xref: /titanic_50/usr/src/uts/common/io/igb/igb_main.c (revision 6e0414acac7fe81ff262fb9d3d83c0700fe9b695)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright(c) 2007-2008 Intel Corporation. All rights reserved.
24  */
25 
26 /*
27  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
28  * Use is subject to license terms.
29  */
30 
31 
32 #include "igb_sw.h"
33 
34 static char ident[] = "Intel 1Gb Ethernet";
35 
36 /*
37  * Local function protoypes
38  */
39 static int igb_register_mac(igb_t *);
40 static int igb_identify_hardware(igb_t *);
41 static int igb_regs_map(igb_t *);
42 static void igb_init_properties(igb_t *);
43 static int igb_init_driver_settings(igb_t *);
44 static void igb_init_locks(igb_t *);
45 static void igb_destroy_locks(igb_t *);
46 static int igb_init(igb_t *);
47 static int igb_chip_start(igb_t *);
48 static void igb_chip_stop(igb_t *);
49 static int igb_reset(igb_t *);
50 static void igb_tx_clean(igb_t *);
51 static boolean_t igb_tx_drain(igb_t *);
52 static boolean_t igb_rx_drain(igb_t *);
53 static int igb_alloc_rings(igb_t *);
54 static int igb_init_rings(igb_t *);
55 static void igb_free_rings(igb_t *);
56 static void igb_fini_rings(igb_t *);
57 static void igb_setup_rings(igb_t *);
58 static void igb_setup_rx(igb_t *);
59 static void igb_setup_tx(igb_t *);
60 static void igb_setup_rx_ring(igb_rx_ring_t *);
61 static void igb_setup_tx_ring(igb_tx_ring_t *);
62 static void igb_setup_rss(igb_t *);
63 static void igb_setup_mac_rss_classify(igb_t *);
64 static void igb_setup_mac_classify(igb_t *);
65 static void igb_init_unicst(igb_t *);
66 static void igb_setup_multicst(igb_t *);
67 static void igb_get_phy_state(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_arm_watchdog_timer(igb_t *);
74 static void igb_start_watchdog_timer(igb_t *);
75 static void igb_restart_watchdog_timer(igb_t *);
76 static void igb_stop_watchdog_timer(igb_t *);
77 static void igb_disable_adapter_interrupts(igb_t *);
78 static void igb_enable_adapter_interrupts(igb_t *);
79 static boolean_t is_valid_mac_addr(uint8_t *);
80 static boolean_t igb_stall_check(igb_t *);
81 static boolean_t igb_set_loopback_mode(igb_t *, uint32_t);
82 static void igb_set_external_loopback(igb_t *);
83 static void igb_set_internal_mac_loopback(igb_t *);
84 static void igb_set_internal_phy_loopback(igb_t *);
85 static void igb_set_internal_serdes_loopback(igb_t *);
86 static boolean_t igb_find_mac_address(igb_t *);
87 static int igb_alloc_intrs(igb_t *);
88 static int igb_alloc_intr_handles(igb_t *, int);
89 static int igb_add_intr_handlers(igb_t *);
90 static void igb_rem_intr_handlers(igb_t *);
91 static void igb_rem_intrs(igb_t *);
92 static int igb_enable_intrs(igb_t *);
93 static int igb_disable_intrs(igb_t *);
94 static void igb_setup_adapter_msix(igb_t *);
95 static uint_t igb_intr_legacy(void *, void *);
96 static uint_t igb_intr_msi(void *, void *);
97 static uint_t igb_intr_rx(void *, void *);
98 static uint_t igb_intr_tx(void *, void *);
99 static uint_t igb_intr_tx_other(void *, void *);
100 static void igb_intr_rx_work(igb_rx_ring_t *);
101 static void igb_intr_tx_work(igb_tx_ring_t *);
102 static void igb_intr_link_work(igb_t *);
103 static void igb_get_driver_control(struct e1000_hw *);
104 static void igb_release_driver_control(struct e1000_hw *);
105 
106 static int igb_attach(dev_info_t *, ddi_attach_cmd_t);
107 static int igb_detach(dev_info_t *, ddi_detach_cmd_t);
108 static int igb_resume(dev_info_t *);
109 static int igb_suspend(dev_info_t *);
110 static int igb_quiesce(dev_info_t *);
111 static void igb_unconfigure(dev_info_t *, igb_t *);
112 static int igb_fm_error_cb(dev_info_t *, ddi_fm_error_t *,
113     const void *);
114 static void igb_fm_init(igb_t *);
115 static void igb_fm_fini(igb_t *);
116 
117 
118 static struct cb_ops igb_cb_ops = {
119 	nulldev,		/* cb_open */
120 	nulldev,		/* cb_close */
121 	nodev,			/* cb_strategy */
122 	nodev,			/* cb_print */
123 	nodev,			/* cb_dump */
124 	nodev,			/* cb_read */
125 	nodev,			/* cb_write */
126 	nodev,			/* cb_ioctl */
127 	nodev,			/* cb_devmap */
128 	nodev,			/* cb_mmap */
129 	nodev,			/* cb_segmap */
130 	nochpoll,		/* cb_chpoll */
131 	ddi_prop_op,		/* cb_prop_op */
132 	NULL,			/* cb_stream */
133 	D_MP | D_HOTPLUG,	/* cb_flag */
134 	CB_REV,			/* cb_rev */
135 	nodev,			/* cb_aread */
136 	nodev			/* cb_awrite */
137 };
138 
139 static struct dev_ops igb_dev_ops = {
140 	DEVO_REV,		/* devo_rev */
141 	0,			/* devo_refcnt */
142 	NULL,			/* devo_getinfo */
143 	nulldev,		/* devo_identify */
144 	nulldev,		/* devo_probe */
145 	igb_attach,		/* devo_attach */
146 	igb_detach,		/* devo_detach */
147 	nodev,			/* devo_reset */
148 	&igb_cb_ops,		/* devo_cb_ops */
149 	NULL,			/* devo_bus_ops */
150 	ddi_power,		/* devo_power */
151 	igb_quiesce,	/* devo_quiesce */
152 };
153 
154 static struct modldrv igb_modldrv = {
155 	&mod_driverops,		/* Type of module.  This one is a driver */
156 	ident,			/* Discription string */
157 	&igb_dev_ops,		/* driver ops */
158 };
159 
160 static struct modlinkage igb_modlinkage = {
161 	MODREV_1, &igb_modldrv, NULL
162 };
163 
164 /* Access attributes for register mapping */
165 ddi_device_acc_attr_t igb_regs_acc_attr = {
166 	DDI_DEVICE_ATTR_V0,
167 	DDI_STRUCTURE_LE_ACC,
168 	DDI_STRICTORDER_ACC,
169 	DDI_FLAGERR_ACC
170 };
171 
172 #define	IGB_M_CALLBACK_FLAGS	(MC_IOCTL | MC_GETCAPAB)
173 
174 static mac_callbacks_t igb_m_callbacks = {
175 	IGB_M_CALLBACK_FLAGS,
176 	igb_m_stat,
177 	igb_m_start,
178 	igb_m_stop,
179 	igb_m_promisc,
180 	igb_m_multicst,
181 	NULL,
182 	NULL,
183 	igb_m_ioctl,
184 	igb_m_getcapab
185 };
186 
187 /*
188  * Module Initialization Functions
189  */
190 
191 int
192 _init(void)
193 {
194 	int status;
195 
196 	mac_init_ops(&igb_dev_ops, MODULE_NAME);
197 
198 	status = mod_install(&igb_modlinkage);
199 
200 	if (status != DDI_SUCCESS) {
201 		mac_fini_ops(&igb_dev_ops);
202 	}
203 
204 	return (status);
205 }
206 
207 int
208 _fini(void)
209 {
210 	int status;
211 
212 	status = mod_remove(&igb_modlinkage);
213 
214 	if (status == DDI_SUCCESS) {
215 		mac_fini_ops(&igb_dev_ops);
216 	}
217 
218 	return (status);
219 
220 }
221 
222 int
223 _info(struct modinfo *modinfop)
224 {
225 	int status;
226 
227 	status = mod_info(&igb_modlinkage, modinfop);
228 
229 	return (status);
230 }
231 
232 /*
233  * igb_attach - driver attach
234  *
235  * This function is the device specific initialization entry
236  * point. This entry point is required and must be written.
237  * The DDI_ATTACH command must be provided in the attach entry
238  * point. When attach() is called with cmd set to DDI_ATTACH,
239  * all normal kernel services (such as kmem_alloc(9F)) are
240  * available for use by the driver.
241  *
242  * The attach() function will be called once for each instance
243  * of  the  device  on  the  system with cmd set to DDI_ATTACH.
244  * Until attach() succeeds, the only driver entry points which
245  * may be called are open(9E) and getinfo(9E).
246  */
247 static int
248 igb_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
249 {
250 	igb_t *igb;
251 	struct igb_osdep *osdep;
252 	struct e1000_hw *hw;
253 	int instance;
254 
255 	/*
256 	 * Check the command and perform corresponding operations
257 	 */
258 	switch (cmd) {
259 	default:
260 		return (DDI_FAILURE);
261 
262 	case DDI_RESUME:
263 		return (igb_resume(devinfo));
264 
265 	case DDI_ATTACH:
266 		break;
267 	}
268 
269 	/* Get the device instance */
270 	instance = ddi_get_instance(devinfo);
271 
272 	/* Allocate memory for the instance data structure */
273 	igb = kmem_zalloc(sizeof (igb_t), KM_SLEEP);
274 
275 	igb->dip = devinfo;
276 	igb->instance = instance;
277 
278 	hw = &igb->hw;
279 	osdep = &igb->osdep;
280 	hw->back = osdep;
281 	osdep->igb = igb;
282 
283 	/* Attach the instance pointer to the dev_info data structure */
284 	ddi_set_driver_private(devinfo, igb);
285 
286 
287 	/* Initialize for fma support */
288 	igb->fm_capabilities = igb_get_prop(igb, "fm-capable",
289 	    0, 0x0f,
290 	    DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
291 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
292 	igb_fm_init(igb);
293 	igb->attach_progress |= ATTACH_PROGRESS_FMINIT;
294 
295 	/*
296 	 * Map PCI config space registers
297 	 */
298 	if (pci_config_setup(devinfo, &osdep->cfg_handle) != DDI_SUCCESS) {
299 		igb_error(igb, "Failed to map PCI configurations");
300 		goto attach_fail;
301 	}
302 	igb->attach_progress |= ATTACH_PROGRESS_PCI_CONFIG;
303 
304 	/*
305 	 * Identify the chipset family
306 	 */
307 	if (igb_identify_hardware(igb) != IGB_SUCCESS) {
308 		igb_error(igb, "Failed to identify hardware");
309 		goto attach_fail;
310 	}
311 
312 	/*
313 	 * Map device registers
314 	 */
315 	if (igb_regs_map(igb) != IGB_SUCCESS) {
316 		igb_error(igb, "Failed to map device registers");
317 		goto attach_fail;
318 	}
319 	igb->attach_progress |= ATTACH_PROGRESS_REGS_MAP;
320 
321 	/*
322 	 * Initialize driver parameters
323 	 */
324 	igb_init_properties(igb);
325 	igb->attach_progress |= ATTACH_PROGRESS_PROPS;
326 
327 	/*
328 	 * Allocate interrupts
329 	 */
330 	if (igb_alloc_intrs(igb) != IGB_SUCCESS) {
331 		igb_error(igb, "Failed to allocate interrupts");
332 		goto attach_fail;
333 	}
334 	igb->attach_progress |= ATTACH_PROGRESS_ALLOC_INTR;
335 
336 	/*
337 	 * Allocate rx/tx rings based on the ring numbers.
338 	 * The actual numbers of rx/tx rings are decided by the number of
339 	 * allocated interrupt vectors, so we should allocate the rings after
340 	 * interrupts are allocated.
341 	 */
342 	if (igb_alloc_rings(igb) != IGB_SUCCESS) {
343 		igb_error(igb, "Failed to allocate rx/tx rings or groups");
344 		goto attach_fail;
345 	}
346 	igb->attach_progress |= ATTACH_PROGRESS_ALLOC_RINGS;
347 
348 	/*
349 	 * Add interrupt handlers
350 	 */
351 	if (igb_add_intr_handlers(igb) != IGB_SUCCESS) {
352 		igb_error(igb, "Failed to add interrupt handlers");
353 		goto attach_fail;
354 	}
355 	igb->attach_progress |= ATTACH_PROGRESS_ADD_INTR;
356 
357 	/*
358 	 * Initialize driver parameters
359 	 */
360 	if (igb_init_driver_settings(igb) != IGB_SUCCESS) {
361 		igb_error(igb, "Failed to initialize driver settings");
362 		goto attach_fail;
363 	}
364 
365 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK) {
366 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
367 		goto attach_fail;
368 	}
369 
370 	/*
371 	 * Initialize mutexes for this device.
372 	 * Do this before enabling the interrupt handler and
373 	 * register the softint to avoid the condition where
374 	 * interrupt handler can try using uninitialized mutex
375 	 */
376 	igb_init_locks(igb);
377 	igb->attach_progress |= ATTACH_PROGRESS_LOCKS;
378 
379 	/*
380 	 * Initialize chipset hardware
381 	 */
382 	mutex_enter(&igb->gen_lock);
383 	if (igb_init(igb) != IGB_SUCCESS) {
384 		mutex_exit(&igb->gen_lock);
385 		igb_error(igb, "Failed to initialize adapter");
386 		goto attach_fail;
387 	}
388 	mutex_exit(&igb->gen_lock);
389 	igb->attach_progress |= ATTACH_PROGRESS_INIT;
390 
391 	/*
392 	 * Initialize DMA and hardware settings for rx/tx rings
393 	 */
394 	if (igb_init_rings(igb) != IGB_SUCCESS) {
395 		igb_error(igb, "Failed to initialize rings");
396 		goto attach_fail;
397 	}
398 	igb->attach_progress |= ATTACH_PROGRESS_INIT_RINGS;
399 
400 	/*
401 	 * Initialize statistics
402 	 */
403 	if (igb_init_stats(igb) != IGB_SUCCESS) {
404 		igb_error(igb, "Failed to initialize statistics");
405 		goto attach_fail;
406 	}
407 	igb->attach_progress |= ATTACH_PROGRESS_STATS;
408 
409 	/*
410 	 * Initialize NDD parameters
411 	 */
412 	if (igb_nd_init(igb) != IGB_SUCCESS) {
413 		igb_error(igb, "Failed to initialize ndd");
414 		goto attach_fail;
415 	}
416 	igb->attach_progress |= ATTACH_PROGRESS_NDD;
417 
418 	/*
419 	 * Register the driver to the MAC
420 	 */
421 	if (igb_register_mac(igb) != IGB_SUCCESS) {
422 		igb_error(igb, "Failed to register MAC");
423 		goto attach_fail;
424 	}
425 	igb->attach_progress |= ATTACH_PROGRESS_MAC;
426 
427 	/*
428 	 * Now that mutex locks are initialized, and the chip is also
429 	 * initialized, enable interrupts.
430 	 */
431 	if (igb_enable_intrs(igb) != IGB_SUCCESS) {
432 		igb_error(igb, "Failed to enable DDI interrupts");
433 		goto attach_fail;
434 	}
435 	igb->attach_progress |= ATTACH_PROGRESS_ENABLE_INTR;
436 
437 	igb->igb_state |= IGB_INITIALIZED;
438 
439 	return (DDI_SUCCESS);
440 
441 attach_fail:
442 	igb_unconfigure(devinfo, igb);
443 	return (DDI_FAILURE);
444 }
445 
446 /*
447  * igb_detach - driver detach
448  *
449  * The detach() function is the complement of the attach routine.
450  * If cmd is set to DDI_DETACH, detach() is used to remove  the
451  * state  associated  with  a  given  instance of a device node
452  * prior to the removal of that instance from the system.
453  *
454  * The detach() function will be called once for each  instance
455  * of the device for which there has been a successful attach()
456  * once there are no longer  any  opens  on  the  device.
457  *
458  * Interrupts routine are disabled, All memory allocated by this
459  * driver are freed.
460  */
461 static int
462 igb_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
463 {
464 	igb_t *igb;
465 
466 	/*
467 	 * Check detach command
468 	 */
469 	switch (cmd) {
470 	default:
471 		return (DDI_FAILURE);
472 
473 	case DDI_SUSPEND:
474 		return (igb_suspend(devinfo));
475 
476 	case DDI_DETACH:
477 		break;
478 	}
479 
480 
481 	/*
482 	 * Get the pointer to the driver private data structure
483 	 */
484 	igb = (igb_t *)ddi_get_driver_private(devinfo);
485 	if (igb == NULL)
486 		return (DDI_FAILURE);
487 
488 	/*
489 	 * Unregister MAC. If failed, we have to fail the detach
490 	 */
491 	if (mac_unregister(igb->mac_hdl) != 0) {
492 		igb_error(igb, "Failed to unregister MAC");
493 		return (DDI_FAILURE);
494 	}
495 	igb->attach_progress &= ~ATTACH_PROGRESS_MAC;
496 
497 	/*
498 	 * If the device is still running, it needs to be stopped first.
499 	 * This check is necessary because under some specific circumstances,
500 	 * the detach routine can be called without stopping the interface
501 	 * first.
502 	 */
503 	mutex_enter(&igb->gen_lock);
504 	if (igb->igb_state & IGB_STARTED) {
505 		igb->igb_state &= ~IGB_STARTED;
506 		igb_stop(igb);
507 		mutex_exit(&igb->gen_lock);
508 		/* Disable and stop the watchdog timer */
509 		igb_disable_watchdog_timer(igb);
510 	} else
511 		mutex_exit(&igb->gen_lock);
512 
513 	/*
514 	 * Check if there are still rx buffers held by the upper layer.
515 	 * If so, fail the detach.
516 	 */
517 	if (!igb_rx_drain(igb))
518 		return (DDI_FAILURE);
519 
520 	/*
521 	 * Do the remaining unconfigure routines
522 	 */
523 	igb_unconfigure(devinfo, igb);
524 
525 	return (DDI_SUCCESS);
526 }
527 
528 /*
529  * quiesce(9E) entry point.
530  *
531  * This function is called when the system is single-threaded at high
532  * PIL with preemption disabled. Therefore, this function must not be
533  * blocked.
534  *
535  * This function returns DDI_SUCCESS on success, or DDI_FAILURE on failure.
536  * DDI_FAILURE indicates an error condition and should almost never happen.
