xref: /titanic_44/usr/src/uts/common/io/igb/igb_main.c (revision 72ca8cc954e6ca79da549736925d4569dc23b239)
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_init_unicst(igb_t *);
64 static void igb_setup_multicst(igb_t *);
65 static void igb_get_phy_state(igb_t *);
66 static void igb_get_conf(igb_t *);
67 static int igb_get_prop(igb_t *, char *, int, int, int);
68 static boolean_t igb_is_link_up(igb_t *);
69 static boolean_t igb_link_check(igb_t *);
70 static void igb_local_timer(void *);
71 static void igb_arm_watchdog_timer(igb_t *);
72 static void igb_start_watchdog_timer(igb_t *);
73 static void igb_restart_watchdog_timer(igb_t *);
74 static void igb_stop_watchdog_timer(igb_t *);
75 static void igb_disable_adapter_interrupts(igb_t *);
76 static void igb_enable_adapter_interrupts(igb_t *);
77 static boolean_t is_valid_mac_addr(uint8_t *);
78 static boolean_t igb_stall_check(igb_t *);
79 static boolean_t igb_set_loopback_mode(igb_t *, uint32_t);
80 static void igb_set_external_loopback(igb_t *);
81 static void igb_set_internal_mac_loopback(igb_t *);
82 static void igb_set_internal_phy_loopback(igb_t *);
83 static void igb_set_internal_serdes_loopback(igb_t *);
84 static boolean_t igb_find_mac_address(igb_t *);
85 static int igb_alloc_intrs(igb_t *);
86 static int igb_alloc_intr_handles(igb_t *, int);
87 static int igb_add_intr_handlers(igb_t *);
88 static void igb_rem_intr_handlers(igb_t *);
89 static void igb_rem_intrs(igb_t *);
90 static int igb_enable_intrs(igb_t *);
91 static int igb_disable_intrs(igb_t *);
92 static void igb_setup_adapter_msix(igb_t *);
93 static uint_t igb_intr_legacy(void *, void *);
94 static uint_t igb_intr_msi(void *, void *);
95 static uint_t igb_intr_rx(void *, void *);
96 static uint_t igb_intr_tx_other(void *, void *);
97 static void igb_intr_rx_work(igb_rx_ring_t *);
98 static void igb_intr_tx_work(igb_tx_ring_t *);
99 static void igb_intr_other_work(igb_t *);
100 static void igb_get_driver_control(struct e1000_hw *);
101 static void igb_release_driver_control(struct e1000_hw *);
102 
103 static int igb_attach(dev_info_t *, ddi_attach_cmd_t);
104 static int igb_detach(dev_info_t *, ddi_detach_cmd_t);
105 static int igb_resume(dev_info_t *);
106 static int igb_suspend(dev_info_t *);
107 static int igb_quiesce(dev_info_t *);
108 static void igb_unconfigure(dev_info_t *, igb_t *);
109 static int igb_fm_error_cb(dev_info_t *, ddi_fm_error_t *,
110     const void *);
111 static void igb_fm_init(igb_t *);
112 static void igb_fm_fini(igb_t *);
113 
114 
115 static struct cb_ops igb_cb_ops = {
116 	nulldev,		/* cb_open */
117 	nulldev,		/* cb_close */
118 	nodev,			/* cb_strategy */
119 	nodev,			/* cb_print */
120 	nodev,			/* cb_dump */
121 	nodev,			/* cb_read */
122 	nodev,			/* cb_write */
123 	nodev,			/* cb_ioctl */
124 	nodev,			/* cb_devmap */
125 	nodev,			/* cb_mmap */
126 	nodev,			/* cb_segmap */
127 	nochpoll,		/* cb_chpoll */
128 	ddi_prop_op,		/* cb_prop_op */
129 	NULL,			/* cb_stream */
130 	D_MP | D_HOTPLUG,	/* cb_flag */
131 	CB_REV,			/* cb_rev */
132 	nodev,			/* cb_aread */
133 	nodev			/* cb_awrite */
134 };
135 
136 static struct dev_ops igb_dev_ops = {
137 	DEVO_REV,		/* devo_rev */
138 	0,			/* devo_refcnt */
139 	NULL,			/* devo_getinfo */
140 	nulldev,		/* devo_identify */
141 	nulldev,		/* devo_probe */
142 	igb_attach,		/* devo_attach */
143 	igb_detach,		/* devo_detach */
144 	nodev,			/* devo_reset */
145 	&igb_cb_ops,		/* devo_cb_ops */
146 	NULL,			/* devo_bus_ops */
147 	ddi_power,		/* devo_power */
148 	igb_quiesce,	/* devo_quiesce */
149 };
150 
151 static struct modldrv igb_modldrv = {
152 	&mod_driverops,		/* Type of module.  This one is a driver */
153 	ident,			/* Discription string */
154 	&igb_dev_ops,		/* driver ops */
155 };
156 
157 static struct modlinkage igb_modlinkage = {
158 	MODREV_1, &igb_modldrv, NULL
159 };
160 
161 /* Access attributes for register mapping */
162 ddi_device_acc_attr_t igb_regs_acc_attr = {
163 	DDI_DEVICE_ATTR_V0,
164 	DDI_STRUCTURE_LE_ACC,
165 	DDI_STRICTORDER_ACC,
166 	DDI_FLAGERR_ACC
167 };
168 
169 #define	IGB_M_CALLBACK_FLAGS	(MC_IOCTL | MC_GETCAPAB)
170 
171 static mac_callbacks_t igb_m_callbacks = {
172 	IGB_M_CALLBACK_FLAGS,
173 	igb_m_stat,
174 	igb_m_start,
175 	igb_m_stop,
176 	igb_m_promisc,
177 	igb_m_multicst,
178 	igb_m_unicst,
179 	igb_m_tx,
180 	NULL,
181 	igb_m_ioctl,
182 	igb_m_getcapab
183 };
184 
185 
186 /*
187  * Module Initialization Functions
188  */
189 
190 int
191 _init(void)
192 {
193 	int status;
194 
195 	mac_init_ops(&igb_dev_ops, MODULE_NAME);
196 
197 	status = mod_install(&igb_modlinkage);
198 
199 	if (status != DDI_SUCCESS) {
200 		mac_fini_ops(&igb_dev_ops);
201 	}
202 
203 	return (status);
204 }
205 
206 int
207 _fini(void)
208 {
209 	int status;
210 
211 	status = mod_remove(&igb_modlinkage);
212 
213 	if (status == DDI_SUCCESS) {
214 		mac_fini_ops(&igb_dev_ops);
215 	}
216 
217 	return (status);
218 
219 }
220 
221 int
222 _info(struct modinfo *modinfop)
223 {
224 	int status;
225 
226 	status = mod_info(&igb_modlinkage, modinfop);
227 
228 	return (status);
229 }
230 
231 /*
232  * igb_attach - driver attach
233  *
234  * This function is the device specific initialization entry
235  * point. This entry point is required and must be written.
236  * The DDI_ATTACH command must be provided in the attach entry
237  * point. When attach() is called with cmd set to DDI_ATTACH,
238  * all normal kernel services (such as kmem_alloc(9F)) are
239  * available for use by the driver.
240  *
241  * The attach() function will be called once for each instance
242  * of  the  device  on  the  system with cmd set to DDI_ATTACH.
243  * Until attach() succeeds, the only driver entry points which
244  * may be called are open(9E) and getinfo(9E).
245  */
246 static int
247 igb_attach(dev_info_t *devinfo, ddi_attach_cmd_t cmd)
248 {
249 	igb_t *igb;
250 	struct igb_osdep *osdep;
251 	struct e1000_hw *hw;
252 	int instance;
253 
254 	/*
255 	 * Check the command and perform corresponding operations
256 	 */
257 	switch (cmd) {
258 	default:
259 		return (DDI_FAILURE);
260 
261 	case DDI_RESUME:
262 		return (igb_resume(devinfo));
263 
264 	case DDI_ATTACH:
265 		break;
266 	}
267 
268 	/* Get the device instance */
269 	instance = ddi_get_instance(devinfo);
270 
271 	/* Allocate memory for the instance data structure */
272 	igb = kmem_zalloc(sizeof (igb_t), KM_SLEEP);
273 
274 	igb->dip = devinfo;
275 	igb->instance = instance;
276 
277 	hw = &igb->hw;
278 	osdep = &igb->osdep;
279 	hw->back = osdep;
280 	osdep->igb = igb;
281 
282 	/* Attach the instance pointer to the dev_info data structure */
283 	ddi_set_driver_private(devinfo, igb);
284 
285 
286 	/* Initialize for fma support */
287 	igb->fm_capabilities = igb_get_prop(igb, "fm-capable",
288 	    0, 0x0f,
289 	    DDI_FM_EREPORT_CAPABLE | DDI_FM_ACCCHK_CAPABLE |
290 	    DDI_FM_DMACHK_CAPABLE | DDI_FM_ERRCB_CAPABLE);
291 	igb_fm_init(igb);
292 	igb->attach_progress |= ATTACH_PROGRESS_FMINIT;
293 
294 	/*
295 	 * Map PCI config space registers
296 	 */
297 	if (pci_config_setup(devinfo, &osdep->cfg_handle) != DDI_SUCCESS) {
298 		igb_error(igb, "Failed to map PCI configurations");
299 		goto attach_fail;
300 	}
301 	igb->attach_progress |= ATTACH_PROGRESS_PCI_CONFIG;
302 
303 	/*
304 	 * Identify the chipset family
305 	 */
306 	if (igb_identify_hardware(igb) != IGB_SUCCESS) {
307 		igb_error(igb, "Failed to identify hardware");
308 		goto attach_fail;
309 	}
310 
311 	/*
312 	 * Map device registers
313 	 */
314 	if (igb_regs_map(igb) != IGB_SUCCESS) {
315 		igb_error(igb, "Failed to map device registers");
316 		goto attach_fail;
317 	}
318 	igb->attach_progress |= ATTACH_PROGRESS_REGS_MAP;
319 
320 	/*
321 	 * Initialize driver parameters
322 	 */
323 	igb_init_properties(igb);
324 	igb->attach_progress |= ATTACH_PROGRESS_PROPS;
325 
326 	/*
327 	 * Allocate interrupts
328 	 */
329 	if (igb_alloc_intrs(igb) != IGB_SUCCESS) {
330 		igb_error(igb, "Failed to allocate interrupts");
331 		goto attach_fail;
332 	}
333 	igb->attach_progress |= ATTACH_PROGRESS_ALLOC_INTR;
334 
335 	/*
336 	 * Allocate rx/tx rings based on the ring numbers.
337 	 * The actual numbers of rx/tx rings are decided by the number of
338 	 * allocated interrupt vectors, so we should allocate the rings after
339 	 * interrupts are allocated.
340 	 */
341 	if (igb_alloc_rings(igb) != IGB_SUCCESS) {
342 		igb_error(igb, "Failed to allocate rx and tx rings");
343 		goto attach_fail;
344 	}
345 	igb->attach_progress |= ATTACH_PROGRESS_ALLOC_RINGS;
346 
347 	/*
348 	 * Add interrupt handlers
349 	 */
350 	if (igb_add_intr_handlers(igb) != IGB_SUCCESS) {
351 		igb_error(igb, "Failed to add interrupt handlers");
352 		goto attach_fail;
353 	}
354 	igb->attach_progress |= ATTACH_PROGRESS_ADD_INTR;
355 
356 	/*
357 	 * Initialize driver parameters
358 	 */
359 	if (igb_init_driver_settings(igb) != IGB_SUCCESS) {
360 		igb_error(igb, "Failed to initialize driver settings");
361 		goto attach_fail;
362 	}
363 
364 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK) {
365 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
366 		goto attach_fail;
367 	}
368 
369 	/*
370 	 * Initialize mutexes for this device.
371 	 * Do this before enabling the interrupt handler and
372 	 * register the softint to avoid the condition where
373 	 * interrupt handler can try using uninitialized mutex
374 	 */
375 	igb_init_locks(igb);
376 	igb->attach_progress |= ATTACH_PROGRESS_LOCKS;
377 
378 	/*
379 	 * Initialize chipset hardware
380 	 */
381 	if (igb_init(igb) != IGB_SUCCESS) {
382 		igb_error(igb, "Failed to initialize adapter");
383 		goto attach_fail;
384 	}
385 	igb->attach_progress |= ATTACH_PROGRESS_INIT;
386 
387 	/*
388 	 * Initialize DMA and hardware settings for rx/tx rings
389 	 */
390 	if (igb_init_rings(igb) != IGB_SUCCESS) {
391 		igb_error(igb, "Failed to initialize rings");
392 		goto attach_fail;
393 	}
394 	igb->attach_progress |= ATTACH_PROGRESS_INIT_RINGS;
395 
396 	/*
397 	 * Initialize statistics
398 	 */
399 	if (igb_init_stats(igb) != IGB_SUCCESS) {
400 		igb_error(igb, "Failed to initialize statistics");
401 		goto attach_fail;
402 	}
403 	igb->attach_progress |= ATTACH_PROGRESS_STATS;
404 
405 	/*
406 	 * Initialize NDD parameters
407 	 */
408 	if (igb_nd_init(igb) != IGB_SUCCESS) {
409 		igb_error(igb, "Failed to initialize ndd");
410 		goto attach_fail;
411 	}
412 	igb->attach_progress |= ATTACH_PROGRESS_NDD;
413 
414 	/*
415 	 * Register the driver to the MAC
416 	 */
417 	if (igb_register_mac(igb) != IGB_SUCCESS) {
418 		igb_error(igb, "Failed to register MAC");
419 		goto attach_fail;
420 	}
421 	igb->attach_progress |= ATTACH_PROGRESS_MAC;
422 
423 	/*
424 	 * Now that mutex locks are initialized, and the chip is also
425 	 * initialized, enable interrupts.
426 	 */
427 	if (igb_enable_intrs(igb) != IGB_SUCCESS) {
428 		igb_error(igb, "Failed to enable DDI interrupts");
429 		goto attach_fail;
430 	}
431 	igb->attach_progress |= ATTACH_PROGRESS_ENABLE_INTR;
432 
433 	igb->igb_state |= IGB_INITIALIZED;
434 
435 	return (DDI_SUCCESS);
436 
437 attach_fail:
438 	igb_unconfigure(devinfo, igb);
439 	return (DDI_FAILURE);
440 }
441 
442 /*
443  * igb_detach - driver detach
444  *
445  * The detach() function is the complement of the attach routine.
446  * If cmd is set to DDI_DETACH, detach() is used to remove  the
447  * state  associated  with  a  given  instance of a device node
448  * prior to the removal of that instance from the system.
449  *
450  * The detach() function will be called once for each  instance
451  * of the device for which there has been a successful attach()
452  * once there are no longer  any  opens  on  the  device.
453  *
454  * Interrupts routine are disabled, All memory allocated by this
455  * driver are freed.
456  */
457 static int
458 igb_detach(dev_info_t *devinfo, ddi_detach_cmd_t cmd)
459 {
460 	igb_t *igb;
461 
462 	/*
463 	 * Check detach command
464 	 */
465 	switch (cmd) {
466 	default:
467 		return (DDI_FAILURE);
468 
469 	case DDI_SUSPEND:
470 		return (igb_suspend(devinfo));
471 
472 	case DDI_DETACH:
473 		break;
474 	}
475 
476 
477 	/*
478 	 * Get the pointer to the driver private data structure
479 	 */
480 	igb = (igb_t *)ddi_get_driver_private(devinfo);
481 	if (igb == NULL)
482 		return (DDI_FAILURE);
483 
484 	/*
485 	 * Unregister MAC. If failed, we have to fail the detach
486 	 */
487 	if (mac_unregister(igb->mac_hdl) != 0) {
488 		igb_error(igb, "Failed to unregister MAC");
489 		return (DDI_FAILURE);
490 	}
491 	igb->attach_progress &= ~ATTACH_PROGRESS_MAC;
492 
493 	/*
494 	 * If the device is still running, it needs to be stopped first.
495 	 * This check is necessary because under some specific circumstances,
496 	 * the detach routine can be called without stopping the interface
497 	 * first.
498 	 */
499 	mutex_enter(&igb->gen_lock);
500 	if (igb->igb_state & IGB_STARTED) {
501 		igb->igb_state &= ~IGB_STARTED;
502 		igb_stop(igb);
503 		mutex_exit(&igb->gen_lock);
504 		/* Disable and stop the watchdog timer */
505 		igb_disable_watchdog_timer(igb);
506 	} else
507 		mutex_exit(&igb->gen_lock);
508 
509 	/*
510 	 * Check if there are still rx buffers held by the upper layer.
511 	 * If so, fail the detach.
512 	 */
513 	if (!igb_rx_drain(igb))
514 		return (DDI_FAILURE);
515 
516 	/*
517 	 * Do the remaining unconfigure routines
518 	 */
519 	igb_unconfigure(devinfo, igb);
520 
521 	return (DDI_SUCCESS);
522 }
523 
524 /*
525  * quiesce(9E) entry point.
526  *
527  * This function is called when the system is single-threaded at high
528  * PIL with preemption disabled. Therefore, this function must not be
529  * blocked.
530  *
531  * This function returns DDI_SUCCESS on success, or DDI_FAILURE on failure.
532  * DDI_FAILURE indicates an error condition and should almost never happen.
