max_intr_throttle,
2839 	    igb->capab->def_intr_throttle);
2840 
2841 	/*
2842 	 * Max number of multicast addresses
2843 	 */
2844 	igb->mcast_max_num =
2845 	    igb_get_prop(igb, PROP_MCAST_MAX_NUM,
2846 	    MIN_MCAST_NUM, MAX_MCAST_NUM, DEFAULT_MCAST_NUM);
2847 }
2848 
2849 /*
2850  * igb_get_prop - Get a property value out of the configuration file igb.conf
2851  *
2852  * Caller provides the name of the property, a default value, a minimum
2853  * value, and a maximum value.
2854  *
2855  * Return configured value of the property, with default, minimum and
2856  * maximum properly applied.
2857  */
2858 static int
2859 igb_get_prop(igb_t *igb,
2860     char *propname,	/* name of the property */
2861     int minval,		/* minimum acceptable value */
2862     int maxval,		/* maximim acceptable value */
2863     int defval)		/* default value */
2864 {
2865 	int value;
2866 
2867 	/*
2868 	 * Call ddi_prop_get_int() to read the conf settings
2869 	 */
2870 	value = ddi_prop_get_int(DDI_DEV_T_ANY, igb->dip,
2871 	    DDI_PROP_DONTPASS, propname, defval);
2872 
2873 	if (value > maxval)
2874 		value = maxval;
2875 
2876 	if (value < minval)
2877 		value = minval;
2878 
2879 	return (value);
2880 }
2881 
2882 /*
2883  * igb_setup_link - Using the link properties to setup the link
2884  */
2885 int
2886 igb_setup_link(igb_t *igb, boolean_t setup_hw)
2887 {
2888 	struct e1000_mac_info *mac;
2889 	struct e1000_phy_info *phy;
2890 	boolean_t invalid;
2891 
2892 	mac = &igb->hw.mac;
2893 	phy = &igb->hw.phy;
2894 	invalid = B_FALSE;
2895 
2896 	if (igb->param_adv_autoneg_cap == 1) {
2897 		mac->autoneg = B_TRUE;
2898 		phy->autoneg_advertised = 0;
2899 
2900 		/*
2901 		 * 1000hdx is not supported for autonegotiation
2902 		 */
2903 		if (igb->param_adv_1000fdx_cap == 1)
2904 			phy->autoneg_advertised |= ADVERTISE_1000_FULL;
2905 
2906 		if (igb->param_adv_100fdx_cap == 1)
2907 			phy->autoneg_advertised |= ADVERTISE_100_FULL;
2908 
2909 		if (igb->param_adv_100hdx_cap == 1)
2910 			phy->autoneg_advertised |= ADVERTISE_100_HALF;
2911 
2912 		if (igb->param_adv_10fdx_cap == 1)
2913 			phy->autoneg_advertised |= ADVERTISE_10_FULL;
2914 
2915 		if (igb->param_adv_10hdx_cap == 1)
2916 			phy->autoneg_advertised |= ADVERTISE_10_HALF;
2917 
2918 		if (phy->autoneg_advertised == 0)
2919 			invalid = B_TRUE;
2920 	} else {
2921 		mac->autoneg = B_FALSE;
2922 
2923 		/*
2924 		 * 1000fdx and 1000hdx are not supported for forced link
2925 		 */
2926 		if (igb->param_adv_100fdx_cap == 1)
2927 			mac->forced_speed_duplex = ADVERTISE_100_FULL;
2928 		else if (igb->param_adv_100hdx_cap == 1)
2929 			mac->forced_speed_duplex = ADVERTISE_100_HALF;
2930 		else if (igb->param_adv_10fdx_cap == 1)
2931 			mac->forced_speed_duplex = ADVERTISE_10_FULL;
2932 		else if (igb->param_adv_10hdx_cap == 1)
2933 			mac->forced_speed_duplex = ADVERTISE_10_HALF;
2934 		else
2935 			invalid = B_TRUE;
2936 	}
2937 
2938 	if (invalid) {
2939 		igb_notice(igb, "Invalid link settings. Setup link to "
2940 		    "autonegotiation with full link capabilities.");
2941 		mac->autoneg = B_TRUE;
2942 		phy->autoneg_advertised = ADVERTISE_1000_FULL |
2943 		    ADVERTISE_100_FULL | ADVERTISE_100_HALF |
2944 		    ADVERTISE_10_FULL | ADVERTISE_10_HALF;
2945 	}
2946 
2947 	if (setup_hw) {
2948 		if (e1000_setup_link(&igb->hw) != E1000_SUCCESS)
2949 			return (IGB_FAILURE);
2950 	}
2951 
2952 	return (IGB_SUCCESS);
2953 }
2954 
2955 
2956 /*
2957  * igb_is_link_up - Check if the link is up
2958  */
2959 static boolean_t
2960 igb_is_link_up(igb_t *igb)
2961 {
2962 	struct e1000_hw *hw = &igb->hw;
2963 	boolean_t link_up = B_FALSE;
2964 
2965 	ASSERT(mutex_owned(&igb->gen_lock));
2966 
2967 	/*
2968 	 * get_link_status is set in the interrupt handler on link-status-change
2969 	 * or rx sequence error interrupt.  get_link_status will stay
2970 	 * false until the e1000_check_for_link establishes link only
2971 	 * for copper adapters.
2972 	 */
2973 	switch (hw->phy.media_type) {
2974 	case e1000_media_type_copper:
2975 		if (hw->mac.get_link_status) {
2976 			(void) e1000_check_for_link(hw);
2977 			link_up = !hw->mac.get_link_status;
2978 		} else {
2979 			link_up = B_TRUE;
2980 		}
2981 		break;
2982 	case e1000_media_type_fiber:
2983 		(void) e1000_check_for_link(hw);
2984 		link_up = (E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU);
2985 		break;
2986 	case e1000_media_type_internal_serdes:
2987 		(void) e1000_check_for_link(hw);
2988 		link_up = hw->mac.serdes_has_link;
2989 		break;
2990 	}
2991 
2992 	return (link_up);
2993 }
2994 
2995 /*
2996  * igb_link_check - Link status processing
2997  */
2998 static boolean_t
2999 igb_link_check(igb_t *igb)
3000 {
3001 	struct e1000_hw *hw = &igb->hw;
3002 	uint16_t speed = 0, duplex = 0;
3003 	boolean_t link_changed = B_FALSE;
3004 
3005 	ASSERT(mutex_owned(&igb->gen_lock));
3006 
3007 	if (igb_is_link_up(igb)) {
3008 		/*
3009 		 * The Link is up, check whether it was marked as down earlier
3010 		 */
3011 		if (igb->link_state != LINK_STATE_UP) {
3012 			(void) e1000_get_speed_and_duplex(hw, &speed, &duplex);
3013 			igb->link_speed = speed;
3014 			igb->link_duplex = duplex;
3015 			igb->link_state = LINK_STATE_UP;
3016 			igb->link_down_timeout = 0;
3017 			link_changed = B_TRUE;
3018 			if (!igb->link_complete)
3019 				igb_stop_link_timer(igb);
3020 		}
3021 	} else if (igb->link_complete) {
3022 		if (igb->link_state != LINK_STATE_DOWN) {
3023 			igb->link_speed = 0;
3024 			igb->link_duplex = 0;
3025 			igb->link_state = LINK_STATE_DOWN;
3026 			link_changed = B_TRUE;
3027 		}
3028 
3029 		if (igb->igb_state & IGB_STARTED) {
3030 			if (igb->link_down_timeout < MAX_LINK_DOWN_TIMEOUT) {
3031 				igb->link_down_timeout++;
3032 			} else if (igb->link_down_timeout ==
3033 			    MAX_LINK_DOWN_TIMEOUT) {
3034 				igb_tx_clean(igb);
3035 				igb->link_down_timeout++;
3036 			}
3037 		}
3038 	}
3039 
3040 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
3041 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
3042 		return (B_FALSE);
3043 	}
3044 
3045 	return (link_changed);
3046 }
3047 
3048 /*
3049  * igb_local_timer - driver watchdog function
3050  *
3051  * This function will handle the hardware stall check, link status
3052  * check and other routines.
3053  */
3054 static void
3055 igb_local_timer(void *arg)
3056 {
3057 	igb_t *igb = (igb_t *)arg;
3058 	boolean_t link_changed = B_FALSE;
3059 
3060 	if (igb->igb_state & IGB_ERROR) {
3061 		igb->reset_count++;
3062 		if (igb_reset(igb) == IGB_SUCCESS)
3063 			ddi_fm_service_impact(igb->dip, DDI_SERVICE_RESTORED);
3064 
3065 		igb_restart_watchdog_timer(igb);
3066 		return;
3067 	}
3068 
3069 	if (igb_stall_check(igb) || (igb->igb_state & IGB_STALL)) {
3070 		igb_fm_ereport(igb, DDI_FM_DEVICE_STALL);
3071 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
3072 		igb->reset_count++;
3073 		if (igb_reset(igb) == IGB_SUCCESS)
3074 			ddi_fm_service_impact(igb->dip, DDI_SERVICE_RESTORED);
3075 
3076 		igb_restart_watchdog_timer(igb);
3077 		return;
3078 	}
3079 
3080 	mutex_enter(&igb->gen_lock);
3081 	if (!(igb->igb_state & IGB_SUSPENDED) && (igb->igb_state & IGB_STARTED))
3082 		link_changed = igb_link_check(igb);
3083 	mutex_exit(&igb->gen_lock);
3084 
3085 	if (link_changed)
3086 		mac_link_update(igb->mac_hdl, igb->link_state);
3087 
3088 	igb_restart_watchdog_timer(igb);
3089 }
3090 
3091 /*
3092  * igb_link_timer - link setup timer function
3093  *
3094  * It is called when the timer for link setup is expired, which indicates
3095  * the completion of the link setup. The link state will not be updated
3096  * until the link setup is completed. And the link state will not be sent
3097  * to the upper layer through mac_link_update() in this function. It will
3098  * be updated in the local timer routine or the interrupts service routine
3099  * after the interface is started (plumbed).
3100  */
3101 static void
3102 igb_link_timer(void *arg)
3103 {
3104 	igb_t *igb = (igb_t *)arg;
3105 
3106 	mutex_enter(&igb->link_lock);
3107 	igb->link_complete = B_TRUE;
3108 	igb->link_tid = 0;
3109 	mutex_exit(&igb->link_lock);
3110 }
3111 /*
3112  * igb_stall_check - check for transmit stall
3113  *
3114  * This function checks if the adapter is stalled (in transmit).
3115  *
3116  * It is called each time the watchdog timeout is invoked.
3117  * If the transmit descriptor reclaim continuously fails,
3118  * the watchdog value will increment by 1. If the watchdog
3119  * value exceeds the threshold, the igb is assumed to
3120  * have stalled and need to be reset.