537  */
538 static int
539 igb_quiesce(dev_info_t *devinfo)
540 {
541 	igb_t *igb;
542 	struct e1000_hw *hw;
543 
544 	igb = (igb_t *)ddi_get_driver_private(devinfo);
545 
546 	if (igb == NULL)
547 		return (DDI_FAILURE);
548 
549 	hw = &igb->hw;
550 
551 	/*
552 	 * Disable the adapter interrupts
553 	 */
554 	igb_disable_adapter_interrupts(igb);
555 
556 	/* Tell firmware driver is no longer in control */
557 	igb_release_driver_control(hw);
558 
559 	/*
560 	 * Reset the chipset
561 	 */
562 	(void) e1000_reset_hw(hw);
563 
564 	/*
565 	 * Reset PHY if possible
566 	 */
567 	if (e1000_check_reset_block(hw) == E1000_SUCCESS)
568 		(void) e1000_phy_hw_reset(hw);
569 
570 	return (DDI_SUCCESS);
571 }
572 
573 
574 static void
575 igb_unconfigure(dev_info_t *devinfo, igb_t *igb)
576 {
577 	/*
578 	 * Disable interrupt
579 	 */
580 	if (igb->attach_progress & ATTACH_PROGRESS_ENABLE_INTR) {
581 		(void) igb_disable_intrs(igb);
582 	}
583 
584 	/*
585 	 * Unregister MAC
586 	 */
587 	if (igb->attach_progress & ATTACH_PROGRESS_MAC) {
588 		(void) mac_unregister(igb->mac_hdl);
589 	}
590 
591 	/*
592 	 * Free ndd parameters
593 	 */
594 	if (igb->attach_progress & ATTACH_PROGRESS_NDD) {
595 		igb_nd_cleanup(igb);
596 	}
597 
598 	/*
599 	 * Free statistics
600 	 */
601 	if (igb->attach_progress & ATTACH_PROGRESS_STATS) {
602 		kstat_delete((kstat_t *)igb->igb_ks);
603 	}
604 
605 	/*
606 	 * Remove interrupt handlers
607 	 */
608 	if (igb->attach_progress & ATTACH_PROGRESS_ADD_INTR) {
609 		igb_rem_intr_handlers(igb);
610 	}
611 
612 	/*
613 	 * Remove interrupts
614 	 */
615 	if (igb->attach_progress & ATTACH_PROGRESS_ALLOC_INTR) {
616 		igb_rem_intrs(igb);
617 	}
618 
619 	/*
620 	 * Remove driver properties
621 	 */
622 	if (igb->attach_progress & ATTACH_PROGRESS_PROPS) {
623 		(void) ddi_prop_remove_all(devinfo);
624 	}
625 
626 	/*
627 	 * Release the DMA resources of rx/tx rings
628 	 */
629 	if (igb->attach_progress & ATTACH_PROGRESS_INIT_RINGS) {
630 		igb_fini_rings(igb);
631 	}
632 
633 	/*
634 	 * Stop the chipset
635 	 */
636 	if (igb->attach_progress & ATTACH_PROGRESS_INIT) {
637 		mutex_enter(&igb->gen_lock);
638 		igb_chip_stop(igb);
639 		mutex_exit(&igb->gen_lock);
640 		if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
641 			ddi_fm_service_impact(igb->dip, DDI_SERVICE_UNAFFECTED);
642 	}
643 
644 	/*
645 	 * Free register handle
646 	 */
647 	if (igb->attach_progress & ATTACH_PROGRESS_REGS_MAP) {
648 		if (igb->osdep.reg_handle != NULL)
649 			ddi_regs_map_free(&igb->osdep.reg_handle);
650 	}
651 
652 	/*
653 	 * Free PCI config handle
654 	 */
655 	if (igb->attach_progress & ATTACH_PROGRESS_PCI_CONFIG) {
656 		if (igb->osdep.cfg_handle != NULL)
657 			pci_config_teardown(&igb->osdep.cfg_handle);
658 	}
659 
660 	/*
661 	 * Free locks
662 	 */
663 	if (igb->attach_progress & ATTACH_PROGRESS_LOCKS) {
664 		igb_destroy_locks(igb);
665 	}
666 
667 	/*
668 	 * Free the rx/tx rings
669 	 */
670 	if (igb->attach_progress & ATTACH_PROGRESS_ALLOC_RINGS) {
671 		igb_free_rings(igb);
672 	}
673 
674 	/*
675 	 * Remove FMA
676 	 */
677 	if (igb->attach_progress & ATTACH_PROGRESS_FMINIT) {
678 		igb_fm_fini(igb);
679 	}
680 
681 	/*
682 	 * Free device specific structure
683 	 */
684 	e1000_remove_device(&igb->hw);
685 
686 	/*
687 	 * Free the driver data structure
688 	 */
689 	kmem_free(igb, sizeof (igb_t));
690 
691 	ddi_set_driver_private(devinfo, NULL);
692 }
693 
694 /*
695  * igb_register_mac - Register the driver and its function pointers with
696  * the GLD interface
697  */
698 static int
699 igb_register_mac(igb_t *igb)
700 {
701 	struct e1000_hw *hw = &igb->hw;
702 	mac_register_t *mac;
703 	int status;
704 
705 	if ((mac = mac_alloc(MAC_VERSION)) == NULL)
706 		return (IGB_FAILURE);
707 
708 	mac->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
709 	mac->m_driver = igb;
710 	mac->m_dip = igb->dip;
711 	mac->m_src_addr = hw->mac.addr;
712 	mac->m_callbacks = &igb_m_callbacks;
713 	mac->m_min_sdu = 0;
714 	mac->m_max_sdu = igb->max_frame_size -
715 	    sizeof (struct ether_vlan_header) - ETHERFCSL;
716 	mac->m_margin = VLAN_TAGSZ;
717 	mac->m_v12n = MAC_VIRT_LEVEL1;
718 
719 	status = mac_register(mac, &igb->mac_hdl);
720 
721 	mac_free(mac);
722 
723 	return ((status == 0) ? IGB_SUCCESS : IGB_FAILURE);
724 }
725 
726 /*
727  * igb_identify_hardware - Identify the type of the chipset
728  */
729 static int
730 igb_identify_hardware(igb_t *igb)
731 {
732 	struct e1000_hw *hw = &igb->hw;
733 	struct igb_osdep *osdep = &igb->osdep;
734 
735 	/*
736 	 * Get the device id
737 	 */
738 	hw->vendor_id =
739 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_VENID);
740 	hw->device_id =
741 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_DEVID);
742 	hw->revision_id =
743 	    pci_config_get8(osdep->cfg_handle, PCI_CONF_REVID);
744 	hw->subsystem_device_id =
745 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBSYSID);
746 	hw->subsystem_vendor_id =
747 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBVENID);
748 
749 	/*
750 	 * Set the mac type of the adapter based on the device id
751 	 */
752 	if (e1000_set_mac_type(hw) != E1000_SUCCESS) {
753 		return (IGB_FAILURE);
754 	}
755 
756 	return (IGB_SUCCESS);
757 }
758 
759 /*
760  * igb_regs_map - Map the device registers
761  */
762 static int
763 igb_regs_map(igb_t *igb)
764 {
765 	dev_info_t *devinfo = igb->dip;
766 	struct e1000_hw *hw = &igb->hw;
767 	struct igb_osdep *osdep = &igb->osdep;
768 	off_t mem_size;
769 
770 	/*
771 	 * First get the size of device registers to be mapped.
772 	 */
773 	if (ddi_dev_regsize(devinfo, 1, &mem_size) != DDI_SUCCESS) {
774 		return (IGB_FAILURE);
775 	}
776 
777 	/*
778 	 * Call ddi_regs_map_setup() to map registers
779 	 */
780 	if ((ddi_regs_map_setup(devinfo, 1,
781 	    (caddr_t *)&hw->hw_addr, 0,
782 	    mem_size, &igb_regs_acc_attr,
783 	    &osdep->reg_handle)) != DDI_SUCCESS) {
784 		return (IGB_FAILURE);
785 	}
786 
787 	return (IGB_SUCCESS);
788 }
789 
790 /*
791  * igb_init_properties - Initialize driver properties
792  */
793 static void
794 igb_init_properties(igb_t *igb)
795 {
796 	/*
797 	 * Get conf file properties, including link settings
798 	 * jumbo frames, ring number, descriptor number, etc.
799 	 */
800 	igb_get_conf(igb);
801 }
802 
803 /*
804  * igb_init_driver_settings - Initialize driver settings
805  *
806  * The settings include hardware function pointers, bus information,
807  * rx/tx rings settings, link state, and any other parameters that
808  * need to be setup during driver initialization.
809  */
810 static int
811 igb_init_driver_settings(igb_t *igb)
812 {
813 	struct e1000_hw *hw = &igb->hw;
814 	igb_rx_ring_t *rx_ring;
815 	igb_tx_ring_t *tx_ring;
816 	uint32_t rx_size;
817 	uint32_t tx_size;
818 	int i;
819 
820 	/*
821 	 * Initialize chipset specific hardware function pointers
822 	 */
823 	if (e1000_setup_init_funcs(hw, B_TRUE) != E1000_SUCCESS) {
824 		return (IGB_FAILURE);
825 	}
826 
827 	/*
828 	 * Get bus information
829 	 */
830 	if (e1000_get_bus_info(hw) != E1000_SUCCESS) {
831 		return (IGB_FAILURE);
832 	}
833 
834 	/*
835 	 * Set rx buffer size
836 	 * The IP header alignment room is counted in the calculation.
837 	 * The rx buffer size is in unit of 1K that is required by the
838 	 * chipset hardware.
839 	 */
840 	rx_size = igb->max_frame_size + IPHDR_ALIGN_ROOM;
841 	igb->rx_buf_size = ((rx_size >> 10) +
842 	    ((rx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
843 
844 	/*
845 	 * Set tx buffer size
846 	 */
847 	tx_size = igb->max_frame_size;
848 	igb->tx_buf_size = ((tx_size >> 10) +
849 	    ((tx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
850 
851 	/*
852 	 * Initialize rx/tx rings parameters
853 	 */
854 	for (i = 0; i < igb->num_rx_rings; i++) {
855 		rx_ring = &igb->rx_rings[i];
856 		rx_ring->index = i;
857 		rx_ring->igb = igb;
858 
859 		rx_ring->ring_size = igb->rx_ring_size;
860 		rx_ring->free_list_size = igb->rx_ring_size;
861 		rx_ring->copy_thresh = igb->rx_copy_thresh;
862 		rx_ring->limit_per_intr = igb->rx_limit_per_intr;
863 	}
864 
865 	for (i = 0; i < igb->num_tx_rings; i++) {
866 		tx_ring = &igb->tx_rings[i];
867 		tx_ring->index = i;
868 		tx_ring->igb = igb;
869 		if (igb->tx_head_wb_enable)
870 			tx_ring->tx_recycle = igb_tx_recycle_head_wb;
871 		else
872 			tx_ring->tx_recycle = igb_tx_recycle_legacy;
873 
874 		tx_ring->ring_size = igb->tx_ring_size;
875 		tx_ring->free_list_size = igb->tx_ring_size +
876 		    (igb->tx_ring_size >> 1);
877 		tx_ring->copy_thresh = igb->tx_copy_thresh;
878 		tx_ring->recycle_thresh = igb->tx_recycle_thresh;
879 		tx_ring->overload_thresh = igb->tx_overload_thresh;
880 		tx_ring->resched_thresh = igb->tx_resched_thresh;
881 	}
882 
883 	/*
884 	 * Initialize values of interrupt throttling rate
885 	 */
886 	for (i = 1; i < MAX_NUM_EITR; i++)
887 		igb->intr_throttling[i] = igb->intr_throttling[0];
888 
889 	/*
890 	 * The initial link state should be "unknown"
891 	 */
892 	igb->link_state = LINK_STATE_UNKNOWN;
893 
894 	return (IGB_SUCCESS);
895 }
896 
897 /*
898  * igb_init_locks - Initialize locks
899  */
900 static void
901 igb_init_locks(igb_t *igb)
902 {
903 	igb_rx_ring_t *rx_ring;
904 	igb_tx_ring_t *tx_ring;
905 	int i;
906 
907 	for (i = 0; i < igb->num_rx_rings; i++) {
908 		rx_ring = &igb->rx_rings[i];
909 		mutex_init(&rx_ring->rx_lock, NULL,
910 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
911 		mutex_init(&rx_ring->recycle_lock, NULL,
912 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
913 	}
914 
915 	for (i = 0; i < igb->num_tx_rings; i++) {
916 		tx_ring = &igb->tx_rings[i];
917 		mutex_init(&tx_ring->tx_lock, NULL,
918 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
919 		mutex_init(&tx_ring->recycle_lock, NULL,
920 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
921 		mutex_init(&tx_ring->tcb_head_lock, NULL,
922 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
923 		mutex_init(&tx_ring->tcb_tail_lock, NULL,
924 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
925 	}
926 
927 	mutex_init(&igb->gen_lock, NULL,
928 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
929 
930 	mutex_init(&igb->watchdog_lock, NULL,
931 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
932 }
933 
934 /*
935  * igb_destroy_locks - Destroy locks
936  */
937 static void
938 igb_destroy_locks(igb_t *igb)
939 {
940 	igb_rx_ring_t *rx_ring;
941 	igb_tx_ring_t *tx_ring;
942 	int i;
943 
944 	for (i = 0; i < igb->num_rx_rings; i++) {
945 		rx_ring = &igb->rx_rings[i];
946 		mutex_destroy(&rx_ring->rx_lock);
947 		mutex_destroy(&rx_ring->recycle_lock);
948 	}
949 
950 	for (i = 0; i < igb->num_tx_rings; i++) {
951 		tx_ring = &igb->tx_rings[i];
952 		mutex_destroy(&tx_ring->tx_lock);
953 		mutex_destroy(&tx_ring->recycle_lock);
954 		mutex_destroy(&tx_ring->tcb_head_lock);
955 		mutex_destroy(&tx_ring->tcb_tail_lock);
956 	}
957 
958 	mutex_destroy(&igb->gen_lock);
959 	mutex_destroy(&igb->watchdog_lock);
960 }
961 
962 static int
963 igb_resume(dev_info_t *devinfo)
964 {
965 	igb_t *igb;
966 
967 	igb = (igb_t *)ddi_get_driver_private(devinfo);
968 	if (igb == NULL)
969 		return (DDI_FAILURE);
970 
971 	mutex_enter(&igb->gen_lock);
972 
973 	if (igb->igb_state & IGB_STARTED) {
974 		if (igb_start(igb) != IGB_SUCCESS) {
975 			mutex_exit(&igb->gen_lock);
976 			return (DDI_FAILURE);
977 		}
978 
979 		/*
980 		 * Enable and start the watchdog timer
981 		 */
982 		igb_enable_watchdog_timer(igb);
983 	}
984 
985 	igb->igb_state &= ~IGB_SUSPENDED;
986 
987 	mutex_exit(&igb->gen_lock);
988 
989 	return (DDI_SUCCESS);
990 }
991 
992 static int
993 igb_suspend(dev_info_t *devinfo)
994 {
995 	igb_t *igb;
996 
997 	igb = (igb_t *)ddi_get_driver_private(devinfo);
998 	if (igb == NULL)
999 		return (DDI_FAILURE);
1000 
1001 	mutex_enter(&igb->gen_lock);
1002 
1003 	igb->igb_state |= IGB_SUSPENDED;
1004 
1005 	igb_stop(igb);
1006 
1007 	mutex_exit(&igb->gen_lock);
1008 
1009 	/*
1010 	 * Disable and stop the watchdog timer
1011 	 */
1012 	igb_disable_watchdog_timer(igb);
1013 
1014 	return (DDI_SUCCESS);
1015 }
1016 
1017 /*
1018  * igb_init - Initialize the device
1019  */
1020 static int
1021 igb_init(igb_t *igb)
1022 {
1023 	struct e1000_hw *hw = &igb->hw;
1024 	uint32_t pba;
1025 	uint32_t high_water;
1026 
1027 	ASSERT(mutex_owned(&igb->gen_lock));
1028 
1029 	/*
1030 	 * Reset chipset to put the hardware in a known state
1031 	 * before we try to do anything with the eeprom
1032 	 */
1033 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1034 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1035 		goto init_fail;
1036 	}
1037 
1038 	/*
1039 	 * NVM validation
1040 	 */
1041 	if (e1000_validate_nvm_checksum(hw) < 0) {
1042 		/*
1043 		 * Some PCI-E parts fail the first check due to
1044 		 * the link being in sleep state.  Call it again,
1045 		 * if it fails a second time its a real issue.
1046 		 */
1047 		if (e1000_validate_nvm_checksum(hw) < 0) {
1048 			igb_error(igb,
1049 			    "Invalid NVM checksum. Please contact "
1050 			    "the vendor to update the NVM.");
1051 			igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1052 			goto init_fail;
1053 		}
1054 	}
1055 
1056 	/*
1057 	 * Set the FIFO size
1058 	 */
1059 	pba = E1000_PBA_32K;	/* 32K for Rx, 16K for Tx */
1060 	E1000_WRITE_REG(hw, E1000_PBA, pba);
1061 
1062 	/*
1063 	 * Setup flow control
1064 	 *
1065 	 * These parameters set thresholds for the adapter's generation(Tx)
1066 	 * and response(Rx) to Ethernet PAUSE frames.  These are just threshold
1067 	 * settings.  Flow control is enabled or disabled in the configuration
1068 	 * file.
1069 	 * High-water mark is set down from the top of the rx fifo (not
1070 	 * sensitive to max_frame_size) and low-water is set just below
1071 	 * high-water mark.
1072 	 * The high water mark must be low enough to fit one full frame above
1073 	 * it in the rx FIFO.  Should be the lower of:
1074 	 * 90% of the Rx FIFO size, or the full Rx FIFO size minus one full
1075 	 * frame.
1076 	 */
1077 	high_water = min(((pba << 10) * 9 / 10),
1078 	    ((pba << 10) - igb->max_frame_size));
1079 
1080 	hw->fc.high_water = high_water & 0xFFF8;
1081 	hw->fc.low_water = hw->fc.high_water - 8;
1082 	hw->fc.pause_time = E1000_FC_PAUSE_TIME;
1083 	hw->fc.send_xon = B_TRUE;
1084 
1085 	/*
1086 	 * Reset the chipset hardware the second time to validate
1087 	 * the PBA setting.
1088 	 */
1089 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1090 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1091 		goto init_fail;
1092 	}
1093 
1094 	/*
1095 	 * Don't wait for auto-negotiation to complete
1096 	 */
1097 	hw->phy.autoneg_wait_to_complete = B_FALSE;
1098 
1099 	/*
1100 	 * Copper options
1101 	 */
1102 	if (hw->phy.media_type == e1000_media_type_copper) {
1103 		hw->phy.mdix = 0;	/* AUTO_ALL_MODES */
1104 		hw->phy.disable_polarity_correction = B_FALSE;
1105 		hw->phy.ms_type = e1000_ms_hw_default; /* E1000_MASTER_SLAVE */
1106 	}
1107 
1108 	/*
1109 	 * Initialize link settings
1110 	 */
1111 	(void) igb_setup_link(igb, B_FALSE);
1112 
1113 	/*
1114 	 * Initialize the chipset hardware
1115 	 */
1116 	if (igb_chip_start(igb) != IGB_SUCCESS) {
1117 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1118 		goto init_fail;
1119 	}
1120 
1121 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK) {
1122 		goto init_fail;
1123 	}
1124 
1125 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
1126 		goto init_fail;
1127 	}
1128 
1129 	return (IGB_SUCCESS);
1130 
1131 init_fail:
1132 	/*
1133 	 * Reset PHY if possible
1134 	 */
1135 	if (e1000_check_reset_block(hw) == E1000_SUCCESS)
1136 		(void) e1000_phy_hw_reset(hw);
1137 
1138 	ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1139 
1140 	return (IGB_FAILURE);
1141 }
1142 
1143 /*
1144  * igb_init_rings - Allocate DMA resources for all rx/tx rings and
1145  * initialize relevant hardware settings.
1146  */
1147 static int
1148 igb_init_rings(igb_t *igb)
1149 {
1150 	int i;
1151 
1152 	/*
1153 	 * Allocate buffers for all the rx/tx rings
1154 	 */
1155 	if (igb_alloc_dma(igb) != IGB_SUCCESS)
1156 		return (IGB_FAILURE);
1157 
1158 	/*
1159 	 * Setup the rx/tx rings
1160 	 */
1161 	mutex_enter(&igb->gen_lock);
1162 
1163 	for (i = 0; i < igb->num_rx_rings; i++)
1164 		mutex_enter(&igb->rx_rings[i].rx_lock);
1165 	for (i = 0; i < igb->num_tx_rings; i++)
1166 		mutex_enter(&igb->tx_rings[i].tx_lock);
1167 
1168 	igb_setup_rings(igb);
1169 
1170 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1171 		mutex_exit(&igb->tx_rings[i].tx_lock);
1172 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1173 		mutex_exit(&igb->rx_rings[i].rx_lock);
1174 
1175 	mutex_exit(&igb->gen_lock);
1176 
1177 	return (IGB_SUCCESS);
1178 }
1179 
1180 /*
1181  * igb_fini_rings - Release DMA resources of all rx/tx rings
1182  */
1183 static void
1184 igb_fini_rings(igb_t *igb)
1185 {
1186 	/*
1187 	 * Release the DMA/memory resources of rx/tx rings
1188 	 */
1189 	igb_free_dma(igb);
1190 }
1191 
1192 /*
1193  * igb_chip_start - Initialize and start the chipset hardware
1194  */
1195 static int
1196 igb_chip_start(igb_t *igb)
1197 {
1198 	struct e1000_hw *hw = &igb->hw;
1199 	int i;
1200 
1201 	ASSERT(mutex_owned(&igb->gen_lock));
1202 
1203 	/*
1204 	 * Get the mac address
1205 	 * This function should handle SPARC case correctly.
1206 	 */
1207 	if (!igb_find_mac_address(igb)) {
1208 		igb_error(igb, "Failed to get the mac address");
1209 		return (IGB_FAILURE);
1210 	}
1211 
1212 	/* Validate mac address */
1213 	if (!is_valid_mac_addr(hw->mac.addr)) {
1214 		igb_error(igb, "Invalid mac address");
1215 		return (IGB_FAILURE);
1216 	}
1217 
1218 	/* Disable wakeup control by default */
1219 	E1000_WRITE_REG(hw, E1000_WUC, 0);
1220 
1221 	/*
1222 	 * Configure/Initialize hardware
1223 	 */
1224 	if (e1000_init_hw(hw) != E1000_SUCCESS) {
1225 		igb_error(igb, "Failed to initialize hardware");
1226 		return (IGB_FAILURE);
1227 	}
1228 
1229 	/*
1230 	 * Make sure driver has control
1231 	 */
1232 	igb_get_driver_control(hw);
1233 
1234 	/*
1235 	 * Setup MSI-X interrupts
1236 	 */
1237 	if (igb->intr_type == DDI_INTR_TYPE_MSIX)
1238 		igb_setup_adapter_msix(igb);
1239 
1240 	/*
1241 	 * Initialize unicast addresses.