533  */
534 static int
535 igb_quiesce(dev_info_t *devinfo)
536 {
537 	igb_t *igb;
538 	struct e1000_hw *hw;
539 
540 	igb = (igb_t *)ddi_get_driver_private(devinfo);
541 
542 	if (igb == NULL)
543 		return (DDI_FAILURE);
544 
545 	hw = &igb->hw;
546 
547 	/*
548 	 * Disable the adapter interrupts
549 	 */
550 	igb_disable_adapter_interrupts(igb);
551 
552 	/* Tell firmware driver is no longer in control */
553 	igb_release_driver_control(hw);
554 
555 	/*
556 	 * Reset the chipset
557 	 */
558 	(void) e1000_reset_hw(hw);
559 
560 	/*
561 	 * Reset PHY if possible
562 	 */
563 	if (e1000_check_reset_block(hw) == E1000_SUCCESS)
564 		(void) e1000_phy_hw_reset(hw);
565 
566 	return (DDI_SUCCESS);
567 }
568 
569 
570 static void
571 igb_unconfigure(dev_info_t *devinfo, igb_t *igb)
572 {
573 	/*
574 	 * Disable interrupt
575 	 */
576 	if (igb->attach_progress & ATTACH_PROGRESS_ENABLE_INTR) {
577 		(void) igb_disable_intrs(igb);
578 	}
579 
580 	/*
581 	 * Unregister MAC
582 	 */
583 	if (igb->attach_progress & ATTACH_PROGRESS_MAC) {
584 		(void) mac_unregister(igb->mac_hdl);
585 	}
586 
587 	/*
588 	 * Free ndd parameters
589 	 */
590 	if (igb->attach_progress & ATTACH_PROGRESS_NDD) {
591 		igb_nd_cleanup(igb);
592 	}
593 
594 	/*
595 	 * Free statistics
596 	 */
597 	if (igb->attach_progress & ATTACH_PROGRESS_STATS) {
598 		kstat_delete((kstat_t *)igb->igb_ks);
599 	}
600 
601 	/*
602 	 * Remove interrupt handlers
603 	 */
604 	if (igb->attach_progress & ATTACH_PROGRESS_ADD_INTR) {
605 		igb_rem_intr_handlers(igb);
606 	}
607 
608 	/*
609 	 * Remove interrupts
610 	 */
611 	if (igb->attach_progress & ATTACH_PROGRESS_ALLOC_INTR) {
612 		igb_rem_intrs(igb);
613 	}
614 
615 	/*
616 	 * Remove driver properties
617 	 */
618 	if (igb->attach_progress & ATTACH_PROGRESS_PROPS) {
619 		(void) ddi_prop_remove_all(devinfo);
620 	}
621 
622 	/*
623 	 * Release the DMA resources of rx/tx rings
624 	 */
625 	if (igb->attach_progress & ATTACH_PROGRESS_INIT_RINGS) {
626 		igb_fini_rings(igb);
627 	}
628 
629 	/*
630 	 * Stop the chipset
631 	 */
632 	if (igb->attach_progress & ATTACH_PROGRESS_INIT) {
633 		mutex_enter(&igb->gen_lock);
634 		igb_chip_stop(igb);
635 		mutex_exit(&igb->gen_lock);
636 		if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
637 			ddi_fm_service_impact(igb->dip, DDI_SERVICE_UNAFFECTED);
638 	}
639 
640 	/*
641 	 * Free register handle
642 	 */
643 	if (igb->attach_progress & ATTACH_PROGRESS_REGS_MAP) {
644 		if (igb->osdep.reg_handle != NULL)
645 			ddi_regs_map_free(&igb->osdep.reg_handle);
646 	}
647 
648 	/*
649 	 * Free PCI config handle
650 	 */
651 	if (igb->attach_progress & ATTACH_PROGRESS_PCI_CONFIG) {
652 		if (igb->osdep.cfg_handle != NULL)
653 			pci_config_teardown(&igb->osdep.cfg_handle);
654 	}
655 
656 	/*
657 	 * Free locks
658 	 */
659 	if (igb->attach_progress & ATTACH_PROGRESS_LOCKS) {
660 		igb_destroy_locks(igb);
661 	}
662 
663 	/*
664 	 * Free the rx/tx rings
665 	 */
666 	if (igb->attach_progress & ATTACH_PROGRESS_ALLOC_RINGS) {
667 		igb_free_rings(igb);
668 	}
669 
670 	/*
671 	 * Remove FMA
672 	 */
673 	if (igb->attach_progress & ATTACH_PROGRESS_FMINIT) {
674 		igb_fm_fini(igb);
675 	}
676 
677 	/*
678 	 * Free device specific structure
679 	 */
680 	e1000_remove_device(&igb->hw);
681 
682 	/*
683 	 * Free the driver data structure
684 	 */
685 	kmem_free(igb, sizeof (igb_t));
686 
687 	ddi_set_driver_private(devinfo, NULL);
688 }
689 
690 /*
691  * igb_register_mac - Register the driver and its function pointers with
692  * the GLD interface
693  */
694 static int
695 igb_register_mac(igb_t *igb)
696 {
697 	struct e1000_hw *hw = &igb->hw;
698 	mac_register_t *mac;
699 	int status;
700 
701 	if ((mac = mac_alloc(MAC_VERSION)) == NULL)
702 		return (IGB_FAILURE);
703 
704 	mac->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
705 	mac->m_driver = igb;
706 	mac->m_dip = igb->dip;
707 	mac->m_src_addr = hw->mac.addr;
708 	mac->m_callbacks = &igb_m_callbacks;
709 	mac->m_min_sdu = 0;
710 	mac->m_max_sdu = igb->max_frame_size -
711 	    sizeof (struct ether_vlan_header) - ETHERFCSL;
712 	mac->m_margin = VLAN_TAGSZ;
713 
714 	status = mac_register(mac, &igb->mac_hdl);
715 
716 	mac_free(mac);
717 
718 	return ((status == 0) ? IGB_SUCCESS : IGB_FAILURE);
719 }
720 
721 /*
722  * igb_identify_hardware - Identify the type of the chipset
723  */
724 static int
725 igb_identify_hardware(igb_t *igb)
726 {
727 	struct e1000_hw *hw = &igb->hw;
728 	struct igb_osdep *osdep = &igb->osdep;
729 
730 	/*
731 	 * Get the device id
732 	 */
733 	hw->vendor_id =
734 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_VENID);
735 	hw->device_id =
736 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_DEVID);
737 	hw->revision_id =
738 	    pci_config_get8(osdep->cfg_handle, PCI_CONF_REVID);
739 	hw->subsystem_device_id =
740 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBSYSID);
741 	hw->subsystem_vendor_id =
742 	    pci_config_get16(osdep->cfg_handle, PCI_CONF_SUBVENID);
743 
744 	/*
745 	 * Set the mac type of the adapter based on the device id
746 	 */
747 	if (e1000_set_mac_type(hw) != E1000_SUCCESS) {
748 		return (IGB_FAILURE);
749 	}
750 
751 	return (IGB_SUCCESS);
752 }
753 
754 /*
755  * igb_regs_map - Map the device registers
756  */
757 static int
758 igb_regs_map(igb_t *igb)
759 {
760 	dev_info_t *devinfo = igb->dip;
761 	struct e1000_hw *hw = &igb->hw;
762 	struct igb_osdep *osdep = &igb->osdep;
763 	off_t mem_size;
764 
765 	/*
766 	 * First get the size of device registers to be mapped.
767 	 */
768 	if (ddi_dev_regsize(devinfo, 1, &mem_size) != DDI_SUCCESS) {
769 		return (IGB_FAILURE);
770 	}
771 
772 	/*
773 	 * Call ddi_regs_map_setup() to map registers
774 	 */
775 	if ((ddi_regs_map_setup(devinfo, 1,
776 	    (caddr_t *)&hw->hw_addr, 0,
777 	    mem_size, &igb_regs_acc_attr,
778 	    &osdep->reg_handle)) != DDI_SUCCESS) {
779 		return (IGB_FAILURE);
780 	}
781 
782 	return (IGB_SUCCESS);
783 }
784 
785 /*
786  * igb_init_properties - Initialize driver properties
787  */
788 static void
789 igb_init_properties(igb_t *igb)
790 {
791 	/*
792 	 * Get conf file properties, including link settings
793 	 * jumbo frames, ring number, descriptor number, etc.
794 	 */
795 	igb_get_conf(igb);
796 }
797 
798 /*
799  * igb_init_driver_settings - Initialize driver settings
800  *
801  * The settings include hardware function pointers, bus information,
802  * rx/tx rings settings, link state, and any other parameters that
803  * need to be setup during driver initialization.
804  */
805 static int
806 igb_init_driver_settings(igb_t *igb)
807 {
808 	struct e1000_hw *hw = &igb->hw;
809 	igb_rx_ring_t *rx_ring;
810 	igb_tx_ring_t *tx_ring;
811 	uint32_t rx_size;
812 	uint32_t tx_size;
813 	int i;
814 
815 	/*
816 	 * Initialize chipset specific hardware function pointers
817 	 */
818 	if (e1000_setup_init_funcs(hw, B_TRUE) != E1000_SUCCESS) {
819 		return (IGB_FAILURE);
820 	}
821 
822 	/*
823 	 * Get bus information
824 	 */
825 	if (e1000_get_bus_info(hw) != E1000_SUCCESS) {
826 		return (IGB_FAILURE);
827 	}
828 
829 	/*
830 	 * Set rx buffer size
831 	 * The IP header alignment room is counted in the calculation.
832 	 * The rx buffer size is in unit of 1K that is required by the
833 	 * chipset hardware.
834 	 */
835 	rx_size = igb->max_frame_size + IPHDR_ALIGN_ROOM;
836 	igb->rx_buf_size = ((rx_size >> 10) +
837 	    ((rx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
838 
839 	/*
840 	 * Set tx buffer size
841 	 */
842 	tx_size = igb->max_frame_size;
843 	igb->tx_buf_size = ((tx_size >> 10) +
844 	    ((tx_size & (((uint32_t)1 << 10) - 1)) > 0 ? 1 : 0)) << 10;
845 
846 	/*
847 	 * Initialize rx/tx rings parameters
848 	 */
849 	for (i = 0; i < igb->num_rx_rings; i++) {
850 		rx_ring = &igb->rx_rings[i];
851 		rx_ring->index = i;
852 		rx_ring->igb = igb;
853 
854 		rx_ring->ring_size = igb->rx_ring_size;
855 		rx_ring->free_list_size = igb->rx_ring_size;
856 		rx_ring->copy_thresh = igb->rx_copy_thresh;
857 		rx_ring->limit_per_intr = igb->rx_limit_per_intr;
858 	}
859 
860 	for (i = 0; i < igb->num_tx_rings; i++) {
861 		tx_ring = &igb->tx_rings[i];
862 		tx_ring->index = i;
863 		tx_ring->igb = igb;
864 		if (igb->tx_head_wb_enable)
865 			tx_ring->tx_recycle = igb_tx_recycle_head_wb;
866 		else
867 			tx_ring->tx_recycle = igb_tx_recycle_legacy;
868 
869 		tx_ring->ring_size = igb->tx_ring_size;
870 		tx_ring->free_list_size = igb->tx_ring_size +
871 		    (igb->tx_ring_size >> 1);
872 		tx_ring->copy_thresh = igb->tx_copy_thresh;
873 		tx_ring->recycle_thresh = igb->tx_recycle_thresh;
874 		tx_ring->overload_thresh = igb->tx_overload_thresh;
875 		tx_ring->resched_thresh = igb->tx_resched_thresh;
876 	}
877 
878 	/*
879 	 * Initialize values of interrupt throttling rate
880 	 */
881 	for (i = 1; i < MAX_NUM_EITR; i++)
882 		igb->intr_throttling[i] = igb->intr_throttling[0];
883 
884 	/*
885 	 * The initial link state should be "unknown"
886 	 */
887 	igb->link_state = LINK_STATE_UNKNOWN;
888 
889 	return (IGB_SUCCESS);
890 }
891 
892 /*
893  * igb_init_locks - Initialize locks
894  */
895 static void
896 igb_init_locks(igb_t *igb)
897 {
898 	igb_rx_ring_t *rx_ring;
899 	igb_tx_ring_t *tx_ring;
900 	int i;
901 
902 	for (i = 0; i < igb->num_rx_rings; i++) {
903 		rx_ring = &igb->rx_rings[i];
904 		mutex_init(&rx_ring->rx_lock, NULL,
905 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
906 		mutex_init(&rx_ring->recycle_lock, NULL,
907 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
908 	}
909 
910 	for (i = 0; i < igb->num_tx_rings; i++) {
911 		tx_ring = &igb->tx_rings[i];
912 		mutex_init(&tx_ring->tx_lock, NULL,
913 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
914 		mutex_init(&tx_ring->recycle_lock, NULL,
915 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
916 		mutex_init(&tx_ring->tcb_head_lock, NULL,
917 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
918 		mutex_init(&tx_ring->tcb_tail_lock, NULL,
919 		    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
920 	}
921 
922 	mutex_init(&igb->gen_lock, NULL,
923 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
924 
925 	mutex_init(&igb->watchdog_lock, NULL,
926 	    MUTEX_DRIVER, DDI_INTR_PRI(igb->intr_pri));
927 }
928 
929 /*
930  * igb_destroy_locks - Destroy locks
931  */
932 static void
933 igb_destroy_locks(igb_t *igb)
934 {
935 	igb_rx_ring_t *rx_ring;
936 	igb_tx_ring_t *tx_ring;
937 	int i;
938 
939 	for (i = 0; i < igb->num_rx_rings; i++) {
940 		rx_ring = &igb->rx_rings[i];
941 		mutex_destroy(&rx_ring->rx_lock);
942 		mutex_destroy(&rx_ring->recycle_lock);
943 	}
944 
945 	for (i = 0; i < igb->num_tx_rings; i++) {
946 		tx_ring = &igb->tx_rings[i];
947 		mutex_destroy(&tx_ring->tx_lock);
948 		mutex_destroy(&tx_ring->recycle_lock);
949 		mutex_destroy(&tx_ring->tcb_head_lock);
950 		mutex_destroy(&tx_ring->tcb_tail_lock);
951 	}
952 
953 	mutex_destroy(&igb->gen_lock);
954 	mutex_destroy(&igb->watchdog_lock);
955 }
956 
957 static int
958 igb_resume(dev_info_t *devinfo)
959 {
960 	igb_t *igb;
961 
962 	igb = (igb_t *)ddi_get_driver_private(devinfo);
963 	if (igb == NULL)
964 		return (DDI_FAILURE);
965 
966 	mutex_enter(&igb->gen_lock);
967 
968 	if (igb->igb_state & IGB_STARTED) {
969 		if (igb_start(igb) != IGB_SUCCESS) {
970 			mutex_exit(&igb->gen_lock);
971 			return (DDI_FAILURE);
972 		}
973 
974 		/*
975 		 * Enable and start the watchdog timer
976 		 */
977 		igb_enable_watchdog_timer(igb);
978 	}
979 
980 	igb->igb_state &= ~IGB_SUSPENDED;
981 
982 	mutex_exit(&igb->gen_lock);
983 
984 	return (DDI_SUCCESS);
985 }
986 
987 static int
988 igb_suspend(dev_info_t *devinfo)
989 {
990 	igb_t *igb;
991 
992 	igb = (igb_t *)ddi_get_driver_private(devinfo);
993 	if (igb == NULL)
994 		return (DDI_FAILURE);
995 
996 	mutex_enter(&igb->gen_lock);
997 
998 	igb->igb_state |= IGB_SUSPENDED;
999 
1000 	igb_stop(igb);
1001 
1002 	mutex_exit(&igb->gen_lock);
1003 
1004 	/*
1005 	 * Disable and stop the watchdog timer
1006 	 */
1007 	igb_disable_watchdog_timer(igb);
1008 
1009 	return (DDI_SUCCESS);
1010 }
1011 
1012 /*
1013  * igb_init - Initialize the device
1014  */
1015 static int
1016 igb_init(igb_t *igb)
1017 {
1018 	struct e1000_hw *hw = &igb->hw;
1019 	uint32_t pba;
1020 	uint32_t high_water;
1021 
1022 	mutex_enter(&igb->gen_lock);
1023 
1024 	/*
1025 	 * Reset chipset to put the hardware in a known state
1026 	 * before we try to do anything with the eeprom
1027 	 */
1028 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1029 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1030 		goto init_fail;
1031 	}
1032 
1033 	/*
1034 	 * NVM validation
1035 	 */
1036 	if (e1000_validate_nvm_checksum(hw) < 0) {
1037 		/*
1038 		 * Some PCI-E parts fail the first check due to
1039 		 * the link being in sleep state.  Call it again,
1040 		 * if it fails a second time its a real issue.
1041 		 */
1042 		if (e1000_validate_nvm_checksum(hw) < 0) {
1043 			igb_error(igb,
1044 			    "Invalid NVM checksum. Please contact "
1045 			    "the vendor to update the NVM.");
1046 			igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1047 			goto init_fail;
1048 		}
1049 	}
1050 
1051 	/*
1052 	 * Set the FIFO size
1053 	 */
1054 	pba = E1000_PBA_32K;	/* 32K for Rx, 16K for Tx */
1055 	E1000_WRITE_REG(hw, E1000_PBA, pba);
1056 
1057 	/*
1058 	 * Setup flow control
1059 	 *
1060 	 * These parameters set thresholds for the adapter's generation(Tx)
1061 	 * and response(Rx) to Ethernet PAUSE frames.  These are just threshold
1062 	 * settings.  Flow control is enabled or disabled in the configuration
1063 	 * file.
1064 	 * High-water mark is set down from the top of the rx fifo (not
1065 	 * sensitive to max_frame_size) and low-water is set just below
1066 	 * high-water mark.
1067 	 * The high water mark must be low enough to fit one full frame above
1068 	 * it in the rx FIFO.  Should be the lower of:
1069 	 * 90% of the Rx FIFO size, or the full Rx FIFO size minus one full
1070 	 * frame.
1071 	 */
1072 	high_water = min(((pba << 10) * 9 / 10),
1073 	    ((pba << 10) - igb->max_frame_size));
1074 
1075 	hw->fc.high_water = high_water & 0xFFF8;
1076 	hw->fc.low_water = hw->fc.high_water - 8;
1077 	hw->fc.pause_time = E1000_FC_PAUSE_TIME;
1078 	hw->fc.send_xon = B_TRUE;
1079 
1080 	/*
1081 	 * Reset the chipset hardware the second time to validate
1082 	 * the PBA setting.
1083 	 */
1084 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1085 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1086 		goto init_fail;
1087 	}
1088 
1089 	/*
1090 	 * Don't wait for auto-negotiation to complete
1091 	 */
1092 	hw->phy.autoneg_wait_to_complete = B_FALSE;
1093 
1094 	/*
1095 	 * Copper options
1096 	 */
1097 	if (hw->phy.media_type == e1000_media_type_copper) {
1098 		hw->phy.mdix = 0;	/* AUTO_ALL_MODES */
1099 		hw->phy.disable_polarity_correction = B_FALSE;
1100 		hw->phy.ms_type = e1000_ms_hw_default; /* E1000_MASTER_SLAVE */
1101 	}
1102 
1103 	/*
1104 	 * Initialize link settings
1105 	 */
1106 	(void) igb_setup_link(igb, B_FALSE);
1107 
1108 	/*
1109 	 * Initialize the chipset hardware
1110 	 */
1111 	if (igb_chip_start(igb) != IGB_SUCCESS) {
1112 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1113 		goto init_fail;
1114 	}
1115 
1116 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK) {
1117 		goto init_fail;
1118 	}
1119 
1120 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
1121 		goto init_fail;
1122 	}
1123 
1124 	mutex_exit(&igb->gen_lock);
1125 	return (IGB_SUCCESS);
1126 
1127 init_fail:
1128 	/*
1129 	 * Reset PHY if possible
1130 	 */
1131 	if (e1000_check_reset_block(hw) == E1000_SUCCESS)
1132 		(void) e1000_phy_hw_reset(hw);
1133 
1134 	mutex_exit(&igb->gen_lock);
1135 
1136 	ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1137 
1138 	return (IGB_FAILURE);
1139 }
1140 
1141 /*
1142  * igb_init_rings - Allocate DMA resources for all rx/tx rings and
1143  * initialize relevant hardware settings.
1144  */
1145 static int
1146 igb_init_rings(igb_t *igb)
1147 {
1148 	int i;
1149 
1150 	/*
1151 	 * Allocate buffers for all the rx/tx rings
1152 	 */
1153 	if (igb_alloc_dma(igb) != IGB_SUCCESS)
1154 		return (IGB_FAILURE);
1155 
1156 	/*
1157 	 * Setup the rx/tx rings
1158 	 */
1159 	mutex_enter(&igb->gen_lock);
1160 
1161 	for (i = 0; i < igb->num_rx_rings; i++)
1162 		mutex_enter(&igb->rx_rings[i].rx_lock);
1163 	for (i = 0; i < igb->num_tx_rings; i++)
1164 		mutex_enter(&igb->tx_rings[i].tx_lock);
1165 
1166 	igb_setup_rings(igb);
1167 
1168 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1169 		mutex_exit(&igb->tx_rings[i].tx_lock);
1170 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1171 		mutex_exit(&igb->rx_rings[i].rx_lock);
1172 
1173 	mutex_exit(&igb->gen_lock);
1174 
1175 	return (IGB_SUCCESS);
1176 }
1177 
1178 /*
1179  * igb_fini_rings - Release DMA resources of all rx/tx rings
1180  */
1181 static void
1182 igb_fini_rings(igb_t *igb)
1183 {
1184 	/*
1185 	 * Release the DMA/memory resources of rx/tx rings
1186 	 */
1187 	igb_free_dma(igb);
1188 }
1189 
1190 /*
1191  * igb_chip_start - Initialize and start the chipset hardware
1192  */
1193 static int
1194 igb_chip_start(igb_t *igb)
1195 {
1196 	struct e1000_hw *hw = &igb->hw;
1197 	int i;
1198 
1199 	ASSERT(mutex_owned(&igb->gen_lock));
1200 
1201 	/*
1202 	 * Get the mac address
1203 	 * This function should handle SPARC case correctly.