3121  */
3122 static boolean_t
3123 igb_stall_check(igb_t *igb)
3124 {
3125 	igb_tx_ring_t *tx_ring;
3126 	struct e1000_hw *hw = &igb->hw;
3127 	boolean_t result;
3128 	int i;
3129 
3130 	if (igb->link_state != LINK_STATE_UP)
3131 		return (B_FALSE);
3132 
3133 	/*
3134 	 * If any tx ring is stalled, we'll reset the chipset
3135 	 */
3136 	result = B_FALSE;
3137 	for (i = 0; i < igb->num_tx_rings; i++) {
3138 		tx_ring = &igb->tx_rings[i];
3139 
3140 		if (tx_ring->recycle_fail > 0)
3141 			tx_ring->stall_watchdog++;
3142 		else
3143 			tx_ring->stall_watchdog = 0;
3144 
3145 		if (tx_ring->stall_watchdog >= STALL_WATCHDOG_TIMEOUT) {
3146 			result = B_TRUE;
3147 			if (hw->mac.type == e1000_82580) {
3148 				hw->dev_spec._82575.global_device_reset
3149 				    = B_TRUE;
3150 			}
3151 			break;
3152 		}
3153 	}
3154 
3155 	if (result) {
3156 		tx_ring->stall_watchdog = 0;
3157 		tx_ring->recycle_fail = 0;
3158 	}
3159 
3160 	return (result);
3161 }
3162 
3163 
3164 /*
3165  * is_valid_mac_addr - Check if the mac address is valid
3166  */
3167 static boolean_t
3168 is_valid_mac_addr(uint8_t *mac_addr)
3169 {
3170 	const uint8_t addr_test1[6] = { 0, 0, 0, 0, 0, 0 };
3171 	const uint8_t addr_test2[6] =
3172 	    { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF };
3173 
3174 	if (!(bcmp(addr_test1, mac_addr, ETHERADDRL)) ||
3175 	    !(bcmp(addr_test2, mac_addr, ETHERADDRL)))
3176 		return (B_FALSE);
3177 
3178 	return (B_TRUE);
3179 }
3180 
3181 static boolean_t
3182 igb_find_mac_address(igb_t *igb)
3183 {
3184 	struct e1000_hw *hw = &igb->hw;
3185 #ifdef __sparc
3186 	uchar_t *bytes;
3187 	struct ether_addr sysaddr;
3188 	uint_t nelts;
3189 	int err;
3190 	boolean_t found = B_FALSE;
3191 
3192 	/*
3193 	 * The "vendor's factory-set address" may already have
3194 	 * been extracted from the chip, but if the property
3195 	 * "local-mac-address" is set we use that instead.
3196 	 *
3197 	 * We check whether it looks like an array of 6
3198 	 * bytes (which it should, if OBP set it).  If we can't
3199 	 * make sense of it this way, we'll ignore it.
3200 	 */
3201 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip,
3202 	    DDI_PROP_DONTPASS, "local-mac-address", &bytes, &nelts);
3203 	if (err == DDI_PROP_SUCCESS) {
3204 		if (nelts == ETHERADDRL) {
3205 			while (nelts--)
3206 				hw->mac.addr[nelts] = bytes[nelts];
3207 			found = B_TRUE;
3208 		}
3209 		ddi_prop_free(bytes);
3210 	}
3211 
3212 	/*
3213 	 * Look up the OBP property "local-mac-address?". If the user has set
3214 	 * 'local-mac-address? = false', use "the system address" instead.
3215 	 */
3216 	if (ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip, 0,
3217 	    "local-mac-address?", &bytes, &nelts) == DDI_PROP_SUCCESS) {
3218 		if (strncmp("false", (caddr_t)bytes, (size_t)nelts) == 0) {
3219 			if (localetheraddr(NULL, &sysaddr) != 0) {
3220 				bcopy(&sysaddr, hw->mac.addr, ETHERADDRL);
3221 				found = B_TRUE;
3222 			}
3223 		}
3224 		ddi_prop_free(bytes);
3225 	}
3226 
3227 	/*
3228 	 * Finally(!), if there's a valid "mac-address" property (created
3229 	 * if we netbooted from this interface), we must use this instead
3230 	 * of any of the above to ensure that the NFS/install server doesn't
3231 	 * get confused by the address changing as Solaris takes over!
3232 	 */
3233 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip,
3234 	    DDI_PROP_DONTPASS, "mac-address", &bytes, &nelts);
3235 	if (err == DDI_PROP_SUCCESS) {
3236 		if (nelts == ETHERADDRL) {
3237 			while (nelts--)
3238 				hw->mac.addr[nelts] = bytes[nelts];
3239 			found = B_TRUE;
3240 		}
3241 		ddi_prop_free(bytes);
3242 	}
3243 
3244 	if (found) {
3245 		bcopy(hw->mac.addr, hw->mac.perm_addr, ETHERADDRL);
3246 		return (B_TRUE);
3247 	}
3248 #endif
3249 
3250 	/*
3251 	 * Read the device MAC address from the EEPROM
3252 	 */
3253 	if (e1000_read_mac_addr(hw) != E1000_SUCCESS)
3254 		return (B_FALSE);
3255 
3256 	return (B_TRUE);
3257 }
3258 
3259 #pragma inline(igb_arm_watchdog_timer)
3260 
3261 static void
3262 igb_arm_watchdog_timer(igb_t *igb)
3263 {
3264 	/*
3265 	 * Fire a watchdog timer
3266 	 */
3267 	igb->watchdog_tid =
3268 	    timeout(igb_local_timer,
3269 	    (void *)igb, 1 * drv_usectohz(1000000));
3270 
3271 }
3272 
3273 /*
3274  * igb_enable_watchdog_timer - Enable and start the driver watchdog timer
3275  */
3276 void
3277 igb_enable_watchdog_timer(igb_t *igb)
3278 {
3279 	mutex_enter(&igb->watchdog_lock);
3280 
3281 	if (!igb->watchdog_enable) {
3282 		igb->watchdog_enable = B_TRUE;
3283 		igb->watchdog_start = B_TRUE;
3284 		igb_arm_watchdog_timer(igb);
3285 	}
3286 
3287 	mutex_exit(&igb->watchdog_lock);
3288 
3289 }
3290 
3291 /*
3292  * igb_disable_watchdog_timer - Disable and stop the driver watchdog timer
3293  */
3294 void
3295 igb_disable_watchdog_timer(igb_t *igb)
3296 {
3297 	timeout_id_t tid;
3298 
3299 	mutex_enter(&igb->watchdog_lock);
3300 
3301 	igb->watchdog_enable = B_FALSE;
3302 	igb->watchdog_start = B_FALSE;
3303 	tid = igb->watchdog_tid;
3304 	igb->watchdog_tid = 0;
3305 
3306 	mutex_exit(&igb->watchdog_lock);
3307 
3308 	if (tid != 0)
3309 		(void) untimeout(tid);
3310 
3311 }
3312 
3313 /*
3314  * igb_start_watchdog_timer - Start the driver watchdog timer
3315  */
3316 static void
3317 igb_start_watchdog_timer(igb_t *igb)
3318 {
3319 	mutex_enter(&igb->watchdog_lock);
3320 
3321 	if (igb->watchdog_enable) {
3322 		if (!igb->watchdog_start) {
3323 			igb->watchdog_start = B_TRUE;
3324 			igb_arm_watchdog_timer(igb);
3325 		}
3326 	}
3327 
3328 	mutex_exit(&igb->watchdog_lock);
3329 }
3330 
3331 /*
3332  * igb_restart_watchdog_timer - Restart the driver watchdog timer
3333  */
3334 static void
3335 igb_restart_watchdog_timer(igb_t *igb)
3336 {
3337 	mutex_enter(&igb->watchdog_lock);
3338 
3339 	if (igb->watchdog_start)
3340 		igb_arm_watchdog_timer(igb);
3341 
3342 	mutex_exit(&igb->watchdog_lock);
3343 }
3344 
3345 /*
3346  * igb_stop_watchdog_timer - Stop the driver watchdog timer
3347  */
3348 static void
3349 igb_stop_watchdog_timer(igb_t *igb)
3350 {
3351 	timeout_id_t tid;
3352 
3353 	mutex_enter(&igb->watchdog_lock);
3354 
3355 	igb->watchdog_start = B_FALSE;
3356 	tid = igb->watchdog_tid;
3357 	igb->watchdog_tid = 0;
3358 
3359 	mutex_exit(&igb->watchdog_lock);
3360 
3361 	if (tid != 0)
3362 		(void) untimeout(tid);
3363 }
3364 
3365 /*
3366  * igb_start_link_timer - Start the link setup timer
3367  */
3368 static void
3369 igb_start_link_timer(struct igb *igb)
3370 {
3371 	struct e1000_hw *hw = &igb->hw;
3372 	clock_t link_timeout;
3373 
3374 	if (hw->mac.autoneg)
3375 		link_timeout = PHY_AUTO_NEG_LIMIT *
3376 		    drv_usectohz(100000);
3377 	else
3378 		link_timeout = PHY_FORCE_LIMIT * drv_usectohz(100000);
3379 
3380 	mutex_enter(&igb->link_lock);
3381 	if (hw->phy.autoneg_wait_to_complete) {
3382 		igb->link_complete = B_TRUE;
3383 	} else {
3384 		igb->link_complete = B_FALSE;
3385 		igb->link_tid = timeout(igb_link_timer, (void *)igb,
3386 		    link_timeout);
3387 	}
3388 	mutex_exit(&igb->link_lock);
3389 }
3390 
3391 /*
3392  * igb_stop_link_timer - Stop the link setup timer
3393  */
3394 static void
3395 igb_stop_link_timer(struct igb *igb)
3396 {
3397 	timeout_id_t tid;
3398 
3399 	mutex_enter(&igb->link_lock);
3400 	igb->link_complete = B_TRUE;
3401 	tid = igb->link_tid;
3402 	igb->link_tid = 0;
3403 	mutex_exit(&igb->link_lock);
3404 
3405 	if (tid != 0)
3406 		(void) untimeout(tid);
3407 }
3408 
3409 /*
3410  * igb_disable_adapter_interrupts - Clear/disable all hardware interrupts
3411  */
3412 static void
3413 igb_disable_adapter_interrupts(igb_t *igb)
3414 {
3415 	struct e1000_hw *hw = &igb->hw;
3416 
3417 	/*
3418 	 * Set the IMC register to mask all the interrupts,
3419 	 * including the tx interrupts.