1242 	 */
1243 	igb_init_unicst(igb);
1244 
1245 	/*
1246 	 * Setup and initialize the mctable structures.
1247 	 */
1248 	igb_setup_multicst(igb);
1249 
1250 	/*
1251 	 * Set interrupt throttling rate
1252 	 */
1253 	for (i = 0; i < igb->intr_cnt; i++)
1254 		E1000_WRITE_REG(hw, E1000_EITR(i), igb->intr_throttling[i]);
1255 
1256 	/* Enable PCI-E master */
1257 	if (hw->bus.type == e1000_bus_type_pci_express) {
1258 		e1000_enable_pciex_master(hw);
1259 	}
1260 
1261 	/*
1262 	 * Save the state of the phy
1263 	 */
1264 	igb_get_phy_state(igb);
1265 
1266 	return (IGB_SUCCESS);
1267 }
1268 
1269 /*
1270  * igb_chip_stop - Stop the chipset hardware
1271  */
1272 static void
1273 igb_chip_stop(igb_t *igb)
1274 {
1275 	struct e1000_hw *hw = &igb->hw;
1276 
1277 	ASSERT(mutex_owned(&igb->gen_lock));
1278 
1279 	/* Tell firmware driver is no longer in control */
1280 	igb_release_driver_control(hw);
1281 
1282 	/*
1283 	 * Reset the chipset
1284 	 */
1285 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1286 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1287 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1288 	}
1289 
1290 	/*
1291 	 * Reset PHY if possible
1292 	 */
1293 	if (e1000_check_reset_block(hw) == E1000_SUCCESS)
1294 		(void) e1000_phy_hw_reset(hw);
1295 }
1296 
1297 /*
1298  * igb_reset - Reset the chipset and restart the driver.
1299  *
1300  * It involves stopping and re-starting the chipset,
1301  * and re-configuring the rx/tx rings.
1302  */
1303 static int
1304 igb_reset(igb_t *igb)
1305 {
1306 	int i;
1307 
1308 	mutex_enter(&igb->gen_lock);
1309 
1310 	ASSERT(igb->igb_state & IGB_STARTED);
1311 
1312 	/*
1313 	 * Disable the adapter interrupts to stop any rx/tx activities
1314 	 * before draining pending data and resetting hardware.
1315 	 */
1316 	igb_disable_adapter_interrupts(igb);
1317 
1318 	/*
1319 	 * Drain the pending transmit packets
1320 	 */
1321 	(void) igb_tx_drain(igb);
1322 
1323 	for (i = 0; i < igb->num_rx_rings; i++)
1324 		mutex_enter(&igb->rx_rings[i].rx_lock);
1325 	for (i = 0; i < igb->num_tx_rings; i++)
1326 		mutex_enter(&igb->tx_rings[i].tx_lock);
1327 
1328 	/*
1329 	 * Stop the chipset hardware
1330 	 */
1331 	igb_chip_stop(igb);
1332 
1333 	/*
1334 	 * Clean the pending tx data/resources
1335 	 */
1336 	igb_tx_clean(igb);
1337 
1338 	/*
1339 	 * Start the chipset hardware
1340 	 */
1341 	if (igb_chip_start(igb) != IGB_SUCCESS) {
1342 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1343 		goto reset_failure;
1344 	}
1345 
1346 	/*
1347 	 * Setup the rx/tx rings
1348 	 */
1349 	igb_setup_rings(igb);
1350 
1351 	/*
1352 	 * Enable adapter interrupts
1353 	 * The interrupts must be enabled after the driver state is START
1354 	 */
1355 	igb_enable_adapter_interrupts(igb);
1356 
1357 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK)
1358 		goto reset_failure;
1359 
1360 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1361 		goto reset_failure;
1362 
1363 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1364 		mutex_exit(&igb->tx_rings[i].tx_lock);
1365 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1366 		mutex_exit(&igb->rx_rings[i].rx_lock);
1367 
1368 	mutex_exit(&igb->gen_lock);
1369 
1370 	return (IGB_SUCCESS);
1371 
1372 reset_failure:
1373 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1374 		mutex_exit(&igb->tx_rings[i].tx_lock);
1375 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1376 		mutex_exit(&igb->rx_rings[i].rx_lock);
1377 
1378 	mutex_exit(&igb->gen_lock);
1379 
1380 	ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1381 
1382 	return (IGB_FAILURE);
1383 }
1384 
1385 /*
1386  * igb_tx_clean - Clean the pending transmit packets and DMA resources
1387  */
1388 static void
1389 igb_tx_clean(igb_t *igb)
1390 {
1391 	igb_tx_ring_t *tx_ring;
1392 	tx_control_block_t *tcb;
1393 	link_list_t pending_list;
1394 	uint32_t desc_num;
1395 	int i, j;
1396 
1397 	LINK_LIST_INIT(&pending_list);
1398 
1399 	for (i = 0; i < igb->num_tx_rings; i++) {
1400 		tx_ring = &igb->tx_rings[i];
1401 
1402 		mutex_enter(&tx_ring->recycle_lock);
1403 
1404 		/*
1405 		 * Clean the pending tx data - the pending packets in the
1406 		 * work_list that have no chances to be transmitted again.
1407 		 *
1408 		 * We must ensure the chipset is stopped or the link is down
1409 		 * before cleaning the transmit packets.
1410 		 */
1411 		desc_num = 0;
1412 		for (j = 0; j < tx_ring->ring_size; j++) {
1413 			tcb = tx_ring->work_list[j];
1414 			if (tcb != NULL) {
1415 				desc_num += tcb->desc_num;
1416 
1417 				tx_ring->work_list[j] = NULL;
1418 
1419 				igb_free_tcb(tcb);
1420 
1421 				LIST_PUSH_TAIL(&pending_list, &tcb->link);
1422 			}
1423 		}
1424 
1425 		if (desc_num > 0) {
1426 			atomic_add_32(&tx_ring->tbd_free, desc_num);
1427 			ASSERT(tx_ring->tbd_free == tx_ring->ring_size);
1428 
1429 			/*
1430 			 * Reset the head and tail pointers of the tbd ring;
1431 			 * Reset the head write-back if it is enabled.
1432 			 */
1433 			tx_ring->tbd_head = 0;
1434 			tx_ring->tbd_tail = 0;
1435 			if (igb->tx_head_wb_enable)
1436 				*tx_ring->tbd_head_wb = 0;
1437 
1438 			E1000_WRITE_REG(&igb->hw, E1000_TDH(tx_ring->index), 0);
1439 			E1000_WRITE_REG(&igb->hw, E1000_TDT(tx_ring->index), 0);
1440 		}
1441 
1442 		mutex_exit(&tx_ring->recycle_lock);
1443 
1444 		/*
1445 		 * Add the tx control blocks in the pending list to
1446 		 * the free list.
1447 		 */
1448 		igb_put_free_list(tx_ring, &pending_list);
1449 	}
1450 }
1451 
1452 /*
1453  * igb_tx_drain - Drain the tx rings to allow pending packets to be transmitted
1454  */
1455 static boolean_t
1456 igb_tx_drain(igb_t *igb)
1457 {
1458 	igb_tx_ring_t *tx_ring;
1459 	boolean_t done;
1460 	int i, j;
1461 
1462 	/*
1463 	 * Wait for a specific time to allow pending tx packets
1464 	 * to be transmitted.
1465 	 *
1466 	 * Check the counter tbd_free to see if transmission is done.
1467 	 * No lock protection is needed here.
1468 	 *
1469 	 * Return B_TRUE if all pending packets have been transmitted;
1470 	 * Otherwise return B_FALSE;
1471 	 */
1472 	for (i = 0; i < TX_DRAIN_TIME; i++) {
1473 
1474 		done = B_TRUE;
1475 		for (j = 0; j < igb->num_tx_rings; j++) {
1476 			tx_ring = &igb->tx_rings[j];
1477 			done = done &&
1478 			    (tx_ring->tbd_free == tx_ring->ring_size);
1479 		}
1480 
1481 		if (done)
1482 			break;
1483 
1484 		msec_delay(1);
1485 	}
1486 
1487 	return (done);
1488 }
1489 
1490 /*
1491  * igb_rx_drain - Wait for all rx buffers to be released by upper layer
1492  */
1493 static boolean_t
1494 igb_rx_drain(igb_t *igb)
1495 {
1496 	igb_rx_ring_t *rx_ring;
1497 	boolean_t done;
1498 	int i, j;
1499 
1500 	/*
1501 	 * Polling the rx free list to check if those rx buffers held by
1502 	 * the upper layer are released.
1503 	 *
1504 	 * Check the counter rcb_free to see if all pending buffers are
1505 	 * released. No lock protection is needed here.
1506 	 *
1507 	 * Return B_TRUE if all pending buffers have been released;
1508 	 * Otherwise return B_FALSE;
1509 	 */
1510 	for (i = 0; i < RX_DRAIN_TIME; i++) {
1511 
1512 		done = B_TRUE;
1513 		for (j = 0; j < igb->num_rx_rings; j++) {
1514 			rx_ring = &igb->rx_rings[j];
1515 			done = done &&
1516 			    (rx_ring->rcb_free == rx_ring->free_list_size);
1517 		}
1518 
1519 		if (done)
1520 			break;
1521 
1522 		msec_delay(1);
1523 	}
1524 
1525 	return (done);
1526 }
1527 
1528 /*
1529  * igb_start - Start the driver/chipset
1530  */
1531 int
1532 igb_start(igb_t *igb)
1533 {
1534 	int i;
1535 
1536 	ASSERT(mutex_owned(&igb->gen_lock));
1537 
1538 	for (i = 0; i < igb->num_rx_rings; i++)
1539 		mutex_enter(&igb->rx_rings[i].rx_lock);
1540 	for (i = 0; i < igb->num_tx_rings; i++)
1541 		mutex_enter(&igb->tx_rings[i].tx_lock);
1542 
1543 	/*
1544 	 * Start the chipset hardware
1545 	 */
1546 	if (!(igb->attach_progress & ATTACH_PROGRESS_INIT)) {
1547 		if (igb_init(igb) != IGB_SUCCESS) {
1548 			igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1549 			goto start_failure;
1550 		}
1551 		igb->attach_progress |= ATTACH_PROGRESS_INIT;
1552 	}
1553 
1554 	/*
1555 	 * Setup the rx/tx rings
1556 	 */
1557 	igb_setup_rings(igb);
1558 
1559 	/*
1560 	 * Enable adapter interrupts
1561 	 * The interrupts must be enabled after the driver state is START
1562 	 */
1563 	igb_enable_adapter_interrupts(igb);
1564 
1565 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK)
1566 		goto start_failure;
1567 
1568 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1569 		goto start_failure;
1570 
1571 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1572 		mutex_exit(&igb->tx_rings[i].tx_lock);
1573 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1574 		mutex_exit(&igb->rx_rings[i].rx_lock);
1575 
1576 	return (IGB_SUCCESS);
1577 
1578 start_failure:
1579 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1580 		mutex_exit(&igb->tx_rings[i].tx_lock);
1581 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1582 		mutex_exit(&igb->rx_rings[i].rx_lock);
1583 
1584 	ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1585 
1586 	return (IGB_FAILURE);
1587 }
1588 
1589 /*
1590  * igb_stop - Stop the driver/chipset
1591  */
1592 void
1593 igb_stop(igb_t *igb)
1594 {
1595 	int i;
1596 
1597 	ASSERT(mutex_owned(&igb->gen_lock));
1598 
1599 	igb->attach_progress &= ~ ATTACH_PROGRESS_INIT;
1600 
1601 	/*
1602 	 * Disable the adapter interrupts
1603 	 */
1604 	igb_disable_adapter_interrupts(igb);
1605 
1606 	/*
1607 	 * Drain the pending tx packets
1608 	 */
1609 	(void) igb_tx_drain(igb);
1610 
1611 	for (i = 0; i < igb->num_rx_rings; i++)
1612 		mutex_enter(&igb->rx_rings[i].rx_lock);
1613 	for (i = 0; i < igb->num_tx_rings; i++)
1614 		mutex_enter(&igb->tx_rings[i].tx_lock);
1615 
1616 	/*
1617 	 * Stop the chipset hardware
1618 	 */
1619 	igb_chip_stop(igb);
1620 
1621 	/*
1622 	 * Clean the pending tx data/resources
1623 	 */
1624 	igb_tx_clean(igb);
1625 
1626 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1627 		mutex_exit(&igb->tx_rings[i].tx_lock);
1628 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1629 		mutex_exit(&igb->rx_rings[i].rx_lock);
1630 
1631 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1632 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1633 }
1634 
1635 /*
1636  * igb_alloc_rings - Allocate memory space for rx/tx rings
1637  */
1638 static int
1639 igb_alloc_rings(igb_t *igb)
1640 {
1641 	/*
1642 	 * Allocate memory space for rx rings
1643 	 */
1644 	igb->rx_rings = kmem_zalloc(
1645 	    sizeof (igb_rx_ring_t) * igb->num_rx_rings,
1646 	    KM_NOSLEEP);
1647 
1648 	if (igb->rx_rings == NULL) {
1649 		return (IGB_FAILURE);
1650 	}
1651 
1652 	/*
1653 	 * Allocate memory space for tx rings
1654 	 */
1655 	igb->tx_rings = kmem_zalloc(
1656 	    sizeof (igb_tx_ring_t) * igb->num_tx_rings,
1657 	    KM_NOSLEEP);
1658 
1659 	if (igb->tx_rings == NULL) {
1660 		kmem_free(igb->rx_rings,
1661 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1662 		igb->rx_rings = NULL;
1663 		return (IGB_FAILURE);
1664 	}
1665 
1666 	/*
1667 	 * Allocate memory space for rx ring groups
1668 	 */
1669 	igb->rx_groups = kmem_zalloc(
1670 	    sizeof (igb_rx_group_t) * igb->num_rx_groups,
1671 	    KM_NOSLEEP);
1672 
1673 	if (igb->rx_groups == NULL) {
1674 		kmem_free(igb->rx_rings,
1675 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1676 		kmem_free(igb->tx_rings,
1677 		    sizeof (igb_tx_ring_t) * igb->num_tx_rings);
1678 		igb->rx_rings = NULL;
1679 		igb->tx_rings = NULL;
1680 		return (IGB_FAILURE);
1681 	}
1682 
1683 	return (IGB_SUCCESS);
1684 }
1685 
1686 /*
1687  * igb_free_rings - Free the memory space of rx/tx rings.
1688  */
1689 static void
1690 igb_free_rings(igb_t *igb)
1691 {
1692 	if (igb->rx_rings != NULL) {
1693 		kmem_free(igb->rx_rings,
1694 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1695 		igb->rx_rings = NULL;
1696 	}
1697 
1698 	if (igb->tx_rings != NULL) {
1699 		kmem_free(igb->tx_rings,
1700 		    sizeof (igb_tx_ring_t) * igb->num_tx_rings);
1701 		igb->tx_rings = NULL;
1702 	}
1703 
1704 	if (igb->rx_groups != NULL) {
1705 		kmem_free(igb->rx_groups,
1706 		    sizeof (igb_rx_group_t) * igb->num_rx_groups);
1707 		igb->rx_groups = NULL;
1708 	}
1709 }
1710 
1711 /*
1712  * igb_setup_rings - Setup rx/tx rings
1713  */
1714 static void
1715 igb_setup_rings(igb_t *igb)
1716 {
1717 	/*
1718 	 * Setup the rx/tx rings, including the following:
1719 	 *
1720 	 * 1. Setup the descriptor ring and the control block buffers;
1721 	 * 2. Initialize necessary registers for receive/transmit;
1722 	 * 3. Initialize software pointers/parameters for receive/transmit;
1723 	 */
1724 	igb_setup_rx(igb);
1725 
1726 	igb_setup_tx(igb);
1727 
1728 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1729 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1730 }
1731 
1732 static void
1733 igb_setup_rx_ring(igb_rx_ring_t *rx_ring)
1734 {
1735 	igb_t *igb = rx_ring->igb;
1736 	struct e1000_hw *hw = &igb->hw;
1737 	rx_control_block_t *rcb;
1738 	union e1000_adv_rx_desc	*rbd;
1739 	uint32_t size;
1740 	uint32_t buf_low;
1741 	uint32_t buf_high;
1742 	uint32_t reg_val;
1743 	int i;
1744 
1745 	ASSERT(mutex_owned(&rx_ring->rx_lock));
1746 	ASSERT(mutex_owned(&igb->gen_lock));
1747 
1748 	for (i = 0; i < igb->rx_ring_size; i++) {
1749 		rcb = rx_ring->work_list[i];
1750 		rbd = &rx_ring->rbd_ring[i];
1751 
1752 		rbd->read.pkt_addr = rcb->rx_buf.dma_address;
1753 		rbd->read.hdr_addr = NULL;
1754 	}
1755 
1756 	/*
1757 	 * Initialize the length register
1758 	 */
1759 	size = rx_ring->ring_size * sizeof (union e1000_adv_rx_desc);
1760 	E1000_WRITE_REG(hw, E1000_RDLEN(rx_ring->index), size);
1761 
1762 	/*
1763 	 * Initialize the base address registers
1764 	 */
1765 	buf_low = (uint32_t)rx_ring->rbd_area.dma_address;
1766 	buf_high = (uint32_t)(rx_ring->rbd_area.dma_address >> 32);
1767 	E1000_WRITE_REG(hw, E1000_RDBAH(rx_ring->index), buf_high);
1768 	E1000_WRITE_REG(hw, E1000_RDBAL(rx_ring->index), buf_low);
1769 
1770 	/*
1771 	 * Setup head & tail pointers
1772 	 */
1773 	E1000_WRITE_REG(hw, E1000_RDT(rx_ring->index), rx_ring->ring_size - 1);
1774 	E1000_WRITE_REG(hw, E1000_RDH(rx_ring->index), 0);
1775 
1776 	rx_ring->rbd_next = 0;
1777 
1778 	/*
1779 	 * Note: Considering the case that the chipset is being reset
1780 	 * and there are still some buffers held by the upper layer,
1781 	 * we should not reset the values of rcb_head, rcb_tail and
1782 	 * rcb_free;
1783 	 */
1784 	if (igb->igb_state == IGB_UNKNOWN) {
1785 		rx_ring->rcb_head = 0;
1786 		rx_ring->rcb_tail = 0;
1787 		rx_ring->rcb_free = rx_ring->free_list_size;
1788 	}
1789 
1790 	/*
1791 	 * Setup the Receive Descriptor Control Register (RXDCTL)
1792 	 */
1793 	reg_val = E1000_READ_REG(hw, E1000_RXDCTL(rx_ring->index));
1794 	reg_val |= E1000_RXDCTL_QUEUE_ENABLE;
1795 	reg_val &= 0xFFF00000;
1796 	reg_val |= 16;		/* pthresh */
1797 	reg_val |= 8 << 8;	/* hthresh */
1798 	reg_val |= 1 << 16;	/* wthresh */
1799 	E1000_WRITE_REG(hw, E1000_RXDCTL(rx_ring->index), reg_val);
1800 
1801 	/*
1802 	 * Setup the Split and Replication Receive Control Register.
1803 	 * Set the rx buffer size and the advanced descriptor type.
1804 	 */
1805 	reg_val = (igb->rx_buf_size >> E1000_SRRCTL_BSIZEPKT_SHIFT) |
1806 	    E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1807 
1808 	E1000_WRITE_REG(hw, E1000_SRRCTL(rx_ring->index), reg_val);
1809 }
1810 
1811 static void
1812 igb_setup_rx(igb_t *igb)
1813 {
1814 	igb_rx_ring_t *rx_ring;
1815 	igb_rx_group_t *rx_group;
1816 	struct e1000_hw *hw = &igb->hw;
1817 	uint32_t reg_val;
1818 	uint32_t ring_per_group;
1819 	int i;
1820 
1821 	/*
1822 	 * Setup the Receive Control Register (RCTL), and ENABLE the
1823 	 * receiver. The initial configuration is to: Enable the receiver,
1824 	 * accept broadcasts, discard bad packets (and long packets),
1825 	 * disable VLAN filter checking, set the receive descriptor
1826 	 * minimum threshold size to 1/2, and the receive buffer size to
1827 	 * 2k.