1204 	 */
1205 	if (!igb_find_mac_address(igb)) {
1206 		igb_error(igb, "Failed to get the mac address");
1207 		return (IGB_FAILURE);
1208 	}
1209 
1210 	/* Validate mac address */
1211 	if (!is_valid_mac_addr(hw->mac.addr)) {
1212 		igb_error(igb, "Invalid mac address");
1213 		return (IGB_FAILURE);
1214 	}
1215 
1216 	/* Disable wakeup control by default */
1217 	E1000_WRITE_REG(hw, E1000_WUC, 0);
1218 
1219 	/*
1220 	 * Configure/Initialize hardware
1221 	 */
1222 	if (e1000_init_hw(hw) != E1000_SUCCESS) {
1223 		igb_error(igb, "Failed to initialize hardware");
1224 		return (IGB_FAILURE);
1225 	}
1226 
1227 	/*
1228 	 * Make sure driver has control
1229 	 */
1230 	igb_get_driver_control(hw);
1231 
1232 	/*
1233 	 * Setup MSI-X interrupts
1234 	 */
1235 	if (igb->intr_type == DDI_INTR_TYPE_MSIX)
1236 		igb_setup_adapter_msix(igb);
1237 
1238 	/*
1239 	 * Initialize unicast addresses.
1240 	 */
1241 	igb_init_unicst(igb);
1242 
1243 	/*
1244 	 * Setup and initialize the mctable structures.
1245 	 */
1246 	igb_setup_multicst(igb);
1247 
1248 	/*
1249 	 * Set interrupt throttling rate
1250 	 */
1251 	for (i = 0; i < igb->intr_cnt; i++)
1252 		E1000_WRITE_REG(hw, E1000_EITR(i), igb->intr_throttling[i]);
1253 
1254 	/* Enable PCI-E master */
1255 	if (hw->bus.type == e1000_bus_type_pci_express) {
1256 		e1000_enable_pciex_master(hw);
1257 	}
1258 
1259 	/*
1260 	 * Save the state of the phy
1261 	 */
1262 	igb_get_phy_state(igb);
1263 
1264 	return (IGB_SUCCESS);
1265 }
1266 
1267 /*
1268  * igb_chip_stop - Stop the chipset hardware
1269  */
1270 static void
1271 igb_chip_stop(igb_t *igb)
1272 {
1273 	struct e1000_hw *hw = &igb->hw;
1274 
1275 	ASSERT(mutex_owned(&igb->gen_lock));
1276 
1277 	/* Tell firmware driver is no longer in control */
1278 	igb_release_driver_control(hw);
1279 
1280 	/*
1281 	 * Reset the chipset
1282 	 */
1283 	if (e1000_reset_hw(hw) != E1000_SUCCESS) {
1284 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1285 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1286 	}
1287 
1288 	/*
1289 	 * Reset PHY if possible
1290 	 */
1291 	if (e1000_check_reset_block(hw) == E1000_SUCCESS)
1292 		(void) e1000_phy_hw_reset(hw);
1293 }
1294 
1295 /*
1296  * igb_reset - Reset the chipset and restart the driver.
1297  *
1298  * It involves stopping and re-starting the chipset,
1299  * and re-configuring the rx/tx rings.
1300  */
1301 static int
1302 igb_reset(igb_t *igb)
1303 {
1304 	int i;
1305 
1306 	mutex_enter(&igb->gen_lock);
1307 
1308 	ASSERT(igb->igb_state & IGB_STARTED);
1309 
1310 	/*
1311 	 * Disable the adapter interrupts to stop any rx/tx activities
1312 	 * before draining pending data and resetting hardware.
1313 	 */
1314 	igb_disable_adapter_interrupts(igb);
1315 
1316 	/*
1317 	 * Drain the pending transmit packets
1318 	 */
1319 	(void) igb_tx_drain(igb);
1320 
1321 	for (i = 0; i < igb->num_rx_rings; i++)
1322 		mutex_enter(&igb->rx_rings[i].rx_lock);
1323 	for (i = 0; i < igb->num_tx_rings; i++)
1324 		mutex_enter(&igb->tx_rings[i].tx_lock);
1325 
1326 	/*
1327 	 * Stop the chipset hardware
1328 	 */
1329 	igb_chip_stop(igb);
1330 
1331 	/*
1332 	 * Clean the pending tx data/resources
1333 	 */
1334 	igb_tx_clean(igb);
1335 
1336 	/*
1337 	 * Start the chipset hardware
1338 	 */
1339 	if (igb_chip_start(igb) != IGB_SUCCESS) {
1340 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1341 		goto reset_failure;
1342 	}
1343 
1344 	/*
1345 	 * Setup the rx/tx rings
1346 	 */
1347 	igb_setup_rings(igb);
1348 
1349 	/*
1350 	 * Enable adapter interrupts
1351 	 * The interrupts must be enabled after the driver state is START
1352 	 */
1353 	igb_enable_adapter_interrupts(igb);
1354 
1355 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK)
1356 		goto reset_failure;
1357 
1358 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1359 		goto reset_failure;
1360 
1361 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1362 		mutex_exit(&igb->tx_rings[i].tx_lock);
1363 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1364 		mutex_exit(&igb->rx_rings[i].rx_lock);
1365 
1366 	mutex_exit(&igb->gen_lock);
1367 
1368 	return (IGB_SUCCESS);
1369 
1370 reset_failure:
1371 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1372 		mutex_exit(&igb->tx_rings[i].tx_lock);
1373 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1374 		mutex_exit(&igb->rx_rings[i].rx_lock);
1375 
1376 	mutex_exit(&igb->gen_lock);
1377 
1378 	ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1379 
1380 	return (IGB_FAILURE);
1381 }
1382 
1383 /*
1384  * igb_tx_clean - Clean the pending transmit packets and DMA resources
1385  */
1386 static void
1387 igb_tx_clean(igb_t *igb)
1388 {
1389 	igb_tx_ring_t *tx_ring;
1390 	tx_control_block_t *tcb;
1391 	link_list_t pending_list;
1392 	uint32_t desc_num;
1393 	int i, j;
1394 
1395 	LINK_LIST_INIT(&pending_list);
1396 
1397 	for (i = 0; i < igb->num_tx_rings; i++) {
1398 		tx_ring = &igb->tx_rings[i];
1399 
1400 		mutex_enter(&tx_ring->recycle_lock);
1401 
1402 		/*
1403 		 * Clean the pending tx data - the pending packets in the
1404 		 * work_list that have no chances to be transmitted again.
1405 		 *
1406 		 * We must ensure the chipset is stopped or the link is down
1407 		 * before cleaning the transmit packets.
1408 		 */
1409 		desc_num = 0;
1410 		for (j = 0; j < tx_ring->ring_size; j++) {
1411 			tcb = tx_ring->work_list[j];
1412 			if (tcb != NULL) {
1413 				desc_num += tcb->desc_num;
1414 
1415 				tx_ring->work_list[j] = NULL;
1416 
1417 				igb_free_tcb(tcb);
1418 
1419 				LIST_PUSH_TAIL(&pending_list, &tcb->link);
1420 			}
1421 		}
1422 
1423 		if (desc_num > 0) {
1424 			atomic_add_32(&tx_ring->tbd_free, desc_num);
1425 			ASSERT(tx_ring->tbd_free == tx_ring->ring_size);
1426 
1427 			/*
1428 			 * Reset the head and tail pointers of the tbd ring;
1429 			 * Reset the head write-back if it is enabled.
1430 			 */
1431 			tx_ring->tbd_head = 0;
1432 			tx_ring->tbd_tail = 0;
1433 			if (igb->tx_head_wb_enable)
1434 				*tx_ring->tbd_head_wb = 0;
1435 
1436 			E1000_WRITE_REG(&igb->hw, E1000_TDH(tx_ring->index), 0);
1437 			E1000_WRITE_REG(&igb->hw, E1000_TDT(tx_ring->index), 0);
1438 		}
1439 
1440 		mutex_exit(&tx_ring->recycle_lock);
1441 
1442 		/*
1443 		 * Add the tx control blocks in the pending list to
1444 		 * the free list.
1445 		 */
1446 		igb_put_free_list(tx_ring, &pending_list);
1447 	}
1448 }
1449 
1450 /*
1451  * igb_tx_drain - Drain the tx rings to allow pending packets to be transmitted
1452  */
1453 static boolean_t
1454 igb_tx_drain(igb_t *igb)
1455 {
1456 	igb_tx_ring_t *tx_ring;
1457 	boolean_t done;
1458 	int i, j;
1459 
1460 	/*
1461 	 * Wait for a specific time to allow pending tx packets
1462 	 * to be transmitted.
1463 	 *
1464 	 * Check the counter tbd_free to see if transmission is done.
1465 	 * No lock protection is needed here.
1466 	 *
1467 	 * Return B_TRUE if all pending packets have been transmitted;
1468 	 * Otherwise return B_FALSE;
1469 	 */
1470 	for (i = 0; i < TX_DRAIN_TIME; i++) {
1471 
1472 		done = B_TRUE;
1473 		for (j = 0; j < igb->num_tx_rings; j++) {
1474 			tx_ring = &igb->tx_rings[j];
1475 			done = done &&
1476 			    (tx_ring->tbd_free == tx_ring->ring_size);
1477 		}
1478 
1479 		if (done)
1480 			break;
1481 
1482 		msec_delay(1);
1483 	}
1484 
1485 	return (done);
1486 }
1487 
1488 /*
1489  * igb_rx_drain - Wait for all rx buffers to be released by upper layer
1490  */
1491 static boolean_t
1492 igb_rx_drain(igb_t *igb)
1493 {
1494 	igb_rx_ring_t *rx_ring;
1495 	boolean_t done;
1496 	int i, j;
1497 
1498 	/*
1499 	 * Polling the rx free list to check if those rx buffers held by
1500 	 * the upper layer are released.
1501 	 *
1502 	 * Check the counter rcb_free to see if all pending buffers are
1503 	 * released. No lock protection is needed here.
1504 	 *
1505 	 * Return B_TRUE if all pending buffers have been released;
1506 	 * Otherwise return B_FALSE;
1507 	 */
1508 	for (i = 0; i < RX_DRAIN_TIME; i++) {
1509 
1510 		done = B_TRUE;
1511 		for (j = 0; j < igb->num_rx_rings; j++) {
1512 			rx_ring = &igb->rx_rings[j];
1513 			done = done &&
1514 			    (rx_ring->rcb_free == rx_ring->free_list_size);
1515 		}
1516 
1517 		if (done)
1518 			break;
1519 
1520 		msec_delay(1);
1521 	}
1522 
1523 	return (done);
1524 }
1525 
1526 /*
1527  * igb_start - Start the driver/chipset
1528  */
1529 int
1530 igb_start(igb_t *igb)
1531 {
1532 	int i;
1533 
1534 	ASSERT(mutex_owned(&igb->gen_lock));
1535 
1536 	for (i = 0; i < igb->num_rx_rings; i++)
1537 		mutex_enter(&igb->rx_rings[i].rx_lock);
1538 	for (i = 0; i < igb->num_tx_rings; i++)
1539 		mutex_enter(&igb->tx_rings[i].tx_lock);
1540 
1541 	/*
1542 	 * Start the chipset hardware
1543 	 */
1544 	if (igb_chip_start(igb) != IGB_SUCCESS) {
1545 		igb_fm_ereport(igb, DDI_FM_DEVICE_INVAL_STATE);
1546 		goto start_failure;
1547 	}
1548 
1549 	/*
1550 	 * Setup the rx/tx rings
1551 	 */
1552 	igb_setup_rings(igb);
1553 
1554 	/*
1555 	 * Enable adapter interrupts
1556 	 * The interrupts must be enabled after the driver state is START
1557 	 */
1558 	igb_enable_adapter_interrupts(igb);
1559 
1560 	if (igb_check_acc_handle(igb->osdep.cfg_handle) != DDI_FM_OK)
1561 		goto start_failure;
1562 
1563 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1564 		goto start_failure;
1565 
1566 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1567 		mutex_exit(&igb->tx_rings[i].tx_lock);
1568 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1569 		mutex_exit(&igb->rx_rings[i].rx_lock);
1570 
1571 	return (IGB_SUCCESS);
1572 
1573 start_failure:
1574 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1575 		mutex_exit(&igb->tx_rings[i].tx_lock);
1576 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1577 		mutex_exit(&igb->rx_rings[i].rx_lock);
1578 
1579 	ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1580 
1581 	return (IGB_FAILURE);
1582 }
1583 
1584 /*
1585  * igb_stop - Stop the driver/chipset
1586  */
1587 void
1588 igb_stop(igb_t *igb)
1589 {
1590 	int i;
1591 
1592 	ASSERT(mutex_owned(&igb->gen_lock));
1593 
1594 	/*
1595 	 * Disable the adapter interrupts
1596 	 */
1597 	igb_disable_adapter_interrupts(igb);
1598 
1599 	/*
1600 	 * Drain the pending tx packets
1601 	 */
1602 	(void) igb_tx_drain(igb);
1603 
1604 	for (i = 0; i < igb->num_rx_rings; i++)
1605 		mutex_enter(&igb->rx_rings[i].rx_lock);
1606 	for (i = 0; i < igb->num_tx_rings; i++)
1607 		mutex_enter(&igb->tx_rings[i].tx_lock);
1608 
1609 	/*
1610 	 * Stop the chipset hardware
1611 	 */
1612 	igb_chip_stop(igb);
1613 
1614 	/*
1615 	 * Clean the pending tx data/resources
1616 	 */
1617 	igb_tx_clean(igb);
1618 
1619 	for (i = igb->num_tx_rings - 1; i >= 0; i--)
1620 		mutex_exit(&igb->tx_rings[i].tx_lock);
1621 	for (i = igb->num_rx_rings - 1; i >= 0; i--)
1622 		mutex_exit(&igb->rx_rings[i].rx_lock);
1623 
1624 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1625 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1626 }
1627 
1628 /*
1629  * igb_alloc_rings - Allocate memory space for rx/tx rings
1630  */
1631 static int
1632 igb_alloc_rings(igb_t *igb)
1633 {
1634 	/*
1635 	 * Allocate memory space for rx rings
1636 	 */
1637 	igb->rx_rings = kmem_zalloc(
1638 	    sizeof (igb_rx_ring_t) * igb->num_rx_rings,
1639 	    KM_NOSLEEP);
1640 
1641 	if (igb->rx_rings == NULL) {
1642 		return (IGB_FAILURE);
1643 	}
1644 
1645 	/*
1646 	 * Allocate memory space for tx rings
1647 	 */
1648 	igb->tx_rings = kmem_zalloc(
1649 	    sizeof (igb_tx_ring_t) * igb->num_tx_rings,
1650 	    KM_NOSLEEP);
1651 
1652 	if (igb->tx_rings == NULL) {
1653 		kmem_free(igb->rx_rings,
1654 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1655 		igb->rx_rings = NULL;
1656 		return (IGB_FAILURE);
1657 	}
1658 
1659 	return (IGB_SUCCESS);
1660 }
1661 
1662 /*
1663  * igb_free_rings - Free the memory space of rx/tx rings.
1664  */
1665 static void
1666 igb_free_rings(igb_t *igb)
1667 {
1668 	if (igb->rx_rings != NULL) {
1669 		kmem_free(igb->rx_rings,
1670 		    sizeof (igb_rx_ring_t) * igb->num_rx_rings);
1671 		igb->rx_rings = NULL;
1672 	}
1673 
1674 	if (igb->tx_rings != NULL) {
1675 		kmem_free(igb->tx_rings,
1676 		    sizeof (igb_tx_ring_t) * igb->num_tx_rings);
1677 		igb->tx_rings = NULL;
1678 	}
1679 }
1680 
1681 /*
1682  * igb_setup_rings - Setup rx/tx rings
1683  */
1684 static void
1685 igb_setup_rings(igb_t *igb)
1686 {
1687 	/*
1688 	 * Setup the rx/tx rings, including the following:
1689 	 *
1690 	 * 1. Setup the descriptor ring and the control block buffers;
1691 	 * 2. Initialize necessary registers for receive/transmit;
1692 	 * 3. Initialize software pointers/parameters for receive/transmit;
1693 	 */
1694 	igb_setup_rx(igb);
1695 
1696 	igb_setup_tx(igb);
1697 
1698 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
1699 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_LOST);
1700 }
1701 
1702 static void
1703 igb_setup_rx_ring(igb_rx_ring_t *rx_ring)
1704 {
1705 	igb_t *igb = rx_ring->igb;
1706 	struct e1000_hw *hw = &igb->hw;
1707 	rx_control_block_t *rcb;
1708 	union e1000_adv_rx_desc	*rbd;
1709 	uint32_t size;
1710 	uint32_t buf_low;
1711 	uint32_t buf_high;
1712 	uint32_t reg_val;
1713 	int i;
1714 
1715 	ASSERT(mutex_owned(&rx_ring->rx_lock));
1716 	ASSERT(mutex_owned(&igb->gen_lock));
1717 
1718 	for (i = 0; i < igb->rx_ring_size; i++) {
1719 		rcb = rx_ring->work_list[i];
1720 		rbd = &rx_ring->rbd_ring[i];
1721 
1722 		rbd->read.pkt_addr = rcb->rx_buf.dma_address;
1723 		rbd->read.hdr_addr = NULL;
1724 	}
1725 
1726 	/*
1727 	 * Initialize the length register
1728 	 */
1729 	size = rx_ring->ring_size * sizeof (union e1000_adv_rx_desc);
1730 	E1000_WRITE_REG(hw, E1000_RDLEN(rx_ring->index), size);
1731 
1732 	/*
1733 	 * Initialize the base address registers
1734 	 */
1735 	buf_low = (uint32_t)rx_ring->rbd_area.dma_address;
1736 	buf_high = (uint32_t)(rx_ring->rbd_area.dma_address >> 32);
1737 	E1000_WRITE_REG(hw, E1000_RDBAH(rx_ring->index), buf_high);
1738 	E1000_WRITE_REG(hw, E1000_RDBAL(rx_ring->index), buf_low);
1739 
1740 	/*
1741 	 * Setup head & tail pointers
1742 	 */
1743 	E1000_WRITE_REG(hw, E1000_RDT(rx_ring->index), rx_ring->ring_size - 1);
1744 	E1000_WRITE_REG(hw, E1000_RDH(rx_ring->index), 0);
1745 
1746 	rx_ring->rbd_next = 0;
1747 
1748 	/*
1749 	 * Note: Considering the case that the chipset is being reset
1750 	 * and there are still some buffers held by the upper layer,
1751 	 * we should not reset the values of rcb_head, rcb_tail and
1752 	 * rcb_free;
1753 	 */
1754 	if (igb->igb_state == IGB_UNKNOWN) {
1755 		rx_ring->rcb_head = 0;
1756 		rx_ring->rcb_tail = 0;
1757 		rx_ring->rcb_free = rx_ring->free_list_size;
1758 	}
1759 
1760 	/*
1761 	 * Setup the Receive Descriptor Control Register (RXDCTL)
1762 	 */
1763 	reg_val = E1000_READ_REG(hw, E1000_RXDCTL(rx_ring->index));
1764 	reg_val |= E1000_RXDCTL_QUEUE_ENABLE;
1765 	reg_val &= 0xFFF00000;
1766 	reg_val |= 16;		/* pthresh */
1767 	reg_val |= 8 << 8;	/* hthresh */
1768 	reg_val |= 1 << 16;	/* wthresh */
1769 	E1000_WRITE_REG(hw, E1000_RXDCTL(rx_ring->index), reg_val);
1770 
1771 	/*
1772 	 * Setup the Split and Replication Receive Control Register.