3420 	 */
3421 	E1000_WRITE_REG(hw, E1000_IMC, ~0);
3422 	E1000_WRITE_REG(hw, E1000_IAM, 0);
3423 
3424 	/*
3425 	 * Additional disabling for MSI-X
3426 	 */
3427 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3428 		E1000_WRITE_REG(hw, E1000_EIMC, ~0);
3429 		E1000_WRITE_REG(hw, E1000_EIAC, 0);
3430 		E1000_WRITE_REG(hw, E1000_EIAM, 0);
3431 	}
3432 
3433 	E1000_WRITE_FLUSH(hw);
3434 }
3435 
3436 /*
3437  * igb_enable_adapter_interrupts_82580 - Enable NIC interrupts for 82580
3438  */
3439 static void
3440 igb_enable_adapter_interrupts_82580(igb_t *igb)
3441 {
3442 	struct e1000_hw *hw = &igb->hw;
3443 
3444 	/* Clear any pending interrupts */
3445 	(void) E1000_READ_REG(hw, E1000_ICR);
3446 	igb->ims_mask |= E1000_IMS_DRSTA;
3447 
3448 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3449 
3450 		/* Interrupt enabling for MSI-X */
3451 		E1000_WRITE_REG(hw, E1000_EIMS, igb->eims_mask);
3452 		E1000_WRITE_REG(hw, E1000_EIAC, igb->eims_mask);
3453 		igb->ims_mask = (E1000_IMS_LSC | E1000_IMS_DRSTA);
3454 		E1000_WRITE_REG(hw, E1000_IMS, igb->ims_mask);
3455 	} else { /* Interrupt enabling for MSI and legacy */
3456 		E1000_WRITE_REG(hw, E1000_IVAR0, E1000_IVAR_VALID);
3457 		igb->ims_mask = IMS_ENABLE_MASK | E1000_IMS_TXQE;
3458 		igb->ims_mask |= E1000_IMS_DRSTA;
3459 		E1000_WRITE_REG(hw, E1000_IMS, igb->ims_mask);
3460 	}
3461 
3462 	/* Disable auto-mask for ICR interrupt bits */
3463 	E1000_WRITE_REG(hw, E1000_IAM, 0);
3464 
3465 	E1000_WRITE_FLUSH(hw);
3466 }
3467 
3468 /*
3469  * igb_enable_adapter_interrupts_82576 - Enable NIC interrupts for 82576
3470  */
3471 static void
3472 igb_enable_adapter_interrupts_82576(igb_t *igb)
3473 {
3474 	struct e1000_hw *hw = &igb->hw;
3475 
3476 	/* Clear any pending interrupts */
3477 	(void) E1000_READ_REG(hw, E1000_ICR);
3478 
3479 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3480 
3481 		/* Interrupt enabling for MSI-X */
3482 		E1000_WRITE_REG(hw, E1000_EIMS, igb->eims_mask);
3483 		E1000_WRITE_REG(hw, E1000_EIAC, igb->eims_mask);
3484 		igb->ims_mask = E1000_IMS_LSC;
3485 		E1000_WRITE_REG(hw, E1000_IMS, E1000_IMS_LSC);
3486 	} else {
3487 		/* Interrupt enabling for MSI and legacy */
3488 		E1000_WRITE_REG(hw, E1000_IVAR0, E1000_IVAR_VALID);
3489 		igb->ims_mask = IMS_ENABLE_MASK | E1000_IMS_TXQE;
3490 		E1000_WRITE_REG(hw, E1000_IMS,
3491 		    (IMS_ENABLE_MASK | E1000_IMS_TXQE));
3492 	}
3493 
3494 	/* Disable auto-mask for ICR interrupt bits */
3495 	E1000_WRITE_REG(hw, E1000_IAM, 0);
3496 
3497 	E1000_WRITE_FLUSH(hw);
3498 }
3499 
3500 /*
3501  * igb_enable_adapter_interrupts_82575 - Enable NIC interrupts for 82575
3502  */
3503 static void
3504 igb_enable_adapter_interrupts_82575(igb_t *igb)
3505 {
3506 	struct e1000_hw *hw = &igb->hw;
3507 	uint32_t reg;
3508 
3509 	/* Clear any pending interrupts */
3510 	(void) E1000_READ_REG(hw, E1000_ICR);
3511 
3512 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3513 		/* Interrupt enabling for MSI-X */
3514 		E1000_WRITE_REG(hw, E1000_EIMS, igb->eims_mask);
3515 		E1000_WRITE_REG(hw, E1000_EIAC, igb->eims_mask);
3516 		igb->ims_mask = E1000_IMS_LSC;
3517 		E1000_WRITE_REG(hw, E1000_IMS, E1000_IMS_LSC);
3518 
3519 		/* Enable MSI-X PBA support */
3520 		reg = E1000_READ_REG(hw, E1000_CTRL_EXT);
3521 		reg |= E1000_CTRL_EXT_PBA_CLR;
3522 
3523 		/* Non-selective interrupt clear-on-read */
3524 		reg |= E1000_CTRL_EXT_IRCA;	/* Called NSICR in the EAS */
3525 
3526 		E1000_WRITE_REG(hw, E1000_CTRL_EXT, reg);
3527 	} else {
3528 		/* Interrupt enabling for MSI and legacy */
3529 		igb->ims_mask = IMS_ENABLE_MASK;
3530 		E1000_WRITE_REG(hw, E1000_IMS, IMS_ENABLE_MASK);
3531 	}
3532 
3533 	E1000_WRITE_FLUSH(hw);
3534 }
3535 
3536 /*
3537  * Loopback Support
3538  */
3539 static lb_property_t lb_normal =
3540 	{ normal,	"normal",	IGB_LB_NONE		};
3541 static lb_property_t lb_external =
3542 	{ external,	"External",	IGB_LB_EXTERNAL		};
3543 static lb_property_t lb_phy =
3544 	{ internal,	"PHY",		IGB_LB_INTERNAL_PHY	};
3545 static lb_property_t lb_serdes =
3546 	{ internal,	"SerDes",	IGB_LB_INTERNAL_SERDES	};
3547 
3548 enum ioc_reply
3549 igb_loopback_ioctl(igb_t *igb, struct iocblk *iocp, mblk_t *mp)
3550 {
3551 	lb_info_sz_t *lbsp;
3552 	lb_property_t *lbpp;
3553 	struct e1000_hw *hw;
3554 	uint32_t *lbmp;
3555 	uint32_t size;
3556 	uint32_t value;
3557 
3558 	hw = &igb->hw;
3559 
3560 	if (mp->b_cont == NULL)
3561 		return (IOC_INVAL);
3562 
3563 	switch (iocp->ioc_cmd) {
3564 	default:
3565 		return (IOC_INVAL);
3566 
3567 	case LB_GET_INFO_SIZE:
3568 		size = sizeof (lb_info_sz_t);
3569 		if (iocp->ioc_count != size)
3570 			return (IOC_INVAL);
3571 
3572 		value = sizeof (lb_normal);
3573 		if (hw->phy.media_type == e1000_media_type_copper)
3574 			value += sizeof (lb_phy);
3575 		else
3576 			value += sizeof (lb_serdes);
3577 		value += sizeof (lb_external);
3578 
3579 		lbsp = (lb_info_sz_t *)(uintptr_t)mp->b_cont->b_rptr;
3580 		*lbsp = value;
3581 		break;
3582 
3583 	case LB_GET_INFO:
3584 		value = sizeof (lb_normal);
3585 		if (hw->phy.media_type == e1000_media_type_copper)
3586 			value += sizeof (lb_phy);
3587 		else
3588 			value += sizeof (lb_serdes);
3589 		value += sizeof (lb_external);
3590 
3591 		size = value;
3592 		if (iocp->ioc_count != size)
3593 			return (IOC_INVAL);
3594 
3595 		value = 0;
3596 		lbpp = (lb_property_t *)(uintptr_t)mp->b_cont->b_rptr;
3597 
3598 		lbpp[value++] = lb_normal;
3599 		if (hw->phy.media_type == e1000_media_type_copper)
3600 			lbpp[value++] = lb_phy;
3601 		else
3602 			lbpp[value++] = lb_serdes;
3603 		lbpp[value++] = lb_external;
3604 		break;
3605 
3606 	case LB_GET_MODE:
3607 		size = sizeof (uint32_t);
3608 		if (iocp->ioc_count != size)
3609 			return (IOC_INVAL);
3610 
3611 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
3612 		*lbmp = igb->loopback_mode;
3613 		break;
3614 
3615 	case LB_SET_MODE:
3616 		size = 0;
3617 		if (iocp->ioc_count != sizeof (uint32_t))
3618 			return (IOC_INVAL);
3619 
3620 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
3621 		if (!igb_set_loopback_mode(igb, *lbmp))
3622 			return (IOC_INVAL);
3623 		break;
3624 	}
3625 
3626 	iocp->ioc_count = size;
3627 	iocp->ioc_error = 0;
3628 
3629 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
3630 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
3631 		return (IOC_INVAL);
3632 	}
3633 
3634 	return (IOC_REPLY);
3635 }
3636 
3637 /*
3638  * igb_set_loopback_mode - Setup loopback based on the loopback mode
3639  */
3640 static boolean_t
3641 igb_set_loopback_mode(igb_t *igb, uint32_t mode)
3642 {
3643 	struct e1000_hw *hw;
3644 	int i;
3645 
3646 	if (mode == igb->loopback_mode)
3647 		return (B_TRUE);
3648 
3649 	hw = &igb->hw;
3650 
3651 	igb->loopback_mode = mode;
3652 
3653 	if (mode == IGB_LB_NONE) {
3654 		/* Reset the chip */
3655 		hw->phy.autoneg_wait_to_complete = B_TRUE;
3656 		(void) igb_reset(igb);
3657 		hw->phy.autoneg_wait_to_complete = B_FALSE;
3658 		return (B_TRUE);
3659 	}
3660 
3661 	mutex_enter(&igb->gen_lock);
3662 
3663 	switch (mode) {
3664 	default:
3665 		mutex_exit(&igb->gen_lock);
3666 		return (B_FALSE);
3667 
3668 	case IGB_LB_EXTERNAL:
3669 		igb_set_external_loopback(igb);
3670 		break;
3671 
3672 	case IGB_LB_INTERNAL_PHY:
3673 		igb_set_internal_phy_loopback(igb);
3674 		break;
3675 
3676 	case IGB_LB_INTERNAL_SERDES:
3677 		igb_set_internal_serdes_loopback(igb);
3678 		break;
3679 	}
3680 
3681 	mutex_exit(&igb->gen_lock);
3682 
3683 	/*
3684 	 * When external loopback is set, wait up to 1000ms to get the link up.
3685 	 * According to test, 1000ms can work and it's an experimental value.
3686 	 */
3687 	if (mode == IGB_LB_EXTERNAL) {
3688 		for (i = 0; i <= 10; i++) {
3689 			mutex_enter(&igb->gen_lock);
3690 			(void) igb_link_check(igb);
3691 			mutex_exit(&igb->gen_lock);
3692 
3693 			if (igb->link_state == LINK_STATE_UP)
3694 				break;
3695 
3696 			msec_delay(100);
3697 		}
3698 
3699 		if (igb->link_state != LINK_STATE_UP) {
3700 			/*
3701 			 * Does not support external loopback.
3702 			 * Reset driver to loopback none.