1828 	 */
1829 	reg_val = E1000_RCTL_EN |	/* Enable Receive Unit */
1830 	    E1000_RCTL_BAM |		/* Accept Broadcast Packets */
1831 	    E1000_RCTL_LPE |		/* Large Packet Enable bit */
1832 	    (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT) |
1833 	    E1000_RCTL_RDMTS_HALF |
1834 	    E1000_RCTL_SECRC |		/* Strip Ethernet CRC */
1835 	    E1000_RCTL_LBM_NO;		/* Loopback Mode = none */
1836 
1837 	E1000_WRITE_REG(hw, E1000_RCTL, reg_val);
1838 
1839 	for (i = 0; i < igb->num_rx_groups; i++) {
1840 		rx_group = &igb->rx_groups[i];
1841 		rx_group->index = i;
1842 		rx_group->igb = igb;
1843 	}
1844 
1845 	/*
1846 	 * igb_setup_rx_ring must be called after configuring RCTL
1847 	 */
1848 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
1849 	for (i = 0; i < igb->num_rx_rings; i++) {
1850 		rx_ring = &igb->rx_rings[i];
1851 		igb_setup_rx_ring(rx_ring);
1852 
1853 		/*
1854 		 * Map a ring to a group by assigning a group index
1855 		 */
1856 		rx_ring->group_index = i / ring_per_group;
1857 	}
1858 
1859 	/*
1860 	 * Setup the Rx Long Packet Max Length register
1861 	 */
1862 	E1000_WRITE_REG(hw, E1000_RLPML, igb->max_frame_size);
1863 
1864 	/*
1865 	 * Hardware checksum settings
1866 	 */
1867 	if (igb->rx_hcksum_enable) {
1868 		reg_val =
1869 		    E1000_RXCSUM_TUOFL |	/* TCP/UDP checksum */
1870 		    E1000_RXCSUM_IPOFL;		/* IP checksum */
1871 
1872 		E1000_WRITE_REG(hw, E1000_RXCSUM, reg_val);
1873 	}
1874 
1875 	/*
1876 	 * Setup classify and RSS for multiple receive queues
1877 	 */
1878 	switch (igb->vmdq_mode) {
1879 	case E1000_VMDQ_OFF:
1880 		/*
1881 		 * One ring group, only RSS is needed when more than
1882 		 * one ring enabled.
1883 		 */
1884 		if (igb->num_rx_rings > 1)
1885 			igb_setup_rss(igb);
1886 		break;
1887 	case E1000_VMDQ_MAC:
1888 		/*
1889 		 * Multiple groups, each group has one ring,
1890 		 * only the MAC classification is needed.
1891 		 */
1892 		igb_setup_mac_classify(igb);
1893 		break;
1894 	case E1000_VMDQ_MAC_RSS:
1895 		/*
1896 		 * Multiple groups and multiple rings, both
1897 		 * MAC classification and RSS are needed.
1898 		 */
1899 		igb_setup_mac_rss_classify(igb);
1900 		break;
1901 	}
1902 }
1903 
1904 static void
1905 igb_setup_tx_ring(igb_tx_ring_t *tx_ring)
1906 {
1907 	igb_t *igb = tx_ring->igb;
1908 	struct e1000_hw *hw = &igb->hw;
1909 	uint32_t size;
1910 	uint32_t buf_low;
1911 	uint32_t buf_high;
1912 	uint32_t reg_val;
1913 
1914 	ASSERT(mutex_owned(&tx_ring->tx_lock));
1915 	ASSERT(mutex_owned(&igb->gen_lock));
1916 
1917 
1918 	/*
1919 	 * Initialize the length register
1920 	 */
1921 	size = tx_ring->ring_size * sizeof (union e1000_adv_tx_desc);
1922 	E1000_WRITE_REG(hw, E1000_TDLEN(tx_ring->index), size);
1923 
1924 	/*
1925 	 * Initialize the base address registers
1926 	 */
1927 	buf_low = (uint32_t)tx_ring->tbd_area.dma_address;
1928 	buf_high = (uint32_t)(tx_ring->tbd_area.dma_address >> 32);
1929 	E1000_WRITE_REG(hw, E1000_TDBAL(tx_ring->index), buf_low);
1930 	E1000_WRITE_REG(hw, E1000_TDBAH(tx_ring->index), buf_high);
1931 
1932 	/*
1933 	 * Setup head & tail pointers
1934 	 */
1935 	E1000_WRITE_REG(hw, E1000_TDH(tx_ring->index), 0);
1936 	E1000_WRITE_REG(hw, E1000_TDT(tx_ring->index), 0);
1937 
1938 	/*
1939 	 * Setup head write-back
1940 	 */
1941 	if (igb->tx_head_wb_enable) {
1942 		/*
1943 		 * The memory of the head write-back is allocated using
1944 		 * the extra tbd beyond the tail of the tbd ring.
1945 		 */
1946 		tx_ring->tbd_head_wb = (uint32_t *)
1947 		    ((uintptr_t)tx_ring->tbd_area.address + size);
1948 		*tx_ring->tbd_head_wb = 0;
1949 
1950 		buf_low = (uint32_t)
1951 		    (tx_ring->tbd_area.dma_address + size);
1952 		buf_high = (uint32_t)
1953 		    ((tx_ring->tbd_area.dma_address + size) >> 32);
1954 
1955 		/* Set the head write-back enable bit */
1956 		buf_low |= E1000_TX_HEAD_WB_ENABLE;
1957 
1958 		E1000_WRITE_REG(hw, E1000_TDWBAL(tx_ring->index), buf_low);
1959 		E1000_WRITE_REG(hw, E1000_TDWBAH(tx_ring->index), buf_high);
1960 
1961 		/*
1962 		 * Turn off relaxed ordering for head write back or it will
1963 		 * cause problems with the tx recycling
1964 		 */
1965 		reg_val = E1000_READ_REG(hw,
1966 		    E1000_DCA_TXCTRL(tx_ring->index));
1967 		reg_val &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1968 		E1000_WRITE_REG(hw,
1969 		    E1000_DCA_TXCTRL(tx_ring->index), reg_val);
1970 	} else {
1971 		tx_ring->tbd_head_wb = NULL;
1972 	}
1973 
1974 	tx_ring->tbd_head = 0;
1975 	tx_ring->tbd_tail = 0;
1976 	tx_ring->tbd_free = tx_ring->ring_size;
1977 
1978 	/*
1979 	 * Note: Considering the case that the chipset is being reset,
1980 	 * and there are still some buffers held by the upper layer,
1981 	 * we should not reset the values of tcb_head, tcb_tail.
1982 	 */
1983 	if (igb->igb_state == IGB_UNKNOWN) {
1984 		tx_ring->tcb_head = 0;
1985 		tx_ring->tcb_tail = 0;
1986 		tx_ring->tcb_free = tx_ring->free_list_size;
1987 	} else {
1988 		ASSERT(tx_ring->tcb_free == tx_ring->free_list_size);
1989 	}
1990 
1991 	/*
1992 	 * Enable specific tx ring, it is required by multiple tx
1993 	 * ring support.
1994 	 */
1995 	reg_val = E1000_READ_REG(hw, E1000_TXDCTL(tx_ring->index));
1996 	reg_val |= E1000_TXDCTL_QUEUE_ENABLE;
1997 	E1000_WRITE_REG(hw, E1000_TXDCTL(tx_ring->index), reg_val);
1998 
1999 	/*
2000 	 * Initialize hardware checksum offload settings
2001 	 */
2002 	tx_ring->hcksum_context.hcksum_flags = 0;
2003 	tx_ring->hcksum_context.ip_hdr_len = 0;
2004 	tx_ring->hcksum_context.mac_hdr_len = 0;
2005 	tx_ring->hcksum_context.l4_proto = 0;
2006 }
2007 
2008 static void
2009 igb_setup_tx(igb_t *igb)
2010 {
2011 	igb_tx_ring_t *tx_ring;
2012 	struct e1000_hw *hw = &igb->hw;
2013 	uint32_t reg_val;
2014 	int i;
2015 
2016 	for (i = 0; i < igb->num_tx_rings; i++) {
2017 		tx_ring = &igb->tx_rings[i];
2018 		igb_setup_tx_ring(tx_ring);
2019 	}
2020 
2021 	/*
2022 	 * Setup the Transmit Control Register (TCTL)
2023 	 */
2024 	reg_val = E1000_TCTL_PSP | E1000_TCTL_EN |
2025 	    (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT) |
2026 	    (E1000_COLLISION_DISTANCE << E1000_COLD_SHIFT) |
2027 	    E1000_TCTL_RTLC;
2028 
2029 	/* Enable the MULR bit */
2030 	if (hw->bus.type == e1000_bus_type_pci_express)
2031 		reg_val |= E1000_TCTL_MULR;
2032 
2033 	E1000_WRITE_REG(hw, E1000_TCTL, reg_val);
2034 
2035 	/*
2036 	 * Set the default values for the Tx Inter Packet Gap timer
2037 	 */
2038 	if (hw->phy.media_type == e1000_media_type_fiber)
2039 		reg_val = DEFAULT_82543_TIPG_IPGT_FIBER;
2040 	else
2041 		reg_val = DEFAULT_82543_TIPG_IPGT_COPPER;
2042 	reg_val |=
2043 	    DEFAULT_82543_TIPG_IPGR1 << E1000_TIPG_IPGR1_SHIFT;
2044 	reg_val |=
2045 	    DEFAULT_82543_TIPG_IPGR2 << E1000_TIPG_IPGR2_SHIFT;
2046 
2047 	E1000_WRITE_REG(hw, E1000_TIPG, reg_val);
2048 }
2049 
2050 /*
2051  * igb_setup_rss - Setup receive-side scaling feature
2052  */
2053 static void
2054 igb_setup_rss(igb_t *igb)
2055 {
2056 	struct e1000_hw *hw = &igb->hw;
2057 	uint32_t i, mrqc, rxcsum;
2058 	int shift;
2059 	uint32_t random;
2060 	union e1000_reta {
2061 		uint32_t	dword;
2062 		uint8_t		bytes[4];
2063 	} reta;
2064 
2065 	/* Setup the Redirection Table */
2066 	shift = 6;
2067 	for (i = 0; i < (32 * 4); i++) {
2068 		reta.bytes[i & 3] = (i % igb->num_rx_rings) << shift;
2069 		if ((i & 3) == 3) {
2070 			E1000_WRITE_REG(hw,
2071 			    (E1000_RETA(0) + (i & ~3)), reta.dword);
2072 		}
2073 	}
2074 
2075 	/* Fill out hash function seeds */
2076 	for (i = 0; i < 10; i++) {
2077 		(void) random_get_pseudo_bytes((uint8_t *)&random,
2078 		    sizeof (uint32_t));
2079 		E1000_WRITE_REG(hw, E1000_RSSRK(i), random);
2080 	}
2081 
2082 	/* Setup the Multiple Receive Queue Control register */
2083 	mrqc = E1000_MRQC_ENABLE_RSS_4Q;
2084 	mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2085 	    E1000_MRQC_RSS_FIELD_IPV4_TCP |
2086 	    E1000_MRQC_RSS_FIELD_IPV6 |
2087 	    E1000_MRQC_RSS_FIELD_IPV6_TCP |
2088 	    E1000_MRQC_RSS_FIELD_IPV4_UDP |
2089 	    E1000_MRQC_RSS_FIELD_IPV6_UDP |
2090 	    E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2091 	    E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2092 
2093 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2094 
2095 	/*
2096 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2097 	 *
2098 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2099 	 * checksum offloading provided by the 82575 chipset besides the IP
2100 	 * header checksum offloading and the TCP/UDP checksum offloading.
2101 	 * The Packet Checksum is by default computed over the entire packet
2102 	 * from the first byte of the DA through the last byte of the CRC,
2103 	 * including the Ethernet and IP headers.
2104 	 *
2105 	 * It is a hardware limitation that Packet Checksum is mutually
2106 	 * exclusive with RSS.
2107 	 */
2108 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2109 	rxcsum |= E1000_RXCSUM_PCSD;
2110 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2111 }
2112 
2113 /*
2114  * igb_setup_mac_rss_classify - Setup MAC classification and rss
2115  */
2116 static void
2117 igb_setup_mac_rss_classify(igb_t *igb)
2118 {
2119 	struct e1000_hw *hw = &igb->hw;
2120 	uint32_t i, mrqc, vmdctl, rxcsum;
2121 	uint32_t ring_per_group;
2122 	int shift_group0, shift_group1;
2123 	uint32_t random;
2124 	union e1000_reta {
2125 		uint32_t	dword;
2126 		uint8_t		bytes[4];
2127 	} reta;
2128 
2129 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
2130 
2131 	/* Setup the Redirection Table, it is shared between two groups */
2132 	shift_group0 = 2;
2133 	shift_group1 = 6;
2134 	for (i = 0; i < (32 * 4); i++) {
2135 		reta.bytes[i & 3] = ((i % ring_per_group) << shift_group0) |
2136 		    ((ring_per_group + (i % ring_per_group)) << shift_group1);
2137 		if ((i & 3) == 3) {
2138 			E1000_WRITE_REG(hw,
2139 			    (E1000_RETA(0) + (i & ~3)), reta.dword);
2140 		}
2141 	}
2142 
2143 	/* Fill out hash function seeds */
2144 	for (i = 0; i < 10; i++) {
2145 		(void) random_get_pseudo_bytes((uint8_t *)&random,
2146 		    sizeof (uint32_t));
2147 		E1000_WRITE_REG(hw, E1000_RSSRK(i), random);
2148 	}
2149 
2150 	/*
2151 	 * Setup the Multiple Receive Queue Control register,
2152 	 * enable VMDq based on packet destination MAC address and RSS.
2153 	 */
2154 	mrqc = E1000_MRQC_ENABLE_VMDQ_MAC_RSS_GROUP;
2155 	mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2156 	    E1000_MRQC_RSS_FIELD_IPV4_TCP |
2157 	    E1000_MRQC_RSS_FIELD_IPV6 |
2158 	    E1000_MRQC_RSS_FIELD_IPV6_TCP |
2159 	    E1000_MRQC_RSS_FIELD_IPV4_UDP |
2160 	    E1000_MRQC_RSS_FIELD_IPV6_UDP |
2161 	    E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2162 	    E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2163 
2164 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2165 
2166 
2167 	/* Define the default group and default queues */
2168 	vmdctl = E1000_VMDQ_MAC_GROUP_DEFAULT_QUEUE;
2169 	E1000_WRITE_REG(hw, E1000_VMD_CTL, vmdctl);
2170 
2171 	/*
2172 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2173 	 *
2174 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2175 	 * checksum offloading provided by the 82575 chipset besides the IP
2176 	 * header checksum offloading and the TCP/UDP checksum offloading.
2177 	 * The Packet Checksum is by default computed over the entire packet
2178 	 * from the first byte of the DA through the last byte of the CRC,
2179 	 * including the Ethernet and IP headers.
2180 	 *
2181 	 * It is a hardware limitation that Packet Checksum is mutually
2182 	 * exclusive with RSS.
2183 	 */
2184 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2185 	rxcsum |= E1000_RXCSUM_PCSD;
2186 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2187 }
2188 
2189 /*
2190  * igb_setup_mac_classify - Setup MAC classification feature
2191  */
2192 static void
2193 igb_setup_mac_classify(igb_t *igb)
2194 {
2195 	struct e1000_hw *hw = &igb->hw;
2196 	uint32_t mrqc, rxcsum;
2197 
2198 	/*
2199 	 * Setup the Multiple Receive Queue Control register,
2200 	 * enable VMDq based on packet destination MAC address.
2201 	 */
2202 	mrqc = E1000_MRQC_ENABLE_VMDQ_MAC_GROUP;
2203 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2204 
2205 	/*
2206 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2207 	 *
2208 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2209 	 * checksum offloading provided by the 82575 chipset besides the IP
2210 	 * header checksum offloading and the TCP/UDP checksum offloading.
2211 	 * The Packet Checksum is by default computed over the entire packet
2212 	 * from the first byte of the DA through the last byte of the CRC,
2213 	 * including the Ethernet and IP headers.
2214 	 *
2215 	 * It is a hardware limitation that Packet Checksum is mutually
2216 	 * exclusive with RSS.
2217 	 */
2218 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2219 	rxcsum |= E1000_RXCSUM_PCSD;
2220 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2221 
2222 }
2223 
2224 /*
2225  * igb_init_unicst - Initialize the unicast addresses
2226  */
2227 static void
2228 igb_init_unicst(igb_t *igb)
2229 {
2230 	struct e1000_hw *hw = &igb->hw;
2231 	int slot;
2232 
2233 	/*
2234 	 * Here we should consider two situations:
2235 	 *
2236 	 * 1. Chipset is initialized the first time
2237 	 *    Initialize the multiple unicast addresses, and
2238 	 *    save the default MAC address.
2239 	 *
2240 	 * 2. Chipset is reset
2241 	 *    Recover the multiple unicast addresses from the
2242 	 *    software data structure to the RAR registers.
2243 	 */
2244 
2245 	/*
2246 	 * Clear the default MAC address in the RAR0 rgister,
2247 	 * which is loaded from EEPROM when system boot or chipreset,
2248 	 * this will cause the conficts with add_mac/rem_mac entry
2249 	 * points when VMDq is enabled. For this reason, the RAR0
2250 	 * must be cleared for both cases mentioned above.
2251 	 */
2252 	e1000_rar_clear(hw, 0);
2253 
2254 	if (!igb->unicst_init) {
2255 
2256 		/* Initialize the multiple unicast addresses */
2257 		igb->unicst_total = MAX_NUM_UNICAST_ADDRESSES;
2258 		igb->unicst_avail = igb->unicst_total;
2259 
2260 		for (slot = 0; slot < igb->unicst_total; slot++)
2261 			igb->unicst_addr[slot].mac.set = 0;
2262 
2263 		igb->unicst_init = B_TRUE;
2264 	} else {
2265 		/* Re-configure the RAR registers */
2266 		for (slot = 0; slot < igb->unicst_total; slot++) {
2267 			e1000_rar_set_vmdq(hw, igb->unicst_addr[slot].mac.addr,
2268 			    slot, igb->vmdq_mode,
2269 			    igb->unicst_addr[slot].mac.group_index);
2270 		}
2271 	}
2272 
2273 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
2274 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2275 }
2276 
2277 /*
2278  * igb_unicst_find - Find the slot for the specified unicast address
2279  */
2280 int
2281 igb_unicst_find(igb_t *igb, const uint8_t *mac_addr)
2282 {
2283 	int slot;
2284 
2285 	ASSERT(mutex_owned(&igb->gen_lock));
2286 
2287 	for (slot = 0; slot < igb->unicst_total; slot++) {
2288 		if (bcmp(igb->unicst_addr[slot].mac.addr,
2289 		    mac_addr, ETHERADDRL) == 0)
2290 			return (slot);
2291 	}
2292 
2293 	return (-1);
2294 }
2295 
2296 /*
2297  * igb_unicst_set - Set the unicast address to the specified slot
2298  */
2299 int
2300 igb_unicst_set(igb_t *igb, const uint8_t *mac_addr,
2301     int slot)
2302 {
2303 	struct e1000_hw *hw = &igb->hw;
2304 
2305 	ASSERT(mutex_owned(&igb->gen_lock));
2306 
2307 	/*
2308 	 * Save the unicast address in the software data structure
2309 	 */
2310 	bcopy(mac_addr, igb->unicst_addr[slot].mac.addr, ETHERADDRL);
2311 
2312 	/*
2313 	 * Set the unicast address to the RAR register
2314 	 */
2315 	e1000_rar_set(hw, (uint8_t *)mac_addr, slot);
2316 
2317 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2318 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2319 		return (EIO);
2320 	}
2321 
2322 	return (0);
2323 }
2324 
2325 /*
2326  * igb_multicst_add - Add a multicst address
2327  */
2328 int
2329 igb_multicst_add(igb_t *igb, const uint8_t *multiaddr)
2330 {
2331 	ASSERT(mutex_owned(&igb->gen_lock));
2332 
2333 	if ((multiaddr[0] & 01) == 0) {
2334 		return (EINVAL);
2335 	}
2336 
2337 	if (igb->mcast_count >= MAX_NUM_MULTICAST_ADDRESSES) {
2338 		return (ENOENT);
2339 	}
2340 
2341 	bcopy(multiaddr,
2342 	    &igb->mcast_table[igb->mcast_count], ETHERADDRL);
2343 	igb->mcast_count++;
2344 
2345 	/*
2346 	 * Update the multicast table in the hardware
2347 	 */
2348 	igb_setup_multicst(igb);
2349 
2350 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2351 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2352 		return (EIO);
2353 	}
2354 
2355 	return (0);
2356 }
2357 
2358 /*
2359  * igb_multicst_remove - Remove a multicst address
2360  */
2361 int
2362 igb_multicst_remove(igb_t *igb, const uint8_t *multiaddr)
2363 {
2364 	int i;
2365 
2366 	ASSERT(mutex_owned(&igb->gen_lock));
2367 
2368 	for (i = 0; i < igb->mcast_count; i++) {
2369 		if (bcmp(multiaddr, &igb->mcast_table[i],
2370 		    ETHERADDRL) == 0) {
2371 			for (i++; i < igb->mcast_count; i++) {
2372 				igb->mcast_table[i - 1] =
2373 				    igb->mcast_table[i];
2374 			}
2375 			igb->mcast_count--;
2376 			break;
2377 		}
2378 	}
2379 
2380 	/*
2381 	 * Update the multicast table in the hardware
2382 	 */
2383 	igb_setup_multicst(igb);
2384 
2385 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2386 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2387 		return (EIO);
2388 	}
2389 
2390 	return (0);
2391 }
2392 
2393 /*
2394  * igb_setup_multicast - setup multicast data structures
2395  *
2396  * This routine initializes all of the multicast related structures
2397  * and save them in the hardware registers.