1773 	 * Set the rx buffer size and the advanced descriptor type.
1774 	 */
1775 	reg_val = (igb->rx_buf_size >> E1000_SRRCTL_BSIZEPKT_SHIFT) |
1776 	    E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1777 
1778 	E1000_WRITE_REG(hw, E1000_SRRCTL(rx_ring->index), reg_val);
1779 }
1780 
1781 static void
1782 igb_setup_rx(igb_t *igb)
1783 {
1784 	igb_rx_ring_t *rx_ring;
1785 	struct e1000_hw *hw = &igb->hw;
1786 	uint32_t reg_val;
1787 	int i;
1788 
1789 	/*
1790 	 * Setup the Receive Control Register (RCTL), and ENABLE the
1791 	 * receiver. The initial configuration is to: Enable the receiver,
1792 	 * accept broadcasts, discard bad packets (and long packets),
1793 	 * disable VLAN filter checking, set the receive descriptor
1794 	 * minimum threshold size to 1/2, and the receive buffer size to
1795 	 * 2k.
1796 	 */
1797 	reg_val = E1000_RCTL_EN |	/* Enable Receive Unit */
1798 	    E1000_RCTL_BAM |		/* Accept Broadcast Packets */
1799 	    E1000_RCTL_LPE |		/* Large Packet Enable bit */
1800 	    (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT) |
1801 	    E1000_RCTL_RDMTS_HALF |
1802 	    E1000_RCTL_SECRC |		/* Strip Ethernet CRC */
1803 	    E1000_RCTL_LBM_NO;		/* Loopback Mode = none */
1804 
1805 	E1000_WRITE_REG(hw, E1000_RCTL, reg_val);
1806 
1807 	/*
1808 	 * igb_setup_rx_ring must be called after configuring RCTL
1809 	 */
1810 	for (i = 0; i < igb->num_rx_rings; i++) {
1811 		rx_ring = &igb->rx_rings[i];
1812 		igb_setup_rx_ring(rx_ring);
1813 	}
1814 
1815 	/*
1816 	 * Setup the Rx Long Packet Max Length register
1817 	 */
1818 	E1000_WRITE_REG(hw, E1000_RLPML, igb->max_frame_size);
1819 
1820 	/*
1821 	 * Hardware checksum settings
1822 	 */
1823 	if (igb->rx_hcksum_enable) {
1824 		reg_val =
1825 		    E1000_RXCSUM_TUOFL |	/* TCP/UDP checksum */
1826 		    E1000_RXCSUM_IPOFL;		/* IP checksum */
1827 
1828 		E1000_WRITE_REG(hw, E1000_RXCSUM, reg_val);
1829 	}
1830 
1831 	/*
1832 	 * Setup RSS for multiple receive queues
1833 	 */
1834 	if (igb->num_rx_rings > 1)
1835 		igb_setup_rss(igb);
1836 }
1837 
1838 static void
1839 igb_setup_tx_ring(igb_tx_ring_t *tx_ring)
1840 {
1841 	igb_t *igb = tx_ring->igb;
1842 	struct e1000_hw *hw = &igb->hw;
1843 	uint32_t size;
1844 	uint32_t buf_low;
1845 	uint32_t buf_high;
1846 	uint32_t reg_val;
1847 
1848 	ASSERT(mutex_owned(&tx_ring->tx_lock));
1849 	ASSERT(mutex_owned(&igb->gen_lock));
1850 
1851 	/*
1852 	 * Initialize the length register
1853 	 */
1854 	size = tx_ring->ring_size * sizeof (union e1000_adv_tx_desc);
1855 	E1000_WRITE_REG(hw, E1000_TDLEN(tx_ring->index), size);
1856 
1857 	/*
1858 	 * Initialize the base address registers
1859 	 */
1860 	buf_low = (uint32_t)tx_ring->tbd_area.dma_address;
1861 	buf_high = (uint32_t)(tx_ring->tbd_area.dma_address >> 32);
1862 	E1000_WRITE_REG(hw, E1000_TDBAL(tx_ring->index), buf_low);
1863 	E1000_WRITE_REG(hw, E1000_TDBAH(tx_ring->index), buf_high);
1864 
1865 	/*
1866 	 * Setup head & tail pointers
1867 	 */
1868 	E1000_WRITE_REG(hw, E1000_TDH(tx_ring->index), 0);
1869 	E1000_WRITE_REG(hw, E1000_TDT(tx_ring->index), 0);
1870 
1871 	/*
1872 	 * Setup head write-back
1873 	 */
1874 	if (igb->tx_head_wb_enable) {
1875 		/*
1876 		 * The memory of the head write-back is allocated using
1877 		 * the extra tbd beyond the tail of the tbd ring.
1878 		 */
1879 		tx_ring->tbd_head_wb = (uint32_t *)
1880 		    ((uintptr_t)tx_ring->tbd_area.address + size);
1881 		*tx_ring->tbd_head_wb = 0;
1882 
1883 		buf_low = (uint32_t)
1884 		    (tx_ring->tbd_area.dma_address + size);
1885 		buf_high = (uint32_t)
1886 		    ((tx_ring->tbd_area.dma_address + size) >> 32);
1887 
1888 		/* Set the head write-back enable bit */
1889 		buf_low |= E1000_TX_HEAD_WB_ENABLE;
1890 
1891 		E1000_WRITE_REG(hw, E1000_TDWBAL(tx_ring->index), buf_low);
1892 		E1000_WRITE_REG(hw, E1000_TDWBAH(tx_ring->index), buf_high);
1893 
1894 		/*
1895 		 * Turn off relaxed ordering for head write back or it will
1896 		 * cause problems with the tx recycling
1897 		 */
1898 		reg_val = E1000_READ_REG(hw,
1899 		    E1000_DCA_TXCTRL(tx_ring->index));
1900 		reg_val &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1901 		E1000_WRITE_REG(hw,
1902 		    E1000_DCA_TXCTRL(tx_ring->index), reg_val);
1903 	} else {
1904 		tx_ring->tbd_head_wb = NULL;
1905 	}
1906 
1907 	tx_ring->tbd_head = 0;
1908 	tx_ring->tbd_tail = 0;
1909 	tx_ring->tbd_free = tx_ring->ring_size;
1910 
1911 	/*
1912 	 * Note: Considering the case that the chipset is being reset,
1913 	 * and there are still some buffers held by the upper layer,
1914 	 * we should not reset the values of tcb_head, tcb_tail.
1915 	 */
1916 	if (igb->igb_state == IGB_UNKNOWN) {
1917 		tx_ring->tcb_head = 0;
1918 		tx_ring->tcb_tail = 0;
1919 		tx_ring->tcb_free = tx_ring->free_list_size;
1920 	} else {
1921 		ASSERT(tx_ring->tcb_free == tx_ring->free_list_size);
1922 	}
1923 
1924 	/*
1925 	 * Initialize hardware checksum offload settings
1926 	 */
1927 	tx_ring->hcksum_context.hcksum_flags = 0;
1928 	tx_ring->hcksum_context.ip_hdr_len = 0;
1929 	tx_ring->hcksum_context.mac_hdr_len = 0;
1930 	tx_ring->hcksum_context.l4_proto = 0;
1931 }
1932 
1933 static void
1934 igb_setup_tx(igb_t *igb)
1935 {
1936 	igb_tx_ring_t *tx_ring;
1937 	struct e1000_hw *hw = &igb->hw;
1938 	uint32_t reg_val;
1939 	int i;
1940 
1941 	for (i = 0; i < igb->num_tx_rings; i++) {
1942 		tx_ring = &igb->tx_rings[i];
1943 		igb_setup_tx_ring(tx_ring);
1944 	}
1945 
1946 	/*
1947 	 * Setup the Transmit Control Register (TCTL)
1948 	 */
1949 	reg_val = E1000_TCTL_PSP | E1000_TCTL_EN |
1950 	    (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT) |
1951 	    (E1000_COLLISION_DISTANCE << E1000_COLD_SHIFT) |
1952 	    E1000_TCTL_RTLC;
1953 
1954 	/* Enable the MULR bit */
1955 	if (hw->bus.type == e1000_bus_type_pci_express)
1956 		reg_val |= E1000_TCTL_MULR;
1957 
1958 	E1000_WRITE_REG(hw, E1000_TCTL, reg_val);
1959 
1960 	/*
1961 	 * Set the default values for the Tx Inter Packet Gap timer
1962 	 */
1963 	if (hw->phy.media_type == e1000_media_type_fiber)
1964 		reg_val = DEFAULT_82543_TIPG_IPGT_FIBER;
1965 	else
1966 		reg_val = DEFAULT_82543_TIPG_IPGT_COPPER;
1967 	reg_val |=
1968 	    DEFAULT_82543_TIPG_IPGR1 << E1000_TIPG_IPGR1_SHIFT;
1969 	reg_val |=
1970 	    DEFAULT_82543_TIPG_IPGR2 << E1000_TIPG_IPGR2_SHIFT;
1971 
1972 	E1000_WRITE_REG(hw, E1000_TIPG, reg_val);
1973 }
1974 
1975 /*
1976  * igb_setup_rss - Setup receive-side scaling feature
1977  */
1978 static void
1979 igb_setup_rss(igb_t *igb)
1980 {
1981 	struct e1000_hw *hw = &igb->hw;
1982 	uint32_t i, mrqc, rxcsum;
1983 	int shift;
1984 	uint32_t random;
1985 	union e1000_reta {
1986 		uint32_t	dword;
1987 		uint8_t		bytes[4];
1988 	} reta;
1989 
1990 	/* Setup the Redirection Table */
1991 	shift = 6;
1992 	for (i = 0; i < (32 * 4); i++) {
1993 		reta.bytes[i & 3] = (i % igb->num_rx_rings) << shift;
1994 		if ((i & 3) == 3) {
1995 			E1000_WRITE_REG(hw,
1996 			    (E1000_RETA(0) + (i & ~3)), reta.dword);
1997 		}
1998 	}
1999 
2000 	/* Fill out hash function seeds */
2001 	for (i = 0; i < 10; i++) {
2002 		(void) random_get_pseudo_bytes((uint8_t *)&random,
2003 		    sizeof (uint32_t));
2004 		E1000_WRITE_REG(hw, E1000_RSSRK(i), random);
2005 	}
2006 
2007 	/* Setup the Multiple Receive Queue Control register */
2008 	mrqc = E1000_MRQC_ENABLE_RSS_4Q;
2009 	mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2010 	    E1000_MRQC_RSS_FIELD_IPV4_TCP |
2011 	    E1000_MRQC_RSS_FIELD_IPV6 |
2012 	    E1000_MRQC_RSS_FIELD_IPV6_TCP |
2013 	    E1000_MRQC_RSS_FIELD_IPV4_UDP |
2014 	    E1000_MRQC_RSS_FIELD_IPV6_UDP |
2015 	    E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2016 	    E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2017 
2018 	E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
2019 
2020 	/*
2021 	 * Disable Packet Checksum to enable RSS for multiple receive queues.
2022 	 *
2023 	 * The Packet Checksum is not ethernet CRC. It is another kind of
2024 	 * checksum offloading provided by the 82575 chipset besides the IP
2025 	 * header checksum offloading and the TCP/UDP checksum offloading.
2026 	 * The Packet Checksum is by default computed over the entire packet
2027 	 * from the first byte of the DA through the last byte of the CRC,
2028 	 * including the Ethernet and IP headers.
2029 	 *
2030 	 * It is a hardware limitation that Packet Checksum is mutually
2031 	 * exclusive with RSS.
2032 	 */
2033 	rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
2034 	rxcsum |= E1000_RXCSUM_PCSD;
2035 	E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
2036 }
2037 
2038 /*
2039  * igb_init_unicst - Initialize the unicast addresses
2040  */
2041 static void
2042 igb_init_unicst(igb_t *igb)
2043 {
2044 	struct e1000_hw *hw = &igb->hw;
2045 	int slot;
2046 
2047 	/*
2048 	 * Here we should consider two situations:
2049 	 *
2050 	 * 1. Chipset is initialized the first time
2051 	 *    Initialize the multiple unicast addresses, and
2052 	 *    save the default mac address.
2053 	 *
2054 	 * 2. Chipset is reset
2055 	 *    Recover the multiple unicast addresses from the
2056 	 *    software data structure to the RAR registers.
2057 	 */
2058 	if (!igb->unicst_init) {
2059 		/* Initialize the multiple unicast addresses */
2060 		igb->unicst_total = MAX_NUM_UNICAST_ADDRESSES;
2061 
2062 		igb->unicst_avail = igb->unicst_total - 1;
2063 
2064 		/* Store the default mac address */
2065 		e1000_rar_set(hw, hw->mac.addr, 0);
2066 
2067 		bcopy(hw->mac.addr, igb->unicst_addr[0].mac.addr,
2068 		    ETHERADDRL);
2069 		igb->unicst_addr[0].mac.set = 1;
2070 
2071 		for (slot = 1; slot < igb->unicst_total; slot++)
2072 			igb->unicst_addr[slot].mac.set = 0;
2073 
2074 		igb->unicst_init = B_TRUE;
2075 	} else {
2076 		/* Recover the default mac address */
2077 		bcopy(igb->unicst_addr[0].mac.addr, hw->mac.addr,
2078 		    ETHERADDRL);
2079 
2080 		/* Store the default mac address */
2081 		e1000_rar_set(hw, hw->mac.addr, 0);
2082 
2083 		/* Re-configure the RAR registers */
2084 		for (slot = 1; slot < igb->unicst_total; slot++)
2085 			e1000_rar_set(hw,
2086 			    igb->unicst_addr[slot].mac.addr, slot);
2087 	}
2088 
2089 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
2090 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2091 }
2092 
2093 /*
2094  * igb_unicst_set - Set the unicast address to the specified slot
2095  */
2096 int
2097 igb_unicst_set(igb_t *igb, const uint8_t *mac_addr,
2098     mac_addr_slot_t slot)
2099 {
2100 	struct e1000_hw *hw = &igb->hw;
2101 
2102 	ASSERT(mutex_owned(&igb->gen_lock));
2103 
2104 	/*
2105 	 * Save the unicast address in the software data structure
2106 	 */
2107 	bcopy(mac_addr, igb->unicst_addr[slot].mac.addr, ETHERADDRL);
2108 
2109 	/*
2110 	 * Set the unicast address to the RAR register
2111 	 */
2112 	e1000_rar_set(hw, (uint8_t *)mac_addr, slot);
2113 
2114 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2115 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2116 		return (EIO);
2117 	}
2118 
2119 	return (0);
2120 }
2121 
2122 /*
2123  * igb_multicst_add - Add a multicst address
2124  */
2125 int
2126 igb_multicst_add(igb_t *igb, const uint8_t *multiaddr)
2127 {
2128 	ASSERT(mutex_owned(&igb->gen_lock));
2129 
2130 	if ((multiaddr[0] & 01) == 0) {
2131 		return (EINVAL);
2132 	}
2133 
2134 	if (igb->mcast_count >= MAX_NUM_MULTICAST_ADDRESSES) {
2135 		return (ENOENT);
2136 	}
2137 
2138 	bcopy(multiaddr,
2139 	    &igb->mcast_table[igb->mcast_count], ETHERADDRL);
2140 	igb->mcast_count++;
2141 
2142 	/*
2143 	 * Update the multicast table in the hardware
2144 	 */
2145 	igb_setup_multicst(igb);
2146 
2147 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2148 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2149 		return (EIO);
2150 	}
2151 
2152 	return (0);
2153 }
2154 
2155 /*
2156  * igb_multicst_remove - Remove a multicst address
2157  */
2158 int
2159 igb_multicst_remove(igb_t *igb, const uint8_t *multiaddr)
2160 {
2161 	int i;
2162 
2163 	ASSERT(mutex_owned(&igb->gen_lock));
2164 
2165 	for (i = 0; i < igb->mcast_count; i++) {
2166 		if (bcmp(multiaddr, &igb->mcast_table[i],
2167 		    ETHERADDRL) == 0) {
2168 			for (i++; i < igb->mcast_count; i++) {
2169 				igb->mcast_table[i - 1] =
2170 				    igb->mcast_table[i];
2171 			}
2172 			igb->mcast_count--;
2173 			break;
2174 		}
2175 	}
2176 
2177 	/*
2178 	 * Update the multicast table in the hardware
2179 	 */
2180 	igb_setup_multicst(igb);
2181 
2182 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2183 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2184 		return (EIO);
2185 	}
2186 
2187 	return (0);
2188 }
2189 
2190 /*
2191  * igb_setup_multicast - setup multicast data structures
2192  *
2193  * This routine initializes all of the multicast related structures
2194  * and save them in the hardware registers.
2195  */
2196 static void
2197 igb_setup_multicst(igb_t *igb)
2198 {
2199 	uint8_t *mc_addr_list;
2200 	uint32_t mc_addr_count;
2201 	struct e1000_hw *hw = &igb->hw;
2202 
2203 	ASSERT(mutex_owned(&igb->gen_lock));
2204 
2205 	ASSERT(igb->mcast_count <= MAX_NUM_MULTICAST_ADDRESSES);
2206 
2207 	mc_addr_list = (uint8_t *)igb->mcast_table;
2208 	mc_addr_count = igb->mcast_count;
2209 
2210 	/*
2211 	 * Update the multicase addresses to the MTA registers
2212 	 */
2213 	e1000_update_mc_addr_list(hw, mc_addr_list, mc_addr_count,
2214 	    igb->unicst_total, hw->mac.rar_entry_count);
2215 }
2216 
2217 /*
2218  * igb_get_conf - Get driver configurations set in driver.conf
2219  *
2220  * This routine gets user-configured values out of the configuration
2221  * file igb.conf.
2222  *
2223  * For each configurable value, there is a minimum, a maximum, and a
2224  * default.
2225  * If user does not configure a value, use the default.
2226  * If user configures below the minimum, use the minumum.
2227  * If user configures above the maximum, use the maxumum.
2228  */
2229 static void
2230 igb_get_conf(igb_t *igb)
2231 {
2232 	struct e1000_hw *hw = &igb->hw;
2233 	uint32_t default_mtu;
2234 	uint32_t flow_control;
2235 
2236 	/*
2237 	 * igb driver supports the following user configurations:
2238 	 *
2239 	 * Link configurations:
2240 	 *    adv_autoneg_cap
2241 	 *    adv_1000fdx_cap
2242 	 *    adv_100fdx_cap
2243 	 *    adv_100hdx_cap
2244 	 *    adv_10fdx_cap
2245 	 *    adv_10hdx_cap
2246 	 * Note: 1000hdx is not supported.