3703 			 */
3704 			igb->loopback_mode = IGB_LB_NONE;
3705 
3706 			/* Reset the chip */
3707 			hw->phy.autoneg_wait_to_complete = B_TRUE;
3708 			(void) igb_reset(igb);
3709 			hw->phy.autoneg_wait_to_complete = B_FALSE;
3710 
3711 			IGB_DEBUGLOG_0(igb, "Set external loopback failed, "
3712 			    "reset to loopback none.");
3713 
3714 			return (B_FALSE);
3715 		}
3716 	}
3717 
3718 	return (B_TRUE);
3719 }
3720 
3721 /*
3722  * igb_set_external_loopback - Set the external loopback mode
3723  */
3724 static void
3725 igb_set_external_loopback(igb_t *igb)
3726 {
3727 	struct e1000_hw *hw;
3728 
3729 	hw = &igb->hw;
3730 
3731 	/* Set phy to known state */
3732 	(void) e1000_phy_hw_reset(hw);
3733 
3734 	(void) e1000_write_phy_reg(hw, 0x0, 0x0140);
3735 	(void) e1000_write_phy_reg(hw, 0x9, 0x1b00);
3736 	(void) e1000_write_phy_reg(hw, 0x12, 0x1610);
3737 	(void) e1000_write_phy_reg(hw, 0x1f37, 0x3f1c);
3738 }
3739 
3740 /*
3741  * igb_set_internal_phy_loopback - Set the internal PHY loopback mode
3742  */
3743 static void
3744 igb_set_internal_phy_loopback(igb_t *igb)
3745 {
3746 	struct e1000_hw *hw;
3747 	uint32_t ctrl_ext;
3748 	uint16_t phy_ctrl;
3749 	uint16_t phy_pconf;
3750 
3751 	hw = &igb->hw;
3752 
3753 	/* Set link mode to PHY (00b) in the Extended Control register */
3754 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3755 	ctrl_ext &= ~E1000_CTRL_EXT_LINK_MODE_MASK;
3756 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3757 
3758 	/*
3759 	 * Set PHY control register (0x4140):
3760 	 *    Set full duplex mode
3761 	 *    Set loopback bit
3762 	 *    Clear auto-neg enable bit
3763 	 *    Set PHY speed
3764 	 */
3765 	phy_ctrl = MII_CR_FULL_DUPLEX | MII_CR_SPEED_1000 | MII_CR_LOOPBACK;
3766 	(void) e1000_write_phy_reg(hw, PHY_CONTROL, phy_ctrl);
3767 
3768 	/* Set the link disable bit in the Port Configuration register */
3769 	(void) e1000_read_phy_reg(hw, 0x10, &phy_pconf);
3770 	phy_pconf |= (uint16_t)1 << 14;
3771 	(void) e1000_write_phy_reg(hw, 0x10, phy_pconf);
3772 }
3773 
3774 /*
3775  * igb_set_internal_serdes_loopback - Set the internal SerDes loopback mode
3776  */
3777 static void
3778 igb_set_internal_serdes_loopback(igb_t *igb)
3779 {
3780 	struct e1000_hw *hw;
3781 	uint32_t ctrl_ext;
3782 	uint32_t ctrl;
3783 	uint32_t pcs_lctl;
3784 	uint32_t connsw;
3785 
3786 	hw = &igb->hw;
3787 
3788 	/* Set link mode to SerDes (11b) in the Extended Control register */
3789 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3790 	ctrl_ext |= E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
3791 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3792 
3793 	/* Configure the SerDes to loopback */
3794 	E1000_WRITE_REG(hw, E1000_SCTL, 0x410);
3795 
3796 	/* Set Device Control register */
3797 	ctrl = E1000_READ_REG(hw, E1000_CTRL);
3798 	ctrl |= (E1000_CTRL_FD |	/* Force full duplex */
3799 	    E1000_CTRL_SLU);		/* Force link up */
3800 	ctrl &= ~(E1000_CTRL_RFCE |	/* Disable receive flow control */
3801 	    E1000_CTRL_TFCE |		/* Disable transmit flow control */
3802 	    E1000_CTRL_LRST);		/* Clear link reset */
3803 	E1000_WRITE_REG(hw, E1000_CTRL, ctrl);
3804 
3805 	/* Set PCS Link Control register */
3806 	pcs_lctl = E1000_READ_REG(hw, E1000_PCS_LCTL);
3807 	pcs_lctl |= (E1000_PCS_LCTL_FORCE_LINK |
3808 	    E1000_PCS_LCTL_FSD |
3809 	    E1000_PCS_LCTL_FDV_FULL |
3810 	    E1000_PCS_LCTL_FLV_LINK_UP);
3811 	pcs_lctl &= ~E1000_PCS_LCTL_AN_ENABLE;
3812 	E1000_WRITE_REG(hw, E1000_PCS_LCTL, pcs_lctl);
3813 
3814 	/* Set the Copper/Fiber Switch Control - CONNSW register */
3815 	connsw = E1000_READ_REG(hw, E1000_CONNSW);
3816 	connsw &= ~E1000_CONNSW_ENRGSRC;
3817 	E1000_WRITE_REG(hw, E1000_CONNSW, connsw);
3818 }
3819 
3820 #pragma inline(igb_intr_rx_work)
3821 /*
3822  * igb_intr_rx_work - rx processing of ISR
3823  */
3824 static void
3825 igb_intr_rx_work(igb_rx_ring_t *rx_ring)
3826 {
3827 	mblk_t *mp;
3828 
3829 	mutex_enter(&rx_ring->rx_lock);
3830 	mp = igb_rx(rx_ring, IGB_NO_POLL);
3831 	mutex_exit(&rx_ring->rx_lock);
3832 
3833 	if (mp != NULL)
3834 		mac_rx_ring(rx_ring->igb->mac_hdl, rx_ring->ring_handle, mp,
3835 		    rx_ring->ring_gen_num);
3836 }
3837 
3838 #pragma inline(igb_intr_tx_work)
3839 /*
3840  * igb_intr_tx_work - tx processing of ISR
3841  */
3842 static void
3843 igb_intr_tx_work(igb_tx_ring_t *tx_ring)
3844 {
3845 	igb_t *igb = tx_ring->igb;
3846 
3847 	/* Recycle the tx descriptors */
3848 	tx_ring->tx_recycle(tx_ring);
3849 
3850 	/* Schedule the re-transmit */
3851 	if (tx_ring->reschedule &&
3852 	    (tx_ring->tbd_free >= igb->tx_resched_thresh)) {
3853 		tx_ring->reschedule = B_FALSE;
3854 		mac_tx_ring_update(tx_ring->igb->mac_hdl, tx_ring->ring_handle);
3855 		IGB_DEBUG_STAT(tx_ring->stat_reschedule);
3856 	}
3857 }
3858 
3859 #pragma inline(igb_intr_link_work)
3860 /*
3861  * igb_intr_link_work - link-status-change processing of ISR
3862  */
3863 static void
3864 igb_intr_link_work(igb_t *igb)
3865 {
3866 	boolean_t link_changed;
3867 
3868 	igb_stop_watchdog_timer(igb);
3869 
3870 	mutex_enter(&igb->gen_lock);
3871 
3872 	/*
3873 	 * Because we got a link-status-change interrupt, force
3874 	 * e1000_check_for_link() to look at phy
3875 	 */
3876 	igb->hw.mac.get_link_status = B_TRUE;
3877 
3878 	/* igb_link_check takes care of link status change */
3879 	link_changed = igb_link_check(igb);
3880 
3881 	/* Get new phy state */
3882 	igb_get_phy_state(igb);
3883 
3884 	mutex_exit(&igb->gen_lock);
3885 
3886 	if (link_changed)
3887 		mac_link_update(igb->mac_hdl, igb->link_state);
3888 
3889 	igb_start_watchdog_timer(igb);
3890 }
3891 
3892 /*
3893  * igb_intr_legacy - Interrupt handler for legacy interrupts
3894  */
3895 static uint_t
3896 igb_intr_legacy(void *arg1, void *arg2)
3897 {
3898 	igb_t *igb = (igb_t *)arg1;
3899 	igb_tx_ring_t *tx_ring;
3900 	uint32_t icr;
3901 	mblk_t *mp;
3902 	boolean_t tx_reschedule;
3903 	boolean_t link_changed;
3904 	uint_t result;
3905 
3906 	_NOTE(ARGUNUSED(arg2));
3907 
3908 	mutex_enter(&igb->gen_lock);
3909 
3910 	if (igb->igb_state & IGB_SUSPENDED) {
3911 		mutex_exit(&igb->gen_lock);
3912 		return (DDI_INTR_UNCLAIMED);
3913 	}
3914 
3915 	mp = NULL;
3916 	tx_reschedule = B_FALSE;
3917 	link_changed = B_FALSE;
3918 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
3919 
3920 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
3921 		mutex_exit(&igb->gen_lock);
3922 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
3923 		atomic_or_32(&igb->igb_state, IGB_ERROR);
3924 		return (DDI_INTR_UNCLAIMED);
3925 	}
3926 
3927 	if (icr & E1000_ICR_INT_ASSERTED) {
3928 		/*
3929 		 * E1000_ICR_INT_ASSERTED bit was set:
3930 		 * Read(Clear) the ICR, claim this interrupt,
3931 		 * look for work to do.
3932 		 */
3933 		ASSERT(igb->num_rx_rings == 1);
3934 		ASSERT(igb->num_tx_rings == 1);
3935 
3936 		/* Make sure all interrupt causes cleared */
3937 		(void) E1000_READ_REG(&igb->hw, E1000_EICR);
3938 
3939 		if (icr & E1000_ICR_RXT0) {
3940 			mp = igb_rx(&igb->rx_rings[0], IGB_NO_POLL);
3941 		}
3942 
3943 		if (icr & E1000_ICR_TXDW) {
3944 			tx_ring = &igb->tx_rings[0];
3945 
3946 			/* Recycle the tx descriptors */
3947 			tx_ring->tx_recycle(tx_ring);
3948 
3949 			/* Schedule the re-transmit */
3950 			tx_reschedule = (tx_ring->reschedule &&
3951 			    (tx_ring->tbd_free >= igb->tx_resched_thresh));
3952 		}
3953 
3954 		if (icr & E1000_ICR_LSC) {
3955 			/*
3956 			 * Because we got a link-status-change interrupt, force
3957 			 * e1000_check_for_link() to look at phy
3958 			 */
3959 			igb->hw.mac.get_link_status = B_TRUE;
3960 
3961 			/* igb_link_check takes care of link status change */
3962 			link_changed = igb_link_check(igb);
3963 
3964 			/* Get new phy state */
3965 			igb_get_phy_state(igb);
3966 		}
3967 
3968 		if (icr & E1000_ICR_DRSTA) {
3969 			/* 82580 Full Device Reset needed */
3970 			atomic_or_32(&igb->igb_state, IGB_STALL);
3971 		}
3972 
3973 		result = DDI_INTR_CLAIMED;
3974 	} else {
3975 		/*
3976 		 * E1000_ICR_INT_ASSERTED bit was not set:
3977 		 * Don't claim this interrupt.
3978 		 */
3979 		result = DDI_INTR_UNCLAIMED;
3980 	}
3981 
3982 	mutex_exit(&igb->gen_lock);
3983 
3984 	/*
3985 	 * Do the following work outside of the gen_lock
3986 	 */
3987 	if (mp != NULL)
3988 		mac_rx(igb->mac_hdl, NULL, mp);
3989 
3990 	if (tx_reschedule)  {
3991 		tx_ring->reschedule = B_FALSE;
3992 		mac_tx_ring_update(igb->mac_hdl, tx_ring->ring_handle);
3993 		IGB_DEBUG_STAT(tx_ring->stat_reschedule);
3994 	}
3995 
3996 	if (link_changed)
3997 		mac_link_update(igb->mac_hdl, igb->link_state);
3998 
3999 	return (result);
4000 }
4001 
4002 /*
4003  * igb_intr_msi - Interrupt handler for MSI
4004  */
4005 static uint_t
4006 igb_intr_msi(void *arg1, void *arg2)
4007 {
4008 	igb_t *igb = (igb_t *)arg1;
4009 	uint32_t icr;
4010 
4011 	_NOTE(ARGUNUSED(arg2));
4012 
4013 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
4014 
4015 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
4016 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
4017 		atomic_or_32(&igb->igb_state, IGB_ERROR);
4018 		return (DDI_INTR_CLAIMED);
4019 	}
4020 
4021 	/* Make sure all interrupt causes cleared */
4022 	(void) E1000_READ_REG(&igb->hw, E1000_EICR);
4023 
4024 	/*
4025 	 * For MSI interrupt, we have only one vector,
4026 	 * so we have only one rx ring and one tx ring enabled.