2398  */
2399 static void
2400 igb_setup_multicst(igb_t *igb)
2401 {
2402 	uint8_t *mc_addr_list;
2403 	uint32_t mc_addr_count;
2404 	struct e1000_hw *hw = &igb->hw;
2405 
2406 	ASSERT(mutex_owned(&igb->gen_lock));
2407 
2408 	ASSERT(igb->mcast_count <= MAX_NUM_MULTICAST_ADDRESSES);
2409 
2410 	mc_addr_list = (uint8_t *)igb->mcast_table;
2411 	mc_addr_count = igb->mcast_count;
2412 
2413 	/*
2414 	 * Update the multicase addresses to the MTA registers
2415 	 */
2416 	e1000_update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2417 	    igb->unicst_total, hw->mac.rar_entry_count);
2418 }
2419 
2420 /*
2421  * igb_get_conf - Get driver configurations set in driver.conf
2422  *
2423  * This routine gets user-configured values out of the configuration
2424  * file igb.conf.
2425  *
2426  * For each configurable value, there is a minimum, a maximum, and a
2427  * default.
2428  * If user does not configure a value, use the default.
2429  * If user configures below the minimum, use the minumum.
2430  * If user configures above the maximum, use the maxumum.
2431  */
2432 static void
2433 igb_get_conf(igb_t *igb)
2434 {
2435 	struct e1000_hw *hw = &igb->hw;
2436 	uint32_t default_mtu;
2437 	uint32_t flow_control;
2438 	uint32_t ring_per_group;
2439 	int i;
2440 
2441 	/*
2442 	 * igb driver supports the following user configurations:
2443 	 *
2444 	 * Link configurations:
2445 	 *    adv_autoneg_cap
2446 	 *    adv_1000fdx_cap
2447 	 *    adv_100fdx_cap
2448 	 *    adv_100hdx_cap
2449 	 *    adv_10fdx_cap
2450 	 *    adv_10hdx_cap
2451 	 * Note: 1000hdx is not supported.
2452 	 *
2453 	 * Jumbo frame configuration:
2454 	 *    default_mtu
2455 	 *
2456 	 * Ethernet flow control configuration:
2457 	 *    flow_control
2458 	 *
2459 	 * Multiple rings configurations:
2460 	 *    tx_queue_number
2461 	 *    tx_ring_size
2462 	 *    rx_queue_number
2463 	 *    rx_ring_size
2464 	 *
2465 	 * Call igb_get_prop() to get the value for a specific
2466 	 * configuration parameter.
2467 	 */
2468 
2469 	/*
2470 	 * Link configurations
2471 	 */
2472 	igb->param_adv_autoneg_cap = igb_get_prop(igb,
2473 	    PROP_ADV_AUTONEG_CAP, 0, 1, 1);
2474 	igb->param_adv_1000fdx_cap = igb_get_prop(igb,
2475 	    PROP_ADV_1000FDX_CAP, 0, 1, 1);
2476 	igb->param_adv_100fdx_cap = igb_get_prop(igb,
2477 	    PROP_ADV_100FDX_CAP, 0, 1, 1);
2478 	igb->param_adv_100hdx_cap = igb_get_prop(igb,
2479 	    PROP_ADV_100HDX_CAP, 0, 1, 1);
2480 	igb->param_adv_10fdx_cap = igb_get_prop(igb,
2481 	    PROP_ADV_10FDX_CAP, 0, 1, 1);
2482 	igb->param_adv_10hdx_cap = igb_get_prop(igb,
2483 	    PROP_ADV_10HDX_CAP, 0, 1, 1);
2484 
2485 	/*
2486 	 * Jumbo frame configurations
2487 	 */
2488 	default_mtu = igb_get_prop(igb, PROP_DEFAULT_MTU,
2489 	    MIN_MTU, MAX_MTU, DEFAULT_MTU);
2490 
2491 	igb->max_frame_size = default_mtu +
2492 	    sizeof (struct ether_vlan_header) + ETHERFCSL;
2493 
2494 	/*
2495 	 * Ethernet flow control configuration
2496 	 */
2497 	flow_control = igb_get_prop(igb, PROP_FLOW_CONTROL,
2498 	    e1000_fc_none, 4, e1000_fc_full);
2499 	if (flow_control == 4)
2500 		flow_control = e1000_fc_default;
2501 
2502 	hw->fc.type = flow_control;
2503 
2504 	/*
2505 	 * Multiple rings configurations
2506 	 */
2507 	igb->tx_ring_size = igb_get_prop(igb, PROP_TX_RING_SIZE,
2508 	    MIN_TX_RING_SIZE, MAX_TX_RING_SIZE, DEFAULT_TX_RING_SIZE);
2509 	igb->rx_ring_size = igb_get_prop(igb, PROP_RX_RING_SIZE,
2510 	    MIN_RX_RING_SIZE, MAX_RX_RING_SIZE, DEFAULT_RX_RING_SIZE);
2511 
2512 	igb->mr_enable = igb_get_prop(igb, PROP_MR_ENABLE, 0, 1, 1);
2513 	igb->num_rx_groups = igb_get_prop(igb, PROP_RX_GROUP_NUM,
2514 	    MIN_RX_GROUP_NUM, MAX_RX_GROUP_NUM, DEFAULT_RX_GROUP_NUM);
2515 
2516 	if (igb->mr_enable) {
2517 		igb->num_tx_rings = DEFAULT_TX_QUEUE_NUM;
2518 		igb->num_rx_rings = DEFAULT_RX_QUEUE_NUM;
2519 	} else {
2520 		igb->num_tx_rings = 1;
2521 		igb->num_rx_rings = 1;
2522 
2523 		if (igb->num_rx_groups > 1) {
2524 			igb_error(igb,
2525 			    "Invalid rx groups number. Please enable multiple "
2526 			    "rings first");
2527 			igb->num_rx_groups = 1;
2528 		}
2529 	}
2530 
2531 	/*
2532 	 * Check the divisibility between rx rings and rx groups.
2533 	 */
2534 	for (i = igb->num_rx_groups; i > 0; i--) {
2535 		if ((igb->num_rx_rings % i) == 0)
2536 			break;
2537 	}
2538 	if (i != igb->num_rx_groups) {
2539 		igb_error(igb,
2540 		    "Invalid rx groups number. Downgrade the rx group "
2541 		    "number to %d.", i);
2542 		igb->num_rx_groups = i;
2543 	}
2544 
2545 	/*
2546 	 * Get the ring number per group.
2547 	 */
2548 	ring_per_group = igb->num_rx_rings / igb->num_rx_groups;
2549 
2550 	if (igb->num_rx_groups == 1) {
2551 		/*
2552 		 * One rx ring group, the rx ring number is num_rx_rings.
2553 		 */
2554 		igb->vmdq_mode = E1000_VMDQ_OFF;
2555 	} else if (ring_per_group == 1) {
2556 		/*
2557 		 * Multiple rx groups, each group has one rx ring.
2558 		 */
2559 		igb->vmdq_mode = E1000_VMDQ_MAC;
2560 	} else {
2561 		/*
2562 		 * Multiple groups and multiple rings.
2563 		 */
2564 		igb->vmdq_mode = E1000_VMDQ_MAC_RSS;
2565 	}
2566 
2567 	/*
2568 	 * Tunable used to force an interrupt type. The only use is
2569 	 * for testing of the lesser interrupt types.
2570 	 * 0 = don't force interrupt type
2571 	 * 1 = force interrupt type MSIX
2572 	 * 2 = force interrupt type MSI
2573 	 * 3 = force interrupt type Legacy
2574 	 */
2575 	igb->intr_force = igb_get_prop(igb, PROP_INTR_FORCE,
2576 	    IGB_INTR_NONE, IGB_INTR_LEGACY, IGB_INTR_NONE);
2577 
2578 	igb->tx_hcksum_enable = igb_get_prop(igb, PROP_TX_HCKSUM_ENABLE,
2579 	    0, 1, 1);
2580 	igb->rx_hcksum_enable = igb_get_prop(igb, PROP_RX_HCKSUM_ENABLE,
2581 	    0, 1, 1);
2582 	igb->lso_enable = igb_get_prop(igb, PROP_LSO_ENABLE,
2583 	    0, 1, 0);
2584 	igb->tx_head_wb_enable = igb_get_prop(igb, PROP_TX_HEAD_WB_ENABLE,
2585 	    0, 1, 1);
2586 
2587 	igb->tx_copy_thresh = igb_get_prop(igb, PROP_TX_COPY_THRESHOLD,
2588 	    MIN_TX_COPY_THRESHOLD, MAX_TX_COPY_THRESHOLD,
2589 	    DEFAULT_TX_COPY_THRESHOLD);
2590 	igb->tx_recycle_thresh = igb_get_prop(igb, PROP_TX_RECYCLE_THRESHOLD,
2591 	    MIN_TX_RECYCLE_THRESHOLD, MAX_TX_RECYCLE_THRESHOLD,
2592 	    DEFAULT_TX_RECYCLE_THRESHOLD);
2593 	igb->tx_overload_thresh = igb_get_prop(igb, PROP_TX_OVERLOAD_THRESHOLD,
2594 	    MIN_TX_OVERLOAD_THRESHOLD, MAX_TX_OVERLOAD_THRESHOLD,
2595 	    DEFAULT_TX_OVERLOAD_THRESHOLD);
2596 	igb->tx_resched_thresh = igb_get_prop(igb, PROP_TX_RESCHED_THRESHOLD,
2597 	    MIN_TX_RESCHED_THRESHOLD, MAX_TX_RESCHED_THRESHOLD,
2598 	    DEFAULT_TX_RESCHED_THRESHOLD);
2599 
2600 	igb->rx_copy_thresh = igb_get_prop(igb, PROP_RX_COPY_THRESHOLD,
2601 	    MIN_RX_COPY_THRESHOLD, MAX_RX_COPY_THRESHOLD,
2602 	    DEFAULT_RX_COPY_THRESHOLD);
2603 	igb->rx_limit_per_intr = igb_get_prop(igb, PROP_RX_LIMIT_PER_INTR,
2604 	    MIN_RX_LIMIT_PER_INTR, MAX_RX_LIMIT_PER_INTR,
2605 	    DEFAULT_RX_LIMIT_PER_INTR);
2606 
2607 	igb->intr_throttling[0] = igb_get_prop(igb, PROP_INTR_THROTTLING,
2608 	    MIN_INTR_THROTTLING, MAX_INTR_THROTTLING,
2609 	    DEFAULT_INTR_THROTTLING);
2610 }
2611 
2612 /*
2613  * igb_get_prop - Get a property value out of the configuration file igb.conf
2614  *
2615  * Caller provides the name of the property, a default value, a minimum
2616  * value, and a maximum value.
2617  *
2618  * Return configured value of the property, with default, minimum and
2619  * maximum properly applied.
2620  */
2621 static int
2622 igb_get_prop(igb_t *igb,
2623     char *propname,	/* name of the property */
2624     int minval,		/* minimum acceptable value */
2625     int maxval,		/* maximim acceptable value */
2626     int defval)		/* default value */
2627 {
2628 	int value;
2629 
2630 	/*
2631 	 * Call ddi_prop_get_int() to read the conf settings
2632 	 */
2633 	value = ddi_prop_get_int(DDI_DEV_T_ANY, igb->dip,
2634 	    DDI_PROP_DONTPASS, propname, defval);
2635 
2636 	if (value > maxval)
2637 		value = maxval;
2638 
2639 	if (value < minval)
2640 		value = minval;
2641 
2642 	return (value);
2643 }
2644 
2645 /*
2646  * igb_setup_link - Using the link properties to setup the link
2647  */
2648 int
2649 igb_setup_link(igb_t *igb, boolean_t setup_hw)
2650 {
2651 	struct e1000_mac_info *mac;
2652 	struct e1000_phy_info *phy;
2653 	boolean_t invalid;
2654 
2655 	mac = &igb->hw.mac;
2656 	phy = &igb->hw.phy;
2657 	invalid = B_FALSE;
2658 
2659 	if (igb->param_adv_autoneg_cap == 1) {
2660 		mac->autoneg = B_TRUE;
2661 		phy->autoneg_advertised = 0;
2662 
2663 		/*
2664 		 * 1000hdx is not supported for autonegotiation
2665 		 */
2666 		if (igb->param_adv_1000fdx_cap == 1)
2667 			phy->autoneg_advertised |= ADVERTISE_1000_FULL;
2668 
2669 		if (igb->param_adv_100fdx_cap == 1)
2670 			phy->autoneg_advertised |= ADVERTISE_100_FULL;
2671 
2672 		if (igb->param_adv_100hdx_cap == 1)
2673 			phy->autoneg_advertised |= ADVERTISE_100_HALF;
2674 
2675 		if (igb->param_adv_10fdx_cap == 1)
2676 			phy->autoneg_advertised |= ADVERTISE_10_FULL;
2677 
2678 		if (igb->param_adv_10hdx_cap == 1)
2679 			phy->autoneg_advertised |= ADVERTISE_10_HALF;
2680 
2681 		if (phy->autoneg_advertised == 0)
2682 			invalid = B_TRUE;
2683 	} else {
2684 		mac->autoneg = B_FALSE;
2685 
2686 		/*
2687 		 * 1000fdx and 1000hdx are not supported for forced link
2688 		 */
2689 		if (igb->param_adv_100fdx_cap == 1)
2690 			mac->forced_speed_duplex = ADVERTISE_100_FULL;
2691 		else if (igb->param_adv_100hdx_cap == 1)
2692 			mac->forced_speed_duplex = ADVERTISE_100_HALF;
2693 		else if (igb->param_adv_10fdx_cap == 1)
2694 			mac->forced_speed_duplex = ADVERTISE_10_FULL;
2695 		else if (igb->param_adv_10hdx_cap == 1)
2696 			mac->forced_speed_duplex = ADVERTISE_10_HALF;
2697 		else
2698 			invalid = B_TRUE;
2699 	}
2700 
2701 	if (invalid) {
2702 		igb_notice(igb, "Invalid link settings. Setup link to "
2703 		    "autonegotiation with full link capabilities.");
2704 		mac->autoneg = B_TRUE;
2705 		phy->autoneg_advertised = ADVERTISE_1000_FULL |
2706 		    ADVERTISE_100_FULL | ADVERTISE_100_HALF |
2707 		    ADVERTISE_10_FULL | ADVERTISE_10_HALF;
2708 	}
2709 
2710 	if (setup_hw) {
2711 		if (e1000_setup_link(&igb->hw) != E1000_SUCCESS)
2712 			return (IGB_FAILURE);
2713 	}
2714 
2715 	return (IGB_SUCCESS);
2716 }
2717 
2718 
2719 /*
2720  * igb_is_link_up - Check if the link is up
2721  */
2722 static boolean_t
2723 igb_is_link_up(igb_t *igb)
2724 {
2725 	struct e1000_hw *hw = &igb->hw;
2726 	boolean_t link_up;
2727 
2728 	ASSERT(mutex_owned(&igb->gen_lock));
2729 
2730 	(void) e1000_check_for_link(hw);
2731 
2732 	if ((E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU) ||
2733 	    ((hw->phy.media_type == e1000_media_type_internal_serdes) &&
2734 	    (hw->mac.serdes_has_link))) {
2735 		link_up = B_TRUE;
2736 	} else {
2737 		link_up = B_FALSE;
2738 	}
2739 
2740 	return (link_up);
2741 }
2742 
2743 /*
2744  * igb_link_check - Link status processing
2745  */
2746 static boolean_t
2747 igb_link_check(igb_t *igb)
2748 {
2749 	struct e1000_hw *hw = &igb->hw;
2750 	uint16_t speed = 0, duplex = 0;
2751 	boolean_t link_changed = B_FALSE;
2752 
2753 	ASSERT(mutex_owned(&igb->gen_lock));
2754 
2755 	if (igb_is_link_up(igb)) {
2756 		/*
2757 		 * The Link is up, check whether it was marked as down earlier
2758 		 */
2759 		if (igb->link_state != LINK_STATE_UP) {
2760 			(void) e1000_get_speed_and_duplex(hw, &speed, &duplex);
2761 			igb->link_speed = speed;
2762 			igb->link_duplex = duplex;
2763 			igb->link_state = LINK_STATE_UP;
2764 			igb->link_down_timeout = 0;
2765 			link_changed = B_TRUE;
2766 		}
2767 	} else {
2768 		if (igb->link_state != LINK_STATE_DOWN) {
2769 			igb->link_speed = 0;
2770 			igb->link_duplex = 0;
2771 			igb->link_state = LINK_STATE_DOWN;
2772 			link_changed = B_TRUE;
2773 		}
2774 
2775 		if (igb->igb_state & IGB_STARTED) {
2776 			if (igb->link_down_timeout < MAX_LINK_DOWN_TIMEOUT) {
2777 				igb->link_down_timeout++;
2778 			} else if (igb->link_down_timeout ==
2779 			    MAX_LINK_DOWN_TIMEOUT) {
2780 				igb_tx_clean(igb);
2781 				igb->link_down_timeout++;
2782 			}
2783 		}
2784 	}
2785 
2786 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
2787 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2788 
2789 	return (link_changed);
2790 }
2791 
2792 /*
2793  * igb_local_timer - driver watchdog function
2794  *
2795  * This function will handle the transmit stall check, link status check and
2796  * other routines.
2797  */
2798 static void
2799 igb_local_timer(void *arg)
2800 {
2801 	igb_t *igb = (igb_t *)arg;
2802 	struct e1000_hw *hw = &igb->hw;
2803 	boolean_t link_changed;
2804 
2805 	if (igb_stall_check(igb)) {
2806 		igb_fm_ereport(igb, DDI_FM_DEVICE_STALL);
2807 		igb->reset_count++;
2808 		if (igb_reset(igb) == IGB_SUCCESS)
2809 			ddi_fm_service_impact(igb->dip,
2810 			    DDI_SERVICE_RESTORED);
2811 	}
2812 
2813 	mutex_enter(&igb->gen_lock);
2814 	link_changed = igb_link_check(igb);
2815 	mutex_exit(&igb->gen_lock);
2816 
2817 	if (link_changed)
2818 		mac_link_update(igb->mac_hdl, igb->link_state);
2819 
2820 	/*
2821 	 * Set Timer Interrupts
2822 	 */
2823 	if (igb->intr_type != DDI_INTR_TYPE_MSIX)
2824 		E1000_WRITE_REG(hw, E1000_ICS, E1000_IMS_RXT0);
2825 
2826 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
2827 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2828 
2829 	igb_restart_watchdog_timer(igb);
2830 }
2831 
2832 /*
2833  * igb_stall_check - check for transmit stall
2834  *
2835  * This function checks if the adapter is stalled (in transmit).
2836  *
2837  * It is called each time the watchdog timeout is invoked.
2838  * If the transmit descriptor reclaim continuously fails,
2839  * the watchdog value will increment by 1. If the watchdog
2840  * value exceeds the threshold, the igb is assumed to
2841  * have stalled and need to be reset.
2842  */
2843 static boolean_t
2844 igb_stall_check(igb_t *igb)
2845 {
2846 	igb_tx_ring_t *tx_ring;
2847 	boolean_t result;
2848 	int i;
2849 
2850 	if (igb->link_state != LINK_STATE_UP)
2851 		return (B_FALSE);
2852 
2853 	/*
2854 	 * If any tx ring is stalled, we'll reset the chipset
2855 	 */
2856 	result = B_FALSE;
2857 	for (i = 0; i < igb->num_tx_rings; i++) {
2858 		tx_ring = &igb->tx_rings[i];
2859 
2860 		if (tx_ring->recycle_fail > 0)
2861 			tx_ring->stall_watchdog++;
2862 		else
2863 			tx_ring->stall_watchdog = 0;
2864 
2865 		if (tx_ring->stall_watchdog >= STALL_WATCHDOG_TIMEOUT) {
2866 			result = B_TRUE;
2867 			break;
2868 		}
2869 	}
2870 
2871 	if (result) {
2872 		tx_ring->stall_watchdog = 0;
2873 		tx_ring->recycle_fail = 0;
2874 	}
2875 
2876 	return (result);
2877 }
2878 
2879 
2880 /*
2881  * is_valid_mac_addr - Check if the mac address is valid
2882  */
2883 static boolean_t
2884 is_valid_mac_addr(uint8_t *mac_addr)
2885 {
2886 	const uint8_t addr_test1[6] = { 0, 0, 0, 0, 0, 0 };
2887 	const uint8_t addr_test2[6] =
2888 	    { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF };
2889 
2890 	if (!(bcmp(addr_test1, mac_addr, ETHERADDRL)) ||
2891 	    !(bcmp(addr_test2, mac_addr, ETHERADDRL)))
2892 		return (B_FALSE);
2893 
2894 	return (B_TRUE);
2895 }
2896 
2897 static boolean_t
2898 igb_find_mac_address(igb_t *igb)
2899 {
2900 	struct e1000_hw *hw = &igb->hw;
2901 #ifdef __sparc
2902 	uchar_t *bytes;
2903 	struct ether_addr sysaddr;
2904 	uint_t nelts;
2905 	int err;
2906 	boolean_t found = B_FALSE;
2907 
2908 	/*
2909 	 * The "vendor's factory-set address" may already have
2910 	 * been extracted from the chip, but if the property
2911 	 * "local-mac-address" is set we use that instead.