2247 	 *
2248 	 * Jumbo frame configuration:
2249 	 *    default_mtu
2250 	 *
2251 	 * Ethernet flow control configuration:
2252 	 *    flow_control
2253 	 *
2254 	 * Multiple rings configurations:
2255 	 *    tx_queue_number
2256 	 *    tx_ring_size
2257 	 *    rx_queue_number
2258 	 *    rx_ring_size
2259 	 *
2260 	 * Call igb_get_prop() to get the value for a specific
2261 	 * configuration parameter.
2262 	 */
2263 
2264 	/*
2265 	 * Link configurations
2266 	 */
2267 	igb->param_adv_autoneg_cap = igb_get_prop(igb,
2268 	    PROP_ADV_AUTONEG_CAP, 0, 1, 1);
2269 	igb->param_adv_1000fdx_cap = igb_get_prop(igb,
2270 	    PROP_ADV_1000FDX_CAP, 0, 1, 1);
2271 	igb->param_adv_100fdx_cap = igb_get_prop(igb,
2272 	    PROP_ADV_100FDX_CAP, 0, 1, 1);
2273 	igb->param_adv_100hdx_cap = igb_get_prop(igb,
2274 	    PROP_ADV_100HDX_CAP, 0, 1, 1);
2275 	igb->param_adv_10fdx_cap = igb_get_prop(igb,
2276 	    PROP_ADV_10FDX_CAP, 0, 1, 1);
2277 	igb->param_adv_10hdx_cap = igb_get_prop(igb,
2278 	    PROP_ADV_10HDX_CAP, 0, 1, 1);
2279 
2280 	/*
2281 	 * Jumbo frame configurations
2282 	 */
2283 	default_mtu = igb_get_prop(igb, PROP_DEFAULT_MTU,
2284 	    MIN_MTU, MAX_MTU, DEFAULT_MTU);
2285 
2286 	igb->max_frame_size = default_mtu +
2287 	    sizeof (struct ether_vlan_header) + ETHERFCSL;
2288 
2289 	/*
2290 	 * Ethernet flow control configuration
2291 	 */
2292 	flow_control = igb_get_prop(igb, PROP_FLOW_CONTROL,
2293 	    e1000_fc_none, 4, e1000_fc_full);
2294 	if (flow_control == 4)
2295 		flow_control = e1000_fc_default;
2296 
2297 	hw->fc.type = flow_control;
2298 
2299 	/*
2300 	 * Multiple rings configurations
2301 	 */
2302 	igb->num_tx_rings = igb_get_prop(igb, PROP_TX_QUEUE_NUM,
2303 	    MIN_TX_QUEUE_NUM, MAX_TX_QUEUE_NUM, DEFAULT_TX_QUEUE_NUM);
2304 	igb->tx_ring_size = igb_get_prop(igb, PROP_TX_RING_SIZE,
2305 	    MIN_TX_RING_SIZE, MAX_TX_RING_SIZE, DEFAULT_TX_RING_SIZE);
2306 
2307 	igb->num_rx_rings = igb_get_prop(igb, PROP_RX_QUEUE_NUM,
2308 	    MIN_RX_QUEUE_NUM, MAX_RX_QUEUE_NUM, DEFAULT_RX_QUEUE_NUM);
2309 	igb->rx_ring_size = igb_get_prop(igb, PROP_RX_RING_SIZE,
2310 	    MIN_RX_RING_SIZE, MAX_RX_RING_SIZE, DEFAULT_RX_RING_SIZE);
2311 
2312 	/*
2313 	 * Tunable used to force an interrupt type. The only use is
2314 	 * for testing of the lesser interrupt types.
2315 	 * 0 = don't force interrupt type
2316 	 * 1 = force interrupt type MSIX
2317 	 * 2 = force interrupt type MSI
2318 	 * 3 = force interrupt type Legacy
2319 	 */
2320 	igb->intr_force = igb_get_prop(igb, PROP_INTR_FORCE,
2321 	    IGB_INTR_NONE, IGB_INTR_LEGACY, IGB_INTR_NONE);
2322 
2323 	igb->tx_hcksum_enable = igb_get_prop(igb, PROP_TX_HCKSUM_ENABLE,
2324 	    0, 1, 1);
2325 	igb->rx_hcksum_enable = igb_get_prop(igb, PROP_RX_HCKSUM_ENABLE,
2326 	    0, 1, 1);
2327 	igb->lso_enable = igb_get_prop(igb, PROP_LSO_ENABLE,
2328 	    0, 1, 0);
2329 	igb->tx_head_wb_enable = igb_get_prop(igb, PROP_TX_HEAD_WB_ENABLE,
2330 	    0, 1, 1);
2331 
2332 	igb->tx_copy_thresh = igb_get_prop(igb, PROP_TX_COPY_THRESHOLD,
2333 	    MIN_TX_COPY_THRESHOLD, MAX_TX_COPY_THRESHOLD,
2334 	    DEFAULT_TX_COPY_THRESHOLD);
2335 	igb->tx_recycle_thresh = igb_get_prop(igb, PROP_TX_RECYCLE_THRESHOLD,
2336 	    MIN_TX_RECYCLE_THRESHOLD, MAX_TX_RECYCLE_THRESHOLD,
2337 	    DEFAULT_TX_RECYCLE_THRESHOLD);
2338 	igb->tx_overload_thresh = igb_get_prop(igb, PROP_TX_OVERLOAD_THRESHOLD,
2339 	    MIN_TX_OVERLOAD_THRESHOLD, MAX_TX_OVERLOAD_THRESHOLD,
2340 	    DEFAULT_TX_OVERLOAD_THRESHOLD);
2341 	igb->tx_resched_thresh = igb_get_prop(igb, PROP_TX_RESCHED_THRESHOLD,
2342 	    MIN_TX_RESCHED_THRESHOLD, MAX_TX_RESCHED_THRESHOLD,
2343 	    DEFAULT_TX_RESCHED_THRESHOLD);
2344 
2345 	igb->rx_copy_thresh = igb_get_prop(igb, PROP_RX_COPY_THRESHOLD,
2346 	    MIN_RX_COPY_THRESHOLD, MAX_RX_COPY_THRESHOLD,
2347 	    DEFAULT_RX_COPY_THRESHOLD);
2348 	igb->rx_limit_per_intr = igb_get_prop(igb, PROP_RX_LIMIT_PER_INTR,
2349 	    MIN_RX_LIMIT_PER_INTR, MAX_RX_LIMIT_PER_INTR,
2350 	    DEFAULT_RX_LIMIT_PER_INTR);
2351 
2352 	igb->intr_throttling[0] = igb_get_prop(igb, PROP_INTR_THROTTLING,
2353 	    MIN_INTR_THROTTLING, MAX_INTR_THROTTLING,
2354 	    DEFAULT_INTR_THROTTLING);
2355 }
2356 
2357 /*
2358  * igb_get_prop - Get a property value out of the configuration file igb.conf
2359  *
2360  * Caller provides the name of the property, a default value, a minimum
2361  * value, and a maximum value.
2362  *
2363  * Return configured value of the property, with default, minimum and
2364  * maximum properly applied.
2365  */
2366 static int
2367 igb_get_prop(igb_t *igb,
2368     char *propname,	/* name of the property */
2369     int minval,		/* minimum acceptable value */
2370     int maxval,		/* maximim acceptable value */
2371     int defval)		/* default value */
2372 {
2373 	int value;
2374 
2375 	/*
2376 	 * Call ddi_prop_get_int() to read the conf settings
2377 	 */
2378 	value = ddi_prop_get_int(DDI_DEV_T_ANY, igb->dip,
2379 	    DDI_PROP_DONTPASS, propname, defval);
2380 
2381 	if (value > maxval)
2382 		value = maxval;
2383 
2384 	if (value < minval)
2385 		value = minval;
2386 
2387 	return (value);
2388 }
2389 
2390 /*
2391  * igb_setup_link - Using the link properties to setup the link
2392  */
2393 int
2394 igb_setup_link(igb_t *igb, boolean_t setup_hw)
2395 {
2396 	struct e1000_mac_info *mac;
2397 	struct e1000_phy_info *phy;
2398 	boolean_t invalid;
2399 
2400 	mac = &igb->hw.mac;
2401 	phy = &igb->hw.phy;
2402 	invalid = B_FALSE;
2403 
2404 	if (igb->param_adv_autoneg_cap == 1) {
2405 		mac->autoneg = B_TRUE;
2406 		phy->autoneg_advertised = 0;
2407 
2408 		/*
2409 		 * 1000hdx is not supported for autonegotiation
2410 		 */
2411 		if (igb->param_adv_1000fdx_cap == 1)
2412 			phy->autoneg_advertised |= ADVERTISE_1000_FULL;
2413 
2414 		if (igb->param_adv_100fdx_cap == 1)
2415 			phy->autoneg_advertised |= ADVERTISE_100_FULL;
2416 
2417 		if (igb->param_adv_100hdx_cap == 1)
2418 			phy->autoneg_advertised |= ADVERTISE_100_HALF;
2419 
2420 		if (igb->param_adv_10fdx_cap == 1)
2421 			phy->autoneg_advertised |= ADVERTISE_10_FULL;
2422 
2423 		if (igb->param_adv_10hdx_cap == 1)
2424 			phy->autoneg_advertised |= ADVERTISE_10_HALF;
2425 
2426 		if (phy->autoneg_advertised == 0)
2427 			invalid = B_TRUE;
2428 	} else {
2429 		mac->autoneg = B_FALSE;
2430 
2431 		/*
2432 		 * 1000fdx and 1000hdx are not supported for forced link
2433 		 */
2434 		if (igb->param_adv_100fdx_cap == 1)
2435 			mac->forced_speed_duplex = ADVERTISE_100_FULL;
2436 		else if (igb->param_adv_100hdx_cap == 1)
2437 			mac->forced_speed_duplex = ADVERTISE_100_HALF;
2438 		else if (igb->param_adv_10fdx_cap == 1)
2439 			mac->forced_speed_duplex = ADVERTISE_10_FULL;
2440 		else if (igb->param_adv_10hdx_cap == 1)
2441 			mac->forced_speed_duplex = ADVERTISE_10_HALF;
2442 		else
2443 			invalid = B_TRUE;
2444 	}
2445 
2446 	if (invalid) {
2447 		igb_notice(igb, "Invalid link settings. Setup link to "
2448 		    "autonegotiation with full link capabilities.");
2449 		mac->autoneg = B_TRUE;
2450 		phy->autoneg_advertised = ADVERTISE_1000_FULL |
2451 		    ADVERTISE_100_FULL | ADVERTISE_100_HALF |
2452 		    ADVERTISE_10_FULL | ADVERTISE_10_HALF;
2453 	}
2454 
2455 	if (setup_hw) {
2456 		if (e1000_setup_link(&igb->hw) != E1000_SUCCESS)
2457 			return (IGB_FAILURE);
2458 	}
2459 
2460 	return (IGB_SUCCESS);
2461 }
2462 
2463 
2464 /*
2465  * igb_is_link_up - Check if the link is up
2466  */
2467 static boolean_t
2468 igb_is_link_up(igb_t *igb)
2469 {
2470 	struct e1000_hw *hw = &igb->hw;
2471 	boolean_t link_up;
2472 
2473 	ASSERT(mutex_owned(&igb->gen_lock));
2474 
2475 	(void) e1000_check_for_link(hw);
2476 
2477 	if ((E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU) ||
2478 	    ((hw->phy.media_type == e1000_media_type_internal_serdes) &&
2479 	    (hw->mac.serdes_has_link))) {
2480 		link_up = B_TRUE;
2481 	} else {
2482 		link_up = B_FALSE;
2483 	}
2484 
2485 	return (link_up);
2486 }
2487 
2488 /*
2489  * igb_link_check - Link status processing
2490  */
2491 static boolean_t
2492 igb_link_check(igb_t *igb)
2493 {
2494 	struct e1000_hw *hw = &igb->hw;
2495 	uint16_t speed = 0, duplex = 0;
2496 	boolean_t link_changed = B_FALSE;
2497 
2498 	ASSERT(mutex_owned(&igb->gen_lock));
2499 
2500 	if (igb_is_link_up(igb)) {
2501 		/*
2502 		 * The Link is up, check whether it was marked as down earlier
2503 		 */
2504 		if (igb->link_state != LINK_STATE_UP) {
2505 			(void) e1000_get_speed_and_duplex(hw, &speed, &duplex);
2506 			igb->link_speed = speed;
2507 			igb->link_duplex = duplex;
2508 			igb->link_state = LINK_STATE_UP;
2509 			igb->link_down_timeout = 0;
2510 			link_changed = B_TRUE;
2511 		}
2512 	} else {
2513 		if (igb->link_state != LINK_STATE_DOWN) {
2514 			igb->link_speed = 0;
2515 			igb->link_duplex = 0;
2516 			igb->link_state = LINK_STATE_DOWN;
2517 			link_changed = B_TRUE;
2518 		}
2519 
2520 		if (igb->igb_state & IGB_STARTED) {
2521 			if (igb->link_down_timeout < MAX_LINK_DOWN_TIMEOUT) {
2522 				igb->link_down_timeout++;
2523 			} else if (igb->link_down_timeout ==
2524 			    MAX_LINK_DOWN_TIMEOUT) {
2525 				igb_tx_clean(igb);
2526 				igb->link_down_timeout++;
2527 			}
2528 		}
2529 	}
2530 
2531 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
2532 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2533 
2534 	return (link_changed);
2535 }
2536 
2537 /*
2538  * igb_local_timer - driver watchdog function
2539  *
2540  * This function will handle the transmit stall check, link status check and
2541  * other routines.
2542  */
2543 static void
2544 igb_local_timer(void *arg)
2545 {
2546 	igb_t *igb = (igb_t *)arg;
2547 	struct e1000_hw *hw = &igb->hw;
2548 	boolean_t link_changed;
2549 
2550 	if (igb_stall_check(igb)) {
2551 		igb_fm_ereport(igb, DDI_FM_DEVICE_STALL);
2552 		igb->reset_count++;
2553 		if (igb_reset(igb) == IGB_SUCCESS)
2554 			ddi_fm_service_impact(igb->dip,
2555 			    DDI_SERVICE_RESTORED);
2556 	}
2557 
2558 	mutex_enter(&igb->gen_lock);
2559 	link_changed = igb_link_check(igb);
2560 	mutex_exit(&igb->gen_lock);
2561 
2562 	if (link_changed)
2563 		mac_link_update(igb->mac_hdl, igb->link_state);
2564 
2565 	/*
2566 	 * Set Timer Interrupts
2567 	 */
2568 	if (igb->intr_type != DDI_INTR_TYPE_MSIX)
2569 		E1000_WRITE_REG(hw, E1000_ICS, E1000_IMS_RXT0);
2570 
2571 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK)
2572 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2573 
2574 	igb_restart_watchdog_timer(igb);
2575 }
2576 
2577 /*
2578  * igb_stall_check - check for transmit stall
2579  *
2580  * This function checks if the adapter is stalled (in transmit).
2581  *
2582  * It is called each time the watchdog timeout is invoked.
2583  * If the transmit descriptor reclaim continuously fails,
2584  * the watchdog value will increment by 1. If the watchdog
2585  * value exceeds the threshold, the igb is assumed to
2586  * have stalled and need to be reset.
2587  */
2588 static boolean_t
2589 igb_stall_check(igb_t *igb)
2590 {
2591 	igb_tx_ring_t *tx_ring;
2592 	boolean_t result;
2593 	int i;
2594 
2595 	if (igb->link_state != LINK_STATE_UP)
2596 		return (B_FALSE);
2597 
2598 	/*
2599 	 * If any tx ring is stalled, we'll reset the chipset
2600 	 */
2601 	result = B_FALSE;
2602 	for (i = 0; i < igb->num_tx_rings; i++) {
2603 		tx_ring = &igb->tx_rings[i];
2604 
2605 		if (tx_ring->recycle_fail > 0)
2606 			tx_ring->stall_watchdog++;
2607 		else
2608 			tx_ring->stall_watchdog = 0;
2609 
2610 		if (tx_ring->stall_watchdog >= STALL_WATCHDOG_TIMEOUT) {
2611 			result = B_TRUE;
2612 			break;
2613 		}
2614 	}
2615 
2616 	if (result) {
2617 		tx_ring->stall_watchdog = 0;
2618 		tx_ring->recycle_fail = 0;
2619 	}
2620 
2621 	return (result);
2622 }
2623 
2624 
2625 /*
2626  * is_valid_mac_addr - Check if the mac address is valid
2627  */
2628 static boolean_t
2629 is_valid_mac_addr(uint8_t *mac_addr)
2630 {
2631 	const uint8_t addr_test1[6] = { 0, 0, 0, 0, 0, 0 };
2632 	const uint8_t addr_test2[6] =
2633 	    { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF };
2634 
2635 	if (!(bcmp(addr_test1, mac_addr, ETHERADDRL)) ||
2636 	    !(bcmp(addr_test2, mac_addr, ETHERADDRL)))
2637 		return (B_FALSE);
2638 
2639 	return (B_TRUE);
2640 }
2641 
2642 static boolean_t
2643 igb_find_mac_address(igb_t *igb)
2644 {
2645 	struct e1000_hw *hw = &igb->hw;
2646 #ifdef __sparc
2647 	uchar_t *bytes;
2648 	struct ether_addr sysaddr;
2649 	uint_t nelts;
2650 	int err;
2651 	boolean_t found = B_FALSE;
2652 
2653 	/*
2654 	 * The "vendor's factory-set address" may already have
2655 	 * been extracted from the chip, but if the property
2656 	 * "local-mac-address" is set we use that instead.
2657 	 *
2658 	 * We check whether it looks like an array of 6
2659 	 * bytes (which it should, if OBP set it).  If we can't
2660 	 * make sense of it this way, we'll ignore it.
2661 	 */
2662 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip,
2663 	    DDI_PROP_DONTPASS, "local-mac-address", &bytes, &nelts);
2664 	if (err == DDI_PROP_SUCCESS) {
2665 		if (nelts == ETHERADDRL) {
2666 			while (nelts--)
2667 				hw->mac.addr[nelts] = bytes[nelts];
2668 			found = B_TRUE;
2669 		}
2670 		ddi_prop_free(bytes);
2671 	}
2672 
2673 	/*
2674 	 * Look up the OBP property "local-mac-address?". If the user has set
2675 	 * 'local-mac-address? = false', use "the system address" instead.
2676 	 */
2677 	if (ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip, 0,
2678 	    "local-mac-address?", &bytes, &nelts) == DDI_PROP_SUCCESS) {
2679 		if (strncmp("false", (caddr_t)bytes, (size_t)nelts) == 0) {
2680 			if (localetheraddr(NULL, &sysaddr) != 0) {
2681 				bcopy(&sysaddr, hw->mac.addr, ETHERADDRL);
2682 				found = B_TRUE;
2683 			}
2684 		}
2685 		ddi_prop_free(bytes);
2686 	}
2687 
2688 	/*
2689 	 * Finally(!), if there's a valid "mac-address" property (created
2690 	 * if we netbooted from this interface), we must use this instead
2691 	 * of any of the above to ensure that the NFS/install server doesn't
2692 	 * get confused by the address changing as Solaris takes over!