4027 	 */
4028 	ASSERT(igb->num_rx_rings == 1);
4029 	ASSERT(igb->num_tx_rings == 1);
4030 
4031 	if (icr & E1000_ICR_RXT0) {
4032 		igb_intr_rx_work(&igb->rx_rings[0]);
4033 	}
4034 
4035 	if (icr & E1000_ICR_TXDW) {
4036 		igb_intr_tx_work(&igb->tx_rings[0]);
4037 	}
4038 
4039 	if (icr & E1000_ICR_LSC) {
4040 		igb_intr_link_work(igb);
4041 	}
4042 
4043 	if (icr & E1000_ICR_DRSTA) {
4044 		/* 82580 Full Device Reset needed */
4045 		atomic_or_32(&igb->igb_state, IGB_STALL);
4046 	}
4047 
4048 	return (DDI_INTR_CLAIMED);
4049 }
4050 
4051 /*
4052  * igb_intr_rx - Interrupt handler for rx
4053  */
4054 static uint_t
4055 igb_intr_rx(void *arg1, void *arg2)
4056 {
4057 	igb_rx_ring_t *rx_ring = (igb_rx_ring_t *)arg1;
4058 
4059 	_NOTE(ARGUNUSED(arg2));
4060 
4061 	/*
4062 	 * Only used via MSI-X vector so don't check cause bits
4063 	 * and only clean the given ring.
4064 	 */
4065 	igb_intr_rx_work(rx_ring);
4066 
4067 	return (DDI_INTR_CLAIMED);
4068 }
4069 
4070 /*
4071  * igb_intr_tx - Interrupt handler for tx
4072  */
4073 static uint_t
4074 igb_intr_tx(void *arg1, void *arg2)
4075 {
4076 	igb_tx_ring_t *tx_ring = (igb_tx_ring_t *)arg1;
4077 
4078 	_NOTE(ARGUNUSED(arg2));
4079 
4080 	/*
4081 	 * Only used via MSI-X vector so don't check cause bits
4082 	 * and only clean the given ring.
4083 	 */
4084 	igb_intr_tx_work(tx_ring);
4085 
4086 	return (DDI_INTR_CLAIMED);
4087 }
4088 
4089 /*
4090  * igb_intr_tx_other - Interrupt handler for both tx and other
4091  *
4092  */
4093 static uint_t
4094 igb_intr_tx_other(void *arg1, void *arg2)
4095 {
4096 	igb_t *igb = (igb_t *)arg1;
4097 	uint32_t icr;
4098 
4099 	_NOTE(ARGUNUSED(arg2));
4100 
4101 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
4102 
4103 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
4104 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
4105 		atomic_or_32(&igb->igb_state, IGB_ERROR);
4106 		return (DDI_INTR_CLAIMED);
4107 	}
4108 
4109 	/*
4110 	 * Look for tx reclaiming work first. Remember, in the
4111 	 * case of only interrupt sharing, only one tx ring is
4112 	 * used
4113 	 */
4114 	igb_intr_tx_work(&igb->tx_rings[0]);
4115 
4116 	/*
4117 	 * Check for "other" causes.
4118 	 */
4119 	if (icr & E1000_ICR_LSC) {
4120 		igb_intr_link_work(igb);
4121 	}
4122 
4123 	/*
4124 	 * The DOUTSYNC bit indicates a tx packet dropped because
4125 	 * DMA engine gets "out of sync". There isn't a real fix
4126 	 * for this. The Intel recommendation is to count the number
4127 	 * of occurrences so user can detect when it is happening.
4128 	 * The issue is non-fatal and there's no recovery action
4129 	 * available.
4130 	 */
4131 	if (icr & E1000_ICR_DOUTSYNC) {
4132 		IGB_STAT(igb->dout_sync);
4133 	}
4134 
4135 	if (icr & E1000_ICR_DRSTA) {
4136 		/* 82580 Full Device Reset needed */
4137 		atomic_or_32(&igb->igb_state, IGB_STALL);
4138 	}
4139 
4140 	return (DDI_INTR_CLAIMED);
4141 }
4142 
4143 /*
4144  * igb_alloc_intrs - Allocate interrupts for the driver
4145  *
4146  * Normal sequence is to try MSI-X; if not sucessful, try MSI;
4147  * if not successful, try Legacy.
4148  * igb->intr_force can be used to force sequence to start with
4149  * any of the 3 types.
4150  * If MSI-X is not used, number of tx/rx rings is forced to 1.
4151  */
4152 static int
4153 igb_alloc_intrs(igb_t *igb)
4154 {
4155 	dev_info_t *devinfo;
4156 	int intr_types;
4157 	int rc;
4158 
4159 	devinfo = igb->dip;
4160 
4161 	/* Get supported interrupt types */
4162 	rc = ddi_intr_get_supported_types(devinfo, &intr_types);
4163 
4164 	if (rc != DDI_SUCCESS) {
4165 		igb_log(igb,
4166 		    "Get supported interrupt types failed: %d", rc);
4167 		return (IGB_FAILURE);
4168 	}
4169 	IGB_DEBUGLOG_1(igb, "Supported interrupt types: %x", intr_types);
4170 
4171 	igb->intr_type = 0;
4172 
4173 	/* Install MSI-X interrupts */
4174 	if ((intr_types & DDI_INTR_TYPE_MSIX) &&
4175 	    (igb->intr_force <= IGB_INTR_MSIX)) {
4176 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_MSIX);
4177 
4178 		if (rc == IGB_SUCCESS)
4179 			return (IGB_SUCCESS);
4180 
4181 		igb_log(igb,
4182 		    "Allocate MSI-X failed, trying MSI interrupts...");
4183 	}
4184 
4185 	/* MSI-X not used, force rings to 1 */
4186 	igb->num_rx_rings = 1;
4187 	igb->num_tx_rings = 1;
4188 	igb_log(igb,
4189 	    "MSI-X not used, force rx and tx queue number to 1");
4190 
4191 	/* Install MSI interrupts */
4192 	if ((intr_types & DDI_INTR_TYPE_MSI) &&
4193 	    (igb->intr_force <= IGB_INTR_MSI)) {
4194 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_MSI);
4195 
4196 		if (rc == IGB_SUCCESS)
4197 			return (IGB_SUCCESS);
4198 
4199 		igb_log(igb,
4200 		    "Allocate MSI failed, trying Legacy interrupts...");
4201 	}
4202 
4203 	/* Install legacy interrupts */
4204 	if (intr_types & DDI_INTR_TYPE_FIXED) {
4205 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_FIXED);
4206 
4207 		if (rc == IGB_SUCCESS)
4208 			return (IGB_SUCCESS);
4209 
4210 		igb_log(igb,
4211 		    "Allocate Legacy interrupts failed");
4212 	}
4213 
4214 	/* If none of the 3 types succeeded, return failure */
4215 	return (IGB_FAILURE);
4216 }
4217 
4218 /*
4219  * igb_alloc_intr_handles - Allocate interrupt handles.
4220  *
4221  * For legacy and MSI, only 1 handle is needed.  For MSI-X,
4222  * if fewer than 2 handles are available, return failure.
4223  * Upon success, this sets the number of Rx rings to a number that
4224  * matches the handles available for Rx interrupts.
4225  */
4226 static int
4227 igb_alloc_intr_handles(igb_t *igb, int intr_type)
4228 {
4229 	dev_info_t *devinfo;
4230 	int orig, request, count, avail, actual;
4231 	int diff, minimum;
4232 	int rc;
4233 
4234 	devinfo = igb->dip;
4235 
4236 	switch (intr_type) {
4237 	case DDI_INTR_TYPE_FIXED:
4238 		request = 1;	/* Request 1 legacy interrupt handle */
4239 		minimum = 1;
4240 		IGB_DEBUGLOG_0(igb, "interrupt type: legacy");
4241 		break;
4242 
4243 	case DDI_INTR_TYPE_MSI:
4244 		request = 1;	/* Request 1 MSI interrupt handle */
4245 		minimum = 1;
4246 		IGB_DEBUGLOG_0(igb, "interrupt type: MSI");
4247 		break;
4248 
4249 	case DDI_INTR_TYPE_MSIX:
4250 		/*
4251 		 * Number of vectors for the adapter is
4252 		 * # rx rings + # tx rings
4253 		 * One of tx vectors is for tx & other
4254 		 */
4255 		request = igb->num_rx_rings + igb->num_tx_rings;
4256 		orig = request;
4257 		minimum = 2;
4258 		IGB_DEBUGLOG_0(igb, "interrupt type: MSI-X");
4259 		break;
4260 
4261 	default:
4262 		igb_log(igb,
4263 		    "invalid call to igb_alloc_intr_handles(): %d\n",
4264 		    intr_type);
4265 		return (IGB_FAILURE);
4266 	}
4267 	IGB_DEBUGLOG_2(igb, "interrupt handles requested: %d  minimum: %d",
4268 	    request, minimum);
4269 
4270 	/*
4271 	 * Get number of supported interrupts
4272 	 */
4273 	rc = ddi_intr_get_nintrs(devinfo, intr_type, &count);
4274 	if ((rc != DDI_SUCCESS) || (count < minimum)) {
4275 		igb_log(igb,
4276 		    "Get supported interrupt number failed. "
4277 		    "Return: %d, count: %d", rc, count);
4278 		return (IGB_FAILURE);
4279 	}
4280 	IGB_DEBUGLOG_1(igb, "interrupts supported: %d", count);
4281 
4282 	/*
4283 	 * Get number of available interrupts
4284 	 */
4285 	rc = ddi_intr_get_navail(devinfo, intr_type, &avail);
4286 	if ((rc != DDI_SUCCESS) || (avail < minimum)) {
4287 		igb_log(igb,
4288 		    "Get available interrupt number failed. "
4289 		    "Return: %d, available: %d", rc, avail);
4290 		return (IGB_FAILURE);
4291 	}
4292 	IGB_DEBUGLOG_1(igb, "interrupts available: %d", avail);
4293 
4294 	if (avail < request) {
4295 		igb_log(igb, "Request %d handles, %d available",
4296 		    request, avail);
4297 		request = avail;
4298 	}
4299 
4300 	actual = 0;
4301 	igb->intr_cnt = 0;
4302 
4303 	/*
4304 	 * Allocate an array of interrupt handles
4305 	 */
4306 	igb->intr_size = request * sizeof (ddi_intr_handle_t);
4307 	igb->htable = kmem_alloc(igb->intr_size, KM_SLEEP);
4308 
4309 	rc = ddi_intr_alloc(devinfo, igb->htable, intr_type, 0,
4310 	    request, &actual, DDI_INTR_ALLOC_NORMAL);
4311 	if (rc != DDI_SUCCESS) {
4312 		igb_log(igb, "Allocate interrupts failed. "
4313 		    "return: %d, request: %d, actual: %d",
4314 		    rc, request, actual);
4315 		goto alloc_handle_fail;
4316 	}
4317 	IGB_DEBUGLOG_1(igb, "interrupts actually allocated: %d", actual);
4318 
4319 	igb->intr_cnt = actual;
4320 
4321 	if (actual < minimum) {
4322 		igb_log(igb, "Insufficient interrupt handles allocated: %d",
4323 		    actual);
4324 		goto alloc_handle_fail;
4325 	}
4326 
4327 	/*
4328 	 * For MSI-X, actual might force us to reduce number of tx & rx rings
4329 	 */
4330 	if ((intr_type == DDI_INTR_TYPE_MSIX) && (orig > actual)) {
4331 		diff = orig - actual;
4332 		if (diff < igb->num_tx_rings) {
4333 			igb_log(igb,
4334 			    "MSI-X vectors force Tx queue number to %d",
4335 			    igb->num_tx_rings - diff);
4336 			igb->num_tx_rings -= diff;
4337 		} else {
4338 			igb_log(igb,
4339 			    "MSI-X vectors force Tx queue number to 1");
4340 			igb->num_tx_rings = 1;
4341 
4342 			igb_log(igb,
4343 			    "MSI-X vectors force Rx queue number to %d",
4344 			    actual - 1);
4345 			igb->num_rx_rings = actual - 1;
4346 		}
4347 	}
4348 
4349 	/*
4350 	 * Get priority for first vector, assume remaining are all the same
4351 	 */
4352 	rc = ddi_intr_get_pri(igb->htable[0], &igb->intr_pri);
4353 	if (rc != DDI_SUCCESS) {
4354 		igb_log(igb,
4355 		    "Get interrupt priority failed: %d", rc);
4356 		goto alloc_handle_fail;
4357 	}
4358 
4359 	rc = ddi_intr_get_cap(igb->htable[0], &igb->intr_cap);
4360 	if (rc != DDI_SUCCESS) {
4361 		igb_log(igb,
4362 		    "Get interrupt cap failed: %d", rc);
4363 		goto alloc_handle_fail;
4364 	}
4365 
4366 	igb->intr_type = intr_type;
4367 
4368 	return (IGB_SUCCESS);
4369 
4370 alloc_handle_fail:
4371 	igb_rem_intrs(igb);
4372 
4373 	return (IGB_FAILURE);
4374 }
4375 
4376 /*
4377  * igb_add_intr_handlers - Add interrupt handlers based on the interrupt type
4378  *
4379  * Before adding the interrupt handlers, the interrupt vectors have
4380  * been allocated, and the rx/tx rings have also been allocated.