2912 	 *
2913 	 * We check whether it looks like an array of 6
2914 	 * bytes (which it should, if OBP set it).  If we can't
2915 	 * make sense of it this way, we'll ignore it.
2916 	 */
2917 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip,
2918 	    DDI_PROP_DONTPASS, "local-mac-address", &bytes, &nelts);
2919 	if (err == DDI_PROP_SUCCESS) {
2920 		if (nelts == ETHERADDRL) {
2921 			while (nelts--)
2922 				hw->mac.addr[nelts] = bytes[nelts];
2923 			found = B_TRUE;
2924 		}
2925 		ddi_prop_free(bytes);
2926 	}
2927 
2928 	/*
2929 	 * Look up the OBP property "local-mac-address?". If the user has set
2930 	 * 'local-mac-address? = false', use "the system address" instead.
2931 	 */
2932 	if (ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip, 0,
2933 	    "local-mac-address?", &bytes, &nelts) == DDI_PROP_SUCCESS) {
2934 		if (strncmp("false", (caddr_t)bytes, (size_t)nelts) == 0) {
2935 			if (localetheraddr(NULL, &sysaddr) != 0) {
2936 				bcopy(&sysaddr, hw->mac.addr, ETHERADDRL);
2937 				found = B_TRUE;
2938 			}
2939 		}
2940 		ddi_prop_free(bytes);
2941 	}
2942 
2943 	/*
2944 	 * Finally(!), if there's a valid "mac-address" property (created
2945 	 * if we netbooted from this interface), we must use this instead
2946 	 * of any of the above to ensure that the NFS/install server doesn't
2947 	 * get confused by the address changing as Solaris takes over!
2948 	 */
2949 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip,
2950 	    DDI_PROP_DONTPASS, "mac-address", &bytes, &nelts);
2951 	if (err == DDI_PROP_SUCCESS) {
2952 		if (nelts == ETHERADDRL) {
2953 			while (nelts--)
2954 				hw->mac.addr[nelts] = bytes[nelts];
2955 			found = B_TRUE;
2956 		}
2957 		ddi_prop_free(bytes);
2958 	}
2959 
2960 	if (found) {
2961 		bcopy(hw->mac.addr, hw->mac.perm_addr, ETHERADDRL);
2962 		return (B_TRUE);
2963 	}
2964 #endif
2965 
2966 	/*
2967 	 * Read the device MAC address from the EEPROM
2968 	 */
2969 	if (e1000_read_mac_addr(hw) != E1000_SUCCESS)
2970 		return (B_FALSE);
2971 
2972 	return (B_TRUE);
2973 }
2974 
2975 #pragma inline(igb_arm_watchdog_timer)
2976 
2977 static void
2978 igb_arm_watchdog_timer(igb_t *igb)
2979 {
2980 	/*
2981 	 * Fire a watchdog timer
2982 	 */
2983 	igb->watchdog_tid =
2984 	    timeout(igb_local_timer,
2985 	    (void *)igb, 1 * drv_usectohz(1000000));
2986 
2987 }
2988 
2989 /*
2990  * igb_enable_watchdog_timer - Enable and start the driver watchdog timer
2991  */
2992 void
2993 igb_enable_watchdog_timer(igb_t *igb)
2994 {
2995 	mutex_enter(&igb->watchdog_lock);
2996 
2997 	if (!igb->watchdog_enable) {
2998 		igb->watchdog_enable = B_TRUE;
2999 		igb->watchdog_start = B_TRUE;
3000 		igb_arm_watchdog_timer(igb);
3001 	}
3002 
3003 	mutex_exit(&igb->watchdog_lock);
3004 
3005 }
3006 
3007 /*
3008  * igb_disable_watchdog_timer - Disable and stop the driver watchdog timer
3009  */
3010 void
3011 igb_disable_watchdog_timer(igb_t *igb)
3012 {
3013 	timeout_id_t tid;
3014 
3015 	mutex_enter(&igb->watchdog_lock);
3016 
3017 	igb->watchdog_enable = B_FALSE;
3018 	igb->watchdog_start = B_FALSE;
3019 	tid = igb->watchdog_tid;
3020 	igb->watchdog_tid = 0;
3021 
3022 	mutex_exit(&igb->watchdog_lock);
3023 
3024 	if (tid != 0)
3025 		(void) untimeout(tid);
3026 
3027 }
3028 
3029 /*
3030  * igb_start_watchdog_timer - Start the driver watchdog timer
3031  */
3032 static void
3033 igb_start_watchdog_timer(igb_t *igb)
3034 {
3035 	mutex_enter(&igb->watchdog_lock);
3036 
3037 	if (igb->watchdog_enable) {
3038 		if (!igb->watchdog_start) {
3039 			igb->watchdog_start = B_TRUE;
3040 			igb_arm_watchdog_timer(igb);
3041 		}
3042 	}
3043 
3044 	mutex_exit(&igb->watchdog_lock);
3045 }
3046 
3047 /*
3048  * igb_restart_watchdog_timer - Restart the driver watchdog timer
3049  */
3050 static void
3051 igb_restart_watchdog_timer(igb_t *igb)
3052 {
3053 	mutex_enter(&igb->watchdog_lock);
3054 
3055 	if (igb->watchdog_start)
3056 		igb_arm_watchdog_timer(igb);
3057 
3058 	mutex_exit(&igb->watchdog_lock);
3059 }
3060 
3061 /*
3062  * igb_stop_watchdog_timer - Stop the driver watchdog timer
3063  */
3064 static void
3065 igb_stop_watchdog_timer(igb_t *igb)
3066 {
3067 	timeout_id_t tid;
3068 
3069 	mutex_enter(&igb->watchdog_lock);
3070 
3071 	igb->watchdog_start = B_FALSE;
3072 	tid = igb->watchdog_tid;
3073 	igb->watchdog_tid = 0;
3074 
3075 	mutex_exit(&igb->watchdog_lock);
3076 
3077 	if (tid != 0)
3078 		(void) untimeout(tid);
3079 }
3080 
3081 /*
3082  * igb_disable_adapter_interrupts - Clear/disable all hardware interrupts
3083  */
3084 static void
3085 igb_disable_adapter_interrupts(igb_t *igb)
3086 {
3087 	struct e1000_hw *hw = &igb->hw;
3088 
3089 	/*
3090 	 * Set the IMC register to mask all the interrupts,
3091 	 * including the tx interrupts.
3092 	 */
3093 	E1000_WRITE_REG(hw, E1000_IMC, 0xffffffff);
3094 
3095 	/*
3096 	 * Additional disabling for MSI-X
3097 	 */
3098 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3099 		E1000_WRITE_REG(hw, E1000_EIMC, 0xffffffff);
3100 		E1000_WRITE_REG(hw, E1000_EIAC, 0x0);
3101 	}
3102 
3103 	E1000_WRITE_FLUSH(hw);
3104 }
3105 
3106 /*
3107  * igb_enable_adapter_interrupts - Mask/enable all hardware interrupts
3108  */
3109 static void
3110 igb_enable_adapter_interrupts(igb_t *igb)
3111 {
3112 	struct e1000_hw *hw = &igb->hw;
3113 	uint32_t reg;
3114 
3115 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
3116 		/* Interrupt enabling for MSI-X */
3117 		E1000_WRITE_REG(hw, E1000_EIMS, igb->eims_mask);
3118 		E1000_WRITE_REG(hw, E1000_EIAC, igb->eims_mask);
3119 		igb->ims_mask = E1000_IMS_LSC;
3120 		E1000_WRITE_REG(hw, E1000_IMS, E1000_IMS_LSC);
3121 
3122 		/* Enable MSI-X PBA support */
3123 		reg = E1000_READ_REG(hw, E1000_CTRL_EXT);
3124 		reg |= E1000_CTRL_EXT_PBA_CLR;
3125 
3126 		/* Non-selective interrupt clear-on-read */
3127 		reg |= E1000_CTRL_EXT_IRCA;	/* Called NSICR in the EAS */
3128 
3129 		E1000_WRITE_REG(hw, E1000_CTRL_EXT, reg);
3130 	} else {
3131 		/* Interrupt enabling for MSI and legacy */
3132 		igb->ims_mask = IMS_ENABLE_MASK;
3133 		E1000_WRITE_REG(hw, E1000_IMS, IMS_ENABLE_MASK);
3134 	}
3135 
3136 	E1000_WRITE_FLUSH(hw);
3137 }
3138 
3139 /*
3140  * Loopback Support
3141  */
3142 static lb_property_t lb_normal =
3143 	{ normal,	"normal",	IGB_LB_NONE		};
3144 static lb_property_t lb_external =
3145 	{ external,	"External",	IGB_LB_EXTERNAL		};
3146 static lb_property_t lb_mac =
3147 	{ internal,	"MAC",		IGB_LB_INTERNAL_MAC	};
3148 static lb_property_t lb_phy =
3149 	{ internal,	"PHY",		IGB_LB_INTERNAL_PHY	};
3150 static lb_property_t lb_serdes =
3151 	{ internal,	"SerDes",	IGB_LB_INTERNAL_SERDES	};
3152 
3153 enum ioc_reply
3154 igb_loopback_ioctl(igb_t *igb, struct iocblk *iocp, mblk_t *mp)
3155 {
3156 	lb_info_sz_t *lbsp;
3157 	lb_property_t *lbpp;
3158 	struct e1000_hw *hw;
3159 	uint32_t *lbmp;
3160 	uint32_t size;
3161 	uint32_t value;
3162 
3163 	hw = &igb->hw;
3164 
3165 	if (mp->b_cont == NULL)
3166 		return (IOC_INVAL);
3167 
3168 	switch (iocp->ioc_cmd) {
3169 	default:
3170 		return (IOC_INVAL);
3171 
3172 	case LB_GET_INFO_SIZE:
3173 		size = sizeof (lb_info_sz_t);
3174 		if (iocp->ioc_count != size)
3175 			return (IOC_INVAL);
3176 
3177 		value = sizeof (lb_normal);
3178 		value += sizeof (lb_mac);
3179 		if (hw->phy.media_type == e1000_media_type_copper)
3180 			value += sizeof (lb_phy);
3181 		else
3182 			value += sizeof (lb_serdes);
3183 		value += sizeof (lb_external);
3184 
3185 		lbsp = (lb_info_sz_t *)(uintptr_t)mp->b_cont->b_rptr;
3186 		*lbsp = value;
3187 		break;
3188 
3189 	case LB_GET_INFO:
3190 		value = sizeof (lb_normal);
3191 		value += sizeof (lb_mac);
3192 		if (hw->phy.media_type == e1000_media_type_copper)
3193 			value += sizeof (lb_phy);
3194 		else
3195 			value += sizeof (lb_serdes);
3196 		value += sizeof (lb_external);
3197 
3198 		size = value;
3199 		if (iocp->ioc_count != size)
3200 			return (IOC_INVAL);
3201 
3202 		value = 0;
3203 		lbpp = (lb_property_t *)(uintptr_t)mp->b_cont->b_rptr;
3204 
3205 		lbpp[value++] = lb_normal;
3206 		lbpp[value++] = lb_mac;
3207 		if (hw->phy.media_type == e1000_media_type_copper)
3208 			lbpp[value++] = lb_phy;
3209 		else
3210 			lbpp[value++] = lb_serdes;
3211 		lbpp[value++] = lb_external;
3212 		break;
3213 
3214 	case LB_GET_MODE:
3215 		size = sizeof (uint32_t);
3216 		if (iocp->ioc_count != size)
3217 			return (IOC_INVAL);
3218 
3219 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
3220 		*lbmp = igb->loopback_mode;
3221 		break;
3222 
3223 	case LB_SET_MODE:
3224 		size = 0;
3225 		if (iocp->ioc_count != sizeof (uint32_t))
3226 			return (IOC_INVAL);
3227 
3228 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
3229 		if (!igb_set_loopback_mode(igb, *lbmp))
3230 			return (IOC_INVAL);
3231 		break;
3232 	}
3233 
3234 	iocp->ioc_count = size;
3235 	iocp->ioc_error = 0;
3236 
3237 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
3238 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
3239 		return (IOC_INVAL);
3240 	}
3241 
3242 	return (IOC_REPLY);
3243 }
3244 
3245 /*
3246  * igb_set_loopback_mode - Setup loopback based on the loopback mode
3247  */
3248 static boolean_t
3249 igb_set_loopback_mode(igb_t *igb, uint32_t mode)
3250 {
3251 	struct e1000_hw *hw;
3252 
3253 	if (mode == igb->loopback_mode)
3254 		return (B_TRUE);
3255 
3256 	hw = &igb->hw;
3257 
3258 	igb->loopback_mode = mode;
3259 
3260 	if (mode == IGB_LB_NONE) {
3261 		/* Reset the chip */
3262 		hw->phy.autoneg_wait_to_complete = B_TRUE;
3263 		(void) igb_reset(igb);
3264 		hw->phy.autoneg_wait_to_complete = B_FALSE;
3265 		return (B_TRUE);
3266 	}
3267 
3268 	mutex_enter(&igb->gen_lock);
3269 
3270 	switch (mode) {
3271 	default:
3272 		mutex_exit(&igb->gen_lock);
3273 		return (B_FALSE);
3274 
3275 	case IGB_LB_EXTERNAL:
3276 		igb_set_external_loopback(igb);
3277 		break;
3278 
3279 	case IGB_LB_INTERNAL_MAC:
3280 		igb_set_internal_mac_loopback(igb);
3281 		break;
3282 
3283 	case IGB_LB_INTERNAL_PHY:
3284 		igb_set_internal_phy_loopback(igb);
3285 		break;
3286 
3287 	case IGB_LB_INTERNAL_SERDES:
3288 		igb_set_internal_serdes_loopback(igb);
3289 		break;
3290 	}
3291 
3292 	mutex_exit(&igb->gen_lock);
3293 
3294 	return (B_TRUE);
3295 }
3296 
3297 /*
3298  * igb_set_external_loopback - Set the external loopback mode
3299  */
3300 static void
3301 igb_set_external_loopback(igb_t *igb)
3302 {
3303 	struct e1000_hw *hw;
3304 
3305 	hw = &igb->hw;
3306 
3307 	/* Set phy to known state */
3308 	(void) e1000_phy_hw_reset(hw);
3309 
3310 	(void) e1000_write_phy_reg(hw, 0x0, 0x0140);
3311 	(void) e1000_write_phy_reg(hw, 0x9, 0x1b00);
3312 	(void) e1000_write_phy_reg(hw, 0x12, 0x1610);
3313 	(void) e1000_write_phy_reg(hw, 0x1f37, 0x3f1c);
3314 }
3315 
3316 /*
3317  * igb_set_internal_mac_loopback - Set the internal MAC loopback mode
3318  */
3319 static void
3320 igb_set_internal_mac_loopback(igb_t *igb)
3321 {
3322 	struct e1000_hw *hw;
3323 	uint32_t ctrl;
3324 	uint32_t rctl;
3325 
3326 	hw = &igb->hw;
3327 
3328 	/* Set the Receive Control register */
3329 	rctl = E1000_READ_REG(hw, E1000_RCTL);
3330 	rctl &= ~E1000_RCTL_LBM_TCVR;
3331 	rctl |= E1000_RCTL_LBM_MAC;
3332 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
3333 
3334 	/* Set the Device Control register */
3335 	ctrl = E1000_READ_REG(hw, E1000_CTRL);
3336 	ctrl &= ~E1000_CTRL_SPD_SEL;	/* Clear the speed sel bits */
3337 	ctrl |= (E1000_CTRL_SLU |	/* Force link up */
3338 	    E1000_CTRL_FRCSPD |		/* Force speed */
3339 	    E1000_CTRL_FRCDPX |		/* Force duplex */
3340 	    E1000_CTRL_SPD_1000 |	/* Force speed to 1000 */
3341 	    E1000_CTRL_FD);		/* Force full duplex */
3342 	ctrl &= ~E1000_CTRL_ILOS;	/* Clear ILOS when there's a link */
3343 	E1000_WRITE_REG(hw, E1000_CTRL, ctrl);
3344 }
3345 
3346 /*
3347  * igb_set_internal_phy_loopback - Set the internal PHY loopback mode
3348  */
3349 static void
3350 igb_set_internal_phy_loopback(igb_t *igb)
3351 {
3352 	struct e1000_hw *hw;
3353 	uint32_t ctrl_ext;
3354 	uint16_t phy_ctrl;
3355 	uint16_t phy_pconf;
3356 
3357 	hw = &igb->hw;
3358 
3359 	/* Set link mode to PHY (00b) in the Extended Control register */
3360 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3361 	ctrl_ext &= ~E1000_CTRL_EXT_LINK_MODE_MASK;
3362 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3363 
3364 	/*
3365 	 * Set PHY control register (0x4140):
3366 	 *    Set full duplex mode
3367 	 *    Set loopback bit
3368 	 *    Clear auto-neg enable bit
3369 	 *    Set PHY speed
3370 	 */
3371 	phy_ctrl = MII_CR_FULL_DUPLEX | MII_CR_SPEED_1000 | MII_CR_LOOPBACK;
3372 	(void) e1000_write_phy_reg(hw, PHY_CONTROL, phy_ctrl);
3373 
3374 	/* Set the link disable bit in the Port Configuration register */
3375 	(void) e1000_read_phy_reg(hw, 0x10, &phy_pconf);
3376 	phy_pconf |= (uint16_t)1 << 14;
3377 	(void) e1000_write_phy_reg(hw, 0x10, phy_pconf);
3378 }
3379 
3380 /*
3381  * igb_set_internal_serdes_loopback - Set the internal SerDes loopback mode
3382  */
3383 static void
3384 igb_set_internal_serdes_loopback(igb_t *igb)
3385 {
3386 	struct e1000_hw *hw;
3387 	uint32_t ctrl_ext;
3388 	uint32_t ctrl;
3389 	uint32_t pcs_lctl;
3390 	uint32_t connsw;
3391 
3392 	hw = &igb->hw;
3393 
3394 	/* Set link mode to SerDes (11b) in the Extended Control register */
3395 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3396 	ctrl_ext |= E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
3397 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3398 
3399 	/* Configure the SerDes to loopback */
3400 	E1000_WRITE_REG(hw, E1000_SCTL, 0x410);
3401 
3402 	/* Set Device Control register */
3403 	ctrl = E1000_READ_REG(hw, E1000_CTRL);
3404 	ctrl |= (E1000_CTRL_FD |	/* Force full duplex */
3405 	    E1000_CTRL_SLU);		/* Force link up */
3406 	ctrl &= ~(E1000_CTRL_RFCE |	/* Disable receive flow control */
3407 	    E1000_CTRL_TFCE |		/* Disable transmit flow control */
3408 	    E1000_CTRL_LRST);		/* Clear link reset */
3409 	E1000_WRITE_REG(hw, E1000_CTRL, ctrl);
3410 
3411 	/* Set PCS Link Control register */
3412 	pcs_lctl = E1000_READ_REG(hw, E1000_PCS_LCTL);
3413 	pcs_lctl |= (E1000_PCS_LCTL_FORCE_LINK |
3414 	    E1000_PCS_LCTL_FSD |
3415 	    E1000_PCS_LCTL_FDV_FULL |
3416 	    E1000_PCS_LCTL_FLV_LINK_UP);
3417 	pcs_lctl &= ~E1000_PCS_LCTL_AN_ENABLE;
3418 	E1000_WRITE_REG(hw, E1000_PCS_LCTL, pcs_lctl);
3419 
3420 	/* Set the Copper/Fiber Switch Control - CONNSW register */
3421 	connsw = E1000_READ_REG(hw, E1000_CONNSW);
3422 	connsw &= ~E1000_CONNSW_ENRGSRC;
3423 	E1000_WRITE_REG(hw, E1000_CONNSW, connsw);
3424 }
3425 
3426 #pragma inline(igb_intr_rx_work)
3427 /*
3428  * igb_intr_rx_work - rx processing of ISR
3429  */
3430 static void
3431 igb_intr_rx_work(igb_rx_ring_t *rx_ring)
3432 {
3433 	mblk_t *mp;
3434 
3435 	mutex_enter(&rx_ring->rx_lock);
3436 	mp = igb_rx(rx_ring, IGB_NO_POLL);
3437 	mutex_exit(&rx_ring->rx_lock);
3438 
3439 	if (mp != NULL)
3440 		mac_rx_ring(rx_ring->igb->mac_hdl, rx_ring->ring_handle, mp,
3441 		    rx_ring->ring_gen_num);
3442 }
3443 
3444 #pragma inline(igb_intr_tx_work)
3445 /*
3446  * igb_intr_tx_work - tx processing of ISR
3447  */
3448 static void
3449 igb_intr_tx_work(igb_tx_ring_t *tx_ring)
3450 {
3451 	/* Recycle the tx descriptors */
3452 	tx_ring->tx_recycle(tx_ring);
3453 
3454 	/* Schedule the re-transmit */
3455 	if (tx_ring->reschedule &&
3456 	    (tx_ring->tbd_free >= tx_ring->resched_thresh)) {
3457 		tx_ring->reschedule = B_FALSE;
3458 		mac_tx_ring_update(tx_ring->igb->mac_hdl, tx_ring->ring_handle);
3459 		IGB_DEBUG_STAT(tx_ring->stat_reschedule);
3460 	}
3461 }
3462 
3463 #pragma inline(igb_intr_link_work)
3464 /*
3465  * igb_intr_link_work - link-status-change processing of ISR
3466  */
3467 static void
3468 igb_intr_link_work(igb_t *igb)
3469 {
3470 	boolean_t link_changed;
3471 
3472 	igb_stop_watchdog_timer(igb);
3473 
3474 	mutex_enter(&igb->gen_lock);
3475 
3476 	/*
3477 	 * Because we got a link-status-change interrupt, force
3478 	 * e1000_check_for_link() to look at phy
3479 	 */
3480 	igb->hw.mac.get_link_status = B_TRUE;
3481 
3482 	/* igb_link_check takes care of link status change */
3483 	link_changed = igb_link_check(igb);
3484 
3485 	/* Get new phy state */
3486 	igb_get_phy_state(igb);
3487 
3488 	mutex_exit(&igb->gen_lock);
3489 
3490 	if (link_changed)
3491 		mac_link_update(igb->mac_hdl, igb->link_state);
3492 
3493 	igb_start_watchdog_timer(igb);
3494 }
3495 
3496 /*
3497  * igb_intr_legacy - Interrupt handler for legacy interrupts
3498  */
3499 static uint_t
3500 igb_intr_legacy(void *arg1, void *arg2)
3501 {
3502 	igb_t *igb = (igb_t *)arg1;
3503 	igb_tx_ring_t *tx_ring;
3504 	uint32_t icr;
3505 	mblk_t *mp;
3506 	boolean_t tx_reschedule;
3507 	boolean_t link_changed;
3508 	uint_t result;
3509 
3510 	_NOTE(ARGUNUSED(arg2));
3511 
3512 	mutex_enter(&igb->gen_lock);
3513 
3514 	if (igb->igb_state & IGB_SUSPENDED) {
3515 		mutex_exit(&igb->gen_lock);
3516 		return (DDI_INTR_UNCLAIMED);
3517 	}
3518 
3519 	mp = NULL;
3520 	tx_reschedule = B_FALSE;
3521 	link_changed = B_FALSE;
3522 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
3523 
3524 	if (icr & E1000_ICR_INT_ASSERTED) {
3525 		/*
3526 		 * E1000_ICR_INT_ASSERTED bit was set:
3527 		 * Read(Clear) the ICR, claim this interrupt,
3528 		 * look for work to do.