2693 	 */
2694 	err = ddi_prop_lookup_byte_array(DDI_DEV_T_ANY, igb->dip,
2695 	    DDI_PROP_DONTPASS, "mac-address", &bytes, &nelts);
2696 	if (err == DDI_PROP_SUCCESS) {
2697 		if (nelts == ETHERADDRL) {
2698 			while (nelts--)
2699 				hw->mac.addr[nelts] = bytes[nelts];
2700 			found = B_TRUE;
2701 		}
2702 		ddi_prop_free(bytes);
2703 	}
2704 
2705 	if (found) {
2706 		bcopy(hw->mac.addr, hw->mac.perm_addr, ETHERADDRL);
2707 		return (B_TRUE);
2708 	}
2709 #endif
2710 
2711 	/*
2712 	 * Read the device MAC address from the EEPROM
2713 	 */
2714 	if (e1000_read_mac_addr(hw) != E1000_SUCCESS)
2715 		return (B_FALSE);
2716 
2717 	return (B_TRUE);
2718 }
2719 
2720 #pragma inline(igb_arm_watchdog_timer)
2721 
2722 static void
2723 igb_arm_watchdog_timer(igb_t *igb)
2724 {
2725 	/*
2726 	 * Fire a watchdog timer
2727 	 */
2728 	igb->watchdog_tid =
2729 	    timeout(igb_local_timer,
2730 	    (void *)igb, 1 * drv_usectohz(1000000));
2731 
2732 }
2733 
2734 /*
2735  * igb_enable_watchdog_timer - Enable and start the driver watchdog timer
2736  */
2737 void
2738 igb_enable_watchdog_timer(igb_t *igb)
2739 {
2740 	mutex_enter(&igb->watchdog_lock);
2741 
2742 	if (!igb->watchdog_enable) {
2743 		igb->watchdog_enable = B_TRUE;
2744 		igb->watchdog_start = B_TRUE;
2745 		igb_arm_watchdog_timer(igb);
2746 	}
2747 
2748 	mutex_exit(&igb->watchdog_lock);
2749 
2750 }
2751 
2752 /*
2753  * igb_disable_watchdog_timer - Disable and stop the driver watchdog timer
2754  */
2755 void
2756 igb_disable_watchdog_timer(igb_t *igb)
2757 {
2758 	timeout_id_t tid;
2759 
2760 	mutex_enter(&igb->watchdog_lock);
2761 
2762 	igb->watchdog_enable = B_FALSE;
2763 	igb->watchdog_start = B_FALSE;
2764 	tid = igb->watchdog_tid;
2765 	igb->watchdog_tid = 0;
2766 
2767 	mutex_exit(&igb->watchdog_lock);
2768 
2769 	if (tid != 0)
2770 		(void) untimeout(tid);
2771 
2772 }
2773 
2774 /*
2775  * igb_start_watchdog_timer - Start the driver watchdog timer
2776  */
2777 static void
2778 igb_start_watchdog_timer(igb_t *igb)
2779 {
2780 	mutex_enter(&igb->watchdog_lock);
2781 
2782 	if (igb->watchdog_enable) {
2783 		if (!igb->watchdog_start) {
2784 			igb->watchdog_start = B_TRUE;
2785 			igb_arm_watchdog_timer(igb);
2786 		}
2787 	}
2788 
2789 	mutex_exit(&igb->watchdog_lock);
2790 }
2791 
2792 /*
2793  * igb_restart_watchdog_timer - Restart the driver watchdog timer
2794  */
2795 static void
2796 igb_restart_watchdog_timer(igb_t *igb)
2797 {
2798 	mutex_enter(&igb->watchdog_lock);
2799 
2800 	if (igb->watchdog_start)
2801 		igb_arm_watchdog_timer(igb);
2802 
2803 	mutex_exit(&igb->watchdog_lock);
2804 }
2805 
2806 /*
2807  * igb_stop_watchdog_timer - Stop the driver watchdog timer
2808  */
2809 static void
2810 igb_stop_watchdog_timer(igb_t *igb)
2811 {
2812 	timeout_id_t tid;
2813 
2814 	mutex_enter(&igb->watchdog_lock);
2815 
2816 	igb->watchdog_start = B_FALSE;
2817 	tid = igb->watchdog_tid;
2818 	igb->watchdog_tid = 0;
2819 
2820 	mutex_exit(&igb->watchdog_lock);
2821 
2822 	if (tid != 0)
2823 		(void) untimeout(tid);
2824 }
2825 
2826 /*
2827  * igb_disable_adapter_interrupts - Clear/disable all hardware interrupts
2828  */
2829 static void
2830 igb_disable_adapter_interrupts(igb_t *igb)
2831 {
2832 	struct e1000_hw *hw = &igb->hw;
2833 
2834 	/*
2835 	 * Set the IMC register to mask all the interrupts,
2836 	 * including the tx interrupts.
2837 	 */
2838 	E1000_WRITE_REG(hw, E1000_IMC, 0xffffffff);
2839 
2840 	/*
2841 	 * Additional disabling for MSI-X
2842 	 */
2843 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
2844 		E1000_WRITE_REG(hw, E1000_EIMC, 0xffffffff);
2845 		E1000_WRITE_REG(hw, E1000_EIAC, 0x0);
2846 	}
2847 
2848 	E1000_WRITE_FLUSH(hw);
2849 }
2850 
2851 /*
2852  * igb_enable_adapter_interrupts - Mask/enable all hardware interrupts
2853  */
2854 static void
2855 igb_enable_adapter_interrupts(igb_t *igb)
2856 {
2857 	struct e1000_hw *hw = &igb->hw;
2858 	uint32_t reg;
2859 
2860 	if (igb->intr_type == DDI_INTR_TYPE_MSIX) {
2861 		/* Interrupt enabling for MSI-X */
2862 		E1000_WRITE_REG(hw, E1000_EIMS, igb->eims_mask);
2863 		E1000_WRITE_REG(hw, E1000_EIAC, igb->eims_mask);
2864 		E1000_WRITE_REG(hw, E1000_IMS, E1000_IMS_LSC);
2865 
2866 		/* Enable MSI-X PBA support */
2867 		reg = E1000_READ_REG(hw, E1000_CTRL_EXT);
2868 		reg |= E1000_CTRL_EXT_PBA_CLR;
2869 
2870 		/* Non-selective interrupt clear-on-read */
2871 		reg |= E1000_CTRL_EXT_IRCA;	/* Called NSICR in the EAS */
2872 
2873 		E1000_WRITE_REG(hw, E1000_CTRL_EXT, reg);
2874 	} else {
2875 		/* Interrupt enabling for MSI and legacy */
2876 		E1000_WRITE_REG(hw, E1000_IMS, IMS_ENABLE_MASK);
2877 	}
2878 
2879 	E1000_WRITE_FLUSH(hw);
2880 }
2881 
2882 /*
2883  * Loopback Support
2884  */
2885 static lb_property_t lb_normal =
2886 	{ normal,	"normal",	IGB_LB_NONE		};
2887 static lb_property_t lb_external =
2888 	{ external,	"External",	IGB_LB_EXTERNAL		};
2889 static lb_property_t lb_mac =
2890 	{ internal,	"MAC",		IGB_LB_INTERNAL_MAC	};
2891 static lb_property_t lb_phy =
2892 	{ internal,	"PHY",		IGB_LB_INTERNAL_PHY	};
2893 static lb_property_t lb_serdes =
2894 	{ internal,	"SerDes",	IGB_LB_INTERNAL_SERDES	};
2895 
2896 enum ioc_reply
2897 igb_loopback_ioctl(igb_t *igb, struct iocblk *iocp, mblk_t *mp)
2898 {
2899 	lb_info_sz_t *lbsp;
2900 	lb_property_t *lbpp;
2901 	struct e1000_hw *hw;
2902 	uint32_t *lbmp;
2903 	uint32_t size;
2904 	uint32_t value;
2905 
2906 	hw = &igb->hw;
2907 
2908 	if (mp->b_cont == NULL)
2909 		return (IOC_INVAL);
2910 
2911 	switch (iocp->ioc_cmd) {
2912 	default:
2913 		return (IOC_INVAL);
2914 
2915 	case LB_GET_INFO_SIZE:
2916 		size = sizeof (lb_info_sz_t);
2917 		if (iocp->ioc_count != size)
2918 			return (IOC_INVAL);
2919 
2920 		value = sizeof (lb_normal);
2921 		value += sizeof (lb_mac);
2922 		if (hw->phy.media_type == e1000_media_type_copper)
2923 			value += sizeof (lb_phy);
2924 		else
2925 			value += sizeof (lb_serdes);
2926 		value += sizeof (lb_external);
2927 
2928 		lbsp = (lb_info_sz_t *)(uintptr_t)mp->b_cont->b_rptr;
2929 		*lbsp = value;
2930 		break;
2931 
2932 	case LB_GET_INFO:
2933 		value = sizeof (lb_normal);
2934 		value += sizeof (lb_mac);
2935 		if (hw->phy.media_type == e1000_media_type_copper)
2936 			value += sizeof (lb_phy);
2937 		else
2938 			value += sizeof (lb_serdes);
2939 		value += sizeof (lb_external);
2940 
2941 		size = value;
2942 		if (iocp->ioc_count != size)
2943 			return (IOC_INVAL);
2944 
2945 		value = 0;
2946 		lbpp = (lb_property_t *)(uintptr_t)mp->b_cont->b_rptr;
2947 
2948 		lbpp[value++] = lb_normal;
2949 		lbpp[value++] = lb_mac;
2950 		if (hw->phy.media_type == e1000_media_type_copper)
2951 			lbpp[value++] = lb_phy;
2952 		else
2953 			lbpp[value++] = lb_serdes;
2954 		lbpp[value++] = lb_external;
2955 		break;
2956 
2957 	case LB_GET_MODE:
2958 		size = sizeof (uint32_t);
2959 		if (iocp->ioc_count != size)
2960 			return (IOC_INVAL);
2961 
2962 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
2963 		*lbmp = igb->loopback_mode;
2964 		break;
2965 
2966 	case LB_SET_MODE:
2967 		size = 0;
2968 		if (iocp->ioc_count != sizeof (uint32_t))
2969 			return (IOC_INVAL);
2970 
2971 		lbmp = (uint32_t *)(uintptr_t)mp->b_cont->b_rptr;
2972 		if (!igb_set_loopback_mode(igb, *lbmp))
2973 			return (IOC_INVAL);
2974 		break;
2975 	}
2976 
2977 	iocp->ioc_count = size;
2978 	iocp->ioc_error = 0;
2979 
2980 	if (igb_check_acc_handle(igb->osdep.reg_handle) != DDI_FM_OK) {
2981 		ddi_fm_service_impact(igb->dip, DDI_SERVICE_DEGRADED);
2982 		return (IOC_INVAL);
2983 	}
2984 
2985 	return (IOC_REPLY);
2986 }
2987 
2988 /*
2989  * igb_set_loopback_mode - Setup loopback based on the loopback mode
2990  */
2991 static boolean_t
2992 igb_set_loopback_mode(igb_t *igb, uint32_t mode)
2993 {
2994 	struct e1000_hw *hw;
2995 
2996 	if (mode == igb->loopback_mode)
2997 		return (B_TRUE);
2998 
2999 	hw = &igb->hw;
3000 
3001 	igb->loopback_mode = mode;
3002 
3003 	if (mode == IGB_LB_NONE) {
3004 		/* Reset the chip */
3005 		hw->phy.autoneg_wait_to_complete = B_TRUE;
3006 		(void) igb_reset(igb);
3007 		hw->phy.autoneg_wait_to_complete = B_FALSE;
3008 		return (B_TRUE);
3009 	}
3010 
3011 	mutex_enter(&igb->gen_lock);
3012 
3013 	switch (mode) {
3014 	default:
3015 		mutex_exit(&igb->gen_lock);
3016 		return (B_FALSE);
3017 
3018 	case IGB_LB_EXTERNAL:
3019 		igb_set_external_loopback(igb);
3020 		break;
3021 
3022 	case IGB_LB_INTERNAL_MAC:
3023 		igb_set_internal_mac_loopback(igb);
3024 		break;
3025 
3026 	case IGB_LB_INTERNAL_PHY:
3027 		igb_set_internal_phy_loopback(igb);
3028 		break;
3029 
3030 	case IGB_LB_INTERNAL_SERDES:
3031 		igb_set_internal_serdes_loopback(igb);
3032 		break;
3033 	}
3034 
3035 	mutex_exit(&igb->gen_lock);
3036 
3037 	return (B_TRUE);
3038 }
3039 
3040 /*
3041  * igb_set_external_loopback - Set the external loopback mode
3042  */
3043 static void
3044 igb_set_external_loopback(igb_t *igb)
3045 {
3046 	struct e1000_hw *hw;
3047 
3048 	hw = &igb->hw;
3049 
3050 	/* Set phy to known state */
3051 	(void) e1000_phy_hw_reset(hw);
3052 
3053 	(void) e1000_write_phy_reg(hw, 0x0, 0x0140);
3054 	(void) e1000_write_phy_reg(hw, 0x9, 0x1b00);
3055 	(void) e1000_write_phy_reg(hw, 0x12, 0x1610);
3056 	(void) e1000_write_phy_reg(hw, 0x1f37, 0x3f1c);
3057 }
3058 
3059 /*
3060  * igb_set_internal_mac_loopback - Set the internal MAC loopback mode
3061  */
3062 static void
3063 igb_set_internal_mac_loopback(igb_t *igb)
3064 {
3065 	struct e1000_hw *hw;
3066 	uint32_t ctrl;
3067 	uint32_t rctl;
3068 
3069 	hw = &igb->hw;
3070 
3071 	/* Set the Receive Control register */
3072 	rctl = E1000_READ_REG(hw, E1000_RCTL);
3073 	rctl &= ~E1000_RCTL_LBM_TCVR;
3074 	rctl |= E1000_RCTL_LBM_MAC;
3075 	E1000_WRITE_REG(hw, E1000_RCTL, rctl);
3076 
3077 	/* Set the Device Control register */
3078 	ctrl = E1000_READ_REG(hw, E1000_CTRL);
3079 	ctrl &= ~E1000_CTRL_SPD_SEL;	/* Clear the speed sel bits */
3080 	ctrl |= (E1000_CTRL_SLU |	/* Force link up */
3081 	    E1000_CTRL_FRCSPD |		/* Force speed */
3082 	    E1000_CTRL_FRCDPX |		/* Force duplex */
3083 	    E1000_CTRL_SPD_1000 |	/* Force speed to 1000 */
3084 	    E1000_CTRL_FD);		/* Force full duplex */
3085 	ctrl &= ~E1000_CTRL_ILOS;	/* Clear ILOS when there's a link */
3086 	E1000_WRITE_REG(hw, E1000_CTRL, ctrl);
3087 }
3088 
3089 /*
3090  * igb_set_internal_phy_loopback - Set the internal PHY loopback mode
3091  */
3092 static void
3093 igb_set_internal_phy_loopback(igb_t *igb)
3094 {
3095 	struct e1000_hw *hw;
3096 	uint32_t ctrl_ext;
3097 	uint16_t phy_ctrl;
3098 	uint16_t phy_pconf;
3099 
3100 	hw = &igb->hw;
3101 
3102 	/* Set link mode to PHY (00b) in the Extended Control register */
3103 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3104 	ctrl_ext &= ~E1000_CTRL_EXT_LINK_MODE_MASK;
3105 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3106 
3107 	/*
3108 	 * Set PHY control register (0x4140):
3109 	 *    Set full duplex mode
3110 	 *    Set loopback bit
3111 	 *    Clear auto-neg enable bit
3112 	 *    Set PHY speed
3113 	 */
3114 	phy_ctrl = MII_CR_FULL_DUPLEX | MII_CR_SPEED_1000 | MII_CR_LOOPBACK;
3115 	(void) e1000_write_phy_reg(hw, PHY_CONTROL, phy_ctrl);
3116 
3117 	/* Set the link disable bit in the Port Configuration register */
3118 	(void) e1000_read_phy_reg(hw, 0x10, &phy_pconf);
3119 	phy_pconf |= (uint16_t)1 << 14;
3120 	(void) e1000_write_phy_reg(hw, 0x10, phy_pconf);
3121 }
3122 
3123 /*
3124  * igb_set_internal_serdes_loopback - Set the internal SerDes loopback mode
3125  */
3126 static void
3127 igb_set_internal_serdes_loopback(igb_t *igb)
3128 {
3129 	struct e1000_hw *hw;
3130 	uint32_t ctrl_ext;
3131 	uint32_t ctrl;
3132 	uint32_t pcs_lctl;
3133 	uint32_t connsw;
3134 
3135 	hw = &igb->hw;
3136 
3137 	/* Set link mode to SerDes (11b) in the Extended Control register */
3138 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
3139 	ctrl_ext |= E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
3140 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
3141 
3142 	/* Configure the SerDes to loopback */
3143 	E1000_WRITE_REG(hw, E1000_SCTL, 0x410);
3144 
3145 	/* Set Device Control register */
3146 	ctrl = E1000_READ_REG(hw, E1000_CTRL);
3147 	ctrl |= (E1000_CTRL_FD |	/* Force full duplex */
3148 	    E1000_CTRL_SLU);		/* Force link up */
3149 	ctrl &= ~(E1000_CTRL_RFCE |	/* Disable receive flow control */
3150 	    E1000_CTRL_TFCE |		/* Disable transmit flow control */
3151 	    E1000_CTRL_LRST);		/* Clear link reset */
3152 	E1000_WRITE_REG(hw, E1000_CTRL, ctrl);
3153 
3154 	/* Set PCS Link Control register */
3155 	pcs_lctl = E1000_READ_REG(hw, E1000_PCS_LCTL);
3156 	pcs_lctl |= (E1000_PCS_LCTL_FORCE_LINK |
3157 	    E1000_PCS_LCTL_FSD |
3158 	    E1000_PCS_LCTL_FDV_FULL |
3159 	    E1000_PCS_LCTL_FLV_LINK_UP);
3160 	pcs_lctl &= ~E1000_PCS_LCTL_AN_ENABLE;
3161 	E1000_WRITE_REG(hw, E1000_PCS_LCTL, pcs_lctl);
3162 
3163 	/* Set the Copper/Fiber Switch Control - CONNSW register */
3164 	connsw = E1000_READ_REG(hw, E1000_CONNSW);
3165 	connsw &= ~E1000_CONNSW_ENRGSRC;
3166 	E1000_WRITE_REG(hw, E1000_CONNSW, connsw);
3167 }
3168 
3169 #pragma inline(igb_intr_rx_work)
3170 /*
3171  * igb_intr_rx_work - rx processing of ISR
3172  */
3173 static void
3174 igb_intr_rx_work(igb_rx_ring_t *rx_ring)
3175 {
3176 	mblk_t *mp;
3177 
3178 	mutex_enter(&rx_ring->rx_lock);
3179 	mp = igb_rx(rx_ring);
3180 	mutex_exit(&rx_ring->rx_lock);
3181 
3182 	if (mp != NULL)
3183 		mac_rx(rx_ring->igb->mac_hdl, NULL, mp);
3184 }
3185 
3186 #pragma inline(igb_intr_tx_work)
3187 /*
3188  * igb_intr_tx_work - tx processing of ISR
3189  */
3190 static void
3191 igb_intr_tx_work(igb_tx_ring_t *tx_ring)
3192 {
3193 	/* Recycle the tx descriptors */
3194 	tx_ring->tx_recycle(tx_ring);
3195 
3196 	/* Schedule the re-transmit */
3197 	if (tx_ring->reschedule &&
3198 	    (tx_ring->tbd_free >= tx_ring->resched_thresh)) {
3199 		tx_ring->reschedule = B_FALSE;
3200 		mac_tx_update(tx_ring->igb->mac_hdl);
3201 		IGB_DEBUG_STAT(tx_ring->stat_reschedule);
3202 	}
3203 }
3204 
3205 #pragma inline(igb_intr_other_work)
3206 /*
3207  * igb_intr_other_work - other processing of ISR
3208  */
3209 static void
3210 igb_intr_other_work(igb_t *igb)
3211 {
3212 	boolean_t link_changed;
3213 
3214 	igb_stop_watchdog_timer(igb);
3215 
3216 	mutex_enter(&igb->gen_lock);
3217 
3218 	/*
3219 	 * Because we got a link-status-change interrupt, force
3220 	 * e1000_check_for_link() to look at phy
3221 	 */
3222 	igb->hw.mac.get_link_status = B_TRUE;
3223 
3224 	/* igb_link_check takes care of link status change */
3225 	link_changed = igb_link_check(igb);
3226 
3227 	/* Get new phy state */
3228 	igb_get_phy_state(igb);
3229 
3230 	mutex_exit(&igb->gen_lock);
3231 
3232 	if (link_changed)
3233 		mac_link_update(igb->mac_hdl, igb->link_state);
3234 
3235 	igb_start_watchdog_timer(igb);
3236 }
3237 
3238 /*
3239  * igb_intr_legacy - Interrupt handler for legacy interrupts
3240  */
3241 static uint_t
3242 igb_intr_legacy(void *arg1, void *arg2)
3243 {
3244 	igb_t *igb = (igb_t *)arg1;
3245 	igb_tx_ring_t *tx_ring;
3246 	uint32_t icr;
3247 	mblk_t *mp;
3248 	boolean_t tx_reschedule;
3249 	boolean_t link_changed;
3250 	uint_t result;
3251 
3252 	_NOTE(ARGUNUSED(arg2));
3253 
3254 	mutex_enter(&igb->gen_lock);
3255 
3256 	if (igb->igb_state & IGB_SUSPENDED) {
3257 		mutex_exit(&igb->gen_lock);
3258 		return (DDI_INTR_UNCLAIMED);
3259 	}
3260 
3261 	mp = NULL;
3262 	tx_reschedule = B_FALSE;
3263 	link_changed = B_FALSE;
3264 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
3265 
3266 	if (icr & E1000_ICR_INT_ASSERTED) {
3267 		/*
3268 		 * E1000_ICR_INT_ASSERTED bit was set:
3269 		 * Read(Clear) the ICR, claim this interrupt,
3270 		 * look for work to do.