4381  */
4382 static int
4383 igb_add_intr_handlers(igb_t *igb)
4384 {
4385 	igb_rx_ring_t *rx_ring;
4386 	igb_tx_ring_t *tx_ring;
4387 	int vector;
4388 	int rc;
4389 	int i;
4390 
4391 	vector = 0;
4392 
4393 	switch (igb->intr_type) {
4394 	case DDI_INTR_TYPE_MSIX:
4395 		/* Add interrupt handler for tx + other */
4396 		tx_ring = &igb->tx_rings[0];
4397 		rc = ddi_intr_add_handler(igb->htable[vector],
4398 		    (ddi_intr_handler_t *)igb_intr_tx_other,
4399 		    (void *)igb, NULL);
4400 
4401 		if (rc != DDI_SUCCESS) {
4402 			igb_log(igb,
4403 			    "Add tx/other interrupt handler failed: %d", rc);
4404 			return (IGB_FAILURE);
4405 		}
4406 		tx_ring->intr_vector = vector;
4407 		vector++;
4408 
4409 		/* Add interrupt handler for each rx ring */
4410 		for (i = 0; i < igb->num_rx_rings; i++) {
4411 			rx_ring = &igb->rx_rings[i];
4412 
4413 			rc = ddi_intr_add_handler(igb->htable[vector],
4414 			    (ddi_intr_handler_t *)igb_intr_rx,
4415 			    (void *)rx_ring, NULL);
4416 
4417 			if (rc != DDI_SUCCESS) {
4418 				igb_log(igb,
4419 				    "Add rx interrupt handler failed. "
4420 				    "return: %d, rx ring: %d", rc, i);
4421 				for (vector--; vector >= 0; vector--) {
4422 					(void) ddi_intr_remove_handler(
4423 					    igb->htable[vector]);
4424 				}
4425 				return (IGB_FAILURE);
4426 			}
4427 
4428 			rx_ring->intr_vector = vector;
4429 
4430 			vector++;
4431 		}
4432 
4433 		/* Add interrupt handler for each tx ring from 2nd ring */
4434 		for (i = 1; i < igb->num_tx_rings; i++) {
4435 			tx_ring = &igb->tx_rings[i];
4436 
4437 			rc = ddi_intr_add_handler(igb->htable[vector],
4438 			    (ddi_intr_handler_t *)igb_intr_tx,
4439 			    (void *)tx_ring, NULL);
4440 
4441 			if (rc != DDI_SUCCESS) {
4442 				igb_log(igb,
4443 				    "Add tx interrupt handler failed. "
4444 				    "return: %d, tx ring: %d", rc, i);
4445 				for (vector--; vector >= 0; vector--) {
4446 					(void) ddi_intr_remove_handler(
4447 					    igb->htable[vector]);
4448 				}
4449 				return (IGB_FAILURE);
4450 			}
4451 
4452 			tx_ring->intr_vector = vector;
4453 
4454 			vector++;
4455 		}
4456 
4457 		break;
4458 
4459 	case DDI_INTR_TYPE_MSI:
4460 		/* Add interrupt handlers for the only vector */
4461 		rc = ddi_intr_add_handler(igb->htable[vector],
4462 		    (ddi_intr_handler_t *)igb_intr_msi,
4463 		    (void *)igb, NULL);
4464 
4465 		if (rc != DDI_SUCCESS) {
4466 			igb_log(igb,
4467 			    "Add MSI interrupt handler failed: %d", rc);
4468 			return (IGB_FAILURE);
4469 		}
4470 
4471 		rx_ring = &igb->rx_rings[0];
4472 		rx_ring->intr_vector = vector;
4473 
4474 		vector++;
4475 		break;
4476 
4477 	case DDI_INTR_TYPE_FIXED:
4478 		/* Add interrupt handlers for the only vector */
4479 		rc = ddi_intr_add_handler(igb->htable[vector],
4480 		    (ddi_intr_handler_t *)igb_intr_legacy,
4481 		    (void *)igb, NULL);
4482 
4483 		if (rc != DDI_SUCCESS) {
4484 			igb_log(igb,
4485 			    "Add legacy interrupt handler failed: %d", rc);
4486 			return (IGB_FAILURE);
4487 		}
4488 
4489 		rx_ring = &igb->rx_rings[0];
4490 		rx_ring->intr_vector = vector;
4491 
4492 		vector++;
4493 		break;
4494 
4495 	default:
4496 		return (IGB_FAILURE);
4497 	}
4498 
4499 	ASSERT(vector == igb->intr_cnt);
4500 
4501 	return (IGB_SUCCESS);
4502 }
4503 
4504 /*
4505  * igb_setup_msix_82575 - setup 82575 adapter to use MSI-X interrupts
4506  *
4507  * For each vector enabled on the adapter, Set the MSIXBM register accordingly
4508  */
4509 static void
4510 igb_setup_msix_82575(igb_t *igb)
4511 {
4512 	uint32_t eims = 0;
4513 	int i, vector;
4514 	struct e1000_hw *hw = &igb->hw;
4515 
4516 	/*
4517 	 * Set vector for tx ring 0 and other causes.
4518 	 * NOTE assumption that it is vector 0.
4519 	 */
4520 	vector = 0;
4521 
4522 	igb->eims_mask = E1000_EICR_TX_QUEUE0 | E1000_EICR_OTHER;
4523 	E1000_WRITE_REG(hw, E1000_MSIXBM(vector), igb->eims_mask);
4524 	vector++;
4525 
4526 	for (i = 0; i < igb->num_rx_rings; i++) {
4527 		/*
4528 		 * Set vector for each rx ring
4529 		 */
4530 		eims = (E1000_EICR_RX_QUEUE0 << i);
4531 		E1000_WRITE_REG(hw, E1000_MSIXBM(vector), eims);
4532 
4533 		/*
4534 		 * Accumulate bits to enable in
4535 		 * igb_enable_adapter_interrupts_82575()
4536 		 */
4537 		igb->eims_mask |= eims;
4538 
4539 		vector++;
4540 	}
4541 
4542 	for (i = 1; i < igb->num_tx_rings; i++) {
4543 		/*
4544 		 * Set vector for each tx ring from 2nd tx ring
4545 		 */
4546 		eims = (E1000_EICR_TX_QUEUE0 << i);
4547 		E1000_WRITE_REG(hw, E1000_MSIXBM(vector), eims);
4548 
4549 		/*
4550 		 * Accumulate bits to enable in
4551 		 * igb_enable_adapter_interrupts_82575()
4552 		 */
4553 		igb->eims_mask |= eims;
4554 
4555 		vector++;
4556 	}
4557 
4558 	ASSERT(vector == igb->intr_cnt);
4559 
4560 	/*
4561 	 * Disable IAM for ICR interrupt bits
4562 	 */
4563 	E1000_WRITE_REG(hw, E1000_IAM, 0);
4564 	E1000_WRITE_FLUSH(hw);
4565 }
4566 
4567 /*
4568  * igb_setup_msix_82576 - setup 82576 adapter to use MSI-X interrupts
4569  *
4570  * 82576 uses a table based method for assigning vectors.  Each queue has a
4571  * single entry in the table to which we write a vector number along with a
4572  * "valid" bit.  The entry is a single byte in a 4-byte register.  Vectors
4573  * take a different position in the 4-byte register depending on whether
4574  * they are numbered above or below 8.
4575  */
4576 static void
4577 igb_setup_msix_82576(igb_t *igb)
4578 {
4579 	struct e1000_hw *hw = &igb->hw;
4580 	uint32_t ivar, index, vector;
4581 	int i;
4582 
4583 	/* must enable msi-x capability before IVAR settings */
4584 	E1000_WRITE_REG(hw, E1000_GPIE,
4585 	    (E1000_GPIE_MSIX_MODE | E1000_GPIE_PBA | E1000_GPIE_NSICR));
4586 
4587 	/*
4588 	 * Set vector for tx ring 0 and other causes.
4589 	 * NOTE assumption that it is vector 0.
4590 	 * This is also interdependent with installation of interrupt service
4591 	 * routines in igb_add_intr_handlers().
4592 	 */
4593 
4594 	/* assign "other" causes to vector 0 */
4595 	vector = 0;
4596 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4597 	E1000_WRITE_REG(hw, E1000_IVAR_MISC, ivar);
4598 
4599 	/* assign tx ring 0 to vector 0 */
4600 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4601 	E1000_WRITE_REG(hw, E1000_IVAR0, ivar);
4602 
4603 	/* prepare to enable tx & other interrupt causes */
4604 	igb->eims_mask = (1 << vector);
4605 
4606 	vector ++;
4607 	for (i = 0; i < igb->num_rx_rings; i++) {
4608 		/*
4609 		 * Set vector for each rx ring
4610 		 */
4611 		index = (i & 0x7);
4612 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4613 
4614 		if (i < 8) {
4615 			/* vector goes into low byte of register */
4616 			ivar = ivar & 0xFFFFFF00;
4617 			ivar |= (vector | E1000_IVAR_VALID);
4618 		} else {
4619 			/* vector goes into third byte of register */
4620 			ivar = ivar & 0xFF00FFFF;
4621 			ivar |= ((vector | E1000_IVAR_VALID) << 16);
4622 		}
4623 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4624 
4625 		/* Accumulate interrupt-cause bits to enable */
4626 		igb->eims_mask |= (1 << vector);
4627 
4628 		vector ++;
4629 	}
4630 
4631 	for (i = 1; i < igb->num_tx_rings; i++) {
4632 		/*
4633 		 * Set vector for each tx ring from 2nd tx ring.
4634 		 * Note assumption that tx vectors numericall follow rx vectors.
4635 		 */
4636 		index = (i & 0x7);
4637 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4638 
4639 		if (i < 8) {
4640 			/* vector goes into second byte of register */
4641 			ivar = ivar & 0xFFFF00FF;
4642 			ivar |= ((vector | E1000_IVAR_VALID) << 8);
4643 		} else {
4644 			/* vector goes into fourth byte of register */
4645 			ivar = ivar & 0x00FFFFFF;
4646 			ivar |= (vector | E1000_IVAR_VALID) << 24;
4647 		}
4648 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4649 
4650 		/* Accumulate interrupt-cause bits to enable */
4651 		igb->eims_mask |= (1 << vector);
4652 
4653 		vector ++;
4654 	}
4655 
4656 	ASSERT(vector == igb->intr_cnt);
4657 }
4658 
4659 /*
4660  * igb_setup_msix_82580 - setup 82580 adapter to use MSI-X interrupts
4661  *
4662  * 82580 uses same table approach at 82576 but has fewer entries.  Each
4663  * queue has a single entry in the table to which we write a vector number
4664  * along with a "valid" bit.  Vectors take a different position in the
4665  * register depending on * whether * they are numbered above or below 4.