3529 		 */
3530 		ASSERT(igb->num_rx_rings == 1);
3531 		ASSERT(igb->num_tx_rings == 1);
3532 
3533 		if (icr & E1000_ICR_RXT0) {
3534 			mp = igb_rx(&igb->rx_rings[0], IGB_NO_POLL);
3535 		}
3536 
3537 		if (icr & E1000_ICR_TXDW) {
3538 			tx_ring = &igb->tx_rings[0];
3539 
3540 			/* Recycle the tx descriptors */
3541 			tx_ring->tx_recycle(tx_ring);
3542 
3543 			/* Schedule the re-transmit */
3544 			tx_reschedule = (tx_ring->reschedule &&
3545 			    (tx_ring->tbd_free >= tx_ring->resched_thresh));
3546 		}
3547 
3548 		if (icr & E1000_ICR_LSC) {
3549 			/*
3550 			 * Because we got a link-status-change interrupt, force
3551 			 * e1000_check_for_link() to look at phy
3552 			 */
3553 			igb->hw.mac.get_link_status = B_TRUE;
3554 
3555 			/* igb_link_check takes care of link status change */
3556 			link_changed = igb_link_check(igb);
3557 
3558 			/* Get new phy state */
3559 			igb_get_phy_state(igb);
3560 		}
3561 
3562 		result = DDI_INTR_CLAIMED;
3563 	} else {
3564 		/*
3565 		 * E1000_ICR_INT_ASSERTED bit was not set:
3566 		 * Don't claim this interrupt.
3567 		 */
3568 		result = DDI_INTR_UNCLAIMED;
3569 	}
3570 
3571 	mutex_exit(&igb->gen_lock);
3572 
3573 	/*
3574 	 * Do the following work outside of the gen_lock
3575 	 */
3576 	if (mp != NULL)
3577 		mac_rx(igb->mac_hdl, NULL, mp);
3578 
3579 	if (tx_reschedule)  {
3580 		tx_ring->reschedule = B_FALSE;
3581 		mac_tx_ring_update(igb->mac_hdl, tx_ring->ring_handle);
3582 		IGB_DEBUG_STAT(tx_ring->stat_reschedule);
3583 	}
3584 
3585 	if (link_changed)
3586 		mac_link_update(igb->mac_hdl, igb->link_state);
3587 
3588 	return (result);
3589 }
3590 
3591 /*
3592  * igb_intr_msi - Interrupt handler for MSI
3593  */
3594 static uint_t
3595 igb_intr_msi(void *arg1, void *arg2)
3596 {
3597 	igb_t *igb = (igb_t *)arg1;
3598 	uint32_t icr;
3599 
3600 	_NOTE(ARGUNUSED(arg2));
3601 
3602 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
3603 
3604 	/*
3605 	 * For MSI interrupt, we have only one vector,
3606 	 * so we have only one rx ring and one tx ring enabled.
3607 	 */
3608 	ASSERT(igb->num_rx_rings == 1);
3609 	ASSERT(igb->num_tx_rings == 1);
3610 
3611 	if (icr & E1000_ICR_RXT0) {
3612 		igb_intr_rx_work(&igb->rx_rings[0]);
3613 	}
3614 
3615 	if (icr & E1000_ICR_TXDW) {
3616 		igb_intr_tx_work(&igb->tx_rings[0]);
3617 	}
3618 
3619 	if (icr & E1000_ICR_LSC) {
3620 		igb_intr_link_work(igb);
3621 	}
3622 
3623 	return (DDI_INTR_CLAIMED);
3624 }
3625 
3626 /*
3627  * igb_intr_rx - Interrupt handler for rx
3628  */
3629 static uint_t
3630 igb_intr_rx(void *arg1, void *arg2)
3631 {
3632 	igb_rx_ring_t *rx_ring = (igb_rx_ring_t *)arg1;
3633 
3634 	_NOTE(ARGUNUSED(arg2));
3635 
3636 	/*
3637 	 * Only used via MSI-X vector so don't check cause bits
3638 	 * and only clean the given ring.
3639 	 */
3640 	igb_intr_rx_work(rx_ring);
3641 
3642 	return (DDI_INTR_CLAIMED);
3643 }
3644 
3645 /*
3646  * igb_intr_tx - Interrupt handler for tx
3647  */
3648 static uint_t
3649 igb_intr_tx(void *arg1, void *arg2)
3650 {
3651 	igb_tx_ring_t *tx_ring = (igb_tx_ring_t *)arg1;
3652 
3653 	_NOTE(ARGUNUSED(arg2));
3654 
3655 	/*
3656 	 * Only used via MSI-X vector so don't check cause bits
3657 	 * and only clean the given ring.
3658 	 */
3659 	igb_intr_tx_work(tx_ring);
3660 
3661 	return (DDI_INTR_CLAIMED);
3662 }
3663 
3664 /*
3665  * igb_intr_tx_other - Interrupt handler for both tx and other
3666  *
3667  */
3668 static uint_t
3669 igb_intr_tx_other(void *arg1, void *arg2)
3670 {
3671 	igb_t *igb = (igb_t *)arg1;
3672 	uint32_t icr;
3673 
3674 	_NOTE(ARGUNUSED(arg2));
3675 
3676 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
3677 
3678 	/*
3679 	 * Look for tx reclaiming work first. Remember, in the
3680 	 * case of only interrupt sharing, only one tx ring is
3681 	 * used
3682 	 */
3683 	igb_intr_tx_work(&igb->tx_rings[0]);
3684 
3685 	/*
3686 	 * Need check cause bits and only link change will
3687 	 * be processed
3688 	 */
3689 	if (icr & E1000_ICR_LSC) {
3690 		igb_intr_link_work(igb);
3691 	}
3692 
3693 	return (DDI_INTR_CLAIMED);
3694 }
3695 
3696 /*
3697  * igb_alloc_intrs - Allocate interrupts for the driver
3698  *
3699  * Normal sequence is to try MSI-X; if not sucessful, try MSI;
3700  * if not successful, try Legacy.
3701  * igb->intr_force can be used to force sequence to start with
3702  * any of the 3 types.
3703  * If MSI-X is not used, number of tx/rx rings is forced to 1.
3704  */
3705 static int
3706 igb_alloc_intrs(igb_t *igb)
3707 {
3708 	dev_info_t *devinfo;
3709 	int intr_types;
3710 	int rc;
3711 
3712 	devinfo = igb->dip;
3713 
3714 	/* Get supported interrupt types */
3715 	rc = ddi_intr_get_supported_types(devinfo, &intr_types);
3716 
3717 	if (rc != DDI_SUCCESS) {
3718 		igb_log(igb,
3719 		    "Get supported interrupt types failed: %d", rc);
3720 		return (IGB_FAILURE);
3721 	}
3722 	IGB_DEBUGLOG_1(igb, "Supported interrupt types: %x", intr_types);
3723 
3724 	igb->intr_type = 0;
3725 
3726 	/* Install MSI-X interrupts */
3727 	if ((intr_types & DDI_INTR_TYPE_MSIX) &&
3728 	    (igb->intr_force <= IGB_INTR_MSIX)) {
3729 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_MSIX);
3730 
3731 		if (rc == IGB_SUCCESS)
3732 			return (IGB_SUCCESS);
3733 
3734 		igb_log(igb,
3735 		    "Allocate MSI-X failed, trying MSI interrupts...");
3736 	}
3737 
3738 	/* MSI-X not used, force rings to 1 */
3739 	igb->num_rx_rings = 1;
3740 	igb->num_tx_rings = 1;
3741 	igb_log(igb,
3742 	    "MSI-X not used, force rx and tx queue number to 1");
3743 
3744 	/* Install MSI interrupts */
3745 	if ((intr_types & DDI_INTR_TYPE_MSI) &&
3746 	    (igb->intr_force <= IGB_INTR_MSI)) {
3747 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_MSI);
3748 
3749 		if (rc == IGB_SUCCESS)
3750 			return (IGB_SUCCESS);
3751 
3752 		igb_log(igb,
3753 		    "Allocate MSI failed, trying Legacy interrupts...");
3754 	}
3755 
3756 	/* Install legacy interrupts */
3757 	if (intr_types & DDI_INTR_TYPE_FIXED) {
3758 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_FIXED);
3759 
3760 		if (rc == IGB_SUCCESS)
3761 			return (IGB_SUCCESS);
3762 
3763 		igb_log(igb,
3764 		    "Allocate Legacy interrupts failed");
3765 	}
3766 
3767 	/* If none of the 3 types succeeded, return failure */
3768 	return (IGB_FAILURE);
3769 }
3770 
3771 /*
3772  * igb_alloc_intr_handles - Allocate interrupt handles.
3773  *
3774  * For legacy and MSI, only 1 handle is needed.  For MSI-X,
3775  * if fewer than 2 handles are available, return failure.
3776  * Upon success, this sets the number of Rx rings to a number that
3777  * matches the handles available for Rx interrupts.
3778  */
3779 static int
3780 igb_alloc_intr_handles(igb_t *igb, int intr_type)
3781 {
3782 	dev_info_t *devinfo;
3783 	int orig, request, count, avail, actual;
3784 	int diff, minimum;
3785 	int rc;
3786 
3787 	devinfo = igb->dip;
3788 
3789 	switch (intr_type) {
3790 	case DDI_INTR_TYPE_FIXED:
3791 		request = 1;	/* Request 1 legacy interrupt handle */
3792 		minimum = 1;
3793 		IGB_DEBUGLOG_0(igb, "interrupt type: legacy");
3794 		break;
3795 
3796 	case DDI_INTR_TYPE_MSI:
3797 		request = 1;	/* Request 1 MSI interrupt handle */
3798 		minimum = 1;
3799 		IGB_DEBUGLOG_0(igb, "interrupt type: MSI");
3800 		break;
3801 
3802 	case DDI_INTR_TYPE_MSIX:
3803 		/*
3804 		 * Number of vectors for the adapter is
3805 		 * # rx rings + # tx rings
3806 		 * One of tx vectors is for tx & other
3807 		 */
3808 		request = igb->num_rx_rings + igb->num_tx_rings;
3809 		orig = request;
3810 		minimum = 2;
3811 		IGB_DEBUGLOG_0(igb, "interrupt type: MSI-X");
3812 		break;
3813 
3814 	default:
3815 		igb_log(igb,
3816 		    "invalid call to igb_alloc_intr_handles(): %d\n",
3817 		    intr_type);
3818 		return (IGB_FAILURE);
3819 	}
3820 	IGB_DEBUGLOG_2(igb, "interrupt handles requested: %d  minimum: %d",
3821 	    request, minimum);
3822 
3823 	/*
3824 	 * Get number of supported interrupts
3825 	 */
3826 	rc = ddi_intr_get_nintrs(devinfo, intr_type, &count);
3827 	if ((rc != DDI_SUCCESS) || (count < minimum)) {
3828 		igb_log(igb,
3829 		    "Get supported interrupt number failed. "
3830 		    "Return: %d, count: %d", rc, count);
3831 		return (IGB_FAILURE);
3832 	}
3833 	IGB_DEBUGLOG_1(igb, "interrupts supported: %d", count);
3834 
3835 	/*
3836 	 * Get number of available interrupts
3837 	 */
3838 	rc = ddi_intr_get_navail(devinfo, intr_type, &avail);
3839 	if ((rc != DDI_SUCCESS) || (avail < minimum)) {
3840 		igb_log(igb,
3841 		    "Get available interrupt number failed. "
3842 		    "Return: %d, available: %d", rc, avail);
3843 		return (IGB_FAILURE);
3844 	}
3845 	IGB_DEBUGLOG_1(igb, "interrupts available: %d", avail);
3846 
3847 	if (avail < request) {
3848 		igb_log(igb, "Request %d handles, %d available",
3849 		    request, avail);
3850 		request = avail;
3851 	}
3852 
3853 	actual = 0;
3854 	igb->intr_cnt = 0;
3855 
3856 	/*
3857 	 * Allocate an array of interrupt handles
3858 	 */
3859 	igb->intr_size = request * sizeof (ddi_intr_handle_t);
3860 	igb->htable = kmem_alloc(igb->intr_size, KM_SLEEP);
3861 
3862 	rc = ddi_intr_alloc(devinfo, igb->htable, intr_type, 0,
3863 	    request, &actual, DDI_INTR_ALLOC_NORMAL);
3864 	if (rc != DDI_SUCCESS) {
3865 		igb_log(igb, "Allocate interrupts failed. "
3866 		    "return: %d, request: %d, actual: %d",
3867 		    rc, request, actual);
3868 		goto alloc_handle_fail;
3869 	}
3870 	IGB_DEBUGLOG_1(igb, "interrupts actually allocated: %d", actual);
3871 
3872 	igb->intr_cnt = actual;
3873 
3874 	if (actual < minimum) {
3875 		igb_log(igb, "Insufficient interrupt handles allocated: %d",
3876 		    actual);
3877 		goto alloc_handle_fail;
3878 	}
3879 
3880 	/*
3881 	 * For MSI-X, actual might force us to reduce number of tx & rx rings
3882 	 */
3883 	if ((intr_type == DDI_INTR_TYPE_MSIX) && (orig > actual)) {
3884 		diff = orig - actual;
3885 		if (diff < igb->num_tx_rings) {
3886 			igb_log(igb,
3887 			    "MSI-X vectors force Tx queue number to %d",
3888 			    igb->num_tx_rings - diff);
3889 			igb->num_tx_rings -= diff;
3890 		} else {
3891 			igb_log(igb,
3892 			    "MSI-X vectors force Tx queue number to 1");
3893 			igb->num_tx_rings = 1;
3894 
3895 			igb_log(igb,
3896 			    "MSI-X vectors force Rx queue number to %d",
3897 			    actual - 1);
3898 			igb->num_rx_rings = actual - 1;
3899 		}
3900 	}
3901 
3902 	/*
3903 	 * Get priority for first vector, assume remaining are all the same
3904 	 */
3905 	rc = ddi_intr_get_pri(igb->htable[0], &igb->intr_pri);
3906 	if (rc != DDI_SUCCESS) {
3907 		igb_log(igb,
3908 		    "Get interrupt priority failed: %d", rc);
3909 		goto alloc_handle_fail;
3910 	}
3911 
3912 	rc = ddi_intr_get_cap(igb->htable[0], &igb->intr_cap);
3913 	if (rc != DDI_SUCCESS) {
3914 		igb_log(igb,
3915 		    "Get interrupt cap failed: %d", rc);
3916 		goto alloc_handle_fail;
3917 	}
3918 
3919 	igb->intr_type = intr_type;
3920 
3921 	return (IGB_SUCCESS);
3922 
3923 alloc_handle_fail:
3924 	igb_rem_intrs(igb);
3925 
3926 	return (IGB_FAILURE);
3927 }
3928 
3929 /*
3930  * igb_add_intr_handlers - Add interrupt handlers based on the interrupt type
3931  *
3932  * Before adding the interrupt handlers, the interrupt vectors have
3933  * been allocated, and the rx/tx rings have also been allocated.