3271 		 */
3272 		ASSERT(igb->num_rx_rings == 1);
3273 		ASSERT(igb->num_tx_rings == 1);
3274 
3275 		if (icr & E1000_ICR_RXT0) {
3276 			mp = igb_rx(&igb->rx_rings[0]);
3277 		}
3278 
3279 		if (icr & E1000_ICR_TXDW) {
3280 			tx_ring = &igb->tx_rings[0];
3281 
3282 			/* Recycle the tx descriptors */
3283 			tx_ring->tx_recycle(tx_ring);
3284 
3285 			/* Schedule the re-transmit */
3286 			tx_reschedule = (tx_ring->reschedule &&
3287 			    (tx_ring->tbd_free >= tx_ring->resched_thresh));
3288 		}
3289 
3290 		if (icr & E1000_ICR_LSC) {
3291 			/*
3292 			 * Because we got a link-status-change interrupt, force
3293 			 * e1000_check_for_link() to look at phy
3294 			 */
3295 			igb->hw.mac.get_link_status = B_TRUE;
3296 
3297 			/* igb_link_check takes care of link status change */
3298 			link_changed = igb_link_check(igb);
3299 
3300 			/* Get new phy state */
3301 			igb_get_phy_state(igb);
3302 		}
3303 
3304 		result = DDI_INTR_CLAIMED;
3305 	} else {
3306 		/*
3307 		 * E1000_ICR_INT_ASSERTED bit was not set:
3308 		 * Don't claim this interrupt.
3309 		 */
3310 		result = DDI_INTR_UNCLAIMED;
3311 	}
3312 
3313 	mutex_exit(&igb->gen_lock);
3314 
3315 	/*
3316 	 * Do the following work outside of the gen_lock
3317 	 */
3318 	if (mp != NULL)
3319 		mac_rx(igb->mac_hdl, NULL, mp);
3320 
3321 	if (tx_reschedule)  {
3322 		tx_ring->reschedule = B_FALSE;
3323 		mac_tx_update(igb->mac_hdl);
3324 		IGB_DEBUG_STAT(tx_ring->stat_reschedule);
3325 	}
3326 
3327 	if (link_changed)
3328 		mac_link_update(igb->mac_hdl, igb->link_state);
3329 
3330 	return (result);
3331 }
3332 
3333 /*
3334  * igb_intr_msi - Interrupt handler for MSI
3335  */
3336 static uint_t
3337 igb_intr_msi(void *arg1, void *arg2)
3338 {
3339 	igb_t *igb = (igb_t *)arg1;
3340 	uint32_t icr;
3341 
3342 	_NOTE(ARGUNUSED(arg2));
3343 
3344 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
3345 
3346 	/*
3347 	 * For MSI interrupt, we have only one vector,
3348 	 * so we have only one rx ring and one tx ring enabled.
3349 	 */
3350 	ASSERT(igb->num_rx_rings == 1);
3351 	ASSERT(igb->num_tx_rings == 1);
3352 
3353 	if (icr & E1000_ICR_RXT0) {
3354 		igb_intr_rx_work(&igb->rx_rings[0]);
3355 	}
3356 
3357 	if (icr & E1000_ICR_TXDW) {
3358 		igb_intr_tx_work(&igb->tx_rings[0]);
3359 	}
3360 
3361 	if (icr & E1000_ICR_LSC) {
3362 		igb_intr_other_work(igb);
3363 	}
3364 
3365 	return (DDI_INTR_CLAIMED);
3366 }
3367 
3368 /*
3369  * igb_intr_rx - Interrupt handler for rx
3370  */
3371 static uint_t
3372 igb_intr_rx(void *arg1, void *arg2)
3373 {
3374 	igb_rx_ring_t *rx_ring = (igb_rx_ring_t *)arg1;
3375 
3376 	_NOTE(ARGUNUSED(arg2));
3377 
3378 	/*
3379 	 * Only used via MSI-X vector so don't check cause bits
3380 	 * and only clean the given ring.
3381 	 */
3382 	igb_intr_rx_work(rx_ring);
3383 
3384 	return (DDI_INTR_CLAIMED);
3385 }
3386 
3387 /*
3388  * igb_intr_tx_other - Interrupt handler for both tx and other
3389  *
3390  * Always look for Tx cleanup work.  Only look for other work if the right
3391  * bits are set in the Interrupt Cause Register.
3392  */
3393 static uint_t
3394 igb_intr_tx_other(void *arg1, void *arg2)
3395 {
3396 	igb_t *igb = (igb_t *)arg1;
3397 	uint32_t icr;
3398 
3399 	_NOTE(ARGUNUSED(arg2));
3400 
3401 	icr = E1000_READ_REG(&igb->hw, E1000_ICR);
3402 
3403 	/*
3404 	 * Always look for Tx cleanup work.  We don't have separate
3405 	 * transmit vectors, so we have only one tx ring enabled.
3406 	 */
3407 	ASSERT(igb->num_tx_rings == 1);
3408 	igb_intr_tx_work(&igb->tx_rings[0]);
3409 
3410 	/*
3411 	 * Check for "other" causes.
3412 	 */
3413 	if (icr & E1000_ICR_LSC) {
3414 		igb_intr_other_work(igb);
3415 	}
3416 
3417 	return (DDI_INTR_CLAIMED);
3418 }
3419 
3420 /*
3421  * igb_alloc_intrs - Allocate interrupts for the driver
3422  *
3423  * Normal sequence is to try MSI-X; if not sucessful, try MSI;
3424  * if not successful, try Legacy.
3425  * igb->intr_force can be used to force sequence to start with
3426  * any of the 3 types.
3427  * If MSI-X is not used, number of tx/rx rings is forced to 1.
3428  */
3429 static int
3430 igb_alloc_intrs(igb_t *igb)
3431 {
3432 	dev_info_t *devinfo;
3433 	int intr_types;
3434 	int rc;
3435 
3436 	devinfo = igb->dip;
3437 
3438 	/* Get supported interrupt types */
3439 	rc = ddi_intr_get_supported_types(devinfo, &intr_types);
3440 
3441 	if (rc != DDI_SUCCESS) {
3442 		igb_log(igb,
3443 		    "Get supported interrupt types failed: %d", rc);
3444 		return (IGB_FAILURE);
3445 	}
3446 	IGB_DEBUGLOG_1(igb, "Supported interrupt types: %x", intr_types);
3447 
3448 	igb->intr_type = 0;
3449 
3450 	/* Install MSI-X interrupts */
3451 	if ((intr_types & DDI_INTR_TYPE_MSIX) &&
3452 	    (igb->intr_force <= IGB_INTR_MSIX)) {
3453 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_MSIX);
3454 
3455 		if (rc == IGB_SUCCESS)
3456 			return (IGB_SUCCESS);
3457 
3458 		igb_log(igb,
3459 		    "Allocate MSI-X failed, trying MSI interrupts...");
3460 	}
3461 
3462 	/* MSI-X not used, force rings to 1 */
3463 	igb->num_rx_rings = 1;
3464 	igb->num_tx_rings = 1;
3465 	igb_log(igb,
3466 	    "MSI-X not used, force rx and tx queue number to 1");
3467 
3468 	/* Install MSI interrupts */
3469 	if ((intr_types & DDI_INTR_TYPE_MSI) &&
3470 	    (igb->intr_force <= IGB_INTR_MSI)) {
3471 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_MSI);
3472 
3473 		if (rc == IGB_SUCCESS)
3474 			return (IGB_SUCCESS);
3475 
3476 		igb_log(igb,
3477 		    "Allocate MSI failed, trying Legacy interrupts...");
3478 	}
3479 
3480 	/* Install legacy interrupts */
3481 	if (intr_types & DDI_INTR_TYPE_FIXED) {
3482 		rc = igb_alloc_intr_handles(igb, DDI_INTR_TYPE_FIXED);
3483 
3484 		if (rc == IGB_SUCCESS)
3485 			return (IGB_SUCCESS);
3486 
3487 		igb_log(igb,
3488 		    "Allocate Legacy interrupts failed");
3489 	}
3490 
3491 	/* If none of the 3 types succeeded, return failure */
3492 	return (IGB_FAILURE);
3493 }
3494 
3495 /*
3496  * igb_alloc_intr_handles - Allocate interrupt handles.
3497  *
3498  * For legacy and MSI, only 1 handle is needed.  For MSI-X,
3499  * if fewer than 2 handles are available, return failure.
3500  * Upon success, this sets the number of Rx rings to a number that
3501  * matches the handles available for Rx interrupts.
3502  */
3503 static int
3504 igb_alloc_intr_handles(igb_t *igb, int intr_type)
3505 {
3506 	dev_info_t *devinfo;
3507 	int request, count, avail, actual;
3508 	int rx_rings, minimum;
3509 	int rc;
3510 
3511 	devinfo = igb->dip;
3512 
3513 	/*
3514 	 * Currently only 1 tx ring is supported. More tx rings
3515 	 * will be supported with future enhancement.
3516 	 */
3517 	if (igb->num_tx_rings > 1) {
3518 		igb->num_tx_rings = 1;
3519 		igb_log(igb,
3520 		    "Use only 1 MSI-X vector for tx, "
3521 		    "force tx queue number to 1");
3522 	}
3523 
3524 	switch (intr_type) {
3525 	case DDI_INTR_TYPE_FIXED:
3526 		request = 1;	/* Request 1 legacy interrupt handle */
3527 		minimum = 1;
3528 		IGB_DEBUGLOG_0(igb, "interrupt type: legacy");
3529 		break;
3530 
3531 	case DDI_INTR_TYPE_MSI:
3532 		request = 1;	/* Request 1 MSI interrupt handle */
3533 		minimum = 1;
3534 		IGB_DEBUGLOG_0(igb, "interrupt type: MSI");
3535 		break;
3536 
3537 	case DDI_INTR_TYPE_MSIX:
3538 		/*
3539 		 * Best number of vectors for the adapter is
3540 		 * # rx rings + # tx rings + 1 for other
3541 		 * But currently we only support number of vectors of
3542 		 * # rx rings + 1 for tx & other
3543 		 */
3544 		request = igb->num_rx_rings + 1;
3545 		minimum = 2;
3546 		IGB_DEBUGLOG_0(igb, "interrupt type: MSI-X");
3547 		break;
3548 
3549 	default:
3550 		igb_log(igb,
3551 		    "invalid call to igb_alloc_intr_handles(): %d\n",
3552 		    intr_type);
3553 		return (IGB_FAILURE);
3554 	}
3555 	IGB_DEBUGLOG_2(igb, "interrupt handles requested: %d  minimum: %d",
3556 	    request, minimum);
3557 
3558 	/*
3559 	 * Get number of supported interrupts
3560 	 */
3561 	rc = ddi_intr_get_nintrs(devinfo, intr_type, &count);
3562 	if ((rc != DDI_SUCCESS) || (count < minimum)) {
3563 		igb_log(igb,
3564 		    "Get supported interrupt number failed. "
3565 		    "Return: %d, count: %d", rc, count);
3566 		return (IGB_FAILURE);
3567 	}
3568 	IGB_DEBUGLOG_1(igb, "interrupts supported: %d", count);
3569 
3570 	/*
3571 	 * Get number of available interrupts
3572 	 */
3573 	rc = ddi_intr_get_navail(devinfo, intr_type, &avail);
3574 	if ((rc != DDI_SUCCESS) || (avail < minimum)) {
3575 		igb_log(igb,
3576 		    "Get available interrupt number failed. "
3577 		    "Return: %d, available: %d", rc, avail);
3578 		return (IGB_FAILURE);
3579 	}
3580 	IGB_DEBUGLOG_1(igb, "interrupts available: %d", avail);
3581 
3582 	if (avail < request) {
3583 		igb_log(igb, "Request %d handles, %d available",
3584 		    request, avail);
3585 		request = avail;
3586 	}
3587 
3588 	actual = 0;
3589 	igb->intr_cnt = 0;
3590 
3591 	/*
3592 	 * Allocate an array of interrupt handles
3593 	 */
3594 	igb->intr_size = request * sizeof (ddi_intr_handle_t);
3595 	igb->htable = kmem_alloc(igb->intr_size, KM_SLEEP);
3596 
3597 	rc = ddi_intr_alloc(devinfo, igb->htable, intr_type, 0,
3598 	    request, &actual, DDI_INTR_ALLOC_NORMAL);
3599 	if (rc != DDI_SUCCESS) {
3600 		igb_log(igb, "Allocate interrupts failed. "
3601 		    "return: %d, request: %d, actual: %d",
3602 		    rc, request, actual);
3603 		goto alloc_handle_fail;
3604 	}
3605 	IGB_DEBUGLOG_1(igb, "interrupts actually allocated: %d", actual);
3606 
3607 	igb->intr_cnt = actual;
3608 
3609 	if (actual < minimum) {
3610 		igb_log(igb, "Insufficient interrupt handles allocated: %d",
3611 		    actual);
3612 		goto alloc_handle_fail;
3613 	}
3614 
3615 	/*
3616 	 * For MSI-X, actual might force us to reduce number of rx rings
3617 	 */
3618 	if (intr_type == DDI_INTR_TYPE_MSIX) {
3619 		rx_rings = actual - 1;
3620 		if (rx_rings < igb->num_rx_rings) {
3621 			igb_log(igb,
3622 			    "MSI-X vectors force Rx queue number to %d",
3623 			    rx_rings);
3624 			igb->num_rx_rings = rx_rings;
3625 		}
3626 	}
3627 
3628 	/*
3629 	 * Get priority for first vector, assume remaining are all the same
3630 	 */
3631 	rc = ddi_intr_get_pri(igb->htable[0], &igb->intr_pri);
3632 	if (rc != DDI_SUCCESS) {
3633 		igb_log(igb,
3634 		    "Get interrupt priority failed: %d", rc);
3635 		goto alloc_handle_fail;
3636 	}
3637 
3638 	rc = ddi_intr_get_cap(igb->htable[0], &igb->intr_cap);
3639 	if (rc != DDI_SUCCESS) {
3640 		igb_log(igb,
3641 		    "Get interrupt cap failed: %d", rc);
3642 		goto alloc_handle_fail;
3643 	}
3644 
3645 	igb->intr_type = intr_type;
3646 
3647 	return (IGB_SUCCESS);
3648 
3649 alloc_handle_fail:
3650 	igb_rem_intrs(igb);
3651 
3652 	return (IGB_FAILURE);
3653 }
3654 
3655 /*
3656  * igb_add_intr_handlers - Add interrupt handlers based on the interrupt type
3657  *
3658  * Before adding the interrupt handlers, the interrupt vectors have
3659  * been allocated, and the rx/tx rings have also been allocated.