4666  */
4667 static void
4668 igb_setup_msix_82580(igb_t *igb)
4669 {
4670 	struct e1000_hw *hw = &igb->hw;
4671 	uint32_t ivar, index, vector;
4672 	int i;
4673 
4674 	/* must enable msi-x capability before IVAR settings */
4675 	E1000_WRITE_REG(hw, E1000_GPIE, (E1000_GPIE_MSIX_MODE |
4676 	    E1000_GPIE_PBA | E1000_GPIE_NSICR | E1000_GPIE_EIAME));
4677 	/*
4678 	 * Set vector for tx ring 0 and other causes.
4679 	 * NOTE assumption that it is vector 0.
4680 	 * This is also interdependent with installation of interrupt service
4681 	 * routines in igb_add_intr_handlers().
4682 	 */
4683 
4684 	/* assign "other" causes to vector 0 */
4685 	vector = 0;
4686 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4687 	E1000_WRITE_REG(hw, E1000_IVAR_MISC, ivar);
4688 
4689 	/* assign tx ring 0 to vector 0 */
4690 	ivar = ((vector | E1000_IVAR_VALID) << 8);
4691 	E1000_WRITE_REG(hw, E1000_IVAR0, ivar);
4692 
4693 	/* prepare to enable tx & other interrupt causes */
4694 	igb->eims_mask = (1 << vector);
4695 
4696 	vector ++;
4697 
4698 	for (i = 0; i < igb->num_rx_rings; i++) {
4699 		/*
4700 		 * Set vector for each rx ring
4701 		 */
4702 		index = (i >> 1);
4703 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4704 
4705 		if (i & 1) {
4706 			/* vector goes into third byte of register */
4707 			ivar = ivar & 0xFF00FFFF;
4708 			ivar |= ((vector | E1000_IVAR_VALID) << 16);
4709 		} else {
4710 			/* vector goes into low byte of register */
4711 			ivar = ivar & 0xFFFFFF00;
4712 			ivar |= (vector | E1000_IVAR_VALID);
4713 		}
4714 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4715 
4716 		/* Accumulate interrupt-cause bits to enable */
4717 		igb->eims_mask |= (1 << vector);
4718 
4719 		vector ++;
4720 	}
4721 
4722 	for (i = 1; i < igb->num_tx_rings; i++) {
4723 		/*
4724 		 * Set vector for each tx ring from 2nd tx ring.
4725 		 * Note assumption that tx vectors numericall follow rx vectors.
4726 		 */
4727 		index = (i >> 1);
4728 		ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
4729 
4730 		if (i & 1) {
4731 			/* vector goes into high byte of register */
4732 			ivar = ivar & 0x00FFFFFF;
4733 			ivar |= ((vector | E1000_IVAR_VALID) << 24);
4734 		} else {
4735 			/* vector goes into second byte of register */
4736 			ivar = ivar & 0xFFFF00FF;
4737 			ivar |= (vector | E1000_IVAR_VALID) << 8;
4738 		}
4739 		E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
4740 
4741 		/* Accumulate interrupt-cause bits to enable */
4742 		igb->eims_mask |= (1 << vector);
4743 
4744 		vector ++;
4745 	}
4746 	ASSERT(vector == igb->intr_cnt);
4747 }
4748 
4749 /*
4750  * igb_rem_intr_handlers - remove the interrupt handlers
4751  */
4752 static void
4753 igb_rem_intr_handlers(igb_t *igb)
4754 {
4755 	int i;
4756 	int rc;
4757 
4758 	for (i = 0; i < igb->intr_cnt; i++) {
4759 		rc = ddi_intr_remove_handler(igb->htable[i]);
4760 		if (rc != DDI_SUCCESS) {
4761 			IGB_DEBUGLOG_1(igb,
4762 			    "Remove intr handler failed: %d", rc);
4763 		}
4764 	}
4765 }
4766 
4767 /*
4768  * igb_rem_intrs - remove the allocated interrupts
4769  */
4770 static void
4771 igb_rem_intrs(igb_t *igb)
4772 {
4773 	int i;
4774 	int rc;
4775 
4776 	for (i = 0; i < igb->intr_cnt; i++) {
4777 		rc = ddi_intr_free(igb->htable[i]);
4778 		if (rc != DDI_SUCCESS) {
4779 			IGB_DEBUGLOG_1(igb,
4780 			    "Free intr failed: %d", rc);
4781 		}
4782 	}
4783 
4784 	kmem_free(igb->htable, igb->intr_size);
4785 	igb->htable = NULL;
4786 }
4787 
4788 /*
4789  * igb_enable_intrs - enable all the ddi interrupts
4790  */
4791 static int
4792 igb_enable_intrs(igb_t *igb)
4793 {
4794 	int i;
4795 	int rc;
4796 
4797 	/* Enable interrupts */
4798 	if (igb->intr_cap & DDI_INTR_FLAG_BLOCK) {
4799 		/* Call ddi_intr_block_enable() for MSI */
4800 		rc = ddi_intr_block_enable(igb->htable, igb->intr_cnt);
4801 		if (rc != DDI_SUCCESS) {
4802 			igb_log(igb,
4803 			    "Enable block intr failed: %d", rc);
4804 			return (IGB_FAILURE);
4805 		}
4806 	} else {
4807 		/* Call ddi_intr_enable() for Legacy/MSI non block enable */
4808 		for (i = 0; i < igb->intr_cnt; i++) {
4809 			rc = ddi_intr_enable(igb->htable[i]);
4810 			if (rc != DDI_SUCCESS) {
4811 				igb_log(igb,
4812 				    "Enable intr failed: %d", rc);
4813 				return (IGB_FAILURE);
4814 			}
4815 		}
4816 	}
4817 
4818 	return (IGB_SUCCESS);
4819 }
4820 
4821 /*
4822  * igb_disable_intrs - disable all the ddi interrupts
4823  */
4824 static int
4825 igb_disable_intrs(igb_t *igb)
4826 {
4827 	int i;
4828 	int rc;
4829 
4830 	/* Disable all interrupts */
4831 	if (igb->intr_cap & DDI_INTR_FLAG_BLOCK) {
4832 		rc = ddi_intr_block_disable(igb->htable, igb->intr_cnt);
4833 		if (rc != DDI_SUCCESS) {
4834 			igb_log(igb,
4835 			    "Disable block intr failed: %d", rc);
4836 			return (IGB_FAILURE);
4837 		}
4838 	} else {
4839 		for (i = 0; i < igb->intr_cnt; i++) {
4840 			rc = ddi_intr_disable(igb->htable[i]);
4841 			if (rc != DDI_SUCCESS) {
4842 				igb_log(igb,
4843 				    "Disable intr failed: %d", rc);
4844 				return (IGB_FAILURE);
4845 			}
4846 		}
4847 	}
4848 
4849 	return (IGB_SUCCESS);
4850 }
4851 
4852 /*
4853  * igb_get_phy_state - Get and save the parameters read from PHY registers
4854  */
4855 static void
4856 igb_get_phy_state(igb_t *igb)
4857 {
4858 	struct e1000_hw *hw = &igb->hw;
4859 	uint16_t phy_ctrl;
4860 	uint16_t phy_status;
4861 	uint16_t phy_an_adv;
4862 	uint16_t phy_an_exp;
4863 	uint16_t phy_ext_status;
4864 	uint16_t phy_1000t_ctrl;
4865 	uint16_t phy_1000t_status;
4866 	uint16_t phy_lp_able;
4867 
4868 	ASSERT(mutex_owned(&igb->gen_lock));
4869 
4870 	if (hw->phy.media_type == e1000_media_type_copper) {
4871 		(void) e1000_read_phy_reg(hw, PHY_CONTROL, &phy_ctrl);
4872 		(void) e1000_read_phy_reg(hw, PHY_STATUS, &phy_status);
4873 		(void) e1000_read_phy_reg(hw, PHY_AUTONEG_ADV, &phy_an_adv);
4874 		(void) e1000_read_phy_reg(hw, PHY_AUTONEG_EXP, &phy_an_exp);
4875 		(void) e1000_read_phy_reg(hw, PHY_EXT_STATUS, &phy_ext_status);
4876 		(void) e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_1000t_ctrl);
4877 		(void) e1000_read_phy_reg(hw,
4878 		    PHY_1000T_STATUS, &phy_1000t_status);
4879 		(void) e1000_read_phy_reg(hw, PHY_LP_ABILITY, &phy_lp_able);
4880 
4881 		igb->param_autoneg_cap =
4882 		    (phy_status & MII_SR_AUTONEG_CAPS) ? 1 : 0;
4883 		igb->param_pause_cap =
4884 		    (phy_an_adv & NWAY_AR_PAUSE) ? 1 : 0;
4885 		igb->param_asym_pause_cap =
4886 		    (phy_an_adv & NWAY_AR_ASM_DIR) ? 1 : 0;
4887 		igb->param_1000fdx_cap =
4888 		    ((phy_ext_status & IEEE_ESR_1000T_FD_CAPS) ||
4889 		    (phy_ext_status & IEEE_ESR_1000X_FD_CAPS)) ? 1 : 0;
4890 		igb->param_1000hdx_cap =
4891 		    ((phy_ext_status & IEEE_ESR_1000T_HD_CAPS) ||
4892 		    (phy_ext_status & IEEE_ESR_1000X_HD_CAPS)) ? 1 : 0;
4893 		igb->param_100t4_cap =
4894 		    (phy_status & MII_SR_100T4_CAPS) ? 1 : 0;
4895 		igb->param_100fdx_cap = ((phy_status & MII_SR_100X_FD_CAPS) ||
4896 		    (phy_status & MII_SR_100T2_FD_CAPS)) ? 1 : 0;
4897 		igb->param_100hdx_cap = ((phy_status & MII_SR_100X_HD_CAPS) ||
4898 		    (phy_status & MII_SR_100T2_HD_CAPS)) ? 1 : 0;
4899 		igb->param_10fdx_cap =
4900 		    (phy_status & MII_SR_10T_FD_CAPS) ? 1 : 0;
4901 		igb->param_10hdx_cap =
4902 		    (phy_status & MII_SR_10T_HD_CAPS) ? 1 : 0;
4903 		igb->param_rem_fault =
4904 		    (phy_status & MII_SR_REMOTE_FAULT) ? 1 : 0;
4905 
4906 		igb->param_adv_autoneg_cap = hw->mac.autoneg;
4907 		igb->param_adv_pause_cap =
4908 		    (phy_an_adv & NWAY_AR_PAUSE) ? 1 : 0;
4909 		igb->param_adv_asym_pause_cap =
4910 		    (phy_an_adv & NWAY_AR_ASM_DIR) ? 1 : 0;
4911 		igb->param_adv_1000hdx_cap =
4912 		    (phy_1000t_ctrl & CR_1000T_HD_CAPS) ? 1 : 0;
4913 		igb->param_adv_100t4_cap =
4914 		    (phy_an_adv & NWAY_AR_100T4_CAPS) ? 1 : 0;
4915 		igb->param_adv_rem_fault =
4916 		    (phy_an_adv & NWAY_AR_REMOTE_FAULT) ? 1 : 0;
4917 		if (igb->param_adv_autoneg_cap == 1) {
4918 			igb->param_adv_1000fdx_cap =
4919 			    (phy_1000t_ctrl & CR_1000T_FD_CAPS) ? 1 : 0;
4920 			igb->param_adv_100fdx_cap =
4921 			    (phy_an_adv & NWAY_AR_100TX_FD_CAPS) ? 1 : 0;
4922 			igb->param_adv_100hdx_cap =
4923 			    (phy_an_adv & NWAY_AR_100TX_HD_CAPS) ? 1 : 0;
4924 			igb->param_adv_10fdx_cap =
4925 			    (phy_an_adv & NWAY_AR_10T_FD_CAPS) ? 1 : 0;
4926 			igb->param_adv_10hdx_cap =
4927 			    (phy_an_adv & NWAY_AR_10T_HD_CAPS) ? 1 : 0;
4928 		}
4929 
4930 		igb->param_lp_autoneg_cap =
4931 		    (phy_an_exp & NWAY_ER_LP_NWAY_CAPS) ? 1 : 0;
4932 		igb->param_lp_pause_cap =
4933 		    (phy_lp_able & NWAY_LPAR_PAUSE) ? 1 : 0;
4934 		igb->param_lp_asym_pause_cap =
4935 		    (phy_lp_able & NWAY_LPAR_ASM_DIR) ? 1 : 0;
4936 		igb->param_lp_1000fdx_cap =
4937 		    (phy_1000t_status & SR_1000T_LP_FD_CAPS) ? 1 : 0;
4938 		igb->param_lp_1000hdx_cap =
4939 		    (phy_1000t_status & SR_1000T_LP_HD_CAPS) ? 1 : 0;
4940 		igb->param_lp_100t4_cap =
4941 		    (phy_lp_able & NWAY_LPAR_100T4_CAPS) ? 1 : 0;
4942 		igb->param_lp_100fdx_cap =
4943 		    (phy_lp_able & NWAY_LPAR_100TX_FD_CAPS) ? 1 : 0;
4944 		igb->param_lp_100hdx_cap =
4945 		    (phy_lp_able & NWAY_LPAR_100TX_HD_CAPS) ? 1 : 0;
4946 		igb->param_lp_10fdx_cap =
4947 		    (phy_lp_able & NWAY_LPAR_10T_FD_CAPS) ? 1 : 0;
4948 		igb->param_lp_10hdx_cap =
4949 		    (phy_lp_able & NWAY_LPAR_10T_HD_CAPS) ? 1 : 0;
4950 		igb->param_lp_rem_fault =
4951 		    (phy_lp_able & NWAY_LPAR_REMOTE_FAULT) ? 1 : 0;
4952 	} else {
4953 		/*
4954 		 * 1Gig Fiber adapter only offers 1Gig Full Duplex.