3934  */
3935 static int
3936 igb_add_intr_handlers(igb_t *igb)
3937 {
3938 	igb_rx_ring_t *rx_ring;
3939 	igb_tx_ring_t *tx_ring;
3940 	int vector;
3941 	int rc;
3942 	int i;
3943 
3944 	vector = 0;
3945 
3946 	switch (igb->intr_type) {
3947 	case DDI_INTR_TYPE_MSIX:
3948 		/* Add interrupt handler for tx + other */
3949 		tx_ring = &igb->tx_rings[0];
3950 		rc = ddi_intr_add_handler(igb->htable[vector],
3951 		    (ddi_intr_handler_t *)igb_intr_tx_other,
3952 		    (void *)igb, NULL);
3953 
3954 		if (rc != DDI_SUCCESS) {
3955 			igb_log(igb,
3956 			    "Add tx/other interrupt handler failed: %d", rc);
3957 			return (IGB_FAILURE);
3958 		}
3959 		tx_ring->intr_vector = vector;
3960 		vector++;
3961 
3962 		/* Add interrupt handler for each rx ring */
3963 		for (i = 0; i < igb->num_rx_rings; i++) {
3964 			rx_ring = &igb->rx_rings[i];
3965 
3966 			rc = ddi_intr_add_handler(igb->htable[vector],
3967 			    (ddi_intr_handler_t *)igb_intr_rx,
3968 			    (void *)rx_ring, NULL);
3969 
3970 			if (rc != DDI_SUCCESS) {
3971 				igb_log(igb,
3972 				    "Add rx interrupt handler failed. "
3973 				    "return: %d, rx ring: %d", rc, i);
3974 				for (vector--; vector >= 0; vector--) {
3975 					(void) ddi_intr_remove_handler(
3976 					    igb->htable[vector]);
3977 				}
3978 				return (IGB_FAILURE);
3979 			}
3980 
3981 			rx_ring->intr_vector = vector;
3982 
3983 			vector++;
3984 		}
3985 
3986 		/* Add interrupt handler for each tx ring from 2nd ring */
3987 		for (i = 1; i < igb->num_tx_rings; i++) {
3988 			tx_ring = &igb->tx_rings[i];
3989 
3990 			rc = ddi_intr_add_handler(igb->htable[vector],
3991 			    (ddi_intr_handler_t *)igb_intr_tx,
3992 			    (void *)tx_ring, NULL);
3993 
3994 			if (rc != DDI_SUCCESS) {
3995 				igb_log(igb,
3996 				    "Add tx interrupt handler failed. "
3997 				    "return: %d, tx ring: %d", rc, i);
3998 				for (vector--; vector >= 0; vector--) {
3999 					(void) ddi_intr_remove_handler(
4000 					    igb->htable[vector]);
4001 				}
4002 				return (IGB_FAILURE);
4003 			}
4004 
4005 			tx_ring->intr_vector = vector;
4006 
4007 			vector++;
4008 		}
4009 
4010 		break;
4011 
4012 	case DDI_INTR_TYPE_MSI:
4013 		/* Add interrupt handlers for the only vector */
4014 		rc = ddi_intr_add_handler(igb->htable[vector],
4015 		    (ddi_intr_handler_t *)igb_intr_msi,
4016 		    (void *)igb, NULL);
4017 
4018 		if (rc != DDI_SUCCESS) {
4019 			igb_log(igb,
4020 			    "Add MSI interrupt handler failed: %d", rc);
4021 			return (IGB_FAILURE);
4022 		}
4023 
4024 		rx_ring = &igb->rx_rings[0];
4025 		rx_ring->intr_vector = vector;
4026 
4027 		vector++;
4028 		break;
4029 
4030 	case DDI_INTR_TYPE_FIXED:
4031 		/* Add interrupt handlers for the only vector */
4032 		rc = ddi_intr_add_handler(igb->htable[vector],
4033 		    (ddi_intr_handler_t *)igb_intr_legacy,
4034 		    (void *)igb, NULL);
4035 
4036 		if (rc != DDI_SUCCESS) {
4037 			igb_log(igb,
4038 			    "Add legacy interrupt handler failed: %d", rc);
4039 			return (IGB_FAILURE);
4040 		}
4041 
4042 		rx_ring = &igb->rx_rings[0];
4043 		rx_ring->intr_vector = vector;
4044 
4045 		vector++;
4046 		break;
4047 
4048 	default:
4049 		return (IGB_FAILURE);
4050 	}
4051 
4052 	ASSERT(vector == igb->intr_cnt);
4053 
4054 	return (IGB_SUCCESS);
4055 }
4056 
4057 /*
4058  * igb_setup_adapter_msix - setup the adapter to use MSI-X interrupts
4059  *
4060  * For each vector enabled on the adapter, Set the MSIXBM register accordingly
4061  */
4062 static void
4063 igb_setup_adapter_msix(igb_t *igb)
4064 {
4065 	uint32_t eims = 0;
4066 	int i, vector;
4067 	struct e1000_hw *hw = &igb->hw;
4068 
4069 	/*
4070 	 * Set vector for other causes, NOTE assumption that it is vector 0
4071 	 */
4072 	vector = 0;
4073 
4074 	igb->eims_mask = E1000_EICR_TX_QUEUE0 | E1000_EICR_OTHER;
4075 	E1000_WRITE_REG(hw, E1000_MSIXBM(vector), igb->eims_mask);
4076 	vector++;
4077 
4078 	for (i = 0; i < igb->num_rx_rings; i++) {
4079 		/*
4080 		 * Set vector for each rx ring
4081 		 */
4082 		eims = (E1000_EICR_RX_QUEUE0 << i);
4083 		E1000_WRITE_REG(hw, E1000_MSIXBM(vector), eims);
4084 
4085 		/*
4086 		 * Accumulate bits to enable in igb_enable_adapter_interrupts()
4087 		 */
4088 		igb->eims_mask |= eims;
4089 
4090 		vector++;
4091 	}
4092 
4093 	for (i = 1; i < igb->num_tx_rings; i++) {
4094 		/*
4095 		 * Set vector for each tx ring from 2nd tx ring
4096 		 */
4097 		eims = (E1000_EICR_TX_QUEUE0 << i);
4098 		E1000_WRITE_REG(hw, E1000_MSIXBM(vector), eims);
4099 
4100 		/*
4101 		 * Accumulate bits to enable in igb_enable_adapter_interrupts()
4102 		 */
4103 		igb->eims_mask |= eims;
4104 
4105 		vector++;
4106 	}
4107 
4108 	ASSERT(vector == igb->intr_cnt);
4109 
4110 	/*
4111 	 * Disable IAM for ICR interrupt bits
4112 	 */
4113 	E1000_WRITE_REG(hw, E1000_IAM, 0);
4114 	E1000_WRITE_FLUSH(hw);
4115 }
4116 
4117 /*
4118  * igb_rem_intr_handlers - remove the interrupt handlers
4119  */
4120 static void
4121 igb_rem_intr_handlers(igb_t *igb)
4122 {
4123 	int i;
4124 	int rc;
4125 
4126 	for (i = 0; i < igb->intr_cnt; i++) {
4127 		rc = ddi_intr_remove_handler(igb->htable[i]);
4128 		if (rc != DDI_SUCCESS) {
4129 			IGB_DEBUGLOG_1(igb,
4130 			    "Remove intr handler failed: %d", rc);
4131 		}
4132 	}
4133 }
4134 
4135 /*
4136  * igb_rem_intrs - remove the allocated interrupts
4137  */
4138 static void
4139 igb_rem_intrs(igb_t *igb)
4140 {
4141 	int i;
4142 	int rc;
4143 
4144 	for (i = 0; i < igb->intr_cnt; i++) {
4145 		rc = ddi_intr_free(igb->htable[i]);
4146 		if (rc != DDI_SUCCESS) {
4147 			IGB_DEBUGLOG_1(igb,
4148 			    "Free intr failed: %d", rc);
4149 		}
4150 	}
4151 
4152 	kmem_free(igb->htable, igb->intr_size);
4153 	igb->htable = NULL;
4154 }
4155 
4156 /*
4157  * igb_enable_intrs - enable all the ddi interrupts
4158  */
4159 static int
4160 igb_enable_intrs(igb_t *igb)
4161 {
4162 	int i;
4163 	int rc;
4164 
4165 	/* Enable interrupts */
4166 	if (igb->intr_cap & DDI_INTR_FLAG_BLOCK) {
4167 		/* Call ddi_intr_block_enable() for MSI */
4168 		rc = ddi_intr_block_enable(igb->htable, igb->intr_cnt);
4169 		if (rc != DDI_SUCCESS) {
4170 			igb_log(igb,
4171 			    "Enable block intr failed: %d", rc);
4172 			return (IGB_FAILURE);
4173 		}
4174 	} else {
4175 		/* Call ddi_intr_enable() for Legacy/MSI non block enable */
4176 		for (i = 0; i < igb->intr_cnt; i++) {
4177 			rc = ddi_intr_enable(igb->htable[i]);
4178 			if (rc != DDI_SUCCESS) {
4179 				igb_log(igb,
4180 				    "Enable intr failed: %d", rc);
4181 				return (IGB_FAILURE);
4182 			}
4183 		}
4184 	}
4185 
4186 	return (IGB_SUCCESS);
4187 }
4188 
4189 /*
4190  * igb_disable_intrs - disable all the ddi interrupts
4191  */
4192 static int
4193 igb_disable_intrs(igb_t *igb)
4194 {
4195 	int i;
4196 	int rc;
4197 
4198 	/* Disable all interrupts */
4199 	if (igb->intr_cap & DDI_INTR_FLAG_BLOCK) {
4200 		rc = ddi_intr_block_disable(igb->htable, igb->intr_cnt);
4201 		if (rc != DDI_SUCCESS) {
4202 			igb_log(igb,
4203 			    "Disable block intr failed: %d", rc);
4204 			return (IGB_FAILURE);
4205 		}
4206 	} else {
4207 		for (i = 0; i < igb->intr_cnt; i++) {
4208 			rc = ddi_intr_disable(igb->htable[i]);
4209 			if (rc != DDI_SUCCESS) {
4210 				igb_log(igb,
4211 				    "Disable intr failed: %d", rc);
4212 				return (IGB_FAILURE);
4213 			}
4214 		}
4215 	}
4216 
4217 	return (IGB_SUCCESS);
4218 }
4219 
4220 /*
4221  * igb_get_phy_state - Get and save the parameters read from PHY registers
4222  */
4223 static void
4224 igb_get_phy_state(igb_t *igb)
4225 {
4226 	struct e1000_hw *hw = &igb->hw;
4227 	uint16_t phy_ctrl;
4228 	uint16_t phy_status;
4229 	uint16_t phy_an_adv;
4230 	uint16_t phy_an_exp;
4231 	uint16_t phy_ext_status;
4232 	uint16_t phy_1000t_ctrl;
4233 	uint16_t phy_1000t_status;
4234 	uint16_t phy_lp_able;
4235 
4236 	ASSERT(mutex_owned(&igb->gen_lock));
4237 
4238 	(void) e1000_read_phy_reg(hw, PHY_CONTROL, &phy_ctrl);
4239 	(void) e1000_read_phy_reg(hw, PHY_STATUS, &phy_status);
4240 	(void) e1000_read_phy_reg(hw, PHY_AUTONEG_ADV, &phy_an_adv);
4241 	(void) e1000_read_phy_reg(hw, PHY_AUTONEG_EXP, &phy_an_exp);
4242 	(void) e1000_read_phy_reg(hw, PHY_EXT_STATUS, &phy_ext_status);
4243 	(void) e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_1000t_ctrl);
4244 	(void) e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_1000t_status);
4245 	(void) e1000_read_phy_reg(hw, PHY_LP_ABILITY, &phy_lp_able);
4246 
4247 	igb->param_autoneg_cap =
4248 	    (phy_status & MII_SR_AUTONEG_CAPS) ? 1 : 0;
4249 	igb->param_pause_cap =
4250 	    (phy_an_adv & NWAY_AR_PAUSE) ? 1 : 0;
4251 	igb->param_asym_pause_cap =
4252 	    (phy_an_adv & NWAY_AR_ASM_DIR) ? 1 : 0;
4253 	igb->param_1000fdx_cap = ((phy_ext_status & IEEE_ESR_1000T_FD_CAPS) ||
4254 	    (phy_ext_status & IEEE_ESR_1000X_FD_CAPS)) ? 1 : 0;
4255 	igb->param_1000hdx_cap = ((phy_ext_status & IEEE_ESR_1000T_HD_CAPS) ||
4256 	    (phy_ext_status & IEEE_ESR_1000X_HD_CAPS)) ? 1 : 0;
4257 	igb->param_100t4_cap =
4258 	    (phy_status & MII_SR_100T4_CAPS) ? 1 : 0;
4259 	igb->param_100fdx_cap = ((phy_status & MII_SR_100X_FD_CAPS) ||
4260 	    (phy_status & MII_SR_100T2_FD_CAPS)) ? 1 : 0;
4261 	igb->param_100hdx_cap = ((phy_status & MII_SR_100X_HD_CAPS) ||
4262 	    (phy_status & MII_SR_100T2_HD_CAPS)) ? 1 : 0;
4263 	igb->param_10fdx_cap =
4264 	    (phy_status & MII_SR_10T_FD_CAPS) ? 1 : 0;
4265 	igb->param_10hdx_cap =
4266 	    (phy_status & MII_SR_10T_HD_CAPS) ? 1 : 0;
4267 	igb->param_rem_fault =
4268 	    (phy_status & MII_SR_REMOTE_FAULT) ? 1 : 0;
4269 
4270 	igb->param_adv_autoneg_cap = hw->mac.autoneg;
4271 	igb->param_adv_pause_cap =
4272 	    (phy_an_adv & NWAY_AR_PAUSE) ? 1 : 0;
4273 	igb->param_adv_asym_pause_cap =
4274 	    (phy_an_adv & NWAY_AR_ASM_DIR) ? 1 : 0;
4275 	igb->param_adv_1000hdx_cap =
4276 	    (phy_1000t_ctrl & CR_1000T_HD_CAPS) ? 1 : 0;
4277 	igb->param_adv_100t4_cap =
4278 	    (phy_an_adv & NWAY_AR_100T4_CAPS) ? 1 : 0;
4279 	igb->param_adv_rem_fault =
4280 	    (phy_an_adv & NWAY_AR_REMOTE_FAULT) ? 1 : 0;
4281 	if (igb->param_adv_autoneg_cap == 1) {
4282 		igb->param_adv_1000fdx_cap =
4283 		    (phy_1000t_ctrl & CR_1000T_FD_CAPS) ? 1 : 0;
4284 		igb->param_adv_100fdx_cap =
4285 		    (phy_an_adv & NWAY_AR_100TX_FD_CAPS) ? 1 : 0;
4286 		igb->param_adv_100hdx_cap =
4287 		    (phy_an_adv & NWAY_AR_100TX_HD_CAPS) ? 1 : 0;
4288 		igb->param_adv_10fdx_cap =
4289 		    (phy_an_adv & NWAY_AR_10T_FD_CAPS) ? 1 : 0;
4290 		igb->param_adv_10hdx_cap =
4291 		    (phy_an_adv & NWAY_AR_10T_HD_CAPS) ? 1 : 0;
4292 	}
4293 
4294 	igb->param_lp_autoneg_cap =
4295 	    (phy_an_exp & NWAY_ER_LP_NWAY_CAPS) ? 1 : 0;
4296 	igb->param_lp_pause_cap =
4297 	    (phy_lp_able & NWAY_LPAR_PAUSE) ? 1 : 0;
4298 	igb->param_lp_asym_pause_cap =
4299 	    (phy_lp_able & NWAY_LPAR_ASM_DIR) ? 1 : 0;
4300 	igb->param_lp_1000fdx_cap =
4301 	    (phy_1000t_status & SR_1000T_LP_FD_CAPS) ? 1 : 0;
4302 	igb->param_lp_1000hdx_cap =
4303 	    (phy_1000t_status & SR_1000T_LP_HD_CAPS) ? 1 : 0;
4304 	igb->param_lp_100t4_cap =
4305 	    (phy_lp_able & NWAY_LPAR_100T4_CAPS) ? 1 : 0;
4306 	igb->param_lp_100fdx_cap =
4307 	    (phy_lp_able & NWAY_LPAR_100TX_FD_CAPS) ? 1 : 0;
4308 	igb->param_lp_100hdx_cap =
4309 	    (phy_lp_able & NWAY_LPAR_100TX_HD_CAPS) ? 1 : 0;
4310 	igb->param_lp_10fdx_cap =
4311 	    (phy_lp_able & NWAY_LPAR_10T_FD_CAPS) ? 1 : 0;
4312 	igb->param_lp_10hdx_cap =
4313 	    (phy_lp_able & NWAY_LPAR_10T_HD_CAPS) ? 1 : 0;
4314 	igb->param_lp_rem_fault =
4315 	    (phy_lp_able & NWAY_LPAR_REMOTE_FAULT) ? 1 : 0;
4316 }
4317 
4318 /*
4319  * igb_get_driver_control
4320  */
4321 static void
4322 igb_get_driver_control(struct e1000_hw *hw)
4323 {
4324 	uint32_t ctrl_ext;
4325 
4326 	/* Notify firmware that driver is in control of device */
4327 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
4328 	ctrl_ext |= E1000_CTRL_EXT_DRV_LOAD;
4329 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
4330 }
4331 
4332 /*
4333  * igb_release_driver_control
4334  */
4335 static void
4336 igb_release_driver_control(struct e1000_hw *hw)
4337 {
4338 	uint32_t ctrl_ext;
4339 
4340 	/* Notify firmware that driver is no longer in control of device */
4341 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
4342 	ctrl_ext &= ~E1000_CTRL_EXT_DRV_LOAD;
4343 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
4344 }
4345 
4346 /*
4347  * igb_atomic_reserve - Atomic decrease operation
4348  */
4349 int
4350 igb_atomic_reserve(uint32_t *count_p, uint32_t n)
4351 {
4352 	uint32_t oldval;
4353 	uint32_t newval;
4354 
4355 	/* ATOMICALLY */
4356 	do {
4357 		oldval = *count_p;
4358 		if (oldval < n)
4359 			return (-1);
4360 		newval = oldval - n;
4361 	} while (atomic_cas_32(count_p, oldval, newval) != oldval);
4362 
4363 	return (newval);
4364 }
4365 
4366 /*
4367  * FMA support
4368  */
4369 
4370 int
4371 igb_check_acc_handle(ddi_acc_handle_t handle)
4372 {
4373 	ddi_fm_error_t de;
4374 
4375 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
4376 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
4377 	return (de.fme_status);
4378 }
4379 
4380 int
4381 igb_check_dma_handle(ddi_dma_handle_t handle)
4382 {
4383 	ddi_fm_error_t de;
4384 
4385 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
4386 	return (de.fme_status);
4387 }
4388 
4389 /*
4390  * The IO fault service error handling callback function
4391  */
4392 /*ARGSUSED*/
4393 static int
4394 igb_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
4395 {
4396 	/*
4397 	 * as the driver can always deal with an error in any dma or
4398 	 * access handle, we can just return the fme_status value.
4399 	 */
4400 	pci_ereport_post(dip, err, NULL);
4401 	return (err->fme_status);
4402 }
4403 
4404 static void
4405 igb_fm_init(igb_t *igb)
4406 {
4407 	ddi_iblock_cookie_t iblk;
4408 	int fma_acc_flag, fma_dma_flag;
4409 
4410 	/* Only register with IO Fault Services if we have some capability */
4411 	if (igb->fm_capabilities & DDI_FM_ACCCHK_CAPABLE) {
4412 		igb_regs_acc_attr.devacc_attr_access = DDI_FLAGERR_ACC;
4413 		fma_acc_flag = 1;
4414 	} else {
4415 		igb_regs_acc_attr.devacc_attr_access = DDI_DEFAULT_ACC;
4416 		fma_acc_flag = 0;
4417 	}
4418 
4419 	if (igb->fm_capabilities & DDI_FM_DMACHK_CAPABLE) {
4420 		fma_dma_flag = 1;
4421 	} else {
4422 		fma_dma_flag = 0;
4423 	}
4424 
4425 	(void) igb_set_fma_flags(fma_acc_flag, fma_dma_flag);
4426 
4427 	if (igb->fm_capabilities) {
4428 
4429 		/* Register capabilities with IO Fault Services */
4430 		ddi_fm_init(igb->dip, &igb->fm_capabilities, &iblk);
4431 
4432 		/*
4433 		 * Initialize pci ereport capabilities if ereport capable
4434 		 */
4435 		if (DDI_FM_EREPORT_CAP(igb->fm_capabilities) ||
4436 		    DDI_FM_ERRCB_CAP(igb->fm_capabilities))
4437 			pci_ereport_setup(igb->dip);
4438 
4439 		/*
4440 		 * Register error callback if error callback capable
4441 		 */
4442 		if (DDI_FM_ERRCB_CAP(igb->fm_capabilities))
4443 			ddi_fm_handler_register(igb->dip,
4444 			    igb_fm_error_cb, (void*) igb);
4445 	}
4446 }
4447 
4448 static void
4449 igb_fm_fini(igb_t *igb)
4450 {
4451 	/* Only unregister FMA capabilities if we registered some */
4452 	if (igb->fm_capabilities) {
4453 
4454 		/*
4455 		 * Release any resources allocated by pci_ereport_setup()
4456 		 */
4457 		if (DDI_FM_EREPORT_CAP(igb->fm_capabilities) ||
4458 		    DDI_FM_ERRCB_CAP(igb->fm_capabilities))
4459 			pci_ereport_teardown(igb->dip);
4460 
4461 		/*
4462 		 * Un-register error callback if error callback capable
4463 		 */
4464 		if (DDI_FM_ERRCB_CAP(igb->fm_capabilities))
4465 			ddi_fm_handler_unregister(igb->dip);
4466 
4467 		/* Unregister from IO Fault Services */
4468 		ddi_fm_fini(igb->dip);
4469 	}
4470 }
4471 
4472 void
4473 igb_fm_ereport(igb_t *igb, char *detail)
4474 {
4475 	uint64_t ena;
4476 	char buf[FM_MAX_CLASS];
4477 
4478 	(void) snprintf(buf, FM_MAX_CLASS, "%s.%s", DDI_FM_DEVICE, detail);
4479 	ena = fm_ena_generate(0, FM_ENA_FMT1);
4480 	if (DDI_FM_EREPORT_CAP(igb->fm_capabilities)) {
4481 		ddi_fm_ereport_post(igb->dip, buf, ena, DDI_NOSLEEP,
4482 		    FM_VERSION, DATA_TYPE_UINT8, FM_EREPORT_VERS0, NULL);
4483 	}
4484 }
4485