3660  */
3661 static int
3662 igb_add_intr_handlers(igb_t *igb)
3663 {
3664 	igb_rx_ring_t *rx_ring;
3665 	int vector;
3666 	int rc;
3667 	int i;
3668 
3669 	vector = 0;
3670 
3671 	switch (igb->intr_type) {
3672 	case DDI_INTR_TYPE_MSIX:
3673 		/* Add interrupt handler for tx + other */
3674 		rc = ddi_intr_add_handler(igb->htable[vector],
3675 		    (ddi_intr_handler_t *)igb_intr_tx_other,
3676 		    (void *)igb, NULL);
3677 		if (rc != DDI_SUCCESS) {
3678 			igb_log(igb,
3679 			    "Add tx/other interrupt handler failed: %d", rc);
3680 			return (IGB_FAILURE);
3681 		}
3682 		vector++;
3683 
3684 		/* Add interrupt handler for each rx ring */
3685 		for (i = 0; i < igb->num_rx_rings; i++) {
3686 			rx_ring = &igb->rx_rings[i];
3687 
3688 			rc = ddi_intr_add_handler(igb->htable[vector],
3689 			    (ddi_intr_handler_t *)igb_intr_rx,
3690 			    (void *)rx_ring, NULL);
3691 
3692 			if (rc != DDI_SUCCESS) {
3693 				igb_log(igb,
3694 				    "Add rx interrupt handler failed. "
3695 				    "return: %d, rx ring: %d", rc, i);
3696 				for (vector--; vector >= 0; vector--) {
3697 					(void) ddi_intr_remove_handler(
3698 					    igb->htable[vector]);
3699 				}
3700 				return (IGB_FAILURE);
3701 			}
3702 
3703 			rx_ring->intr_vector = vector;
3704 
3705 			vector++;
3706 		}
3707 		break;
3708 
3709 	case DDI_INTR_TYPE_MSI:
3710 		/* Add interrupt handlers for the only vector */
3711 		rc = ddi_intr_add_handler(igb->htable[vector],
3712 		    (ddi_intr_handler_t *)igb_intr_msi,
3713 		    (void *)igb, NULL);
3714 
3715 		if (rc != DDI_SUCCESS) {
3716 			igb_log(igb,
3717 			    "Add MSI interrupt handler failed: %d", rc);
3718 			return (IGB_FAILURE);
3719 		}
3720 
3721 		rx_ring = &igb->rx_rings[0];
3722 		rx_ring->intr_vector = vector;
3723 
3724 		vector++;
3725 		break;
3726 
3727 	case DDI_INTR_TYPE_FIXED:
3728 		/* Add interrupt handlers for the only vector */
3729 		rc = ddi_intr_add_handler(igb->htable[vector],
3730 		    (ddi_intr_handler_t *)igb_intr_legacy,
3731 		    (void *)igb, NULL);
3732 
3733 		if (rc != DDI_SUCCESS) {
3734 			igb_log(igb,
3735 			    "Add legacy interrupt handler failed: %d", rc);
3736 			return (IGB_FAILURE);
3737 		}
3738 
3739 		rx_ring = &igb->rx_rings[0];
3740 		rx_ring->intr_vector = vector;
3741 
3742 		vector++;
3743 		break;
3744 
3745 	default:
3746 		return (IGB_FAILURE);
3747 	}
3748 
3749 	ASSERT(vector == igb->intr_cnt);
3750 
3751 	return (IGB_SUCCESS);
3752 }
3753 
3754 /*
3755  * igb_setup_adapter_msix - setup the adapter to use MSI-X interrupts
3756  *
3757  * For each vector enabled on the adapter, Set the MSIXBM register accordingly
3758  */
3759 static void
3760 igb_setup_adapter_msix(igb_t *igb)
3761 {
3762 	uint32_t eims = 0;
3763 	int i, vector;
3764 	struct e1000_hw *hw = &igb->hw;
3765 
3766 	/*
3767 	 * Set vector for Tx + Other causes
3768 	 * NOTE assumption that there is only one of these and it is vector 0
3769 	 */
3770 	vector = 0;
3771 	igb->eims_mask = E1000_EICR_TX_QUEUE0 | E1000_EICR_OTHER;
3772 	E1000_WRITE_REG(hw, E1000_MSIXBM(vector), igb->eims_mask);
3773 
3774 	vector++;
3775 	for (i = 0; i < igb->num_rx_rings; i++) {
3776 		/*
3777 		 * Set vector for each rx ring
3778 		 */
3779 		eims = (E1000_EICR_RX_QUEUE0 << i);
3780 		E1000_WRITE_REG(hw, E1000_MSIXBM(vector), eims);
3781 
3782 		/*
3783 		 * Accumulate bits to enable in igb_enable_adapter_interrupts()
3784 		 */
3785 		igb->eims_mask |= eims;
3786 
3787 		vector++;
3788 	}
3789 
3790 	ASSERT(vector == igb->intr_cnt);
3791 
3792 	/*
3793 	 * Disable IAM for ICR interrupt bits
3794 	 */
3795 	E1000_WRITE_REG(hw, E1000_IAM, 0);
3796 	E1000_WRITE_FLUSH(hw);
3797 }
3798 
3799 /*
3800  * igb_rem_intr_handlers - remove the interrupt handlers
3801  */
3802 static void
3803 igb_rem_intr_handlers(igb_t *igb)
3804 {
3805 	int i;
3806 	int rc;
3807 
3808 	for (i = 0; i < igb->intr_cnt; i++) {
3809 		rc = ddi_intr_remove_handler(igb->htable[i]);
3810 		if (rc != DDI_SUCCESS) {
3811 			IGB_DEBUGLOG_1(igb,
3812 			    "Remove intr handler failed: %d", rc);
3813 		}
3814 	}
3815 }
3816 
3817 /*
3818  * igb_rem_intrs - remove the allocated interrupts
3819  */
3820 static void
3821 igb_rem_intrs(igb_t *igb)
3822 {
3823 	int i;
3824 	int rc;
3825 
3826 	for (i = 0; i < igb->intr_cnt; i++) {
3827 		rc = ddi_intr_free(igb->htable[i]);
3828 		if (rc != DDI_SUCCESS) {
3829 			IGB_DEBUGLOG_1(igb,
3830 			    "Free intr failed: %d", rc);
3831 		}
3832 	}
3833 
3834 	kmem_free(igb->htable, igb->intr_size);
3835 	igb->htable = NULL;
3836 }
3837 
3838 /*
3839  * igb_enable_intrs - enable all the ddi interrupts
3840  */
3841 static int
3842 igb_enable_intrs(igb_t *igb)
3843 {
3844 	int i;
3845 	int rc;
3846 
3847 	/* Enable interrupts */
3848 	if (igb->intr_cap & DDI_INTR_FLAG_BLOCK) {
3849 		/* Call ddi_intr_block_enable() for MSI */
3850 		rc = ddi_intr_block_enable(igb->htable, igb->intr_cnt);
3851 		if (rc != DDI_SUCCESS) {
3852 			igb_log(igb,
3853 			    "Enable block intr failed: %d", rc);
3854 			return (IGB_FAILURE);
3855 		}
3856 	} else {
3857 		/* Call ddi_intr_enable() for Legacy/MSI non block enable */
3858 		for (i = 0; i < igb->intr_cnt; i++) {
3859 			rc = ddi_intr_enable(igb->htable[i]);
3860 			if (rc != DDI_SUCCESS) {
3861 				igb_log(igb,
3862 				    "Enable intr failed: %d", rc);
3863 				return (IGB_FAILURE);
3864 			}
3865 		}
3866 	}
3867 
3868 	return (IGB_SUCCESS);
3869 }
3870 
3871 /*
3872  * igb_disable_intrs - disable all the ddi interrupts
3873  */
3874 static int
3875 igb_disable_intrs(igb_t *igb)
3876 {
3877 	int i;
3878 	int rc;
3879 
3880 	/* Disable all interrupts */
3881 	if (igb->intr_cap & DDI_INTR_FLAG_BLOCK) {
3882 		rc = ddi_intr_block_disable(igb->htable, igb->intr_cnt);
3883 		if (rc != DDI_SUCCESS) {
3884 			igb_log(igb,
3885 			    "Disable block intr failed: %d", rc);
3886 			return (IGB_FAILURE);
3887 		}
3888 	} else {
3889 		for (i = 0; i < igb->intr_cnt; i++) {
3890 			rc = ddi_intr_disable(igb->htable[i]);
3891 			if (rc != DDI_SUCCESS) {
3892 				igb_log(igb,
3893 				    "Disable intr failed: %d", rc);
3894 				return (IGB_FAILURE);
3895 			}
3896 		}
3897 	}
3898 
3899 	return (IGB_SUCCESS);
3900 }
3901 
3902 /*
3903  * igb_get_phy_state - Get and save the parameters read from PHY registers
3904  */
3905 static void
3906 igb_get_phy_state(igb_t *igb)
3907 {
3908 	struct e1000_hw *hw = &igb->hw;
3909 	uint16_t phy_ctrl;
3910 	uint16_t phy_status;
3911 	uint16_t phy_an_adv;
3912 	uint16_t phy_an_exp;
3913 	uint16_t phy_ext_status;
3914 	uint16_t phy_1000t_ctrl;
3915 	uint16_t phy_1000t_status;
3916 	uint16_t phy_lp_able;
3917 
3918 	ASSERT(mutex_owned(&igb->gen_lock));
3919 
3920 	(void) e1000_read_phy_reg(hw, PHY_CONTROL, &phy_ctrl);
3921 	(void) e1000_read_phy_reg(hw, PHY_STATUS, &phy_status);
3922 	(void) e1000_read_phy_reg(hw, PHY_AUTONEG_ADV, &phy_an_adv);
3923 	(void) e1000_read_phy_reg(hw, PHY_AUTONEG_EXP, &phy_an_exp);
3924 	(void) e1000_read_phy_reg(hw, PHY_EXT_STATUS, &phy_ext_status);
3925 	(void) e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_1000t_ctrl);
3926 	(void) e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_1000t_status);
3927 	(void) e1000_read_phy_reg(hw, PHY_LP_ABILITY, &phy_lp_able);
3928 
3929 	igb->param_autoneg_cap =
3930 	    (phy_status & MII_SR_AUTONEG_CAPS) ? 1 : 0;
3931 	igb->param_pause_cap =
3932 	    (phy_an_adv & NWAY_AR_PAUSE) ? 1 : 0;
3933 	igb->param_asym_pause_cap =
3934 	    (phy_an_adv & NWAY_AR_ASM_DIR) ? 1 : 0;
3935 	igb->param_1000fdx_cap = ((phy_ext_status & IEEE_ESR_1000T_FD_CAPS) ||
3936 	    (phy_ext_status & IEEE_ESR_1000X_FD_CAPS)) ? 1 : 0;
3937 	igb->param_1000hdx_cap = ((phy_ext_status & IEEE_ESR_1000T_HD_CAPS) ||
3938 	    (phy_ext_status & IEEE_ESR_1000X_HD_CAPS)) ? 1 : 0;
3939 	igb->param_100t4_cap =
3940 	    (phy_status & MII_SR_100T4_CAPS) ? 1 : 0;
3941 	igb->param_100fdx_cap = ((phy_status & MII_SR_100X_FD_CAPS) ||
3942 	    (phy_status & MII_SR_100T2_FD_CAPS)) ? 1 : 0;
3943 	igb->param_100hdx_cap = ((phy_status & MII_SR_100X_HD_CAPS) ||
3944 	    (phy_status & MII_SR_100T2_HD_CAPS)) ? 1 : 0;
3945 	igb->param_10fdx_cap =
3946 	    (phy_status & MII_SR_10T_FD_CAPS) ? 1 : 0;
3947 	igb->param_10hdx_cap =
3948 	    (phy_status & MII_SR_10T_HD_CAPS) ? 1 : 0;
3949 	igb->param_rem_fault =
3950 	    (phy_status & MII_SR_REMOTE_FAULT) ? 1 : 0;
3951 
3952 	igb->param_adv_autoneg_cap = hw->mac.autoneg;
3953 	igb->param_adv_pause_cap =
3954 	    (phy_an_adv & NWAY_AR_PAUSE) ? 1 : 0;
3955 	igb->param_adv_asym_pause_cap =
3956 	    (phy_an_adv & NWAY_AR_ASM_DIR) ? 1 : 0;
3957 	igb->param_adv_1000hdx_cap =
3958 	    (phy_1000t_ctrl & CR_1000T_HD_CAPS) ? 1 : 0;
3959 	igb->param_adv_100t4_cap =
3960 	    (phy_an_adv & NWAY_AR_100T4_CAPS) ? 1 : 0;
3961 	igb->param_adv_rem_fault =
3962 	    (phy_an_adv & NWAY_AR_REMOTE_FAULT) ? 1 : 0;
3963 	if (igb->param_adv_autoneg_cap == 1) {
3964 		igb->param_adv_1000fdx_cap =
3965 		    (phy_1000t_ctrl & CR_1000T_FD_CAPS) ? 1 : 0;
3966 		igb->param_adv_100fdx_cap =
3967 		    (phy_an_adv & NWAY_AR_100TX_FD_CAPS) ? 1 : 0;
3968 		igb->param_adv_100hdx_cap =
3969 		    (phy_an_adv & NWAY_AR_100TX_HD_CAPS) ? 1 : 0;
3970 		igb->param_adv_10fdx_cap =
3971 		    (phy_an_adv & NWAY_AR_10T_FD_CAPS) ? 1 : 0;
3972 		igb->param_adv_10hdx_cap =
3973 		    (phy_an_adv & NWAY_AR_10T_HD_CAPS) ? 1 : 0;
3974 	}
3975 
3976 	igb->param_lp_autoneg_cap =
3977 	    (phy_an_exp & NWAY_ER_LP_NWAY_CAPS) ? 1 : 0;
3978 	igb->param_lp_pause_cap =
3979 	    (phy_lp_able & NWAY_LPAR_PAUSE) ? 1 : 0;
3980 	igb->param_lp_asym_pause_cap =
3981 	    (phy_lp_able & NWAY_LPAR_ASM_DIR) ? 1 : 0;
3982 	igb->param_lp_1000fdx_cap =
3983 	    (phy_1000t_status & SR_1000T_LP_FD_CAPS) ? 1 : 0;
3984 	igb->param_lp_1000hdx_cap =
3985 	    (phy_1000t_status & SR_1000T_LP_HD_CAPS) ? 1 : 0;
3986 	igb->param_lp_100t4_cap =
3987 	    (phy_lp_able & NWAY_LPAR_100T4_CAPS) ? 1 : 0;
3988 	igb->param_lp_100fdx_cap =
3989 	    (phy_lp_able & NWAY_LPAR_100TX_FD_CAPS) ? 1 : 0;
3990 	igb->param_lp_100hdx_cap =
3991 	    (phy_lp_able & NWAY_LPAR_100TX_HD_CAPS) ? 1 : 0;
3992 	igb->param_lp_10fdx_cap =
3993 	    (phy_lp_able & NWAY_LPAR_10T_FD_CAPS) ? 1 : 0;
3994 	igb->param_lp_10hdx_cap =
3995 	    (phy_lp_able & NWAY_LPAR_10T_HD_CAPS) ? 1 : 0;
3996 	igb->param_lp_rem_fault =
3997 	    (phy_lp_able & NWAY_LPAR_REMOTE_FAULT) ? 1 : 0;
3998 }
3999 
4000 /*
4001  * igb_get_driver_control
4002  */
4003 static void
4004 igb_get_driver_control(struct e1000_hw *hw)
4005 {
4006 	uint32_t ctrl_ext;
4007 
4008 	/* Notify firmware that driver is in control of device */
4009 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
4010 	ctrl_ext |= E1000_CTRL_EXT_DRV_LOAD;
4011 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
4012 }
4013 
4014 /*
4015  * igb_release_driver_control
4016  */
4017 static void
4018 igb_release_driver_control(struct e1000_hw *hw)
4019 {
4020 	uint32_t ctrl_ext;
4021 
4022 	/* Notify firmware that driver is no longer in control of device */
4023 	ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
4024 	ctrl_ext &= ~E1000_CTRL_EXT_DRV_LOAD;
4025 	E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
4026 }
4027 
4028 /*
4029  * igb_atomic_reserve - Atomic decrease operation
4030  */
4031 int
4032 igb_atomic_reserve(uint32_t *count_p, uint32_t n)
4033 {
4034 	uint32_t oldval;
4035 	uint32_t newval;
4036 
4037 	/* ATOMICALLY */
4038 	do {
4039 		oldval = *count_p;
4040 		if (oldval < n)
4041 			return (-1);
4042 		newval = oldval - n;
4043 	} while (atomic_cas_32(count_p, oldval, newval) != oldval);
4044 
4045 	return (newval);
4046 }
4047 
4048 /*
4049  * FMA support
4050  */
4051 
4052 int
4053 igb_check_acc_handle(ddi_acc_handle_t handle)
4054 {
4055 	ddi_fm_error_t de;
4056 
4057 	ddi_fm_acc_err_get(handle, &de, DDI_FME_VERSION);
4058 	ddi_fm_acc_err_clear(handle, DDI_FME_VERSION);
4059 	return (de.fme_status);
4060 }
4061 
4062 int
4063 igb_check_dma_handle(ddi_dma_handle_t handle)
4064 {
4065 	ddi_fm_error_t de;
4066 
4067 	ddi_fm_dma_err_get(handle, &de, DDI_FME_VERSION);
4068 	return (de.fme_status);
4069 }
4070 
4071 /*
4072  * The IO fault service error handling callback function
4073  */
4074 /*ARGSUSED*/
4075 static int
4076 igb_fm_error_cb(dev_info_t *dip, ddi_fm_error_t *err, const void *impl_data)
4077 {
4078 	/*
4079 	 * as the driver can always deal with an error in any dma or
4080 	 * access handle, we can just return the fme_status value.
4081 	 */
4082 	pci_ereport_post(dip, err, NULL);
4083 	return (err->fme_status);
4084 }
4085 
4086 static void
4087 igb_fm_init(igb_t *igb)
4088 {
4089 	ddi_iblock_cookie_t iblk;
4090 	int fma_acc_flag, fma_dma_flag;
4091 
4092 	/* Only register with IO Fault Services if we have some capability */
4093 	if (igb->fm_capabilities & DDI_FM_ACCCHK_CAPABLE) {
4094 		igb_regs_acc_attr.devacc_attr_access = DDI_FLAGERR_ACC;
4095 		fma_acc_flag = 1;
4096 	} else {
4097 		igb_regs_acc_attr.devacc_attr_access = DDI_DEFAULT_ACC;
4098 		fma_acc_flag = 0;
4099 	}
4100 
4101 	if (igb->fm_capabilities & DDI_FM_DMACHK_CAPABLE) {
4102 		fma_dma_flag = 1;
4103 	} else {
4104 		fma_dma_flag = 0;
4105 	}
4106 
4107 	(void) igb_set_fma_flags(fma_acc_flag, fma_dma_flag);
4108 
4109 	if (igb->fm_capabilities) {
4110 
4111 		/* Register capabilities with IO Fault Services */
4112 		ddi_fm_init(igb->dip, &igb->fm_capabilities, &iblk);
4113 
4114 		/*
4115 		 * Initialize pci ereport capabilities if ereport capable
4116 		 */
4117 		if (DDI_FM_EREPORT_CAP(igb->fm_capabilities) ||
4118 		    DDI_FM_ERRCB_CAP(igb->fm_capabilities))
4119 			pci_ereport_setup(igb->dip);
4120 
4121 		/*
4122 		 * Register error callback if error callback capable
4123 		 */
4124 		if (DDI_FM_ERRCB_CAP(igb->fm_capabilities))
4125 			ddi_fm_handler_register(igb->dip,
4126 			    igb_fm_error_cb, (void*) igb);
4127 	}
4128 }
4129 
4130 static void
4131 igb_fm_fini(igb_t *igb)
4132 {
4133 	/* Only unregister FMA capabilities if we registered some */
4134 	if (igb->fm_capabilities) {
4135 
4136 		/*
4137 		 * Release any resources allocated by pci_ereport_setup()
4138 		 */
4139 		if (DDI_FM_EREPORT_CAP(igb->fm_capabilities) ||
4140 		    DDI_FM_ERRCB_CAP(igb->fm_capabilities))
4141 			pci_ereport_teardown(igb->dip);
4142 
4143 		/*
4144 		 * Un-register error callback if error callback capable
4145 		 */
4146 		if (DDI_FM_ERRCB_CAP(igb->fm_capabilities))
4147 			ddi_fm_handler_unregister(igb->dip);
4148 
4149 		/* Unregister from IO Fault Services */
4150 		ddi_fm_fini(igb->dip);
4151 	}
4152 }
4153 
4154 void
4155 igb_fm_ereport(igb_t *igb, char *detail)
4156 {
4157 	uint64_t ena;
4158 	char buf[FM_MAX_CLASS];
4159 
4160 	(void) snprintf(buf, FM_MAX_CLASS, "%s.%s", DDI_FM_DEVICE, detail);
4161 	ena = fm_ena_generate(0, FM_ENA_FMT1);
4162 	if (DDI_FM_EREPORT_CAP(igb->fm_capabilities)) {
4163 		ddi_fm_ereport_post(igb->dip, buf, ena, DDI_NOSLEEP,
4164 		    FM_VERSION, DATA_TYPE_UINT8, FM_EREPORT_VERS0, NULL);
4165 	}
4166 }
4167