4955 		 */
4956 		igb->param_autoneg_cap = 0;
4957 		igb->param_pause_cap = 1;
4958 		igb->param_asym_pause_cap = 1;
4959 		igb->param_1000fdx_cap = 1;
4960 		igb->param_1000hdx_cap = 0;
4961 		igb->param_100t4_cap = 0;
4962 		igb->param_100fdx_cap = 0;
4963 		igb->param_100hdx_cap = 0;
4964 		igb->param_10fdx_cap = 0;
4965 		igb->param_10hdx_cap = 0;
4966 
4967 		igb->param_adv_autoneg_cap = 0;
4968 		igb->param_adv_pause_cap = 1;
4969 		igb->param_adv_asym_pause_cap = 1;
4970 		igb->param_adv_1000fdx_cap = 1;
4971 		igb->param_adv_1000hdx_cap = 0;
4972 		igb->param_adv_100t4_cap = 0;
4973 		igb->param_adv_100fdx_cap = 0;
4974 		igb->param_adv_100hdx_cap = 0;
4975 		igb->param_adv_10fdx_cap = 0;
4976 		igb->param_adv_10hdx_cap = 0;
4977 
4978 		igb->param_lp_autoneg_cap = 0;
4979 		igb->param_lp_pause_cap = 0;
4980 		igb->param_lp_asym_pause_cap = 0;
4981 		igb->param_lp_1000fdx_cap = 0;
4982 		igb->param_lp_1000hdx_cap = 0;
4983 		igb->param_lp_100t4_cap = 0;
4984 		igb->param_lp_100fdx_cap = 0;
4985 		igb->param_lp_100hdx_cap = 0;
4986 		igb->param_lp_10fdx_cap = 0;
4987 		igb->param_lp_10hdx_cap = 0;
4988 		igb->param_lp_rem_fault = 0;
4989 	}
4990 }
4991 
4992 /*
4993  * synchronize the adv* and en* parameters.
4994  *
4995  * See comments in <sys/dld.h> for details of the *_en_*
4996  * parameters. The usage of ndd for setting adv parameters will
4997  * synchronize all the en parameters with the e1000g parameters,
4998  * implicitly disabling any settings made via dladm.
4999  */
5000 static void
5001 igb_param_sync(igb_t *igb)
5002 {
5003 	igb->param_en_1000fdx_cap = igb->param_adv_1000fdx_cap;
5004 	igb->param_en_1000hdx_cap = igb->param_adv_1000hdx_cap;
5005 	igb->param_en_100t4_cap = igb->param_adv_100t4_cap;
5006 	igb->param_en_100fdx_cap = igb->param_adv_100fdx_cap;
5007 	igb->param_en_100hdx_cap = igb->param_adv_100hdx_cap;
5008 	igb->param_en_10fdx_cap = igb->param_adv_10fdx_cap;
5009 	igb->param_en_10hdx_cap = igb->param_adv_10hdx_cap;
5010 }
5011 
5012 /*
5013  * igb_get_driver_control
5014  */
5015 static void
5016 igb_get_driver_control(struct e1000_hw *hw)
5017 {
5018 	uint32_t ctrl_ext;
5019 
5020 	/* Notify firmware that driver is in control of device */
5021 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
5022 	ctrl_ext |= E1000_CTRL_EXT_DRV_LOAD;
5023 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
5024 }
5025 
5026 /*
5027  * igb_release_driver_control
5028  */
5029 static void
5030 igb_release_driver_control(struct e1000_hw *hw)
5031 {
5032 	uint32_t ctrl_ext;
5033 
5034 	/* Notify firmware that driver is no longer in control of device */
5035 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
5036 	ctrl_ext &= ~E1000_CTRL_EXT_DRV_LOAD;
5037 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
5038 }
5039 
5040 /*
5041  * igb_atomic_reserve - Atomic decrease operation
5042  */
5043 int
5044 igb_atomic_reserve(uint32_t *count_p, uint32_t n)
5045 {
5046 	uint32_t oldval;
5047 	uint32_t newval;
5048 
5049 	/* ATOMICALLY */
5050 	do {
5051 		oldval = *count_p;
5052 		if (oldval < n)
5053 			return (-1);
5054 		newval = oldval - n;
5055 	} while (atomic_cas_32(count_p, oldval, newval) != oldval);
5056 
5057 	return (newval);
5058 }
5059 
5060 /*
5061  * FMA support
5062  */
5063 
5064 int
5065 igb_check_acc_handle(ddi_acc_handle_t handle)
5066 {
5067 	ddi_fm_error_t de;
5068 
5069 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
5070 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
5071 	return (de.fme_status);
5072 }
5073 
5074 int
5075 igb_check_dma_handle(ddi_dma_handle_t handle)
5076 {
5077 	ddi_fm_error_t de;
5078 
5079 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
5080 	return (de.fme_status);
5081 }
5082 
5083 /*
5084  * The IO fault service error handling callback function
5085  */
5086 /*ARGSUSED*/
5087 static int
5088 igb_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
5089 {
5090 	/*
5091 	 * as the driver can always deal with an error in any dma or
5092 	 * access handle, we can just return the fme_status value.
5093 	 */
5094 	pci_ereport_post(dip, err, NULL);
5095 	return (err->fme_status);
5096 }
5097 
5098 static void
5099 igb_fm_init(igb_t *igb)
5100 {
5101 	ddi_iblock_cookie_t iblk;
5102 	int fma_dma_flag;
5103 
5104 	/* Only register with IO Fault Services if we have some capability */
5105 	if (igb->fm_capabilities & DDI_FM_ACCCHK_CAPABLE) {
5106 		igb_regs_acc_attr.devacc_attr_access = DDI_FLAGERR_ACC;
5107 	} else {
5108 		igb_regs_acc_attr.devacc_attr_access = DDI_DEFAULT_ACC;
5109 	}
5110 
5111 	if (igb->fm_capabilities & DDI_FM_DMACHK_CAPABLE) {
5112 		fma_dma_flag = 1;
5113 	} else {
5114 		fma_dma_flag = 0;
5115 	}
5116 
5117 	(void) igb_set_fma_flags(fma_dma_flag);
5118 
5119 	if (igb->fm_capabilities) {
5120 
5121 		/* Register capabilities with IO Fault Services */
5122 		ddi_fm_init(igb->dip, &igb->fm_capabilities, &iblk);
5123 
5124 		/*
5125 		 * Initialize pci ereport capabilities if ereport capable
5126 		 */
5127 		if (DDI_FM_EREPORT_CAP(igb->fm_capabilities) ||
5128 		    DDI_FM_ERRCB_CAP(igb->fm_capabilities))
5129 			pci_ereport_setup(igb->dip);
5130 
5131 		/*
5132 		 * Register error callback if error callback capable
5133 		 */
5134 		if (DDI_FM_ERRCB_CAP(igb->fm_capabilities))
5135 			ddi_fm_handler_register(igb->dip,
5136 			    igb_fm_error_cb, (void*) igb);
5137 	}
5138 }
5139 
5140 static void
5141 igb_fm_fini(igb_t *igb)
5142 {
5143 	/* Only unregister FMA capabilities if we registered some */
5144 	if (igb->fm_capabilities) {
5145 
5146 		/*
5147 		 * Release any resources allocated by pci_ereport_setup()
5148 		 */
5149 		if (DDI_FM_EREPORT_CAP(igb->fm_capabilities) ||
5150 		    DDI_FM_ERRCB_CAP(igb->fm_capabilities))
5151 			pci_ereport_teardown(igb->dip);
5152 
5153 		/*
5154 		 * Un-register error callback if error callback capable
5155 		 */
5156 		if (DDI_FM_ERRCB_CAP(igb->fm_capabilities))
5157 			ddi_fm_handler_unregister(igb->dip);
5158 
5159 		/* Unregister from IO Fault Services */
5160 		ddi_fm_fini(igb->dip);
5161 	}
5162 }
5163 
5164 void
5165 igb_fm_ereport(igb_t *igb, char *detail)
5166 {
5167 	uint64_t ena;
5168 	char buf[FM_MAX_CLASS];
5169 
5170 	(void) snprintf(buf, FM_MAX_CLASS, "%s.%s", DDI_FM_DEVICE, detail);
5171 	ena = fm_ena_generate(0, FM_ENA_FMT1);
5172 	if (DDI_FM_EREPORT_CAP(igb->fm_capabilities)) {
5173 		ddi_fm_ereport_post(igb->dip, buf, ena, DDI_NOSLEEP,
5174 		    FM_VERSION, DATA_TYPE_UINT8, FM_EREPORT_VERS0, NULL);
5175 	}
5176 }
5177